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 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
  63 #include <linux/falloc.h>
  64 #endif
  65 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
  66 #include <sys/disk.h>
  67 #include <sys/cdio.h>
  68 #endif
  69
  70 #ifdef __OpenBSD__
  71 #include <sys/ioctl.h>
  72 #include <sys/disklabel.h>
  73 #include <sys/dkio.h>
  74 #endif
  75
  76 #ifdef __NetBSD__
  77 #include <sys/ioctl.h>
  78 #include <sys/disklabel.h>
  79 #include <sys/dkio.h>
  80 #include <sys/disk.h>
  81 #endif
  82
  83 #ifdef __DragonFly__
  84 #include <sys/ioctl.h>
  85 #include <sys/diskslice.h>
  86 #endif
  87
  88 #ifdef CONFIG_XFS
  89 #include <xfs/xfs.h>
  90 #endif
  91
  92 //#define DEBUG_FLOPPY
  93
  94 //#define DEBUG_BLOCK
  95 #if defined(DEBUG_BLOCK)
  96 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
  97     { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
  98 #else
  99 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
 100 #endif
 101
 102 /* OS X does not have O_DSYNC */
 103 #ifndef O_DSYNC
 104 #ifdef O_SYNC
 105 #define O_DSYNC O_SYNC
 106 #elif defined(O_FSYNC)
 107 #define O_DSYNC O_FSYNC
 108 #endif
 109 #endif
 110
 111 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
 112 #ifndef O_DIRECT
 113 #define O_DIRECT O_DSYNC
 114 #endif
 115
 116 #define FTYPE_FILE   0
 117 #define FTYPE_CD     1
 118 #define FTYPE_FD     2
 119
 120 /* if the FD is not accessed during that time (in ns), we try to
 121    reopen it to see if the disk has been changed */
 122 #define FD_OPEN_TIMEOUT (1000000000)
 123
 124 #define MAX_BLOCKSIZE   4096
 125
 126 typedef struct BDRVRawState {
 127     int fd;
 128     int type;
 129     int open_flags;
 130 #if defined(__linux__)
 131     /* linux floppy specific */
 132     int64_t fd_open_time;
 133     int64_t fd_error_time;
 134     int fd_got_error;
 135     int fd_media_changed;
 136 #endif
 137 #ifdef CONFIG_LINUX_AIO
 138     int use_aio;
 139     void *aio_ctx;
 140 #endif
 141 #ifdef CONFIG_XFS
 142     bool is_xfs : 1;
 143 #endif
 144     bool has_discard : 1;
 145 } BDRVRawState;
 146
 147 typedef struct BDRVRawReopenState {
 148     int fd;
 149     int open_flags;
 150 #ifdef CONFIG_LINUX_AIO
 151     int use_aio;
 152 #endif
 153 } BDRVRawReopenState;
 154
 155 static int fd_open(BlockDriverState *bs);
 156 static int64_t raw_getlength(BlockDriverState *bs);
 157
 158 typedef struct RawPosixAIOData {
 159     BlockDriverState *bs;
 160     int aio_fildes;
 161     union {
 162         struct iovec *aio_iov;
 163         void *aio_ioctl_buf;
 164     };
 165     int aio_niov;
 166     uint64_t aio_nbytes;
 167 #define aio_ioctl_cmd   aio_nbytes /* for QEMU_AIO_IOCTL */
 168     off_t aio_offset;
 169     int aio_type;
 170 } RawPosixAIOData;
 171
 172 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
 173 static int cdrom_reopen(BlockDriverState *bs);
 174 #endif
 175
 176 #if defined(__NetBSD__)
 177 static int raw_normalize_devicepath(const char **filename)
 178 {
 179     static char namebuf[PATH_MAX];
 180     const char *dp, *fname;
 181     struct stat sb;
 182
 183     fname = *filename;
 184     dp = strrchr(fname, '/');
 185     if (lstat(fname, &sb) < 0) {
 186         fprintf(stderr, "%s: stat failed: %s\n",
 187             fname, strerror(errno));
 188         return -errno;
 189     }
 190
 191     if (!S_ISBLK(sb.st_mode)) {
 192         return 0;
 193     }
 194
 195     if (dp == NULL) {
 196         snprintf(namebuf, PATH_MAX, "r%s", fname);
 197     } else {
 198         snprintf(namebuf, PATH_MAX, "%.*s/r%s",
 199             (int)(dp - fname), fname, dp + 1);
 200     }
 201     fprintf(stderr, "%s is a block device", fname);
 202     *filename = namebuf;
 203     fprintf(stderr, ", using %s\n", *filename);
 204
 205     return 0;
 206 }
 207 #else
 208 static int raw_normalize_devicepath(const char **filename)
 209 {
 210     return 0;
 211 }
 212 #endif
 213
 214 static void raw_parse_flags(int bdrv_flags, int *open_flags)
 215 {
 216     assert(open_flags != NULL);
 217
 218     *open_flags |= O_BINARY;
 219     *open_flags &= ~O_ACCMODE;
 220     if (bdrv_flags & BDRV_O_RDWR) {
 221         *open_flags |= O_RDWR;
 222     } else {
 223         *open_flags |= O_RDONLY;
 224     }
 225
 226     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 227      * and O_DIRECT for no caching. */
 228     if ((bdrv_flags & BDRV_O_NOCACHE)) {
 229         *open_flags |= O_DIRECT;
 230     }
 231 }
 232
 233 #ifdef CONFIG_LINUX_AIO
 234 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
 235 {
 236     int ret = -1;
 237     assert(aio_ctx != NULL);
 238     assert(use_aio != NULL);
 239     /*
 240      * Currently Linux do AIO only for files opened with O_DIRECT
 241      * specified so check NOCACHE flag too
 242      */
 243     if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
 244                       (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
 245
 246         /* if non-NULL, laio_init() has already been run */
 247         if (*aio_ctx == NULL) {
 248             *aio_ctx = laio_init();
 249             if (!*aio_ctx) {
 250                 goto error;
 251             }
 252         }
 253         *use_aio = 1;
 254     } else {
 255         *use_aio = 0;
 256     }
 257
 258     ret = 0;
 259
 260 error:
 261     return ret;
 262 }
 263 #endif
 264
 265 static QemuOptsList raw_runtime_opts = {
 266     .name = "raw",
 267     .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
 268     .desc = {
 269         {
 270             .name = "filename",
 271             .type = QEMU_OPT_STRING,
 272             .help = "File name of the image",
 273         },
 274         { /* end of list */ }
 275     },
 276 };
 277
 278 static int raw_open_common(BlockDriverState *bs, QDict *options,
 279                            int bdrv_flags, int open_flags)
 280 {
 281     BDRVRawState *s = bs->opaque;
 282     QemuOpts *opts;
 283     Error *local_err = NULL;
 284     const char *filename;
 285     int fd, ret;
 286
 287     opts = qemu_opts_create_nofail(&raw_runtime_opts);
 288     qemu_opts_absorb_qdict(opts, options, &local_err);
 289     if (error_is_set(&local_err)) {
 290         qerror_report_err(local_err);
 291         error_free(local_err);
 292         ret = -EINVAL;
 293         goto fail;
 294     }
 295
 296     filename = qemu_opt_get(opts, "filename");
 297
 298     ret = raw_normalize_devicepath(&filename);
 299     if (ret != 0) {
 300         goto fail;
 301     }
 302
 303     s->open_flags = open_flags;
 304     raw_parse_flags(bdrv_flags, &s->open_flags);
 305
 306     s->fd = -1;
 307     fd = qemu_open(filename, s->open_flags, 0644);
 308     if (fd < 0) {
 309         ret = -errno;
