block-raw-posix.c

   1 /*
   2  * Block driver for RAW files (posix)
   3  *
   4  * Copyright (c) 2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu-common.h"
  25 #include "qemu-timer.h"
  26 #include "qemu-char.h"
  27 #include "block_int.h"
  28 #include <assert.h>
  29 #ifdef CONFIG_AIO
  30 #include <aio.h>
  31 #endif
  32
  33 #ifdef CONFIG_COCOA
  34 #include <paths.h>
  35 #include <sys/param.h>
  36 #include <IOKit/IOKitLib.h>
  37 #include <IOKit/IOBSD.h>
  38 #include <IOKit/storage/IOMediaBSDClient.h>
  39 #include <IOKit/storage/IOMedia.h>
  40 #include <IOKit/storage/IOCDMedia.h>
  41 //#include <IOKit/storage/IOCDTypes.h>
  42 #include <CoreFoundation/CoreFoundation.h>
  43 #endif
  44
  45 #ifdef __sun__
  46 #define _POSIX_PTHREAD_SEMANTICS 1
  47 #include <signal.h>
  48 #include <sys/dkio.h>
  49 #endif
  50 #ifdef __linux__
  51 #include <sys/ioctl.h>
  52 #include <linux/cdrom.h>
  53 #include <linux/fd.h>
  54 #endif
  55 #ifdef __FreeBSD__
  56 #include <signal.h>
  57 #include <sys/disk.h>
  58 #endif
  59
  60 #ifdef __OpenBSD__
  61 #include <sys/ioctl.h>
  62 #include <sys/disklabel.h>
  63 #include <sys/dkio.h>
  64 #endif
  65
  66 //#define DEBUG_FLOPPY
  67
  68 //#define DEBUG_BLOCK
  69 #if defined(DEBUG_BLOCK)
  70 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (loglevel != 0)  \
  71     { fprintf(logfile, formatCstr, ##args); fflush(logfile); } } while (0)
  72 #else
  73 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
  74 #endif
  75
  76 #define FTYPE_FILE   0
  77 #define FTYPE_CD     1
  78 #define FTYPE_FD     2
  79
  80 #define ALIGNED_BUFFER_SIZE (32 * 512)
  81
  82 /* if the FD is not accessed during that time (in ms), we try to
  83    reopen it to see if the disk has been changed */
  84 #define FD_OPEN_TIMEOUT 1000
  85
  86 /* posix-aio doesn't allow multiple outstanding requests to a single file
  87  * descriptor.  we implement a pool of dup()'d file descriptors to work
  88  * around this */
  89 #define RAW_FD_POOL_SIZE        64
  90
  91 typedef struct BDRVRawState {
  92     int fd;
  93     int type;
  94     unsigned int lseek_err_cnt;
  95     int fd_pool[RAW_FD_POOL_SIZE];
  96 #if defined(__linux__)
  97     /* linux floppy specific */
  98     int fd_open_flags;
  99     int64_t fd_open_time;
 100     int64_t fd_error_time;
 101     int fd_got_error;
 102     int fd_media_changed;
 103 #endif
 104 #if defined(O_DIRECT)
 105     uint8_t* aligned_buf;
 106 #endif
 107 } BDRVRawState;
 108
 109 static int posix_aio_init(void);
 110
 111 static int fd_open(BlockDriverState *bs);
 112
 113 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 114 {
 115     BDRVRawState *s = bs->opaque;
 116     int fd, open_flags, ret;
 117     int i;
 118
 119     posix_aio_init();
 120
 121     s->lseek_err_cnt = 0;
 122
 123     open_flags = O_BINARY;
 124     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 125         open_flags |= O_RDWR;
 126     } else {
 127         open_flags |= O_RDONLY;
 128         bs->read_only = 1;
 129     }
 130     if (flags & BDRV_O_CREAT)
 131         open_flags |= O_CREAT | O_TRUNC;
 132 #ifdef O_DIRECT
 133     if (flags & BDRV_O_DIRECT)
 134         open_flags |= O_DIRECT;
 135 #endif
 136
 137     s->type = FTYPE_FILE;
 138
 139     fd = open(filename, open_flags, 0644);
 140     if (fd < 0) {
 141         ret = -errno;
 142         if (ret == -EROFS)
 143             ret = -EACCES;
 144         return ret;
 145     }
 146     s->fd = fd;
 147     for (i = 0; i < RAW_FD_POOL_SIZE; i++)
 148         s->fd_pool[i] = -1;
 149 #if defined(O_DIRECT)
 150     s->aligned_buf = NULL;
 151     if (flags & BDRV_O_DIRECT) {
 152         s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
 153         if (s->aligned_buf == NULL) {
 154             ret = -errno;
 155             close(fd);
 156             return ret;
 157         }
 158     }
 159 #endif
 160     return 0;
 161 }
 162
 163 /* XXX: use host sector size if necessary with:
 164 #ifdef DIOCGSECTORSIZE
 165         {
 166             unsigned int sectorsize = 512;
 167             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 168                 sectorsize > bufsize)
 169                 bufsize = sectorsize;
