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/log.h"
  27 #include "block/block_int.h"
  28 #include "qemu/module.h"
  29 #include "trace.h"
  30 #include "block/thread-pool.h"
  31 #include "qemu/iov.h"
  32 #include "raw-aio.h"
  33
  34 #if defined(__APPLE__) && (__MACH__)
  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 <sys/dkio.h>
  49 #endif
  50 #ifdef __linux__
  51 #include <sys/types.h>
  52 #include <sys/stat.h>
  53 #include <sys/ioctl.h>
  54 #include <sys/param.h>
  55 #include <linux/cdrom.h>
  56 #include <linux/fd.h>
  57 #include <linux/fs.h>
  58 #endif
  59 #ifdef CONFIG_FIEMAP
  60 #include <linux/fiemap.h>
  61 #endif
  62 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
  63 #include <sys/disk.h>
  64 #include <sys/cdio.h>
  65 #endif
  66
  67 #ifdef __OpenBSD__
  68 #include <sys/ioctl.h>
  69 #include <sys/disklabel.h>
  70 #include <sys/dkio.h>
  71 #endif
  72
  73 #ifdef __NetBSD__
  74 #include <sys/ioctl.h>
  75 #include <sys/disklabel.h>
  76 #include <sys/dkio.h>
  77 #include <sys/disk.h>
  78 #endif
  79
  80 #ifdef __DragonFly__
  81 #include <sys/ioctl.h>
  82 #include <sys/diskslice.h>
  83 #endif
  84
  85 #ifdef CONFIG_XFS
  86 #include <xfs/xfs.h>
  87 #endif
  88
  89 //#define DEBUG_FLOPPY
  90
  91 //#define DEBUG_BLOCK
  92 #if defined(DEBUG_BLOCK)
  93 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
  94     { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
  95 #else
  96 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
  97 #endif
  98
  99 /* OS X does not have O_DSYNC */
 100 #ifndef O_DSYNC
 101 #ifdef O_SYNC
 102 #define O_DSYNC O_SYNC
 103 #elif defined(O_FSYNC)
 104 #define O_DSYNC O_FSYNC
 105 #endif
 106 #endif
 107
 108 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
 109 #ifndef O_DIRECT
 110 #define O_DIRECT O_DSYNC
 111 #endif
 112
 113 #define FTYPE_FILE   0
 114 #define FTYPE_CD     1
 115 #define FTYPE_FD     2
 116
 117 /* if the FD is not accessed during that time (in ns), we try to
 118    reopen it to see if the disk has been changed */
 119 #define FD_OPEN_TIMEOUT (1000000000)
 120
 121 #define MAX_BLOCKSIZE   4096
 122
 123 typedef struct BDRVRawState {
 124     int fd;
 125     int type;
 126     int open_flags;
 127 #if defined(__linux__)
 128     /* linux floppy specific */
 129     int64_t fd_open_time;
 130     int64_t fd_error_time;
 131     int fd_got_error;
 132     int fd_media_changed;
 133 #endif
 134 #ifdef CONFIG_LINUX_AIO
 135     int use_aio;
 136     void *aio_ctx;
 137 #endif
 138 #ifdef CONFIG_XFS
 139     bool is_xfs : 1;
 140 #endif
 141 } BDRVRawState;
 142
 143 typedef struct BDRVRawReopenState {
 144     int fd;
 145     int open_flags;
 146 #ifdef CONFIG_LINUX_AIO
 147     int use_aio;
 148 #endif
 149 } BDRVRawReopenState;
 150
 151 static int fd_open(BlockDriverState *bs);
 152 static int64_t raw_getlength(BlockDriverState *bs);
 153
 154 typedef struct RawPosixAIOData {
 155     BlockDriverState *bs;
 156     int aio_fildes;
 157     union {
 158         struct iovec *aio_iov;
 159         void *aio_ioctl_buf;
 160     };
 161     int aio_niov;
 162     size_t aio_nbytes;
 163 #define aio_ioctl_cmd   aio_nbytes /* for QEMU_AIO_IOCTL */
 164     off_t aio_offset;
 165     int aio_type;
 166 } RawPosixAIOData;
 167
 168 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
 169 static int cdrom_reopen(BlockDriverState *bs);
 170 #endif
 171
 172 #if defined(__NetBSD__)
 173 static int raw_normalize_devicepath(const char **filename)
 174 {
 175     static char namebuf[PATH_MAX];
 176     const char *dp, *fname;
 177     struct stat sb;
 178
 179     fname = *filename;
 180     dp = strrchr(fname, '/');
 181     if (lstat(fname, &sb) < 0) {
 182         fprintf(stderr, "%s: stat failed: %s\n",
 183             fname, strerror(errno));
 184         return -errno;
 185     }
 186
 187     if (!S_ISBLK(sb.st_mode)) {
 188         return 0;
 189     }
 190
 191     if (dp == NULL) {
 192         snprintf(namebuf, PATH_MAX, "r%s", fname);
 193     } else {
 194         snprintf(namebuf, PATH_MAX, "%.*s/r%s",
 195             (int)(dp - fname), fname, dp + 1);
 196     }
 197     fprintf(stderr, "%s is a block device", fname);
 198     *filename = namebuf;
 199     fprintf(stderr, ", using %s\n", *filename);
 200
 201     return 0;
 202 }
 203 #else
 204 static int raw_normalize_devicepath(const char **filename)
 205 {
 206     return 0;
 207 }
 208 #endif
 209
 210 static void raw_parse_flags(int bdrv_flags, int *open_flags)
 211 {
 212     assert(open_flags != NULL);
 213
 214     *open_flags |= O_BINARY;
 215     *open_flags &= ~O_ACCMODE;
 216     if (bdrv_flags & BDRV_O_RDWR) {
 217         *open_flags |= O_RDWR;
 218     } else {
 219         *open_flags |= O_RDONLY;
 220     }
 221
 222     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 223      * and O_DIRECT for no caching. */
 224     if ((bdrv_flags & BDRV_O_NOCACHE)) {
 225         *open_flags |= O_DIRECT;
 226     }
 227 }
 228
 229 #ifdef CONFIG_LINUX_AIO
 230 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
 231 {
 232     int ret = -1;
 233     assert(aio_ctx != NULL);
 234     assert(use_aio != NULL);
 235     /*
 236      * Currently Linux do AIO only for files opened with O_DIRECT
 237      * specified so check NOCACHE flag too
 238      */
 239     if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
 240                       (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
 241
 242         /* if non-NULL, laio_init() has already been run */
 243         if (*aio_ctx == NULL) {
 244             *aio_ctx = laio_init();
 245             if (!*aio_ctx) {
 246                 goto error;
 247             }
 248         }
 249         *use_aio = 1;
 250     } else {
 251         *use_aio = 0;
 252     }
 253
 254     ret = 0;
 255
 256 error:
 257     return ret;
 258 }
 259 #endif
 260
 261 static int raw_open_common(BlockDriverState *bs, const char *filename,
 262                            int bdrv_flags, int open_flags)
 263 {
 264     BDRVRawState *s = bs->opaque;
 265     int fd, ret;
 266
 267     ret = raw_normalize_devicepath(&filename);
 268     if (ret != 0) {
 269         return ret;
 270     }
 271
 272     s->open_flags = open_flags;
 273     raw_parse_flags(bdrv_flags, &s->open_flags);
 274
 275     s->fd = -1;
