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         error_setg_errno(errp, -ret, "Could not open file");
 314         goto fail;
 315     }
 316     s->fd = fd;
 317
 318 #ifdef CONFIG_LINUX_AIO
 319     if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
 320         qemu_close(fd);
 321         ret = -errno;
 322         error_setg_errno(errp, -ret, "Could not set AIO state");
 323         goto fail;
 324     }
 325 #endif
 326
 327     s->has_discard = 1;
 328 #ifdef CONFIG_XFS
 329     if (platform_test_xfs_fd(s->fd)) {
 330         s->is_xfs = 1;
 331     }
 332 #endif
 333
 334     ret = 0;
 335 fail:
 336     qemu_opts_del(opts);
 337     return ret;
 338 }
 339
 340 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
 341                     Error **errp)
 342 {
 343     BDRVRawState *s = bs->opaque;
 344     Error *local_err = NULL;
 345     int ret;
 346
 347     s->type = FTYPE_FILE;
 348     ret = raw_open_common(bs, options, flags, 0, &local_err);
 349     if (error_is_set(&local_err)) {
 350         error_propagate(errp, local_err);
 351     }
 352     return ret;
 353 }
 354
 355 static int raw_reopen_prepare(BDRVReopenState *state,
 356                               BlockReopenQueue *queue, Error **errp)
 357 {
 358     BDRVRawState *s;
 359     BDRVRawReopenState *raw_s;
 360     int ret = 0;
 361
 362     assert(state != NULL);
 363     assert(state->bs != NULL);
 364
 365     s = state->bs->opaque;
 366
 367     state->opaque = g_malloc0(sizeof(BDRVRawReopenState));
 368     raw_s = state->opaque;
 369
 370 #ifdef CONFIG_LINUX_AIO
 371     raw_s->use_aio = s->use_aio;
 372
 373     /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
 374      * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
 375      * won't override aio_ctx if aio_ctx is non-NULL */
 376     if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
 377         error_setg(errp, "Could not set AIO state");
 378         return -1;
 379     }
 380 #endif
 381
 382     if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
 383         raw_s->open_flags |= O_NONBLOCK;
 384     }
 385
 386     raw_parse_flags(state->flags, &raw_s->open_flags);
 387
 388     raw_s->fd = -1;
 389
 390     int fcntl_flags = O_APPEND | O_NONBLOCK;
 391 #ifdef O_NOATIME
 392     fcntl_flags |= O_NOATIME;
 393 #endif
 394
 395 #ifdef O_ASYNC
 396     /* Not all operating systems have O_ASYNC, and those that don't
 397      * will not let us track the state into raw_s->open_flags (typically
 398      * you achieve the same effect with an ioctl, for example I_SETSIG
 399      * on Solaris). But we do not use O_ASYNC, so that's fine.
 400      */
 401     assert((s->open_flags & O_ASYNC) == 0);
 402 #endif
 403
 404     if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
 405         /* dup the original fd */
 406         /* TODO: use qemu fcntl wrapper */
 407 #ifdef F_DUPFD_CLOEXEC
 408         raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
 409 #else
 410         raw_s->fd = dup(s->fd);
 411         if (raw_s->fd != -1) {
 412             qemu_set_cloexec(raw_s->fd);
 413         }
 414 #endif
 415         if (raw_s->fd >= 0) {
 416             ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
 417             if (ret) {
 418                 qemu_close(raw_s->fd);
 419                 raw_s->fd = -1;
 420             }
 421         }
 422     }
 423
 424     /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
 425     if (raw_s->fd == -1) {
 426         assert(!(raw_s->open_flags & O_CREAT));
 427         raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
 428         if (raw_s->fd == -1) {
 429             error_setg_errno(errp, errno, "Could not reopen file");
 430             ret = -1;
 431         }
 432     }
 433     return ret;
 434 }
 435
 436
 437 static void raw_reopen_commit(BDRVReopenState *state)
 438 {
 439     BDRVRawReopenState *raw_s = state->opaque;
 440     BDRVRawState *s = state->bs->opaque;
 441
 442     s->open_flags = raw_s->open_flags;
 443
 444     qemu_close(s->fd);
 445     s->fd = raw_s->fd;
 446 #ifdef CONFIG_LINUX_AIO
 447     s->use_aio = raw_s->use_aio;
 448 #endif
 449
 450     g_free(state->opaque);
 451     state->opaque = NULL;
 452 }
 453
 454
 455 static void raw_reopen_abort(BDRVReopenState *state)
 456 {
 457     BDRVRawReopenState *raw_s = state->opaque;
 458
 459      /* nothing to do if NULL, we didn't get far enough */
 460     if (raw_s == NULL) {
 461         return;
 462     }
 463
 464     if (raw_s->fd >= 0) {
 465         qemu_close(raw_s->fd);
 466         raw_s->fd = -1;
 467     }
 468     g_free(state->opaque);
 469     state->opaque = NULL;
 470 }
 471
 472
 473 /* XXX: use host sector size if necessary with:
 474 #ifdef DIOCGSECTORSIZE
 475         {
 476             unsigned int sectorsize = 512;
 477             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 478                 sectorsize > bufsize)
 479                 bufsize = sectorsize;
 480         }
 481 #endif
 482 #ifdef CONFIG_COCOA
 483         uint32_t blockSize = 512;
 484         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 485             bufsize = blockSize;
 486         }
 487 #endif
 488 */
 489
 490 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
 491 {
 492     int ret;
 493
 494     ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
 495     if (ret == -1) {
 496         return -errno;
 497     }
 498
 499     return 0;
 500 }
 501
 502 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
 503 {
 504     int ret;
 505
 506     ret = qemu_fdatasync(aiocb->aio_fildes);
 507     if (ret == -1) {
 508         return -errno;
 509     }
 510     return 0;
 511 }
 512
 513 #ifdef CONFIG_PREADV
 514
 515 static bool preadv_present = true;
 516
 517 static ssize_t
 518 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 519 {
 520     return preadv(fd, iov, nr_iov, offset);
 521 }
 522
 523 static ssize_t
 524 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 525 {
 526     return pwritev(fd, iov, nr_iov, offset);
 527 }
 528
 529 #else
 530
 531 static bool preadv_present = false;
 532
 533 static ssize_t
 534 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 535 {
 536     return -ENOSYS;
 537 }
 538
 539 static ssize_t
 540 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 541 {
 542     return -ENOSYS;
 543 }
 544
 545 #endif
 546
 547 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
 548 {
 549     ssize_t len;
 550
 551     do {
 552         if (aiocb->aio_type & QEMU_AIO_WRITE)
 553             len = qemu_pwritev(aiocb->aio_fildes,
 554                                aiocb->aio_iov,
 555                                aiocb->aio_niov,
 556                                aiocb->aio_offset);
 557          else
 558             len = qemu_preadv(aiocb->aio_fildes,
 559                               aiocb->aio_iov,
 560                               aiocb->aio_niov,
 561                               aiocb->aio_offset);
 562     } while (len == -1 && errno == EINTR);
 563
 564     if (len == -1) {
 565         return -errno;
 566     }
 567     return len;
 568 }
 569
 570 /*
 571  * Read/writes the data to/from a given linear buffer.
