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     bool has_write_zeroes:1;
 146     bool discard_zeroes:1;
 147 } BDRVRawState;
 148
 149 typedef struct BDRVRawReopenState {
 150     int fd;
 151     int open_flags;
 152 #ifdef CONFIG_LINUX_AIO
 153     int use_aio;
 154 #endif
 155 } BDRVRawReopenState;
 156
 157 static int fd_open(BlockDriverState *bs);
 158 static int64_t raw_getlength(BlockDriverState *bs);
 159
 160 typedef struct RawPosixAIOData {
 161     BlockDriverState *bs;
 162     int aio_fildes;
 163     union {
 164         struct iovec *aio_iov;
 165         void *aio_ioctl_buf;
 166     };
 167     int aio_niov;
 168     uint64_t aio_nbytes;
 169 #define aio_ioctl_cmd   aio_nbytes /* for QEMU_AIO_IOCTL */
 170     off_t aio_offset;
 171     int aio_type;
 172 } RawPosixAIOData;
 173
 174 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
 175 static int cdrom_reopen(BlockDriverState *bs);
 176 #endif
 177
 178 #if defined(__NetBSD__)
 179 static int raw_normalize_devicepath(const char **filename)
 180 {
 181     static char namebuf[PATH_MAX];
 182     const char *dp, *fname;
 183     struct stat sb;
 184
 185     fname = *filename;
 186     dp = strrchr(fname, '/');
 187     if (lstat(fname, &sb) < 0) {
 188         fprintf(stderr, "%s: stat failed: %s\n",
 189             fname, strerror(errno));
 190         return -errno;
 191     }
 192
 193     if (!S_ISBLK(sb.st_mode)) {
 194         return 0;
 195     }
 196
 197     if (dp == NULL) {
 198         snprintf(namebuf, PATH_MAX, "r%s", fname);
 199     } else {
 200         snprintf(namebuf, PATH_MAX, "%.*s/r%s",
 201             (int)(dp - fname), fname, dp + 1);
 202     }
 203     fprintf(stderr, "%s is a block device", fname);
 204     *filename = namebuf;
 205     fprintf(stderr, ", using %s\n", *filename);
 206
 207     return 0;
 208 }
 209 #else
 210 static int raw_normalize_devicepath(const char **filename)
 211 {
 212     return 0;
 213 }
 214 #endif
 215
 216 static void raw_parse_flags(int bdrv_flags, int *open_flags)
 217 {
 218     assert(open_flags != NULL);
 219
 220     *open_flags |= O_BINARY;
 221     *open_flags &= ~O_ACCMODE;
 222     if (bdrv_flags & BDRV_O_RDWR) {
 223         *open_flags |= O_RDWR;
 224     } else {
 225         *open_flags |= O_RDONLY;
 226     }
 227
 228     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 229      * and O_DIRECT for no caching. */
 230     if ((bdrv_flags & BDRV_O_NOCACHE)) {
 231         *open_flags |= O_DIRECT;
 232     }
 233 }
 234
 235 #ifdef CONFIG_LINUX_AIO
 236 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
 237 {
 238     int ret = -1;
 239     assert(aio_ctx != NULL);
 240     assert(use_aio != NULL);
 241     /*
 242      * Currently Linux do AIO only for files opened with O_DIRECT
 243      * specified so check NOCACHE flag too
 244      */
 245     if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
 246                       (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
 247
 248         /* if non-NULL, laio_init() has already been run */
 249         if (*aio_ctx == NULL) {
 250             *aio_ctx = laio_init();
 251             if (!*aio_ctx) {
 252                 goto error;
 253             }
 254         }
 255         *use_aio = 1;
 256     } else {
 257         *use_aio = 0;
 258     }
 259
 260     ret = 0;
 261
 262 error:
 263     return ret;
 264 }
 265 #endif
 266
 267 static QemuOptsList raw_runtime_opts = {
 268     .name = "raw",
 269     .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
 270     .desc = {
 271         {
 272             .name = "filename",
 273             .type = QEMU_OPT_STRING,
 274             .help = "File name of the image",
 275         },
 276         { /* end of list */ }
 277     },
 278 };
 279
 280 static int raw_open_common(BlockDriverState *bs, QDict *options,
 281                            int bdrv_flags, int open_flags, Error **errp)
 282 {
 283     BDRVRawState *s = bs->opaque;
 284     QemuOpts *opts;
 285     Error *local_err = NULL;
 286     const char *filename;
 287     int fd, ret;
 288     struct stat st;
 289
 290     opts = qemu_opts_create_nofail(&raw_runtime_opts);
 291     qemu_opts_absorb_qdict(opts, options, &local_err);
 292     if (error_is_set(&local_err)) {
 293         error_propagate(errp, local_err);
 294         ret = -EINVAL;
 295         goto fail;
 296     }
 297
 298     filename = qemu_opt_get(opts, "filename");
 299
 300     ret = raw_normalize_devicepath(&filename);
 301     if (ret != 0) {
 302         error_setg_errno(errp, -ret, "Could not normalize device path");
 303         goto fail;
 304     }
 305
 306     s->open_flags = open_flags;
 307     raw_parse_flags(bdrv_flags, &s->open_flags);
 308
 309     s->fd = -1;
 310     fd = qemu_open(filename, s->open_flags, 0644);
 311     if (fd < 0) {
 312         ret = -errno;
 313         if (ret == -EROFS) {
 314             ret = -EACCES;
 315         }
 316         goto fail;
 317     }
 318     s->fd = fd;
 319
 320 #ifdef CONFIG_LINUX_AIO
 321     if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
 322         qemu_close(fd);
 323         ret = -errno;
 324         error_setg_errno(errp, -ret, "Could not set AIO state");
 325         goto fail;
 326     }
 327 #endif
 328
 329     s->has_discard = true;
 330     s->has_write_zeroes = true;
 331
 332     if (fstat(s->fd, &st) < 0) {
 333         error_setg_errno(errp, errno, "Could not stat file");
 334         goto fail;
 335     }
 336     if (S_ISREG(st.st_mode)) {
 337         s->discard_zeroes = true;
 338     }
 339     if (S_ISBLK(st.st_mode)) {
 340 #ifdef BLKDISCARDZEROES
 341         unsigned int arg;
 342         if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
 343             s->discard_zeroes = true;
 344         }
 345 #endif
 346 #ifdef __linux__
 347         /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache.  Do
 348          * not rely on the contents of discarded blocks unless using O_DIRECT.
 349          * Same for BLKZEROOUT.