 276     fd = qemu_open(filename, s->open_flags, 0644);
 277     if (fd < 0) {
 278         ret = -errno;
 279         if (ret == -EROFS)
 280             ret = -EACCES;
 281         return ret;
 282     }
 283     s->fd = fd;
 284
 285 #ifdef CONFIG_LINUX_AIO
 286     if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
 287         qemu_close(fd);
 288         return -errno;
 289     }
 290 #endif
 291
 292 #ifdef CONFIG_XFS
 293     if (platform_test_xfs_fd(s->fd)) {
 294         s->is_xfs = 1;
 295     }
 296 #endif
 297
 298     return 0;
 299 }
 300
 301 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 302 {
 303     BDRVRawState *s = bs->opaque;
 304
 305     s->type = FTYPE_FILE;
 306     return raw_open_common(bs, filename, flags, 0);
 307 }
 308
 309 static int raw_reopen_prepare(BDRVReopenState *state,
 310                               BlockReopenQueue *queue, Error **errp)
 311 {
 312     BDRVRawState *s;
 313     BDRVRawReopenState *raw_s;
 314     int ret = 0;
 315
 316     assert(state != NULL);
 317     assert(state->bs != NULL);
 318
 319     s = state->bs->opaque;
 320
 321     state->opaque = g_malloc0(sizeof(BDRVRawReopenState));
 322     raw_s = state->opaque;
 323
 324 #ifdef CONFIG_LINUX_AIO
 325     raw_s->use_aio = s->use_aio;
 326
 327     /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
 328      * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
 329      * won't override aio_ctx if aio_ctx is non-NULL */
 330     if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
 331         return -1;
 332     }
 333 #endif
 334
 335     if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
 336         raw_s->open_flags |= O_NONBLOCK;
 337     }
 338
 339     raw_parse_flags(state->flags, &raw_s->open_flags);
 340
 341     raw_s->fd = -1;
 342
 343     int fcntl_flags = O_APPEND | O_ASYNC | O_NONBLOCK;
 344 #ifdef O_NOATIME
 345     fcntl_flags |= O_NOATIME;
 346 #endif
 347
 348     if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
 349         /* dup the original fd */
 350         /* TODO: use qemu fcntl wrapper */
 351 #ifdef F_DUPFD_CLOEXEC
 352         raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
 353 #else
 354         raw_s->fd = dup(s->fd);
 355         if (raw_s->fd != -1) {
 356             qemu_set_cloexec(raw_s->fd);
 357         }
 358 #endif
 359         if (raw_s->fd >= 0) {
 360             ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
 361             if (ret) {
 362                 qemu_close(raw_s->fd);
 363                 raw_s->fd = -1;
 364             }
 365         }
 366     }
 367
 368     /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
 369     if (raw_s->fd == -1) {
 370         assert(!(raw_s->open_flags & O_CREAT));
 371         raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
 372         if (raw_s->fd == -1) {
 373             ret = -1;
 374         }
 375     }
 376     return ret;
 377 }
 378
 379
 380 static void raw_reopen_commit(BDRVReopenState *state)
 381 {
 382     BDRVRawReopenState *raw_s = state->opaque;
 383     BDRVRawState *s = state->bs->opaque;
 384
 385     s->open_flags = raw_s->open_flags;
 386
 387     qemu_close(s->fd);
 388     s->fd = raw_s->fd;
 389 #ifdef CONFIG_LINUX_AIO
 390     s->use_aio = raw_s->use_aio;
 391 #endif
 392
 393     g_free(state->opaque);
 394     state->opaque = NULL;
 395 }
 396
 397
 398 static void raw_reopen_abort(BDRVReopenState *state)
 399 {
 400     BDRVRawReopenState *raw_s = state->opaque;
 401
 402      /* nothing to do if NULL, we didn't get far enough */
 403     if (raw_s == NULL) {
 404         return;
 405     }
 406
 407     if (raw_s->fd >= 0) {
 408         qemu_close(raw_s->fd);
 409         raw_s->fd = -1;
 410     }
 411     g_free(state->opaque);
 412     state->opaque = NULL;
 413 }
 414
 415
 416 /* XXX: use host sector size if necessary with:
 417 #ifdef DIOCGSECTORSIZE
 418         {
 419             unsigned int sectorsize = 512;
 420             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 421                 sectorsize > bufsize)
 422                 bufsize = sectorsize;
 423         }
 424 #endif
 425 #ifdef CONFIG_COCOA
 426         uint32_t blockSize = 512;
 427         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 428             bufsize = blockSize;
 429         }
 430 #endif
 431 */
 432
 433 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
 434 {
 435     int ret;
 436
 437     ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
 438     if (ret == -1) {
 439         return -errno;
 440     }
 441
 442     return 0;
 443 }
 444
 445 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
 446 {
 447     int ret;
 448
 449     ret = qemu_fdatasync(aiocb->aio_fildes);
 450     if (ret == -1) {
 451         return -errno;
 452     }
 453     return 0;
 454 }
 455
 456 #ifdef CONFIG_PREADV
 457
 458 static bool preadv_present = true;
 459
 460 static ssize_t
 461 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 462 {
 463     return preadv(fd, iov, nr_iov, offset);
 464 }
 465
 466 static ssize_t
 467 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 468 {
 469     return pwritev(fd, iov, nr_iov, offset);
 470 }
 471
 472 #else
 473
 474 static bool preadv_present = false;
 475
 476 static ssize_t
 477 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 478 {
 479     return -ENOSYS;
 480 }
 481
 482 static ssize_t
 483 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 484 {
 485     return -ENOSYS;
 486 }
 487
 488 #endif
 489
 490 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
 491 {
 492     ssize_t len;
 493
 494     do {
 495         if (aiocb->aio_type & QEMU_AIO_WRITE)
 496             len = qemu_pwritev(aiocb->aio_fildes,
 497                                aiocb->aio_iov,
 498                                aiocb->aio_niov,
 499                                aiocb->aio_offset);
 500          else
 501             len = qemu_preadv(aiocb->aio_fildes,
 502                               aiocb->aio_iov,
 503                               aiocb->aio_niov,
 504                               aiocb->aio_offset);
 505     } while (len == -1 && errno == EINTR);
 506
 507     if (len == -1) {
 508         return -errno;
 509     }
 510     return len;
 511 }
 512
 513 /*
 514  * Read/writes the data to/from a given linear buffer.
