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 #include "qapi/util.h"
  34
  35 #if defined(__APPLE__) && (__MACH__)
  36 #include <paths.h>
  37 #include <sys/param.h>
  38 #include <IOKit/IOKitLib.h>
  39 #include <IOKit/IOBSD.h>
  40 #include <IOKit/storage/IOMediaBSDClient.h>
  41 #include <IOKit/storage/IOMedia.h>
  42 #include <IOKit/storage/IOCDMedia.h>
  43 //#include <IOKit/storage/IOCDTypes.h>
  44 #include <CoreFoundation/CoreFoundation.h>
  45 #endif
  46
  47 #ifdef __sun__
  48 #define _POSIX_PTHREAD_SEMANTICS 1
  49 #include <sys/dkio.h>
  50 #endif
  51 #ifdef __linux__
  52 #include <sys/types.h>
  53 #include <sys/stat.h>
  54 #include <sys/ioctl.h>
  55 #include <sys/param.h>
  56 #include <linux/cdrom.h>
  57 #include <linux/fd.h>
  58 #include <linux/fs.h>
  59 #ifndef FS_NOCOW_FL
  60 #define FS_NOCOW_FL                     0x00800000 /* Do not cow file */
  61 #endif
  62 #endif
  63 #ifdef CONFIG_FIEMAP
  64 #include <linux/fiemap.h>
  65 #endif
  66 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
  67 #include <linux/falloc.h>
  68 #endif
  69 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
  70 #include <sys/disk.h>
  71 #include <sys/cdio.h>
  72 #endif
  73
  74 #ifdef __OpenBSD__
  75 #include <sys/ioctl.h>
  76 #include <sys/disklabel.h>
  77 #include <sys/dkio.h>
  78 #endif
  79
  80 #ifdef __NetBSD__
  81 #include <sys/ioctl.h>
  82 #include <sys/disklabel.h>
  83 #include <sys/dkio.h>
  84 #include <sys/disk.h>
  85 #endif
  86
  87 #ifdef __DragonFly__
  88 #include <sys/ioctl.h>
  89 #include <sys/diskslice.h>
  90 #endif
  91
  92 #ifdef CONFIG_XFS
  93 #include <xfs/xfs.h>
  94 #endif
  95
  96 //#define DEBUG_FLOPPY
  97
  98 //#define DEBUG_BLOCK
  99 #if defined(DEBUG_BLOCK)
 100 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
 101     { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
 102 #else
 103 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
 104 #endif
 105
 106 /* OS X does not have O_DSYNC */
 107 #ifndef O_DSYNC
 108 #ifdef O_SYNC
 109 #define O_DSYNC O_SYNC
 110 #elif defined(O_FSYNC)
 111 #define O_DSYNC O_FSYNC
 112 #endif
 113 #endif
 114
 115 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
 116 #ifndef O_DIRECT
 117 #define O_DIRECT O_DSYNC
 118 #endif
 119
 120 #define FTYPE_FILE   0
 121 #define FTYPE_CD     1
 122 #define FTYPE_FD     2
 123
 124 /* if the FD is not accessed during that time (in ns), we try to
 125    reopen it to see if the disk has been changed */
 126 #define FD_OPEN_TIMEOUT (1000000000)
 127
 128 #define MAX_BLOCKSIZE   4096
 129
 130 typedef struct BDRVRawState {
 131     int fd;
 132     int type;
 133     int open_flags;
 134     size_t buf_align;
 135
 136 #if defined(__linux__)
 137     /* linux floppy specific */
 138     int64_t fd_open_time;
 139     int64_t fd_error_time;
 140     int fd_got_error;
 141     int fd_media_changed;
 142 #endif
 143 #ifdef CONFIG_LINUX_AIO
 144     int use_aio;
 145     void *aio_ctx;
 146 #endif
 147 #ifdef CONFIG_XFS
 148     bool is_xfs:1;
 149 #endif
 150     bool has_discard:1;
 151     bool has_write_zeroes:1;
 152     bool discard_zeroes:1;
 153 #ifdef CONFIG_FIEMAP
 154     bool skip_fiemap;
 155 #endif
 156 } BDRVRawState;
 157
 158 typedef struct BDRVRawReopenState {
 159     int fd;
 160     int open_flags;
 161 #ifdef CONFIG_LINUX_AIO
 162     int use_aio;
 163 #endif
 164 } BDRVRawReopenState;
 165
 166 static int fd_open(BlockDriverState *bs);
 167 static int64_t raw_getlength(BlockDriverState *bs);
 168
 169 typedef struct RawPosixAIOData {
 170     BlockDriverState *bs;
 171     int aio_fildes;
 172     union {
 173         struct iovec *aio_iov;
 174         void *aio_ioctl_buf;
 175     };
 176     int aio_niov;
 177     uint64_t aio_nbytes;
 178 #define aio_ioctl_cmd   aio_nbytes /* for QEMU_AIO_IOCTL */
 179     off_t aio_offset;
 180     int aio_type;
 181 } RawPosixAIOData;
 182
 183 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
 184 static int cdrom_reopen(BlockDriverState *bs);
 185 #endif
 186
 187 #if defined(__NetBSD__)
 188 static int raw_normalize_devicepath(const char **filename)
 189 {
 190     static char namebuf[PATH_MAX];
 191     const char *dp, *fname;
 192     struct stat sb;
 193
 194     fname = *filename;
 195     dp = strrchr(fname, '/');
 196     if (lstat(fname, &sb) < 0) {
 197         fprintf(stderr, "%s: stat failed: %s\n",
 198             fname, strerror(errno));
 199         return -errno;
 200     }
 201
 202     if (!S_ISBLK(sb.st_mode)) {
 203         return 0;
 204     }
 205
 206     if (dp == NULL) {
 207         snprintf(namebuf, PATH_MAX, "r%s", fname);
 208     } else {
 209         snprintf(namebuf, PATH_MAX, "%.*s/r%s",
 210             (int)(dp - fname), fname, dp + 1);
 211     }
 212     fprintf(stderr, "%s is a block device", fname);
 213     *filename = namebuf;
 214     fprintf(stderr, ", using %s\n", *filename);
 215
 216     return 0;
 217 }
 218 #else
 219 static int raw_normalize_devicepath(const char **filename)
 220 {
 221     return 0;
 222 }
 223 #endif
 224
 225 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
 226 {
 227     BDRVRawState *s = bs->opaque;
 228     char *buf;
 229     unsigned int sector_size;
 230
 231     /* For /dev/sg devices the alignment is not really used.