 350          */
 351         if (!(bs->open_flags & BDRV_O_NOCACHE)) {
 352             s->discard_zeroes = false;
 353             s->has_write_zeroes = false;
 354         }
 355 #endif
 356     }
 357
 358 #ifdef CONFIG_XFS
 359     if (platform_test_xfs_fd(s->fd)) {
 360         s->is_xfs = true;
 361     }
 362 #endif
 363
 364     ret = 0;
 365 fail:
 366     qemu_opts_del(opts);
 367     return ret;
 368 }
 369
 370 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
 371                     Error **errp)
 372 {
 373     BDRVRawState *s = bs->opaque;
 374     Error *local_err = NULL;
 375     int ret;
 376
 377     s->type = FTYPE_FILE;
 378     ret = raw_open_common(bs, options, flags, 0, &local_err);
 379     if (error_is_set(&local_err)) {
 380         error_propagate(errp, local_err);
 381     }
 382     return ret;
 383 }
 384
 385 static int raw_reopen_prepare(BDRVReopenState *state,
 386                               BlockReopenQueue *queue, Error **errp)
 387 {
 388     BDRVRawState *s;
 389     BDRVRawReopenState *raw_s;
 390     int ret = 0;
 391
 392     assert(state != NULL);
 393     assert(state->bs != NULL);
 394
 395     s = state->bs->opaque;
 396
 397     state->opaque = g_malloc0(sizeof(BDRVRawReopenState));
 398     raw_s = state->opaque;
 399
 400 #ifdef CONFIG_LINUX_AIO
 401     raw_s->use_aio = s->use_aio;
 402
 403     /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
 404      * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
 405      * won't override aio_ctx if aio_ctx is non-NULL */
 406     if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
 407         error_setg(errp, "Could not set AIO state");
 408         return -1;
 409     }
 410 #endif
 411
 412     if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
 413         raw_s->open_flags |= O_NONBLOCK;
 414     }
 415
 416     raw_parse_flags(state->flags, &raw_s->open_flags);
 417
 418     raw_s->fd = -1;
 419
 420     int fcntl_flags = O_APPEND | O_NONBLOCK;
 421 #ifdef O_NOATIME
 422     fcntl_flags |= O_NOATIME;
 423 #endif
 424
 425 #ifdef O_ASYNC
 426     /* Not all operating systems have O_ASYNC, and those that don't
 427      * will not let us track the state into raw_s->open_flags (typically
 428      * you achieve the same effect with an ioctl, for example I_SETSIG
 429      * on Solaris). But we do not use O_ASYNC, so that's fine.
 430      */
 431     assert((s->open_flags & O_ASYNC) == 0);
 432 #endif
 433
 434     if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
 435         /* dup the original fd */
 436         /* TODO: use qemu fcntl wrapper */
 437 #ifdef F_DUPFD_CLOEXEC
 438         raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
 439 #else
 440         raw_s->fd = dup(s->fd);
 441         if (raw_s->fd != -1) {
 442             qemu_set_cloexec(raw_s->fd);
 443         }
 444 #endif
 445         if (raw_s->fd >= 0) {
 446             ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
 447             if (ret) {
 448                 qemu_close(raw_s->fd);
 449                 raw_s->fd = -1;
 450             }
 451         }
 452     }
 453
 454     /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
 455     if (raw_s->fd == -1) {
 456         assert(!(raw_s->open_flags & O_CREAT));
 457         raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
 458         if (raw_s->fd == -1) {
 459             error_setg_errno(errp, errno, "Could not reopen file");
 460             ret = -1;
 461         }
 462     }
 463     return ret;
 464 }
 465
 466
 467 static void raw_reopen_commit(BDRVReopenState *state)
 468 {
 469     BDRVRawReopenState *raw_s = state->opaque;
 470     BDRVRawState *s = state->bs->opaque;
 471
 472     s->open_flags = raw_s->open_flags;
 473
 474     qemu_close(s->fd);
 475     s->fd = raw_s->fd;
 476 #ifdef CONFIG_LINUX_AIO
 477     s->use_aio = raw_s->use_aio;
 478 #endif
 479
 480     g_free(state->opaque);
 481     state->opaque = NULL;
 482 }
 483
 484
 485 static void raw_reopen_abort(BDRVReopenState *state)
 486 {
 487     BDRVRawReopenState *raw_s = state->opaque;
 488
 489      /* nothing to do if NULL, we didn't get far enough */
 490     if (raw_s == NULL) {
 491         return;
 492     }
 493
 494     if (raw_s->fd >= 0) {
 495         qemu_close(raw_s->fd);
 496         raw_s->fd = -1;
 497     }
 498     g_free(state->opaque);
 499     state->opaque = NULL;
 500 }
 501
 502
 503 /* XXX: use host sector size if necessary with:
 504 #ifdef DIOCGSECTORSIZE
 505         {
 506             unsigned int sectorsize = 512;
 507             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 508                 sectorsize > bufsize)
 509                 bufsize = sectorsize;
 510         }
 511 #endif
 512 #ifdef CONFIG_COCOA
 513         uint32_t blockSize = 512;
 514         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 515             bufsize = blockSize;
 516         }
 517 #endif
 518 */
 519
 520 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
 521 {
 522     int ret;
 523
 524     ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
 525     if (ret == -1) {
 526         return -errno;
 527     }
 528
 529     return 0;
 530 }
 531
 532 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
 533 {
 534     int ret;
 535
 536     ret = qemu_fdatasync(aiocb->aio_fildes);
 537     if (ret == -1) {
 538         return -errno;
 539     }
 540     return 0;
 541 }
 542
 543 #ifdef CONFIG_PREADV
 544
 545 static bool preadv_present = true;
 546
 547 static ssize_t
 548 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 549 {
 550     return preadv(fd, iov, nr_iov, offset);
 551 }
 552
 553 static ssize_t
 554 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 555 {
 556     return pwritev(fd, iov, nr_iov, offset);
 557 }
 558
 559 #else
 560
 561 static bool preadv_present = false;
 562
 563 static ssize_t
 564 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 565 {
 566     return -ENOSYS;
 567 }
 568
 569 static ssize_t
 570 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 571 {
 572     return -ENOSYS;
 573 }
 574
 575 #endif
 576
 577 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
 578 {
 579     ssize_t len;
 580
 581     do {
 582         if (aiocb->aio_type & QEMU_AIO_WRITE)
 583             len = qemu_pwritev(aiocb->aio_fildes,
 584                                aiocb->aio_iov,
 585                                aiocb->aio_niov,
 586                                aiocb->aio_offset);
 587          else
 588             len = qemu_preadv(aiocb->aio_fildes,
 589                               aiocb->aio_iov,
 590                               aiocb->aio_niov,
 591                               aiocb->aio_offset);
 592     } while (len == -1 && errno == EINTR);
 593
 594     if (len == -1) {
 595         return -errno;
 596     }
 597     return len;
 598 }
 599
 600 /*
 601  * Read/writes the data to/from a given linear buffer.
 602  *
 603  * Returns the number of bytes handles or -errno in case of an error. Short
 604  * reads are only returned if the end of the file is reached.
 605  */
 606 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
 607 {
 608     ssize_t offset = 0;
 609     ssize_t len;
 610
 611     while (offset < aiocb->aio_nbytes) {
 612         if (aiocb->aio_type & QEMU_AIO_WRITE) {
 613             len = pwrite(aiocb->aio_fildes,
 614                          (const char *)buf + offset,
 615                          aiocb->aio_nbytes - offset,
 616                          aiocb->aio_offset + offset);
 617         } else {
 618             len = pread(aiocb->aio_fildes,
 619                         buf + offset,
 620                         aiocb->aio_nbytes - offset,
 621                         aiocb->aio_offset + offset);
 622         }
 623         if (len == -1 && errno == EINTR) {
 624             continue;
 625         } else if (len == -1) {
 626             offset = -errno;
 627             break;
 628         } else if (len == 0) {
 629             break;
 630         }
 631         offset += len;
 632     }
 633
 634     return offset;
 635 }
 636
 637 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
 638 {
 639     ssize_t nbytes;
 640     char *buf;
 641
 642     if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
 643         /*
 644          * If there is just a single buffer, and it is properly aligned
 645          * we can just use plain pread/pwrite without any problems.
 646          */
 647         if (aiocb->aio_niov == 1) {
 648              return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
 649         }
 650         /*
 651          * We have more than one iovec, and all are properly aligned.
 652          *
 653          * Try preadv/pwritev first and fall back to linearizing the
 654          * buffer if it's not supported.
 655          */
 656         if (preadv_present) {
 657             nbytes = handle_aiocb_rw_vector(aiocb);
 658             if (nbytes == aiocb->aio_nbytes ||
 659                 (nbytes < 0 && nbytes != -ENOSYS)) {
 660                 return nbytes;
 661             }
 662             preadv_present = false;
 663         }
 664
 665         /*
 666          * XXX(hch): short read/write.  no easy way to handle the reminder
 667          * using these interfaces.  For now retry using plain
 668          * pread/pwrite?