 515  *
 516  * Returns the number of bytes handles or -errno in case of an error. Short
 517  * reads are only returned if the end of the file is reached.
 518  */
 519 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
 520 {
 521     ssize_t offset = 0;
 522     ssize_t len;
 523
 524     while (offset < aiocb->aio_nbytes) {
 525         if (aiocb->aio_type & QEMU_AIO_WRITE) {
 526             len = pwrite(aiocb->aio_fildes,
 527                          (const char *)buf + offset,
 528                          aiocb->aio_nbytes - offset,
 529                          aiocb->aio_offset + offset);
 530         } else {
 531             len = pread(aiocb->aio_fildes,
 532                         buf + offset,
 533                         aiocb->aio_nbytes - offset,
 534                         aiocb->aio_offset + offset);
 535         }
 536         if (len == -1 && errno == EINTR) {
 537             continue;
 538         } else if (len == -1) {
 539             offset = -errno;
 540             break;
 541         } else if (len == 0) {
 542             break;
 543         }
 544         offset += len;
 545     }
 546
 547     return offset;
 548 }
 549
 550 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
 551 {
 552     ssize_t nbytes;
 553     char *buf;
 554
 555     if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
 556         /*
 557          * If there is just a single buffer, and it is properly aligned
 558          * we can just use plain pread/pwrite without any problems.
 559          */
 560         if (aiocb->aio_niov == 1) {
 561              return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
 562         }
 563         /*
 564          * We have more than one iovec, and all are properly aligned.
 565          *
 566          * Try preadv/pwritev first and fall back to linearizing the
 567          * buffer if it's not supported.
 568          */
 569         if (preadv_present) {
 570             nbytes = handle_aiocb_rw_vector(aiocb);
 571             if (nbytes == aiocb->aio_nbytes ||
 572                 (nbytes < 0 && nbytes != -ENOSYS)) {
 573                 return nbytes;
 574             }
 575             preadv_present = false;
 576         }
 577
 578         /*
 579          * XXX(hch): short read/write.  no easy way to handle the reminder
 580          * using these interfaces.  For now retry using plain
 581          * pread/pwrite?
 582          */
 583     }
 584
 585     /*
 586      * Ok, we have to do it the hard way, copy all segments into
 587      * a single aligned buffer.
 588      */
 589     buf = qemu_blockalign(aiocb->bs, aiocb->aio_nbytes);
 590     if (aiocb->aio_type & QEMU_AIO_WRITE) {
 591         char *p = buf;
 592         int i;
 593
 594         for (i = 0; i < aiocb->aio_niov; ++i) {
 595             memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
 596             p += aiocb->aio_iov[i].iov_len;
 597         }
 598     }
 599
 600     nbytes = handle_aiocb_rw_linear(aiocb, buf);
 601     if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
 602         char *p = buf;
 603         size_t count = aiocb->aio_nbytes, copy;
 604         int i;
 605
 606         for (i = 0; i < aiocb->aio_niov && count; ++i) {
 607             copy = count;
 608             if (copy > aiocb->aio_iov[i].iov_len) {
 609                 copy = aiocb->aio_iov[i].iov_len;
 610             }
 611             memcpy(aiocb->aio_iov[i].iov_base, p, copy);
 612             p     += copy;
 613             count -= copy;
 614         }
 615     }
 616     qemu_vfree(buf);
 617
 618     return nbytes;
 619 }
 620
 621 static int aio_worker(void *arg)
 622 {
 623     RawPosixAIOData *aiocb = arg;
 624     ssize_t ret = 0;
 625
 626     switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
 627     case QEMU_AIO_READ:
 628         ret = handle_aiocb_rw(aiocb);
 629         if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) {
 630             iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
 631                       0, aiocb->aio_nbytes - ret);
 632
 633             ret = aiocb->aio_nbytes;
 634         }
 635         if (ret == aiocb->aio_nbytes) {
 636             ret = 0;
 637         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 638             ret = -EINVAL;
 639         }
 640         break;
 641     case QEMU_AIO_WRITE:
 642         ret = handle_aiocb_rw(aiocb);
 643         if (ret == aiocb->aio_nbytes) {
 644             ret = 0;
 645         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 646             ret = -EINVAL;
 647         }
 648         break;
 649     case QEMU_AIO_FLUSH:
 650         ret = handle_aiocb_flush(aiocb);
 651         break;
 652     case QEMU_AIO_IOCTL:
 653         ret = handle_aiocb_ioctl(aiocb);
 654         break;
 655     default:
 656         fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
 657         ret = -EINVAL;
 658         break;
 659     }
 660
 661     g_slice_free(RawPosixAIOData, aiocb);
 662     return ret;
 663 }
 664
 665 static BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd,
 666         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 667         BlockDriverCompletionFunc *cb, void *opaque, int type)
 668 {
 669     RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
 670
 671     acb->bs = bs;
 672     acb->aio_type = type;
 673     acb->aio_fildes = fd;
 674
 675     if (qiov) {
 676         acb->aio_iov = qiov->iov;
 677         acb->aio_niov = qiov->niov;
 678     }
 679     acb->aio_nbytes = nb_sectors * 512;
 680     acb->aio_offset = sector_num * 512;
 681
 682     trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
 683     return thread_pool_submit_aio(aio_worker, acb, cb, opaque);
 684 }
 685
 686 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
 687         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 688         BlockDriverCompletionFunc *cb, void *opaque, int type)
 689 {
 690     BDRVRawState *s = bs->opaque;
 691
 692     if (fd_open(bs) < 0)
 693         return NULL;
 694
 695     /*
 696      * If O_DIRECT is used the buffer needs to be aligned on a sector
 697      * boundary.  Check if this is the case or tell the low-level
 698      * driver that it needs to copy the buffer.