 232        With buffered I/O, we don't have any restrictions. */
 233     if (bs->sg || !(s->open_flags & O_DIRECT)) {
 234         bs->request_alignment = 1;
 235         s->buf_align = 1;
 236         return;
 237     }
 238
 239     /* Try a few ioctls to get the right size */
 240     bs->request_alignment = 0;
 241     s->buf_align = 0;
 242
 243 #ifdef BLKSSZGET
 244     if (ioctl(fd, BLKSSZGET, &sector_size) >= 0) {
 245         bs->request_alignment = sector_size;
 246     }
 247 #endif
 248 #ifdef DKIOCGETBLOCKSIZE
 249     if (ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) >= 0) {
 250         bs->request_alignment = sector_size;
 251     }
 252 #endif
 253 #ifdef DIOCGSECTORSIZE
 254     if (ioctl(fd, DIOCGSECTORSIZE, &sector_size) >= 0) {
 255         bs->request_alignment = sector_size;
 256     }
 257 #endif
 258 #ifdef CONFIG_XFS
 259     if (s->is_xfs) {
 260         struct dioattr da;
 261         if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
 262             bs->request_alignment = da.d_miniosz;
 263             /* The kernel returns wrong information for d_mem */
 264             /* s->buf_align = da.d_mem; */
 265         }
 266     }
 267 #endif
 268
 269     /* If we could not get the sizes so far, we can only guess them */
 270     if (!s->buf_align) {
 271         size_t align;
 272         buf = qemu_memalign(MAX_BLOCKSIZE, 2 * MAX_BLOCKSIZE);
 273         for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
 274             if (pread(fd, buf + align, MAX_BLOCKSIZE, 0) >= 0) {
 275                 s->buf_align = align;
 276                 break;
 277             }
 278         }
 279         qemu_vfree(buf);
 280     }
 281
 282     if (!bs->request_alignment) {
 283         size_t align;
 284         buf = qemu_memalign(s->buf_align, MAX_BLOCKSIZE);
 285         for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
 286             if (pread(fd, buf, align, 0) >= 0) {
 287                 bs->request_alignment = align;
 288                 break;
 289             }
 290         }
 291         qemu_vfree(buf);
 292     }
 293
 294     if (!s->buf_align || !bs->request_alignment) {
 295         error_setg(errp, "Could not find working O_DIRECT alignment. "
 296                          "Try cache.direct=off.");
 297     }
 298 }
 299
 300 static void raw_parse_flags(int bdrv_flags, int *open_flags)
 301 {
 302     assert(open_flags != NULL);
 303
 304     *open_flags |= O_BINARY;
 305     *open_flags &= ~O_ACCMODE;
 306     if (bdrv_flags & BDRV_O_RDWR) {
 307         *open_flags |= O_RDWR;
 308     } else {
 309         *open_flags |= O_RDONLY;
 310     }
 311
 312     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 313      * and O_DIRECT for no caching. */
 314     if ((bdrv_flags & BDRV_O_NOCACHE)) {
 315         *open_flags |= O_DIRECT;
 316     }
 317 }
 318
 319 static void raw_detach_aio_context(BlockDriverState *bs)
 320 {
 321 #ifdef CONFIG_LINUX_AIO
 322     BDRVRawState *s = bs->opaque;
 323
 324     if (s->use_aio) {
 325         laio_detach_aio_context(s->aio_ctx, bdrv_get_aio_context(bs));
 326     }
 327 #endif
 328 }
 329
 330 static void raw_attach_aio_context(BlockDriverState *bs,
 331                                    AioContext *new_context)
 332 {
 333 #ifdef CONFIG_LINUX_AIO
 334     BDRVRawState *s = bs->opaque;
 335
 336     if (s->use_aio) {
 337         laio_attach_aio_context(s->aio_ctx, new_context);
 338     }
 339 #endif
 340 }
 341
 342 #ifdef CONFIG_LINUX_AIO
 343 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
 344 {
 345     int ret = -1;
 346     assert(aio_ctx != NULL);
 347     assert(use_aio != NULL);
 348     /*
 349      * Currently Linux do AIO only for files opened with O_DIRECT
 350      * specified so check NOCACHE flag too
 351      */
 352     if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
 353                       (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
 354
 355         /* if non-NULL, laio_init() has already been run */
 356         if (*aio_ctx == NULL) {
 357             *aio_ctx = laio_init();
 358             if (!*aio_ctx) {
 359                 goto error;
 360             }
 361         }
 362         *use_aio = 1;
 363     } else {
 364         *use_aio = 0;
 365     }
 366
 367     ret = 0;
 368
 369 error:
 370     return ret;
 371 }
 372 #endif
 373
 374 static void raw_parse_filename(const char *filename, QDict *options,
 375                                Error **errp)
 376 {
 377     /* The filename does not have to be prefixed by the protocol name, since
 378      * "file" is the default protocol; therefore, the return value of this
 379      * function call can be ignored. */
 380     strstart(filename, "file:", &filename);
 381
 382     qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
 383 }
 384
 385 static QemuOptsList raw_runtime_opts = {
 386     .name = "raw",
 387     .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
 388     .desc = {
 389         {
 390             .name = "filename",
 391             .type = QEMU_OPT_STRING,
 392             .help = "File name of the image",
 393         },
 394         { /* end of list */ }
 395     },
 396 };
 397
 398 static int raw_open_common(BlockDriverState *bs, QDict *options,
 399                            int bdrv_flags, int open_flags, Error **errp)
 400 {
 401     BDRVRawState *s = bs->opaque;
 402     QemuOpts *opts;
 403     Error *local_err = NULL;
 404     const char *filename = NULL;
 405     int fd, ret;
 406     struct stat st;
 407
 408     opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
 409     qemu_opts_absorb_qdict(opts, options, &local_err);
 410     if (local_err) {
 411         error_propagate(errp, local_err);
 412         ret = -EINVAL;
 413         goto fail;
 414     }
 415
 416     filename = qemu_opt_get(opts, "filename");
 417
 418     ret = raw_normalize_devicepath(&filename);
 419     if (ret != 0) {
 420         error_setg_errno(errp, -ret, "Could not normalize device path");
 421         goto fail;
 422     }
 423
 424     s->open_flags = open_flags;
 425     raw_parse_flags(bdrv_flags, &s->open_flags);
 426
 427     s->fd = -1;
 428     fd = qemu_open(filename, s->open_flags, 0644);
 429     if (fd < 0) {
 430         ret = -errno;
 431         if (ret == -EROFS) {
 432             ret = -EACCES;
 433         }
 434         goto fail;
 435     }
 436     s->fd = fd;
 437
 438 #ifdef CONFIG_LINUX_AIO
 439     if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
 440         qemu_close(fd);
 441         ret = -errno;
 442         error_setg_errno(errp, -ret, "Could not set AIO state");
 443         goto fail;
 444     }
 445 #endif
 446
 447     s->has_discard = true;
 448     s->has_write_zeroes = true;
 449
 450     if (fstat(s->fd, &st) < 0) {
 451         error_setg_errno(errp, errno, "Could not stat file");
 452         goto fail;
 453     }
 454     if (S_ISREG(st.st_mode)) {
 455         s->discard_zeroes = true;
 456     }
 457     if (S_ISBLK(st.st_mode)) {
 458 #ifdef BLKDISCARDZEROES
 459         unsigned int arg;
 460         if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
 461             s->discard_zeroes = true;
 462         }
 463 #endif
 464 #ifdef __linux__
 465         /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache.  Do
 466          * not rely on the contents of discarded blocks unless using O_DIRECT.
