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, const char *filename,
 339                     QDict *options, int flags)
 340 {
 341     BDRVRawState *s = bs->opaque;
 342
 343     s->type = FTYPE_FILE;
 344     return raw_open_common(bs, options, flags, 0);
 345 }
 346
 347 static int raw_reopen_prepare(BDRVReopenState *state,
 348                               BlockReopenQueue *queue, Error **errp)
 349 {
 350     BDRVRawState *s;
 351     BDRVRawReopenState *raw_s;
 352     int ret = 0;
 353
 354     assert(state != NULL);
 355     assert(state->bs != NULL);
 356
 357     s = state->bs->opaque;
 358
 359     state->opaque = g_malloc0(sizeof(BDRVRawReopenState));
 360     raw_s = state->opaque;
 361
 362 #ifdef CONFIG_LINUX_AIO
 363     raw_s->use_aio = s->use_aio;
 364
 365     /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
 366      * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
 367      * won't override aio_ctx if aio_ctx is non-NULL */
 368     if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
 369         return -1;
 370     }
 371 #endif
 372
 373     if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
 374         raw_s->open_flags |= O_NONBLOCK;
 375     }
 376
 377     raw_parse_flags(state->flags, &raw_s->open_flags);
 378
 379     raw_s->fd = -1;
 380
 381     int fcntl_flags = O_APPEND | O_NONBLOCK;
 382 #ifdef O_NOATIME
 383     fcntl_flags |= O_NOATIME;
 384 #endif
 385
 386 #ifdef O_ASYNC
 387     /* Not all operating systems have O_ASYNC, and those that don't
 388      * will not let us track the state into raw_s->open_flags (typically
 389      * you achieve the same effect with an ioctl, for example I_SETSIG
 390      * on Solaris). But we do not use O_ASYNC, so that's fine.
 391      */
 392     assert((s->open_flags & O_ASYNC) == 0);
 393 #endif
 394
 395     if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
 396         /* dup the original fd */
 397         /* TODO: use qemu fcntl wrapper */
 398 #ifdef F_DUPFD_CLOEXEC
 399         raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
 400 #else
 401         raw_s->fd = dup(s->fd);
 402         if (raw_s->fd != -1) {
 403             qemu_set_cloexec(raw_s->fd);
 404         }
 405 #endif
 406         if (raw_s->fd >= 0) {
 407             ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
 408             if (ret) {
 409                 qemu_close(raw_s->fd);
 410                 raw_s->fd = -1;
 411             }
 412         }
 413     }
 414
 415     /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
 416     if (raw_s->fd == -1) {
 417         assert(!(raw_s->open_flags & O_CREAT));
 418         raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
 419         if (raw_s->fd == -1) {
 420             ret = -1;
 421         }
 422     }
 423     return ret;
 424 }
 425
 426
 427 static void raw_reopen_commit(BDRVReopenState *state)
 428 {
 429     BDRVRawReopenState *raw_s = state->opaque;
 430     BDRVRawState *s = state->bs->opaque;
 431
 432     s->open_flags = raw_s->open_flags;
 433
 434     qemu_close(s->fd);
 435     s->fd = raw_s->fd;
 436 #ifdef CONFIG_LINUX_AIO
 437     s->use_aio = raw_s->use_aio;
 438 #endif
 439
 440     g_free(state->opaque);
 441     state->opaque = NULL;
 442 }
 443
 444
 445 static void raw_reopen_abort(BDRVReopenState *state)
 446 {
 447     BDRVRawReopenState *raw_s = state->opaque;
 448
 449      /* nothing to do if NULL, we didn't get far enough */
 450     if (raw_s == NULL) {
 451         return;
 452     }
 453
 454     if (raw_s->fd >= 0) {
 455         qemu_close(raw_s->fd);
 456         raw_s->fd = -1;
 457     }
 458     g_free(state->opaque);
 459     state->opaque = NULL;
 460 }
 461
 462
 463 /* XXX: use host sector size if necessary with:
 464 #ifdef DIOCGSECTORSIZE
 465         {
 466             unsigned int sectorsize = 512;
 467             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 468                 sectorsize > bufsize)
 469                 bufsize = sectorsize;
 470         }
 471 #endif
 472 #ifdef CONFIG_COCOA
 473         uint32_t blockSize = 512;
 474         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 475             bufsize = blockSize;
 476         }
 477 #endif
 478 */
 479
 480 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
 481 {
 482     int ret;
 483
 484     ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
 485     if (ret == -1) {
 486         return -errno;
 487     }
 488
 489     return 0;
 490 }
 491
 492 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
 493 {
 494     int ret;
 495
 496     ret = qemu_fdatasync(aiocb->aio_fildes);
 497     if (ret == -1) {
 498         return -errno;
 499     }
 500     return 0;
 501 }
 502
 503 #ifdef CONFIG_PREADV
 504
 505 static bool preadv_present = true;
 506
 507 static ssize_t
 508 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 509 {
 510     return preadv(fd, iov, nr_iov, offset);
 511 }
 512
 513 static ssize_t
 514 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 515 {
 516     return pwritev(fd, iov, nr_iov, offset);
 517 }
 518
 519 #else
 520
 521 static bool preadv_present = false;
 522
 523 static ssize_t
 524 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 525 {
 526     return -ENOSYS;
 527 }
 528
 529 static ssize_t
 530 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 531 {
 532     return -ENOSYS;
 533 }
 534
 535 #endif
 536
 537 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
 538 {
 539     ssize_t len;
 540
 541     do {
 542         if (aiocb->aio_type & QEMU_AIO_WRITE)
 543             len = qemu_pwritev(aiocb->aio_fildes,
 544                                aiocb->aio_iov,
 545                                aiocb->aio_niov,
 546                                aiocb->aio_offset);
 547          else
 548             len = qemu_preadv(aiocb->aio_fildes,
 549                               aiocb->aio_iov,
 550                               aiocb->aio_niov,
 551                               aiocb->aio_offset);
 552     } while (len == -1 && errno == EINTR);
 553
 554     if (len == -1) {
 555         return -errno;
 556     }
 557     return len;
 558 }
 559
 560 /*
 561  * Read/writes the data to/from a given linear buffer.