 669          */
 670     }
 671
 672     /*
 673      * Ok, we have to do it the hard way, copy all segments into
 674      * a single aligned buffer.
 675      */
 676     buf = qemu_blockalign(aiocb->bs, aiocb->aio_nbytes);
 677     if (aiocb->aio_type & QEMU_AIO_WRITE) {
 678         char *p = buf;
 679         int i;
 680
 681         for (i = 0; i < aiocb->aio_niov; ++i) {
 682             memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
 683             p += aiocb->aio_iov[i].iov_len;
 684         }
 685     }
 686
 687     nbytes = handle_aiocb_rw_linear(aiocb, buf);
 688     if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
 689         char *p = buf;
 690         size_t count = aiocb->aio_nbytes, copy;
 691         int i;
 692
 693         for (i = 0; i < aiocb->aio_niov && count; ++i) {
 694             copy = count;
 695             if (copy > aiocb->aio_iov[i].iov_len) {
 696                 copy = aiocb->aio_iov[i].iov_len;
 697             }
 698             memcpy(aiocb->aio_iov[i].iov_base, p, copy);
 699             p     += copy;
 700             count -= copy;
 701         }
 702     }
 703     qemu_vfree(buf);
 704
 705     return nbytes;
 706 }
 707
 708 #ifdef CONFIG_XFS
 709 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
 710 {
 711     struct xfs_flock64 fl;
 712
 713     memset(&fl, 0, sizeof(fl));
 714     fl.l_whence = SEEK_SET;
 715     fl.l_start = offset;
 716     fl.l_len = bytes;
 717
 718     if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
 719         DEBUG_BLOCK_PRINT("cannot write zero range (%s)\n", strerror(errno));
 720         return -errno;
 721     }
 722
 723     return 0;
 724 }
 725
 726 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
 727 {
 728     struct xfs_flock64 fl;
 729
 730     memset(&fl, 0, sizeof(fl));
 731     fl.l_whence = SEEK_SET;
 732     fl.l_start = offset;
 733     fl.l_len = bytes;
 734
 735     if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
 736         DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
 737         return -errno;
 738     }
 739
 740     return 0;
 741 }
 742 #endif
 743
 744 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
 745 {
 746     int ret = -EOPNOTSUPP;
 747     BDRVRawState *s = aiocb->bs->opaque;
 748
 749     if (s->has_write_zeroes == 0) {
 750         return -ENOTSUP;
 751     }
 752
 753     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
 754 #ifdef BLKZEROOUT
 755         do {
 756             uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
 757             if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
 758                 return 0;
 759             }
 760         } while (errno == EINTR);
 761
 762         ret = -errno;
 763 #endif
 764     } else {
 765 #ifdef CONFIG_XFS
 766         if (s->is_xfs) {
 767             return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
 768         }
 769 #endif
 770     }
 771
 772     if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
 773         ret == -ENOTTY) {
 774         s->has_write_zeroes = false;
 775         ret = -ENOTSUP;
 776     }
 777     return ret;
 778 }
 779
 780 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
 781 {
 782     int ret = -EOPNOTSUPP;
 783     BDRVRawState *s = aiocb->bs->opaque;
 784
 785     if (!s->has_discard) {
 786         return -ENOTSUP;
 787     }
 788
 789     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
 790 #ifdef BLKDISCARD
 791         do {
 792             uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
 793             if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
 794                 return 0;
 795             }
 796         } while (errno == EINTR);
 797
 798         ret = -errno;
 799 #endif
 800     } else {
 801 #ifdef CONFIG_XFS
 802         if (s->is_xfs) {
 803             return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
 804         }
 805 #endif
 806
 807 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
 808         do {
 809             if (fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
 810                           aiocb->aio_offset, aiocb->aio_nbytes) == 0) {
 811                 return 0;
 812             }
 813         } while (errno == EINTR);
 814
 815         ret = -errno;
 816 #endif
 817     }
 818
 819     if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
 820         ret == -ENOTTY) {
 821         s->has_discard = false;
 822         ret = -ENOTSUP;
 823     }
 824     return ret;
 825 }
 826
 827 static int aio_worker(void *arg)
 828 {
 829     RawPosixAIOData *aiocb = arg;
 830     ssize_t ret = 0;
 831
 832     switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
 833     case QEMU_AIO_READ:
 834         ret = handle_aiocb_rw(aiocb);
 835         if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) {
 836             iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
 837                       0, aiocb->aio_nbytes - ret);
 838
 839             ret = aiocb->aio_nbytes;
 840         }
 841         if (ret == aiocb->aio_nbytes) {
 842             ret = 0;
 843         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 844             ret = -EINVAL;
 845         }
 846         break;
 847     case QEMU_AIO_WRITE:
 848         ret = handle_aiocb_rw(aiocb);
 849         if (ret == aiocb->aio_nbytes) {
 850             ret = 0;
 851         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 852             ret = -EINVAL;
 853         }
 854         break;
 855     case QEMU_AIO_FLUSH:
 856         ret = handle_aiocb_flush(aiocb);
 857         break;
 858     case QEMU_AIO_IOCTL:
 859         ret = handle_aiocb_ioctl(aiocb);
 860         break;
 861     case QEMU_AIO_DISCARD:
 862         ret = handle_aiocb_discard(aiocb);
 863         break;
 864     case QEMU_AIO_WRITE_ZEROES:
 865         ret = handle_aiocb_write_zeroes(aiocb);
 866         break;
 867     default:
 868         fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
 869         ret = -EINVAL;
 870         break;
 871     }
 872
 873     g_slice_free(RawPosixAIOData, aiocb);
 874     return ret;
 875 }
 876
 877 static int paio_submit_co(BlockDriverState *bs, int fd,
 878         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 879         int type)
 880 {
 881     RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
 882     ThreadPool *pool;
 883
 884     acb->bs = bs;
 885     acb->aio_type = type;
 886     acb->aio_fildes = fd;
 887
 888     if (qiov) {
 889         acb->aio_iov = qiov->iov;
 890         acb->aio_niov = qiov->niov;
 891     }
 892     acb->aio_nbytes = nb_sectors * 512;
 893     acb->aio_offset = sector_num * 512;
 894
 895     trace_paio_submit_co(sector_num, nb_sectors, type);
 896     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
 897     return thread_pool_submit_co(pool, aio_worker, acb);
 898 }
 899
 900 static BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd,
 901         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 902         BlockDriverCompletionFunc *cb, void *opaque, int type)
 903 {
 904     RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
 905     ThreadPool *pool;
 906
 907     acb->bs = bs;
 908     acb->aio_type = type;
 909     acb->aio_fildes = fd;
 910
 911     if (qiov) {
 912         acb->aio_iov = qiov->iov;
 913         acb->aio_niov = qiov->niov;
 914     }
 915     acb->aio_nbytes = nb_sectors * 512;
 916     acb->aio_offset = sector_num * 512;
 917
 918     trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
 919     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
 920     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
 921 }
 922
 923 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
 924         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 925         BlockDriverCompletionFunc *cb, void *opaque, int type)
 926 {
 927     BDRVRawState *s = bs->opaque;
 928
 929     if (fd_open(bs) < 0)
 930         return NULL;
 931
 932     /*
 933      * If O_DIRECT is used the buffer needs to be aligned on a sector
 934      * boundary.  Check if this is the case or tell the low-level
 935      * driver that it needs to copy the buffer.
