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 /* posix-aio doesn't allow multiple outstanding requests to a single file
  88  * descriptor.  we implement a pool of dup()'d file descriptors to work
  89  * around this */
  90 #define RAW_FD_POOL_SIZE        64
  91
  92 typedef struct BDRVRawState {
  93     int fd;
  94     int type;
  95     unsigned int lseek_err_cnt;
  96     int fd_pool[RAW_FD_POOL_SIZE];
  97 #if defined(__linux__)
  98     /* linux floppy specific */
  99     int fd_open_flags;
 100     int64_t fd_open_time;
 101     int64_t fd_error_time;
 102     int fd_got_error;
 103     int fd_media_changed;
 104 #endif
 105 #if defined(O_DIRECT)
 106     uint8_t* aligned_buf;
 107 #endif
 108 } BDRVRawState;
 109
 110 static int posix_aio_init(void);
 111
 112 static int fd_open(BlockDriverState *bs);
 113
 114 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 115 {
 116     BDRVRawState *s = bs->opaque;
 117     int fd, open_flags, ret;
 118     int i;
 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 #ifdef O_DIRECT
 134     if (flags & BDRV_O_DIRECT)
 135         open_flags |= O_DIRECT;
 136 #endif
 137
 138     s->type = FTYPE_FILE;
 139
 140     fd = open(filename, open_flags, 0644);
 141     if (fd < 0) {
 142         ret = -errno;
 143         if (ret == -EROFS)
 144             ret = -EACCES;
 145         return ret;
 146     }
 147     s->fd = fd;
 148     for (i = 0; i < RAW_FD_POOL_SIZE; i++)
 149         s->fd_pool[i] = -1;
 150 #if defined(O_DIRECT)
 151     s->aligned_buf = NULL;
 152     if (flags & BDRV_O_DIRECT) {
 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 #endif
 161     return 0;
 162 }
 163
 164 /* XXX: use host sector size if necessary with:
 165 #ifdef DIOCGSECTORSIZE
 166         {
 167             unsigned int sectorsize = 512;
 168             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 169                 sectorsize > bufsize)
 170                 bufsize = sectorsize;
 171         }
 172 #endif
 173 #ifdef CONFIG_COCOA
 174         u_int32_t   blockSize = 512;
 175         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 176             bufsize = blockSize;
 177         }
 178 #endif
 179 */
 180
 181 /*
 182  * offset and count are in bytes, but must be multiples of 512 for files
 183  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 184  *
 185  * This function may be called without alignment if the caller ensures
 186  * that O_DIRECT is not in effect.
 187  */
 188 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 189                      uint8_t *buf, int count)
 190 {
 191     BDRVRawState *s = bs->opaque;
 192     int ret;
 193
 194     ret = fd_open(bs);
 195     if (ret < 0)
 196         return ret;
 197
 198     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 199         ++(s->lseek_err_cnt);
 200         if(s->lseek_err_cnt <= 10) {
 201             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 202                               "] lseek failed : %d = %s\n",
 203                               s->fd, bs->filename, offset, buf, count,
 204                               bs->total_sectors, errno, strerror(errno));
 205         }
 206         return -1;
 207     }
 208     s->lseek_err_cnt=0;
 209
 210     ret = read(s->fd, buf, count);
 211     if (ret == count)
 212         goto label__raw_read__success;
 213
 214     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 215                       "] read failed %d : %d = %s\n",
 216                       s->fd, bs->filename, offset, buf, count,
 217                       bs->total_sectors, ret, errno, strerror(errno));
 218
 219     /* Try harder for CDrom. */
 220     if (bs->type == BDRV_TYPE_CDROM) {
 221         lseek(s->fd, offset, SEEK_SET);
 222         ret = read(s->fd, buf, count);
 223         if (ret == count)
 224             goto label__raw_read__success;
 225         lseek(s->fd, offset, SEEK_SET);
 226         ret = read(s->fd, buf, count);
 227         if (ret == count)
 228             goto label__raw_read__success;
 229
 230         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 231                           "] retry read failed %d : %d = %s\n",
 232                           s->fd, bs->filename, offset, buf, count,
 233                           bs->total_sectors, ret, errno, strerror(errno));
 234     }
 235
 236 label__raw_read__success:
 237
 238     return ret;
 239 }
 240
 241 /*
 242  * offset and count are in bytes, but must be multiples of 512 for files
 243  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 244  *
 245  * This function may be called without alignment if the caller ensures
 246  * that O_DIRECT is not in effect.
