block-raw-posix.c

   1 /*
   2  * Block driver for RAW files (posix)
   3  *
   4  * Copyright (c) 2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu-common.h"
  25 #include "qemu-timer.h"
  26 #include "qemu-char.h"
  27 #include "block_int.h"
  28 #include "compatfd.h"
  29 #include <assert.h>
  30 #ifdef CONFIG_AIO
  31 #include <aio.h>
  32 #endif
  33
  34 #ifdef CONFIG_COCOA
  35 #include <paths.h>
  36 #include <sys/param.h>
  37 #include <IOKit/IOKitLib.h>
  38 #include <IOKit/IOBSD.h>
  39 #include <IOKit/storage/IOMediaBSDClient.h>
  40 #include <IOKit/storage/IOMedia.h>
  41 #include <IOKit/storage/IOCDMedia.h>
  42 //#include <IOKit/storage/IOCDTypes.h>
  43 #include <CoreFoundation/CoreFoundation.h>
  44 #endif
  45
  46 #ifdef __sun__
  47 #define _POSIX_PTHREAD_SEMANTICS 1
  48 #include <signal.h>
  49 #include <sys/dkio.h>
  50 #endif
  51 #ifdef __linux__
  52 #include <sys/ioctl.h>
  53 #include <linux/cdrom.h>
  54 #include <linux/fd.h>
  55 #endif
  56 #ifdef __FreeBSD__
  57 #include <signal.h>
  58 #include <sys/disk.h>
  59 #endif
  60
  61 #ifdef __OpenBSD__
  62 #include <sys/ioctl.h>
  63 #include <sys/disklabel.h>
  64 #include <sys/dkio.h>
  65 #endif
  66
  67 //#define DEBUG_FLOPPY
  68
  69 //#define DEBUG_BLOCK
  70 #if defined(DEBUG_BLOCK)
  71 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (loglevel != 0)  \
  72     { fprintf(logfile, formatCstr, ##args); fflush(logfile); } } while (0)
  73 #else
  74 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
  75 #endif
  76
  77 #define FTYPE_FILE   0
  78 #define FTYPE_CD     1
  79 #define FTYPE_FD     2
  80
  81 #define ALIGNED_BUFFER_SIZE (32 * 512)
  82
  83 /* if the FD is not accessed during that time (in ms), we try to
  84    reopen it to see if the disk has been changed */
  85 #define FD_OPEN_TIMEOUT 1000
  86
  87 typedef struct BDRVRawState {
  88     int fd;
  89     int type;
  90     unsigned int lseek_err_cnt;
  91 #if defined(__linux__)
  92     /* linux floppy specific */
  93     int fd_open_flags;
  94     int64_t fd_open_time;
  95     int64_t fd_error_time;
  96     int fd_got_error;
  97     int fd_media_changed;
  98 #endif
  99 #if defined(O_DIRECT)
 100     uint8_t* aligned_buf;
 101 #endif
 102 } BDRVRawState;
 103
 104 static int fd_open(BlockDriverState *bs);
 105
 106 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 107 {
 108     BDRVRawState *s = bs->opaque;
 109     int fd, open_flags, ret;
 110
 111     s->lseek_err_cnt = 0;
 112
 113     open_flags = O_BINARY;
 114     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 115         open_flags |= O_RDWR;
 116     } else {
 117         open_flags |= O_RDONLY;
 118         bs->read_only = 1;
 119     }
 120     if (flags & BDRV_O_CREAT)
 121         open_flags |= O_CREAT | O_TRUNC;
 122 #ifdef O_DIRECT
 123     if (flags & BDRV_O_DIRECT)
 124         open_flags |= O_DIRECT;
 125 #endif
 126
 127     s->type = FTYPE_FILE;
 128
 129     fd = open(filename, open_flags, 0644);
 130     if (fd < 0) {
 131         ret = -errno;
 132         if (ret == -EROFS)
 133             ret = -EACCES;
 134         return ret;
 135     }
 136     s->fd = fd;
 137 #if defined(O_DIRECT)
 138     s->aligned_buf = NULL;
 139     if (flags & BDRV_O_DIRECT) {
 140         s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
 141         if (s->aligned_buf == NULL) {
 142             ret = -errno;
 143             close(fd);
 144             return ret;
 145         }
 146     }
 147 #endif
 148     return 0;
 149 }
 150
 151 /* XXX: use host sector size if necessary with:
 152 #ifdef DIOCGSECTORSIZE
 153         {
 154             unsigned int sectorsize = 512;
 155             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 156                 sectorsize > bufsize)
 157                 bufsize = sectorsize;
 158         }
 159 #endif
 160 #ifdef CONFIG_COCOA
 161         u_int32_t   blockSize = 512;
 162         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 163             bufsize = blockSize;