 936      */
 937     if ((bs->open_flags & BDRV_O_NOCACHE)) {
 938         if (!bdrv_qiov_is_aligned(bs, qiov)) {
 939             type |= QEMU_AIO_MISALIGNED;
 940 #ifdef CONFIG_LINUX_AIO
 941         } else if (s->use_aio) {
 942             return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
 943                                nb_sectors, cb, opaque, type);
 944 #endif
 945         }
 946     }
 947
 948     return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
 949                        cb, opaque, type);
 950 }
 951
 952 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
 953         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 954         BlockDriverCompletionFunc *cb, void *opaque)
 955 {
 956     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 957                           cb, opaque, QEMU_AIO_READ);
 958 }
 959
 960 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
 961         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 962         BlockDriverCompletionFunc *cb, void *opaque)
 963 {
 964     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 965                           cb, opaque, QEMU_AIO_WRITE);
 966 }
 967
 968 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
 969         BlockDriverCompletionFunc *cb, void *opaque)
 970 {
 971     BDRVRawState *s = bs->opaque;
 972
 973     if (fd_open(bs) < 0)
 974         return NULL;
 975
 976     return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
 977 }
 978
 979 static void raw_close(BlockDriverState *bs)
 980 {
 981     BDRVRawState *s = bs->opaque;
 982     if (s->fd >= 0) {
 983         qemu_close(s->fd);
 984         s->fd = -1;
 985     }
 986 }
 987
 988 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 989 {
 990     BDRVRawState *s = bs->opaque;
 991     struct stat st;
 992
 993     if (fstat(s->fd, &st)) {
 994         return -errno;
 995     }
 996
 997     if (S_ISREG(st.st_mode)) {
 998         if (ftruncate(s->fd, offset) < 0) {
 999             return -errno;
1000         }
1001     } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1002        if (offset > raw_getlength(bs)) {
1003            return -EINVAL;
1004        }
1005     } else {
1006         return -ENOTSUP;
1007     }
1008
1009     return 0;
1010 }
1011
1012 #ifdef __OpenBSD__
1013 static int64_t raw_getlength(BlockDriverState *bs)
1014 {
1015     BDRVRawState *s = bs->opaque;
1016     int fd = s->fd;
1017     struct stat st;
1018
1019     if (fstat(fd, &st))
1020         return -1;
1021     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1022         struct disklabel dl;
1023
1024         if (ioctl(fd, DIOCGDINFO, &dl))
1025             return -1;
1026         return (uint64_t)dl.d_secsize *
1027             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1028     } else
1029         return st.st_size;
1030 }
1031 #elif defined(__NetBSD__)
1032 static int64_t raw_getlength(BlockDriverState *bs)
1033 {
1034     BDRVRawState *s = bs->opaque;
1035     int fd = s->fd;
1036     struct stat st;
1037
1038     if (fstat(fd, &st))
1039         return -1;
1040     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1041         struct dkwedge_info dkw;
1042
1043         if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1044             return dkw.dkw_size * 512;
1045         } else {
1046             struct disklabel dl;
1047
1048             if (ioctl(fd, DIOCGDINFO, &dl))
1049                 return -1;
1050             return (uint64_t)dl.d_secsize *
1051                 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1052         }
1053     } else
1054         return st.st_size;
1055 }
1056 #elif defined(__sun__)
1057 static int64_t raw_getlength(BlockDriverState *bs)
1058 {
1059     BDRVRawState *s = bs->opaque;
1060     struct dk_minfo minfo;
1061     int ret;
1062
1063     ret = fd_open(bs);
1064     if (ret < 0) {
1065         return ret;
1066     }
1067
1068     /*
1069      * Use the DKIOCGMEDIAINFO ioctl to read the size.
1070      */
1071     ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1072     if (ret != -1) {
1073         return minfo.dki_lbsize * minfo.dki_capacity;
1074     }
1075
1076     /*
1077      * There are reports that lseek on some devices fails, but
1078      * irc discussion said that contingency on contingency was overkill.
1079      */
1080     return lseek(s->fd, 0, SEEK_END);
1081 }
1082 #elif defined(CONFIG_BSD)
1083 static int64_t raw_getlength(BlockDriverState *bs)
1084 {
1085     BDRVRawState *s = bs->opaque;
1086     int fd = s->fd;
1087     int64_t size;
1088     struct stat sb;
1089 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1090     int reopened = 0;
1091 #endif
1092     int ret;
1093
1094     ret = fd_open(bs);
1095     if (ret < 0)
1096         return ret;
1097
1098 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1099 again:
1100 #endif
1101     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1102 #ifdef DIOCGMEDIASIZE
1103         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1104 #elif defined(DIOCGPART)
1105         {
1106                 struct partinfo pi;
1107                 if (ioctl(fd, DIOCGPART, &pi) == 0)
1108                         size = pi.media_size;
1109                 else
1110                         size = 0;
1111         }
1112         if (size == 0)
1113 #endif
1114 #if defined(__APPLE__) && defined(__MACH__)
1115         size = LONG_LONG_MAX;
1116 #else
1117         size = lseek(fd, 0LL, SEEK_END);
1118 #endif
1119 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1120         switch(s->type) {
1121         case FTYPE_CD:
1122             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1123             if (size == 2048LL * (unsigned)-1)
1124                 size = 0;
1125             /* XXX no disc?  maybe we need to reopen... */
1126             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1127                 reopened = 1;
1128                 goto again;
1129             }
1130         }
1131 #endif
1132     } else {
1133         size = lseek(fd, 0, SEEK_END);
1134     }
1135     return size;
1136 }
1137 #else
1138 static int64_t raw_getlength(BlockDriverState *bs)
1139 {
1140     BDRVRawState *s = bs->opaque;
1141     int ret;
1142
1143     ret = fd_open(bs);
1144     if (ret < 0) {
1145         return ret;
1146     }
1147
1148     return lseek(s->fd, 0, SEEK_END);
1149 }
1150 #endif
1151
1152 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1153 {
1154     struct stat st;
1155     BDRVRawState *s = bs->opaque;
1156
1157     if (fstat(s->fd, &st) < 0) {
1158         return -errno;
1159     }
1160     return (int64_t)st.st_blocks * 512;
1161 }
1162
1163 static int raw_create(const char *filename, QEMUOptionParameter *options,
1164                       Error **errp)
1165 {
1166     int fd;
1167     int result = 0;
1168     int64_t total_size = 0;
1169
1170     /* Read out options */
1171     while (options && options->name) {
1172         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1173             total_size = options->value.n / BDRV_SECTOR_SIZE;
1174         }
1175         options++;
1176     }
1177
1178     fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
1179                    0644);
1180     if (fd < 0) {
1181         result = -errno;
1182         error_setg_errno(errp, -result, "Could not create file");
1183     } else {
1184         if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
1185             result = -errno;
1186             error_setg_errno(errp, -result, "Could not resize file");
1187         }
1188         if (qemu_close(fd) != 0) {
1189             result = -errno;
1190             error_setg_errno(errp, -result, "Could not close the new file");
1191         }
1192     }
1193     return result;
1194 }
1195
1196 /*
1197  * Returns true iff the specified sector is present in the disk image. Drivers
1198  * not implementing the functionality are assumed to not support backing files,
1199  * hence all their sectors are reported as allocated.
1200  *
1201  * If 'sector_num' is beyond the end of the disk image the return value is 0
1202  * and 'pnum' is set to 0.
1203  *
1204  * 'pnum' is set to the number of sectors (including and immediately following
1205  * the specified sector) that are known to be in the same
1206  * allocated/unallocated state.
1207  *
1208  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
1209  * beyond the end of the disk image it will be clamped.