 247  */
 248 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 249                       const uint8_t *buf, int count)
 250 {
 251     BDRVRawState *s = bs->opaque;
 252     int ret;
 253
 254     ret = fd_open(bs);
 255     if (ret < 0)
 256         return ret;
 257
 258     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 259         ++(s->lseek_err_cnt);
 260         if(s->lseek_err_cnt) {
 261             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 262                               PRId64 "] lseek failed : %d = %s\n",
 263                               s->fd, bs->filename, offset, buf, count,
 264                               bs->total_sectors, errno, strerror(errno));
 265         }
 266         return -1;
 267     }
 268     s->lseek_err_cnt = 0;
 269
 270     ret = write(s->fd, buf, count);
 271     if (ret == count)
 272         goto label__raw_write__success;
 273
 274     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 275                       "] write failed %d : %d = %s\n",
 276                       s->fd, bs->filename, offset, buf, count,
 277                       bs->total_sectors, ret, errno, strerror(errno));
 278
 279 label__raw_write__success:
 280
 281     return ret;
 282 }
 283
 284
 285 #if defined(O_DIRECT)
 286 /*
 287  * offset and count are in bytes and possibly not aligned. For files opened
 288  * with O_DIRECT, necessary alignments are ensured before calling
 289  * raw_pread_aligned to do the actual read.
 290  */
 291 static int raw_pread(BlockDriverState *bs, int64_t offset,
 292                      uint8_t *buf, int count)
 293 {
 294     BDRVRawState *s = bs->opaque;
 295     int size, ret, shift, sum;
 296
 297     sum = 0;
 298
 299     if (s->aligned_buf != NULL)  {
 300
 301         if (offset & 0x1ff) {
 302             /* align offset on a 512 bytes boundary */
 303
 304             shift = offset & 0x1ff;
 305             size = (shift + count + 0x1ff) & ~0x1ff;
 306             if (size > ALIGNED_BUFFER_SIZE)
 307                 size = ALIGNED_BUFFER_SIZE;
 308             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 309             if (ret < 0)
 310                 return ret;
 311
 312             size = 512 - shift;
 313             if (size > count)
 314                 size = count;
 315             memcpy(buf, s->aligned_buf + shift, size);
 316
 317             buf += size;
 318             offset += size;
 319             count -= size;
 320             sum += size;
 321
 322             if (count == 0)
 323                 return sum;
 324         }
 325         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 326
 327             /* read on aligned buffer */
 328
 329             while (count) {
 330
 331                 size = (count + 0x1ff) & ~0x1ff;
 332                 if (size > ALIGNED_BUFFER_SIZE)
 333                     size = ALIGNED_BUFFER_SIZE;
 334
 335                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 336                 if (ret < 0)
 337                     return ret;
 338
 339                 size = ret;
 340                 if (size > count)
 341                     size = count;
 342
 343                 memcpy(buf, s->aligned_buf, size);
 344
 345                 buf += size;
 346                 offset += size;
 347                 count -= size;
 348                 sum += size;
 349             }
 350
 351             return sum;
 352         }
 353     }
 354
 355     return raw_pread_aligned(bs, offset, buf, count) + sum;
 356 }
 357
 358 /*
 359  * offset and count are in bytes and possibly not aligned. For files opened
 360  * with O_DIRECT, necessary alignments are ensured before calling
 361  * raw_pwrite_aligned to do the actual write.