 164         }
 165 #endif
 166 */
 167
 168 /*
 169  * offset and count are in bytes, but must be multiples of 512 for files
 170  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 171  *
 172  * This function may be called without alignment if the caller ensures
 173  * that O_DIRECT is not in effect.
 174  */
 175 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 176                      uint8_t *buf, int count)
 177 {
 178     BDRVRawState *s = bs->opaque;
 179     int ret;
 180
 181     ret = fd_open(bs);
 182     if (ret < 0)
 183         return ret;
 184
 185     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 186         ++(s->lseek_err_cnt);
 187         if(s->lseek_err_cnt <= 10) {
 188             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 189                               "] lseek failed : %d = %s\n",
 190                               s->fd, bs->filename, offset, buf, count,
 191                               bs->total_sectors, errno, strerror(errno));
 192         }
 193         return -1;
 194     }
 195     s->lseek_err_cnt=0;
 196
 197     ret = read(s->fd, buf, count);
 198     if (ret == count)
 199         goto label__raw_read__success;
 200
 201     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 202                       "] read failed %d : %d = %s\n",
 203                       s->fd, bs->filename, offset, buf, count,
 204                       bs->total_sectors, ret, errno, strerror(errno));
 205
 206     /* Try harder for CDrom. */
 207     if (bs->type == BDRV_TYPE_CDROM) {
 208         lseek(s->fd, offset, SEEK_SET);
 209         ret = read(s->fd, buf, count);
 210         if (ret == count)
 211             goto label__raw_read__success;
 212         lseek(s->fd, offset, SEEK_SET);
 213         ret = read(s->fd, buf, count);
 214         if (ret == count)
 215             goto label__raw_read__success;
 216
 217         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 218                           "] retry read failed %d : %d = %s\n",
 219                           s->fd, bs->filename, offset, buf, count,
 220                           bs->total_sectors, ret, errno, strerror(errno));
 221     }
 222
 223 label__raw_read__success:
 224
 225     return ret;
 226 }
 227
 228 /*
 229  * offset and count are in bytes, but must be multiples of 512 for files
 230  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 231  *
 232  * This function may be called without alignment if the caller ensures
 233  * that O_DIRECT is not in effect.
 234  */
 235 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 236                       const uint8_t *buf, int count)
 237 {
 238     BDRVRawState *s = bs->opaque;
 239     int ret;
 240
 241     ret = fd_open(bs);
 242     if (ret < 0)
 243         return ret;
 244
 245     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 246         ++(s->lseek_err_cnt);
 247         if(s->lseek_err_cnt) {
 248             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 249                               PRId64 "] lseek failed : %d = %s\n",
 250                               s->fd, bs->filename, offset, buf, count,
 251                               bs->total_sectors, errno, strerror(errno));
 252         }
 253         return -1;
 254     }
 255     s->lseek_err_cnt = 0;
 256
 257     ret = write(s->fd, buf, count);
 258     if (ret == count)
 259         goto label__raw_write__success;
 260
 261     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 262                       "] write failed %d : %d = %s\n",
 263                       s->fd, bs->filename, offset, buf, count,
 264                       bs->total_sectors, ret, errno, strerror(errno));
 265
 266 label__raw_write__success:
 267
 268     return ret;
 269 }
 270
 271
 272 #if defined(O_DIRECT)
 273 /*
 274  * offset and count are in bytes and possibly not aligned. For files opened
 275  * with O_DIRECT, necessary alignments are ensured before calling
 276  * raw_pread_aligned to do the actual read.