 562  *
 563  * Returns the number of bytes handles or -errno in case of an error. Short
 564  * reads are only returned if the end of the file is reached.
 565  */
 566 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
 567 {
 568     ssize_t offset = 0;
 569     ssize_t len;
 570
 571     while (offset < aiocb->aio_nbytes) {
 572         if (aiocb->aio_type & QEMU_AIO_WRITE) {
 573             len = pwrite(aiocb->aio_fildes,
 574                          (const char *)buf + offset,
 575                          aiocb->aio_nbytes - offset,
 576                          aiocb->aio_offset + offset);
 577         } else {
 578             len = pread(aiocb->aio_fildes,
 579                         buf + offset,
 580                         aiocb->aio_nbytes - offset,
 581                         aiocb->aio_offset + offset);
 582         }
 583         if (len == -1 && errno == EINTR) {
 584             continue;
 585         } else if (len == -1) {
 586             offset = -errno;
 587             break;
 588         } else if (len == 0) {
 589             break;
 590         }
 591         offset += len;
 592     }
 593
 594     return offset;
 595 }
 596
 597 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
 598 {
 599     ssize_t nbytes;
 600     char *buf;
 601
 602     if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
 603         /*
 604          * If there is just a single buffer, and it is properly aligned
 605          * we can just use plain pread/pwrite without any problems.
 606          */
 607         if (aiocb->aio_niov == 1) {
 608              return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
 609         }
 610         /*
 611          * We have more than one iovec, and all are properly aligned.
 612          *
 613          * Try preadv/pwritev first and fall back to linearizing the
 614          * buffer if it's not supported.
 615          */
 616         if (preadv_present) {
 617             nbytes = handle_aiocb_rw_vector(aiocb);
 618             if (nbytes == aiocb->aio_nbytes ||
 619                 (nbytes < 0 && nbytes != -ENOSYS)) {
 620                 return nbytes;
 621             }
 622             preadv_present = false;
 623         }
 624
 625         /*
 626          * XXX(hch): short read/write.  no easy way to handle the reminder
 627          * using these interfaces.  For now retry using plain
 628          * pread/pwrite?
 629          */
 630     }
 631
 632     /*
 633      * Ok, we have to do it the hard way, copy all segments into
 634      * a single aligned buffer.
 635      */
 636     buf = qemu_blockalign(aiocb->bs, aiocb->aio_nbytes);
 637     if (aiocb->aio_type & QEMU_AIO_WRITE) {
 638         char *p = buf;
 639         int i;
 640
 641         for (i = 0; i < aiocb->aio_niov; ++i) {
 642             memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
 643             p += aiocb->aio_iov[i].iov_len;
 644         }
 645     }
 646
 647     nbytes = handle_aiocb_rw_linear(aiocb, buf);
 648     if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
 649         char *p = buf;
 650         size_t count = aiocb->aio_nbytes, copy;
 651         int i;
 652
 653         for (i = 0; i < aiocb->aio_niov && count; ++i) {
 654             copy = count;
 655             if (copy > aiocb->aio_iov[i].iov_len) {
 656                 copy = aiocb->aio_iov[i].iov_len;
 657             }
 658             memcpy(aiocb->aio_iov[i].iov_base, p, copy);
 659             p     += copy;
 660             count -= copy;
 661         }
 662     }
 663     qemu_vfree(buf);
 664
 665     return nbytes;
 666 }
 667
 668 #ifdef CONFIG_XFS
 669 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
 670 {
 671     struct xfs_flock64 fl;
 672
 673     memset(&fl, 0, sizeof(fl));
 674     fl.l_whence = SEEK_SET;
 675     fl.l_start = offset;
 676     fl.l_len = bytes;
 677
 678     if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
 679         DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
 680         return -errno;
 681     }
 682
 683     return 0;
 684 }
 685 #endif
 686
 687 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
 688 {
 689     int ret = -EOPNOTSUPP;
 690     BDRVRawState *s = aiocb->bs->opaque;
 691
 692     if (s->has_discard == 0) {
 693         return 0;
 694     }
 695
 696     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
 697 #ifdef BLKDISCARD
 698         do {
 699             uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
 700             if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
 701                 return 0;
 702             }
 703         } while (errno == EINTR);
 704
 705         ret = -errno;
 706 #endif
 707     } else {
 708 #ifdef CONFIG_XFS
 709         if (s->is_xfs) {
 710             return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
 711         }
 712 #endif
 713
 714 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
 715         do {
 716             if (fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
 717                           aiocb->aio_offset, aiocb->aio_nbytes) == 0) {
 718                 return 0;
 719             }
 720         } while (errno == EINTR);
 721
 722         ret = -errno;
 723 #endif
 724     }
 725
 726     if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
 727         ret == -ENOTTY) {
 728         s->has_discard = 0;
 729         ret = 0;
 730     }
 731     return ret;
 732 }
 733
 734 static int aio_worker(void *arg)
 735 {
 736     RawPosixAIOData *aiocb = arg;
 737     ssize_t ret = 0;
 738
 739     switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
 740     case QEMU_AIO_READ:
 741         ret = handle_aiocb_rw(aiocb);
 742         if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) {
 743             iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
 744                       0, aiocb->aio_nbytes - ret);
 745
 746             ret = aiocb->aio_nbytes;
 747         }
 748         if (ret == aiocb->aio_nbytes) {
 749             ret = 0;
 750         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 751             ret = -EINVAL;
 752         }
 753         break;
 754     case QEMU_AIO_WRITE:
 755         ret = handle_aiocb_rw(aiocb);
 756         if (ret == aiocb->aio_nbytes) {
 757             ret = 0;
 758         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 759             ret = -EINVAL;
 760         }
 761         break;
 762     case QEMU_AIO_FLUSH:
 763         ret = handle_aiocb_flush(aiocb);
 764         break;
 765     case QEMU_AIO_IOCTL:
 766         ret = handle_aiocb_ioctl(aiocb);
 767         break;
 768     case QEMU_AIO_DISCARD:
 769         ret = handle_aiocb_discard(aiocb);
 770         break;
 771     default:
 772         fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
 773         ret = -EINVAL;
 774         break;
 775     }
 776
 777     g_slice_free(RawPosixAIOData, aiocb);
 778     return ret;
 779 }
 780
 781 static BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd,
 782         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 783         BlockDriverCompletionFunc *cb, void *opaque, int type)
 784 {
 785     RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
 786     ThreadPool *pool;
 787
 788     acb->bs = bs;
 789     acb->aio_type = type;
 790     acb->aio_fildes = fd;
 791
 792     if (qiov) {
 793         acb->aio_iov = qiov->iov;
 794         acb->aio_niov = qiov->niov;
 795     }
 796     acb->aio_nbytes = nb_sectors * 512;
 797     acb->aio_offset = sector_num * 512;
 798
 799     trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
 800     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
 801     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
 802 }
 803
 804 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
 805         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 806         BlockDriverCompletionFunc *cb, void *opaque, int type)
 807 {
 808     BDRVRawState *s = bs->opaque;
 809
 810     if (fd_open(bs) < 0)
 811         return NULL;
 812
 813     /*
 814      * If O_DIRECT is used the buffer needs to be aligned on a sector
 815      * boundary.  Check if this is the case or tell the low-level
 816      * driver that it needs to copy the buffer.
