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, Error **errp)
 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         error_propagate(errp, local_err);
 291         ret = -EINVAL;
 292         goto fail;
 293     }
 294
 295     filename = qemu_opt_get(opts, "filename");
 296
 297     ret = raw_normalize_devicepath(&filename);
 298     if (ret != 0) {
 299         error_setg_errno(errp, -ret, "Could not normalize device path");
 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         error_setg_errno(errp, -ret, "Could not set AIO state");
 322         goto fail;
 323     }
 324 #endif
 325
 326     s->has_discard = 1;
 327 #ifdef CONFIG_XFS
 328     if (platform_test_xfs_fd(s->fd)) {
 329         s->is_xfs = 1;
 330     }
 331 #endif
 332
 333     ret = 0;
 334 fail:
 335     qemu_opts_del(opts);
 336     return ret;
 337 }
 338
 339 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
 340                     Error **errp)
 341 {
 342     BDRVRawState *s = bs->opaque;
 343     Error *local_err = NULL;
 344     int ret;
 345
 346     s->type = FTYPE_FILE;
 347     ret = raw_open_common(bs, options, flags, 0, &local_err);
 348     if (error_is_set(&local_err)) {
 349         error_propagate(errp, local_err);
 350     }
 351     return ret;
 352 }
 353
 354 static int raw_reopen_prepare(BDRVReopenState *state,
 355                               BlockReopenQueue *queue, Error **errp)
 356 {
 357     BDRVRawState *s;
 358     BDRVRawReopenState *raw_s;
 359     int ret = 0;
 360
 361     assert(state != NULL);
 362     assert(state->bs != NULL);
 363
 364     s = state->bs->opaque;
 365
 366     state->opaque = g_malloc0(sizeof(BDRVRawReopenState));
 367     raw_s = state->opaque;
 368
 369 #ifdef CONFIG_LINUX_AIO
 370     raw_s->use_aio = s->use_aio;
 371
 372     /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
 373      * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
 374      * won't override aio_ctx if aio_ctx is non-NULL */
 375     if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
 376         error_setg(errp, "Could not set AIO state");
 377         return -1;
 378     }
 379 #endif
 380
 381     if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
 382         raw_s->open_flags |= O_NONBLOCK;
 383     }
 384
 385     raw_parse_flags(state->flags, &raw_s->open_flags);
 386
 387     raw_s->fd = -1;
 388
 389     int fcntl_flags = O_APPEND | O_NONBLOCK;
 390 #ifdef O_NOATIME
 391     fcntl_flags |= O_NOATIME;
 392 #endif
 393
 394 #ifdef O_ASYNC
 395     /* Not all operating systems have O_ASYNC, and those that don't
 396      * will not let us track the state into raw_s->open_flags (typically
 397      * you achieve the same effect with an ioctl, for example I_SETSIG
 398      * on Solaris). But we do not use O_ASYNC, so that's fine.
 399      */
 400     assert((s->open_flags & O_ASYNC) == 0);
 401 #endif
 402
 403     if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
 404         /* dup the original fd */
 405         /* TODO: use qemu fcntl wrapper */
 406 #ifdef F_DUPFD_CLOEXEC
 407         raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
 408 #else
 409         raw_s->fd = dup(s->fd);
 410         if (raw_s->fd != -1) {
 411             qemu_set_cloexec(raw_s->fd);
 412         }
 413 #endif
 414         if (raw_s->fd >= 0) {
 415             ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
 416             if (ret) {
 417                 qemu_close(raw_s->fd);
 418                 raw_s->fd = -1;
 419             }
 420         }
 421     }
 422
 423     /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
 424     if (raw_s->fd == -1) {
 425         assert(!(raw_s->open_flags & O_CREAT));
 426         raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
 427         if (raw_s->fd == -1) {
 428             error_setg_errno(errp, errno, "Could not reopen file");
 429             ret = -1;
 430         }
 431     }
 432     return ret;
 433 }
 434
 435
 436 static void raw_reopen_commit(BDRVReopenState *state)
 437 {
 438     BDRVRawReopenState *raw_s = state->opaque;
 439     BDRVRawState *s = state->bs->opaque;
 440
 441     s->open_flags = raw_s->open_flags;
 442
 443     qemu_close(s->fd);
 444     s->fd = raw_s->fd;
 445 #ifdef CONFIG_LINUX_AIO
 446     s->use_aio = raw_s->use_aio;
 447 #endif
 448
 449     g_free(state->opaque);
 450     state->opaque = NULL;
 451 }
 452
 453
 454 static void raw_reopen_abort(BDRVReopenState *state)
 455 {
 456     BDRVRawReopenState *raw_s = state->opaque;
 457
 458      /* nothing to do if NULL, we didn't get far enough */
 459     if (raw_s == NULL) {
 460         return;
 461     }
 462
 463     if (raw_s->fd >= 0) {
 464         qemu_close(raw_s->fd);
 465         raw_s->fd = -1;
 466     }
 467     g_free(state->opaque);
 468     state->opaque = NULL;
 469 }
 470
 471
 472 /* XXX: use host sector size if necessary with:
 473 #ifdef DIOCGSECTORSIZE
 474         {
 475             unsigned int sectorsize = 512;
 476             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 477                 sectorsize > bufsize)
 478                 bufsize = sectorsize;
 479         }
 480 #endif
 481 #ifdef CONFIG_COCOA
 482         uint32_t blockSize = 512;
 483         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 484             bufsize = blockSize;
 485         }
 486 #endif
 487 */
 488
 489 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
 490 {
 491     int ret;
 492
 493     ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
 494     if (ret == -1) {
 495         return -errno;
 496     }
 497
 498     return 0;
 499 }
 500
 501 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
 502 {
 503     int ret;
 504
 505     ret = qemu_fdatasync(aiocb->aio_fildes);
 506     if (ret == -1) {
 507         return -errno;
 508     }
 509     return 0;
 510 }
 511
 512 #ifdef CONFIG_PREADV
 513
 514 static bool preadv_present = true;
 515
 516 static ssize_t
 517 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 518 {
 519     return preadv(fd, iov, nr_iov, offset);
 520 }
 521
 522 static ssize_t
 523 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 524 {
 525     return pwritev(fd, iov, nr_iov, offset);
 526 }
 527
 528 #else
 529
 530 static bool preadv_present = false;
 531
 532 static ssize_t
 533 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 534 {
 535     return -ENOSYS;
 536 }
 537
 538 static ssize_t
 539 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 540 {
 541     return -ENOSYS;
 542 }
 543
 544 #endif
 545
 546 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
 547 {
 548     ssize_t len;
 549
 550     do {
 551         if (aiocb->aio_type & QEMU_AIO_WRITE)
 552             len = qemu_pwritev(aiocb->aio_fildes,
 553                                aiocb->aio_iov,
 554                                aiocb->aio_niov,
 555                                aiocb->aio_offset);
 556          else
 557             len = qemu_preadv(aiocb->aio_fildes,
 558                               aiocb->aio_iov,
 559                               aiocb->aio_niov,
 560                               aiocb->aio_offset);
 561     } while (len == -1 && errno == EINTR);
 562
 563     if (len == -1) {
 564         return -errno;
 565     }
 566     return len;
 567 }
 568
 569 /*
 570  * Read/writes the data to/from a given linear buffer.