 277  */
 278 static int raw_pread(BlockDriverState *bs, int64_t offset,
 279                      uint8_t *buf, int count)
 280 {
 281     BDRVRawState *s = bs->opaque;
 282     int size, ret, shift, sum;
 283
 284     sum = 0;
 285
 286     if (s->aligned_buf != NULL)  {
 287
 288         if (offset & 0x1ff) {
 289             /* align offset on a 512 bytes boundary */
 290
 291             shift = offset & 0x1ff;
 292             size = (shift + count + 0x1ff) & ~0x1ff;
 293             if (size > ALIGNED_BUFFER_SIZE)
 294                 size = ALIGNED_BUFFER_SIZE;
 295             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 296             if (ret < 0)
 297                 return ret;
 298
 299             size = 512 - shift;
 300             if (size > count)
 301                 size = count;
 302             memcpy(buf, s->aligned_buf + shift, size);
 303
 304             buf += size;
 305             offset += size;
 306             count -= size;
 307             sum += size;
 308
 309             if (count == 0)
 310                 return sum;
 311         }
 312         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 313
 314             /* read on aligned buffer */
 315
 316             while (count) {
 317
 318                 size = (count + 0x1ff) & ~0x1ff;
 319                 if (size > ALIGNED_BUFFER_SIZE)
 320                     size = ALIGNED_BUFFER_SIZE;
 321
 322                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 323                 if (ret < 0)
 324                     return ret;
 325
 326                 size = ret;
 327                 if (size > count)
 328                     size = count;
 329
 330                 memcpy(buf, s->aligned_buf, size);
 331
 332                 buf += size;
 333                 offset += size;
 334                 count -= size;
 335                 sum += size;
 336             }
 337
 338             return sum;
 339         }
 340     }
 341
 342     return raw_pread_aligned(bs, offset, buf, count) + sum;
 343 }
 344
 345 /*
 346  * offset and count are in bytes and possibly not aligned. For files opened
 347  * with O_DIRECT, necessary alignments are ensured before calling
 348  * raw_pwrite_aligned to do the actual write.
 349  */
 350 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 351                       const uint8_t *buf, int count)
 352 {
 353     BDRVRawState *s = bs->opaque;
 354     int size, ret, shift, sum;
 355
 356     sum = 0;
 357
 358     if (s->aligned_buf != NULL) {
 359
 360         if (offset & 0x1ff) {
 361             /* align offset on a 512 bytes boundary */
 362             shift = offset & 0x1ff;
 363             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 364             if (ret < 0)
 365                 return ret;
 366
 367             size = 512 - shift;
 368             if (size > count)
 369                 size = count;
 370             memcpy(s->aligned_buf + shift, buf, size);
 371
 372             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 373             if (ret < 0)
 374                 return ret;
 375
 376             buf += size;
 377             offset += size;
 378             count -= size;
 379             sum += size;
 380
 381             if (count == 0)
 382                 return sum;
 383         }
 384         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 385
 386             while ((size = (count & ~0x1ff)) != 0) {
 387
 388                 if (size > ALIGNED_BUFFER_SIZE)
 389                     size = ALIGNED_BUFFER_SIZE;
 390
 391                 memcpy(s->aligned_buf, buf, size);
 392
 393                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 394                 if (ret < 0)
 395                     return ret;
 396