 170         }
 171 #endif
 172 #ifdef CONFIG_COCOA
 173         u_int32_t   blockSize = 512;
 174         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 175             bufsize = blockSize;
 176         }
 177 #endif
 178 */
 179
 180 /*
 181  * offset and count are in bytes, but must be multiples of 512 for files
 182  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 183  *
 184  * This function may be called without alignment if the caller ensures
 185  * that O_DIRECT is not in effect.
 186  */
 187 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 188                      uint8_t *buf, int count)
 189 {
 190     BDRVRawState *s = bs->opaque;
 191     int ret;
 192
 193     ret = fd_open(bs);
 194     if (ret < 0)
 195         return ret;
 196
 197     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 198         ++(s->lseek_err_cnt);
 199         if(s->lseek_err_cnt <= 10) {
 200             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 201                               "] lseek failed : %d = %s\n",
 202                               s->fd, bs->filename, offset, buf, count,
 203                               bs->total_sectors, errno, strerror(errno));
 204         }
 205         return -1;
 206     }
 207     s->lseek_err_cnt=0;
 208
 209     ret = read(s->fd, buf, count);
 210     if (ret == count)
 211         goto label__raw_read__success;
 212
 213     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 214                       "] read failed %d : %d = %s\n",
 215                       s->fd, bs->filename, offset, buf, count,
 216                       bs->total_sectors, ret, errno, strerror(errno));
 217
 218     /* Try harder for CDrom. */
 219     if (bs->type == BDRV_TYPE_CDROM) {
 220         lseek(s->fd, offset, SEEK_SET);
 221         ret = read(s->fd, buf, count);
 222         if (ret == count)
 223             goto label__raw_read__success;
 224         lseek(s->fd, offset, SEEK_SET);
 225         ret = read(s->fd, buf, count);
 226         if (ret == count)
 227             goto label__raw_read__success;
 228
 229         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 230                           "] retry read failed %d : %d = %s\n",
 231                           s->fd, bs->filename, offset, buf, count,
 232                           bs->total_sectors, ret, errno, strerror(errno));
 233     }
 234
 235 label__raw_read__success:
 236
 237     return ret;
 238 }
 239
 240 /*
 241  * offset and count are in bytes, but must be multiples of 512 for files
 242  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 243  *
 244  * This function may be called without alignment if the caller ensures
 245  * that O_DIRECT is not in effect.
 246  */
 247 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 248                       const uint8_t *buf, int count)
 249 {
 250     BDRVRawState *s = bs->opaque;
 251     int ret;
 252
 253     ret = fd_open(bs);
 254     if (ret < 0)
 255         return ret;
 256
 257     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 258         ++(s->lseek_err_cnt);
 259         if(s->lseek_err_cnt) {
 260             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 261                               PRId64 "] lseek failed : %d = %s\n",
 262                               s->fd, bs->filename, offset, buf, count,
 263                               bs->total_sectors, errno, strerror(errno));
 264         }
 265         return -1;
 266     }
 267     s->lseek_err_cnt = 0;
 268
 269     ret = write(s->fd, buf, count);
 270     if (ret == count)
 271         goto label__raw_write__success;
 272
 273     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 274                       "] write failed %d : %d = %s\n",
 275                       s->fd, bs->filename, offset, buf, count,
 276                       bs->total_sectors, ret, errno, strerror(errno));
 277
 278 label__raw_write__success:
 279
 280     return ret;
 281 }
 282
 283
 284 #if defined(O_DIRECT)
 285 /*
 286  * offset and count are in bytes and possibly not aligned. For files opened
 287  * with O_DIRECT, necessary alignments are ensured before calling
 288  * raw_pread_aligned to do the actual read.