 571  *
 572  * Returns the number of bytes handles or -errno in case of an error. Short
 573  * reads are only returned if the end of the file is reached.
 574  */
 575 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
 576 {
 577     ssize_t offset = 0;
 578     ssize_t len;
 579
 580     while (offset < aiocb->aio_nbytes) {
 581         if (aiocb->aio_type & QEMU_AIO_WRITE) {
 582             len = pwrite(aiocb->aio_fildes,
 583                          (const char *)buf + offset,
 584                          aiocb->aio_nbytes - offset,
 585                          aiocb->aio_offset + offset);
 586         } else {
 587             len = pread(aiocb->aio_fildes,
 588                         buf + offset,
 589                         aiocb->aio_nbytes - offset,
 590                         aiocb->aio_offset + offset);
 591         }
 592         if (len == -1 && errno == EINTR) {
 593             continue;
 594         } else if (len == -1) {
 595             offset = -errno;
 596             break;
 597         } else if (len == 0) {
 598             break;
 599         }
 600         offset += len;
 601     }
 602
 603     return offset;
 604 }
 605
 606 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
 607 {
 608     ssize_t nbytes;
 609     char *buf;
 610
 611     if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
 612         /*
 613          * If there is just a single buffer, and it is properly aligned
 614          * we can just use plain pread/pwrite without any problems.
 615          */
 616         if (aiocb->aio_niov == 1) {
 617              return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
 618         }
 619         /*
 620          * We have more than one iovec, and all are properly aligned.
 621          *
 622          * Try preadv/pwritev first and fall back to linearizing the
 623          * buffer if it's not supported.
 624          */
 625         if (preadv_present) {
 626             nbytes = handle_aiocb_rw_vector(aiocb);
 627             if (nbytes == aiocb->aio_nbytes ||
 628                 (nbytes < 0 && nbytes != -ENOSYS)) {
 629                 return nbytes;
 630             }
 631             preadv_present = false;
 632         }
 633
 634         /*
 635          * XXX(hch): short read/write.  no easy way to handle the reminder
 636          * using these interfaces.  For now retry using plain
 637          * pread/pwrite?
 638          */
 639     }
 640
 641     /*
 642      * Ok, we have to do it the hard way, copy all segments into
 643      * a single aligned buffer.
 644      */
 645     buf = qemu_blockalign(aiocb->bs, aiocb->aio_nbytes);
 646     if (aiocb->aio_type & QEMU_AIO_WRITE) {
 647         char *p = buf;
 648         int i;
 649
 650         for (i = 0; i < aiocb->aio_niov; ++i) {
 651             memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
 652             p += aiocb->aio_iov[i].iov_len;
 653         }
 654     }
 655
 656     nbytes = handle_aiocb_rw_linear(aiocb, buf);
 657     if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
 658         char *p = buf;
 659         size_t count = aiocb->aio_nbytes, copy;
 660         int i;
 661
 662         for (i = 0; i < aiocb->aio_niov && count; ++i) {
 663             copy = count;
 664             if (copy > aiocb->aio_iov[i].iov_len) {
 665                 copy = aiocb->aio_iov[i].iov_len;
 666             }
 667             memcpy(aiocb->aio_iov[i].iov_base, p, copy);
 668             p     += copy;
 669             count -= copy;
 670         }
 671     }
 672     qemu_vfree(buf);
 673
 674     return nbytes;
 675 }
 676
 677 #ifdef CONFIG_XFS
 678 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
 679 {
 680     struct xfs_flock64 fl;
 681
 682     memset(&fl, 0, sizeof(fl));
 683     fl.l_whence = SEEK_SET;
 684     fl.l_start = offset;
 685     fl.l_len = bytes;
 686
 687     if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
 688         DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
 689         return -errno;
 690     }
 691
 692     return 0;
 693 }
 694 #endif
 695
 696 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
 697 {
 698     int ret = -EOPNOTSUPP;
 699     BDRVRawState *s = aiocb->bs->opaque;
 700
 701     if (s->has_discard == 0) {
 702         return 0;
 703     }
 704
 705     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
 706 #ifdef BLKDISCARD
 707         do {
 708             uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
 709             if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
 710                 return 0;
 711             }
 712         } while (errno == EINTR);
 713
 714         ret = -errno;
 715 #endif
 716     } else {
 717 #ifdef CONFIG_XFS
 718         if (s->is_xfs) {
 719             return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
 720         }
 721 #endif
 722
 723 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
 724         do {
 725             if (fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
 726                           aiocb->aio_offset, aiocb->aio_nbytes) == 0) {
 727                 return 0;
 728             }
 729         } while (errno == EINTR);
 730
 731         ret = -errno;
 732 #endif
 733     }
 734
 735     if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
 736         ret == -ENOTTY) {
 737         s->has_discard = 0;
 738         ret = 0;
 739     }
 740     return ret;
 741 }
 742
 743 static int aio_worker(void *arg)
 744 {
 745     RawPosixAIOData *aiocb = arg;
 746     ssize_t ret = 0;
 747
 748     switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
 749     case QEMU_AIO_READ:
 750         ret = handle_aiocb_rw(aiocb);
 751         if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) {
 752             iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
 753                       0, aiocb->aio_nbytes - ret);
 754
 755             ret = aiocb->aio_nbytes;
 756         }
 757         if (ret == aiocb->aio_nbytes) {
 758             ret = 0;
 759         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 760             ret = -EINVAL;
 761         }
 762         break;
 763     case QEMU_AIO_WRITE:
 764         ret = handle_aiocb_rw(aiocb);
 765         if (ret == aiocb->aio_nbytes) {
 766             ret = 0;
 767         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 768             ret = -EINVAL;
 769         }
 770         break;
 771     case QEMU_AIO_FLUSH:
 772         ret = handle_aiocb_flush(aiocb);
 773         break;
 774     case QEMU_AIO_IOCTL:
 775         ret = handle_aiocb_ioctl(aiocb);
 776         break;
 777     case QEMU_AIO_DISCARD:
 778         ret = handle_aiocb_discard(aiocb);
 779         break;
 780     default:
 781         fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
 782         ret = -EINVAL;
 783         break;
 784     }
 785
 786     g_slice_free(RawPosixAIOData, aiocb);
 787     return ret;
 788 }
 789
 790 static BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd,
 791         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 792         BlockDriverCompletionFunc *cb, void *opaque, int type)
 793 {
 794     RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
 795     ThreadPool *pool;
 796
 797     acb->bs = bs;
 798     acb->aio_type = type;
 799     acb->aio_fildes = fd;
 800
 801     if (qiov) {
 802         acb->aio_iov = qiov->iov;
 803         acb->aio_niov = qiov->niov;
 804     }
 805     acb->aio_nbytes = nb_sectors * 512;
 806     acb->aio_offset = sector_num * 512;
 807
 808     trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
 809     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
 810     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
 811 }
 812
 813 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
 814         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 815         BlockDriverCompletionFunc *cb, void *opaque, int type)
 816 {
 817     BDRVRawState *s = bs->opaque;
 818
 819     if (fd_open(bs) < 0)
 820         return NULL;
 821
 822     /*
 823      * If O_DIRECT is used the buffer needs to be aligned on a sector
 824      * boundary.  Check if this is the case or tell the low-level
 825      * driver that it needs to copy the buffer.