 397                 buf += ret;
 398                 offset += ret;
 399                 count -= ret;
 400                 sum += ret;
 401             }
 402             /* here, count < 512 because (count & ~0x1ff) == 0 */
 403             if (count) {
 404                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 405                 if (ret < 0)
 406                     return ret;
 407                  memcpy(s->aligned_buf, buf, count);
 408
 409                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 410                  if (ret < 0)
 411                      return ret;
 412                  if (count < ret)
 413                      ret = count;
 414
 415                  sum += ret;
 416             }
 417             return sum;
 418         }
 419     }
 420     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 421 }
 422
 423 #else
 424 #define raw_pread raw_pread_aligned
 425 #define raw_pwrite raw_pwrite_aligned
 426 #endif
 427
 428
 429 #ifdef CONFIG_AIO
 430 /***********************************************************/
 431 /* Unix AIO using POSIX AIO */
 432
 433 typedef struct RawAIOCB {
 434     BlockDriverAIOCB common;
 435     struct aiocb aiocb;
 436     struct RawAIOCB *next;
 437     int ret;
 438 } RawAIOCB;
 439
 440 static int aio_sig_fd = -1;
 441 static int aio_sig_num = SIGUSR2;
 442 static RawAIOCB *first_aio; /* AIO issued */
 443 static int aio_initialized = 0;
 444
 445 static void qemu_aio_poll(void *opaque)
 446 {
 447     RawAIOCB *acb, **pacb;
 448     int ret;
 449     size_t offset;
 450     union {
 451         struct qemu_signalfd_siginfo siginfo;
 452         char buf[128];
 453     } sig;
 454
 455     /* try to read from signalfd, don't freak out if we can't read anything */
 456     offset = 0;
 457     while (offset < 128) {
 458         ssize_t len;
 459
 460         len = read(aio_sig_fd, sig.buf + offset, 128 - offset);
 461         if (len == -1 && errno == EINTR)
 462             continue;
 463         if (len == -1 && errno == EAGAIN) {
 464             /* there is no natural reason for this to happen,
 465              * so we'll spin hard until we get everything just
 466              * to be on the safe side. */
 467             if (offset > 0)
 468                 continue;
 469         }
 470
 471         offset += len;
 472     }
 473
 474     for(;;) {
 475         pacb = &first_aio;
 476         for(;;) {
 477             acb = *pacb;
 478             if (!acb)
 479                 goto the_end;
 480             ret = aio_error(&acb->aiocb);
 481             if (ret == ECANCELED) {
 482                 /* remove the request */
 483                 *pacb = acb->next;
 484                 qemu_aio_release(acb);
 485             } else if (ret != EINPROGRESS) {
 486                 /* end of aio */
 487                 if (ret == 0) {
 488                     ret = aio_return(&acb->aiocb);
 489                     if (ret == acb->aiocb.aio_nbytes)
 490                         ret = 0;
 491                     else
 492                         ret = -EINVAL;
 493                 } else {
 494                     ret = -ret;
 495                 }
 496                 /* remove the request */
 497                 *pacb = acb->next;
 498                 /* call the callback */
 499                 acb->common.cb(acb->common.opaque, ret);
 500                 qemu_aio_release(acb);
 501                 break;
 502             } else {
 503                 pacb = &acb->next;
 504             }
 505         }
 506     }
 507  the_end: ;
 508 }
 509
 510 void qemu_aio_init(void)
 511 {
 512     sigset_t mask;
 513
 514     aio_initialized = 1;
 515
 516     /* Make sure to block AIO signal */
 517     sigemptyset(&mask);
 518     sigaddset(&mask, aio_sig_num);