 467          * Same for BLKZEROOUT.
 468          */
 469         if (!(bs->open_flags & BDRV_O_NOCACHE)) {
 470             s->discard_zeroes = false;
 471             s->has_write_zeroes = false;
 472         }
 473 #endif
 474     }
 475
 476 #ifdef CONFIG_XFS
 477     if (platform_test_xfs_fd(s->fd)) {
 478         s->is_xfs = true;
 479     }
 480 #endif
 481
 482     raw_attach_aio_context(bs, bdrv_get_aio_context(bs));
 483
 484     ret = 0;
 485 fail:
 486     if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
 487         unlink(filename);
 488     }
 489     qemu_opts_del(opts);
 490     return ret;
 491 }
 492
 493 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
 494                     Error **errp)
 495 {
 496     BDRVRawState *s = bs->opaque;
 497     Error *local_err = NULL;
 498     int ret;
 499
 500     s->type = FTYPE_FILE;
 501     ret = raw_open_common(bs, options, flags, 0, &local_err);
 502     if (local_err) {
 503         error_propagate(errp, local_err);
 504     }
 505     return ret;
 506 }
 507
 508 static int raw_reopen_prepare(BDRVReopenState *state,
 509                               BlockReopenQueue *queue, Error **errp)
 510 {
 511     BDRVRawState *s;
 512     BDRVRawReopenState *raw_s;
 513     int ret = 0;
 514     Error *local_err = NULL;
 515
 516     assert(state != NULL);
 517     assert(state->bs != NULL);
 518
 519     s = state->bs->opaque;
 520
 521     state->opaque = g_new0(BDRVRawReopenState, 1);
 522     raw_s = state->opaque;
 523
 524 #ifdef CONFIG_LINUX_AIO
 525     raw_s->use_aio = s->use_aio;
 526
 527     /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
 528      * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
 529      * won't override aio_ctx if aio_ctx is non-NULL */
 530     if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
 531         error_setg(errp, "Could not set AIO state");
 532         return -1;
 533     }
 534 #endif
 535
 536     if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
 537         raw_s->open_flags |= O_NONBLOCK;
 538     }
 539
 540     raw_parse_flags(state->flags, &raw_s->open_flags);
 541
 542     raw_s->fd = -1;
 543
 544     int fcntl_flags = O_APPEND | O_NONBLOCK;
 545 #ifdef O_NOATIME
 546     fcntl_flags |= O_NOATIME;
 547 #endif
 548
 549 #ifdef O_ASYNC
 550     /* Not all operating systems have O_ASYNC, and those that don't
 551      * will not let us track the state into raw_s->open_flags (typically
 552      * you achieve the same effect with an ioctl, for example I_SETSIG
 553      * on Solaris). But we do not use O_ASYNC, so that's fine.
 554      */
 555     assert((s->open_flags & O_ASYNC) == 0);
 556 #endif
 557
 558     if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
 559         /* dup the original fd */
 560         /* TODO: use qemu fcntl wrapper */
 561 #ifdef F_DUPFD_CLOEXEC
 562         raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
 563 #else
 564         raw_s->fd = dup(s->fd);
 565         if (raw_s->fd != -1) {
 566             qemu_set_cloexec(raw_s->fd);
 567         }
 568 #endif
 569         if (raw_s->fd >= 0) {
 570             ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
 571             if (ret) {
 572                 qemu_close(raw_s->fd);
 573                 raw_s->fd = -1;
 574             }
 575         }
 576     }
 577
 578     /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
 579     if (raw_s->fd == -1) {
 580         assert(!(raw_s->open_flags & O_CREAT));
 581         raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
 582         if (raw_s->fd == -1) {
 583             error_setg_errno(errp, errno, "Could not reopen file");
 584             ret = -1;
 585         }
 586     }
 587
 588     /* Fail already reopen_prepare() if we can't get a working O_DIRECT
 589      * alignment with the new fd. */
 590     if (raw_s->fd != -1) {
 591         raw_probe_alignment(state->bs, raw_s->fd, &local_err);
 592         if (local_err) {
 593             qemu_close(raw_s->fd);
 594             raw_s->fd = -1;
 595             error_propagate(errp, local_err);
 596             ret = -EINVAL;
 597         }
 598     }
 599
 600     return ret;
 601 }
 602
 603 static void raw_reopen_commit(BDRVReopenState *state)
 604 {
 605     BDRVRawReopenState *raw_s = state->opaque;
 606     BDRVRawState *s = state->bs->opaque;
 607
 608     s->open_flags = raw_s->open_flags;
 609
 610     qemu_close(s->fd);
 611     s->fd = raw_s->fd;
 612 #ifdef CONFIG_LINUX_AIO
 613     s->use_aio = raw_s->use_aio;
 614 #endif
 615
 616     g_free(state->opaque);
 617     state->opaque = NULL;
 618 }
 619
 620
 621 static void raw_reopen_abort(BDRVReopenState *state)
 622 {
 623     BDRVRawReopenState *raw_s = state->opaque;
 624
 625      /* nothing to do if NULL, we didn't get far enough */
 626     if (raw_s == NULL) {
 627         return;
 628     }
 629
 630     if (raw_s->fd >= 0) {
 631         qemu_close(raw_s->fd);
 632         raw_s->fd = -1;
 633     }
 634     g_free(state->opaque);
 635     state->opaque = NULL;
 636 }
 637
 638 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
 639 {
 640     BDRVRawState *s = bs->opaque;
 641
 642     raw_probe_alignment(bs, s->fd, errp);
 643     bs->bl.opt_mem_alignment = s->buf_align;
 644 }
 645
 646 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
 647 {
 648     int ret;
 649
 650     ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
 651     if (ret == -1) {
 652         return -errno;
 653     }
 654
 655     return 0;
 656 }
 657
 658 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
 659 {
 660     int ret;
 661
 662     ret = qemu_fdatasync(aiocb->aio_fildes);
 663     if (ret == -1) {
 664         return -errno;
 665     }
 666     return 0;
 667 }
 668
 669 #ifdef CONFIG_PREADV
 670
 671 static bool preadv_present = true;
 672
 673 static ssize_t
 674 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 675 {
 676     return preadv(fd, iov, nr_iov, offset);
 677 }
 678
 679 static ssize_t
 680 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 681 {
 682     return pwritev(fd, iov, nr_iov, offset);
 683 }
 684
 685 #else
 686
 687 static bool preadv_present = false;
 688
 689 static ssize_t
 690 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 691 {
 692     return -ENOSYS;
 693 }
 694
 695 static ssize_t
 696 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 697 {
 698     return -ENOSYS;
 699 }
 700
 701 #endif
 702
 703 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
 704 {
 705     ssize_t len;
 706
 707     do {
 708         if (aiocb->aio_type & QEMU_AIO_WRITE)
 709             len = qemu_pwritev(aiocb->aio_fildes,
 710                                aiocb->aio_iov,
 711                                aiocb->aio_niov,
 712                                aiocb->aio_offset);
 713          else
 714             len = qemu_preadv(aiocb->aio_fildes,
 715                               aiocb->aio_iov,
 716                               aiocb->aio_niov,
 717                               aiocb->aio_offset);
 718     } while (len == -1 && errno == EINTR);
 719
 720     if (len == -1) {
 721         return -errno;
 722     }
 723     return len;
 724 }
 725
 726 /*
 727  * Read/writes the data to/from a given linear buffer.