 817      */
 818     if ((bs->open_flags & BDRV_O_NOCACHE)) {
 819         if (!bdrv_qiov_is_aligned(bs, qiov)) {
 820             type |= QEMU_AIO_MISALIGNED;
 821 #ifdef CONFIG_LINUX_AIO
 822         } else if (s->use_aio) {
 823             return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
 824                                nb_sectors, cb, opaque, type);
 825 #endif
 826         }
 827     }
 828
 829     return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
 830                        cb, opaque, type);
 831 }
 832
 833 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
 834         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 835         BlockDriverCompletionFunc *cb, void *opaque)
 836 {
 837     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 838                           cb, opaque, QEMU_AIO_READ);
 839 }
 840
 841 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
 842         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 843         BlockDriverCompletionFunc *cb, void *opaque)
 844 {
 845     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 846                           cb, opaque, QEMU_AIO_WRITE);
 847 }
 848
 849 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
 850         BlockDriverCompletionFunc *cb, void *opaque)
 851 {
 852     BDRVRawState *s = bs->opaque;
 853
 854     if (fd_open(bs) < 0)
 855         return NULL;
 856
 857     return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
 858 }
 859
 860 static void raw_close(BlockDriverState *bs)
 861 {
 862     BDRVRawState *s = bs->opaque;
 863     if (s->fd >= 0) {
 864         qemu_close(s->fd);
 865         s->fd = -1;
 866     }
 867 }
 868
 869 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 870 {
 871     BDRVRawState *s = bs->opaque;
 872     struct stat st;
 873
 874     if (fstat(s->fd, &st)) {
 875         return -errno;
 876     }
 877
 878     if (S_ISREG(st.st_mode)) {
 879         if (ftruncate(s->fd, offset) < 0) {
 880             return -errno;
 881         }
 882     } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 883        if (offset > raw_getlength(bs)) {
 884            return -EINVAL;
 885        }
 886     } else {
 887         return -ENOTSUP;
 888     }
 889
 890     return 0;
 891 }
 892
 893 #ifdef __OpenBSD__
 894 static int64_t raw_getlength(BlockDriverState *bs)
 895 {
 896     BDRVRawState *s = bs->opaque;
 897     int fd = s->fd;
 898     struct stat st;
 899
 900     if (fstat(fd, &st))
 901         return -1;
 902     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 903         struct disklabel dl;
 904
 905         if (ioctl(fd, DIOCGDINFO, &dl))
 906             return -1;
 907         return (uint64_t)dl.d_secsize *
 908             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 909     } else
 910         return st.st_size;
 911 }
 912 #elif defined(__NetBSD__)
 913 static int64_t raw_getlength(BlockDriverState *bs)
 914 {
 915     BDRVRawState *s = bs->opaque;
 916     int fd = s->fd;
 917     struct stat st;
 918
 919     if (fstat(fd, &st))
 920         return -1;
 921     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 922         struct dkwedge_info dkw;
 923
 924         if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
 925             return dkw.dkw_size * 512;
 926         } else {
 927             struct disklabel dl;
 928
 929             if (ioctl(fd, DIOCGDINFO, &dl))
 930                 return -1;
 931             return (uint64_t)dl.d_secsize *
 932                 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 933         }
 934     } else
 935         return st.st_size;
 936 }
 937 #elif defined(__sun__)
 938 static int64_t raw_getlength(BlockDriverState *bs)
 939 {
 940     BDRVRawState *s = bs->opaque;
 941     struct dk_minfo minfo;
 942     int ret;
 943
 944     ret = fd_open(bs);
 945     if (ret < 0) {
 946         return ret;
 947     }
 948
 949     /*
 950      * Use the DKIOCGMEDIAINFO ioctl to read the size.
 951      */
 952     ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
 953     if (ret != -1) {
 954         return minfo.dki_lbsize * minfo.dki_capacity;
 955     }
 956
 957     /*
 958      * There are reports that lseek on some devices fails, but
 959      * irc discussion said that contingency on contingency was overkill.
