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 int coroutine_fn raw_co_is_allocated(BlockDriverState *bs,
1088                                             int64_t sector_num,
1089                                             int nb_sectors, int *pnum)
1090 {
1091     off_t start, data, hole;
1092     int ret;
1093
1094     ret = fd_open(bs);
1095     if (ret < 0) {
1096         return ret;
1097     }
1098
1099     start = sector_num * BDRV_SECTOR_SIZE;
1100
1101 #ifdef CONFIG_FIEMAP
1102
1103     BDRVRawState *s = bs->opaque;
1104     struct {
1105         struct fiemap fm;
1106         struct fiemap_extent fe;
1107     } f;
1108
1109     f.fm.fm_start = start;
1110     f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE;
1111     f.fm.fm_flags = 0;
1112     f.fm.fm_extent_count = 1;
1113     f.fm.fm_reserved = 0;
1114     if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) {
1115         /* Assume everything is allocated.  */
1116         *pnum = nb_sectors;
1117         return 1;
1118     }
1119
1120     if (f.fm.fm_mapped_extents == 0) {
1121         /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1122          * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1123          */
1124         off_t length = lseek(s->fd, 0, SEEK_END);
1125         hole = f.fm.fm_start;
1126         data = MIN(f.fm.fm_start + f.fm.fm_length, length);
1127     } else {
1128         data = f.fe.fe_logical;
1129         hole = f.fe.fe_logical + f.fe.fe_length;
1130     }
1131
1132 #elif defined SEEK_HOLE && defined SEEK_DATA
1133
1134     BDRVRawState *s = bs->opaque;
1135
1136     hole = lseek(s->fd, start, SEEK_HOLE);
1137     if (hole == -1) {
1138         /* -ENXIO indicates that sector_num was past the end of the file.
1139          * There is a virtual hole there.  */
1140         assert(errno != -ENXIO);
1141
1142         /* Most likely EINVAL.  Assume everything is allocated.  */
1143         *pnum = nb_sectors;
1144         return 1;
1145     }
1146
1147     if (hole > start) {
1148         data = start;
1149     } else {
1150         /* On a hole.  We need another syscall to find its end.  */
1151         data = lseek(s->fd, start, SEEK_DATA);
1152         if (data == -1) {
1153             data = lseek(s->fd, 0, SEEK_END);
1154         }
1155     }
1156 #else
1157     *pnum = nb_sectors;
1158     return 1;
1159 #endif
1160
1161     if (data <= start) {
1162         /* On a data extent, compute sectors to the end of the extent.  */
1163         *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1164         return 1;
1165     } else {
1166         /* On a hole, compute sectors to the beginning of the next extent.  */
1167         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1168         return 0;
1169     }
1170 }
1171
1172 static coroutine_fn BlockDriverAIOCB *raw_aio_discard(BlockDriverState *bs,
1173     int64_t sector_num, int nb_sectors,
1174     BlockDriverCompletionFunc *cb, void *opaque)
1175 {
1176     BDRVRawState *s = bs->opaque;
1177
1178     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1179                        cb, opaque, QEMU_AIO_DISCARD);
1180 }
1181
1182 static QEMUOptionParameter raw_create_options[] = {
1183     {
1184         .name = BLOCK_OPT_SIZE,
1185         .type = OPT_SIZE,
1186         .help = "Virtual disk size"
1187     },
1188     { NULL }
1189 };
1190
1191 static BlockDriver bdrv_file = {
1192     .format_name = "file",
1193     .protocol_name = "file",
1194     .instance_size = sizeof(BDRVRawState),
1195     .bdrv_probe = NULL, /* no probe for protocols */
1196     .bdrv_file_open = raw_open,
1197     .bdrv_reopen_prepare = raw_reopen_prepare,
1198     .bdrv_reopen_commit = raw_reopen_commit,
1199     .bdrv_reopen_abort = raw_reopen_abort,
1200     .bdrv_close = raw_close,
1201     .bdrv_create = raw_create,
1202     .bdrv_has_zero_init = bdrv_has_zero_init_1,
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, QDict *options, int flags)
1329 {