 826      */
 827     if ((bs->open_flags & BDRV_O_NOCACHE)) {
 828         if (!bdrv_qiov_is_aligned(bs, qiov)) {
 829             type |= QEMU_AIO_MISALIGNED;
 830 #ifdef CONFIG_LINUX_AIO
 831         } else if (s->use_aio) {
 832             return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
 833                                nb_sectors, cb, opaque, type);
 834 #endif
 835         }
 836     }
 837
 838     return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
 839                        cb, opaque, type);
 840 }
 841
 842 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
 843         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 844         BlockDriverCompletionFunc *cb, void *opaque)
 845 {
 846     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 847                           cb, opaque, QEMU_AIO_READ);
 848 }
 849
 850 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
 851         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 852         BlockDriverCompletionFunc *cb, void *opaque)
 853 {
 854     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 855                           cb, opaque, QEMU_AIO_WRITE);
 856 }
 857
 858 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
 859         BlockDriverCompletionFunc *cb, void *opaque)
 860 {
 861     BDRVRawState *s = bs->opaque;
 862
 863     if (fd_open(bs) < 0)
 864         return NULL;
 865
 866     return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
 867 }
 868
 869 static void raw_close(BlockDriverState *bs)
 870 {
 871     BDRVRawState *s = bs->opaque;
 872     if (s->fd >= 0) {
 873         qemu_close(s->fd);
 874         s->fd = -1;
 875     }
 876 }
 877
 878 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 879 {
 880     BDRVRawState *s = bs->opaque;
 881     struct stat st;
 882
 883     if (fstat(s->fd, &st)) {
 884         return -errno;
 885     }
 886
 887     if (S_ISREG(st.st_mode)) {
 888         if (ftruncate(s->fd, offset) < 0) {
 889             return -errno;
 890         }
 891     } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 892        if (offset > raw_getlength(bs)) {
 893            return -EINVAL;
 894        }
 895     } else {
 896         return -ENOTSUP;
 897     }
 898
 899     return 0;
 900 }
 901
 902 #ifdef __OpenBSD__
 903 static int64_t raw_getlength(BlockDriverState *bs)
 904 {
 905     BDRVRawState *s = bs->opaque;
 906     int fd = s->fd;
 907     struct stat st;
 908
 909     if (fstat(fd, &st))
 910         return -1;
 911     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 912         struct disklabel dl;
 913
 914         if (ioctl(fd, DIOCGDINFO, &dl))
 915             return -1;
 916         return (uint64_t)dl.d_secsize *
 917             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 918     } else
 919         return st.st_size;
 920 }
 921 #elif defined(__NetBSD__)
 922 static int64_t raw_getlength(BlockDriverState *bs)
 923 {
 924     BDRVRawState *s = bs->opaque;
 925     int fd = s->fd;
 926     struct stat st;
 927
 928     if (fstat(fd, &st))
 929         return -1;
 930     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 931         struct dkwedge_info dkw;
 932
 933         if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
 934             return dkw.dkw_size * 512;
 935         } else {
 936             struct disklabel dl;
 937
 938             if (ioctl(fd, DIOCGDINFO, &dl))
 939                 return -1;
 940             return (uint64_t)dl.d_secsize *
 941                 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 942         }
 943     } else
 944         return st.st_size;
 945 }
 946 #elif defined(__sun__)
 947 static int64_t raw_getlength(BlockDriverState *bs)
 948 {
 949     BDRVRawState *s = bs->opaque;
 950     struct dk_minfo minfo;
 951     int ret;
 952
 953     ret = fd_open(bs);
 954     if (ret < 0) {
 955         return ret;
 956     }
 957
 958     /*
 959      * Use the DKIOCGMEDIAINFO ioctl to read the size.
 960      */
 961     ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
 962     if (ret != -1) {
 963         return minfo.dki_lbsize * minfo.dki_capacity;
 964     }
 965
 966     /*
 967      * There are reports that lseek on some devices fails, but
 968      * irc discussion said that contingency on contingency was overkill.