 699      */
 700     if ((bs->open_flags & BDRV_O_NOCACHE)) {
 701         if (!bdrv_qiov_is_aligned(bs, qiov)) {
 702             type |= QEMU_AIO_MISALIGNED;
 703 #ifdef CONFIG_LINUX_AIO
 704         } else if (s->use_aio) {
 705             return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
 706                                nb_sectors, cb, opaque, type);
 707 #endif
 708         }
 709     }
 710
 711     return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
 712                        cb, opaque, type);
 713 }
 714
 715 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
 716         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 717         BlockDriverCompletionFunc *cb, void *opaque)
 718 {
 719     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 720                           cb, opaque, QEMU_AIO_READ);
 721 }
 722
 723 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
 724         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 725         BlockDriverCompletionFunc *cb, void *opaque)
 726 {
 727     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 728                           cb, opaque, QEMU_AIO_WRITE);
 729 }
 730
 731 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
 732         BlockDriverCompletionFunc *cb, void *opaque)
 733 {
 734     BDRVRawState *s = bs->opaque;
 735
 736     if (fd_open(bs) < 0)
 737         return NULL;
 738
 739     return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
 740 }
 741
 742 static void raw_close(BlockDriverState *bs)
 743 {
 744     BDRVRawState *s = bs->opaque;
 745     if (s->fd >= 0) {
 746         qemu_close(s->fd);
 747         s->fd = -1;
 748     }
 749 }
 750
 751 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 752 {
 753     BDRVRawState *s = bs->opaque;
 754     struct stat st;
 755
 756     if (fstat(s->fd, &st)) {
 757         return -errno;
 758     }
 759
 760     if (S_ISREG(st.st_mode)) {
 761         if (ftruncate(s->fd, offset) < 0) {
 762             return -errno;
 763         }
 764     } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 765        if (offset > raw_getlength(bs)) {
 766            return -EINVAL;
 767        }
 768     } else {
 769         return -ENOTSUP;
 770     }
 771
 772     return 0;
 773 }
 774
 775 #ifdef __OpenBSD__
 776 static int64_t raw_getlength(BlockDriverState *bs)
 777 {
 778     BDRVRawState *s = bs->opaque;
 779     int fd = s->fd;
 780     struct stat st;
 781
 782     if (fstat(fd, &st))
 783         return -1;
 784     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 785         struct disklabel dl;
 786
 787         if (ioctl(fd, DIOCGDINFO, &dl))
 788             return -1;
 789         return (uint64_t)dl.d_secsize *
 790             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 791     } else
 792         return st.st_size;
 793 }
 794 #elif defined(__NetBSD__)
 795 static int64_t raw_getlength(BlockDriverState *bs)
 796 {
 797     BDRVRawState *s = bs->opaque;
 798     int fd = s->fd;
 799     struct stat st;
 800
 801     if (fstat(fd, &st))
 802         return -1;
 803     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 804         struct dkwedge_info dkw;
 805
 806         if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
 807             return dkw.dkw_size * 512;
 808         } else {
 809             struct disklabel dl;
 810
 811             if (ioctl(fd, DIOCGDINFO, &dl))
 812                 return -1;
 813             return (uint64_t)dl.d_secsize *
 814                 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 815         }
 816     } else
 817         return st.st_size;
 818 }
 819 #elif defined(__sun__)
 820 static int64_t raw_getlength(BlockDriverState *bs)
 821 {
 822     BDRVRawState *s = bs->opaque;
 823     struct dk_minfo minfo;
 824     int ret;
 825
 826     ret = fd_open(bs);
 827     if (ret < 0) {
 828         return ret;
 829     }
 830
 831     /*
 832      * Use the DKIOCGMEDIAINFO ioctl to read the size.
 833      */
 834     ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
 835     if (ret != -1) {
 836         return minfo.dki_lbsize * minfo.dki_capacity;
 837     }
 838
 839     /*
 840      * There are reports that lseek on some devices fails, but
 841      * irc discussion said that contingency on contingency was overkill.
 842      */
 843     return lseek(s->fd, 0, SEEK_END);
 844 }
 845 #elif defined(CONFIG_BSD)
 846 static int64_t raw_getlength(BlockDriverState *bs)
 847 {
 848     BDRVRawState *s = bs->opaque;
 849     int fd = s->fd;
 850     int64_t size;
 851     struct stat sb;
 852 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 853     int reopened = 0;
 854 #endif
 855     int ret;
 856
 857     ret = fd_open(bs);
 858     if (ret < 0)
 859         return ret;
 860
 861 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 862 again:
 863 #endif
 864     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 865 #ifdef DIOCGMEDIASIZE
 866         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 867 #elif defined(DIOCGPART)
 868         {
 869                 struct partinfo pi;
 870                 if (ioctl(fd, DIOCGPART, &pi) == 0)
 871                         size = pi.media_size;
 872                 else
 873                         size = 0;
 874         }
 875         if (size == 0)
 876 #endif
 877 #if defined(__APPLE__) && defined(__MACH__)
 878         size = LONG_LONG_MAX;
 879 #else
 880         size = lseek(fd, 0LL, SEEK_END);
 881 #endif
 882 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
 883         switch(s->type) {
 884         case FTYPE_CD:
 885             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
 886             if (size == 2048LL * (unsigned)-1)
 887                 size = 0;
 888             /* XXX no disc?  maybe we need to reopen... */
 889             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
 890                 reopened = 1;
 891                 goto again;
 892             }
 893         }
 894 #endif
 895     } else {
 896         size = lseek(fd, 0, SEEK_END);
 897     }
 898     return size;
 899 }
 900 #else
 901 static int64_t raw_getlength(BlockDriverState *bs)
 902 {
 903     BDRVRawState *s = bs->opaque;
 904     int ret;
 905
 906     ret = fd_open(bs);
 907     if (ret < 0) {
 908         return ret;
 909     }
 910
 911     return lseek(s->fd, 0, SEEK_END);
 912 }
 913 #endif
 914
 915 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
 916 {
 917     struct stat st;
 918     BDRVRawState *s = bs->opaque;
 919
 920     if (fstat(s->fd, &st) < 0) {
 921         return -errno;
 922     }
 923     return (int64_t)st.st_blocks * 512;
 924 }
 925
 926 static int raw_create(const char *filename, QEMUOptionParameter *options)
 927 {
 928     int fd;
 929     int result = 0;
 930     int64_t total_size = 0;
 931
 932     /* Read out options */
 933     while (options && options->name) {
 934         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 935             total_size = options->value.n / BDRV_SECTOR_SIZE;
 936         }
 937         options++;
 938     }
 939
 940     fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 941                    0644);
 942     if (fd < 0) {
 943         result = -errno;
 944     } else {
 945         if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
 946             result = -errno;
 947         }
 948         if (qemu_close(fd) != 0) {
 949             result = -errno;
 950         }
 951     }
 952     return result;
 953 }
 954
 955 /*
 956  * Returns true iff the specified sector is present in the disk image. Drivers
 957  * not implementing the functionality are assumed to not support backing files,
 958  * hence all their sectors are reported as allocated.
 959  *
 960  * If 'sector_num' is beyond the end of the disk image the return value is 0
 961  * and 'pnum' is set to 0.
 962  *
 963  * 'pnum' is set to the number of sectors (including and immediately following
 964  * the specified sector) that are known to be in the same
 965  * allocated/unallocated state.
 966  *
 967  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
 968  * beyond the end of the disk image it will be clamped.