 362  */
 363 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 364                       const uint8_t *buf, int count)
 365 {
 366     BDRVRawState *s = bs->opaque;
 367     int size, ret, shift, sum;
 368
 369     sum = 0;
 370
 371     if (s->aligned_buf != NULL) {
 372
 373         if (offset & 0x1ff) {
 374             /* align offset on a 512 bytes boundary */
 375             shift = offset & 0x1ff;
 376             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 377             if (ret < 0)
 378                 return ret;
 379
 380             size = 512 - shift;
 381             if (size > count)
 382                 size = count;
 383             memcpy(s->aligned_buf + shift, buf, size);
 384
 385             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 386             if (ret < 0)
 387                 return ret;
 388
 389             buf += size;
 390             offset += size;
 391             count -= size;
 392             sum += size;
 393
 394             if (count == 0)
 395                 return sum;
 396         }
 397         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 398
 399             while ((size = (count & ~0x1ff)) != 0) {
 400
 401                 if (size > ALIGNED_BUFFER_SIZE)
 402                     size = ALIGNED_BUFFER_SIZE;
 403
 404                 memcpy(s->aligned_buf, buf, size);
 405
 406                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 407                 if (ret < 0)
 408                     return ret;
 409
 410                 buf += ret;
 411                 offset += ret;
 412                 count -= ret;
 413                 sum += ret;
 414             }
 415             /* here, count < 512 because (count & ~0x1ff) == 0 */
 416             if (count) {
 417                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 418                 if (ret < 0)
 419                     return ret;
 420                  memcpy(s->aligned_buf, buf, count);
 421
 422                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 423                  if (ret < 0)
 424                      return ret;
 425                  if (count < ret)
 426                      ret = count;
 427
 428                  sum += ret;
 429             }
 430             return sum;
 431         }
 432     }
 433     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 434 }
 435
 436 #else
 437 #define raw_pread raw_pread_aligned
 438 #define raw_pwrite raw_pwrite_aligned
 439 #endif
 440
 441
 442 #ifdef CONFIG_AIO
 443 /***********************************************************/
 444 /* Unix AIO using POSIX AIO */
 445
 446 typedef struct RawAIOCB {
 447     BlockDriverAIOCB common;
 448     int fd;
 449     struct aiocb aiocb;
 450     struct RawAIOCB *next;
 451     int ret;
 452 } RawAIOCB;
 453
 454 typedef struct PosixAioState
 455 {
 456     int fd;
 457     RawAIOCB *first_aio;
 458 } PosixAioState;
 459
 460 static int raw_fd_pool_get(BDRVRawState *s)
 461 {
 462     int i;
 463
 464     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 465         /* already in use */
 466         if (s->fd_pool[i] != -1)
 467             continue;
 468
 469         /* try to dup file descriptor */
 470         s->fd_pool[i] = dup(s->fd);
 471         if (s->fd_pool[i] != -1)
 472             return s->fd_pool[i];
 473     }
 474
 475     /* we couldn't dup the file descriptor so just use the main one */
 476     return s->fd;
 477 }
 478
 479 static void raw_fd_pool_put(RawAIOCB *acb)
 480 {
 481     BDRVRawState *s = acb->common.bs->opaque;
 482     int i;
 483
 484     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 485         if (s->fd_pool[i] == acb->fd) {
 486             close(s->fd_pool[i]);
 487             s->fd_pool[i] = -1;
 488         }
 489     }
 490 }
 491
 492 static void posix_aio_read(void *opaque)
 493 {
 494     PosixAioState *s = opaque;
 495     RawAIOCB *acb, **pacb;
 496     int ret;
 497     size_t offset;
 498     union {
 499         struct qemu_signalfd_siginfo siginfo;
 500         char buf[128];
 501     } sig;
 502
 503     /* try to read from signalfd, don't freak out if we can't read anything */
 504     offset = 0;
 505     while (offset < 128) {
 506         ssize_t len;
 507
 508         len = read(s->fd, sig.buf + offset, 128 - offset);
 509         if (len == -1 && errno == EINTR)
 510             continue;
 511         if (len == -1 && errno == EAGAIN) {
 512             /* there is no natural reason for this to happen,