 969      */
 970     return lseek(s->fd, 0, SEEK_END);
 971 }
 972 #elif defined(CONFIG_BSD)
 973 static int64_t raw_getlength(BlockDriverState *bs)
 974 {
 975     BDRVRawState *s = bs->opaque;
 976     int fd = s->fd;
 977     int64_t size;
 978     struct stat sb;
 979 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 980     int reopened = 0;
 981 #endif
 982     int ret;
 983
 984     ret = fd_open(bs);
 985     if (ret < 0)
 986         return ret;
 987
 988 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 989 again:
 990 #endif
 991     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 992 #ifdef DIOCGMEDIASIZE
 993         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 994 #elif defined(DIOCGPART)
 995         {
 996                 struct partinfo pi;
 997                 if (ioctl(fd, DIOCGPART, &pi) == 0)
 998                         size = pi.media_size;
 999                 else
1000                         size = 0;
1001         }
1002         if (size == 0)
1003 #endif
1004 #if defined(__APPLE__) && defined(__MACH__)
1005         size = LONG_LONG_MAX;
1006 #else
1007         size = lseek(fd, 0LL, SEEK_END);
1008 #endif
1009 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1010         switch(s->type) {
1011         case FTYPE_CD:
1012             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1013             if (size == 2048LL * (unsigned)-1)
1014                 size = 0;
1015             /* XXX no disc?  maybe we need to reopen... */
1016             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1017                 reopened = 1;
1018                 goto again;
1019             }
1020         }
1021 #endif
1022     } else {
1023         size = lseek(fd, 0, SEEK_END);
1024     }
1025     return size;
1026 }
1027 #else
1028 static int64_t raw_getlength(BlockDriverState *bs)
1029 {
1030     BDRVRawState *s = bs->opaque;
1031     int ret;
1032
1033     ret = fd_open(bs);
1034     if (ret < 0) {
1035         return ret;
1036     }
1037
1038     return lseek(s->fd, 0, SEEK_END);
1039 }
1040 #endif
1041
1042 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1043 {
1044     struct stat st;
1045     BDRVRawState *s = bs->opaque;
1046
1047     if (fstat(s->fd, &st) < 0) {
1048         return -errno;
1049     }
1050     return (int64_t)st.st_blocks * 512;
1051 }
1052
1053 static int raw_create(const char *filename, QEMUOptionParameter *options,
1054                       Error **errp)
1055 {
1056     int fd;
1057     int result = 0;
1058     int64_t total_size = 0;
1059
1060     /* Read out options */
1061     while (options && options->name) {
1062         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1063             total_size = options->value.n / BDRV_SECTOR_SIZE;
1064         }
1065         options++;
1066     }
1067
1068     fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
1069                    0644);
1070     if (fd < 0) {
1071         result = -errno;
1072         error_setg_errno(errp, -result, "Could not create file");
1073     } else {
1074         if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
1075             result = -errno;
1076             error_setg_errno(errp, -result, "Could not resize file");
1077         }
1078         if (qemu_close(fd) != 0) {
1079             result = -errno;
1080             error_setg_errno(errp, -result, "Could not close the new file");
1081         }
1082     }
1083     return result;
1084 }
1085
1086 /*
1087  * Returns true iff the specified sector is present in the disk image. Drivers
1088  * not implementing the functionality are assumed to not support backing files,
1089  * hence all their sectors are reported as allocated.
1090  *
1091  * If 'sector_num' is beyond the end of the disk image the return value is 0
1092  * and 'pnum' is set to 0.
1093  *
1094  * 'pnum' is set to the number of sectors (including and immediately following
1095  * the specified sector) that are known to be in the same
1096  * allocated/unallocated state.
1097  *
1098  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
1099  * beyond the end of the disk image it will be clamped.
1100  */
1101 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1102                                             int64_t sector_num,
1103                                             int nb_sectors, int *pnum)
1104 {
1105     off_t start, data, hole;
1106     int64_t ret;
1107
1108     ret = fd_open(bs);
1109     if (ret < 0) {
1110         return ret;
1111     }
1112
1113     start = sector_num * BDRV_SECTOR_SIZE;
1114     ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1115
1116 #ifdef CONFIG_FIEMAP
1117
1118     BDRVRawState *s = bs->opaque;
1119     struct {
1120         struct fiemap fm;
1121         struct fiemap_extent fe;
1122     } f;
1123
1124     f.fm.fm_start = start;
1125     f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE;
1126     f.fm.fm_flags = 0;
1127     f.fm.fm_extent_count = 1;
1128     f.fm.fm_reserved = 0;
1129     if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) {
1130         /* Assume everything is allocated.  */
1131         *pnum = nb_sectors;
1132         return ret;
1133     }
1134
1135     if (f.fm.fm_mapped_extents == 0) {
1136         /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1137          * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1138          */
1139         off_t length = lseek(s->fd, 0, SEEK_END);
1140         hole = f.fm.fm_start;
1141         data = MIN(f.fm.fm_start + f.fm.fm_length, length);
1142     } else {
1143         data = f.fe.fe_logical;
1144         hole = f.fe.fe_logical + f.fe.fe_length;
1145         if (f.fe.fe_flags & FIEMAP_EXTENT_UNWRITTEN) {
1146             ret |= BDRV_BLOCK_ZERO;
1147         }
1148     }
1149
1150 #elif defined SEEK_HOLE && defined SEEK_DATA
1151
1152     BDRVRawState *s = bs->opaque;
1153
1154     hole = lseek(s->fd, start, SEEK_HOLE);
1155     if (hole == -1) {
1156         /* -ENXIO indicates that sector_num was past the end of the file.