 310         if (ret == -EROFS) {
 311             ret = -EACCES;
 312         }
 313         goto fail;
 314     }
 315     s->fd = fd;
 316
 317 #ifdef CONFIG_LINUX_AIO
 318     if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
 319         qemu_close(fd);
 320         ret = -errno;
 321         goto fail;
 322     }
 323 #endif
 324
 325     s->has_discard = 1;
 326 #ifdef CONFIG_XFS
 327     if (platform_test_xfs_fd(s->fd)) {
 328         s->is_xfs = 1;
 329     }
 330 #endif
 331
 332     ret = 0;
 333 fail:
 334     qemu_opts_del(opts);
 335     return ret;
 336 }
 337
 338 static int raw_open(BlockDriverState *bs, QDict *options, int flags)
 339 {
 340     BDRVRawState *s = bs->opaque;
 341
 342     s->type = FTYPE_FILE;
 343     return raw_open_common(bs, options, flags, 0);
 344 }
 345
 346 static int raw_reopen_prepare(BDRVReopenState *state,
 347                               BlockReopenQueue *queue, Error **errp)
 348 {
 349     BDRVRawState *s;
 350     BDRVRawReopenState *raw_s;
 351     int ret = 0;
 352
 353     assert(state != NULL);
 354     assert(state->bs != NULL);
 355
 356     s = state->bs->opaque;
 357
 358     state->opaque = g_malloc0(sizeof(BDRVRawReopenState));
 359     raw_s = state->opaque;
 360
 361 #ifdef CONFIG_LINUX_AIO
 362     raw_s->use_aio = s->use_aio;
 363
 364     /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
 365      * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
 366      * won't override aio_ctx if aio_ctx is non-NULL */
 367     if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
 368         return -1;
 369     }
 370 #endif
 371
 372     if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
 373         raw_s->open_flags |= O_NONBLOCK;
 374     }
 375
 376     raw_parse_flags(state->flags, &raw_s->open_flags);
 377
 378     raw_s->fd = -1;
 379
 380     int fcntl_flags = O_APPEND | O_NONBLOCK;
 381 #ifdef O_NOATIME
 382     fcntl_flags |= O_NOATIME;
 383 #endif
 384
 385 #ifdef O_ASYNC
 386     /* Not all operating systems have O_ASYNC, and those that don't
 387      * will not let us track the state into raw_s->open_flags (typically
 388      * you achieve the same effect with an ioctl, for example I_SETSIG
 389      * on Solaris). But we do not use O_ASYNC, so that's fine.
 390      */
 391     assert((s->open_flags & O_ASYNC) == 0);
 392 #endif
 393
 394     if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
 395         /* dup the original fd */
 396         /* TODO: use qemu fcntl wrapper */
 397 #ifdef F_DUPFD_CLOEXEC
 398         raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
 399 #else
 400         raw_s->fd = dup(s->fd);
 401         if (raw_s->fd != -1) {
 402             qemu_set_cloexec(raw_s->fd);
 403         }
 404 #endif
 405         if (raw_s->fd >= 0) {
 406             ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
 407             if (ret) {
 408                 qemu_close(raw_s->fd);
 409                 raw_s->fd = -1;
 410             }
 411         }
 412     }
 413
 414     /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
 415     if (raw_s->fd == -1) {
 416         assert(!(raw_s->open_flags & O_CREAT));
 417         raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
 418         if (raw_s->fd == -1) {
 419             ret = -1;
 420         }
 421     }
 422     return ret;
 423 }
 424
 425
 426 static void raw_reopen_commit(BDRVReopenState *state)
 427 {
 428     BDRVRawReopenState *raw_s = state->opaque;
 429     BDRVRawState *s = state->bs->opaque;
 430
 431     s->open_flags = raw_s->open_flags;
 432
 433     qemu_close(s->fd);
 434     s->fd = raw_s->fd;
 435 #ifdef CONFIG_LINUX_AIO
 436     s->use_aio = raw_s->use_aio;
 437 #endif
 438
 439     g_free(state->opaque);
 440     state->opaque = NULL;
 441 }
 442
 443
 444 static void raw_reopen_abort(BDRVReopenState *state)
 445 {
 446     BDRVRawReopenState *raw_s = state->opaque;
 447
 448      /* nothing to do if NULL, we didn't get far enough */
 449     if (raw_s == NULL) {
 450         return;
 451     }
 452
 453     if (raw_s->fd >= 0) {
 454         qemu_close(raw_s->fd);
 455         raw_s->fd = -1;
 456     }
 457     g_free(state->opaque);
 458     state->opaque = NULL;
 459 }
 460
 461
 462 /* XXX: use host sector size if necessary with:
 463 #ifdef DIOCGSECTORSIZE
 464         {
 465             unsigned int sectorsize = 512;
 466             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 467                 sectorsize > bufsize)
 468                 bufsize = sectorsize;
 469         }
 470 #endif
 471 #ifdef CONFIG_COCOA
 472         uint32_t blockSize = 512;
 473         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 474             bufsize = blockSize;
 475         }
 476 #endif
 477 */
 478
 479 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
 480 {
 481     int ret;
 482
 483     ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
 484     if (ret == -1) {
 485         return -errno;
 486     }
 487
 488     return 0;
 489 }
 490
 491 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
 492 {
 493     int ret;
 494
 495     ret = qemu_fdatasync(aiocb->aio_fildes);
 496     if (ret == -1) {
 497         return -errno;
 498     }
 499     return 0;
 500 }
 501
 502 #ifdef CONFIG_PREADV
 503
 504 static bool preadv_present = true;
 505
 506 static ssize_t
 507 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 508 {
 509     return preadv(fd, iov, nr_iov, offset);
 510 }
 511
 512 static ssize_t
 513 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 514 {
 515     return pwritev(fd, iov, nr_iov, offset);
 516 }
 517
 518 #else
 519
 520 static bool preadv_present = false;
 521
 522 static ssize_t
 523 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 524 {
 525     return -ENOSYS;
 526 }
 527
 528 static ssize_t
 529 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 530 {
 531     return -ENOSYS;
 532 }
 533
 534 #endif
 535
 536 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
 537 {
 538     ssize_t len;
 539
 540     do {
 541         if (aiocb->aio_type & QEMU_AIO_WRITE)
 542             len = qemu_pwritev(aiocb->aio_fildes,
 543                                aiocb->aio_iov,
 544                                aiocb->aio_niov,
 545                                aiocb->aio_offset);
 546          else
 547             len = qemu_preadv(aiocb->aio_fildes,
 548                               aiocb->aio_iov,
 549                               aiocb->aio_niov,
 550                               aiocb->aio_offset);
 551     } while (len == -1 && errno == EINTR);
 552
 553     if (len == -1) {
 554         return -errno;
 555     }
 556     return len;
 557 }
 558
 559 /*
 560  * Read/writes the data to/from a given linear buffer.