 513              * so we'll spin hard until we get everything just
 514              * to be on the safe side. */
 515             if (offset > 0)
 516                 continue;
 517         }
 518
 519         offset += len;
 520     }
 521
 522     for(;;) {
 523         pacb = &s->first_aio;
 524         for(;;) {
 525             acb = *pacb;
 526             if (!acb)
 527                 goto the_end;
 528             ret = aio_error(&acb->aiocb);
 529             if (ret == ECANCELED) {
 530                 /* remove the request */
 531                 *pacb = acb->next;
 532                 raw_fd_pool_put(acb);
 533                 qemu_aio_release(acb);
 534             } else if (ret != EINPROGRESS) {
 535                 /* end of aio */
 536                 if (ret == 0) {
 537                     ret = aio_return(&acb->aiocb);
 538                     if (ret == acb->aiocb.aio_nbytes)
 539                         ret = 0;
 540                     else
 541                         ret = -EINVAL;
 542                 } else {
 543                     ret = -ret;
 544                 }
 545                 /* remove the request */
 546                 *pacb = acb->next;
 547                 /* call the callback */
 548                 acb->common.cb(acb->common.opaque, ret);
 549                 raw_fd_pool_put(acb);
 550                 qemu_aio_release(acb);
 551                 break;
 552             } else {
 553                 pacb = &acb->next;
 554             }
 555         }
 556     }
 557  the_end: ;
 558 }
 559
 560 static int posix_aio_flush(void *opaque)
 561 {
 562     PosixAioState *s = opaque;
 563     return !!s->first_aio;
 564 }
 565
 566 static PosixAioState *posix_aio_state;
 567
 568 static int posix_aio_init(void)
 569 {
 570     sigset_t mask;
 571     PosixAioState *s;
 572
 573     if (posix_aio_state)
 574         return 0;
 575
 576     s = qemu_malloc(sizeof(PosixAioState));
 577     if (s == NULL)
 578         return -ENOMEM;
 579
 580     /* Make sure to block AIO signal */
 581     sigemptyset(&mask);
 582     sigaddset(&mask, SIGUSR2);
 583     sigprocmask(SIG_BLOCK, &mask, NULL);
 584
 585     s->first_aio = NULL;
 586     s->fd = qemu_signalfd(&mask);
 587     if (s->fd == -1) {
 588         fprintf(stderr, "failed to create signalfd\n");
 589         return -errno;
 590     }
 591
 592     fcntl(s->fd, F_SETFL, O_NONBLOCK);
 593
 594     qemu_aio_set_fd_handler(s->fd, posix_aio_read, NULL, posix_aio_flush, s);
 595
 596 #if defined(__linux__)
 597     {
 598         struct aioinit ai;
 599
 600         memset(&ai, 0, sizeof(ai));
 601 #if defined(__GLIBC_PREREQ) && __GLIBC_PREREQ(2, 4)
 602         ai.aio_threads = 64;
 603         ai.aio_num = 64;
 604 #else
 605         /* XXX: aio thread exit seems to hang on RedHat 9 and this init
 606            seems to fix the problem. */
 607         ai.aio_threads = 1;
 608         ai.aio_num = 1;
 609         ai.aio_idle_time = 365 * 100000;
 610 #endif
 611         aio_init(&ai);
 612     }
 613 #endif
 614     posix_aio_state = s;
 615
 616     return 0;
 617 }
 618
 619 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 620         int64_t sector_num, uint8_t *buf, int nb_sectors,
 621         BlockDriverCompletionFunc *cb, void *opaque)
 622 {
 623     BDRVRawState *s = bs->opaque;
 624     RawAIOCB *acb;
 625
 626     if (fd_open(bs) < 0)
 627         return NULL;
 628
 629     acb = qemu_aio_get(bs, cb, opaque);
 630     if (!acb)
 631         return NULL;
 632     acb->fd = raw_fd_pool_get(s);
 633     acb->aiocb.aio_fildes = acb->fd;
 634     acb->aiocb.aio_sigevent.sigev_signo = SIGUSR2;
 635     acb->aiocb.aio_sigevent.sigev_notify = SIGEV_SIGNAL;
 636     acb->aiocb.aio_buf = buf;
 637     if (nb_sectors < 0)
 638         acb->aiocb.aio_nbytes = -nb_sectors;
 639     else
 640         acb->aiocb.aio_nbytes = nb_sectors * 512;
 641     acb->aiocb.aio_offset = sector_num * 512;
 642     acb->next = posix_aio_state->first_aio;
 643     posix_aio_state->first_aio = acb;
 644     return acb;
 645 }
 646
 647 static void raw_aio_em_cb(void* opaque)
 648 {
 649     RawAIOCB *acb = opaque;
 650     acb->common.cb(acb->common.opaque, acb->ret);
 651     qemu_aio_release(acb);
 652 }
 653
 654 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 655         int64_t sector_num, uint8_t *buf, int nb_sectors,
 656         BlockDriverCompletionFunc *cb, void *opaque)
 657 {
 658     RawAIOCB *acb;
 659
 660     /*
 661      * If O_DIRECT is used and the buffer is not aligned fall back
 662      * to synchronous IO.