 969  */
 970 static int coroutine_fn raw_co_is_allocated(BlockDriverState *bs,
 971                                             int64_t sector_num,
 972                                             int nb_sectors, int *pnum)
 973 {
 974     off_t start, data, hole;
 975     int ret;
 976
 977     ret = fd_open(bs);
 978     if (ret < 0) {
 979         return ret;
 980     }
 981
 982     start = sector_num * BDRV_SECTOR_SIZE;
 983
 984 #ifdef CONFIG_FIEMAP
 985
 986     BDRVRawState *s = bs->opaque;
 987     struct {
 988         struct fiemap fm;
 989         struct fiemap_extent fe;
 990     } f;
 991
 992     f.fm.fm_start = start;
 993     f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE;
 994     f.fm.fm_flags = 0;
 995     f.fm.fm_extent_count = 1;
 996     f.fm.fm_reserved = 0;
 997     if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) {
 998         /* Assume everything is allocated.  */
 999         *pnum = nb_sectors;
1000         return 1;
1001     }
1002
1003     if (f.fm.fm_mapped_extents == 0) {
1004         /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1005          * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1006          */
1007         off_t length = lseek(s->fd, 0, SEEK_END);
1008         hole = f.fm.fm_start;
1009         data = MIN(f.fm.fm_start + f.fm.fm_length, length);
1010     } else {
1011         data = f.fe.fe_logical;
1012         hole = f.fe.fe_logical + f.fe.fe_length;
1013     }
1014
1015 #elif defined SEEK_HOLE && defined SEEK_DATA
1016
1017     BDRVRawState *s = bs->opaque;
1018
1019     hole = lseek(s->fd, start, SEEK_HOLE);
1020     if (hole == -1) {
1021         /* -ENXIO indicates that sector_num was past the end of the file.
1022          * There is a virtual hole there.  */
1023         assert(errno != -ENXIO);
1024
1025         /* Most likely EINVAL.  Assume everything is allocated.  */
1026         *pnum = nb_sectors;
1027         return 1;
1028     }
1029
1030     if (hole > start) {
1031         data = start;
1032     } else {
1033         /* On a hole.  We need another syscall to find its end.  */
1034         data = lseek(s->fd, start, SEEK_DATA);
1035         if (data == -1) {
1036             data = lseek(s->fd, 0, SEEK_END);
1037         }
1038     }
1039 #else
1040     *pnum = nb_sectors;
1041     return 1;
1042 #endif
1043
1044     if (data <= start) {
1045         /* On a data extent, compute sectors to the end of the extent.  */
1046         *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1047         return 1;
1048     } else {
1049         /* On a hole, compute sectors to the beginning of the next extent.  */
1050         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1051         return 0;
1052     }
1053 }
1054
1055 #ifdef CONFIG_XFS
1056 static int xfs_discard(BDRVRawState *s, int64_t sector_num, int nb_sectors)
1057 {
1058     struct xfs_flock64 fl;
1059
1060     memset(&fl, 0, sizeof(fl));
1061     fl.l_whence = SEEK_SET;
1062     fl.l_start = sector_num << 9;
1063     fl.l_len = (int64_t)nb_sectors << 9;
1064
1065     if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
1066         DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
1067         return -errno;
1068     }
1069
1070     return 0;
1071 }
1072 #endif
1073
1074 static coroutine_fn int raw_co_discard(BlockDriverState *bs,
1075     int64_t sector_num, int nb_sectors)
1076 {
1077 #ifdef CONFIG_XFS
1078     BDRVRawState *s = bs->opaque;
1079
1080     if (s->is_xfs) {
1081         return xfs_discard(s, sector_num, nb_sectors);
1082     }
1083 #endif
1084
1085     return 0;
1086 }
1087
1088 static QEMUOptionParameter raw_create_options[] = {
1089     {
1090         .name = BLOCK_OPT_SIZE,
1091         .type = OPT_SIZE,
1092         .help = "Virtual disk size"
1093     },
1094     { NULL }
1095 };
1096
1097 static BlockDriver bdrv_file = {
1098     .format_name = "file",
1099     .protocol_name = "file",
1100     .instance_size = sizeof(BDRVRawState),
1101     .bdrv_probe = NULL, /* no probe for protocols */
1102     .bdrv_file_open = raw_open,
1103     .bdrv_reopen_prepare = raw_reopen_prepare,
1104     .bdrv_reopen_commit = raw_reopen_commit,
1105     .bdrv_reopen_abort = raw_reopen_abort,
1106     .bdrv_close = raw_close,
1107     .bdrv_create = raw_create,
1108     .bdrv_co_discard = raw_co_discard,
1109     .bdrv_co_is_allocated = raw_co_is_allocated,
1110
1111     .bdrv_aio_readv = raw_aio_readv,
1112     .bdrv_aio_writev = raw_aio_writev,
1113     .bdrv_aio_flush = raw_aio_flush,
1114
1115     .bdrv_truncate = raw_truncate,
1116     .bdrv_getlength = raw_getlength,
1117     .bdrv_get_allocated_file_size
1118                         = raw_get_allocated_file_size,
1119
1120     .create_options = raw_create_options,
1121 };
1122
1123 /***********************************************/
1124 /* host device */
1125
1126 #if defined(__APPLE__) && defined(__MACH__)
1127 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1128 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1129
1130 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1131 {
1132     kern_return_t       kernResult;
1133     mach_port_t     masterPort;
1134     CFMutableDictionaryRef  classesToMatch;
1135
1136     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1137     if ( KERN_SUCCESS != kernResult ) {
1138         printf( "IOMasterPort returned %d\n", kernResult );
1139     }
1140
1141     classesToMatch = IOServiceMatching( kIOCDMediaClass );
1142     if ( classesToMatch == NULL ) {
1143         printf( "IOServiceMatching returned a NULL dictionary.\n" );
1144     } else {
1145     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1146     }
1147     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1148     if ( KERN_SUCCESS != kernResult )
1149     {
1150         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1151     }
1152
1153     return kernResult;
1154 }
1155
1156 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
1157 {
1158     io_object_t     nextMedia;
1159     kern_return_t   kernResult = KERN_FAILURE;
1160     *bsdPath = '\0';
1161     nextMedia = IOIteratorNext( mediaIterator );
1162     if ( nextMedia )
1163     {
1164         CFTypeRef   bsdPathAsCFString;
1165     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1166         if ( bsdPathAsCFString ) {
1167             size_t devPathLength;
1168             strcpy( bsdPath, _PATH_DEV );