 289  */
 290 static int raw_pread(BlockDriverState *bs, int64_t offset,
 291                      uint8_t *buf, int count)
 292 {
 293     BDRVRawState *s = bs->opaque;
 294     int size, ret, shift, sum;
 295
 296     sum = 0;
 297
 298     if (s->aligned_buf != NULL)  {
 299
 300         if (offset & 0x1ff) {
 301             /* align offset on a 512 bytes boundary */
 302
 303             shift = offset & 0x1ff;
 304             size = (shift + count + 0x1ff) & ~0x1ff;
 305             if (size > ALIGNED_BUFFER_SIZE)
 306                 size = ALIGNED_BUFFER_SIZE;
 307             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 308             if (ret < 0)
 309                 return ret;
 310
 311             size = 512 - shift;
 312             if (size > count)
 313                 size = count;
 314             memcpy(buf, s->aligned_buf + shift, size);
 315
 316             buf += size;
 317             offset += size;
 318             count -= size;
 319             sum += size;
 320
 321             if (count == 0)
 322                 return sum;
 323         }
 324         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 325
 326             /* read on aligned buffer */
 327
 328             while (count) {
 329
 330                 size = (count + 0x1ff) & ~0x1ff;
 331                 if (size > ALIGNED_BUFFER_SIZE)
 332                     size = ALIGNED_BUFFER_SIZE;
 333
 334                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 335                 if (ret < 0)
 336                     return ret;
 337
 338                 size = ret;
 339                 if (size > count)
 340                     size = count;
 341
 342                 memcpy(buf, s->aligned_buf, size);
 343
 344                 buf += size;
 345                 offset += size;
 346                 count -= size;
 347                 sum += size;
 348             }
 349
 350             return sum;
 351         }
 352     }
 353
 354     return raw_pread_aligned(bs, offset, buf, count) + sum;
 355 }
 356
 357 /*
 358  * offset and count are in bytes and possibly not aligned. For files opened
 359  * with O_DIRECT, necessary alignments are ensured before calling
 360  * raw_pwrite_aligned to do the actual write.
 361  */
 362 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 363                       const uint8_t *buf, int count)
 364 {
 365     BDRVRawState *s = bs->opaque;
 366     int size, ret, shift, sum;
 367
 368     sum = 0;
 369
 370     if (s->aligned_buf != NULL) {
 371
 372         if (offset & 0x1ff) {
 373             /* align offset on a 512 bytes boundary */
 374             shift = offset & 0x1ff;
 375             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 376             if (ret < 0)
 377                 return ret;
 378
 379             size = 512 - shift;
 380             if (size > count)
 381                 size = count;
 382             memcpy(s->aligned_buf + shift, buf, size);
 383
 384             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 385             if (ret < 0)
 386                 return ret;
 387
 388             buf += size;
 389             offset += size;
 390             count -= size;
 391             sum += size;
 392
 393             if (count == 0)
 394                 return sum;
 395         }
 396         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 397
 398             while ((size = (count & ~0x1ff)) != 0) {
 399
 400                 if (size > ALIGNED_BUFFER_SIZE)
 401                     size = ALIGNED_BUFFER_SIZE;
 402
 403                 memcpy(s->aligned_buf, buf, size);
 404
 405                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 406                 if (ret < 0)
 407                     return ret;
 408
 409                 buf += ret;
 410                 offset += ret;