 663      */
 664 #if defined(O_DIRECT)
 665     BDRVRawState *s = bs->opaque;
 666
 667     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 668         QEMUBH *bh;
 669         acb = qemu_aio_get(bs, cb, opaque);
 670         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 671         bh = qemu_bh_new(raw_aio_em_cb, acb);
 672         qemu_bh_schedule(bh);
 673         return &acb->common;
 674     }
 675 #endif
 676
 677     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 678     if (!acb)
 679         return NULL;
 680     if (aio_read(&acb->aiocb) < 0) {
 681         qemu_aio_release(acb);
 682         return NULL;
 683     }
 684     return &acb->common;
 685 }
 686
 687 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 688         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 689         BlockDriverCompletionFunc *cb, void *opaque)
 690 {
 691     RawAIOCB *acb;
 692
 693     /*
 694      * If O_DIRECT is used and the buffer is not aligned fall back
 695      * to synchronous IO.
 696      */
 697 #if defined(O_DIRECT)
 698     BDRVRawState *s = bs->opaque;
 699
 700     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 701         QEMUBH *bh;
 702         acb = qemu_aio_get(bs, cb, opaque);
 703         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 704         bh = qemu_bh_new(raw_aio_em_cb, acb);
 705         qemu_bh_schedule(bh);
 706         return &acb->common;
 707     }
 708 #endif
 709
 710     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 711     if (!acb)
 712         return NULL;
 713     if (aio_write(&acb->aiocb) < 0) {
 714         qemu_aio_release(acb);
 715         return NULL;
 716     }
 717     return &acb->common;
 718 }
 719
 720 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 721 {
 722     int ret;
 723     RawAIOCB *acb = (RawAIOCB *)blockacb;
 724     RawAIOCB **pacb;
 725
 726     ret = aio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 727     if (ret == AIO_NOTCANCELED) {
 728         /* fail safe: if the aio could not be canceled, we wait for
 729            it */
 730         while (aio_error(&acb->aiocb) == EINPROGRESS);
 731     }
 732
 733     /* remove the callback from the queue */
 734     pacb = &posix_aio_state->first_aio;
 735     for(;;) {
 736         if (*pacb == NULL) {
 737             break;
 738         } else if (*pacb == acb) {
 739             *pacb = acb->next;
 740             raw_fd_pool_put(acb);
 741             qemu_aio_release(acb);
 742             break;
 743         }
 744         pacb = &acb->next;
 745     }
 746 }
 747
 748 #else /* CONFIG_AIO */
 749 static int posix_aio_init(void)
 750 {
 751 }
 752 #endif /* CONFIG_AIO */
 753
 754 static void raw_close_fd_pool(BDRVRawState *s)
 755 {
 756     int i;
 757
 758     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 759         if (s->fd_pool[i] != -1) {
 760             close(s->fd_pool[i]);
 761             s->fd_pool[i] = -1;
 762         }
 763     }
 764 }
 765
 766 static void raw_close(BlockDriverState *bs)
 767 {
 768     BDRVRawState *s = bs->opaque;
 769     if (s->fd >= 0) {
 770         close(s->fd);
 771         s->fd = -1;
 772 #if defined(O_DIRECT)
 773         if (s->aligned_buf != NULL)
 774             qemu_free(s->aligned_buf);
 775 #endif
 776     }
 777     raw_close_fd_pool(s);
 778 }
 779
 780 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 781 {
 782     BDRVRawState *s = bs->opaque;
 783     if (s->type != FTYPE_FILE)
 784         return -ENOTSUP;
 785     if (ftruncate(s->fd, offset) < 0)
 786         return -errno;
 787     return 0;
 788 }
 789
 790 #ifdef __OpenBSD__
 791 static int64_t raw_getlength(BlockDriverState *bs)
 792 {
 793     BDRVRawState *s = bs->opaque;
 794     int fd = s->fd;
 795     struct stat st;
 796
 797     if (fstat(fd, &st))
 798         return -1;
 799     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 800         struct disklabel dl;
 801
 802         if (ioctl(fd, DIOCGDINFO, &dl))
 803             return -1;
 804         return (uint64_t)dl.d_secsize *
 805             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 806     } else
 807         return st.st_size;
 808 }
 809 #else /* !__OpenBSD__ */
 810 static int64_t  raw_getlength(BlockDriverState *bs)
 811 {
 812     BDRVRawState *s = bs->opaque;
 813     int fd = s->fd;
 814     int64_t size;
 815 #ifdef _BSD
 816     struct stat sb;
 817 #endif
 818 #ifdef __sun__
 819     struct dk_minfo minfo;
 820     int rv;
 821 #endif
 822     int ret;
 823
 824     ret = fd_open(bs);
 825     if (ret < 0)
 826         return ret;
 827
 828 #ifdef _BSD
 829     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 830 #ifdef DIOCGMEDIASIZE
 831         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 832 #endif
 833 #ifdef CONFIG_COCOA
 834         size = LONG_LONG_MAX;
 835 #else
 836         size = lseek(fd, 0LL, SEEK_END);
 837 #endif
 838     } else
 839 #endif
 840 #ifdef __sun__
 841     /*
 842      * use the DKIOCGMEDIAINFO ioctl to read the size.