1169             strcat( bsdPath, "r" );
1170             devPathLength = strlen( bsdPath );
1171             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1172                 kernResult = KERN_SUCCESS;
1173             }
1174             CFRelease( bsdPathAsCFString );
1175         }
1176         IOObjectRelease( nextMedia );
1177     }
1178
1179     return kernResult;
1180 }
1181
1182 #endif
1183
1184 static int hdev_probe_device(const char *filename)
1185 {
1186     struct stat st;
1187
1188     /* allow a dedicated CD-ROM driver to match with a higher priority */
1189     if (strstart(filename, "/dev/cdrom", NULL))
1190         return 50;
1191
1192     if (stat(filename, &st) >= 0 &&
1193             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
1194         return 100;
1195     }
1196
1197     return 0;
1198 }
1199
1200 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
1201 {
1202     BDRVRawState *s = bs->opaque;
1203
1204 #if defined(__APPLE__) && defined(__MACH__)
1205     if (strstart(filename, "/dev/cdrom", NULL)) {
1206         kern_return_t kernResult;
1207         io_iterator_t mediaIterator;
1208         char bsdPath[ MAXPATHLEN ];
1209         int fd;
1210
1211         kernResult = FindEjectableCDMedia( &mediaIterator );
1212         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1213
1214         if ( bsdPath[ 0 ] != '\0' ) {
1215             strcat(bsdPath,"s0");
1216             /* some CDs don't have a partition 0 */
1217             fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1218             if (fd < 0) {
1219                 bsdPath[strlen(bsdPath)-1] = '1';
1220             } else {
1221                 qemu_close(fd);
1222             }
1223             filename = bsdPath;
1224         }
1225
1226         if ( mediaIterator )
1227             IOObjectRelease( mediaIterator );
1228     }
1229 #endif
1230
1231     s->type = FTYPE_FILE;
1232 #if defined(__linux__)
1233     {
1234         char resolved_path[ MAXPATHLEN ], *temp;
1235
1236         temp = realpath(filename, resolved_path);
1237         if (temp && strstart(temp, "/dev/sg", NULL)) {
1238             bs->sg = 1;
1239         }
1240     }
1241 #endif
1242
1243     return raw_open_common(bs, filename, flags, 0);
1244 }
1245
1246 #if defined(__linux__)
1247 /* Note: we do not have a reliable method to detect if the floppy is
1248    present. The current method is to try to open the floppy at every
1249    I/O and to keep it opened during a few hundreds of ms. */
1250 static int fd_open(BlockDriverState *bs)
1251 {
1252     BDRVRawState *s = bs->opaque;
1253     int last_media_present;
1254
1255     if (s->type != FTYPE_FD)
1256         return 0;
1257     last_media_present = (s->fd >= 0);
1258     if (s->fd >= 0 &&
1259         (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1260         qemu_close(s->fd);
1261         s->fd = -1;
1262 #ifdef DEBUG_FLOPPY
1263         printf("Floppy closed\n");
1264 #endif
1265     }
1266     if (s->fd < 0) {
1267         if (s->fd_got_error &&
1268             (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1269 #ifdef DEBUG_FLOPPY
1270             printf("No floppy (open delayed)\n");
1271 #endif
1272             return -EIO;
1273         }
1274         s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
1275         if (s->fd < 0) {
1276             s->fd_error_time = get_clock();
1277             s->fd_got_error = 1;
1278             if (last_media_present)
1279                 s->fd_media_changed = 1;
1280 #ifdef DEBUG_FLOPPY
1281             printf("No floppy\n");
1282 #endif
1283             return -EIO;
1284         }
1285 #ifdef DEBUG_FLOPPY
1286         printf("Floppy opened\n");
1287 #endif
1288     }
1289     if (!last_media_present)
1290         s->fd_media_changed = 1;
1291     s->fd_open_time = get_clock();
1292     s->fd_got_error = 0;
1293     return 0;
1294 }
1295
1296 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1297 {
1298     BDRVRawState *s = bs->opaque;
1299
1300     return ioctl(s->fd, req, buf);
1301 }
1302
1303 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1304         unsigned long int req, void *buf,
1305         BlockDriverCompletionFunc *cb, void *opaque)
1306 {
1307     BDRVRawState *s = bs->opaque;
1308     RawPosixAIOData *acb;
1309
1310     if (fd_open(bs) < 0)
1311         return NULL;
1312
1313     acb = g_slice_new(RawPosixAIOData);
1314     acb->bs = bs;
1315     acb->aio_type = QEMU_AIO_IOCTL;
1316     acb->aio_fildes = s->fd;
1317     acb->aio_offset = 0;
1318     acb->aio_ioctl_buf = buf;
1319     acb->aio_ioctl_cmd = req;
1320     return thread_pool_submit_aio(aio_worker, acb, cb, opaque);
1321 }
1322
1323 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1324 static int fd_open(BlockDriverState *bs)
1325 {
1326     BDRVRawState *s = bs->opaque;
1327
1328     /* this is just to ensure s->fd is sane (its called by io ops) */
1329     if (s->fd >= 0)
1330         return 0;
1331     return -EIO;
1332 }
1333 #else /* !linux && !FreeBSD */
1334
1335 static int fd_open(BlockDriverState *bs)
1336 {
1337     return 0;
1338 }
1339
1340 #endif /* !linux && !FreeBSD */
1341
1342 static int hdev_create(const char *filename, QEMUOptionParameter *options)
1343 {
1344     int fd;
1345     int ret = 0;
1346     struct stat stat_buf;
1347     int64_t total_size = 0;
1348
1349     /* Read out options */
1350     while (options && options->name) {
1351         if (!strcmp(options->name, "size")) {
1352             total_size = options->value.n / BDRV_SECTOR_SIZE;
1353         }
1354         options++;
1355     }
1356
1357     fd = qemu_open(filename, O_WRONLY | O_BINARY);
1358     if (fd < 0)
1359         return -errno;
1360
1361     if (fstat(fd, &stat_buf) < 0)
1362         ret = -errno;
1363     else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode))
1364         ret = -ENODEV;
1365     else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE)
1366         ret = -ENOSPC;
1367
1368     qemu_close(fd);
1369     return ret;
1370 }
1371
1372 static int hdev_has_zero_init(BlockDriverState *bs)
1373 {
1374     return 0;
1375 }
1376
1377 static BlockDriver bdrv_host_device = {
1378     .format_name        = "host_device",
1379     .protocol_name        = "host_device",
1380     .instance_size      = sizeof(BDRVRawState),
1381     .bdrv_probe_device  = hdev_probe_device,