 960      */
 961     return lseek(s->fd, 0, SEEK_END);
 962 }
 963 #elif defined(CONFIG_BSD)
 964 static int64_t raw_getlength(BlockDriverState *bs)
 965 {
 966     BDRVRawState *s = bs->opaque;
 967     int fd = s->fd;
 968     int64_t size;
 969     struct stat sb;
 970 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 971     int reopened = 0;
 972 #endif
 973     int ret;
 974
 975     ret = fd_open(bs);
 976     if (ret < 0)
 977         return ret;
 978
 979 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 980 again:
 981 #endif
 982     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 983 #ifdef DIOCGMEDIASIZE
 984         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 985 #elif defined(DIOCGPART)
 986         {
 987                 struct partinfo pi;
 988                 if (ioctl(fd, DIOCGPART, &pi) == 0)
 989                         size = pi.media_size;
 990                 else
 991                         size = 0;
 992         }
 993         if (size == 0)
 994 #endif
 995 #if defined(__APPLE__) && defined(__MACH__)
 996         size = LONG_LONG_MAX;
 997 #else
 998         size = lseek(fd, 0LL, SEEK_END);
 999 #endif
1000 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1001         switch(s->type) {
1002         case FTYPE_CD:
1003             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1004             if (size == 2048LL * (unsigned)-1)
1005                 size = 0;
1006             /* XXX no disc?  maybe we need to reopen... */
1007             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1008                 reopened = 1;
1009                 goto again;
1010             }
1011         }
1012 #endif
1013     } else {
1014         size = lseek(fd, 0, SEEK_END);
1015     }
1016     return size;
1017 }
1018 #else
1019 static int64_t raw_getlength(BlockDriverState *bs)
1020 {
1021     BDRVRawState *s = bs->opaque;
1022     int ret;
1023
1024     ret = fd_open(bs);
1025     if (ret < 0) {
1026         return ret;
1027     }
1028
1029     return lseek(s->fd, 0, SEEK_END);
1030 }
1031 #endif
1032
1033 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1034 {
1035     struct stat st;
1036     BDRVRawState *s = bs->opaque;
1037
1038     if (fstat(s->fd, &st) < 0) {
1039         return -errno;
1040     }
1041     return (int64_t)st.st_blocks * 512;
1042 }
1043
1044 static int raw_create(const char *filename, QEMUOptionParameter *options)
1045 {
1046     int fd;
1047     int result = 0;
1048     int64_t total_size = 0;
1049
1050     /* Read out options */
1051     while (options && options->name) {
1052         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1053             total_size = options->value.n / BDRV_SECTOR_SIZE;
1054         }
1055         options++;
1056     }
1057
1058     fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
1059                    0644);
1060     if (fd < 0) {
1061         result = -errno;
1062     } else {
1063         if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
1064             result = -errno;
1065         }
1066         if (qemu_close(fd) != 0) {
1067             result = -errno;
1068         }
1069     }
1070     return result;
1071 }
1072
1073 /*
1074  * Returns true iff the specified sector is present in the disk image. Drivers
1075  * not implementing the functionality are assumed to not support backing files,
1076  * hence all their sectors are reported as allocated.
1077  *
1078  * If 'sector_num' is beyond the end of the disk image the return value is 0
1079  * and 'pnum' is set to 0.
1080  *
1081  * 'pnum' is set to the number of sectors (including and immediately following
1082  * the specified sector) that are known to be in the same
1083  * allocated/unallocated state.
1084  *
1085  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
1086  * beyond the end of the disk image it will be clamped.
1087  */
1088 static int coroutine_fn raw_co_is_allocated(BlockDriverState *bs,
1089                                             int64_t sector_num,
1090                                             int nb_sectors, int *pnum)
1091 {
1092     off_t start, data, hole;
1093     int ret;
1094
1095     ret = fd_open(bs);
1096     if (ret < 0) {
1097         return ret;
1098     }
1099
1100     start = sector_num * BDRV_SECTOR_SIZE;
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 1;
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     }
1132
1133 #elif defined SEEK_HOLE && defined SEEK_DATA
1134
1135     BDRVRawState *s = bs->opaque;
1136
1137     hole = lseek(s->fd, start, SEEK_HOLE);
1138     if (hole == -1) {
1139         /* -ENXIO indicates that sector_num was past the end of the file.
1140          * There is a virtual hole there.  */
1141         assert(errno != -ENXIO);
1142
1143         /* Most likely EINVAL.  Assume everything is allocated.  */
1144         *pnum = nb_sectors;
1145         return 1;
1146     }
1147
1148     if (hole > start) {
1149         data = start;
1150     } else {
1151         /* On a hole.  We need another syscall to find its end.  */
1152         data = lseek(s->fd, start, SEEK_DATA);
1153         if (data == -1) {
1154             data = lseek(s->fd, 0, SEEK_END);
1155         }
1156     }
1157 #else
1158     *pnum = nb_sectors;
1159     return 1;
1160 #endif
1161
1162     if (data <= start) {
1163         /* On a data extent, compute sectors to the end of the extent.  */
1164         *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1165         return 1;
1166     } else {
1167         /* On a hole, compute sectors to the beginning of the next extent.  */
1168         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1169         return 0;
1170     }
1171 }
1172
1173 static coroutine_fn BlockDriverAIOCB *raw_aio_discard(BlockDriverState *bs,
1174     int64_t sector_num, int nb_sectors,
1175     BlockDriverCompletionFunc *cb, void *opaque)
1176 {
1177     BDRVRawState *s = bs->opaque;
1178
1179     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1180                        cb, opaque, QEMU_AIO_DISCARD);
1181 }
1182
1183 static QEMUOptionParameter raw_create_options[] = {
1184     {
1185         .name = BLOCK_OPT_SIZE,
1186         .type = OPT_SIZE,
1187         .help = "Virtual disk size"
1188     },
1189     { NULL }
1190 };
1191
1192 static BlockDriver bdrv_file = {
1193     .format_name = "file",
1194     .protocol_name = "file",
1195     .instance_size = sizeof(BDRVRawState),