 572  *
 573  * Returns the number of bytes handles or -errno in case of an error. Short
 574  * reads are only returned if the end of the file is reached.
 575  */
 576 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
 577 {
 578     ssize_t offset = 0;
 579     ssize_t len;
 580
 581     while (offset < aiocb->aio_nbytes) {
 582         if (aiocb->aio_type & QEMU_AIO_WRITE) {
 583             len = pwrite(aiocb->aio_fildes,
 584                          (const char *)buf + offset,
 585                          aiocb->aio_nbytes - offset,
 586                          aiocb->aio_offset + offset);
 587         } else {
 588             len = pread(aiocb->aio_fildes,
 589                         buf + offset,
 590                         aiocb->aio_nbytes - offset,
 591                         aiocb->aio_offset + offset);
 592         }
 593         if (len == -1 && errno == EINTR) {
 594             continue;
 595         } else if (len == -1) {
 596             offset = -errno;
 597             break;
 598         } else if (len == 0) {
 599             break;
 600         }
 601         offset += len;
 602     }
 603
 604     return offset;
 605 }
 606
 607 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
 608 {
 609     ssize_t nbytes;
 610     char *buf;
 611
 612     if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
 613         /*
 614          * If there is just a single buffer, and it is properly aligned
 615          * we can just use plain pread/pwrite without any problems.
 616          */
 617         if (aiocb->aio_niov == 1) {
 618              return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
 619         }
 620         /*
 621          * We have more than one iovec, and all are properly aligned.
 622          *
 623          * Try preadv/pwritev first and fall back to linearizing the
 624          * buffer if it's not supported.
 625          */
 626         if (preadv_present) {
 627             nbytes = handle_aiocb_rw_vector(aiocb);
 628             if (nbytes == aiocb->aio_nbytes ||
 629                 (nbytes < 0 && nbytes != -ENOSYS)) {
 630                 return nbytes;
 631             }
 632             preadv_present = false;
 633         }
 634
 635         /*
 636          * XXX(hch): short read/write.  no easy way to handle the reminder
 637          * using these interfaces.  For now retry using plain
 638          * pread/pwrite?
 639          */
 640     }
 641
 642     /*
 643      * Ok, we have to do it the hard way, copy all segments into
 644      * a single aligned buffer.
 645      */
 646     buf = qemu_blockalign(aiocb->bs, aiocb->aio_nbytes);
 647     if (aiocb->aio_type & QEMU_AIO_WRITE) {
 648         char *p = buf;
 649         int i;
 650
 651         for (i = 0; i < aiocb->aio_niov; ++i) {
 652             memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
 653             p += aiocb->aio_iov[i].iov_len;
 654         }
 655     }
 656
 657     nbytes = handle_aiocb_rw_linear(aiocb, buf);
 658     if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
 659         char *p = buf;
 660         size_t count = aiocb->aio_nbytes, copy;
 661         int i;
 662
 663         for (i = 0; i < aiocb->aio_niov && count; ++i) {
 664             copy = count;
 665             if (copy > aiocb->aio_iov[i].iov_len) {
 666                 copy = aiocb->aio_iov[i].iov_len;
 667             }
 668             memcpy(aiocb->aio_iov[i].iov_base, p, copy);
 669             p     += copy;
 670             count -= copy;
 671         }
 672     }
 673     qemu_vfree(buf);
 674
 675     return nbytes;
 676 }
 677
 678 #ifdef CONFIG_XFS
 679 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
 680 {
 681     struct xfs_flock64 fl;
 682
 683     memset(&fl, 0, sizeof(fl));
 684     fl.l_whence = SEEK_SET;
 685     fl.l_start = offset;
 686     fl.l_len = bytes;
 687
 688     if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
 689         DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
 690         return -errno;
 691     }
 692
 693     return 0;
 694 }
 695 #endif
 696
 697 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
 698 {
 699     int ret = -EOPNOTSUPP;
 700     BDRVRawState *s = aiocb->bs->opaque;
 701
 702     if (s->has_discard == 0) {
 703         return 0;
 704     }
 705
 706     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
 707 #ifdef BLKDISCARD
 708         do {
 709             uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
 710             if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
 711                 return 0;
 712             }
 713         } while (errno == EINTR);
 714
 715         ret = -errno;
 716 #endif
 717     } else {
 718 #ifdef CONFIG_XFS
 719         if (s->is_xfs) {
 720             return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
 721         }
 722 #endif
 723
 724 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
 725         do {
 726             if (fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
 727                           aiocb->aio_offset, aiocb->aio_nbytes) == 0) {
 728                 return 0;
 729             }
 730         } while (errno == EINTR);
 731
 732         ret = -errno;
 733 #endif
 734     }
 735
 736     if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
 737         ret == -ENOTTY) {
 738         s->has_discard = 0;
 739         ret = 0;
 740     }
 741     return ret;
 742 }
 743
 744 static int aio_worker(void *arg)
 745 {
 746     RawPosixAIOData *aiocb = arg;
 747     ssize_t ret = 0;
 748
 749     switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
 750     case QEMU_AIO_READ:
 751         ret = handle_aiocb_rw(aiocb);
 752         if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) {
 753             iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
 754                       0, aiocb->aio_nbytes - ret);
 755
 756             ret = aiocb->aio_nbytes;
 757         }
 758         if (ret == aiocb->aio_nbytes) {
 759             ret = 0;
 760         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 761             ret = -EINVAL;
 762         }