1382     .bdrv_file_open     = hdev_open,
1383     .bdrv_close         = raw_close,
1384     .bdrv_reopen_prepare = raw_reopen_prepare,
1385     .bdrv_reopen_commit  = raw_reopen_commit,
1386     .bdrv_reopen_abort   = raw_reopen_abort,
1387     .bdrv_create        = hdev_create,
1388     .create_options     = raw_create_options,
1389     .bdrv_has_zero_init = hdev_has_zero_init,
1390
1391     .bdrv_aio_readv     = raw_aio_readv,
1392     .bdrv_aio_writev    = raw_aio_writev,
1393     .bdrv_aio_flush     = raw_aio_flush,
1394
1395     .bdrv_truncate      = raw_truncate,
1396     .bdrv_getlength     = raw_getlength,
1397     .bdrv_get_allocated_file_size
1398                         = raw_get_allocated_file_size,
1399
1400     /* generic scsi device */
1401 #ifdef __linux__
1402     .bdrv_ioctl         = hdev_ioctl,
1403     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1404 #endif
1405 };
1406
1407 #ifdef __linux__
1408 static int floppy_open(BlockDriverState *bs, const char *filename, int flags)
1409 {
1410     BDRVRawState *s = bs->opaque;
1411     int ret;
1412
1413     s->type = FTYPE_FD;
1414
1415     /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1416     ret = raw_open_common(bs, filename, flags, O_NONBLOCK);
1417     if (ret)
1418         return ret;
1419
1420     /* close fd so that we can reopen it as needed */
1421     qemu_close(s->fd);
1422     s->fd = -1;
1423     s->fd_media_changed = 1;
1424
1425     return 0;
1426 }
1427
1428 static int floppy_probe_device(const char *filename)
1429 {
1430     int fd, ret;
1431     int prio = 0;
1432     struct floppy_struct fdparam;
1433     struct stat st;
1434
1435     if (strstart(filename, "/dev/fd", NULL) &&
1436         !strstart(filename, "/dev/fdset/", NULL)) {
1437         prio = 50;
1438     }
1439
1440     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1441     if (fd < 0) {
1442         goto out;
1443     }
1444     ret = fstat(fd, &st);
1445     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1446         goto outc;
1447     }
1448
1449     /* Attempt to detect via a floppy specific ioctl */
1450     ret = ioctl(fd, FDGETPRM, &fdparam);
1451     if (ret >= 0)
1452         prio = 100;
1453
1454 outc:
1455     qemu_close(fd);
1456 out:
1457     return prio;
1458 }
1459
1460
1461 static int floppy_is_inserted(BlockDriverState *bs)
1462 {
1463     return fd_open(bs) >= 0;
1464 }
1465
1466 static int floppy_media_changed(BlockDriverState *bs)
1467 {
1468     BDRVRawState *s = bs->opaque;
1469     int ret;
1470
1471     /*
1472      * XXX: we do not have a true media changed indication.
1473      * It does not work if the floppy is changed without trying to read it.
1474      */
1475     fd_open(bs);
1476     ret = s->fd_media_changed;
1477     s->fd_media_changed = 0;
1478 #ifdef DEBUG_FLOPPY
1479     printf("Floppy changed=%d\n", ret);
1480 #endif
1481     return ret;
1482 }
1483
1484 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
1485 {
1486     BDRVRawState *s = bs->opaque;
1487     int fd;
1488
1489     if (s->fd >= 0) {
1490         qemu_close(s->fd);
1491         s->fd = -1;
1492     }
1493     fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
1494     if (fd >= 0) {
1495         if (ioctl(fd, FDEJECT, 0) < 0)
1496             perror("FDEJECT");
1497         qemu_close(fd);
1498     }
1499 }
1500
1501 static BlockDriver bdrv_host_floppy = {
1502     .format_name        = "host_floppy",
1503     .protocol_name      = "host_floppy",
1504     .instance_size      = sizeof(BDRVRawState),
1505     .bdrv_probe_device  = floppy_probe_device,
1506     .bdrv_file_open     = floppy_open,
1507     .bdrv_close         = raw_close,
1508     .bdrv_reopen_prepare = raw_reopen_prepare,
1509     .bdrv_reopen_commit  = raw_reopen_commit,
1510     .bdrv_reopen_abort   = raw_reopen_abort,
1511     .bdrv_create        = hdev_create,
1512     .create_options     = raw_create_options,
1513     .bdrv_has_zero_init = hdev_has_zero_init,
1514
1515     .bdrv_aio_readv     = raw_aio_readv,
1516     .bdrv_aio_writev    = raw_aio_writev,
1517     .bdrv_aio_flush     = raw_aio_flush,
1518
1519     .bdrv_truncate      = raw_truncate,
1520     .bdrv_getlength     = raw_getlength,
1521     .bdrv_get_allocated_file_size
1522                         = raw_get_allocated_file_size,
1523
1524     /* removable device support */
1525     .bdrv_is_inserted   = floppy_is_inserted,
1526     .bdrv_media_changed = floppy_media_changed,
1527     .bdrv_eject         = floppy_eject,
1528 };
1529
1530 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1531 {
1532     BDRVRawState *s = bs->opaque;
1533
1534     s->type = FTYPE_CD;
1535
1536     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
1537     return raw_open_common(bs, filename, flags, O_NONBLOCK);
1538 }
1539
1540 static int cdrom_probe_device(const char *filename)
1541 {
1542     int fd, ret;
1543     int prio = 0;
1544     struct stat st;
1545
1546     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1547     if (fd < 0) {
1548         goto out;
1549     }
1550     ret = fstat(fd, &st);
1551     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1552         goto outc;
1553     }
1554
1555     /* Attempt to detect via a CDROM specific ioctl */
1556     ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1557     if (ret >= 0)
1558         prio = 100;
1559
1560 outc:
1561     qemu_close(fd);
1562 out:
1563     return prio;
1564 }
1565
1566 static int cdrom_is_inserted(BlockDriverState *bs)
1567 {
1568     BDRVRawState *s = bs->opaque;
1569     int ret;
1570
1571     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1572     if (ret == CDS_DISC_OK)
1573         return 1;
1574     return 0;
1575 }
1576
1577 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1578 {
1579     BDRVRawState *s = bs->opaque;
1580
1581     if (eject_flag) {
1582         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
1583             perror("CDROMEJECT");
1584     } else {
1585         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
1586             perror("CDROMEJECT");
1587     }
1588 }
1589
1590 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1591 {
1592     BDRVRawState *s = bs->opaque;
1593
1594     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
1595         /*
1596          * Note: an error can happen if the distribution automatically
1597          * mounts the CD-ROM
1598          */