 519     sigprocmask(SIG_BLOCK, &mask, NULL);
 520
 521     aio_sig_fd = qemu_signalfd(&mask);
 522
 523     fcntl(aio_sig_fd, F_SETFL, O_NONBLOCK);
 524
 525     qemu_set_fd_handler2(aio_sig_fd, NULL, qemu_aio_poll, NULL, NULL);
 526
 527 #if defined(__GLIBC__) && defined(__linux__)
 528     {
 529         /* XXX: aio thread exit seems to hang on RedHat 9 and this init
 530            seems to fix the problem. */
 531         struct aioinit ai;
 532         memset(&ai, 0, sizeof(ai));
 533         ai.aio_threads = 1;
 534         ai.aio_num = 1;
 535         ai.aio_idle_time = 365 * 100000;
 536         aio_init(&ai);
 537     }
 538 #endif
 539 }
 540
 541 /* Wait for all IO requests to complete.  */
 542 void qemu_aio_flush(void)
 543 {
 544     qemu_aio_poll(NULL);
 545     while (first_aio) {
 546         qemu_aio_wait();
 547     }
 548 }
 549
 550 void qemu_aio_wait(void)
 551 {
 552     int ret;
 553
 554     if (qemu_bh_poll())
 555         return;
 556
 557     if (!first_aio)
 558         return;
 559
 560     do {
 561         fd_set rdfds;
 562
 563         FD_ZERO(&rdfds);
 564         FD_SET(aio_sig_fd, &rdfds);
 565
 566         ret = select(aio_sig_fd + 1, &rdfds, NULL, NULL, NULL);
 567         if (ret == -1 && errno == EINTR)
 568             continue;
 569     } while (ret == 0);
 570
 571     qemu_aio_poll(NULL);
 572 }
 573
 574 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 575         int64_t sector_num, uint8_t *buf, int nb_sectors,
 576         BlockDriverCompletionFunc *cb, void *opaque)
 577 {
 578     BDRVRawState *s = bs->opaque;
 579     RawAIOCB *acb;
 580
 581     if (fd_open(bs) < 0)
 582         return NULL;
 583
 584     acb = qemu_aio_get(bs, cb, opaque);
 585     if (!acb)
 586         return NULL;
 587     acb->aiocb.aio_fildes = s->fd;
 588     acb->aiocb.aio_sigevent.sigev_signo = aio_sig_num;
 589     acb->aiocb.aio_sigevent.sigev_notify = SIGEV_SIGNAL;
 590     acb->aiocb.aio_buf = buf;
 591     if (nb_sectors < 0)
 592         acb->aiocb.aio_nbytes = -nb_sectors;
 593     else
 594         acb->aiocb.aio_nbytes = nb_sectors * 512;
 595     acb->aiocb.aio_offset = sector_num * 512;
 596     acb->next = first_aio;
 597     first_aio = acb;
 598     return acb;
 599 }
 600
 601 static void raw_aio_em_cb(void* opaque)
 602 {
 603     RawAIOCB *acb = opaque;
 604     acb->common.cb(acb->common.opaque, acb->ret);
 605     qemu_aio_release(acb);
 606 }
 607
 608 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 609         int64_t sector_num, uint8_t *buf, int nb_sectors,
 610         BlockDriverCompletionFunc *cb, void *opaque)
 611 {
 612     RawAIOCB *acb;
 613
 614     /*
 615      * If O_DIRECT is used and the buffer is not aligned fall back
 616      * to synchronous IO.
 617      */
 618 #if defined(O_DIRECT)
 619     BDRVRawState *s = bs->opaque;
 620
 621     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 622         QEMUBH *bh;
 623         acb = qemu_aio_get(bs, cb, opaque);
 624         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 625         bh = qemu_bh_new(raw_aio_em_cb, acb);
 626         qemu_bh_schedule(bh);
 627         return &acb->common;
 628     }
 629 #endif
 630
 631     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 632     if (!acb)
 633         return NULL;
 634     if (aio_read(&acb->aiocb) < 0) {
 635         qemu_aio_release(acb);
 636         return NULL;
 637     }
 638     return &acb->common;
 639 }
 640
 641 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 642         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 643         BlockDriverCompletionFunc *cb, void *opaque)
 644 {
 645     RawAIOCB *acb;
 646
 647     /*
 648      * If O_DIRECT is used and the buffer is not aligned fall back
 649      * to synchronous IO.