 561  *
 562  * Returns the number of bytes handles or -errno in case of an error. Short
 563  * reads are only returned if the end of the file is reached.
 564  */
 565 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
 566 {
 567     ssize_t offset = 0;
 568     ssize_t len;
 569
 570     while (offset < aiocb->aio_nbytes) {
 571         if (aiocb->aio_type & QEMU_AIO_WRITE) {
 572             len = pwrite(aiocb->aio_fildes,
 573                          (const char *)buf + offset,
 574                          aiocb->aio_nbytes - offset,
 575                          aiocb->aio_offset + offset);
 576         } else {
 577             len = pread(aiocb->aio_fildes,
 578                         buf + offset,
 579                         aiocb->aio_nbytes - offset,
 580                         aiocb->aio_offset + offset);
 581         }
 582         if (len == -1 && errno == EINTR) {
 583             continue;
 584         } else if (len == -1) {
 585             offset = -errno;
 586             break;
 587         } else if (len == 0) {
 588             break;
 589         }
 590         offset += len;
 591     }
 592
 593     return offset;
 594 }
 595
 596 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
 597 {
 598     ssize_t nbytes;
 599     char *buf;
 600
 601     if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
 602         /*
 603          * If there is just a single buffer, and it is properly aligned
 604          * we can just use plain pread/pwrite without any problems.
 605          */
 606         if (aiocb->aio_niov == 1) {
 607              return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
 608         }
 609         /*
 610          * We have more than one iovec, and all are properly aligned.
 611          *
 612          * Try preadv/pwritev first and fall back to linearizing the
 613          * buffer if it's not supported.
 614          */
 615         if (preadv_present) {
 616             nbytes = handle_aiocb_rw_vector(aiocb);
 617             if (nbytes == aiocb->aio_nbytes ||
 618                 (nbytes < 0 && nbytes != -ENOSYS)) {
 619                 return nbytes;
 620             }
 621             preadv_present = false;
 622         }
 623
 624         /*
 625          * XXX(hch): short read/write.  no easy way to handle the reminder
 626          * using these interfaces.  For now retry using plain
 627          * pread/pwrite?
 628          */
 629     }
 630
 631     /*
 632      * Ok, we have to do it the hard way, copy all segments into
 633      * a single aligned buffer.
 634      */
 635     buf = qemu_blockalign(aiocb->bs, aiocb->aio_nbytes);
 636     if (aiocb->aio_type & QEMU_AIO_WRITE) {
 637         char *p = buf;
 638         int i;
 639
 640         for (i = 0; i < aiocb->aio_niov; ++i) {
 641             memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
 642             p += aiocb->aio_iov[i].iov_len;
 643         }
 644     }
 645
 646     nbytes = handle_aiocb_rw_linear(aiocb, buf);
 647     if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
 648         char *p = buf;
 649         size_t count = aiocb->aio_nbytes, copy;
 650         int i;
 651
 652         for (i = 0; i < aiocb->aio_niov && count; ++i) {
 653             copy = count;
 654             if (copy > aiocb->aio_iov[i].iov_len) {
 655                 copy = aiocb->aio_iov[i].iov_len;
 656             }
 657             memcpy(aiocb->aio_iov[i].iov_base, p, copy);
 658             p     += copy;
 659             count -= copy;
 660         }
 661     }
 662     qemu_vfree(buf);
 663
 664     return nbytes;
 665 }
 666
 667 #ifdef CONFIG_XFS
 668 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
 669 {
 670     struct xfs_flock64 fl;
 671
 672     memset(&fl, 0, sizeof(fl));
 673     fl.l_whence = SEEK_SET;
 674     fl.l_start = offset;
 675     fl.l_len = bytes;
 676
 677     if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
 678         DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
 679         return -errno;
 680     }
 681
 682     return 0;
 683 }
 684 #endif
 685
 686 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
 687 {
 688     int ret = -EOPNOTSUPP;
 689     BDRVRawState *s = aiocb->bs->opaque;
 690
 691     if (s->has_discard == 0) {
 692         return 0;
 693     }
 694
 695     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
 696 #ifdef BLKDISCARD
 697         do {
 698             uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
 699             if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
 700                 return 0;
 701             }
 702         } while (errno == EINTR);
 703
 704         ret = -errno;
 705 #endif
 706     } else {
 707 #ifdef CONFIG_XFS
 708         if (s->is_xfs) {
 709             return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
 710         }
 711 #endif
 712
 713 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
 714         do {
 715             if (fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
 716                           aiocb->aio_offset, aiocb->aio_nbytes) == 0) {
 717                 return 0;
 718             }
 719         } while (errno == EINTR);
 720
 721         ret = -errno;
 722 #endif
 723     }
 724
 725     if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
 726         ret == -ENOTTY) {
 727         s->has_discard = 0;
 728         ret = 0;
 729     }
 730     return ret;
 731 }
 732
 733 static int aio_worker(void *arg)
 734 {
 735     RawPosixAIOData *aiocb = arg;
 736     ssize_t ret = 0;
 737
 738     switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
 739     case QEMU_AIO_READ:
 740         ret = handle_aiocb_rw(aiocb);
 741         if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) {
 742             iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
 743                       0, aiocb->aio_nbytes - ret);
 744
 745             ret = aiocb->aio_nbytes;
 746         }
 747         if (ret == aiocb->aio_nbytes) {
 748             ret = 0;
 749         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 750             ret = -EINVAL;
 751         }
 752         break;
 753     case QEMU_AIO_WRITE:
 754         ret = handle_aiocb_rw(aiocb);
 755         if (ret == aiocb->aio_nbytes) {
 756             ret = 0;
 757         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 758             ret = -EINVAL;
 759         }
 760         break;
 761     case QEMU_AIO_FLUSH:
 762         ret = handle_aiocb_flush(aiocb);
 763         break;
 764     case QEMU_AIO_IOCTL:
 765         ret = handle_aiocb_ioctl(aiocb);
 766         break;
 767     case QEMU_AIO_DISCARD:
 768         ret = handle_aiocb_discard(aiocb);
 769         break;
 770     default:
 771         fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
 772         ret = -EINVAL;
 773         break;
 774     }
 775
 776     g_slice_free(RawPosixAIOData, aiocb);
 777     return ret;
 778 }
 779
 780 static BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd,
 781         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 782         BlockDriverCompletionFunc *cb, void *opaque, int type)
 783 {
 784     RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
 785     ThreadPool *pool;
 786
 787     acb->bs = bs;
 788     acb->aio_type = type;
 789     acb->aio_fildes = fd;
 790
 791     if (qiov) {
 792         acb->aio_iov = qiov->iov;
 793         acb->aio_niov = qiov->niov;
 794     }
 795     acb->aio_nbytes = nb_sectors * 512;
 796     acb->aio_offset = sector_num * 512;
 797
 798     trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
 799     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
 800     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
 801 }
 802
 803 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
 804         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 805         BlockDriverCompletionFunc *cb, void *opaque, int type)
 806 {
 807     BDRVRawState *s = bs->opaque;
 808
 809     if (fd_open(bs) < 0)
 810         return NULL;
 811
 812     /*
 813      * If O_DIRECT is used the buffer needs to be aligned on a sector
 814      * boundary.  Check if this is the case or tell the low-level
 815      * driver that it needs to copy the buffer.