1599         /* perror("CDROM_LOCKDOOR"); */
1600     }
1601 }
1602
1603 static BlockDriver bdrv_host_cdrom = {
1604     .format_name        = "host_cdrom",
1605     .protocol_name      = "host_cdrom",
1606     .instance_size      = sizeof(BDRVRawState),
1607     .bdrv_probe_device  = cdrom_probe_device,
1608     .bdrv_file_open     = cdrom_open,
1609     .bdrv_close         = raw_close,
1610     .bdrv_reopen_prepare = raw_reopen_prepare,
1611     .bdrv_reopen_commit  = raw_reopen_commit,
1612     .bdrv_reopen_abort   = raw_reopen_abort,
1613     .bdrv_create        = hdev_create,
1614     .create_options     = raw_create_options,
1615     .bdrv_has_zero_init = hdev_has_zero_init,
1616
1617     .bdrv_aio_readv     = raw_aio_readv,
1618     .bdrv_aio_writev    = raw_aio_writev,
1619     .bdrv_aio_flush     = raw_aio_flush,
1620
1621     .bdrv_truncate      = raw_truncate,
1622     .bdrv_getlength     = raw_getlength,
1623     .bdrv_get_allocated_file_size
1624                         = raw_get_allocated_file_size,
1625
1626     /* removable device support */
1627     .bdrv_is_inserted   = cdrom_is_inserted,
1628     .bdrv_eject         = cdrom_eject,
1629     .bdrv_lock_medium   = cdrom_lock_medium,
1630
1631     /* generic scsi device */
1632     .bdrv_ioctl         = hdev_ioctl,
1633     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1634 };
1635 #endif /* __linux__ */
1636
1637 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1638 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1639 {
1640     BDRVRawState *s = bs->opaque;
1641     int ret;
1642
1643     s->type = FTYPE_CD;
1644
1645     ret = raw_open_common(bs, filename, flags, 0);
1646     if (ret)
1647         return ret;
1648
1649     /* make sure the door isn't locked at this time */
1650     ioctl(s->fd, CDIOCALLOW);
1651     return 0;
1652 }
1653
1654 static int cdrom_probe_device(const char *filename)
1655 {
1656     if (strstart(filename, "/dev/cd", NULL) ||
1657             strstart(filename, "/dev/acd", NULL))
1658         return 100;
1659     return 0;
1660 }
1661
1662 static int cdrom_reopen(BlockDriverState *bs)
1663 {
1664     BDRVRawState *s = bs->opaque;
1665     int fd;
1666
1667     /*
1668      * Force reread of possibly changed/newly loaded disc,
1669      * FreeBSD seems to not notice sometimes...
1670      */
1671     if (s->fd >= 0)
1672         qemu_close(s->fd);
1673     fd = qemu_open(bs->filename, s->open_flags, 0644);
1674     if (fd < 0) {
1675         s->fd = -1;
1676         return -EIO;
1677     }
1678     s->fd = fd;
1679
1680     /* make sure the door isn't locked at this time */
1681     ioctl(s->fd, CDIOCALLOW);
1682     return 0;
1683 }
1684
1685 static int cdrom_is_inserted(BlockDriverState *bs)
1686 {
1687     return raw_getlength(bs) > 0;
1688 }
1689
1690 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1691 {
1692     BDRVRawState *s = bs->opaque;
1693
1694     if (s->fd < 0)
1695         return;
1696
1697     (void) ioctl(s->fd, CDIOCALLOW);
1698
1699     if (eject_flag) {
1700         if (ioctl(s->fd, CDIOCEJECT) < 0)
1701             perror("CDIOCEJECT");
1702     } else {
1703         if (ioctl(s->fd, CDIOCCLOSE) < 0)
1704             perror("CDIOCCLOSE");
1705     }
1706
1707     cdrom_reopen(bs);
1708 }
1709
1710 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1711 {
1712     BDRVRawState *s = bs->opaque;
1713
1714     if (s->fd < 0)
1715         return;
1716     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1717         /*
1718          * Note: an error can happen if the distribution automatically
1719          * mounts the CD-ROM
1720          */
1721         /* perror("CDROM_LOCKDOOR"); */
1722     }
1723 }
1724
1725 static BlockDriver bdrv_host_cdrom = {
1726     .format_name        = "host_cdrom",
1727     .protocol_name      = "host_cdrom",
1728     .instance_size      = sizeof(BDRVRawState),
1729     .bdrv_probe_device  = cdrom_probe_device,
1730     .bdrv_file_open     = cdrom_open,
1731     .bdrv_close         = raw_close,
1732     .bdrv_reopen_prepare = raw_reopen_prepare,
1733     .bdrv_reopen_commit  = raw_reopen_commit,
1734     .bdrv_reopen_abort   = raw_reopen_abort,
1735     .bdrv_create        = hdev_create,
1736     .create_options     = raw_create_options,
1737     .bdrv_has_zero_init = hdev_has_zero_init,
1738
1739     .bdrv_aio_readv     = raw_aio_readv,
1740     .bdrv_aio_writev    = raw_aio_writev,
1741     .bdrv_aio_flush     = raw_aio_flush,
1742
1743     .bdrv_truncate      = raw_truncate,
1744     .bdrv_getlength     = raw_getlength,
1745     .bdrv_get_allocated_file_size
1746                         = raw_get_allocated_file_size,
1747
1748     /* removable device support */
1749     .bdrv_is_inserted   = cdrom_is_inserted,
1750     .bdrv_eject         = cdrom_eject,
1751     .bdrv_lock_medium   = cdrom_lock_medium,
1752 };
1753 #endif /* __FreeBSD__ */
1754
1755 #ifdef CONFIG_LINUX_AIO
1756 /**
1757  * Return the file descriptor for Linux AIO
1758  *
1759  * This function is a layering violation and should be removed when it becomes
1760  * possible to call the block layer outside the global mutex.  It allows the
1761  * caller to hijack the file descriptor so I/O can be performed outside the
1762  * block layer.
1763  */
1764 int raw_get_aio_fd(BlockDriverState *bs)
1765 {
1766     BDRVRawState *s;
1767
1768     if (!bs->drv) {
1769         return -ENOMEDIUM;
1770     }
1771
1772     if (bs->drv == bdrv_find_format("raw")) {
1773         bs = bs->file;
1774     }
1775
1776     /* raw-posix has several protocols so just check for raw_aio_readv */
1777     if (bs->drv->bdrv_aio_readv != raw_aio_readv) {
1778         return -ENOTSUP;
1779     }
1780
1781     s = bs->opaque;
1782     if (!s->use_aio) {
1783         return -ENOTSUP;
1784     }
1785     return s->fd;
1786 }
1787 #endif /* CONFIG_LINUX_AIO */
1788
1789 static void bdrv_file_init(void)
1790 {
1791     /*
1792      * Register all the drivers.  Note that order is important, the driver
1793      * registered last will get probed first.
1794      */
1795     bdrv_register(&bdrv_file);
1796     bdrv_register(&bdrv_host_device);
1797 #ifdef __linux__
1798     bdrv_register(&bdrv_host_floppy);
1799     bdrv_register(&bdrv_host_cdrom);
1800 #endif
1801 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1802     bdrv_register(&bdrv_host_cdrom);
1803 #endif
1804 }
1805
1806 block_init(bdrv_file_init);