 650      */
 651 #if defined(O_DIRECT)
 652     BDRVRawState *s = bs->opaque;
 653
 654     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 655         QEMUBH *bh;
 656         acb = qemu_aio_get(bs, cb, opaque);
 657         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 658         bh = qemu_bh_new(raw_aio_em_cb, acb);
 659         qemu_bh_schedule(bh);
 660         return &acb->common;
 661     }
 662 #endif
 663
 664     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 665     if (!acb)
 666         return NULL;
 667     if (aio_write(&acb->aiocb) < 0) {
 668         qemu_aio_release(acb);
 669         return NULL;
 670     }
 671     return &acb->common;
 672 }
 673
 674 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 675 {
 676     int ret;
 677     RawAIOCB *acb = (RawAIOCB *)blockacb;
 678     RawAIOCB **pacb;
 679
 680     ret = aio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 681     if (ret == AIO_NOTCANCELED) {
 682         /* fail safe: if the aio could not be canceled, we wait for
 683            it */
 684         while (aio_error(&acb->aiocb) == EINPROGRESS);
 685     }
 686
 687     /* remove the callback from the queue */
 688     pacb = &first_aio;
 689     for(;;) {
 690         if (*pacb == NULL) {
 691             break;
 692         } else if (*pacb == acb) {
 693             *pacb = acb->next;
 694             qemu_aio_release(acb);
 695             break;
 696         }
 697         pacb = &acb->next;
 698     }
 699 }
 700
 701 # else /* CONFIG_AIO */
 702
 703 void qemu_aio_init(void)
 704 {
 705 }
 706
 707 void qemu_aio_flush(void)
 708 {
 709 }
 710
 711 void qemu_aio_wait(void)
 712 {
 713     qemu_bh_poll();
 714 }
 715
 716 #endif /* CONFIG_AIO */
 717
 718 static void raw_close(BlockDriverState *bs)
 719 {
 720     BDRVRawState *s = bs->opaque;
 721     if (s->fd >= 0) {
 722         close(s->fd);
 723         s->fd = -1;
 724 #if defined(O_DIRECT)
 725         if (s->aligned_buf != NULL)
 726             qemu_free(s->aligned_buf);
 727 #endif
 728     }
 729 }
 730
 731 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 732 {
 733     BDRVRawState *s = bs->opaque;
 734     if (s->type != FTYPE_FILE)
 735         return -ENOTSUP;
 736     if (ftruncate(s->fd, offset) < 0)
 737         return -errno;
 738     return 0;
 739 }
 740
 741 #ifdef __OpenBSD__
 742 static int64_t raw_getlength(BlockDriverState *bs)
 743 {
 744     BDRVRawState *s = bs->opaque;
 745     int fd = s->fd;
 746     struct stat st;
 747
 748     if (fstat(fd, &st))
 749         return -1;
 750     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 751         struct disklabel dl;
 752
 753         if (ioctl(fd, DIOCGDINFO, &dl))
 754             return -1;
 755         return (uint64_t)dl.d_secsize *
 756             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 757     } else
 758         return st.st_size;
 759 }
 760 #else /* !__OpenBSD__ */
 761 static int64_t  raw_getlength(BlockDriverState *bs)
 762 {
 763     BDRVRawState *s = bs->opaque;
 764     int fd = s->fd;
 765     int64_t size;
 766 #ifdef _BSD
 767     struct stat sb;
 768 #endif
 769 #ifdef __sun__
 770     struct dk_minfo minfo;
 771     int rv;
 772 #endif
 773     int ret;
 774
 775     ret = fd_open(bs);
 776     if (ret < 0)
 777         return ret;
 778
 779 #ifdef _BSD
 780     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 781 #ifdef DIOCGMEDIASIZE
 782         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 783 #endif
 784 #ifdef CONFIG_COCOA
 785         size = LONG_LONG_MAX;
 786 #else
 787         size = lseek(fd, 0LL, SEEK_END);
 788 #endif
 789     } else
 790 #endif
 791 #ifdef __sun__
 792     /*
 793      * use the DKIOCGMEDIAINFO ioctl to read the size.