1210  */
1211 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1212                                             int64_t sector_num,
1213                                             int nb_sectors, int *pnum)
1214 {
1215     off_t start, data, hole;
1216     int64_t ret;
1217
1218     ret = fd_open(bs);
1219     if (ret < 0) {
1220         return ret;
1221     }
1222
1223     start = sector_num * BDRV_SECTOR_SIZE;
1224     ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1225
1226 #ifdef CONFIG_FIEMAP
1227
1228     BDRVRawState *s = bs->opaque;
1229     struct {
1230         struct fiemap fm;
1231         struct fiemap_extent fe;
1232     } f;
1233
1234     f.fm.fm_start = start;
1235     f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE;
1236     f.fm.fm_flags = 0;
1237     f.fm.fm_extent_count = 1;
1238     f.fm.fm_reserved = 0;
1239     if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) {
1240         /* Assume everything is allocated.  */
1241         *pnum = nb_sectors;
1242         return ret;
1243     }
1244
1245     if (f.fm.fm_mapped_extents == 0) {
1246         /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1247          * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1248          */
1249         off_t length = lseek(s->fd, 0, SEEK_END);
1250         hole = f.fm.fm_start;
1251         data = MIN(f.fm.fm_start + f.fm.fm_length, length);
1252     } else {
1253         data = f.fe.fe_logical;
1254         hole = f.fe.fe_logical + f.fe.fe_length;
1255         if (f.fe.fe_flags & FIEMAP_EXTENT_UNWRITTEN) {
1256             ret |= BDRV_BLOCK_ZERO;
1257         }
1258     }
1259
1260 #elif defined SEEK_HOLE && defined SEEK_DATA
1261
1262     BDRVRawState *s = bs->opaque;
1263
1264     hole = lseek(s->fd, start, SEEK_HOLE);
1265     if (hole == -1) {
1266         /* -ENXIO indicates that sector_num was past the end of the file.
1267          * There is a virtual hole there.  */
1268         assert(errno != -ENXIO);
1269
1270         /* Most likely EINVAL.  Assume everything is allocated.  */
1271         *pnum = nb_sectors;
1272         return ret;
1273     }
1274
1275     if (hole > start) {
1276         data = start;
1277     } else {
1278         /* On a hole.  We need another syscall to find its end.  */
1279         data = lseek(s->fd, start, SEEK_DATA);
1280         if (data == -1) {
1281             data = lseek(s->fd, 0, SEEK_END);
1282         }
1283     }
1284 #else
1285     data = 0;
1286     hole = start + nb_sectors * BDRV_SECTOR_SIZE;
1287 #endif
1288
1289     if (data <= start) {
1290         /* On a data extent, compute sectors to the end of the extent.  */
1291         *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1292     } else {
1293         /* On a hole, compute sectors to the beginning of the next extent.  */
1294         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1295         ret &= ~BDRV_BLOCK_DATA;
1296         ret |= BDRV_BLOCK_ZERO;
1297     }
1298
1299     return ret;
1300 }
1301
1302 static coroutine_fn BlockDriverAIOCB *raw_aio_discard(BlockDriverState *bs,
1303     int64_t sector_num, int nb_sectors,
1304     BlockDriverCompletionFunc *cb, void *opaque)
1305 {
1306     BDRVRawState *s = bs->opaque;
1307
1308     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1309                        cb, opaque, QEMU_AIO_DISCARD);
1310 }
1311
1312 static int coroutine_fn raw_co_write_zeroes(
1313     BlockDriverState *bs, int64_t sector_num,
1314     int nb_sectors, BdrvRequestFlags flags)
1315 {
1316     BDRVRawState *s = bs->opaque;
1317
1318     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1319         return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1320                               QEMU_AIO_WRITE_ZEROES);
1321     } else if (s->discard_zeroes) {
1322         return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1323                               QEMU_AIO_DISCARD);
1324     }
1325     return -ENOTSUP;
1326 }
1327
1328 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1329 {
1330     BDRVRawState *s = bs->opaque;
1331
1332     bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1333     bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1334     return 0;
1335 }
1336
1337 static QEMUOptionParameter raw_create_options[] = {
1338     {
1339         .name = BLOCK_OPT_SIZE,
1340         .type = OPT_SIZE,
1341         .help = "Virtual disk size"
1342     },
1343     { NULL }
1344 };
1345
1346 static BlockDriver bdrv_file = {
1347     .format_name = "file",
1348     .protocol_name = "file",
1349     .instance_size = sizeof(BDRVRawState),
1350     .bdrv_needs_filename = true,
1351     .bdrv_probe = NULL, /* no probe for protocols */
1352     .bdrv_file_open = raw_open,
1353     .bdrv_reopen_prepare = raw_reopen_prepare,
1354     .bdrv_reopen_commit = raw_reopen_commit,
1355     .bdrv_reopen_abort = raw_reopen_abort,
1356     .bdrv_close = raw_close,
1357     .bdrv_create = raw_create,
1358     .bdrv_has_zero_init = bdrv_has_zero_init_1,
1359     .bdrv_co_get_block_status = raw_co_get_block_status,
1360     .bdrv_co_write_zeroes = raw_co_write_zeroes,
1361
1362     .bdrv_aio_readv = raw_aio_readv,
1363     .bdrv_aio_writev = raw_aio_writev,
1364     .bdrv_aio_flush = raw_aio_flush,
1365     .bdrv_aio_discard = raw_aio_discard,
1366
1367     .bdrv_truncate = raw_truncate,
1368     .bdrv_getlength = raw_getlength,
1369     .bdrv_get_info = raw_get_info,
1370     .bdrv_get_allocated_file_size
1371                         = raw_get_allocated_file_size,
1372
1373     .create_options = raw_create_options,
1374 };
1375
1376 /***********************************************/
1377 /* host device */
1378
1379 #if defined(__APPLE__) && defined(__MACH__)
1380 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1381 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1382
1383 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1384 {
1385     kern_return_t       kernResult;
1386     mach_port_t     masterPort;
1387     CFMutableDictionaryRef  classesToMatch;
1388
1389     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1390     if ( KERN_SUCCESS != kernResult ) {
1391         printf( "IOMasterPort returned %d\n", kernResult );
1392     }
1393
1394     classesToMatch = IOServiceMatching( kIOCDMediaClass );
1395     if ( classesToMatch == NULL ) {
1396         printf( "IOServiceMatching returned a NULL dictionary.\n" );
1397     } else {
1398     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1399     }
1400     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1401     if ( KERN_SUCCESS != kernResult )
1402     {
1403         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1404     }
1405
1406     return kernResult;
1407 }
1408
1409 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
1410 {
1411     io_object_t     nextMedia;
1412     kern_return_t   kernResult = KERN_FAILURE;
1413     *bsdPath = '\0';
1414     nextMedia = IOIteratorNext( mediaIterator );
1415     if ( nextMedia )
1416     {
1417         CFTypeRef   bsdPathAsCFString;
1418     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1419         if ( bsdPathAsCFString ) {
1420             size_t devPathLength;
1421             strcpy( bsdPath, _PATH_DEV );
1422             strcat( bsdPath, "r" );
1423             devPathLength = strlen( bsdPath );
1424             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1425                 kernResult = KERN_SUCCESS;
1426             }
1427             CFRelease( bsdPathAsCFString );
1428         }
1429         IOObjectRelease( nextMedia );
1430     }
1431
1432     return kernResult;
1433 }
1434
1435 #endif
1436
1437 static int hdev_probe_device(const char *filename)
1438 {
1439     struct stat st;
1440
1441     /* allow a dedicated CD-ROM driver to match with a higher priority */
1442     if (strstart(filename, "/dev/cdrom", NULL))
1443         return 50;
1444
1445     if (stat(filename, &st) >= 0 &&
1446             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
1447         return 100;
1448     }
1449
1450     return 0;
1451 }
1452
1453 static int check_hdev_writable(BDRVRawState *s)
1454 {
1455 #if defined(BLKROGET)
1456     /* Linux block devices can be configured "read-only" using blockdev(8).