 728  *
 729  * Returns the number of bytes handles or -errno in case of an error. Short
 730  * reads are only returned if the end of the file is reached.
 731  */
 732 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
 733 {
 734     ssize_t offset = 0;
 735     ssize_t len;
 736
 737     while (offset < aiocb->aio_nbytes) {
 738         if (aiocb->aio_type & QEMU_AIO_WRITE) {
 739             len = pwrite(aiocb->aio_fildes,
 740                          (const char *)buf + offset,
 741                          aiocb->aio_nbytes - offset,
 742                          aiocb->aio_offset + offset);
 743         } else {
 744             len = pread(aiocb->aio_fildes,
 745                         buf + offset,
 746                         aiocb->aio_nbytes - offset,
 747                         aiocb->aio_offset + offset);
 748         }
 749         if (len == -1 && errno == EINTR) {
 750             continue;
 751         } else if (len == -1 && errno == EINVAL &&
 752                    (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
 753                    !(aiocb->aio_type & QEMU_AIO_WRITE) &&
 754                    offset > 0) {
 755             /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
 756              * after a short read.  Assume that O_DIRECT short reads only occur
 757              * at EOF.  Therefore this is a short read, not an I/O error.
 758              */
 759             break;
 760         } else if (len == -1) {
 761             offset = -errno;
 762             break;
 763         } else if (len == 0) {
 764             break;
 765         }
 766         offset += len;
 767     }
 768
 769     return offset;
 770 }
 771
 772 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
 773 {
 774     ssize_t nbytes;
 775     char *buf;
 776
 777     if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
 778         /*
 779          * If there is just a single buffer, and it is properly aligned
 780          * we can just use plain pread/pwrite without any problems.
 781          */
 782         if (aiocb->aio_niov == 1) {
 783              return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
 784         }
 785         /*
 786          * We have more than one iovec, and all are properly aligned.
 787          *
 788          * Try preadv/pwritev first and fall back to linearizing the
 789          * buffer if it's not supported.
 790          */
 791         if (preadv_present) {
 792             nbytes = handle_aiocb_rw_vector(aiocb);
 793             if (nbytes == aiocb->aio_nbytes ||
 794                 (nbytes < 0 && nbytes != -ENOSYS)) {
 795                 return nbytes;
 796             }
 797             preadv_present = false;
 798         }
 799
 800         /*
 801          * XXX(hch): short read/write.  no easy way to handle the reminder
 802          * using these interfaces.  For now retry using plain
 803          * pread/pwrite?
 804          */
 805     }
 806
 807     /*
 808      * Ok, we have to do it the hard way, copy all segments into
 809      * a single aligned buffer.
 810      */
 811     buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
 812     if (buf == NULL) {
 813         return -ENOMEM;
 814     }
 815
 816     if (aiocb->aio_type & QEMU_AIO_WRITE) {
 817         char *p = buf;
 818         int i;
 819
 820         for (i = 0; i < aiocb->aio_niov; ++i) {
 821             memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
 822             p += aiocb->aio_iov[i].iov_len;
 823         }
 824         assert(p - buf == aiocb->aio_nbytes);
 825     }
 826
 827     nbytes = handle_aiocb_rw_linear(aiocb, buf);
 828     if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
 829         char *p = buf;
 830         size_t count = aiocb->aio_nbytes, copy;
 831         int i;
 832
 833         for (i = 0; i < aiocb->aio_niov && count; ++i) {
 834             copy = count;
 835             if (copy > aiocb->aio_iov[i].iov_len) {
 836                 copy = aiocb->aio_iov[i].iov_len;
 837             }
 838             memcpy(aiocb->aio_iov[i].iov_base, p, copy);
 839             assert(count >= copy);
 840             p     += copy;
 841             count -= copy;
 842         }
 843         assert(count == 0);
 844     }
 845     qemu_vfree(buf);
 846
 847     return nbytes;
 848 }
 849
 850 #ifdef CONFIG_XFS
 851 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
 852 {
 853     struct xfs_flock64 fl;
 854
 855     memset(&fl, 0, sizeof(fl));
 856     fl.l_whence = SEEK_SET;
 857     fl.l_start = offset;
 858     fl.l_len = bytes;
 859
 860     if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
 861         DEBUG_BLOCK_PRINT("cannot write zero range (%s)\n", strerror(errno));
 862         return -errno;
 863     }
 864
 865     return 0;
 866 }
 867
 868 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
 869 {
 870     struct xfs_flock64 fl;
 871
 872     memset(&fl, 0, sizeof(fl));
 873     fl.l_whence = SEEK_SET;
 874     fl.l_start = offset;
 875     fl.l_len = bytes;
 876
 877     if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
 878         DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
 879         return -errno;
 880     }
 881
 882     return 0;
 883 }
 884 #endif
 885
 886 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
 887 {
 888     int ret = -EOPNOTSUPP;
 889     BDRVRawState *s = aiocb->bs->opaque;
 890
 891     if (s->has_write_zeroes == 0) {
 892         return -ENOTSUP;
 893     }
 894
 895     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
 896 #ifdef BLKZEROOUT
 897         do {
 898             uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
 899             if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
 900                 return 0;
 901             }
 902         } while (errno == EINTR);
 903
 904         ret = -errno;
 905 #endif
 906     } else {
 907 #ifdef CONFIG_XFS
 908         if (s->is_xfs) {
 909             return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
 910         }
 911 #endif
 912     }
 913
 914     if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
 915         ret == -ENOTTY) {
 916         s->has_write_zeroes = false;
 917         ret = -ENOTSUP;
 918     }
 919     return ret;
 920 }
 921
 922 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
 923 {
 924     int ret = -EOPNOTSUPP;
 925     BDRVRawState *s = aiocb->bs->opaque;
 926
 927     if (!s->has_discard) {
 928         return -ENOTSUP;
 929     }
 930
 931     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
 932 #ifdef BLKDISCARD
 933         do {
 934             uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
 935             if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
 936                 return 0;
 937             }
 938         } while (errno == EINTR);
 939
 940         ret = -errno;
 941 #endif
 942     } else {
 943 #ifdef CONFIG_XFS
 944         if (s->is_xfs) {