1196     .bdrv_probe = NULL, /* no probe for protocols */
1197     .bdrv_file_open = raw_open,
1198     .bdrv_reopen_prepare = raw_reopen_prepare,
1199     .bdrv_reopen_commit = raw_reopen_commit,
1200     .bdrv_reopen_abort = raw_reopen_abort,
1201     .bdrv_close = raw_close,
1202     .bdrv_create = raw_create,
1203     .bdrv_co_is_allocated = raw_co_is_allocated,
1204
1205     .bdrv_aio_readv = raw_aio_readv,
1206     .bdrv_aio_writev = raw_aio_writev,
1207     .bdrv_aio_flush = raw_aio_flush,
1208     .bdrv_aio_discard = raw_aio_discard,
1209
1210     .bdrv_truncate = raw_truncate,
1211     .bdrv_getlength = raw_getlength,
1212     .bdrv_get_allocated_file_size
1213                         = raw_get_allocated_file_size,
1214
1215     .create_options = raw_create_options,
1216 };
1217
1218 /***********************************************/
1219 /* host device */
1220
1221 #if defined(__APPLE__) && defined(__MACH__)
1222 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1223 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1224
1225 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1226 {
1227     kern_return_t       kernResult;
1228     mach_port_t     masterPort;
1229     CFMutableDictionaryRef  classesToMatch;
1230
1231     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1232     if ( KERN_SUCCESS != kernResult ) {
1233         printf( "IOMasterPort returned %d\n", kernResult );
1234     }
1235
1236     classesToMatch = IOServiceMatching( kIOCDMediaClass );
1237     if ( classesToMatch == NULL ) {
1238         printf( "IOServiceMatching returned a NULL dictionary.\n" );
1239     } else {
1240     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1241     }
1242     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1243     if ( KERN_SUCCESS != kernResult )
1244     {
1245         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1246     }
1247
1248     return kernResult;
1249 }
1250
1251 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
1252 {
1253     io_object_t     nextMedia;
1254     kern_return_t   kernResult = KERN_FAILURE;
1255     *bsdPath = '\0';
1256     nextMedia = IOIteratorNext( mediaIterator );
1257     if ( nextMedia )
1258     {
1259         CFTypeRef   bsdPathAsCFString;
1260     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1261         if ( bsdPathAsCFString ) {
1262             size_t devPathLength;
1263             strcpy( bsdPath, _PATH_DEV );
1264             strcat( bsdPath, "r" );
1265             devPathLength = strlen( bsdPath );
1266             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1267                 kernResult = KERN_SUCCESS;
1268             }
1269             CFRelease( bsdPathAsCFString );
1270         }
1271         IOObjectRelease( nextMedia );
1272     }
1273
1274     return kernResult;
1275 }
1276
1277 #endif
1278
1279 static int hdev_probe_device(const char *filename)
1280 {
1281     struct stat st;
1282
1283     /* allow a dedicated CD-ROM driver to match with a higher priority */
1284     if (strstart(filename, "/dev/cdrom", NULL))
1285         return 50;
1286
1287     if (stat(filename, &st) >= 0 &&
1288             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
1289         return 100;
1290     }
1291
1292     return 0;
1293 }
1294
1295 static int check_hdev_writable(BDRVRawState *s)
1296 {
1297 #if defined(BLKROGET)
1298     /* Linux block devices can be configured "read-only" using blockdev(8).
1299      * This is independent of device node permissions and therefore open(2)
1300      * with O_RDWR succeeds.  Actual writes fail with EPERM.
1301      *
1302      * bdrv_open() is supposed to fail if the disk is read-only.  Explicitly
1303      * check for read-only block devices so that Linux block devices behave
1304      * properly.
1305      */
1306     struct stat st;
1307     int readonly = 0;
1308
1309     if (fstat(s->fd, &st)) {
1310         return -errno;
1311     }
1312
1313     if (!S_ISBLK(st.st_mode)) {
1314         return 0;
1315     }
1316
1317     if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
1318         return -errno;
1319     }
1320
1321     if (readonly) {
1322         return -EACCES;
1323     }
1324 #endif /* defined(BLKROGET) */
1325     return 0;
1326 }
1327
1328 static int hdev_open(BlockDriverState *bs, const char *dummy,
1329                      QDict *options, int flags)
1330 {
1331     BDRVRawState *s = bs->opaque;
1332     int ret;
1333     const char *filename = qdict_get_str(options, "filename");
1334
1335 #if defined(__APPLE__) && defined(__MACH__)
1336     if (strstart(filename, "/dev/cdrom", NULL)) {
1337         kern_return_t kernResult;
1338         io_iterator_t mediaIterator;
1339         char bsdPath[ MAXPATHLEN ];
1340         int fd;
1341
1342         kernResult = FindEjectableCDMedia( &mediaIterator );
1343         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1344
1345         if ( bsdPath[ 0 ] != '\0' ) {
1346             strcat(bsdPath,"s0");
1347             /* some CDs don't have a partition 0 */
1348             fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1349             if (fd < 0) {
1350                 bsdPath[strlen(bsdPath)-1] = '1';
1351             } else {
1352                 qemu_close(fd);
1353             }
1354             filename = bsdPath;
1355         }
1356
1357         if ( mediaIterator )
1358             IOObjectRelease( mediaIterator );
1359     }
1360 #endif
1361
1362     s->type = FTYPE_FILE;
1363 #if defined(__linux__)
1364     {
1365         char resolved_path[ MAXPATHLEN ], *temp;
1366
1367         temp = realpath(filename, resolved_path);
1368         if (temp && strstart(temp, "/dev/sg", NULL)) {
1369             bs->sg = 1;
1370         }
1371     }
1372 #endif
1373
1374     ret = raw_open_common(bs, options, flags, 0);
1375     if (ret < 0) {
1376         return ret;
1377     }
1378
1379     if (flags & BDRV_O_RDWR) {
1380         ret = check_hdev_writable(s);
1381         if (ret < 0) {
1382             raw_close(bs);
1383             return ret;
1384         }
1385     }
1386
1387     return ret;
1388 }
1389
1390 #if defined(__linux__)
1391 /* Note: we do not have a reliable method to detect if the floppy is
1392    present. The current method is to try to open the floppy at every
1393    I/O and to keep it opened during a few hundreds of ms. */
1394 static int fd_open(BlockDriverState *bs)
1395 {
