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