 794      */
 795     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 796     if ( rv != -1 ) {
 797         size = minfo.dki_lbsize * minfo.dki_capacity;
 798     } else /* there are reports that lseek on some devices
 799               fails, but irc discussion said that contingency
 800               on contingency was overkill */
 801 #endif
 802     {
 803         size = lseek(fd, 0, SEEK_END);
 804     }
 805     return size;
 806 }
 807 #endif
 808
 809 static int raw_create(const char *filename, int64_t total_size,
 810                       const char *backing_file, int flags)
 811 {
 812     int fd;
 813
 814     if (flags || backing_file)
 815         return -ENOTSUP;
 816
 817     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 818               0644);
 819     if (fd < 0)
 820         return -EIO;
 821     ftruncate(fd, total_size * 512);
 822     close(fd);
 823     return 0;
 824 }
 825
 826 static void raw_flush(BlockDriverState *bs)
 827 {
 828     BDRVRawState *s = bs->opaque;
 829     fsync(s->fd);
 830 }
 831
 832 BlockDriver bdrv_raw = {
 833     "raw",
 834     sizeof(BDRVRawState),
 835     NULL, /* no probe for protocols */
 836     raw_open,
 837     NULL,
 838     NULL,
 839     raw_close,
 840     raw_create,
 841     raw_flush,
 842
 843 #ifdef CONFIG_AIO
 844     .bdrv_aio_read = raw_aio_read,
 845     .bdrv_aio_write = raw_aio_write,
 846     .bdrv_aio_cancel = raw_aio_cancel,
 847     .aiocb_size = sizeof(RawAIOCB),
 848 #endif
 849     .bdrv_pread = raw_pread,
 850     .bdrv_pwrite = raw_pwrite,
 851     .bdrv_truncate = raw_truncate,
 852     .bdrv_getlength = raw_getlength,
 853 };
 854
 855 /***********************************************/
 856 /* host device */
 857
 858 #ifdef CONFIG_COCOA
 859 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 860 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 861
 862 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 863 {
 864     kern_return_t       kernResult;
 865     mach_port_t     masterPort;
 866     CFMutableDictionaryRef  classesToMatch;
 867
 868     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 869     if ( KERN_SUCCESS != kernResult ) {
 870         printf( "IOMasterPort returned %d\n", kernResult );
 871     }
 872
 873     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 874     if ( classesToMatch == NULL ) {
 875         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 876     } else {
 877     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 878     }
 879     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 880     if ( KERN_SUCCESS != kernResult )
 881     {
 882         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 883     }
 884
 885     return kernResult;
 886 }
 887
 888 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 889 {
 890     io_object_t     nextMedia;
 891     kern_return_t   kernResult = KERN_FAILURE;
 892     *bsdPath = '\0';
 893     nextMedia = IOIteratorNext( mediaIterator );
 894     if ( nextMedia )
 895     {
 896         CFTypeRef   bsdPathAsCFString;
 897     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 898         if ( bsdPathAsCFString ) {
 899             size_t devPathLength;
 900             strcpy( bsdPath, _PATH_DEV );
 901             strcat( bsdPath, "r" );
 902             devPathLength = strlen( bsdPath );
 903             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 904                 kernResult = KERN_SUCCESS;
 905             }