1457      * This is independent of device node permissions and therefore open(2)
1458      * with O_RDWR succeeds.  Actual writes fail with EPERM.
1459      *
1460      * bdrv_open() is supposed to fail if the disk is read-only.  Explicitly
1461      * check for read-only block devices so that Linux block devices behave
1462      * properly.
1463      */
1464     struct stat st;
1465     int readonly = 0;
1466
1467     if (fstat(s->fd, &st)) {
1468         return -errno;
1469     }
1470
1471     if (!S_ISBLK(st.st_mode)) {
1472         return 0;
1473     }
1474
1475     if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
1476         return -errno;
1477     }
1478
1479     if (readonly) {
1480         return -EACCES;
1481     }
1482 #endif /* defined(BLKROGET) */
1483     return 0;
1484 }
1485
1486 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
1487                      Error **errp)
1488 {
1489     BDRVRawState *s = bs->opaque;
1490     Error *local_err = NULL;
1491     int ret;
1492     const char *filename = qdict_get_str(options, "filename");
1493
1494 #if defined(__APPLE__) && defined(__MACH__)
1495     if (strstart(filename, "/dev/cdrom", NULL)) {
1496         kern_return_t kernResult;
1497         io_iterator_t mediaIterator;
1498         char bsdPath[ MAXPATHLEN ];
1499         int fd;
1500
1501         kernResult = FindEjectableCDMedia( &mediaIterator );
1502         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1503
1504         if ( bsdPath[ 0 ] != '\0' ) {
1505             strcat(bsdPath,"s0");
1506             /* some CDs don't have a partition 0 */
1507             fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1508             if (fd < 0) {
1509                 bsdPath[strlen(bsdPath)-1] = '1';
1510             } else {
1511                 qemu_close(fd);
1512             }
1513             filename = bsdPath;
1514             qdict_put(options, "filename", qstring_from_str(filename));
1515         }
1516
1517         if ( mediaIterator )
1518             IOObjectRelease( mediaIterator );
1519     }
1520 #endif
1521
1522     s->type = FTYPE_FILE;
1523 #if defined(__linux__)
1524     {
1525         char resolved_path[ MAXPATHLEN ], *temp;
1526
1527         temp = realpath(filename, resolved_path);
1528         if (temp && strstart(temp, "/dev/sg", NULL)) {
1529             bs->sg = 1;
1530         }
1531     }
1532 #endif
1533
1534     ret = raw_open_common(bs, options, flags, 0, &local_err);
1535     if (ret < 0) {
1536         if (error_is_set(&local_err)) {
1537             error_propagate(errp, local_err);
1538         }
1539         return ret;
1540     }
1541
1542     if (flags & BDRV_O_RDWR) {
1543         ret = check_hdev_writable(s);
1544         if (ret < 0) {
1545             raw_close(bs);
1546             error_setg_errno(errp, -ret, "The device is not writable");
1547             return ret;
1548         }
1549     }
1550
1551     return ret;
1552 }
1553
1554 #if defined(__linux__)
1555 /* Note: we do not have a reliable method to detect if the floppy is
1556    present. The current method is to try to open the floppy at every
1557    I/O and to keep it opened during a few hundreds of ms. */
1558 static int fd_open(BlockDriverState *bs)
1559 {
1560     BDRVRawState *s = bs->opaque;
1561     int last_media_present;
1562
1563     if (s->type != FTYPE_FD)
1564         return 0;
1565     last_media_present = (s->fd >= 0);
1566     if (s->fd >= 0 &&
1567         (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1568         qemu_close(s->fd);
1569         s->fd = -1;
1570 #ifdef DEBUG_FLOPPY
1571         printf("Floppy closed\n");
1572 #endif
1573     }
1574     if (s->fd < 0) {
1575         if (s->fd_got_error &&
1576             (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1577 #ifdef DEBUG_FLOPPY
1578             printf("No floppy (open delayed)\n");
1579 #endif
1580             return -EIO;
1581         }
1582         s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
1583         if (s->fd < 0) {
1584             s->fd_error_time = get_clock();
1585             s->fd_got_error = 1;
1586             if (last_media_present)
1587                 s->fd_media_changed = 1;
1588 #ifdef DEBUG_FLOPPY
1589             printf("No floppy\n");
1590 #endif
1591             return -EIO;
1592         }
1593 #ifdef DEBUG_FLOPPY
1594         printf("Floppy opened\n");
1595 #endif
1596     }
1597     if (!last_media_present)
1598         s->fd_media_changed = 1;
1599     s->fd_open_time = get_clock();
1600     s->fd_got_error = 0;
1601     return 0;
1602 }
1603
1604 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1605 {
1606     BDRVRawState *s = bs->opaque;
1607
1608     return ioctl(s->fd, req, buf);
1609 }
1610
1611 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1612         unsigned long int req, void *buf,
1613         BlockDriverCompletionFunc *cb, void *opaque)
1614 {
1615     BDRVRawState *s = bs->opaque;
1616     RawPosixAIOData *acb;
1617     ThreadPool *pool;
1618
1619     if (fd_open(bs) < 0)
1620         return NULL;
1621
1622     acb = g_slice_new(RawPosixAIOData);
1623     acb->bs = bs;
1624     acb->aio_type = QEMU_AIO_IOCTL;
1625     acb->aio_fildes = s->fd;
1626     acb->aio_offset = 0;
1627     acb->aio_ioctl_buf = buf;
1628     acb->aio_ioctl_cmd = req;
1629     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1630     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1631 }
1632
1633 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1634 static int fd_open(BlockDriverState *bs)
1635 {
1636     BDRVRawState *s = bs->opaque;
1637
1638     /* this is just to ensure s->fd is sane (its called by io ops) */
1639     if (s->fd >= 0)
1640         return 0;
1641     return -EIO;
1642 }
1643 #else /* !linux && !FreeBSD */
1644
1645 static int fd_open(BlockDriverState *bs)
1646 {
1647     return 0;
1648 }
1649
1650 #endif /* !linux && !FreeBSD */
1651
1652 static coroutine_fn BlockDriverAIOCB *hdev_aio_discard(BlockDriverState *bs,
1653     int64_t sector_num, int nb_sectors,
1654     BlockDriverCompletionFunc *cb, void *opaque)
1655 {
1656     BDRVRawState *s = bs->opaque;
1657
1658     if (fd_open(bs) < 0) {
1659         return NULL;
1660     }
1661     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1662                        cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1663 }
1664
1665 static coroutine_fn int hdev_co_write_zeroes(BlockDriverState *bs,
1666     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
1667 {
1668     BDRVRawState *s = bs->opaque;
1669     int rc;
1670
1671     rc = fd_open(bs);
1672     if (rc < 0) {
1673         return rc;
1674     }
1675     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1676         return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1677                               QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
1678     } else if (s->discard_zeroes) {
1679         return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1680                               QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1681     }
1682     return -ENOTSUP;
1683 }
1684
1685 static int hdev_create(const char *filename, QEMUOptionParameter *options,
1686                        Error **errp)
1687 {
1688     int fd;
1689     int ret = 0;
1690     struct stat stat_buf;
1691     int64_t total_size = 0;
1692
1693     /* Read out options */
1694     while (options && options->name) {