 763         break;
 764     case QEMU_AIO_WRITE:
 765         ret = handle_aiocb_rw(aiocb);
 766         if (ret == aiocb->aio_nbytes) {
 767             ret = 0;
 768         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 769             ret = -EINVAL;
 770         }
 771         break;
 772     case QEMU_AIO_FLUSH:
 773         ret = handle_aiocb_flush(aiocb);
 774         break;
 775     case QEMU_AIO_IOCTL:
 776         ret = handle_aiocb_ioctl(aiocb);
 777         break;
 778     case QEMU_AIO_DISCARD:
 779         ret = handle_aiocb_discard(aiocb);
 780         break;
 781     default:
 782         fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
 783         ret = -EINVAL;
 784         break;
 785     }
 786
 787     g_slice_free(RawPosixAIOData, aiocb);
 788     return ret;
 789 }
 790
 791 static BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd,
 792         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 793         BlockDriverCompletionFunc *cb, void *opaque, int type)
 794 {
 795     RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
 796     ThreadPool *pool;
 797
 798     acb->bs = bs;
 799     acb->aio_type = type;
 800     acb->aio_fildes = fd;
 801
 802     if (qiov) {
 803         acb->aio_iov = qiov->iov;
 804         acb->aio_niov = qiov->niov;
 805     }
 806     acb->aio_nbytes = nb_sectors * 512;
 807     acb->aio_offset = sector_num * 512;
 808
 809     trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
 810     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
 811     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
 812 }
 813
 814 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
 815         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 816         BlockDriverCompletionFunc *cb, void *opaque, int type)
 817 {
 818     BDRVRawState *s = bs->opaque;
 819
 820     if (fd_open(bs) < 0)
 821         return NULL;
 822
 823     /*
 824      * If O_DIRECT is used the buffer needs to be aligned on a sector
 825      * boundary.  Check if this is the case or tell the low-level
 826      * driver that it needs to copy the buffer.
 827      */
 828     if ((bs->open_flags & BDRV_O_NOCACHE)) {
 829         if (!bdrv_qiov_is_aligned(bs, qiov)) {
 830             type |= QEMU_AIO_MISALIGNED;
 831 #ifdef CONFIG_LINUX_AIO
 832         } else if (s->use_aio) {
 833             return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
 834                                nb_sectors, cb, opaque, type);
 835 #endif
 836         }
 837     }
 838
 839     return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
 840                        cb, opaque, type);
 841 }
 842
 843 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
 844         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 845         BlockDriverCompletionFunc *cb, void *opaque)
 846 {
 847     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 848                           cb, opaque, QEMU_AIO_READ);
 849 }
 850
 851 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
 852         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 853         BlockDriverCompletionFunc *cb, void *opaque)
 854 {
 855     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 856                           cb, opaque, QEMU_AIO_WRITE);
 857 }
 858
 859 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
 860         BlockDriverCompletionFunc *cb, void *opaque)
 861 {
 862     BDRVRawState *s = bs->opaque;
 863
 864     if (fd_open(bs) < 0)
 865         return NULL;
 866
 867     return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
 868 }
 869
 870 static void raw_close(BlockDriverState *bs)
 871 {
 872     BDRVRawState *s = bs->opaque;
 873     if (s->fd >= 0) {
 874         qemu_close(s->fd);
 875         s->fd = -1;
 876     }
 877 }
 878
 879 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 880 {
 881     BDRVRawState *s = bs->opaque;
 882     struct stat st;
 883
 884     if (fstat(s->fd, &st)) {
 885         return -errno;
 886     }
 887
 888     if (S_ISREG(st.st_mode)) {
 889         if (ftruncate(s->fd, offset) < 0) {
 890             return -errno;
 891         }
 892     } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 893        if (offset > raw_getlength(bs)) {
 894            return -EINVAL;
 895        }
 896     } else {
 897         return -ENOTSUP;
 898     }
 899
 900     return 0;
 901 }
 902
 903 #ifdef __OpenBSD__
 904 static int64_t raw_getlength(BlockDriverState *bs)
 905 {
 906     BDRVRawState *s = bs->opaque;
 907     int fd = s->fd;
 908     struct stat st;
 909
 910     if (fstat(fd, &st))
 911         return -1;
 912     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 913         struct disklabel dl;
 914
 915         if (ioctl(fd, DIOCGDINFO, &dl))
 916             return -1;
 917         return (uint64_t)dl.d_secsize *
 918             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 919     } else
 920         return st.st_size;
 921 }
 922 #elif defined(__NetBSD__)
 923 static int64_t raw_getlength(BlockDriverState *bs)
 924 {
 925     BDRVRawState *s = bs->opaque;
 926     int fd = s->fd;
 927     struct stat st;
 928
 929     if (fstat(fd, &st))
 930         return -1;
 931     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 932         struct dkwedge_info dkw;
 933
 934         if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
 935             return dkw.dkw_size * 512;
 936         } else {
 937             struct disklabel dl;
 938
 939             if (ioctl(fd, DIOCGDINFO, &dl))
 940                 return -1;
 941             return (uint64_t)dl.d_secsize *
 942                 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 943         }
 944     } else
 945         return st.st_size;
 946 }
 947 #elif defined(__sun__)
 948 static int64_t raw_getlength(BlockDriverState *bs)
 949 {
 950     BDRVRawState *s = bs->opaque;
 951     struct dk_minfo minfo;
 952     int ret;
 953
 954     ret = fd_open(bs);
 955     if (ret < 0) {
 956         return ret;
 957     }
 958
 959     /*
 960      * Use the DKIOCGMEDIAINFO ioctl to read the size.