 411                 count -= ret;
 412                 sum += ret;
 413             }
 414             /* here, count < 512 because (count & ~0x1ff) == 0 */
 415             if (count) {
 416                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 417                 if (ret < 0)
 418                     return ret;
 419                  memcpy(s->aligned_buf, buf, count);
 420
 421                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 422                  if (ret < 0)
 423                      return ret;
 424                  if (count < ret)
 425                      ret = count;
 426
 427                  sum += ret;
 428             }
 429             return sum;
 430         }
 431     }
 432     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 433 }
 434
 435 #else
 436 #define raw_pread raw_pread_aligned
 437 #define raw_pwrite raw_pwrite_aligned
 438 #endif
 439
 440
 441 #ifdef CONFIG_AIO
 442 /***********************************************************/
 443 /* Unix AIO using POSIX AIO */
 444
 445 typedef struct RawAIOCB {
 446     BlockDriverAIOCB common;
 447     int fd;
 448     struct aiocb aiocb;
 449     struct RawAIOCB *next;
 450     int ret;
 451 } RawAIOCB;
 452
 453 typedef struct PosixAioState
 454 {
 455     int rfd, wfd;
 456     RawAIOCB *first_aio;
 457 } PosixAioState;
 458
 459 static int raw_fd_pool_get(BDRVRawState *s)
 460 {
 461     int i;
 462
 463     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 464         /* already in use */
 465         if (s->fd_pool[i] != -1)
 466             continue;
 467
 468         /* try to dup file descriptor */
 469         s->fd_pool[i] = dup(s->fd);
 470         if (s->fd_pool[i] != -1)
 471             return s->fd_pool[i];
 472     }
 473
 474     /* we couldn't dup the file descriptor so just use the main one */
 475     return s->fd;
 476 }
 477
 478 static void raw_fd_pool_put(RawAIOCB *acb)
 479 {
 480     BDRVRawState *s = acb->common.bs->opaque;
 481     int i;
 482
 483     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 484         if (s->fd_pool[i] == acb->fd) {
 485             close(s->fd_pool[i]);
 486             s->fd_pool[i] = -1;
 487         }
 488     }
 489 }
 490
 491 static void posix_aio_read(void *opaque)
 492 {
 493     PosixAioState *s = opaque;
 494     RawAIOCB *acb, **pacb;
 495     int ret;
 496     ssize_t len;
 497
 498     do {
 499         char byte;
 500
 501         len = read(s->rfd, &byte, 1);
 502         if (len == -1 && errno == EINTR)
 503             continue;
 504         if (len == -1 && errno == EAGAIN)
 505             break;
 506     } while (len == -1);
 507
 508     for(;;) {
 509         pacb = &s->first_aio;
 510         for(;;) {
 511             acb = *pacb;
 512             if (!acb)
 513                 goto the_end;
 514             ret = aio_error(&acb->aiocb);
 515             if (ret == ECANCELED) {
 516                 /* remove the request */
 517                 *pacb = acb->next;
 518                 raw_fd_pool_put(acb);
 519                 qemu_aio_release(acb);
 520             } else if (ret != EINPROGRESS) {
 521                 /* end of aio */
 522                 if (ret == 0) {
 523                     ret = aio_return(&acb->aiocb);
 524                     if (ret == acb->aiocb.aio_nbytes)
 525                         ret = 0;
 526                     else
 527                         ret = -EINVAL;
 528                 } else {
 529                     ret = -ret;
 530                 }
 531                 /* remove the request */
 532                 *pacb = acb->next;
 533                 /* call the callback */
 534                 acb->common.cb(acb->common.opaque, ret);
 535                 raw_fd_pool_put(acb);
 536                 qemu_aio_release(acb);