 945             return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
 946         }
 947 #endif
 948
 949 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
 950         do {
 951             if (fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
 952                           aiocb->aio_offset, aiocb->aio_nbytes) == 0) {
 953                 return 0;
 954             }
 955         } while (errno == EINTR);
 956
 957         ret = -errno;
 958 #endif
 959     }
 960
 961     if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
 962         ret == -ENOTTY) {
 963         s->has_discard = false;
 964         ret = -ENOTSUP;
 965     }
 966     return ret;
 967 }
 968
 969 static int aio_worker(void *arg)
 970 {
 971     RawPosixAIOData *aiocb = arg;
 972     ssize_t ret = 0;
 973
 974     switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
 975     case QEMU_AIO_READ:
 976         ret = handle_aiocb_rw(aiocb);
 977         if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) {
 978             iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
 979                       0, aiocb->aio_nbytes - ret);
 980
 981             ret = aiocb->aio_nbytes;
 982         }
 983         if (ret == aiocb->aio_nbytes) {
 984             ret = 0;
 985         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 986             ret = -EINVAL;
 987         }
 988         break;
 989     case QEMU_AIO_WRITE:
 990         ret = handle_aiocb_rw(aiocb);
 991         if (ret == aiocb->aio_nbytes) {
 992             ret = 0;
 993         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 994             ret = -EINVAL;
 995         }
 996         break;
 997     case QEMU_AIO_FLUSH:
 998         ret = handle_aiocb_flush(aiocb);
 999         break;
1000     case QEMU_AIO_IOCTL:
1001         ret = handle_aiocb_ioctl(aiocb);
1002         break;
1003     case QEMU_AIO_DISCARD:
1004         ret = handle_aiocb_discard(aiocb);
1005         break;
1006     case QEMU_AIO_WRITE_ZEROES:
1007         ret = handle_aiocb_write_zeroes(aiocb);
1008         break;
1009     default:
1010         fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1011         ret = -EINVAL;
1012         break;
1013     }
1014
1015     g_slice_free(RawPosixAIOData, aiocb);
1016     return ret;
1017 }
1018
1019 static int paio_submit_co(BlockDriverState *bs, int fd,
1020         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1021         int type)
1022 {
1023     RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
1024     ThreadPool *pool;
1025
1026     acb->bs = bs;
1027     acb->aio_type = type;
1028     acb->aio_fildes = fd;
1029
1030     acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1031     acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1032
1033     if (qiov) {
1034         acb->aio_iov = qiov->iov;
1035         acb->aio_niov = qiov->niov;
1036         assert(qiov->size == acb->aio_nbytes);
1037     }
1038
1039     trace_paio_submit_co(sector_num, nb_sectors, type);
1040     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1041     return thread_pool_submit_co(pool, aio_worker, acb);
1042 }
1043
1044 static BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd,
1045         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1046         BlockDriverCompletionFunc *cb, void *opaque, int type)
1047 {
1048     RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
1049     ThreadPool *pool;
1050
1051     acb->bs = bs;
1052     acb->aio_type = type;
1053     acb->aio_fildes = fd;
1054
1055     acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1056     acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1057
1058     if (qiov) {
1059         acb->aio_iov = qiov->iov;
1060         acb->aio_niov = qiov->niov;
1061         assert(qiov->size == acb->aio_nbytes);
1062     }
1063
1064     trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
1065     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1066     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1067 }
1068
1069 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
1070         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1071         BlockDriverCompletionFunc *cb, void *opaque, int type)
1072 {
1073     BDRVRawState *s = bs->opaque;
1074
1075     if (fd_open(bs) < 0)
1076         return NULL;
1077
1078     /*
1079      * If O_DIRECT is used the buffer needs to be aligned on a sector
1080      * boundary.  Check if this is the case or tell the low-level
1081      * driver that it needs to copy the buffer.
1082      */
1083     if ((bs->open_flags & BDRV_O_NOCACHE)) {
1084         if (!bdrv_qiov_is_aligned(bs, qiov)) {
1085             type |= QEMU_AIO_MISALIGNED;
1086 #ifdef CONFIG_LINUX_AIO
1087         } else if (s->use_aio) {
1088             return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
1089                                nb_sectors, cb, opaque, type);
1090 #endif
1091         }
1092     }
1093
1094     return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
1095                        cb, opaque, type);
1096 }
1097
1098 static void raw_aio_plug(BlockDriverState *bs)
1099 {
1100 #ifdef CONFIG_LINUX_AIO
1101     BDRVRawState *s = bs->opaque;
1102     if (s->use_aio) {
1103         laio_io_plug(bs, s->aio_ctx);
1104     }
1105 #endif
1106 }
1107
1108 static void raw_aio_unplug(BlockDriverState *bs)
1109 {
1110 #ifdef CONFIG_LINUX_AIO
1111     BDRVRawState *s = bs->opaque;
1112     if (s->use_aio) {
1113         laio_io_unplug(bs, s->aio_ctx, true);
1114     }
1115 #endif
1116 }
1117
1118 static void raw_aio_flush_io_queue(BlockDriverState *bs)
1119 {
1120 #ifdef CONFIG_LINUX_AIO
1121     BDRVRawState *s = bs->opaque;
1122     if (s->use_aio) {
1123         laio_io_unplug(bs, s->aio_ctx, false);
1124     }
1125 #endif
1126 }
1127
1128 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
1129         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1130         BlockDriverCompletionFunc *cb, void *opaque)
1131 {
1132     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1133                           cb, opaque, QEMU_AIO_READ);
1134 }
1135
1136 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
1137         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1138         BlockDriverCompletionFunc *cb, void *opaque)
1139 {
1140     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1141                           cb, opaque, QEMU_AIO_WRITE);
1142 }
1143
1144 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
1145         BlockDriverCompletionFunc *cb, void *opaque)
1146 {
1147     BDRVRawState *s = bs->opaque;
1148
1149     if (fd_open(bs) < 0)
1150         return NULL;
1151
1152     return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1153 }
1154
1155 static void raw_close(BlockDriverState *bs)
1156 {
1157     BDRVRawState *s = bs->opaque;
1158
1159     raw_detach_aio_context(bs);
1160
1161 #ifdef CONFIG_LINUX_AIO
1162     if (s->use_aio) {
1163         laio_cleanup(s->aio_ctx);
1164     }