 961      */
 962     ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
 963     if (ret != -1) {
 964         return minfo.dki_lbsize * minfo.dki_capacity;
 965     }
 966
 967     /*
 968      * There are reports that lseek on some devices fails, but
 969      * irc discussion said that contingency on contingency was overkill.
 970      */
 971     return lseek(s->fd, 0, SEEK_END);
 972 }
 973 #elif defined(CONFIG_BSD)
 974 static int64_t raw_getlength(BlockDriverState *bs)
 975 {
 976     BDRVRawState *s = bs->opaque;
 977     int fd = s->fd;
 978     int64_t size;
 979     struct stat sb;
 980 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 981     int reopened = 0;
 982 #endif
 983     int ret;
 984
 985     ret = fd_open(bs);
 986     if (ret < 0)
 987         return ret;
 988
 989 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 990 again:
 991 #endif
 992     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 993 #ifdef DIOCGMEDIASIZE
 994         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 995 #elif defined(DIOCGPART)
 996         {
 997                 struct partinfo pi;
 998                 if (ioctl(fd, DIOCGPART, &pi) == 0)
 999                         size = pi.media_size;
1000                 else
1001                         size = 0;
1002         }
1003         if (size == 0)
1004 #endif
1005 #if defined(__APPLE__) && defined(__MACH__)
1006         size = LONG_LONG_MAX;
1007 #else
1008         size = lseek(fd, 0LL, SEEK_END);
1009 #endif
1010 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1011         switch(s->type) {
1012         case FTYPE_CD:
1013             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1014             if (size == 2048LL * (unsigned)-1)
1015                 size = 0;
1016             /* XXX no disc?  maybe we need to reopen... */
1017             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1018                 reopened = 1;
1019                 goto again;
1020             }
1021         }
1022 #endif
1023     } else {
1024         size = lseek(fd, 0, SEEK_END);
1025     }
1026     return size;
1027 }
1028 #else
1029 static int64_t raw_getlength(BlockDriverState *bs)
1030 {
1031     BDRVRawState *s = bs->opaque;
1032     int ret;
1033
1034     ret = fd_open(bs);
1035     if (ret < 0) {
1036         return ret;
1037     }
1038
1039     return lseek(s->fd, 0, SEEK_END);
1040 }
1041 #endif
1042
1043 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1044 {
1045     struct stat st;
1046     BDRVRawState *s = bs->opaque;
1047
1048     if (fstat(s->fd, &st) < 0) {
1049         return -errno;
1050     }
1051     return (int64_t)st.st_blocks * 512;
1052 }
1053
1054 static int raw_create(const char *filename, QEMUOptionParameter *options,
1055                       Error **errp)
1056 {
1057     int fd;
1058     int result = 0;
1059     int64_t total_size = 0;
1060
1061     /* Read out options */
1062     while (options && options->name) {
1063         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1064             total_size = options->value.n / BDRV_SECTOR_SIZE;
1065         }
1066         options++;
1067     }
1068
1069     fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
1070                    0644);
1071     if (fd < 0) {
1072         result = -errno;
1073         error_setg_errno(errp, -result, "Could not create file");
1074     } else {
1075         if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
1076             result = -errno;
1077             error_setg_errno(errp, -result, "Could not resize file");
1078         }
1079         if (qemu_close(fd) != 0) {
1080             result = -errno;
1081             error_setg_errno(errp, -result, "Could not close the new file");
1082         }
1083     }
1084     return result;
1085 }
1086
1087 /*
1088  * Returns true iff the specified sector is present in the disk image. Drivers
1089  * not implementing the functionality are assumed to not support backing files,
1090  * hence all their sectors are reported as allocated.
1091  *
1092  * If 'sector_num' is beyond the end of the disk image the return value is 0
1093  * and 'pnum' is set to 0.
1094  *
1095  * 'pnum' is set to the number of sectors (including and immediately following
1096  * the specified sector) that are known to be in the same
1097  * allocated/unallocated state.
1098  *
1099  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
1100  * beyond the end of the disk image it will be clamped.
1101  */
1102 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1103                                             int64_t sector_num,
1104                                             int nb_sectors, int *pnum)
1105 {
1106     off_t start, data, hole;
1107     int64_t ret;
1108
1109     ret = fd_open(bs);
1110     if (ret < 0) {
1111         return ret;
1112     }
1113
1114     start = sector_num * BDRV_SECTOR_SIZE;
1115     ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1116
1117 #ifdef CONFIG_FIEMAP
1118
1119     BDRVRawState *s = bs->opaque;
1120     struct {
1121         struct fiemap fm;
1122         struct fiemap_extent fe;
1123     } f;
1124
1125     f.fm.fm_start = start;
1126     f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE;
1127     f.fm.fm_flags = 0;
1128     f.fm.fm_extent_count = 1;
1129     f.fm.fm_reserved = 0;
1130     if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) {
1131         /* Assume everything is allocated.  */
1132         *pnum = nb_sectors;
1133         return ret;
1134     }
1135
1136     if (f.fm.fm_mapped_extents == 0) {
1137         /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1138          * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1139          */
1140         off_t length = lseek(s->fd, 0, SEEK_END);
1141         hole = f.fm.fm_start;
1142         data = MIN(f.fm.fm_start + f.fm.fm_length, length);
1143     } else {
1144         data = f.fe.fe_logical;
1145         hole = f.fe.fe_logical + f.fe.fe_length;
1146         if (f.fe.fe_flags & FIEMAP_EXTENT_UNWRITTEN) {
1147             ret |= BDRV_BLOCK_ZERO;
1148         }
1149     }
1150
1151 #elif defined SEEK_HOLE && defined SEEK_DATA
1152
1153     BDRVRawState *s = bs->opaque;
1154
1155     hole = lseek(s->fd, start, SEEK_HOLE);
1156     if (hole == -1) {
1157         /* -ENXIO indicates that sector_num was past the end of the file.