 816      */
 817     if ((bs->open_flags & BDRV_O_NOCACHE)) {
 818         if (!bdrv_qiov_is_aligned(bs, qiov)) {
 819             type |= QEMU_AIO_MISALIGNED;
 820 #ifdef CONFIG_LINUX_AIO
 821         } else if (s->use_aio) {
 822             return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
 823                                nb_sectors, cb, opaque, type);
 824 #endif
 825         }
 826     }
 827
 828     return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
 829                        cb, opaque, type);
 830 }
 831
 832 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
 833         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 834         BlockDriverCompletionFunc *cb, void *opaque)
 835 {
 836     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 837                           cb, opaque, QEMU_AIO_READ);
 838 }
 839
 840 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
 841         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 842         BlockDriverCompletionFunc *cb, void *opaque)
 843 {
 844     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 845                           cb, opaque, QEMU_AIO_WRITE);
 846 }
 847
 848 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
 849         BlockDriverCompletionFunc *cb, void *opaque)
 850 {
 851     BDRVRawState *s = bs->opaque;
 852
 853     if (fd_open(bs) < 0)
 854         return NULL;
 855
 856     return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
 857 }
 858
 859 static void raw_close(BlockDriverState *bs)
 860 {
 861     BDRVRawState *s = bs->opaque;
 862     if (s->fd >= 0) {
 863         qemu_close(s->fd);
 864         s->fd = -1;
 865     }
 866 }
 867
 868 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 869 {
 870     BDRVRawState *s = bs->opaque;
 871     struct stat st;
 872
 873     if (fstat(s->fd, &st)) {
 874         return -errno;
 875     }
 876
 877     if (S_ISREG(st.st_mode)) {
 878         if (ftruncate(s->fd, offset) < 0) {
 879             return -errno;
 880         }
 881     } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 882        if (offset > raw_getlength(bs)) {
 883            return -EINVAL;
 884        }
 885     } else {
 886         return -ENOTSUP;
 887     }
 888
 889     return 0;
 890 }
 891
 892 #ifdef __OpenBSD__
 893 static int64_t raw_getlength(BlockDriverState *bs)
 894 {
 895     BDRVRawState *s = bs->opaque;
 896     int fd = s->fd;
 897     struct stat st;
 898
 899     if (fstat(fd, &st))
 900         return -1;
 901     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 902         struct disklabel dl;
 903
 904         if (ioctl(fd, DIOCGDINFO, &dl))
 905             return -1;
 906         return (uint64_t)dl.d_secsize *
 907             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 908     } else
 909         return st.st_size;
 910 }
 911 #elif defined(__NetBSD__)
 912 static int64_t raw_getlength(BlockDriverState *bs)
 913 {
 914     BDRVRawState *s = bs->opaque;
 915     int fd = s->fd;
 916     struct stat st;
 917
 918     if (fstat(fd, &st))
 919         return -1;
 920     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 921         struct dkwedge_info dkw;
 922
 923         if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
 924             return dkw.dkw_size * 512;
 925         } else {
 926             struct disklabel dl;
 927
 928             if (ioctl(fd, DIOCGDINFO, &dl))
 929                 return -1;
 930             return (uint64_t)dl.d_secsize *
 931                 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 932         }
 933     } else
 934         return st.st_size;
 935 }
 936 #elif defined(__sun__)
 937 static int64_t raw_getlength(BlockDriverState *bs)
 938 {
 939     BDRVRawState *s = bs->opaque;
 940     struct dk_minfo minfo;
 941     int ret;
 942
 943     ret = fd_open(bs);
 944     if (ret < 0) {
 945         return ret;
 946     }
 947
 948     /*
 949      * Use the DKIOCGMEDIAINFO ioctl to read the size.
 950      */
 951     ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
 952     if (ret != -1) {
 953         return minfo.dki_lbsize * minfo.dki_capacity;
 954     }
 955
 956     /*
 957      * There are reports that lseek on some devices fails, but
 958      * irc discussion said that contingency on contingency was overkill.
 959      */
 960     return lseek(s->fd, 0, SEEK_END);
 961 }
 962 #elif defined(CONFIG_BSD)
 963 static int64_t raw_getlength(BlockDriverState *bs)
 964 {
 965     BDRVRawState *s = bs->opaque;
 966     int fd = s->fd;
 967     int64_t size;
 968     struct stat sb;
 969 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 970     int reopened = 0;
 971 #endif
 972     int ret;
 973
 974     ret = fd_open(bs);
 975     if (ret < 0)
 976         return ret;
 977
 978 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 979 again:
 980 #endif
 981     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 982 #ifdef DIOCGMEDIASIZE
 983         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 984 #elif defined(DIOCGPART)
 985         {
 986                 struct partinfo pi;
 987                 if (ioctl(fd, DIOCGPART, &pi) == 0)
 988                         size = pi.media_size;
 989                 else
 990                         size = 0;
 991         }
 992         if (size == 0)
 993 #endif
 994 #if defined(__APPLE__) && defined(__MACH__)
 995         size = LONG_LONG_MAX;
 996 #else
 997         size = lseek(fd, 0LL, SEEK_END);
 998 #endif
 999 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1000         switch(s->type) {
1001         case FTYPE_CD:
1002             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1003             if (size == 2048LL * (unsigned)-1)
1004                 size = 0;
1005             /* XXX no disc?  maybe we need to reopen... */
1006             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1007                 reopened = 1;
1008                 goto again;
1009             }
1010         }
1011 #endif
1012     } else {
1013         size = lseek(fd, 0, SEEK_END);
1014     }
1015     return size;
1016 }
1017 #else
1018 static int64_t raw_getlength(BlockDriverState *bs)
1019 {
1020     BDRVRawState *s = bs->opaque;
1021     int ret;
1022
1023     ret = fd_open(bs);
1024     if (ret < 0) {
1025         return ret;
1026     }
1027
1028     return lseek(s->fd, 0, SEEK_END);
1029 }
1030 #endif
1031
1032 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1033 {
1034     struct stat st;
1035     BDRVRawState *s = bs->opaque;
1036
1037     if (fstat(s->fd, &st) < 0) {
1038         return -errno;
1039     }
1040     return (int64_t)st.st_blocks * 512;
1041 }
1042
1043 static int raw_create(const char *filename, QEMUOptionParameter *options)
1044 {
1045     int fd;
1046     int result = 0;
1047     int64_t total_size = 0;
1048
1049     /* Read out options */
1050     while (options && options->name) {
1051         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1052             total_size = options->value.n / BDRV_SECTOR_SIZE;
1053         }
1054         options++;
1055     }
1056
1057     fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
1058                    0644);
1059     if (fd < 0) {
1060         result = -errno;
1061     } else {
1062         if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
1063             result = -errno;
1064         }
1065         if (qemu_close(fd) != 0) {
1066             result = -errno;
1067         }
1068     }
1069     return result;
1070 }
1071
1072 /*
1073  * Returns true iff the specified sector is present in the disk image. Drivers
1074  * not implementing the functionality are assumed to not support backing files,
1075  * hence all their sectors are reported as allocated.