 906             CFRelease( bsdPathAsCFString );
 907         }
 908         IOObjectRelease( nextMedia );
 909     }
 910
 911     return kernResult;
 912 }
 913
 914 #endif
 915
 916 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 917 {
 918     BDRVRawState *s = bs->opaque;
 919     int fd, open_flags, ret;
 920
 921 #ifdef CONFIG_COCOA
 922     if (strstart(filename, "/dev/cdrom", NULL)) {
 923         kern_return_t kernResult;
 924         io_iterator_t mediaIterator;
 925         char bsdPath[ MAXPATHLEN ];
 926         int fd;
 927
 928         kernResult = FindEjectableCDMedia( &mediaIterator );
 929         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 930
 931         if ( bsdPath[ 0 ] != '\0' ) {
 932             strcat(bsdPath,"s0");
 933             /* some CDs don't have a partition 0 */
 934             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 935             if (fd < 0) {
 936                 bsdPath[strlen(bsdPath)-1] = '1';
 937             } else {
 938                 close(fd);
 939             }
 940             filename = bsdPath;
 941         }
 942
 943         if ( mediaIterator )
 944             IOObjectRelease( mediaIterator );
 945     }
 946 #endif
 947     open_flags = O_BINARY;
 948     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 949         open_flags |= O_RDWR;
 950     } else {
 951         open_flags |= O_RDONLY;
 952         bs->read_only = 1;
 953     }
 954 #ifdef O_DIRECT
 955     if (flags & BDRV_O_DIRECT)
 956         open_flags |= O_DIRECT;
 957 #endif
 958
 959     s->type = FTYPE_FILE;
 960 #if defined(__linux__)
 961     if (strstart(filename, "/dev/cd", NULL)) {
 962         /* open will not fail even if no CD is inserted */
 963         open_flags |= O_NONBLOCK;
 964         s->type = FTYPE_CD;
 965     } else if (strstart(filename, "/dev/fd", NULL)) {
 966         s->type = FTYPE_FD;
 967         s->fd_open_flags = open_flags;
 968         /* open will not fail even if no floppy is inserted */
 969         open_flags |= O_NONBLOCK;
 970     } else if (strstart(filename, "/dev/sg", NULL)) {
 971         bs->sg = 1;
 972     }
 973 #endif
 974     fd = open(filename, open_flags, 0644);
 975     if (fd < 0) {
 976         ret = -errno;
 977         if (ret == -EROFS)
 978             ret = -EACCES;
 979         return ret;
 980     }
 981     s->fd = fd;
 982 #if defined(__linux__)
 983     /* close fd so that we can reopen it as needed */
 984     if (s->type == FTYPE_FD) {
 985         close(s->fd);
 986         s->fd = -1;
 987         s->fd_media_changed = 1;
 988     }
 989 #endif
 990     return 0;
 991 }
 992
 993 #if defined(__linux__)
 994
 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     "host_device",
1184     sizeof(BDRVRawState),
1185     NULL, /* no probe for protocols */
1186     hdev_open,
1187     NULL,
1188     NULL,
1189     raw_close,
1190     NULL,
1191     raw_flush,
1192
1193 #ifdef CONFIG_AIO
1194     .bdrv_aio_read = raw_aio_read,
1195     .bdrv_aio_write = raw_aio_write,
1196     .bdrv_aio_cancel = raw_aio_cancel,
1197     .aiocb_size = sizeof(RawAIOCB),
1198 #endif
1199     .bdrv_pread = raw_pread,
1200     .bdrv_pwrite = raw_pwrite,
1201     .bdrv_getlength = raw_getlength,
1202
1203     /* removable device support */
1204     .bdrv_is_inserted = raw_is_inserted,
1205     .bdrv_media_changed = raw_media_changed,
1206     .bdrv_eject = raw_eject,
1207     .bdrv_set_locked = raw_set_locked,
1208     /* generic scsi device */
1209     .bdrv_ioctl = raw_ioctl,
1210 };