 843      */
 844     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 845     if ( rv != -1 ) {
 846         size = minfo.dki_lbsize * minfo.dki_capacity;
 847     } else /* there are reports that lseek on some devices
 848               fails, but irc discussion said that contingency
 849               on contingency was overkill */
 850 #endif
 851     {
 852         size = lseek(fd, 0, SEEK_END);
 853     }
 854     return size;
 855 }
 856 #endif
 857
 858 static int raw_create(const char *filename, int64_t total_size,
 859                       const char *backing_file, int flags)
 860 {
 861     int fd;
 862
 863     if (flags || backing_file)
 864         return -ENOTSUP;
 865
 866     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 867               0644);
 868     if (fd < 0)
 869         return -EIO;
 870     ftruncate(fd, total_size * 512);
 871     close(fd);
 872     return 0;
 873 }
 874
 875 static void raw_flush(BlockDriverState *bs)
 876 {
 877     BDRVRawState *s = bs->opaque;
 878     fsync(s->fd);
 879 }
 880
 881 BlockDriver bdrv_raw = {
 882     "raw",
 883     sizeof(BDRVRawState),
 884     NULL, /* no probe for protocols */
 885     raw_open,
 886     NULL,
 887     NULL,
 888     raw_close,
 889     raw_create,
 890     raw_flush,
 891
 892 #ifdef CONFIG_AIO
 893     .bdrv_aio_read = raw_aio_read,
 894     .bdrv_aio_write = raw_aio_write,
 895     .bdrv_aio_cancel = raw_aio_cancel,
 896     .aiocb_size = sizeof(RawAIOCB),
 897 #endif
 898     .bdrv_pread = raw_pread,
 899     .bdrv_pwrite = raw_pwrite,
 900     .bdrv_truncate = raw_truncate,
 901     .bdrv_getlength = raw_getlength,
 902 };
 903
 904 /***********************************************/
 905 /* host device */
 906
 907 #ifdef CONFIG_COCOA
 908 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 909 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 910
 911 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 912 {
 913     kern_return_t       kernResult;
 914     mach_port_t     masterPort;
 915     CFMutableDictionaryRef  classesToMatch;
 916
 917     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 918     if ( KERN_SUCCESS != kernResult ) {
 919         printf( "IOMasterPort returned %d\n", kernResult );
 920     }
 921
 922     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 923     if ( classesToMatch == NULL ) {
 924         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 925     } else {
 926     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 927     }
 928     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 929     if ( KERN_SUCCESS != kernResult )
 930     {
 931         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 932     }
 933
 934     return kernResult;
 935 }
 936
 937 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 938 {
 939     io_object_t     nextMedia;
 940     kern_return_t   kernResult = KERN_FAILURE;
 941     *bsdPath = '\0';
 942     nextMedia = IOIteratorNext( mediaIterator );
 943     if ( nextMedia )
 944     {
 945         CFTypeRef   bsdPathAsCFString;
 946     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 947         if ( bsdPathAsCFString ) {
 948             size_t devPathLength;
 949             strcpy( bsdPath, _PATH_DEV );
 950             strcat( bsdPath, "r" );
 951             devPathLength = strlen( bsdPath );
 952             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 953                 kernResult = KERN_SUCCESS;
 954             }
 955             CFRelease( bsdPathAsCFString );
 956         }
 957         IOObjectRelease( nextMedia );
 958     }
 959
 960     return kernResult;
 961 }
 962
 963 #endif
 964
 965 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 966 {
 967     BDRVRawState *s = bs->opaque;
 968     int fd, open_flags, ret, i;
 969
 970     posix_aio_init();
 971
 972 #ifdef CONFIG_COCOA
 973     if (strstart(filename, "/dev/cdrom", NULL)) {
 974         kern_return_t kernResult;
 975         io_iterator_t mediaIterator;
 976         char bsdPath[ MAXPATHLEN ];
 977         int fd;
 978
 979         kernResult = FindEjectableCDMedia( &mediaIterator );