 537                 break;
 538             } else {
 539                 pacb = &acb->next;
 540             }
 541         }
 542     }
 543  the_end: ;
 544 }
 545
 546 static int posix_aio_flush(void *opaque)
 547 {
 548     PosixAioState *s = opaque;
 549     return !!s->first_aio;
 550 }
 551
 552 static PosixAioState *posix_aio_state;
 553
 554 static void aio_signal_handler(int signum)
 555 {
 556     if (posix_aio_state) {
 557         char byte = 0;
 558
 559         write(posix_aio_state->wfd, &byte, sizeof(byte));
 560     }
 561
 562     qemu_service_io();
 563 }
 564
 565 static int posix_aio_init(void)
 566 {
 567     struct sigaction act;
 568     PosixAioState *s;
 569     int fds[2];
 570
 571     if (posix_aio_state)
 572         return 0;
 573
 574     s = qemu_malloc(sizeof(PosixAioState));
 575     if (s == NULL)
 576         return -ENOMEM;
 577
 578     sigfillset(&act.sa_mask);
 579     act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
 580     act.sa_handler = aio_signal_handler;
 581     sigaction(SIGUSR2, &act, NULL);
 582
 583     s->first_aio = NULL;
 584     if (pipe(fds) == -1) {
 585         fprintf(stderr, "failed to create pipe\n");
 586         return -errno;
 587     }
 588
 589     s->rfd = fds[0];
 590     s->wfd = fds[1];
 591
 592     fcntl(s->wfd, F_SETFL, O_NONBLOCK);
 593
 594     qemu_aio_set_fd_handler(s->rfd, 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     return 0;
 752 }
 753 #endif /* CONFIG_AIO */
 754
 755 static void raw_close_fd_pool(BDRVRawState *s)
 756 {
 757     int i;
 758
 759     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 760         if (s->fd_pool[i] != -1) {
 761             close(s->fd_pool[i]);
 762             s->fd_pool[i] = -1;
 763         }
 764     }
 765 }
 766
 767 static void raw_close(BlockDriverState *bs)
 768 {
 769     BDRVRawState *s = bs->opaque;
 770     if (s->fd >= 0) {
 771         close(s->fd);
 772         s->fd = -1;
 773 #if defined(O_DIRECT)
 774         if (s->aligned_buf != NULL)
 775             qemu_free(s->aligned_buf);
 776 #endif
 777     }
 778     raw_close_fd_pool(s);
 779 }
 780
 781 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 782 {
 783     BDRVRawState *s = bs->opaque;
 784     if (s->type != FTYPE_FILE)
 785         return -ENOTSUP;
 786     if (ftruncate(s->fd, offset) < 0)
 787         return -errno;
 788     return 0;
 789 }
 790
 791 #ifdef __OpenBSD__
 792 static int64_t raw_getlength(BlockDriverState *bs)
 793 {
 794     BDRVRawState *s = bs->opaque;
 795     int fd = s->fd;
 796     struct stat st;
 797
 798     if (fstat(fd, &st))
 799         return -1;
 800     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 801         struct disklabel dl;
 802
 803         if (ioctl(fd, DIOCGDINFO, &dl))
 804             return -1;
 805         return (uint64_t)dl.d_secsize *
 806             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 807     } else
 808         return st.st_size;
 809 }
 810 #else /* !__OpenBSD__ */
 811 static int64_t  raw_getlength(BlockDriverState *bs)
 812 {
 813     BDRVRawState *s = bs->opaque;
 814     int fd = s->fd;
 815     int64_t size;
 816 #ifdef _BSD
 817     struct stat sb;
 818 #endif
 819 #ifdef __sun__
 820     struct dk_minfo minfo;
 821     int rv;
 822 #endif
 823     int ret;
 824
 825     ret = fd_open(bs);
 826     if (ret < 0)
 827         return ret;
 828
 829 #ifdef _BSD
 830     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 831 #ifdef DIOCGMEDIASIZE
 832         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 833 #endif
 834 #ifdef CONFIG_COCOA
 835         size = LONG_LONG_MAX;
 836 #else
 837         size = lseek(fd, 0LL, SEEK_END);
 838 #endif
 839     } else
 840 #endif
 841 #ifdef __sun__
 842     /*
 843      * use the DKIOCGMEDIAINFO ioctl to read the size.