1157          * There is a virtual hole there.  */
1158         assert(errno != -ENXIO);
1159
1160         /* Most likely EINVAL.  Assume everything is allocated.  */
1161         *pnum = nb_sectors;
1162         return ret;
1163     }
1164
1165     if (hole > start) {
1166         data = start;
1167     } else {
1168         /* On a hole.  We need another syscall to find its end.  */
1169         data = lseek(s->fd, start, SEEK_DATA);
1170         if (data == -1) {
1171             data = lseek(s->fd, 0, SEEK_END);
1172         }
1173     }
1174 #else
1175     data = 0;
1176     hole = start + nb_sectors * BDRV_SECTOR_SIZE;
1177 #endif
1178
1179     if (data <= start) {
1180         /* On a data extent, compute sectors to the end of the extent.  */
1181         *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1182     } else {
1183         /* On a hole, compute sectors to the beginning of the next extent.  */
1184         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1185         ret &= ~BDRV_BLOCK_DATA;
1186         ret |= BDRV_BLOCK_ZERO;
1187     }
1188
1189     return ret;
1190 }
1191
1192 static coroutine_fn BlockDriverAIOCB *raw_aio_discard(BlockDriverState *bs,
1193     int64_t sector_num, int nb_sectors,
1194     BlockDriverCompletionFunc *cb, void *opaque)
1195 {
1196     BDRVRawState *s = bs->opaque;
1197
1198     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1199                        cb, opaque, QEMU_AIO_DISCARD);
1200 }
1201
1202 static QEMUOptionParameter raw_create_options[] = {
1203     {
1204         .name = BLOCK_OPT_SIZE,
1205         .type = OPT_SIZE,
1206         .help = "Virtual disk size"
1207     },
1208     { NULL }
1209 };
1210
1211 static BlockDriver bdrv_file = {
1212     .format_name = "file",
1213     .protocol_name = "file",
1214     .instance_size = sizeof(BDRVRawState),
1215     .bdrv_needs_filename = true,
1216     .bdrv_probe = NULL, /* no probe for protocols */
1217     .bdrv_file_open = raw_open,
1218     .bdrv_reopen_prepare = raw_reopen_prepare,
1219     .bdrv_reopen_commit = raw_reopen_commit,
1220     .bdrv_reopen_abort = raw_reopen_abort,
1221     .bdrv_close = raw_close,
1222     .bdrv_create = raw_create,
1223     .bdrv_has_zero_init = bdrv_has_zero_init_1,
1224     .bdrv_co_get_block_status = raw_co_get_block_status,
1225
1226     .bdrv_aio_readv = raw_aio_readv,
1227     .bdrv_aio_writev = raw_aio_writev,
1228     .bdrv_aio_flush = raw_aio_flush,
1229     .bdrv_aio_discard = raw_aio_discard,
1230
1231     .bdrv_truncate = raw_truncate,
1232     .bdrv_getlength = raw_getlength,
1233     .bdrv_get_allocated_file_size
1234                         = raw_get_allocated_file_size,
1235
1236     .create_options = raw_create_options,
1237 };
1238
1239 /***********************************************/
1240 /* host device */
1241
1242 #if defined(__APPLE__) && defined(__MACH__)
1243 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1244 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1245
1246 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1247 {
1248     kern_return_t       kernResult;
1249     mach_port_t     masterPort;
1250     CFMutableDictionaryRef  classesToMatch;
1251
1252     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1253     if ( KERN_SUCCESS != kernResult ) {
1254         printf( "IOMasterPort returned %d\n", kernResult );
1255     }
1256
1257     classesToMatch = IOServiceMatching( kIOCDMediaClass );
1258     if ( classesToMatch == NULL ) {
1259         printf( "IOServiceMatching returned a NULL dictionary.\n" );
1260     } else {
1261     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1262     }
1263     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1264     if ( KERN_SUCCESS != kernResult )
1265     {
1266         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1267     }
1268
1269     return kernResult;
1270 }
1271
1272 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
1273 {
1274     io_object_t     nextMedia;
1275     kern_return_t   kernResult = KERN_FAILURE;
1276     *bsdPath = '\0';
1277     nextMedia = IOIteratorNext( mediaIterator );
1278     if ( nextMedia )
1279     {
1280         CFTypeRef   bsdPathAsCFString;
1281     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1282         if ( bsdPathAsCFString ) {
1283             size_t devPathLength;
1284             strcpy( bsdPath, _PATH_DEV );
1285             strcat( bsdPath, "r" );
1286             devPathLength = strlen( bsdPath );
1287             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1288                 kernResult = KERN_SUCCESS;
1289             }
1290             CFRelease( bsdPathAsCFString );
1291         }
1292         IOObjectRelease( nextMedia );
1293     }
1294
1295     return kernResult;
1296 }
1297
1298 #endif
1299
1300 static int hdev_probe_device(const char *filename)
1301 {
1302     struct stat st;
1303
1304     /* allow a dedicated CD-ROM driver to match with a higher priority */
1305     if (strstart(filename, "/dev/cdrom", NULL))
1306         return 50;
1307
1308     if (stat(filename, &st) >= 0 &&
1309             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
1310         return 100;
1311     }
1312
1313     return 0;
1314 }
1315
1316 static int check_hdev_writable(BDRVRawState *s)
1317 {
1318 #if defined(BLKROGET)
1319     /* Linux block devices can be configured "read-only" using blockdev(8).
1320      * This is independent of device node permissions and therefore open(2)
1321      * with O_RDWR succeeds.  Actual writes fail with EPERM.
1322      *
1323      * bdrv_open() is supposed to fail if the disk is read-only.  Explicitly
1324      * check for read-only block devices so that Linux block devices behave
1325      * properly.
1326      */
1327     struct stat st;
1328     int readonly = 0;
1329
1330     if (fstat(s->fd, &st)) {
1331         return -errno;
1332     }
1333
1334     if (!S_ISBLK(st.st_mode)) {
1335         return 0;
1336     }
1337
1338     if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
1339         return -errno;
1340     }
1341
1342     if (readonly) {
1343         return -EACCES;
1344     }
1345 #endif /* defined(BLKROGET) */
1346     return 0;
1347 }
1348
1349 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
1350                      Error **errp)
1351 {
1352     BDRVRawState *s = bs->opaque;
1353     Error *local_err = NULL;
1354     int ret;
1355     const char *filename = qdict_get_str(options, "filename");
1356
1357 #if defined(__APPLE__) && defined(__MACH__)
1358     if (strstart(filename, "/dev/cdrom", NULL)) {
1359         kern_return_t kernResult;
1360         io_iterator_t mediaIterator;
1361         char bsdPath[ MAXPATHLEN ];
1362         int fd;
1363
1364         kernResult = FindEjectableCDMedia( &mediaIterator );
1365         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1366
1367         if ( bsdPath[ 0 ] != '\0' ) {
1368             strcat(bsdPath,"s0");
1369             /* some CDs don't have a partition 0 */
1370             fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1371             if (fd < 0) {
1372                 bsdPath[strlen(bsdPath)-1] = '1';
1373             } else {
1374                 qemu_close(fd);
1375             }
1376             filename = bsdPath;
1377             qdict_put(options, "filename", qstring_from_str(filename));
1378         }
1379
1380         if ( mediaIterator )
1381             IOObjectRelease( mediaIterator );
1382     }
1383 #endif
1384
1385     s->type = FTYPE_FILE;
1386 #if defined(__linux__)
1387     {
1388         char resolved_path[ MAXPATHLEN ], *temp;
1389
1390         temp = realpath(filename, resolved_path);
1391         if (temp && strstart(temp, "/dev/sg", NULL)) {
1392             bs->sg = 1;
1393         }
1394     }
1395 #endif
1396
1397     ret = raw_open_common(bs, options, flags, 0, &local_err);
1398     if (ret < 0) {
1399         if (error_is_set(&local_err)) {
1400             error_propagate(errp, local_err);
1401         }
1402         return ret;
1403     }
1404
1405     if (flags & BDRV_O_RDWR) {