1165 #endif
1166     if (s->fd >= 0) {
1167         qemu_close(s->fd);
1168         s->fd = -1;
1169     }
1170 }
1171
1172 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1173 {
1174     BDRVRawState *s = bs->opaque;
1175     struct stat st;
1176
1177     if (fstat(s->fd, &st)) {
1178         return -errno;
1179     }
1180
1181     if (S_ISREG(st.st_mode)) {
1182         if (ftruncate(s->fd, offset) < 0) {
1183             return -errno;
1184         }
1185     } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1186        if (offset > raw_getlength(bs)) {
1187            return -EINVAL;
1188        }
1189     } else {
1190         return -ENOTSUP;
1191     }
1192
1193     return 0;
1194 }
1195
1196 #ifdef __OpenBSD__
1197 static int64_t raw_getlength(BlockDriverState *bs)
1198 {
1199     BDRVRawState *s = bs->opaque;
1200     int fd = s->fd;
1201     struct stat st;
1202
1203     if (fstat(fd, &st))
1204         return -errno;
1205     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1206         struct disklabel dl;
1207
1208         if (ioctl(fd, DIOCGDINFO, &dl))
1209             return -errno;
1210         return (uint64_t)dl.d_secsize *
1211             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1212     } else
1213         return st.st_size;
1214 }
1215 #elif defined(__NetBSD__)
1216 static int64_t raw_getlength(BlockDriverState *bs)
1217 {
1218     BDRVRawState *s = bs->opaque;
1219     int fd = s->fd;
1220     struct stat st;
1221
1222     if (fstat(fd, &st))
1223         return -errno;
1224     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1225         struct dkwedge_info dkw;
1226
1227         if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1228             return dkw.dkw_size * 512;
1229         } else {
1230             struct disklabel dl;
1231
1232             if (ioctl(fd, DIOCGDINFO, &dl))
1233                 return -errno;
1234             return (uint64_t)dl.d_secsize *
1235                 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1236         }
1237     } else
1238         return st.st_size;
1239 }
1240 #elif defined(__sun__)
1241 static int64_t raw_getlength(BlockDriverState *bs)
1242 {
1243     BDRVRawState *s = bs->opaque;
1244     struct dk_minfo minfo;
1245     int ret;
1246     int64_t size;
1247
1248     ret = fd_open(bs);
1249     if (ret < 0) {
1250         return ret;
1251     }
1252
1253     /*
1254      * Use the DKIOCGMEDIAINFO ioctl to read the size.
1255      */
1256     ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1257     if (ret != -1) {
1258         return minfo.dki_lbsize * minfo.dki_capacity;
1259     }
1260
1261     /*
1262      * There are reports that lseek on some devices fails, but
1263      * irc discussion said that contingency on contingency was overkill.
1264      */
1265     size = lseek(s->fd, 0, SEEK_END);
1266     if (size < 0) {
1267         return -errno;
1268     }
1269     return size;
1270 }
1271 #elif defined(CONFIG_BSD)
1272 static int64_t raw_getlength(BlockDriverState *bs)
1273 {
1274     BDRVRawState *s = bs->opaque;
1275     int fd = s->fd;
1276     int64_t size;
1277     struct stat sb;
1278 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1279     int reopened = 0;
1280 #endif
1281     int ret;
1282
1283     ret = fd_open(bs);
1284     if (ret < 0)
1285         return ret;
1286
1287 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1288 again:
1289 #endif
1290     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1291 #ifdef DIOCGMEDIASIZE
1292         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1293 #elif defined(DIOCGPART)
1294         {
1295                 struct partinfo pi;
1296                 if (ioctl(fd, DIOCGPART, &pi) == 0)
1297                         size = pi.media_size;
1298                 else
1299                         size = 0;
1300         }
1301         if (size == 0)
1302 #endif
1303 #if defined(__APPLE__) && defined(__MACH__)
1304         size = LLONG_MAX;
1305 #else
1306         size = lseek(fd, 0LL, SEEK_END);
1307         if (size < 0) {
1308             return -errno;
1309         }
1310 #endif
1311 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1312         switch(s->type) {
1313         case FTYPE_CD:
1314             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1315             if (size == 2048LL * (unsigned)-1)
1316                 size = 0;
1317             /* XXX no disc?  maybe we need to reopen... */
1318             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1319                 reopened = 1;
1320                 goto again;
1321             }
1322         }
1323 #endif
1324     } else {
1325         size = lseek(fd, 0, SEEK_END);
1326         if (size < 0) {
1327             return -errno;
1328         }
1329     }
1330     return size;
1331 }
1332 #else
1333 static int64_t raw_getlength(BlockDriverState *bs)
1334 {
1335     BDRVRawState *s = bs->opaque;
1336     int ret;
1337     int64_t size;
1338
1339     ret = fd_open(bs);
1340     if (ret < 0) {
1341         return ret;
1342     }
1343
1344     size = lseek(s->fd, 0, SEEK_END);
1345     if (size < 0) {
1346         return -errno;
1347     }
1348     return size;
1349 }
1350 #endif
1351
1352 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1353 {
1354     struct stat st;
1355     BDRVRawState *s = bs->opaque;
1356
1357     if (fstat(s->fd, &st) < 0) {
1358         return -errno;
1359     }
1360     return (int64_t)st.st_blocks * 512;
1361 }
1362
1363 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
1364 {
1365     int fd;
1366     int result = 0;
1367     int64_t total_size = 0;
1368     bool nocow = false;
1369     PreallocMode prealloc;
1370     char *buf = NULL;
1371     Error *local_err = NULL;
1372
1373     strstart(filename, "file:", &filename);
1374
1375     /* Read out options */
1376     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1377                           BDRV_SECTOR_SIZE);
1378     nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
1379     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1380     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
1381                                PREALLOC_MODE_MAX, PREALLOC_MODE_OFF,
1382                                &local_err);
1383     g_free(buf);
1384     if (local_err) {
1385         error_propagate(errp, local_err);
1386         result = -EINVAL;
1387         goto out;
1388     }
1389
1390     fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
1391                    0644);
1392     if (fd < 0) {
1393         result = -errno;
1394         error_setg_errno(errp, -result, "Could not create file");
1395         goto out;
1396     }
1397
1398     if (nocow) {
1399 #ifdef __linux__
1400         /* Set NOCOW flag to solve performance issue on fs like btrfs.
1401          * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1402          * will be ignored since any failure of this operation should not
1403          * block the left work.