1396     BDRVRawState *s = bs->opaque;
1397     int last_media_present;
1398
1399     if (s->type != FTYPE_FD)
1400         return 0;
1401     last_media_present = (s->fd >= 0);
1402     if (s->fd >= 0 &&
1403         (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1404         qemu_close(s->fd);
1405         s->fd = -1;
1406 #ifdef DEBUG_FLOPPY
1407         printf("Floppy closed\n");
1408 #endif
1409     }
1410     if (s->fd < 0) {
1411         if (s->fd_got_error &&
1412             (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1413 #ifdef DEBUG_FLOPPY
1414             printf("No floppy (open delayed)\n");
1415 #endif
1416             return -EIO;
1417         }
1418         s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
1419         if (s->fd < 0) {
1420             s->fd_error_time = get_clock();
1421             s->fd_got_error = 1;
1422             if (last_media_present)
1423                 s->fd_media_changed = 1;
1424 #ifdef DEBUG_FLOPPY
1425             printf("No floppy\n");
1426 #endif
1427             return -EIO;
1428         }
1429 #ifdef DEBUG_FLOPPY
1430         printf("Floppy opened\n");
1431 #endif
1432     }
1433     if (!last_media_present)
1434         s->fd_media_changed = 1;
1435     s->fd_open_time = get_clock();
1436     s->fd_got_error = 0;
1437     return 0;
1438 }
1439
1440 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1441 {
1442     BDRVRawState *s = bs->opaque;
1443
1444     return ioctl(s->fd, req, buf);
1445 }
1446
1447 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1448         unsigned long int req, void *buf,
1449         BlockDriverCompletionFunc *cb, void *opaque)
1450 {
1451     BDRVRawState *s = bs->opaque;
1452     RawPosixAIOData *acb;
1453     ThreadPool *pool;
1454
1455     if (fd_open(bs) < 0)
1456         return NULL;
1457
1458     acb = g_slice_new(RawPosixAIOData);
1459     acb->bs = bs;
1460     acb->aio_type = QEMU_AIO_IOCTL;
1461     acb->aio_fildes = s->fd;
1462     acb->aio_offset = 0;
1463     acb->aio_ioctl_buf = buf;
1464     acb->aio_ioctl_cmd = req;
1465     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1466     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1467 }
1468
1469 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1470 static int fd_open(BlockDriverState *bs)
1471 {
1472     BDRVRawState *s = bs->opaque;
1473
1474     /* this is just to ensure s->fd is sane (its called by io ops) */
1475     if (s->fd >= 0)
1476         return 0;
1477     return -EIO;
1478 }
1479 #else /* !linux && !FreeBSD */
1480
1481 static int fd_open(BlockDriverState *bs)
1482 {
1483     return 0;
1484 }
1485
1486 #endif /* !linux && !FreeBSD */
1487
1488 static coroutine_fn BlockDriverAIOCB *hdev_aio_discard(BlockDriverState *bs,
1489     int64_t sector_num, int nb_sectors,
1490     BlockDriverCompletionFunc *cb, void *opaque)
1491 {
1492     BDRVRawState *s = bs->opaque;
1493
1494     if (fd_open(bs) < 0) {
1495         return NULL;
1496     }
1497     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1498                        cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1499 }
1500
1501 static int hdev_create(const char *filename, QEMUOptionParameter *options)
1502 {
1503     int fd;
1504     int ret = 0;
1505     struct stat stat_buf;
1506     int64_t total_size = 0;
1507
1508     /* Read out options */
1509     while (options && options->name) {
1510         if (!strcmp(options->name, "size")) {
1511             total_size = options->value.n / BDRV_SECTOR_SIZE;
1512         }
1513         options++;
1514     }
1515
1516     fd = qemu_open(filename, O_WRONLY | O_BINARY);
1517     if (fd < 0)
1518         return -errno;
1519
1520     if (fstat(fd, &stat_buf) < 0)
1521         ret = -errno;
1522     else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode))
1523         ret = -ENODEV;
1524     else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE)
1525         ret = -ENOSPC;
1526
1527     qemu_close(fd);
1528     return ret;
1529 }
1530
1531 static int hdev_has_zero_init(BlockDriverState *bs)
1532 {
1533     return 0;
1534 }
1535
1536 static BlockDriver bdrv_host_device = {
1537     .format_name        = "host_device",
1538     .protocol_name        = "host_device",
1539     .instance_size      = sizeof(BDRVRawState),
1540     .bdrv_probe_device  = hdev_probe_device,
1541     .bdrv_file_open     = hdev_open,
1542     .bdrv_close         = raw_close,
1543     .bdrv_reopen_prepare = raw_reopen_prepare,
1544     .bdrv_reopen_commit  = raw_reopen_commit,
1545     .bdrv_reopen_abort   = raw_reopen_abort,
1546     .bdrv_create        = hdev_create,
1547     .create_options     = raw_create_options,
1548     .bdrv_has_zero_init = hdev_has_zero_init,
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, const char *filename,
1569                        QDict *options, int flags)
1570 {
1571     BDRVRawState *s = bs->opaque;
1572     int ret;
1573
1574     s->type = FTYPE_FD;
1575
1576     /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1577     ret = raw_open_common(bs, options, flags, O_NONBLOCK);
1578     if (ret)
1579         return ret;
1580
1581     /* close fd so that we can reopen it as needed */
1582     qemu_close(s->fd);
1583     s->fd = -1;
1584     s->fd_media_changed = 1;
1585
1586     return 0;
1587 }
1588
1589 static int floppy_probe_device(const char *filename)
1590 {
1591     int fd, ret;
1592     int prio = 0;
1593     struct floppy_struct fdparam;
1594     struct stat st;
1595
1596     if (strstart(filename, "/dev/fd", NULL) &&
1597         !strstart(filename, "/dev/fdset/", NULL)) {
1598         prio = 50;
1599     }
1600
1601     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1602     if (fd < 0) {
1603         goto out;
1604     }
1605     ret = fstat(fd, &st);
1606     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1607         goto outc;
1608     }
1609
1610     /* Attempt to detect via a floppy specific ioctl */
1611     ret = ioctl(fd, FDGETPRM, &fdparam);
1612     if (ret >= 0)
1613         prio = 100;
1614
1615 outc:
1616     qemu_close(fd);
1617 out:
1618     return prio;
1619 }
1620
1621
1622 static int floppy_is_inserted(BlockDriverState *bs)
1623 {
1624     return fd_open(bs) >= 0;
1625 }
1626
1627 static int floppy_media_changed(BlockDriverState *bs)
1628 {
1629     BDRVRawState *s = bs->opaque;
1630     int ret;
1631
1632     /*
1633      * XXX: we do not have a true media changed indication.