1695         if (!strcmp(options->name, "size")) {
1696             total_size = options->value.n / BDRV_SECTOR_SIZE;
1697         }
1698         options++;
1699     }
1700
1701     fd = qemu_open(filename, O_WRONLY | O_BINARY);
1702     if (fd < 0) {
1703         ret = -errno;
1704         error_setg_errno(errp, -ret, "Could not open device");
1705         return ret;
1706     }
1707
1708     if (fstat(fd, &stat_buf) < 0) {
1709         ret = -errno;
1710         error_setg_errno(errp, -ret, "Could not stat device");
1711     } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
1712         error_setg(errp,
1713                    "The given file is neither a block nor a character device");
1714         ret = -ENODEV;
1715     } else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE) {
1716         error_setg(errp, "Device is too small");
1717         ret = -ENOSPC;
1718     }
1719
1720     qemu_close(fd);
1721     return ret;
1722 }
1723
1724 static BlockDriver bdrv_host_device = {
1725     .format_name        = "host_device",
1726     .protocol_name        = "host_device",
1727     .instance_size      = sizeof(BDRVRawState),
1728     .bdrv_needs_filename = true,
1729     .bdrv_probe_device  = hdev_probe_device,
1730     .bdrv_file_open     = hdev_open,
1731     .bdrv_close         = raw_close,
1732     .bdrv_reopen_prepare = raw_reopen_prepare,
1733     .bdrv_reopen_commit  = raw_reopen_commit,
1734     .bdrv_reopen_abort   = raw_reopen_abort,
1735     .bdrv_create        = hdev_create,
1736     .create_options     = raw_create_options,
1737     .bdrv_co_write_zeroes = hdev_co_write_zeroes,
1738
1739     .bdrv_aio_readv     = raw_aio_readv,
1740     .bdrv_aio_writev    = raw_aio_writev,
1741     .bdrv_aio_flush     = raw_aio_flush,
1742     .bdrv_aio_discard   = hdev_aio_discard,
1743
1744     .bdrv_truncate      = raw_truncate,
1745     .bdrv_getlength     = raw_getlength,
1746     .bdrv_get_info = raw_get_info,
1747     .bdrv_get_allocated_file_size
1748                         = raw_get_allocated_file_size,
1749
1750     /* generic scsi device */
1751 #ifdef __linux__
1752     .bdrv_ioctl         = hdev_ioctl,
1753     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1754 #endif
1755 };
1756
1757 #ifdef __linux__
1758 static int floppy_open(BlockDriverState *bs, QDict *options, int flags,
1759                        Error **errp)
1760 {
1761     BDRVRawState *s = bs->opaque;
1762     Error *local_err = NULL;
1763     int ret;
1764
1765     s->type = FTYPE_FD;
1766
1767     /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1768     ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
1769     if (ret) {
1770         if (error_is_set(&local_err)) {
1771             error_propagate(errp, local_err);
1772         }
1773         return ret;
1774     }
1775
1776     /* close fd so that we can reopen it as needed */
1777     qemu_close(s->fd);
1778     s->fd = -1;
1779     s->fd_media_changed = 1;
1780
1781     return 0;
1782 }
1783
1784 static int floppy_probe_device(const char *filename)
1785 {
1786     int fd, ret;
1787     int prio = 0;
1788     struct floppy_struct fdparam;
1789     struct stat st;
1790
1791     if (strstart(filename, "/dev/fd", NULL) &&
1792         !strstart(filename, "/dev/fdset/", NULL)) {
1793         prio = 50;
1794     }
1795
1796     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1797     if (fd < 0) {
1798         goto out;
1799     }
1800     ret = fstat(fd, &st);
1801     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1802         goto outc;
1803     }
1804
1805     /* Attempt to detect via a floppy specific ioctl */
1806     ret = ioctl(fd, FDGETPRM, &fdparam);
1807     if (ret >= 0)
1808         prio = 100;
1809
1810 outc:
1811     qemu_close(fd);
1812 out:
1813     return prio;
1814 }
1815
1816
1817 static int floppy_is_inserted(BlockDriverState *bs)
1818 {
1819     return fd_open(bs) >= 0;
1820 }
1821
1822 static int floppy_media_changed(BlockDriverState *bs)
1823 {
1824     BDRVRawState *s = bs->opaque;
1825     int ret;
1826
1827     /*
1828      * XXX: we do not have a true media changed indication.
1829      * It does not work if the floppy is changed without trying to read it.
1830      */
1831     fd_open(bs);
1832     ret = s->fd_media_changed;
1833     s->fd_media_changed = 0;
1834 #ifdef DEBUG_FLOPPY
1835     printf("Floppy changed=%d\n", ret);
1836 #endif
1837     return ret;
1838 }
1839
1840 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
1841 {
1842     BDRVRawState *s = bs->opaque;
1843     int fd;
1844
1845     if (s->fd >= 0) {
1846         qemu_close(s->fd);
1847         s->fd = -1;
1848     }
1849     fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
1850     if (fd >= 0) {
1851         if (ioctl(fd, FDEJECT, 0) < 0)
1852             perror("FDEJECT");
1853         qemu_close(fd);
1854     }
1855 }
1856
1857 static BlockDriver bdrv_host_floppy = {
1858     .format_name        = "host_floppy",
1859     .protocol_name      = "host_floppy",
1860     .instance_size      = sizeof(BDRVRawState),
1861     .bdrv_needs_filename = true,
1862     .bdrv_probe_device  = floppy_probe_device,
1863     .bdrv_file_open     = floppy_open,
1864     .bdrv_close         = raw_close,
1865     .bdrv_reopen_prepare = raw_reopen_prepare,
1866     .bdrv_reopen_commit  = raw_reopen_commit,
1867     .bdrv_reopen_abort   = raw_reopen_abort,
1868     .bdrv_create        = hdev_create,
1869     .create_options     = raw_create_options,
1870
1871     .bdrv_aio_readv     = raw_aio_readv,
1872     .bdrv_aio_writev    = raw_aio_writev,
1873     .bdrv_aio_flush     = raw_aio_flush,
1874
1875     .bdrv_truncate      = raw_truncate,
1876     .bdrv_getlength      = raw_getlength,
1877     .has_variable_length = true,
1878     .bdrv_get_allocated_file_size
1879                         = raw_get_allocated_file_size,
1880
1881     /* removable device support */
1882     .bdrv_is_inserted   = floppy_is_inserted,
1883     .bdrv_media_changed = floppy_media_changed,
1884     .bdrv_eject         = floppy_eject,
1885 };
1886
1887 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
1888                       Error **errp)
1889 {
1890     BDRVRawState *s = bs->opaque;
1891     Error *local_err = NULL;
1892     int ret;
1893
1894     s->type = FTYPE_CD;
1895
1896     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
1897     ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
1898     if (error_is_set(&local_err)) {
1899         error_propagate(errp, local_err);
1900     }
1901     return ret;
1902 }
1903
1904 static int cdrom_probe_device(const char *filename)
1905 {
1906     int fd, ret;
1907     int prio = 0;
1908     struct stat st;
1909
1910     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1911     if (fd < 0) {
1912         goto out;
1913     }
1914     ret = fstat(fd, &st);
1915     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1916         goto outc;
1917     }
1918
1919     /* Attempt to detect via a CDROM specific ioctl */
1920     ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1921     if (ret >= 0)
1922         prio = 100;
1923
1924 outc:
1925     qemu_close(fd);
1926 out:
1927     return prio;
1928 }
1929
1930 static int cdrom_is_inserted(BlockDriverState *bs)
1931 {
1932     BDRVRawState *s = bs->opaque;
1933     int ret;
1934
1935     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1936     if (ret == CDS_DISC_OK)