1406         ret = check_hdev_writable(s);
1407         if (ret < 0) {
1408             raw_close(bs);
1409             error_setg_errno(errp, -ret, "The device is not writable");
1410             return ret;
1411         }
1412     }
1413
1414     return ret;
1415 }
1416
1417 #if defined(__linux__)
1418 /* Note: we do not have a reliable method to detect if the floppy is
1419    present. The current method is to try to open the floppy at every
1420    I/O and to keep it opened during a few hundreds of ms. */
1421 static int fd_open(BlockDriverState *bs)
1422 {
1423     BDRVRawState *s = bs->opaque;
1424     int last_media_present;
1425
1426     if (s->type != FTYPE_FD)
1427         return 0;
1428     last_media_present = (s->fd >= 0);
1429     if (s->fd >= 0 &&
1430         (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1431         qemu_close(s->fd);
1432         s->fd = -1;
1433 #ifdef DEBUG_FLOPPY
1434         printf("Floppy closed\n");
1435 #endif
1436     }
1437     if (s->fd < 0) {
1438         if (s->fd_got_error &&
1439             (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1440 #ifdef DEBUG_FLOPPY
1441             printf("No floppy (open delayed)\n");
1442 #endif
1443             return -EIO;
1444         }
1445         s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
1446         if (s->fd < 0) {
1447             s->fd_error_time = get_clock();
1448             s->fd_got_error = 1;
1449             if (last_media_present)
1450                 s->fd_media_changed = 1;
1451 #ifdef DEBUG_FLOPPY
1452             printf("No floppy\n");
1453 #endif
1454             return -EIO;
1455         }
1456 #ifdef DEBUG_FLOPPY
1457         printf("Floppy opened\n");
1458 #endif
1459     }
1460     if (!last_media_present)
1461         s->fd_media_changed = 1;
1462     s->fd_open_time = get_clock();
1463     s->fd_got_error = 0;
1464     return 0;
1465 }
1466
1467 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1468 {
1469     BDRVRawState *s = bs->opaque;
1470
1471     return ioctl(s->fd, req, buf);
1472 }
1473
1474 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1475         unsigned long int req, void *buf,
1476         BlockDriverCompletionFunc *cb, void *opaque)
1477 {
1478     BDRVRawState *s = bs->opaque;
1479     RawPosixAIOData *acb;
1480     ThreadPool *pool;
1481
1482     if (fd_open(bs) < 0)
1483         return NULL;
1484
1485     acb = g_slice_new(RawPosixAIOData);
1486     acb->bs = bs;
1487     acb->aio_type = QEMU_AIO_IOCTL;
1488     acb->aio_fildes = s->fd;
1489     acb->aio_offset = 0;
1490     acb->aio_ioctl_buf = buf;
1491     acb->aio_ioctl_cmd = req;
1492     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1493     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1494 }
1495
1496 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1497 static int fd_open(BlockDriverState *bs)
1498 {
1499     BDRVRawState *s = bs->opaque;
1500
1501     /* this is just to ensure s->fd is sane (its called by io ops) */
1502     if (s->fd >= 0)
1503         return 0;
1504     return -EIO;
1505 }
1506 #else /* !linux && !FreeBSD */
1507
1508 static int fd_open(BlockDriverState *bs)
1509 {
1510     return 0;
1511 }
1512
1513 #endif /* !linux && !FreeBSD */
1514
1515 static coroutine_fn BlockDriverAIOCB *hdev_aio_discard(BlockDriverState *bs,
1516     int64_t sector_num, int nb_sectors,
1517     BlockDriverCompletionFunc *cb, void *opaque)
1518 {
1519     BDRVRawState *s = bs->opaque;
1520
1521     if (fd_open(bs) < 0) {
1522         return NULL;
1523     }
1524     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1525                        cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1526 }
1527
1528 static int hdev_create(const char *filename, QEMUOptionParameter *options,
1529                        Error **errp)
1530 {
1531     int fd;
1532     int ret = 0;
1533     struct stat stat_buf;
1534     int64_t total_size = 0;
1535
1536     /* Read out options */
1537     while (options && options->name) {
1538         if (!strcmp(options->name, "size")) {
1539             total_size = options->value.n / BDRV_SECTOR_SIZE;
1540         }
1541         options++;
1542     }
1543
1544     fd = qemu_open(filename, O_WRONLY | O_BINARY);
1545     if (fd < 0) {
1546         ret = -errno;
1547         error_setg_errno(errp, -ret, "Could not open device");
1548         return ret;
1549     }
1550
1551     if (fstat(fd, &stat_buf) < 0) {
1552         ret = -errno;
1553         error_setg_errno(errp, -ret, "Could not stat device");
1554     } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
1555         error_setg(errp,
1556                    "The given file is neither a block nor a character device");
1557         ret = -ENODEV;
1558     } else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE) {
1559         error_setg(errp, "Device is too small");
1560         ret = -ENOSPC;
1561     }
1562
1563     qemu_close(fd);
1564     return ret;
1565 }
1566
1567 static BlockDriver bdrv_host_device = {
1568     .format_name        = "host_device",
1569     .protocol_name        = "host_device",
1570     .instance_size      = sizeof(BDRVRawState),
1571     .bdrv_needs_filename = true,
1572     .bdrv_probe_device  = hdev_probe_device,
1573     .bdrv_file_open     = hdev_open,
1574     .bdrv_close         = raw_close,
1575     .bdrv_reopen_prepare = raw_reopen_prepare,
1576     .bdrv_reopen_commit  = raw_reopen_commit,
1577     .bdrv_reopen_abort   = raw_reopen_abort,
1578     .bdrv_create        = hdev_create,
1579     .create_options     = raw_create_options,
1580
1581     .bdrv_aio_readv     = raw_aio_readv,
1582     .bdrv_aio_writev    = raw_aio_writev,
1583     .bdrv_aio_flush     = raw_aio_flush,
1584     .bdrv_aio_discard   = hdev_aio_discard,
1585
1586     .bdrv_truncate      = raw_truncate,
1587     .bdrv_getlength     = raw_getlength,
1588     .bdrv_get_allocated_file_size
1589                         = raw_get_allocated_file_size,
1590
1591     /* generic scsi device */
1592 #ifdef __linux__
1593     .bdrv_ioctl         = hdev_ioctl,
1594     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1595 #endif
1596 };
1597
1598 #ifdef __linux__
1599 static int floppy_open(BlockDriverState *bs, QDict *options, int flags,
1600                        Error **errp)
1601 {
1602     BDRVRawState *s = bs->opaque;
1603     Error *local_err = NULL;
1604     int ret;
1605
1606     s->type = FTYPE_FD;
1607
1608     /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1609     ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
1610     if (ret) {
1611         if (error_is_set(&local_err)) {
1612             error_propagate(errp, local_err);
1613         }
1614         return ret;
1615     }
1616
1617     /* close fd so that we can reopen it as needed */
1618     qemu_close(s->fd);
1619     s->fd = -1;
1620     s->fd_media_changed = 1;
1621
1622     return 0;
1623 }
1624
1625 static int floppy_probe_device(const char *filename)
1626 {
1627     int fd, ret;
1628     int prio = 0;
1629     struct floppy_struct fdparam;
1630     struct stat st;
1631
1632     if (strstart(filename, "/dev/fd", NULL) &&
1633         !strstart(filename, "/dev/fdset/", NULL)) {
1634         prio = 50;
1635     }
1636
1637     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1638     if (fd < 0) {
1639         goto out;
1640     }
1641     ret = fstat(fd, &st);
1642     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1643         goto outc;
1644     }
1645
1646     /* Attempt to detect via a floppy specific ioctl */
1647     ret = ioctl(fd, FDGETPRM, &fdparam);
1648     if (ret >= 0)
1649         prio = 100;
1650
1651 outc:
1652     qemu_close(fd);
1653 out:
1654     return prio;
1655 }
1656
1657
1658 static int floppy_is_inserted(BlockDriverState *bs)
1659 {
1660     return fd_open(bs) >= 0;
1661 }
1662
1663 static int floppy_media_changed(BlockDriverState *bs)
1664 {
1665     BDRVRawState *s = bs->opaque;
1666     int ret;
1667
1668     /*
1669      * XXX: we do not have a true media changed indication.