1076  *
1077  * If 'sector_num' is beyond the end of the disk image the return value is 0
1078  * and 'pnum' is set to 0.
1079  *
1080  * 'pnum' is set to the number of sectors (including and immediately following
1081  * the specified sector) that are known to be in the same
1082  * allocated/unallocated state.
1083  *
1084  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
1085  * beyond the end of the disk image it will be clamped.
1086  */
1087 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1088                                             int64_t sector_num,
1089                                             int nb_sectors, int *pnum)
1090 {
1091     off_t start, data, hole;
1092     int64_t ret;
1093
1094     ret = fd_open(bs);
1095     if (ret < 0) {
1096         return ret;
1097     }
1098
1099     start = sector_num * BDRV_SECTOR_SIZE;
1100     ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1101
1102 #ifdef CONFIG_FIEMAP
1103
1104     BDRVRawState *s = bs->opaque;
1105     struct {
1106         struct fiemap fm;
1107         struct fiemap_extent fe;
1108     } f;
1109
1110     f.fm.fm_start = start;
1111     f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE;
1112     f.fm.fm_flags = 0;
1113     f.fm.fm_extent_count = 1;
1114     f.fm.fm_reserved = 0;
1115     if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) {
1116         /* Assume everything is allocated.  */
1117         *pnum = nb_sectors;
1118         return ret;
1119     }
1120
1121     if (f.fm.fm_mapped_extents == 0) {
1122         /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1123          * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1124          */
1125         off_t length = lseek(s->fd, 0, SEEK_END);
1126         hole = f.fm.fm_start;
1127         data = MIN(f.fm.fm_start + f.fm.fm_length, length);
1128     } else {
1129         data = f.fe.fe_logical;
1130         hole = f.fe.fe_logical + f.fe.fe_length;
1131         if (f.fe.fe_flags & FIEMAP_EXTENT_UNWRITTEN) {
1132             ret |= BDRV_BLOCK_ZERO;
1133         }
1134     }
1135
1136 #elif defined SEEK_HOLE && defined SEEK_DATA
1137
1138     BDRVRawState *s = bs->opaque;
1139
1140     hole = lseek(s->fd, start, SEEK_HOLE);
1141     if (hole == -1) {
1142         /* -ENXIO indicates that sector_num was past the end of the file.
1143          * There is a virtual hole there.  */
1144         assert(errno != -ENXIO);
1145
1146         /* Most likely EINVAL.  Assume everything is allocated.  */
1147         *pnum = nb_sectors;
1148         return ret;
1149     }
1150
1151     if (hole > start) {
1152         data = start;
1153     } else {
1154         /* On a hole.  We need another syscall to find its end.  */
1155         data = lseek(s->fd, start, SEEK_DATA);
1156         if (data == -1) {
1157             data = lseek(s->fd, 0, SEEK_END);
1158         }
1159     }
1160 #else
1161     data = 0;
1162     hole = start + nb_sectors * BDRV_SECTOR_SIZE;
1163 #endif
1164
1165     if (data <= start) {
1166         /* On a data extent, compute sectors to the end of the extent.  */
1167         *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1168     } else {
1169         /* On a hole, compute sectors to the beginning of the next extent.  */
1170         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1171         ret &= ~BDRV_BLOCK_DATA;
1172         ret |= BDRV_BLOCK_ZERO;
1173     }
1174
1175     return ret;
1176 }
1177
1178 static coroutine_fn BlockDriverAIOCB *raw_aio_discard(BlockDriverState *bs,
1179     int64_t sector_num, int nb_sectors,
1180     BlockDriverCompletionFunc *cb, void *opaque)
1181 {
1182     BDRVRawState *s = bs->opaque;
1183
1184     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1185                        cb, opaque, QEMU_AIO_DISCARD);
1186 }
1187
1188 static QEMUOptionParameter raw_create_options[] = {
1189     {
1190         .name = BLOCK_OPT_SIZE,
1191         .type = OPT_SIZE,
1192         .help = "Virtual disk size"
1193     },
1194     { NULL }
1195 };
1196
1197 static BlockDriver bdrv_file = {
1198     .format_name = "file",
1199     .protocol_name = "file",
1200     .instance_size = sizeof(BDRVRawState),
1201     .bdrv_probe = NULL, /* no probe for protocols */
1202     .bdrv_file_open = raw_open,
1203     .bdrv_reopen_prepare = raw_reopen_prepare,
1204     .bdrv_reopen_commit = raw_reopen_commit,
1205     .bdrv_reopen_abort = raw_reopen_abort,
1206     .bdrv_close = raw_close,
1207     .bdrv_create = raw_create,
1208     .bdrv_has_zero_init = bdrv_has_zero_init_1,
1209     .bdrv_co_get_block_status = raw_co_get_block_status,
1210
1211     .bdrv_aio_readv = raw_aio_readv,
1212     .bdrv_aio_writev = raw_aio_writev,
1213     .bdrv_aio_flush = raw_aio_flush,
1214     .bdrv_aio_discard = raw_aio_discard,
1215
1216     .bdrv_truncate = raw_truncate,
1217     .bdrv_getlength = raw_getlength,
1218     .bdrv_get_allocated_file_size
1219                         = raw_get_allocated_file_size,
1220
1221     .create_options = raw_create_options,
1222 };
1223
1224 /***********************************************/
1225 /* host device */
1226
1227 #if defined(__APPLE__) && defined(__MACH__)
1228 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1229 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1230
1231 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1232 {
1233     kern_return_t       kernResult;
1234     mach_port_t     masterPort;
1235     CFMutableDictionaryRef  classesToMatch;
1236
1237     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1238     if ( KERN_SUCCESS != kernResult ) {
1239         printf( "IOMasterPort returned %d\n", kernResult );
1240     }
1241
1242     classesToMatch = IOServiceMatching( kIOCDMediaClass );
1243     if ( classesToMatch == NULL ) {
1244         printf( "IOServiceMatching returned a NULL dictionary.\n" );
1245     } else {
1246     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1247     }
1248     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1249     if ( KERN_SUCCESS != kernResult )
1250     {
1251         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1252     }
1253
1254     return kernResult;
1255 }
1256
1257 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
1258 {
1259     io_object_t     nextMedia;
1260     kern_return_t   kernResult = KERN_FAILURE;
1261     *bsdPath = '\0';
1262     nextMedia = IOIteratorNext( mediaIterator );
1263     if ( nextMedia )
1264     {
1265         CFTypeRef   bsdPathAsCFString;
1266     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1267         if ( bsdPathAsCFString ) {
1268             size_t devPathLength;
1269             strcpy( bsdPath, _PATH_DEV );
1270             strcat( bsdPath, "r" );
1271             devPathLength = strlen( bsdPath );
1272             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1273                 kernResult = KERN_SUCCESS;
1274             }
1275             CFRelease( bsdPathAsCFString );
1276         }
1277         IOObjectRelease( nextMedia );
1278     }
1279
1280     return kernResult;
1281 }
1282
1283 #endif
1284
1285 static int hdev_probe_device(const char *filename)
1286 {
1287     struct stat st;
1288
1289     /* allow a dedicated CD-ROM driver to match with a higher priority */
1290     if (strstart(filename, "/dev/cdrom", NULL))
1291         return 50;
1292
1293     if (stat(filename, &st) >= 0 &&
1294             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
1295         return 100;
1296     }
1297
1298     return 0;
1299 }
1300
1301 static int check_hdev_writable(BDRVRawState *s)
1302 {
1303 #if defined(BLKROGET)
1304     /* Linux block devices can be configured "read-only" using blockdev(8).