 980         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 981
 982         if ( bsdPath[ 0 ] != '\0' ) {
 983             strcat(bsdPath,"s0");
 984             /* some CDs don't have a partition 0 */
 985             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 986             if (fd < 0) {
 987                 bsdPath[strlen(bsdPath)-1] = '1';
 988             } else {
 989                 close(fd);
 990             }
 991             filename = bsdPath;
 992         }
 993
 994         if ( mediaIterator )
 995             IOObjectRelease( mediaIterator );
 996     }
 997 #endif
 998     open_flags = O_BINARY;
 999     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
1000         open_flags |= O_RDWR;
1001     } else {
1002         open_flags |= O_RDONLY;
1003         bs->read_only = 1;
1004     }
1005 #ifdef O_DIRECT
1006     if (flags & BDRV_O_DIRECT)
1007         open_flags |= O_DIRECT;
1008 #endif
1009
1010     s->type = FTYPE_FILE;
1011 #if defined(__linux__)
1012     if (strstart(filename, "/dev/cd", NULL)) {
1013         /* open will not fail even if no CD is inserted */
1014         open_flags |= O_NONBLOCK;
1015         s->type = FTYPE_CD;
1016     } else if (strstart(filename, "/dev/fd", NULL)) {
1017         s->type = FTYPE_FD;
1018         s->fd_open_flags = open_flags;
1019         /* open will not fail even if no floppy is inserted */
1020         open_flags |= O_NONBLOCK;
1021     } else if (strstart(filename, "/dev/sg", NULL)) {
1022         bs->sg = 1;
1023     }
1024 #endif
1025     fd = open(filename, open_flags, 0644);
1026     if (fd < 0) {
1027         ret = -errno;
1028         if (ret == -EROFS)
1029             ret = -EACCES;
1030         return ret;
1031     }
1032     s->fd = fd;
1033     for (i = 0; i < RAW_FD_POOL_SIZE; i++)
1034         s->fd_pool[i] = -1;
1035 #if defined(__linux__)
1036     /* close fd so that we can reopen it as needed */
1037     if (s->type == FTYPE_FD) {
1038         close(s->fd);
1039         s->fd = -1;
1040         s->fd_media_changed = 1;
1041     }
1042 #endif
1043     return 0;
1044 }
1045
1046 #if defined(__linux__)
1047 /* Note: we do not have a reliable method to detect if the floppy is
1048    present. The current method is to try to open the floppy at every
1049    I/O and to keep it opened during a few hundreds of ms. */
1050 static int fd_open(BlockDriverState *bs)
1051 {
1052     BDRVRawState *s = bs->opaque;
1053     int last_media_present;
1054
1055     if (s->type != FTYPE_FD)
1056         return 0;
1057     last_media_present = (s->fd >= 0);
1058     if (s->fd >= 0 &&
1059         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1060         close(s->fd);
1061         s->fd = -1;
1062         raw_close_fd_pool(s);
1063 #ifdef DEBUG_FLOPPY
1064         printf("Floppy closed\n");
1065 #endif
1066     }
1067     if (s->fd < 0) {
1068         if (s->fd_got_error &&
1069             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1070 #ifdef DEBUG_FLOPPY
1071             printf("No floppy (open delayed)\n");
1072 #endif
1073             return -EIO;
1074         }
1075         s->fd = open(bs->filename, s->fd_open_flags);
1076         if (s->fd < 0) {
1077             s->fd_error_time = qemu_get_clock(rt_clock);
1078             s->fd_got_error = 1;
1079             if (last_media_present)
1080                 s->fd_media_changed = 1;
1081 #ifdef DEBUG_FLOPPY
1082             printf("No floppy\n");
1083 #endif
1084             return -EIO;
1085         }
1086 #ifdef DEBUG_FLOPPY
1087         printf("Floppy opened\n");
1088 #endif
1089     }
1090     if (!last_media_present)
1091         s->fd_media_changed = 1;
1092     s->fd_open_time = qemu_get_clock(rt_clock);
1093     s->fd_got_error = 0;
1094     return 0;
1095 }
1096
1097 static int raw_is_inserted(BlockDriverState *bs)
1098 {
1099     BDRVRawState *s = bs->opaque;
1100     int ret;
1101
1102     switch(s->type) {
1103     case FTYPE_CD:
1104         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1105         if (ret == CDS_DISC_OK)
1106             return 1;
1107         else
1108             return 0;
1109         break;
1110     case FTYPE_FD:
1111         ret = fd_open(bs);
1112         return (ret >= 0);
1113     default:
1114         return 1;
1115     }
1116 }
1117
1118 /* currently only used by fdc.c, but a CD version would be good too */
1119 static int raw_media_changed(BlockDriverState *bs)
1120 {
1121     BDRVRawState *s = bs->opaque;
1122
1123     switch(s->type) {
1124     case FTYPE_FD:
1125         {
1126             int ret;
1127             /* XXX: we do not have a true media changed indication. It
1128                does not work if the floppy is changed without trying
1129                to read it */
1130             fd_open(bs);
1131             ret = s->fd_media_changed;
1132             s->fd_media_changed = 0;
1133 #ifdef DEBUG_FLOPPY
1134             printf("Floppy changed=%d\n", ret);
1135 #endif
1136             return ret;
1137         }
1138     default:
1139         return -ENOTSUP;
1140     }
1141 }
1142
1143 static int raw_eject(BlockDriverState *bs, int eject_flag)
1144 {
1145     BDRVRawState *s = bs->opaque;
1146
1147     switch(s->type) {
1148     case FTYPE_CD:
1149         if (eject_flag) {
1150             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1151                 perror("CDROMEJECT");
1152         } else {
1153             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1154                 perror("CDROMEJECT");
1155         }
1156         break;
1157     case FTYPE_FD:
1158         {
1159             int fd;
1160             if (s->fd >= 0) {
1161                 close(s->fd);
1162                 s->fd = -1;
1163                 raw_close_fd_pool(s);
1164             }
1165             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1166             if (fd >= 0) {
1167                 if (ioctl(fd, FDEJECT, 0) < 0)
1168                     perror("FDEJECT");
1169                 close(fd);
1170             }
1171         }
1172         break;
1173     default:
1174         return -ENOTSUP;
1175     }
1176     return 0;
1177 }
1178
1179 static int raw_set_locked(BlockDriverState *bs, int locked)
1180 {
1181     BDRVRawState *s = bs->opaque;
1182
1183     switch(s->type) {
1184     case FTYPE_CD:
1185         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1186             /* Note: an error can happen if the distribution automatically
1187                mounts the CD-ROM */
1188             //        perror("CDROM_LOCKDOOR");
1189         }
1190         break;
1191     default:
1192         return -ENOTSUP;
1193     }
1194     return 0;
1195 }
1196
1197 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1198 {
1199     BDRVRawState *s = bs->opaque;
1200
1201     return ioctl(s->fd, req, buf);
1202 }
1203 #else
1204
1205 static int fd_open(BlockDriverState *bs)
1206 {
1207     return 0;
1208 }
1209
1210 static int raw_is_inserted(BlockDriverState *bs)
1211 {
1212     return 1;
1213 }
1214
1215 static int raw_media_changed(BlockDriverState *bs)
1216 {
1217     return -ENOTSUP;
1218 }
1219
1220 static int raw_eject(BlockDriverState *bs, int eject_flag)
1221 {
1222     return -ENOTSUP;
1223 }
1224
1225 static int raw_set_locked(BlockDriverState *bs, int locked)
1226 {
1227     return -ENOTSUP;
1228 }
1229
1230 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1231 {
1232     return -ENOTSUP;
1233 }
1234 #endif /* !linux */
1235
1236 BlockDriver bdrv_host_device = {
1237     "host_device",
1238     sizeof(BDRVRawState),
1239     NULL, /* no probe for protocols */
1240     hdev_open,
1241     NULL,
1242     NULL,
1243     raw_close,
1244     NULL,
1245     raw_flush,
1246
1247 #ifdef CONFIG_AIO
1248     .bdrv_aio_read = raw_aio_read,
1249     .bdrv_aio_write = raw_aio_write,
1250     .bdrv_aio_cancel = raw_aio_cancel,
1251     .aiocb_size = sizeof(RawAIOCB),
1252 #endif
1253     .bdrv_pread = raw_pread,
1254     .bdrv_pwrite = raw_pwrite,
1255     .bdrv_getlength = raw_getlength,
1256
1257     /* removable device support */
1258     .bdrv_is_inserted = raw_is_inserted,
1259     .bdrv_media_changed = raw_media_changed,
1260     .bdrv_eject = raw_eject,
1261     .bdrv_set_locked = raw_set_locked,
1262     /* generic scsi device */
1263     .bdrv_ioctl = raw_ioctl,
1264 };