1330     BDRVRawState *s = bs->opaque;
1331     int ret;
1332     const char *filename = qdict_get_str(options, "filename");
1333
1334 #if defined(__APPLE__) && defined(__MACH__)
1335     if (strstart(filename, "/dev/cdrom", NULL)) {
1336         kern_return_t kernResult;
1337         io_iterator_t mediaIterator;
1338         char bsdPath[ MAXPATHLEN ];
1339         int fd;
1340
1341         kernResult = FindEjectableCDMedia( &mediaIterator );
1342         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1343
1344         if ( bsdPath[ 0 ] != '\0' ) {
1345             strcat(bsdPath,"s0");
1346             /* some CDs don't have a partition 0 */
1347             fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1348             if (fd < 0) {
1349                 bsdPath[strlen(bsdPath)-1] = '1';
1350             } else {
1351                 qemu_close(fd);
1352             }
1353             filename = bsdPath;
1354             qdict_put(options, "filename", qstring_from_str(filename));
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 BlockDriver bdrv_host_device = {
1532     .format_name        = "host_device",
1533     .protocol_name        = "host_device",
1534     .instance_size      = sizeof(BDRVRawState),
1535     .bdrv_probe_device  = hdev_probe_device,
1536     .bdrv_file_open     = hdev_open,
1537     .bdrv_close         = raw_close,
1538     .bdrv_reopen_prepare = raw_reopen_prepare,
1539     .bdrv_reopen_commit  = raw_reopen_commit,
1540     .bdrv_reopen_abort   = raw_reopen_abort,
1541     .bdrv_create        = hdev_create,
1542     .create_options     = raw_create_options,
1543
1544     .bdrv_aio_readv     = raw_aio_readv,
1545     .bdrv_aio_writev    = raw_aio_writev,
1546     .bdrv_aio_flush     = raw_aio_flush,
1547     .bdrv_aio_discard   = hdev_aio_discard,
1548
1549     .bdrv_truncate      = raw_truncate,
1550     .bdrv_getlength     = raw_getlength,
1551     .bdrv_get_allocated_file_size
1552                         = raw_get_allocated_file_size,
1553
1554     /* generic scsi device */
1555 #ifdef __linux__
1556     .bdrv_ioctl         = hdev_ioctl,
1557     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1558 #endif
1559 };
1560
1561 #ifdef __linux__
1562 static int floppy_open(BlockDriverState *bs, QDict *options, int flags)
1563 {
1564     BDRVRawState *s = bs->opaque;
1565     int ret;
1566
1567     s->type = FTYPE_FD;
1568
1569     /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1570     ret = raw_open_common(bs, options, flags, O_NONBLOCK);
1571     if (ret)
1572         return ret;
1573
1574     /* close fd so that we can reopen it as needed */
1575     qemu_close(s->fd);
1576     s->fd = -1;
1577     s->fd_media_changed = 1;
1578
1579     return 0;
1580 }
1581
1582 static int floppy_probe_device(const char *filename)
1583 {
1584     int fd, ret;
1585     int prio = 0;
1586     struct floppy_struct fdparam;
1587     struct stat st;
1588
1589     if (strstart(filename, "/dev/fd", NULL) &&
1590         !strstart(filename, "/dev/fdset/", NULL)) {
1591         prio = 50;
1592     }
1593
1594     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1595     if (fd < 0) {
1596         goto out;
1597     }
1598     ret = fstat(fd, &st);
1599     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1600         goto outc;
1601     }
1602
1603     /* Attempt to detect via a floppy specific ioctl */
1604     ret = ioctl(fd, FDGETPRM, &fdparam);
1605     if (ret >= 0)
1606         prio = 100;
1607
1608 outc:
1609     qemu_close(fd);
1610 out:
1611     return prio;
1612 }
1613
1614
1615 static int floppy_is_inserted(BlockDriverState *bs)
1616 {
1617     return fd_open(bs) >= 0;
1618 }
1619
1620 static int floppy_media_changed(BlockDriverState *bs)
1621 {
1622     BDRVRawState *s = bs->opaque;
1623     int ret;
1624
1625     /*
1626      * XXX: we do not have a true media changed indication.
1627      * It does not work if the floppy is changed without trying to read it.