1305      * This is independent of device node permissions and therefore open(2)
1306      * with O_RDWR succeeds.  Actual writes fail with EPERM.
1307      *
1308      * bdrv_open() is supposed to fail if the disk is read-only.  Explicitly
1309      * check for read-only block devices so that Linux block devices behave
1310      * properly.
1311      */
1312     struct stat st;
1313     int readonly = 0;
1314
1315     if (fstat(s->fd, &st)) {
1316         return -errno;
1317     }
1318
1319     if (!S_ISBLK(st.st_mode)) {
1320         return 0;
1321     }
1322
1323     if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
1324         return -errno;
1325     }
1326
1327     if (readonly) {
1328         return -EACCES;
1329     }
1330 #endif /* defined(BLKROGET) */
1331     return 0;
1332 }
1333
1334 static int hdev_open(BlockDriverState *bs, QDict *options, int flags)
1335 {
1336     BDRVRawState *s = bs->opaque;
1337     int ret;
1338     const char *filename = qdict_get_str(options, "filename");
1339
1340 #if defined(__APPLE__) && defined(__MACH__)
1341     if (strstart(filename, "/dev/cdrom", NULL)) {
1342         kern_return_t kernResult;
1343         io_iterator_t mediaIterator;
1344         char bsdPath[ MAXPATHLEN ];
1345         int fd;
1346
1347         kernResult = FindEjectableCDMedia( &mediaIterator );
1348         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1349
1350         if ( bsdPath[ 0 ] != '\0' ) {
1351             strcat(bsdPath,"s0");
1352             /* some CDs don't have a partition 0 */
1353             fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1354             if (fd < 0) {
1355                 bsdPath[strlen(bsdPath)-1] = '1';
1356             } else {
1357                 qemu_close(fd);
1358             }
1359             filename = bsdPath;
1360             qdict_put(options, "filename", qstring_from_str(filename));
1361         }
1362
1363         if ( mediaIterator )
1364             IOObjectRelease( mediaIterator );
1365     }
1366 #endif
1367
1368     s->type = FTYPE_FILE;
1369 #if defined(__linux__)
1370     {
1371         char resolved_path[ MAXPATHLEN ], *temp;
1372
1373         temp = realpath(filename, resolved_path);
1374         if (temp && strstart(temp, "/dev/sg", NULL)) {
1375             bs->sg = 1;
1376         }
1377     }
1378 #endif
1379
1380     ret = raw_open_common(bs, options, flags, 0);
1381     if (ret < 0) {
1382         return ret;
1383     }
1384
1385     if (flags & BDRV_O_RDWR) {
1386         ret = check_hdev_writable(s);
1387         if (ret < 0) {
1388             raw_close(bs);
1389             return ret;
1390         }
1391     }
1392
1393     return ret;
1394 }
1395
1396 #if defined(__linux__)
1397 /* Note: we do not have a reliable method to detect if the floppy is
1398    present. The current method is to try to open the floppy at every
1399    I/O and to keep it opened during a few hundreds of ms. */
1400 static int fd_open(BlockDriverState *bs)
1401 {
1402     BDRVRawState *s = bs->opaque;
1403     int last_media_present;
1404
1405     if (s->type != FTYPE_FD)
1406         return 0;
1407     last_media_present = (s->fd >= 0);
1408     if (s->fd >= 0 &&
1409         (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1410         qemu_close(s->fd);
1411         s->fd = -1;
1412 #ifdef DEBUG_FLOPPY
1413         printf("Floppy closed\n");
1414 #endif
1415     }
1416     if (s->fd < 0) {
1417         if (s->fd_got_error &&
1418             (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1419 #ifdef DEBUG_FLOPPY
1420             printf("No floppy (open delayed)\n");
1421 #endif
1422             return -EIO;
1423         }
1424         s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
1425         if (s->fd < 0) {
1426             s->fd_error_time = get_clock();
1427             s->fd_got_error = 1;
1428             if (last_media_present)
1429                 s->fd_media_changed = 1;
1430 #ifdef DEBUG_FLOPPY
1431             printf("No floppy\n");
1432 #endif
1433             return -EIO;
1434         }
1435 #ifdef DEBUG_FLOPPY
1436         printf("Floppy opened\n");
1437 #endif
1438     }
1439     if (!last_media_present)
1440         s->fd_media_changed = 1;
1441     s->fd_open_time = get_clock();
1442     s->fd_got_error = 0;
1443     return 0;
1444 }
1445
1446 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1447 {
1448     BDRVRawState *s = bs->opaque;
1449
1450     return ioctl(s->fd, req, buf);
1451 }
1452
1453 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1454         unsigned long int req, void *buf,
1455         BlockDriverCompletionFunc *cb, void *opaque)
1456 {
1457     BDRVRawState *s = bs->opaque;
1458     RawPosixAIOData *acb;
1459     ThreadPool *pool;
1460
1461     if (fd_open(bs) < 0)
1462         return NULL;
1463
1464     acb = g_slice_new(RawPosixAIOData);
1465     acb->bs = bs;
1466     acb->aio_type = QEMU_AIO_IOCTL;
1467     acb->aio_fildes = s->fd;
1468     acb->aio_offset = 0;
1469     acb->aio_ioctl_buf = buf;
1470     acb->aio_ioctl_cmd = req;
1471     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1472     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1473 }
1474
1475 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1476 static int fd_open(BlockDriverState *bs)
1477 {
1478     BDRVRawState *s = bs->opaque;
1479
1480     /* this is just to ensure s->fd is sane (its called by io ops) */
1481     if (s->fd >= 0)
1482         return 0;
1483     return -EIO;
1484 }
1485 #else /* !linux && !FreeBSD */
1486
1487 static int fd_open(BlockDriverState *bs)
1488 {
1489     return 0;
1490 }
1491
1492 #endif /* !linux && !FreeBSD */
1493
1494 static coroutine_fn BlockDriverAIOCB *hdev_aio_discard(BlockDriverState *bs,
1495     int64_t sector_num, int nb_sectors,
1496     BlockDriverCompletionFunc *cb, void *opaque)
1497 {
1498     BDRVRawState *s = bs->opaque;
1499
1500     if (fd_open(bs) < 0) {
1501         return NULL;
1502     }
1503     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1504                        cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1505 }
1506
1507 static int hdev_create(const char *filename, QEMUOptionParameter *options)
1508 {
1509     int fd;
1510     int ret = 0;
1511     struct stat stat_buf;
1512     int64_t total_size = 0;
1513
1514     /* Read out options */
1515     while (options && options->name) {
1516         if (!strcmp(options->name, "size")) {
1517             total_size = options->value.n / BDRV_SECTOR_SIZE;
1518         }
1519         options++;
1520     }
1521
1522     fd = qemu_open(filename, O_WRONLY | O_BINARY);
1523     if (fd < 0)