1634      * It does not work if the floppy is changed without trying to read it.
1635      */
1636     fd_open(bs);
1637     ret = s->fd_media_changed;
1638     s->fd_media_changed = 0;
1639 #ifdef DEBUG_FLOPPY
1640     printf("Floppy changed=%d\n", ret);
1641 #endif
1642     return ret;
1643 }
1644
1645 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
1646 {
1647     BDRVRawState *s = bs->opaque;
1648     int fd;
1649
1650     if (s->fd >= 0) {
1651         qemu_close(s->fd);
1652         s->fd = -1;
1653     }
1654     fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
1655     if (fd >= 0) {
1656         if (ioctl(fd, FDEJECT, 0) < 0)
1657             perror("FDEJECT");
1658         qemu_close(fd);
1659     }
1660 }
1661
1662 static BlockDriver bdrv_host_floppy = {
1663     .format_name        = "host_floppy",
1664     .protocol_name      = "host_floppy",
1665     .instance_size      = sizeof(BDRVRawState),
1666     .bdrv_probe_device  = floppy_probe_device,
1667     .bdrv_file_open     = floppy_open,
1668     .bdrv_close         = raw_close,
1669     .bdrv_reopen_prepare = raw_reopen_prepare,
1670     .bdrv_reopen_commit  = raw_reopen_commit,
1671     .bdrv_reopen_abort   = raw_reopen_abort,
1672     .bdrv_create        = hdev_create,
1673     .create_options     = raw_create_options,
1674     .bdrv_has_zero_init = hdev_has_zero_init,
1675
1676     .bdrv_aio_readv     = raw_aio_readv,
1677     .bdrv_aio_writev    = raw_aio_writev,
1678     .bdrv_aio_flush     = raw_aio_flush,
1679
1680     .bdrv_truncate      = raw_truncate,
1681     .bdrv_getlength     = raw_getlength,
1682     .bdrv_get_allocated_file_size
1683                         = raw_get_allocated_file_size,
1684
1685     /* removable device support */
1686     .bdrv_is_inserted   = floppy_is_inserted,
1687     .bdrv_media_changed = floppy_media_changed,
1688     .bdrv_eject         = floppy_eject,
1689 };
1690
1691 static int cdrom_open(BlockDriverState *bs, const char *filename,
1692                       QDict *options, int flags)
1693 {
1694     BDRVRawState *s = bs->opaque;
1695
1696     s->type = FTYPE_CD;
1697
1698     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
1699     return raw_open_common(bs, options, flags, O_NONBLOCK);
1700 }
1701
1702 static int cdrom_probe_device(const char *filename)
1703 {
1704     int fd, ret;
1705     int prio = 0;
1706     struct stat st;
1707
1708     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1709     if (fd < 0) {
1710         goto out;
1711     }
1712     ret = fstat(fd, &st);
1713     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1714         goto outc;
1715     }
1716
1717     /* Attempt to detect via a CDROM specific ioctl */
1718     ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1719     if (ret >= 0)
1720         prio = 100;
1721
1722 outc:
1723     qemu_close(fd);
1724 out:
1725     return prio;
1726 }
1727
1728 static int cdrom_is_inserted(BlockDriverState *bs)
1729 {
1730     BDRVRawState *s = bs->opaque;
1731     int ret;
1732
1733     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1734     if (ret == CDS_DISC_OK)
1735         return 1;
1736     return 0;
1737 }
1738
1739 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1740 {
1741     BDRVRawState *s = bs->opaque;
1742
1743     if (eject_flag) {
1744         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
1745             perror("CDROMEJECT");
1746     } else {
1747         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
1748             perror("CDROMEJECT");
1749     }
1750 }
1751
1752 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1753 {
1754     BDRVRawState *s = bs->opaque;
1755
1756     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
1757         /*
1758          * Note: an error can happen if the distribution automatically
1759          * mounts the CD-ROM
1760          */
1761         /* perror("CDROM_LOCKDOOR"); */
1762     }
1763 }
1764
1765 static BlockDriver bdrv_host_cdrom = {
1766     .format_name        = "host_cdrom",
1767     .protocol_name      = "host_cdrom",
1768     .instance_size      = sizeof(BDRVRawState),
1769     .bdrv_probe_device  = cdrom_probe_device,
1770     .bdrv_file_open     = cdrom_open,
1771     .bdrv_close         = raw_close,
1772     .bdrv_reopen_prepare = raw_reopen_prepare,
1773     .bdrv_reopen_commit  = raw_reopen_commit,
1774     .bdrv_reopen_abort   = raw_reopen_abort,
1775     .bdrv_create        = hdev_create,
1776     .create_options     = raw_create_options,
1777     .bdrv_has_zero_init = hdev_has_zero_init,
1778
1779     .bdrv_aio_readv     = raw_aio_readv,
1780     .bdrv_aio_writev    = raw_aio_writev,
1781     .bdrv_aio_flush     = raw_aio_flush,
1782
1783     .bdrv_truncate      = raw_truncate,
1784     .bdrv_getlength     = raw_getlength,
1785     .bdrv_get_allocated_file_size
1786                         = raw_get_allocated_file_size,
1787
1788     /* removable device support */
1789     .bdrv_is_inserted   = cdrom_is_inserted,
1790     .bdrv_eject         = cdrom_eject,
1791     .bdrv_lock_medium   = cdrom_lock_medium,
1792
1793     /* generic scsi device */
1794     .bdrv_ioctl         = hdev_ioctl,
1795     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1796 };
1797 #endif /* __linux__ */
1798
1799 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1800 static int cdrom_open(BlockDriverState *bs, const char *filename,
1801                       QDict *options, int flags)
1802 {
1803     BDRVRawState *s = bs->opaque;
1804     int ret;
1805
1806     s->type = FTYPE_CD;
1807
1808     ret = raw_open_common(bs, options, flags, 0);
1809     if (ret)
1810         return ret;
1811
1812     /* make sure the door isn't locked at this time */
1813     ioctl(s->fd, CDIOCALLOW);
1814     return 0;
1815 }
1816
1817 static int cdrom_probe_device(const char *filename)
1818 {
1819     if (strstart(filename, "/dev/cd", NULL) ||
1820             strstart(filename, "/dev/acd", NULL))
1821         return 100;
1822     return 0;
1823 }
1824
1825 static int cdrom_reopen(BlockDriverState *bs)
1826 {
1827     BDRVRawState *s = bs->opaque;
1828     int fd;
1829
1830     /*
1831      * Force reread of possibly changed/newly loaded disc,
1832      * FreeBSD seems to not notice sometimes...