1158          * There is a virtual hole there.  */
1159         assert(errno != -ENXIO);
1160
1161         /* Most likely EINVAL.  Assume everything is allocated.  */
1162         *pnum = nb_sectors;
1163         return ret;
1164     }
1165
1166     if (hole > start) {
1167         data = start;
1168     } else {
1169         /* On a hole.  We need another syscall to find its end.  */
1170         data = lseek(s->fd, start, SEEK_DATA);
1171         if (data == -1) {
1172             data = lseek(s->fd, 0, SEEK_END);
1173         }
1174     }
1175 #else
1176     data = 0;
1177     hole = start + nb_sectors * BDRV_SECTOR_SIZE;
1178 #endif
1179
1180     if (data <= start) {
1181         /* On a data extent, compute sectors to the end of the extent.  */
1182         *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1183     } else {
1184         /* On a hole, compute sectors to the beginning of the next extent.  */
1185         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1186         ret &= ~BDRV_BLOCK_DATA;
1187         ret |= BDRV_BLOCK_ZERO;
1188     }
1189
1190     return ret;
1191 }
1192
1193 static coroutine_fn BlockDriverAIOCB *raw_aio_discard(BlockDriverState *bs,
1194     int64_t sector_num, int nb_sectors,
1195     BlockDriverCompletionFunc *cb, void *opaque)
1196 {
1197     BDRVRawState *s = bs->opaque;
1198
1199     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1200                        cb, opaque, QEMU_AIO_DISCARD);
1201 }
1202
1203 static QEMUOptionParameter raw_create_options[] = {
1204     {
1205         .name = BLOCK_OPT_SIZE,
1206         .type = OPT_SIZE,
1207         .help = "Virtual disk size"
1208     },
1209     { NULL }
1210 };
1211
1212 static BlockDriver bdrv_file = {
1213     .format_name = "file",
1214     .protocol_name = "file",
1215     .instance_size = sizeof(BDRVRawState),
1216     .bdrv_needs_filename = true,
1217     .bdrv_probe = NULL, /* no probe for protocols */
1218     .bdrv_file_open = raw_open,
1219     .bdrv_reopen_prepare = raw_reopen_prepare,
1220     .bdrv_reopen_commit = raw_reopen_commit,
1221     .bdrv_reopen_abort = raw_reopen_abort,
1222     .bdrv_close = raw_close,
1223     .bdrv_create = raw_create,
1224     .bdrv_has_zero_init = bdrv_has_zero_init_1,
1225     .bdrv_co_get_block_status = raw_co_get_block_status,
1226
1227     .bdrv_aio_readv = raw_aio_readv,
1228     .bdrv_aio_writev = raw_aio_writev,
1229     .bdrv_aio_flush = raw_aio_flush,
1230     .bdrv_aio_discard = raw_aio_discard,
1231
1232     .bdrv_truncate = raw_truncate,
1233     .bdrv_getlength = raw_getlength,
1234     .bdrv_get_allocated_file_size
1235                         = raw_get_allocated_file_size,
1236
1237     .create_options = raw_create_options,
1238 };
1239
1240 /***********************************************/
1241 /* host device */
1242
1243 #if defined(__APPLE__) && defined(__MACH__)
1244 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1245 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1246
1247 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1248 {
1249     kern_return_t       kernResult;
1250     mach_port_t     masterPort;
1251     CFMutableDictionaryRef  classesToMatch;
1252
1253     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1254     if ( KERN_SUCCESS != kernResult ) {
1255         printf( "IOMasterPort returned %d\n", kernResult );
1256     }
1257
1258     classesToMatch = IOServiceMatching( kIOCDMediaClass );
1259     if ( classesToMatch == NULL ) {
1260         printf( "IOServiceMatching returned a NULL dictionary.\n" );
1261     } else {
1262     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1263     }
1264     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1265     if ( KERN_SUCCESS != kernResult )
1266     {
1267         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1268     }
1269
1270     return kernResult;
1271 }
1272
1273 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
1274 {
1275     io_object_t     nextMedia;
1276     kern_return_t   kernResult = KERN_FAILURE;
1277     *bsdPath = '\0';
1278     nextMedia = IOIteratorNext( mediaIterator );
1279     if ( nextMedia )
1280     {
1281         CFTypeRef   bsdPathAsCFString;
1282     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1283         if ( bsdPathAsCFString ) {
1284             size_t devPathLength;
1285             strcpy( bsdPath, _PATH_DEV );
1286             strcat( bsdPath, "r" );
1287             devPathLength = strlen( bsdPath );
1288             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1289                 kernResult = KERN_SUCCESS;
1290             }
1291             CFRelease( bsdPathAsCFString );
1292         }
1293         IOObjectRelease( nextMedia );
1294     }
1295
1296     return kernResult;
1297 }
1298
1299 #endif
1300
1301 static int hdev_probe_device(const char *filename)
1302 {
1303     struct stat st;
1304
1305     /* allow a dedicated CD-ROM driver to match with a higher priority */
1306     if (strstart(filename, "/dev/cdrom", NULL))
1307         return 50;
1308
1309     if (stat(filename, &st) >= 0 &&
1310             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
1311         return 100;
1312     }
1313
1314     return 0;
1315 }
1316
1317 static int check_hdev_writable(BDRVRawState *s)
1318 {
1319 #if defined(BLKROGET)
1320     /* Linux block devices can be configured "read-only" using blockdev(8).
1321      * This is independent of device node permissions and therefore open(2)
1322      * with O_RDWR succeeds.  Actual writes fail with EPERM.
1323      *
1324      * bdrv_open() is supposed to fail if the disk is read-only.  Explicitly
1325      * check for read-only block devices so that Linux block devices behave
1326      * properly.