 844      */
 845     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 846     if ( rv != -1 ) {
 847         size = minfo.dki_lbsize * minfo.dki_capacity;
 848     } else /* there are reports that lseek on some devices
 849               fails, but irc discussion said that contingency
 850               on contingency was overkill */
 851 #endif
 852     {
 853         size = lseek(fd, 0, SEEK_END);
 854     }
 855     return size;
 856 }
 857 #endif
 858
 859 static int raw_create(const char *filename, int64_t total_size,
 860                       const char *backing_file, int flags)
 861 {
 862     int fd;
 863
 864     if (flags || backing_file)
 865         return -ENOTSUP;
 866
 867     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 868               0644);
 869     if (fd < 0)
 870         return -EIO;
 871     ftruncate(fd, total_size * 512);
 872     close(fd);
 873     return 0;
 874 }
 875
 876 static void raw_flush(BlockDriverState *bs)
 877 {
 878     BDRVRawState *s = bs->opaque;
 879     fsync(s->fd);
 880 }
 881
 882 BlockDriver bdrv_raw = {
 883     "raw",
 884     sizeof(BDRVRawState),
 885     NULL, /* no probe for protocols */
 886     raw_open,
 887     NULL,
 888     NULL,
 889     raw_close,
 890     raw_create,
 891     raw_flush,
 892
 893 #ifdef CONFIG_AIO
 894     .bdrv_aio_read = raw_aio_read,
 895     .bdrv_aio_write = raw_aio_write,
 896     .bdrv_aio_cancel = raw_aio_cancel,
 897     .aiocb_size = sizeof(RawAIOCB),
 898 #endif
 899     .bdrv_pread = raw_pread,
 900     .bdrv_pwrite = raw_pwrite,
 901     .bdrv_truncate = raw_truncate,
 902     .bdrv_getlength = raw_getlength,
 903 };
 904
 905 /***********************************************/
 906 /* host device */
 907
 908 #ifdef CONFIG_COCOA
 909 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 910 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 911
 912 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 913 {
 914     kern_return_t       kernResult;
 915     mach_port_t     masterPort;
 916     CFMutableDictionaryRef  classesToMatch;
 917
 918     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 919     if ( KERN_SUCCESS != kernResult ) {
 920         printf( "IOMasterPort returned %d\n", kernResult );
 921     }
 922
 923     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 924     if ( classesToMatch == NULL ) {
 925         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 926     } else {
 927     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 928     }
 929     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 930     if ( KERN_SUCCESS != kernResult )
 931     {
 932         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 933     }
 934
 935     return kernResult;
 936 }
 937
 938 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 939 {
 940     io_object_t     nextMedia;
 941     kern_return_t   kernResult = KERN_FAILURE;
 942     *bsdPath = '\0';
 943     nextMedia = IOIteratorNext( mediaIterator );
 944     if ( nextMedia )
 945     {
 946         CFTypeRef   bsdPathAsCFString;
 947     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 948         if ( bsdPathAsCFString ) {
 949             size_t devPathLength;
 950             strcpy( bsdPath, _PATH_DEV );
 951             strcat( bsdPath, "r" );
 952             devPathLength = strlen( bsdPath );
 953             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 954                 kernResult = KERN_SUCCESS;
 955             }
 956             CFRelease( bsdPathAsCFString );
 957         }
 958         IOObjectRelease( nextMedia );
 959     }
 960
 961     return kernResult;
 962 }
 963
 964 #endif
 965
 966 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 967 {
 968     BDRVRawState *s = bs->opaque;
 969     int fd, open_flags, ret, i;
 970
 971     posix_aio_init();
 972
 973 #ifdef CONFIG_COCOA
 974     if (strstart(filename, "/dev/cdrom", NULL)) {
 975         kern_return_t kernResult;
 976         io_iterator_t mediaIterator;
 977         char bsdPath[ MAXPATHLEN ];
 978         int fd;
 979
 980         kernResult = FindEjectableCDMedia( &mediaIterator );