1524         return -errno;
1525
1526     if (fstat(fd, &stat_buf) < 0)
1527         ret = -errno;
1528     else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode))
1529         ret = -ENODEV;
1530     else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE)
1531         ret = -ENOSPC;
1532
1533     qemu_close(fd);
1534     return ret;
1535 }
1536
1537 static BlockDriver bdrv_host_device = {
1538     .format_name        = "host_device",
1539     .protocol_name        = "host_device",
1540     .instance_size      = sizeof(BDRVRawState),
1541     .bdrv_probe_device  = hdev_probe_device,
1542     .bdrv_file_open     = hdev_open,
1543     .bdrv_close         = raw_close,
1544     .bdrv_reopen_prepare = raw_reopen_prepare,
1545     .bdrv_reopen_commit  = raw_reopen_commit,
1546     .bdrv_reopen_abort   = raw_reopen_abort,
1547     .bdrv_create        = hdev_create,
1548     .create_options     = raw_create_options,
1549
1550     .bdrv_aio_readv     = raw_aio_readv,
1551     .bdrv_aio_writev    = raw_aio_writev,
1552     .bdrv_aio_flush     = raw_aio_flush,
1553     .bdrv_aio_discard   = hdev_aio_discard,
1554
1555     .bdrv_truncate      = raw_truncate,
1556     .bdrv_getlength     = raw_getlength,
1557     .bdrv_get_allocated_file_size
1558                         = raw_get_allocated_file_size,
1559
1560     /* generic scsi device */
1561 #ifdef __linux__
1562     .bdrv_ioctl         = hdev_ioctl,
1563     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1564 #endif
1565 };
1566
1567 #ifdef __linux__
1568 static int floppy_open(BlockDriverState *bs, QDict *options, int flags)
1569 {
1570     BDRVRawState *s = bs->opaque;
1571     int ret;
1572
1573     s->type = FTYPE_FD;
1574
1575     /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1576     ret = raw_open_common(bs, options, flags, O_NONBLOCK);
1577     if (ret)
1578         return ret;
1579
1580     /* close fd so that we can reopen it as needed */
1581     qemu_close(s->fd);
1582     s->fd = -1;
1583     s->fd_media_changed = 1;
1584
1585     return 0;
1586 }
1587
1588 static int floppy_probe_device(const char *filename)
1589 {
1590     int fd, ret;
1591     int prio = 0;
1592     struct floppy_struct fdparam;
1593     struct stat st;
1594
1595     if (strstart(filename, "/dev/fd", NULL) &&
1596         !strstart(filename, "/dev/fdset/", NULL)) {
1597         prio = 50;
1598     }
1599
1600     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1601     if (fd < 0) {
1602         goto out;
1603     }
1604     ret = fstat(fd, &st);
1605     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1606         goto outc;
1607     }
1608
1609     /* Attempt to detect via a floppy specific ioctl */
1610     ret = ioctl(fd, FDGETPRM, &fdparam);
1611     if (ret >= 0)
1612         prio = 100;
1613
1614 outc:
1615     qemu_close(fd);
1616 out:
1617     return prio;
1618 }
1619
1620
1621 static int floppy_is_inserted(BlockDriverState *bs)
1622 {
1623     return fd_open(bs) >= 0;
1624 }
1625
1626 static int floppy_media_changed(BlockDriverState *bs)
1627 {
1628     BDRVRawState *s = bs->opaque;
1629     int ret;
1630
1631     /*
1632      * XXX: we do not have a true media changed indication.
1633      * It does not work if the floppy is changed without trying to read it.
1634      */
1635     fd_open(bs);
1636     ret = s->fd_media_changed;
1637     s->fd_media_changed = 0;
1638 #ifdef DEBUG_FLOPPY
1639     printf("Floppy changed=%d\n", ret);
1640 #endif
1641     return ret;
1642 }
1643
1644 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
1645 {
1646     BDRVRawState *s = bs->opaque;
1647     int fd;
1648
1649     if (s->fd >= 0) {
1650         qemu_close(s->fd);
1651         s->fd = -1;
1652     }
1653     fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
1654     if (fd >= 0) {
1655         if (ioctl(fd, FDEJECT, 0) < 0)
1656             perror("FDEJECT");
1657         qemu_close(fd);
1658     }
1659 }
1660
1661 static BlockDriver bdrv_host_floppy = {
1662     .format_name        = "host_floppy",
1663     .protocol_name      = "host_floppy",
1664     .instance_size      = sizeof(BDRVRawState),
1665     .bdrv_probe_device  = floppy_probe_device,
1666     .bdrv_file_open     = floppy_open,
1667     .bdrv_close         = raw_close,
1668     .bdrv_reopen_prepare = raw_reopen_prepare,
1669     .bdrv_reopen_commit  = raw_reopen_commit,
1670     .bdrv_reopen_abort   = raw_reopen_abort,
1671     .bdrv_create        = hdev_create,
1672     .create_options     = raw_create_options,
1673
1674     .bdrv_aio_readv     = raw_aio_readv,
1675     .bdrv_aio_writev    = raw_aio_writev,
1676     .bdrv_aio_flush     = raw_aio_flush,
1677
1678     .bdrv_truncate      = raw_truncate,
1679     .bdrv_getlength     = raw_getlength,
1680     .bdrv_get_allocated_file_size
1681                         = raw_get_allocated_file_size,
1682
1683     /* removable device support */
1684     .bdrv_is_inserted   = floppy_is_inserted,
1685     .bdrv_media_changed = floppy_media_changed,
1686     .bdrv_eject         = floppy_eject,
1687 };
1688
1689 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags)
1690 {
1691     BDRVRawState *s = bs->opaque;
1692
1693     s->type = FTYPE_CD;
1694
1695     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
1696     return raw_open_common(bs, options, flags, O_NONBLOCK);
1697 }
1698
1699 static int cdrom_probe_device(const char *filename)
1700 {
1701     int fd, ret;
1702     int prio = 0;
1703     struct stat st;
1704
1705     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1706     if (fd < 0) {
1707         goto out;
1708     }
1709     ret = fstat(fd, &st);
1710     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1711         goto outc;
1712     }
1713
1714     /* Attempt to detect via a CDROM specific ioctl */
1715     ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1716     if (ret >= 0)
1717         prio = 100;
1718
1719 outc:
1720     qemu_close(fd);
1721 out:
1722     return prio;
1723 }
1724
1725 static int cdrom_is_inserted(BlockDriverState *bs)
1726 {
1727     BDRVRawState *s = bs->opaque;
1728     int ret;
1729
1730     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1731     if (ret == CDS_DISC_OK)
1732         return 1;
1733     return 0;
1734 }
1735
1736 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1737 {
1738     BDRVRawState *s = bs->opaque;
1739
1740     if (eject_flag) {
1741         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
1742             perror("CDROMEJECT");
1743     } else {
1744         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
1745             perror("CDROMEJECT");
1746     }