1327      */
1328     struct stat st;
1329     int readonly = 0;
1330
1331     if (fstat(s->fd, &st)) {
1332         return -errno;
1333     }
1334
1335     if (!S_ISBLK(st.st_mode)) {
1336         return 0;
1337     }
1338
1339     if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
1340         return -errno;
1341     }
1342
1343     if (readonly) {
1344         return -EACCES;
1345     }
1346 #endif /* defined(BLKROGET) */
1347     return 0;
1348 }
1349
1350 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
1351                      Error **errp)
1352 {
1353     BDRVRawState *s = bs->opaque;
1354     Error *local_err = NULL;
1355     int ret;
1356     const char *filename = qdict_get_str(options, "filename");
1357
1358 #if defined(__APPLE__) && defined(__MACH__)
1359     if (strstart(filename, "/dev/cdrom", NULL)) {
1360         kern_return_t kernResult;
1361         io_iterator_t mediaIterator;
1362         char bsdPath[ MAXPATHLEN ];
1363         int fd;
1364
1365         kernResult = FindEjectableCDMedia( &mediaIterator );
1366         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1367
1368         if ( bsdPath[ 0 ] != '\0' ) {
1369             strcat(bsdPath,"s0");
1370             /* some CDs don't have a partition 0 */
1371             fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1372             if (fd < 0) {
1373                 bsdPath[strlen(bsdPath)-1] = '1';
1374             } else {
1375                 qemu_close(fd);
1376             }
1377             filename = bsdPath;
1378             qdict_put(options, "filename", qstring_from_str(filename));
1379         }
1380
1381         if ( mediaIterator )
1382             IOObjectRelease( mediaIterator );
1383     }
1384 #endif
1385
1386     s->type = FTYPE_FILE;
1387 #if defined(__linux__)
1388     {
1389         char resolved_path[ MAXPATHLEN ], *temp;
1390
1391         temp = realpath(filename, resolved_path);
1392         if (temp && strstart(temp, "/dev/sg", NULL)) {
1393             bs->sg = 1;
1394         }
1395     }
1396 #endif
1397
1398     ret = raw_open_common(bs, options, flags, 0, &local_err);
1399     if (ret < 0) {
1400         if (error_is_set(&local_err)) {
1401             error_propagate(errp, local_err);
1402         }
1403         return ret;
1404     }
1405
1406     if (flags & BDRV_O_RDWR) {
1407         ret = check_hdev_writable(s);
1408         if (ret < 0) {
1409             raw_close(bs);
1410             error_setg_errno(errp, -ret, "The device is not writable");
1411             return ret;
1412         }
1413     }
1414
1415     return ret;
1416 }
1417
1418 #if defined(__linux__)
1419 /* Note: we do not have a reliable method to detect if the floppy is
1420    present. The current method is to try to open the floppy at every
1421    I/O and to keep it opened during a few hundreds of ms. */
1422 static int fd_open(BlockDriverState *bs)
1423 {
1424     BDRVRawState *s = bs->opaque;
1425     int last_media_present;
1426
1427     if (s->type != FTYPE_FD)
1428         return 0;
1429     last_media_present = (s->fd >= 0);
1430     if (s->fd >= 0 &&
1431         (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1432         qemu_close(s->fd);
1433         s->fd = -1;
1434 #ifdef DEBUG_FLOPPY
1435         printf("Floppy closed\n");
1436 #endif
1437     }
1438     if (s->fd < 0) {
1439         if (s->fd_got_error &&
1440             (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1441 #ifdef DEBUG_FLOPPY
1442             printf("No floppy (open delayed)\n");
1443 #endif
1444             return -EIO;
1445         }
1446         s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
1447         if (s->fd < 0) {
1448             s->fd_error_time = get_clock();
1449             s->fd_got_error = 1;
1450             if (last_media_present)
1451                 s->fd_media_changed = 1;
1452 #ifdef DEBUG_FLOPPY
1453             printf("No floppy\n");
1454 #endif
1455             return -EIO;
1456         }
1457 #ifdef DEBUG_FLOPPY
1458         printf("Floppy opened\n");
1459 #endif
1460     }
1461     if (!last_media_present)
1462         s->fd_media_changed = 1;
1463     s->fd_open_time = get_clock();
1464     s->fd_got_error = 0;
1465     return 0;
1466 }
1467
1468 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1469 {
1470     BDRVRawState *s = bs->opaque;
1471
1472     return ioctl(s->fd, req, buf);
1473 }
1474
1475 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1476         unsigned long int req, void *buf,
1477         BlockDriverCompletionFunc *cb, void *opaque)
1478 {
1479     BDRVRawState *s = bs->opaque;
1480     RawPosixAIOData *acb;
1481     ThreadPool *pool;
1482
1483     if (fd_open(bs) < 0)
1484         return NULL;
1485
1486     acb = g_slice_new(RawPosixAIOData);
1487     acb->bs = bs;
1488     acb->aio_type = QEMU_AIO_IOCTL;
1489     acb->aio_fildes = s->fd;
1490     acb->aio_offset = 0;
1491     acb->aio_ioctl_buf = buf;
1492     acb->aio_ioctl_cmd = req;
1493     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1494     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1495 }
1496
1497 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1498 static int fd_open(BlockDriverState *bs)
1499 {
1500     BDRVRawState *s = bs->opaque;
1501
1502     /* this is just to ensure s->fd is sane (its called by io ops) */
1503     if (s->fd >= 0)
1504         return 0;
1505     return -EIO;
1506 }
1507 #else /* !linux && !FreeBSD */
1508
1509 static int fd_open(BlockDriverState *bs)
1510 {
1511     return 0;
1512 }
1513
1514 #endif /* !linux && !FreeBSD */
1515
1516 static coroutine_fn BlockDriverAIOCB *hdev_aio_discard(BlockDriverState *bs,
1517     int64_t sector_num, int nb_sectors,
1518     BlockDriverCompletionFunc *cb, void *opaque)
1519 {
1520     BDRVRawState *s = bs->opaque;
1521
1522     if (fd_open(bs) < 0) {
1523         return NULL;
1524     }
1525     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1526                        cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1527 }
1528
1529 static int hdev_create(const char *filename, QEMUOptionParameter *options,
1530                        Error **errp)
1531 {
1532     int fd;
1533     int ret = 0;
1534     struct stat stat_buf;
1535     int64_t total_size = 0;
1536
1537     /* Read out options */
1538     while (options && options->name) {
1539         if (!strcmp(options->name, "size")) {
1540             total_size = options->value.n / BDRV_SECTOR_SIZE;
1541         }
1542         options++;
1543     }
1544
1545     fd = qemu_open(filename, O_WRONLY | O_BINARY);
1546     if (fd < 0) {
1547         ret = -errno;
1548         error_setg_errno(errp, -ret, "Could not open device");
1549         return ret;
1550     }
1551
1552     if (fstat(fd, &stat_buf) < 0) {
1553         ret = -errno;