 981         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 982
 983         if ( bsdPath[ 0 ] != '\0' ) {
 984             strcat(bsdPath,"s0");
 985             /* some CDs don't have a partition 0 */
 986             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 987             if (fd < 0) {
 988                 bsdPath[strlen(bsdPath)-1] = '1';
 989             } else {
 990                 close(fd);
 991             }
 992             filename = bsdPath;
 993         }
 994
 995         if ( mediaIterator )
 996             IOObjectRelease( mediaIterator );
 997     }
 998 #endif
 999     open_flags = O_BINARY;
1000     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
1001         open_flags |= O_RDWR;
1002     } else {
1003         open_flags |= O_RDONLY;
1004         bs->read_only = 1;
1005     }
1006 #ifdef O_DIRECT
1007     if (flags & BDRV_O_DIRECT)
1008         open_flags |= O_DIRECT;
1009 #endif
1010
1011     s->type = FTYPE_FILE;
1012 #if defined(__linux__)
1013     if (strstart(filename, "/dev/cd", NULL)) {
1014         /* open will not fail even if no CD is inserted */
1015         open_flags |= O_NONBLOCK;
1016         s->type = FTYPE_CD;
1017     } else if (strstart(filename, "/dev/fd", NULL)) {
1018         s->type = FTYPE_FD;
1019         s->fd_open_flags = open_flags;
1020         /* open will not fail even if no floppy is inserted */
1021         open_flags |= O_NONBLOCK;
1022     } else if (strstart(filename, "/dev/sg", NULL)) {
1023         bs->sg = 1;
1024     }
1025 #endif
1026     fd = open(filename, open_flags, 0644);
1027     if (fd < 0) {
1028         ret = -errno;
1029         if (ret == -EROFS)
1030             ret = -EACCES;
1031         return ret;
1032     }
1033     s->fd = fd;
1034     for (i = 0; i < RAW_FD_POOL_SIZE; i++)
1035         s->fd_pool[i] = -1;
1036 #if defined(__linux__)
1037     /* close fd so that we can reopen it as needed */
1038     if (s->type == FTYPE_FD) {
1039         close(s->fd);
1040         s->fd = -1;
1041         s->fd_media_changed = 1;
1042     }
1043 #endif
1044     return 0;
1045 }
1046
1047 #if defined(__linux__)
1048 /* Note: we do not have a reliable method to detect if the floppy is
1049    present. The current method is to try to open the floppy at every
1050    I/O and to keep it opened during a few hundreds of ms. */
1051 static int fd_open(BlockDriverState *bs)
1052 {
1053     BDRVRawState *s = bs->opaque;
1054     int last_media_present;
1055
1056     if (s->type != FTYPE_FD)
1057         return 0;
1058     last_media_present = (s->fd >= 0);
1059     if (s->fd >= 0 &&
1060         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1061         close(s->fd);
1062         s->fd = -1;
1063         raw_close_fd_pool(s);
1064 #ifdef DEBUG_FLOPPY
1065         printf("Floppy closed\n");
1066 #endif
1067     }
1068     if (s->fd < 0) {
1069         if (s->fd_got_error &&
1070             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1071 #ifdef DEBUG_FLOPPY
1072             printf("No floppy (open delayed)\n");
1073 #endif
1074             return -EIO;
1075         }
1076         s->fd = open(bs->filename, s->fd_open_flags);
1077         if (s->fd < 0) {
1078             s->fd_error_time = qemu_get_clock(rt_clock);
1079             s->fd_got_error = 1;
1080             if (last_media_present)
1081                 s->fd_media_changed = 1;
1082 #ifdef DEBUG_FLOPPY
1083             printf("No floppy\n");
1084 #endif
1085             return -EIO;
1086         }
1087 #ifdef DEBUG_FLOPPY
1088         printf("Floppy opened\n");
1089 #endif
1090     }
1091     if (!last_media_present)
1092         s->fd_media_changed = 1;
1093     s->fd_open_time = qemu_get_clock(rt_clock);
1094     s->fd_got_error = 0;
1095     return 0;
1096 }
1097
1098 static int raw_is_inserted(BlockDriverState *bs)
1099 {
1100     BDRVRawState *s = bs->opaque;
1101     int ret;
1102
1103     switch(s->type) {
1104     case FTYPE_CD:
1105         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1106         if (ret == CDS_DISC_OK)
1107             return 1;
1108         else
1109             return 0;
1110         break;
1111     case FTYPE_FD:
1112         ret = fd_open(bs);
1113         return (ret >= 0);
1114     default:
1115         return 1;
1116     }
1117 }
1118
1119 /* currently only used by fdc.c, but a CD version would be good too */
1120 static int raw_media_changed(BlockDriverState *bs)
1121 {
1122     BDRVRawState *s = bs->opaque;
1123
1124     switch(s->type) {
1125     case FTYPE_FD:
1126         {
1127             int ret;
1128             /* XXX: we do not have a true media changed indication. It
1129                does not work if the floppy is changed without trying
1130                to read it */
1131             fd_open(bs);
1132             ret = s->fd_media_changed;
1133             s->fd_media_changed = 0;
1134 #ifdef DEBUG_FLOPPY
1135             printf("Floppy changed=%d\n", ret);
1136 #endif
1137             return ret;
1138         }
1139     default:
1140         return -ENOTSUP;
1141     }
1142 }
1143
1144 static int raw_eject(BlockDriverState *bs, int eject_flag)
1145 {
1146     BDRVRawState *s = bs->opaque;
1147
1148     switch(s->type) {
1149     case FTYPE_CD:
1150         if (eject_flag) {
1151             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1152                 perror("CDROMEJECT");
1153         } else {
1154             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1155                 perror("CDROMEJECT");
1156         }
1157         break;
1158     case FTYPE_FD:
1159         {
1160             int fd;
1161             if (s->fd >= 0) {
1162                 close(s->fd);
1163                 s->fd = -1;
1164                 raw_close_fd_pool(s);
1165             }
1166             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1167             if (fd >= 0) {
1168                 if (ioctl(fd, FDEJECT, 0) < 0)
1169                     perror("FDEJECT");
1170                 close(fd);
1171             }
1172         }
1173         break;
1174     default:
1175         return -ENOTSUP;
1176     }
1177     return 0;
1178 }
1179
1180 static int raw_set_locked(BlockDriverState *bs, int locked)
1181 {
1182     BDRVRawState *s = bs->opaque;
1183
1184     switch(s->type) {
1185     case FTYPE_CD:
1186         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1187             /* Note: an error can happen if the distribution automatically
1188                mounts the CD-ROM */
1189             //        perror("CDROM_LOCKDOOR");
1190         }
1191         break;
1192     default:
1193         return -ENOTSUP;
1194     }
1195     return 0;
1196 }
1197
1198 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1199 {
1200     BDRVRawState *s = bs->opaque;
1201
1202     return ioctl(s->fd, req, buf);
1203 }
1204 #else
1205
1206 static int fd_open(BlockDriverState *bs)
1207 {
1208     return 0;
1209 }
1210
1211 static int raw_is_inserted(BlockDriverState *bs)
1212 {
1213     return 1;
1214 }
1215
1216 static int raw_media_changed(BlockDriverState *bs)
1217 {
1218     return -ENOTSUP;
1219 }
1220
1221 static int raw_eject(BlockDriverState *bs, int eject_flag)
1222 {
1223     return -ENOTSUP;
1224 }
1225
1226 static int raw_set_locked(BlockDriverState *bs, int locked)
1227 {
1228     return -ENOTSUP;
1229 }
1230
1231 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1232 {
1233     return -ENOTSUP;
1234 }
1235 #endif /* !linux */
1236
1237 BlockDriver bdrv_host_device = {
1238     "host_device",
1239     sizeof(BDRVRawState),
1240     NULL, /* no probe for protocols */
1241     hdev_open,
1242     NULL,
1243     NULL,
1244     raw_close,
1245     NULL,
1246     raw_flush,
1247
1248 #ifdef CONFIG_AIO
1249     .bdrv_aio_read = raw_aio_read,
1250     .bdrv_aio_write = raw_aio_write,
1251     .bdrv_aio_cancel = raw_aio_cancel,
1252     .aiocb_size = sizeof(RawAIOCB),
1253 #endif
1254     .bdrv_pread = raw_pread,
1255     .bdrv_pwrite = raw_pwrite,
1256     .bdrv_getlength = raw_getlength,
1257
1258     /* removable device support */
1259     .bdrv_is_inserted = raw_is_inserted,
1260     .bdrv_media_changed = raw_media_changed,
1261     .bdrv_eject = raw_eject,
1262     .bdrv_set_locked = raw_set_locked,
1263     /* generic scsi device */
1264     .bdrv_ioctl = raw_ioctl,
1265 };