1937         return 1;
1938     return 0;
1939 }
1940
1941 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1942 {
1943     BDRVRawState *s = bs->opaque;
1944
1945     if (eject_flag) {
1946         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
1947             perror("CDROMEJECT");
1948     } else {
1949         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
1950             perror("CDROMEJECT");
1951     }
1952 }
1953
1954 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1955 {
1956     BDRVRawState *s = bs->opaque;
1957
1958     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
1959         /*
1960          * Note: an error can happen if the distribution automatically
1961          * mounts the CD-ROM
1962          */
1963         /* perror("CDROM_LOCKDOOR"); */
1964     }
1965 }
1966
1967 static BlockDriver bdrv_host_cdrom = {
1968     .format_name        = "host_cdrom",
1969     .protocol_name      = "host_cdrom",
1970     .instance_size      = sizeof(BDRVRawState),
1971     .bdrv_needs_filename = true,
1972     .bdrv_probe_device  = cdrom_probe_device,
1973     .bdrv_file_open     = cdrom_open,
1974     .bdrv_close         = raw_close,
1975     .bdrv_reopen_prepare = raw_reopen_prepare,
1976     .bdrv_reopen_commit  = raw_reopen_commit,
1977     .bdrv_reopen_abort   = raw_reopen_abort,
1978     .bdrv_create        = hdev_create,
1979     .create_options     = raw_create_options,
1980
1981     .bdrv_aio_readv     = raw_aio_readv,
1982     .bdrv_aio_writev    = raw_aio_writev,
1983     .bdrv_aio_flush     = raw_aio_flush,
1984
1985     .bdrv_truncate      = raw_truncate,
1986     .bdrv_getlength      = raw_getlength,
1987     .has_variable_length = true,
1988     .bdrv_get_allocated_file_size
1989                         = raw_get_allocated_file_size,
1990
1991     /* removable device support */
1992     .bdrv_is_inserted   = cdrom_is_inserted,
1993     .bdrv_eject         = cdrom_eject,
1994     .bdrv_lock_medium   = cdrom_lock_medium,
1995
1996     /* generic scsi device */
1997     .bdrv_ioctl         = hdev_ioctl,
1998     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1999 };
2000 #endif /* __linux__ */
2001
2002 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2003 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2004                       Error **errp)
2005 {
2006     BDRVRawState *s = bs->opaque;
2007     Error *local_err = NULL;
2008     int ret;
2009
2010     s->type = FTYPE_CD;
2011
2012     ret = raw_open_common(bs, options, flags, 0, &local_err);
2013     if (ret) {
2014         if (error_is_set(&local_err)) {
2015             error_propagate(errp, local_err);
2016         }
2017         return ret;
2018     }
2019
2020     /* make sure the door isn't locked at this time */
2021     ioctl(s->fd, CDIOCALLOW);
2022     return 0;
2023 }
2024
2025 static int cdrom_probe_device(const char *filename)
2026 {
2027     if (strstart(filename, "/dev/cd", NULL) ||
2028             strstart(filename, "/dev/acd", NULL))
2029         return 100;
2030     return 0;
2031 }
2032
2033 static int cdrom_reopen(BlockDriverState *bs)
2034 {
2035     BDRVRawState *s = bs->opaque;
2036     int fd;
2037
2038     /*
2039      * Force reread of possibly changed/newly loaded disc,
2040      * FreeBSD seems to not notice sometimes...
2041      */
2042     if (s->fd >= 0)
2043         qemu_close(s->fd);
2044     fd = qemu_open(bs->filename, s->open_flags, 0644);
2045     if (fd < 0) {
2046         s->fd = -1;
2047         return -EIO;
2048     }
2049     s->fd = fd;
2050
2051     /* make sure the door isn't locked at this time */
2052     ioctl(s->fd, CDIOCALLOW);
2053     return 0;
2054 }
2055
2056 static int cdrom_is_inserted(BlockDriverState *bs)
2057 {
2058     return raw_getlength(bs) > 0;
2059 }
2060
2061 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2062 {
2063     BDRVRawState *s = bs->opaque;
2064
2065     if (s->fd < 0)
2066         return;
2067
2068     (void) ioctl(s->fd, CDIOCALLOW);
2069
2070     if (eject_flag) {
2071         if (ioctl(s->fd, CDIOCEJECT) < 0)
2072             perror("CDIOCEJECT");
2073     } else {
2074         if (ioctl(s->fd, CDIOCCLOSE) < 0)
2075             perror("CDIOCCLOSE");
2076     }
2077
2078     cdrom_reopen(bs);
2079 }
2080
2081 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2082 {
2083     BDRVRawState *s = bs->opaque;
2084
2085     if (s->fd < 0)
2086         return;
2087     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2088         /*
2089          * Note: an error can happen if the distribution automatically
2090          * mounts the CD-ROM
2091          */
2092         /* perror("CDROM_LOCKDOOR"); */
2093     }
2094 }
2095
2096 static BlockDriver bdrv_host_cdrom = {
2097     .format_name        = "host_cdrom",
2098     .protocol_name      = "host_cdrom",
2099     .instance_size      = sizeof(BDRVRawState),
2100     .bdrv_needs_filename = true,
2101     .bdrv_probe_device  = cdrom_probe_device,
2102     .bdrv_file_open     = cdrom_open,
2103     .bdrv_close         = raw_close,
2104     .bdrv_reopen_prepare = raw_reopen_prepare,
2105     .bdrv_reopen_commit  = raw_reopen_commit,
2106     .bdrv_reopen_abort   = raw_reopen_abort,
2107     .bdrv_create        = hdev_create,
2108     .create_options     = raw_create_options,
2109
2110     .bdrv_aio_readv     = raw_aio_readv,
2111     .bdrv_aio_writev    = raw_aio_writev,
2112     .bdrv_aio_flush     = raw_aio_flush,
2113
2114     .bdrv_truncate      = raw_truncate,
2115     .bdrv_getlength      = raw_getlength,
2116     .has_variable_length = true,
2117     .bdrv_get_allocated_file_size
2118                         = raw_get_allocated_file_size,
2119
2120     /* removable device support */
2121     .bdrv_is_inserted   = cdrom_is_inserted,
2122     .bdrv_eject         = cdrom_eject,
2123     .bdrv_lock_medium   = cdrom_lock_medium,
2124 };
2125 #endif /* __FreeBSD__ */
2126
2127 #ifdef CONFIG_LINUX_AIO
2128 /**
2129  * Return the file descriptor for Linux AIO
2130  *
2131  * This function is a layering violation and should be removed when it becomes
2132  * possible to call the block layer outside the global mutex.  It allows the
2133  * caller to hijack the file descriptor so I/O can be performed outside the
2134  * block layer.
2135  */
2136 int raw_get_aio_fd(BlockDriverState *bs)
2137 {
2138     BDRVRawState *s;
2139
2140     if (!bs->drv) {
2141         return -ENOMEDIUM;
2142     }
2143
2144     if (bs->drv == bdrv_find_format("raw")) {
2145         bs = bs->file;
2146     }
2147
2148     /* raw-posix has several protocols so just check for raw_aio_readv */
2149     if (bs->drv->bdrv_aio_readv != raw_aio_readv) {
2150         return -ENOTSUP;
2151     }
2152
2153     s = bs->opaque;
2154     if (!s->use_aio) {
2155         return -ENOTSUP;
2156     }
2157     return s->fd;
2158 }
2159 #endif /* CONFIG_LINUX_AIO */
2160
2161 static void bdrv_file_init(void)
2162 {
2163     /*
2164      * Register all the drivers.  Note that order is important, the driver
2165      * registered last will get probed first.
2166      */
2167     bdrv_register(&bdrv_file);
2168     bdrv_register(&bdrv_host_device);
2169 #ifdef __linux__
2170     bdrv_register(&bdrv_host_floppy);
2171     bdrv_register(&bdrv_host_cdrom);
2172 #endif
2173 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2174     bdrv_register(&bdrv_host_cdrom);
2175 #endif
2176 }
2177
2178 block_init(bdrv_file_init);