1628      */
1629     fd_open(bs);
1630     ret = s->fd_media_changed;
1631     s->fd_media_changed = 0;
1632 #ifdef DEBUG_FLOPPY
1633     printf("Floppy changed=%d\n", ret);
1634 #endif
1635     return ret;
1636 }
1637
1638 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
1639 {
1640     BDRVRawState *s = bs->opaque;
1641     int fd;
1642
1643     if (s->fd >= 0) {
1644         qemu_close(s->fd);
1645         s->fd = -1;
1646     }
1647     fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
1648     if (fd >= 0) {
1649         if (ioctl(fd, FDEJECT, 0) < 0)
1650             perror("FDEJECT");
1651         qemu_close(fd);
1652     }
1653 }
1654
1655 static BlockDriver bdrv_host_floppy = {
1656     .format_name        = "host_floppy",
1657     .protocol_name      = "host_floppy",
1658     .instance_size      = sizeof(BDRVRawState),
1659     .bdrv_probe_device  = floppy_probe_device,
1660     .bdrv_file_open     = floppy_open,
1661     .bdrv_close         = raw_close,
1662     .bdrv_reopen_prepare = raw_reopen_prepare,
1663     .bdrv_reopen_commit  = raw_reopen_commit,
1664     .bdrv_reopen_abort   = raw_reopen_abort,
1665     .bdrv_create        = hdev_create,
1666     .create_options     = raw_create_options,
1667
1668     .bdrv_aio_readv     = raw_aio_readv,
1669     .bdrv_aio_writev    = raw_aio_writev,
1670     .bdrv_aio_flush     = raw_aio_flush,
1671
1672     .bdrv_truncate      = raw_truncate,
1673     .bdrv_getlength     = raw_getlength,
1674     .bdrv_get_allocated_file_size
1675                         = raw_get_allocated_file_size,
1676
1677     /* removable device support */
1678     .bdrv_is_inserted   = floppy_is_inserted,
1679     .bdrv_media_changed = floppy_media_changed,
1680     .bdrv_eject         = floppy_eject,
1681 };
1682
1683 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags)
1684 {
1685     BDRVRawState *s = bs->opaque;
1686
1687     s->type = FTYPE_CD;
1688
1689     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
1690     return raw_open_common(bs, options, flags, O_NONBLOCK);
1691 }
1692
1693 static int cdrom_probe_device(const char *filename)
1694 {
1695     int fd, ret;
1696     int prio = 0;
1697     struct stat st;
1698
1699     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1700     if (fd < 0) {
1701         goto out;
1702     }
1703     ret = fstat(fd, &st);
1704     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1705         goto outc;
1706     }
1707
1708     /* Attempt to detect via a CDROM specific ioctl */
1709     ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1710     if (ret >= 0)
1711         prio = 100;
1712
1713 outc:
1714     qemu_close(fd);
1715 out:
1716     return prio;
1717 }
1718
1719 static int cdrom_is_inserted(BlockDriverState *bs)
1720 {
1721     BDRVRawState *s = bs->opaque;
1722     int ret;
1723
1724     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1725     if (ret == CDS_DISC_OK)
1726         return 1;
1727     return 0;
1728 }
1729
1730 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1731 {
1732     BDRVRawState *s = bs->opaque;
1733
1734     if (eject_flag) {
1735         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
1736             perror("CDROMEJECT");
1737     } else {
1738         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
1739             perror("CDROMEJECT");
1740     }
1741 }
1742
1743 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1744 {
1745     BDRVRawState *s = bs->opaque;
1746
1747     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
1748         /*
1749          * Note: an error can happen if the distribution automatically
1750          * mounts the CD-ROM
1751          */
1752         /* perror("CDROM_LOCKDOOR"); */
1753     }
1754 }
1755
1756 static BlockDriver bdrv_host_cdrom = {
1757     .format_name        = "host_cdrom",
1758     .protocol_name      = "host_cdrom",
1759     .instance_size      = sizeof(BDRVRawState),
1760     .bdrv_probe_device  = cdrom_probe_device,
1761     .bdrv_file_open     = cdrom_open,
1762     .bdrv_close         = raw_close,
1763     .bdrv_reopen_prepare = raw_reopen_prepare,
1764     .bdrv_reopen_commit  = raw_reopen_commit,
1765     .bdrv_reopen_abort   = raw_reopen_abort,
1766     .bdrv_create        = hdev_create,
1767     .create_options     = raw_create_options,
1768
1769     .bdrv_aio_readv     = raw_aio_readv,
1770     .bdrv_aio_writev    = raw_aio_writev,
1771     .bdrv_aio_flush     = raw_aio_flush,
1772
1773     .bdrv_truncate      = raw_truncate,
1774     .bdrv_getlength     = raw_getlength,
1775     .bdrv_get_allocated_file_size
1776                         = raw_get_allocated_file_size,
1777
1778     /* removable device support */
1779     .bdrv_is_inserted   = cdrom_is_inserted,
1780     .bdrv_eject         = cdrom_eject,
1781     .bdrv_lock_medium   = cdrom_lock_medium,
1782
1783     /* generic scsi device */
1784     .bdrv_ioctl         = hdev_ioctl,
1785     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1786 };
1787 #endif /* __linux__ */
1788
1789 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1790 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags)
1791 {
1792     BDRVRawState *s = bs->opaque;
1793     int ret;
1794
1795     s->type = FTYPE_CD;
1796
1797     ret = raw_open_common(bs, options, flags, 0);
1798     if (ret)
1799         return ret;
1800
1801     /* make sure the door isn't locked at this time */
1802     ioctl(s->fd, CDIOCALLOW);
1803     return 0;
1804 }
1805
1806 static int cdrom_probe_device(const char *filename)
1807 {
1808     if (strstart(filename, "/dev/cd", NULL) ||
1809             strstart(filename, "/dev/acd", NULL))
1810         return 100;
1811     return 0;
1812 }
1813
1814 static int cdrom_reopen(BlockDriverState *bs)
1815 {
1816     BDRVRawState *s = bs->opaque;
1817     int fd;
1818
1819     /*
1820      * Force reread of possibly changed/newly loaded disc,
1821      * FreeBSD seems to not notice sometimes...