1404          */
1405         int attr;
1406         if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
1407             attr |= FS_NOCOW_FL;
1408             ioctl(fd, FS_IOC_SETFLAGS, &attr);
1409         }
1410 #endif
1411     }
1412
1413     if (ftruncate(fd, total_size) != 0) {
1414         result = -errno;
1415         error_setg_errno(errp, -result, "Could not resize file");
1416         goto out_close;
1417     }
1418
1419     switch (prealloc) {
1420 #ifdef CONFIG_POSIX_FALLOCATE
1421     case PREALLOC_MODE_FALLOC:
1422         /* posix_fallocate() doesn't set errno. */
1423         result = -posix_fallocate(fd, 0, total_size);
1424         if (result != 0) {
1425             error_setg_errno(errp, -result,
1426                              "Could not preallocate data for the new file");
1427         }
1428         break;
1429 #endif
1430     case PREALLOC_MODE_FULL:
1431     {
1432         int64_t num = 0, left = total_size;
1433         buf = g_malloc0(65536);
1434
1435         while (left > 0) {
1436             num = MIN(left, 65536);
1437             result = write(fd, buf, num);
1438             if (result < 0) {
1439                 result = -errno;
1440                 error_setg_errno(errp, -result,
1441                                  "Could not write to the new file");
1442                 break;
1443             }
1444             left -= num;
1445         }
1446         fsync(fd);
1447         g_free(buf);
1448         break;
1449     }
1450     case PREALLOC_MODE_OFF:
1451         break;
1452     default:
1453         result = -EINVAL;
1454         error_setg(errp, "Unsupported preallocation mode: %s",
1455                    PreallocMode_lookup[prealloc]);
1456         break;
1457     }
1458
1459 out_close:
1460     if (qemu_close(fd) != 0 && result == 0) {
1461         result = -errno;
1462         error_setg_errno(errp, -result, "Could not close the new file");
1463     }
1464 out:
1465     return result;
1466 }
1467
1468 static int64_t try_fiemap(BlockDriverState *bs, off_t start, off_t *data,
1469                           off_t *hole, int nb_sectors, int *pnum)
1470 {
1471 #ifdef CONFIG_FIEMAP
1472     BDRVRawState *s = bs->opaque;
1473     int64_t ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1474     struct {
1475         struct fiemap fm;
1476         struct fiemap_extent fe;
1477     } f;
1478
1479     if (s->skip_fiemap) {
1480         return -ENOTSUP;
1481     }
1482
1483     f.fm.fm_start = start;
1484     f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE;
1485     f.fm.fm_flags = 0;
1486     f.fm.fm_extent_count = 1;
1487     f.fm.fm_reserved = 0;
1488     if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) {
1489         s->skip_fiemap = true;
1490         return -errno;
1491     }
1492
1493     if (f.fm.fm_mapped_extents == 0) {
1494         /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1495          * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1496          */
1497         off_t length = lseek(s->fd, 0, SEEK_END);
1498         *hole = f.fm.fm_start;
1499         *data = MIN(f.fm.fm_start + f.fm.fm_length, length);
1500     } else {
1501         *data = f.fe.fe_logical;
1502         *hole = f.fe.fe_logical + f.fe.fe_length;
1503         if (f.fe.fe_flags & FIEMAP_EXTENT_UNWRITTEN) {
1504             ret |= BDRV_BLOCK_ZERO;
1505         }
1506     }
1507
1508     return ret;
1509 #else
1510     return -ENOTSUP;
1511 #endif
1512 }
1513
1514 static int64_t try_seek_hole(BlockDriverState *bs, off_t start, off_t *data,
1515                              off_t *hole, int *pnum)
1516 {
1517 #if defined SEEK_HOLE && defined SEEK_DATA
1518     BDRVRawState *s = bs->opaque;
1519
1520     *hole = lseek(s->fd, start, SEEK_HOLE);
1521     if (*hole == -1) {
1522         /* -ENXIO indicates that sector_num was past the end of the file.
1523          * There is a virtual hole there.  */
1524         assert(errno != -ENXIO);
1525
1526         return -errno;
1527     }
1528
1529     if (*hole > start) {
1530         *data = start;
1531     } else {
1532         /* On a hole.  We need another syscall to find its end.  */
1533         *data = lseek(s->fd, start, SEEK_DATA);
1534         if (*data == -1) {
1535             *data = lseek(s->fd, 0, SEEK_END);
1536         }
1537     }
1538
1539     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1540 #else
1541     return -ENOTSUP;
1542 #endif
1543 }
1544
1545 /*
1546  * Returns true iff the specified sector is present in the disk image. Drivers
1547  * not implementing the functionality are assumed to not support backing files,
1548  * hence all their sectors are reported as allocated.
1549  *
1550  * If 'sector_num' is beyond the end of the disk image the return value is 0
1551  * and 'pnum' is set to 0.
1552  *
1553  * 'pnum' is set to the number of sectors (including and immediately following
1554  * the specified sector) that are known to be in the same
1555  * allocated/unallocated state.
1556  *
1557  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
1558  * beyond the end of the disk image it will be clamped.