1833      */
1834     if (s->fd >= 0)
1835         qemu_close(s->fd);
1836     fd = qemu_open(bs->filename, s->open_flags, 0644);
1837     if (fd < 0) {
1838         s->fd = -1;
1839         return -EIO;
1840     }
1841     s->fd = fd;
1842
1843     /* make sure the door isn't locked at this time */
1844     ioctl(s->fd, CDIOCALLOW);
1845     return 0;
1846 }
1847
1848 static int cdrom_is_inserted(BlockDriverState *bs)
1849 {
1850     return raw_getlength(bs) > 0;
1851 }
1852
1853 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1854 {
1855     BDRVRawState *s = bs->opaque;
1856
1857     if (s->fd < 0)
1858         return;
1859
1860     (void) ioctl(s->fd, CDIOCALLOW);
1861
1862     if (eject_flag) {
1863         if (ioctl(s->fd, CDIOCEJECT) < 0)
1864             perror("CDIOCEJECT");
1865     } else {
1866         if (ioctl(s->fd, CDIOCCLOSE) < 0)
1867             perror("CDIOCCLOSE");
1868     }
1869
1870     cdrom_reopen(bs);
1871 }
1872
1873 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1874 {
1875     BDRVRawState *s = bs->opaque;
1876
1877     if (s->fd < 0)
1878         return;
1879     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1880         /*
1881          * Note: an error can happen if the distribution automatically
1882          * mounts the CD-ROM
1883          */
1884         /* perror("CDROM_LOCKDOOR"); */
1885     }
1886 }
1887
1888 static BlockDriver bdrv_host_cdrom = {
1889     .format_name        = "host_cdrom",
1890     .protocol_name      = "host_cdrom",
1891     .instance_size      = sizeof(BDRVRawState),
1892     .bdrv_probe_device  = cdrom_probe_device,
1893     .bdrv_file_open     = cdrom_open,
1894     .bdrv_close         = raw_close,
1895     .bdrv_reopen_prepare = raw_reopen_prepare,
1896     .bdrv_reopen_commit  = raw_reopen_commit,
1897     .bdrv_reopen_abort   = raw_reopen_abort,
1898     .bdrv_create        = hdev_create,
1899     .create_options     = raw_create_options,
1900     .bdrv_has_zero_init = hdev_has_zero_init,
1901
1902     .bdrv_aio_readv     = raw_aio_readv,
1903     .bdrv_aio_writev    = raw_aio_writev,
1904     .bdrv_aio_flush     = raw_aio_flush,
1905
1906     .bdrv_truncate      = raw_truncate,
1907     .bdrv_getlength     = raw_getlength,
1908     .bdrv_get_allocated_file_size
1909                         = raw_get_allocated_file_size,
1910
1911     /* removable device support */
1912     .bdrv_is_inserted   = cdrom_is_inserted,
1913     .bdrv_eject         = cdrom_eject,
1914     .bdrv_lock_medium   = cdrom_lock_medium,
1915 };
1916 #endif /* __FreeBSD__ */
1917
1918 #ifdef CONFIG_LINUX_AIO
1919 /**
1920  * Return the file descriptor for Linux AIO
1921  *
1922  * This function is a layering violation and should be removed when it becomes
1923  * possible to call the block layer outside the global mutex.  It allows the
1924  * caller to hijack the file descriptor so I/O can be performed outside the
1925  * block layer.
1926  */
1927 int raw_get_aio_fd(BlockDriverState *bs)
1928 {
1929     BDRVRawState *s;
1930
1931     if (!bs->drv) {
1932         return -ENOMEDIUM;
1933     }
1934
1935     if (bs->drv == bdrv_find_format("raw")) {
1936         bs = bs->file;
1937     }
1938
1939     /* raw-posix has several protocols so just check for raw_aio_readv */
1940     if (bs->drv->bdrv_aio_readv != raw_aio_readv) {
1941         return -ENOTSUP;
1942     }
1943
1944     s = bs->opaque;
1945     if (!s->use_aio) {
1946         return -ENOTSUP;
1947     }
1948     return s->fd;
1949 }
1950 #endif /* CONFIG_LINUX_AIO */
1951
1952 static void bdrv_file_init(void)
1953 {
1954     /*
1955      * Register all the drivers.  Note that order is important, the driver
1956      * registered last will get probed first.
1957      */
1958     bdrv_register(&bdrv_file);
1959     bdrv_register(&bdrv_host_device);
1960 #ifdef __linux__
1961     bdrv_register(&bdrv_host_floppy);
1962     bdrv_register(&bdrv_host_cdrom);
1963 #endif
1964 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1965     bdrv_register(&bdrv_host_cdrom);
1966 #endif
1967 }
1968
1969 block_init(bdrv_file_init);