1822      */
1823     if (s->fd >= 0)
1824         qemu_close(s->fd);
1825     fd = qemu_open(bs->filename, s->open_flags, 0644);
1826     if (fd < 0) {
1827         s->fd = -1;
1828         return -EIO;
1829     }
1830     s->fd = fd;
1831
1832     /* make sure the door isn't locked at this time */
1833     ioctl(s->fd, CDIOCALLOW);
1834     return 0;
1835 }
1836
1837 static int cdrom_is_inserted(BlockDriverState *bs)
1838 {
1839     return raw_getlength(bs) > 0;
1840 }
1841
1842 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1843 {
1844     BDRVRawState *s = bs->opaque;
1845
1846     if (s->fd < 0)
1847         return;
1848
1849     (void) ioctl(s->fd, CDIOCALLOW);
1850
1851     if (eject_flag) {
1852         if (ioctl(s->fd, CDIOCEJECT) < 0)
1853             perror("CDIOCEJECT");
1854     } else {
1855         if (ioctl(s->fd, CDIOCCLOSE) < 0)
1856             perror("CDIOCCLOSE");
1857     }
1858
1859     cdrom_reopen(bs);
1860 }
1861
1862 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1863 {
1864     BDRVRawState *s = bs->opaque;
1865
1866     if (s->fd < 0)
1867         return;
1868     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1869         /*
1870          * Note: an error can happen if the distribution automatically
1871          * mounts the CD-ROM
1872          */
1873         /* perror("CDROM_LOCKDOOR"); */
1874     }
1875 }
1876
1877 static BlockDriver bdrv_host_cdrom = {
1878     .format_name        = "host_cdrom",
1879     .protocol_name      = "host_cdrom",
1880     .instance_size      = sizeof(BDRVRawState),
1881     .bdrv_probe_device  = cdrom_probe_device,
1882     .bdrv_file_open     = cdrom_open,
1883     .bdrv_close         = raw_close,
1884     .bdrv_reopen_prepare = raw_reopen_prepare,
1885     .bdrv_reopen_commit  = raw_reopen_commit,
1886     .bdrv_reopen_abort   = raw_reopen_abort,
1887     .bdrv_create        = hdev_create,
1888     .create_options     = raw_create_options,
1889
1890     .bdrv_aio_readv     = raw_aio_readv,
1891     .bdrv_aio_writev    = raw_aio_writev,
1892     .bdrv_aio_flush     = raw_aio_flush,
1893
1894     .bdrv_truncate      = raw_truncate,
1895     .bdrv_getlength     = raw_getlength,
1896     .bdrv_get_allocated_file_size
1897                         = raw_get_allocated_file_size,
1898
1899     /* removable device support */
1900     .bdrv_is_inserted   = cdrom_is_inserted,
1901     .bdrv_eject         = cdrom_eject,
1902     .bdrv_lock_medium   = cdrom_lock_medium,
1903 };
1904 #endif /* __FreeBSD__ */
1905
1906 #ifdef CONFIG_LINUX_AIO
1907 /**
1908  * Return the file descriptor for Linux AIO
1909  *
1910  * This function is a layering violation and should be removed when it becomes
1911  * possible to call the block layer outside the global mutex.  It allows the
1912  * caller to hijack the file descriptor so I/O can be performed outside the
1913  * block layer.
1914  */
1915 int raw_get_aio_fd(BlockDriverState *bs)
1916 {
1917     BDRVRawState *s;
1918
1919     if (!bs->drv) {
1920         return -ENOMEDIUM;
1921     }
1922
1923     if (bs->drv == bdrv_find_format("raw")) {
1924         bs = bs->file;
1925     }
1926
1927     /* raw-posix has several protocols so just check for raw_aio_readv */
1928     if (bs->drv->bdrv_aio_readv != raw_aio_readv) {
1929         return -ENOTSUP;
1930     }
1931
1932     s = bs->opaque;
1933     if (!s->use_aio) {
1934         return -ENOTSUP;
1935     }
1936     return s->fd;
1937 }
1938 #endif /* CONFIG_LINUX_AIO */
1939
1940 static void bdrv_file_init(void)
1941 {
1942     /*
1943      * Register all the drivers.  Note that order is important, the driver
1944      * registered last will get probed first.
1945      */
1946     bdrv_register(&bdrv_file);
1947     bdrv_register(&bdrv_host_device);
1948 #ifdef __linux__
1949     bdrv_register(&bdrv_host_floppy);
1950     bdrv_register(&bdrv_host_cdrom);
1951 #endif
1952 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1953     bdrv_register(&bdrv_host_cdrom);
1954 #endif
1955 }
1956
1957 block_init(bdrv_file_init);