1747 }
1748
1749 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1750 {
1751     BDRVRawState *s = bs->opaque;
1752
1753     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
1754         /*
1755          * Note: an error can happen if the distribution automatically
1756          * mounts the CD-ROM
1757          */
1758         /* perror("CDROM_LOCKDOOR"); */
1759     }
1760 }
1761
1762 static BlockDriver bdrv_host_cdrom = {
1763     .format_name        = "host_cdrom",
1764     .protocol_name      = "host_cdrom",
1765     .instance_size      = sizeof(BDRVRawState),
1766     .bdrv_probe_device  = cdrom_probe_device,
1767     .bdrv_file_open     = cdrom_open,
1768     .bdrv_close         = raw_close,
1769     .bdrv_reopen_prepare = raw_reopen_prepare,
1770     .bdrv_reopen_commit  = raw_reopen_commit,
1771     .bdrv_reopen_abort   = raw_reopen_abort,
1772     .bdrv_create        = hdev_create,
1773     .create_options     = raw_create_options,
1774
1775     .bdrv_aio_readv     = raw_aio_readv,
1776     .bdrv_aio_writev    = raw_aio_writev,
1777     .bdrv_aio_flush     = raw_aio_flush,
1778
1779     .bdrv_truncate      = raw_truncate,
1780     .bdrv_getlength     = raw_getlength,
1781     .bdrv_get_allocated_file_size
1782                         = raw_get_allocated_file_size,
1783
1784     /* removable device support */
1785     .bdrv_is_inserted   = cdrom_is_inserted,
1786     .bdrv_eject         = cdrom_eject,
1787     .bdrv_lock_medium   = cdrom_lock_medium,
1788
1789     /* generic scsi device */
1790     .bdrv_ioctl         = hdev_ioctl,
1791     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1792 };
1793 #endif /* __linux__ */
1794
1795 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1796 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags)
1797 {
1798     BDRVRawState *s = bs->opaque;
1799     int ret;
1800
1801     s->type = FTYPE_CD;
1802
1803     ret = raw_open_common(bs, options, flags, 0);
1804     if (ret)
1805         return ret;
1806
1807     /* make sure the door isn't locked at this time */
1808     ioctl(s->fd, CDIOCALLOW);
1809     return 0;
1810 }
1811
1812 static int cdrom_probe_device(const char *filename)
1813 {
1814     if (strstart(filename, "/dev/cd", NULL) ||
1815             strstart(filename, "/dev/acd", NULL))
1816         return 100;
1817     return 0;
1818 }
1819
1820 static int cdrom_reopen(BlockDriverState *bs)
1821 {
1822     BDRVRawState *s = bs->opaque;
1823     int fd;
1824
1825     /*
1826      * Force reread of possibly changed/newly loaded disc,
1827      * FreeBSD seems to not notice sometimes...
1828      */
1829     if (s->fd >= 0)
1830         qemu_close(s->fd);
1831     fd = qemu_open(bs->filename, s->open_flags, 0644);
1832     if (fd < 0) {
1833         s->fd = -1;
1834         return -EIO;
1835     }
1836     s->fd = fd;
1837
1838     /* make sure the door isn't locked at this time */
1839     ioctl(s->fd, CDIOCALLOW);
1840     return 0;
1841 }
1842
1843 static int cdrom_is_inserted(BlockDriverState *bs)
1844 {
1845     return raw_getlength(bs) > 0;
1846 }
1847
1848 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1849 {
1850     BDRVRawState *s = bs->opaque;
1851
1852     if (s->fd < 0)
1853         return;
1854
1855     (void) ioctl(s->fd, CDIOCALLOW);
1856
1857     if (eject_flag) {
1858         if (ioctl(s->fd, CDIOCEJECT) < 0)
1859             perror("CDIOCEJECT");
1860     } else {
1861         if (ioctl(s->fd, CDIOCCLOSE) < 0)
1862             perror("CDIOCCLOSE");
1863     }
1864
1865     cdrom_reopen(bs);
1866 }
1867
1868 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1869 {
1870     BDRVRawState *s = bs->opaque;
1871
1872     if (s->fd < 0)
1873         return;
1874     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1875         /*
1876          * Note: an error can happen if the distribution automatically
1877          * mounts the CD-ROM
1878          */
1879         /* perror("CDROM_LOCKDOOR"); */
1880     }
1881 }
1882
1883 static BlockDriver bdrv_host_cdrom = {
1884     .format_name        = "host_cdrom",
1885     .protocol_name      = "host_cdrom",
1886     .instance_size      = sizeof(BDRVRawState),
1887     .bdrv_probe_device  = cdrom_probe_device,
1888     .bdrv_file_open     = cdrom_open,
1889     .bdrv_close         = raw_close,
1890     .bdrv_reopen_prepare = raw_reopen_prepare,
1891     .bdrv_reopen_commit  = raw_reopen_commit,
1892     .bdrv_reopen_abort   = raw_reopen_abort,
1893     .bdrv_create        = hdev_create,
1894     .create_options     = raw_create_options,
1895
1896     .bdrv_aio_readv     = raw_aio_readv,
1897     .bdrv_aio_writev    = raw_aio_writev,
1898     .bdrv_aio_flush     = raw_aio_flush,
1899
1900     .bdrv_truncate      = raw_truncate,
1901     .bdrv_getlength     = raw_getlength,
1902     .bdrv_get_allocated_file_size
1903                         = raw_get_allocated_file_size,
1904
1905     /* removable device support */
1906     .bdrv_is_inserted   = cdrom_is_inserted,
1907     .bdrv_eject         = cdrom_eject,
1908     .bdrv_lock_medium   = cdrom_lock_medium,
1909 };
1910 #endif /* __FreeBSD__ */
1911
1912 #ifdef CONFIG_LINUX_AIO
1913 /**
1914  * Return the file descriptor for Linux AIO
1915  *
1916  * This function is a layering violation and should be removed when it becomes
1917  * possible to call the block layer outside the global mutex.  It allows the
1918  * caller to hijack the file descriptor so I/O can be performed outside the
1919  * block layer.
1920  */
1921 int raw_get_aio_fd(BlockDriverState *bs)
1922 {
1923     BDRVRawState *s;
1924
1925     if (!bs->drv) {
1926         return -ENOMEDIUM;
1927     }
1928
1929     if (bs->drv == bdrv_find_format("raw")) {
1930         bs = bs->file;
1931     }
1932
1933     /* raw-posix has several protocols so just check for raw_aio_readv */
1934     if (bs->drv->bdrv_aio_readv != raw_aio_readv) {
1935         return -ENOTSUP;
1936     }
1937
1938     s = bs->opaque;
1939     if (!s->use_aio) {
1940         return -ENOTSUP;
1941     }
1942     return s->fd;
1943 }
1944 #endif /* CONFIG_LINUX_AIO */
1945
1946 static void bdrv_file_init(void)
1947 {
1948     /*
1949      * Register all the drivers.  Note that order is important, the driver
1950      * registered last will get probed first.
1951      */
1952     bdrv_register(&bdrv_file);
1953     bdrv_register(&bdrv_host_device);
1954 #ifdef __linux__
1955     bdrv_register(&bdrv_host_floppy);
1956     bdrv_register(&bdrv_host_cdrom);
1957 #endif
1958 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1959     bdrv_register(&bdrv_host_cdrom);
1960 #endif
1961 }
1962
1963 block_init(bdrv_file_init);