1670      * It does not work if the floppy is changed without trying to read it.
1671      */
1672     fd_open(bs);
1673     ret = s->fd_media_changed;
1674     s->fd_media_changed = 0;
1675 #ifdef DEBUG_FLOPPY
1676     printf("Floppy changed=%d\n", ret);
1677 #endif
1678     return ret;
1679 }
1680
1681 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
1682 {
1683     BDRVRawState *s = bs->opaque;
1684     int fd;
1685
1686     if (s->fd >= 0) {
1687         qemu_close(s->fd);
1688         s->fd = -1;
1689     }
1690     fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
1691     if (fd >= 0) {
1692         if (ioctl(fd, FDEJECT, 0) < 0)
1693             perror("FDEJECT");
1694         qemu_close(fd);
1695     }
1696 }
1697
1698 static BlockDriver bdrv_host_floppy = {
1699     .format_name        = "host_floppy",
1700     .protocol_name      = "host_floppy",
1701     .instance_size      = sizeof(BDRVRawState),
1702     .bdrv_needs_filename = true,
1703     .bdrv_probe_device  = floppy_probe_device,
1704     .bdrv_file_open     = floppy_open,
1705     .bdrv_close         = raw_close,
1706     .bdrv_reopen_prepare = raw_reopen_prepare,
1707     .bdrv_reopen_commit  = raw_reopen_commit,
1708     .bdrv_reopen_abort   = raw_reopen_abort,
1709     .bdrv_create        = hdev_create,
1710     .create_options     = raw_create_options,
1711
1712     .bdrv_aio_readv     = raw_aio_readv,
1713     .bdrv_aio_writev    = raw_aio_writev,
1714     .bdrv_aio_flush     = raw_aio_flush,
1715
1716     .bdrv_truncate      = raw_truncate,
1717     .bdrv_getlength      = raw_getlength,
1718     .has_variable_length = true,
1719     .bdrv_get_allocated_file_size
1720                         = raw_get_allocated_file_size,
1721
1722     /* removable device support */
1723     .bdrv_is_inserted   = floppy_is_inserted,
1724     .bdrv_media_changed = floppy_media_changed,
1725     .bdrv_eject         = floppy_eject,
1726 };
1727
1728 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
1729                       Error **errp)
1730 {
1731     BDRVRawState *s = bs->opaque;
1732     Error *local_err = NULL;
1733     int ret;
1734
1735     s->type = FTYPE_CD;
1736
1737     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
1738     ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
1739     if (error_is_set(&local_err)) {
1740         error_propagate(errp, local_err);
1741     }
1742     return ret;
1743 }
1744
1745 static int cdrom_probe_device(const char *filename)
1746 {
1747     int fd, ret;
1748     int prio = 0;
1749     struct stat st;
1750
1751     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1752     if (fd < 0) {
1753         goto out;
1754     }
1755     ret = fstat(fd, &st);
1756     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1757         goto outc;
1758     }
1759
1760     /* Attempt to detect via a CDROM specific ioctl */
1761     ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1762     if (ret >= 0)
1763         prio = 100;
1764
1765 outc:
1766     qemu_close(fd);
1767 out:
1768     return prio;
1769 }
1770
1771 static int cdrom_is_inserted(BlockDriverState *bs)
1772 {
1773     BDRVRawState *s = bs->opaque;
1774     int ret;
1775
1776     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1777     if (ret == CDS_DISC_OK)
1778         return 1;
1779     return 0;
1780 }
1781
1782 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1783 {
1784     BDRVRawState *s = bs->opaque;
1785
1786     if (eject_flag) {
1787         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
1788             perror("CDROMEJECT");
1789     } else {
1790         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
1791             perror("CDROMEJECT");
1792     }
1793 }
1794
1795 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1796 {
1797     BDRVRawState *s = bs->opaque;
1798
1799     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
1800         /*
1801          * Note: an error can happen if the distribution automatically
1802          * mounts the CD-ROM
1803          */
1804         /* perror("CDROM_LOCKDOOR"); */
1805     }
1806 }
1807
1808 static BlockDriver bdrv_host_cdrom = {
1809     .format_name        = "host_cdrom",
1810     .protocol_name      = "host_cdrom",
1811     .instance_size      = sizeof(BDRVRawState),
1812     .bdrv_needs_filename = true,
1813     .bdrv_probe_device  = cdrom_probe_device,
1814     .bdrv_file_open     = cdrom_open,
1815     .bdrv_close         = raw_close,
1816     .bdrv_reopen_prepare = raw_reopen_prepare,
1817     .bdrv_reopen_commit  = raw_reopen_commit,
1818     .bdrv_reopen_abort   = raw_reopen_abort,
1819     .bdrv_create        = hdev_create,
1820     .create_options     = raw_create_options,
1821
1822     .bdrv_aio_readv     = raw_aio_readv,
1823     .bdrv_aio_writev    = raw_aio_writev,
1824     .bdrv_aio_flush     = raw_aio_flush,
1825
1826     .bdrv_truncate      = raw_truncate,
1827     .bdrv_getlength      = raw_getlength,
1828     .has_variable_length = true,
1829     .bdrv_get_allocated_file_size
1830                         = raw_get_allocated_file_size,
1831
1832     /* removable device support */
1833     .bdrv_is_inserted   = cdrom_is_inserted,
1834     .bdrv_eject         = cdrom_eject,
1835     .bdrv_lock_medium   = cdrom_lock_medium,
1836
1837     /* generic scsi device */
1838     .bdrv_ioctl         = hdev_ioctl,
1839     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1840 };
1841 #endif /* __linux__ */
1842
1843 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1844 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
1845                       Error **errp)
1846 {
1847     BDRVRawState *s = bs->opaque;
1848     Error *local_err = NULL;
1849     int ret;
1850
1851     s->type = FTYPE_CD;
1852
1853     ret = raw_open_common(bs, options, flags, 0, &local_err);
1854     if (ret) {
1855         if (error_is_set(&local_err)) {
1856             error_propagate(errp, local_err);
1857         }
1858         return ret;
1859     }
1860
1861     /* make sure the door isn't locked at this time */
1862     ioctl(s->fd, CDIOCALLOW);
1863     return 0;
1864 }
1865
1866 static int cdrom_probe_device(const char *filename)
1867 {
1868     if (strstart(filename, "/dev/cd", NULL) ||
1869             strstart(filename, "/dev/acd", NULL))
1870         return 100;
1871     return 0;
1872 }
1873
1874 static int cdrom_reopen(BlockDriverState *bs)
1875 {
1876     BDRVRawState *s = bs->opaque;
1877     int fd;
1878
1879     /*
1880      * Force reread of possibly changed/newly loaded disc,
1881      * FreeBSD seems to not notice sometimes...