1554         error_setg_errno(errp, -ret, "Could not stat device");
1555     } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
1556         error_setg(errp,
1557                    "The given file is neither a block nor a character device");
1558         ret = -ENODEV;
1559     } else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE) {
1560         error_setg(errp, "Device is too small");
1561         ret = -ENOSPC;
1562     }
1563
1564     qemu_close(fd);
1565     return ret;
1566 }
1567
1568 static BlockDriver bdrv_host_device = {
1569     .format_name        = "host_device",
1570     .protocol_name        = "host_device",
1571     .instance_size      = sizeof(BDRVRawState),
1572     .bdrv_needs_filename = true,
1573     .bdrv_probe_device  = hdev_probe_device,
1574     .bdrv_file_open     = hdev_open,
1575     .bdrv_close         = raw_close,
1576     .bdrv_reopen_prepare = raw_reopen_prepare,
1577     .bdrv_reopen_commit  = raw_reopen_commit,
1578     .bdrv_reopen_abort   = raw_reopen_abort,
1579     .bdrv_create        = hdev_create,
1580     .create_options     = raw_create_options,
1581
1582     .bdrv_aio_readv     = raw_aio_readv,
1583     .bdrv_aio_writev    = raw_aio_writev,
1584     .bdrv_aio_flush     = raw_aio_flush,
1585     .bdrv_aio_discard   = hdev_aio_discard,
1586
1587     .bdrv_truncate      = raw_truncate,
1588     .bdrv_getlength     = raw_getlength,
1589     .bdrv_get_allocated_file_size
1590                         = raw_get_allocated_file_size,
1591
1592     /* generic scsi device */
1593 #ifdef __linux__
1594     .bdrv_ioctl         = hdev_ioctl,
1595     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1596 #endif
1597 };
1598
1599 #ifdef __linux__
1600 static int floppy_open(BlockDriverState *bs, QDict *options, int flags,
1601                        Error **errp)
1602 {
1603     BDRVRawState *s = bs->opaque;
1604     Error *local_err = NULL;
1605     int ret;
1606
1607     s->type = FTYPE_FD;
1608
1609     /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1610     ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
1611     if (ret) {
1612         if (error_is_set(&local_err)) {
1613             error_propagate(errp, local_err);
1614         }
1615         return ret;
1616     }
1617
1618     /* close fd so that we can reopen it as needed */
1619     qemu_close(s->fd);
1620     s->fd = -1;
1621     s->fd_media_changed = 1;
1622
1623     return 0;
1624 }
1625
1626 static int floppy_probe_device(const char *filename)
1627 {
1628     int fd, ret;
1629     int prio = 0;
1630     struct floppy_struct fdparam;
1631     struct stat st;
1632
1633     if (strstart(filename, "/dev/fd", NULL) &&
1634         !strstart(filename, "/dev/fdset/", NULL)) {
1635         prio = 50;
1636     }
1637
1638     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1639     if (fd < 0) {
1640         goto out;
1641     }
1642     ret = fstat(fd, &st);
1643     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1644         goto outc;
1645     }
1646
1647     /* Attempt to detect via a floppy specific ioctl */
1648     ret = ioctl(fd, FDGETPRM, &fdparam);
1649     if (ret >= 0)
1650         prio = 100;
1651
1652 outc:
1653     qemu_close(fd);
1654 out:
1655     return prio;
1656 }
1657
1658
1659 static int floppy_is_inserted(BlockDriverState *bs)
1660 {
1661     return fd_open(bs) >= 0;
1662 }
1663
1664 static int floppy_media_changed(BlockDriverState *bs)
1665 {
1666     BDRVRawState *s = bs->opaque;
1667     int ret;
1668
1669     /*
1670      * XXX: we do not have a true media changed indication.
1671      * It does not work if the floppy is changed without trying to read it.
1672      */
1673     fd_open(bs);
1674     ret = s->fd_media_changed;
1675     s->fd_media_changed = 0;
1676 #ifdef DEBUG_FLOPPY
1677     printf("Floppy changed=%d\n", ret);
1678 #endif
1679     return ret;
1680 }
1681
1682 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
1683 {
1684     BDRVRawState *s = bs->opaque;
1685     int fd;
1686
1687     if (s->fd >= 0) {
1688         qemu_close(s->fd);
1689         s->fd = -1;
1690     }
1691     fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
1692     if (fd >= 0) {
1693         if (ioctl(fd, FDEJECT, 0) < 0)
1694             perror("FDEJECT");
1695         qemu_close(fd);
1696     }
1697 }
1698
1699 static BlockDriver bdrv_host_floppy = {
1700     .format_name        = "host_floppy",
1701     .protocol_name      = "host_floppy",
1702     .instance_size      = sizeof(BDRVRawState),
1703     .bdrv_needs_filename = true,
1704     .bdrv_probe_device  = floppy_probe_device,
1705     .bdrv_file_open     = floppy_open,
1706     .bdrv_close         = raw_close,
1707     .bdrv_reopen_prepare = raw_reopen_prepare,
1708     .bdrv_reopen_commit  = raw_reopen_commit,
1709     .bdrv_reopen_abort   = raw_reopen_abort,
1710     .bdrv_create        = hdev_create,
1711     .create_options     = raw_create_options,
1712
1713     .bdrv_aio_readv     = raw_aio_readv,
1714     .bdrv_aio_writev    = raw_aio_writev,
1715     .bdrv_aio_flush     = raw_aio_flush,
1716
1717     .bdrv_truncate      = raw_truncate,
1718     .bdrv_getlength     = raw_getlength,
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     .bdrv_get_allocated_file_size
1829                         = raw_get_allocated_file_size,
1830
1831     /* removable device support */
1832     .bdrv_is_inserted   = cdrom_is_inserted,
1833     .bdrv_eject         = cdrom_eject,
1834     .bdrv_lock_medium   = cdrom_lock_medium,
1835
1836     /* generic scsi device */
1837     .bdrv_ioctl         = hdev_ioctl,
1838     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1839 };
1840 #endif /* __linux__ */
1841
1842 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1843 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags)
1844 {
1845     BDRVRawState *s = bs->opaque;
1846     Error *local_err = NULL;
1847     int ret;
1848
1849     s->type = FTYPE_CD;
1850
1851     ret = raw_open_common(bs, options, flags, 0, &local_err);
1852     if (ret) {
1853         if (error_is_set(&local_err)) {
1854             error_propagate(errp, local_err);
1855         }
1856         return ret;
1857     }
1858
1859     /* make sure the door isn't locked at this time */
1860     ioctl(s->fd, CDIOCALLOW);
1861     return 0;
1862 }
1863
1864 static int cdrom_probe_device(const char *filename)
1865 {
1866     if (strstart(filename, "/dev/cd", NULL) ||
1867             strstart(filename, "/dev/acd", NULL))
1868         return 100;
1869     return 0;
1870 }
1871
1872 static int cdrom_reopen(BlockDriverState *bs)
1873 {
1874     BDRVRawState *s = bs->opaque;
1875     int fd;
1876
1877     /*
1878      * Force reread of possibly changed/newly loaded disc,
1879      * FreeBSD seems to not notice sometimes...
