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