1882      */
1883     if (s->fd >= 0)
1884         qemu_close(s->fd);
1885     fd = qemu_open(bs->filename, s->open_flags, 0644);
1886     if (fd < 0) {
1887         s->fd = -1;
1888         return -EIO;
1889     }
1890     s->fd = fd;
1891
1892     /* make sure the door isn't locked at this time */
1893     ioctl(s->fd, CDIOCALLOW);
1894     return 0;
1895 }
1896
1897 static int cdrom_is_inserted(BlockDriverState *bs)
1898 {
1899     return raw_getlength(bs) > 0;
1900 }
1901
1902 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1903 {
1904     BDRVRawState *s = bs->opaque;
1905
1906     if (s->fd < 0)
1907         return;
1908
1909     (void) ioctl(s->fd, CDIOCALLOW);
1910
1911     if (eject_flag) {
1912         if (ioctl(s->fd, CDIOCEJECT) < 0)
1913             perror("CDIOCEJECT");
1914     } else {
1915         if (ioctl(s->fd, CDIOCCLOSE) < 0)
1916             perror("CDIOCCLOSE");
1917     }
1918
1919     cdrom_reopen(bs);
1920 }
1921
1922 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1923 {
1924     BDRVRawState *s = bs->opaque;
1925
1926     if (s->fd < 0)
1927         return;
1928     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1929         /*
1930          * Note: an error can happen if the distribution automatically
1931          * mounts the CD-ROM
1932          */
1933         /* perror("CDROM_LOCKDOOR"); */
1934     }
1935 }
1936
1937 static BlockDriver bdrv_host_cdrom = {
1938     .format_name        = "host_cdrom",
1939     .protocol_name      = "host_cdrom",
1940     .instance_size      = sizeof(BDRVRawState),
1941     .bdrv_needs_filename = true,
1942     .bdrv_probe_device  = cdrom_probe_device,
1943     .bdrv_file_open     = cdrom_open,
1944     .bdrv_close         = raw_close,
1945     .bdrv_reopen_prepare = raw_reopen_prepare,
1946     .bdrv_reopen_commit  = raw_reopen_commit,
1947     .bdrv_reopen_abort   = raw_reopen_abort,
1948     .bdrv_create        = hdev_create,
1949     .create_options     = raw_create_options,
1950
1951     .bdrv_aio_readv     = raw_aio_readv,
1952     .bdrv_aio_writev    = raw_aio_writev,
1953     .bdrv_aio_flush     = raw_aio_flush,
1954
1955     .bdrv_truncate      = raw_truncate,
1956     .bdrv_getlength      = raw_getlength,
1957     .has_variable_length = true,
1958     .bdrv_get_allocated_file_size
1959                         = raw_get_allocated_file_size,
1960
1961     /* removable device support */
1962     .bdrv_is_inserted   = cdrom_is_inserted,
1963     .bdrv_eject         = cdrom_eject,
1964     .bdrv_lock_medium   = cdrom_lock_medium,
1965 };
1966 #endif /* __FreeBSD__ */
1967
1968 #ifdef CONFIG_LINUX_AIO
1969 /**
1970  * Return the file descriptor for Linux AIO
1971  *
1972  * This function is a layering violation and should be removed when it becomes
1973  * possible to call the block layer outside the global mutex.  It allows the
1974  * caller to hijack the file descriptor so I/O can be performed outside the
1975  * block layer.
1976  */
1977 int raw_get_aio_fd(BlockDriverState *bs)
1978 {
1979     BDRVRawState *s;
1980
1981     if (!bs->drv) {
1982         return -ENOMEDIUM;
1983     }
1984
1985     if (bs->drv == bdrv_find_format("raw")) {
1986         bs = bs->file;
1987     }
1988
1989     /* raw-posix has several protocols so just check for raw_aio_readv */
1990     if (bs->drv->bdrv_aio_readv != raw_aio_readv) {
1991         return -ENOTSUP;
1992     }
1993
1994     s = bs->opaque;
1995     if (!s->use_aio) {
1996         return -ENOTSUP;
1997     }
1998     return s->fd;
1999 }
2000 #endif /* CONFIG_LINUX_AIO */
2001
2002 static void bdrv_file_init(void)
2003 {
2004     /*
2005      * Register all the drivers.  Note that order is important, the driver
2006      * registered last will get probed first.
2007      */
2008     bdrv_register(&bdrv_file);
2009     bdrv_register(&bdrv_host_device);
2010 #ifdef __linux__
2011     bdrv_register(&bdrv_host_floppy);
2012     bdrv_register(&bdrv_host_cdrom);
2013 #endif
2014 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2015     bdrv_register(&bdrv_host_cdrom);
2016 #endif
2017 }
2018
2019 block_init(bdrv_file_init);