1880      */
1881     if (s->fd >= 0)
1882         qemu_close(s->fd);
1883     fd = qemu_open(bs->filename, s->open_flags, 0644);
1884     if (fd < 0) {
1885         s->fd = -1;
1886         return -EIO;
1887     }
1888     s->fd = fd;
1889
1890     /* make sure the door isn't locked at this time */
1891     ioctl(s->fd, CDIOCALLOW);
1892     return 0;
1893 }
1894
1895 static int cdrom_is_inserted(BlockDriverState *bs)
1896 {
1897     return raw_getlength(bs) > 0;
1898 }
1899
1900 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1901 {
1902     BDRVRawState *s = bs->opaque;
1903
1904     if (s->fd < 0)
1905         return;
1906
1907     (void) ioctl(s->fd, CDIOCALLOW);
1908
1909     if (eject_flag) {
1910         if (ioctl(s->fd, CDIOCEJECT) < 0)
1911             perror("CDIOCEJECT");
1912     } else {
1913         if (ioctl(s->fd, CDIOCCLOSE) < 0)
1914             perror("CDIOCCLOSE");
1915     }
1916
1917     cdrom_reopen(bs);
1918 }
1919
1920 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1921 {
1922     BDRVRawState *s = bs->opaque;
1923
1924     if (s->fd < 0)
1925         return;
1926     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1927         /*
1928          * Note: an error can happen if the distribution automatically
1929          * mounts the CD-ROM
1930          */
1931         /* perror("CDROM_LOCKDOOR"); */
1932     }
1933 }
1934
1935 static BlockDriver bdrv_host_cdrom = {
1936     .format_name        = "host_cdrom",
1937     .protocol_name      = "host_cdrom",
1938     .instance_size      = sizeof(BDRVRawState),
1939     .bdrv_needs_filename = true,
1940     .bdrv_probe_device  = cdrom_probe_device,
1941     .bdrv_file_open     = cdrom_open,
1942     .bdrv_close         = raw_close,
1943     .bdrv_reopen_prepare = raw_reopen_prepare,
1944     .bdrv_reopen_commit  = raw_reopen_commit,
1945     .bdrv_reopen_abort   = raw_reopen_abort,
1946     .bdrv_create        = hdev_create,
1947     .create_options     = raw_create_options,
1948
1949     .bdrv_aio_readv     = raw_aio_readv,
1950     .bdrv_aio_writev    = raw_aio_writev,
1951     .bdrv_aio_flush     = raw_aio_flush,
1952
1953     .bdrv_truncate      = raw_truncate,
1954     .bdrv_getlength     = raw_getlength,
1955     .bdrv_get_allocated_file_size
1956                         = raw_get_allocated_file_size,
1957
1958     /* removable device support */
1959     .bdrv_is_inserted   = cdrom_is_inserted,
1960     .bdrv_eject         = cdrom_eject,
1961     .bdrv_lock_medium   = cdrom_lock_medium,
1962 };
1963 #endif /* __FreeBSD__ */
1964
1965 #ifdef CONFIG_LINUX_AIO
1966 /**
1967  * Return the file descriptor for Linux AIO
1968  *
1969  * This function is a layering violation and should be removed when it becomes
1970  * possible to call the block layer outside the global mutex.  It allows the
1971  * caller to hijack the file descriptor so I/O can be performed outside the
1972  * block layer.
1973  */
1974 int raw_get_aio_fd(BlockDriverState *bs)
1975 {
1976     BDRVRawState *s;
1977
1978     if (!bs->drv) {
1979         return -ENOMEDIUM;
1980     }
1981
1982     if (bs->drv == bdrv_find_format("raw")) {
1983         bs = bs->file;
1984     }
1985
1986     /* raw-posix has several protocols so just check for raw_aio_readv */
1987     if (bs->drv->bdrv_aio_readv != raw_aio_readv) {
1988         return -ENOTSUP;
1989     }
1990
1991     s = bs->opaque;
1992     if (!s->use_aio) {
1993         return -ENOTSUP;
1994     }
1995     return s->fd;
1996 }
1997 #endif /* CONFIG_LINUX_AIO */
1998
1999 static void bdrv_file_init(void)
2000 {
2001     /*
2002      * Register all the drivers.  Note that order is important, the driver
2003      * registered last will get probed first.
2004      */
2005     bdrv_register(&bdrv_file);
2006     bdrv_register(&bdrv_host_device);
2007 #ifdef __linux__
2008     bdrv_register(&bdrv_host_floppy);
2009     bdrv_register(&bdrv_host_cdrom);
2010 #endif
2011 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2012     bdrv_register(&bdrv_host_cdrom);
2013 #endif
2014 }
2015
2016 block_init(bdrv_file_init);