1559  */
1560 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1561                                                     int64_t sector_num,
1562                                                     int nb_sectors, int *pnum)
1563 {
1564     off_t start, data = 0, hole = 0;
1565     int64_t ret;
1566
1567     ret = fd_open(bs);
1568     if (ret < 0) {
1569         return ret;
1570     }
1571
1572     start = sector_num * BDRV_SECTOR_SIZE;
1573
1574     ret = try_fiemap(bs, start, &data, &hole, nb_sectors, pnum);
1575     if (ret < 0) {
1576         ret = try_seek_hole(bs, start, &data, &hole, pnum);
1577         if (ret < 0) {
1578             /* Assume everything is allocated. */
1579             data = 0;
1580             hole = start + nb_sectors * BDRV_SECTOR_SIZE;
1581             ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1582         }
1583     }
1584
1585     if (data <= start) {
1586         /* On a data extent, compute sectors to the end of the extent.  */
1587         *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1588     } else {
1589         /* On a hole, compute sectors to the beginning of the next extent.  */
1590         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1591         ret &= ~BDRV_BLOCK_DATA;
1592         ret |= BDRV_BLOCK_ZERO;
1593     }
1594
1595     return ret;
1596 }
1597
1598 static coroutine_fn BlockDriverAIOCB *raw_aio_discard(BlockDriverState *bs,
1599     int64_t sector_num, int nb_sectors,
1600     BlockDriverCompletionFunc *cb, void *opaque)
1601 {
1602     BDRVRawState *s = bs->opaque;
1603
1604     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1605                        cb, opaque, QEMU_AIO_DISCARD);
1606 }
1607
1608 static int coroutine_fn raw_co_write_zeroes(
1609     BlockDriverState *bs, int64_t sector_num,
1610     int nb_sectors, BdrvRequestFlags flags)
1611 {
1612     BDRVRawState *s = bs->opaque;
1613
1614     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1615         return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1616                               QEMU_AIO_WRITE_ZEROES);
1617     } else if (s->discard_zeroes) {
1618         return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1619                               QEMU_AIO_DISCARD);
1620     }
1621     return -ENOTSUP;
1622 }
1623
1624 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1625 {
1626     BDRVRawState *s = bs->opaque;
1627
1628     bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1629     bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1630     return 0;
1631 }
1632
1633 static QemuOptsList raw_create_opts = {
1634     .name = "raw-create-opts",
1635     .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
1636     .desc = {
1637         {
1638             .name = BLOCK_OPT_SIZE,
1639             .type = QEMU_OPT_SIZE,
1640             .help = "Virtual disk size"
1641         },
1642         {
1643             .name = BLOCK_OPT_NOCOW,
1644             .type = QEMU_OPT_BOOL,
1645             .help = "Turn off copy-on-write (valid only on btrfs)"
1646         },
1647         {
1648             .name = BLOCK_OPT_PREALLOC,
1649             .type = QEMU_OPT_STRING,
1650             .help = "Preallocation mode (allowed values: off, falloc, full)"
1651         },
1652         { /* end of list */ }
1653     }
1654 };
1655
1656 static BlockDriver bdrv_file = {
1657     .format_name = "file",
1658     .protocol_name = "file",
1659     .instance_size = sizeof(BDRVRawState),
1660     .bdrv_needs_filename = true,
1661     .bdrv_probe = NULL, /* no probe for protocols */
1662     .bdrv_parse_filename = raw_parse_filename,
1663     .bdrv_file_open = raw_open,
1664     .bdrv_reopen_prepare = raw_reopen_prepare,
1665     .bdrv_reopen_commit = raw_reopen_commit,
1666     .bdrv_reopen_abort = raw_reopen_abort,
1667     .bdrv_close = raw_close,
1668     .bdrv_create = raw_create,
1669     .bdrv_has_zero_init = bdrv_has_zero_init_1,
1670     .bdrv_co_get_block_status = raw_co_get_block_status,
1671     .bdrv_co_write_zeroes = raw_co_write_zeroes,
1672
1673     .bdrv_aio_readv = raw_aio_readv,
1674     .bdrv_aio_writev = raw_aio_writev,
1675     .bdrv_aio_flush = raw_aio_flush,
1676     .bdrv_aio_discard = raw_aio_discard,
1677     .bdrv_refresh_limits = raw_refresh_limits,
1678     .bdrv_io_plug = raw_aio_plug,
1679     .bdrv_io_unplug = raw_aio_unplug,
1680     .bdrv_flush_io_queue = raw_aio_flush_io_queue,
1681
1682     .bdrv_truncate = raw_truncate,
1683     .bdrv_getlength = raw_getlength,
1684     .bdrv_get_info = raw_get_info,
1685     .bdrv_get_allocated_file_size
1686                         = raw_get_allocated_file_size,
1687
1688     .bdrv_detach_aio_context = raw_detach_aio_context,
1689     .bdrv_attach_aio_context = raw_attach_aio_context,
1690
1691     .create_opts = &raw_create_opts,
1692 };
1693
1694 /***********************************************/
1695 /* host device */
1696
1697 #if defined(__APPLE__) && defined(__MACH__)
1698 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1699 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1700
1701 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1702 {
1703     kern_return_t       kernResult;
1704     mach_port_t     masterPort;
1705     CFMutableDictionaryRef  classesToMatch;
1706
1707     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1708     if ( KERN_SUCCESS != kernResult ) {
1709         printf( "IOMasterPort returned %d\n", kernResult );
1710     }
1711
1712     classesToMatch = IOServiceMatching( kIOCDMediaClass );
1713     if ( classesToMatch == NULL ) {
1714         printf( "IOServiceMatching returned a NULL dictionary.\n" );
1715     } else {
1716     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1717     }
1718     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1719     if ( KERN_SUCCESS != kernResult )
1720     {
1721         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1722     }
1723
1724     return kernResult;
1725 }
1726
1727 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
1728 {
1729     io_object_t     nextMedia;
1730     kern_return_t   kernResult = KERN_FAILURE;
1731     *bsdPath = '\0';
1732     nextMedia = IOIteratorNext( mediaIterator );
1733     if ( nextMedia )
1734     {
1735         CFTypeRef   bsdPathAsCFString;
1736     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1737         if ( bsdPathAsCFString ) {
1738             size_t devPathLength;
1739             strcpy( bsdPath, _PATH_DEV );
1740             strcat( bsdPath, "r" );
1741             devPathLength = strlen( bsdPath );
1742             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1743                 kernResult = KERN_SUCCESS;
1744             }
1745             CFRelease( bsdPathAsCFString );
1746         }
1747         IOObjectRelease( nextMedia );
1748     }
1749
1750     return kernResult;
1751 }
1752
1753 #endif
1754
1755 static int hdev_probe_device(const char *filename)
1756 {
1757     struct stat st;
1758
1759     /* allow a dedicated CD-ROM driver to match with a higher priority */
1760     if (strstart(filename, "/dev/cdrom", NULL))
1761         return 50;
1762
1763     if (stat(filename, &st) >= 0 &&
1764             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
1765         return 100;
1766     }
1767
1768     return 0;
1769 }
1770
1771 static int check_hdev_writable(BDRVRawState *s)
1772 {
1773 #if defined(BLKROGET)
1774     /* Linux block devices can be configured "read-only" using blockdev(8).
1775      * This is independent of device node permissions and therefore open(2)
1776      * with O_RDWR succeeds.  Actual writes fail with EPERM.
1777      *
1778      * bdrv_open() is supposed to fail if the disk is read-only.  Explicitly
1779      * check for read-only block devices so that Linux block devices behave
1780      * properly.
1781      */
1782     struct stat st;
1783     int readonly = 0;
1784
1785     if (fstat(s->fd, &st)) {
1786         return -errno;
1787     }
1788
1789     if (!S_ISBLK(st.st_mode)) {
1790         return 0;
1791     }
1792
1793     if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
1794         return -errno;
1795     }
1796
1797     if (readonly) {
1798         return -EACCES;
1799     }
1800 #endif /* defined(BLKROGET) */
1801     return 0;
1802 }
1803
1804 static void hdev_parse_filename(const char *filename, QDict *options,
1805                                 Error **errp)
1806 {
1807     /* The prefix is optional, just as for "file". */
1808     strstart(filename, "host_device:", &filename);
1809
1810     qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
1811 }
1812
1813 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
1814                      Error **errp)
1815 {
1816     BDRVRawState *s = bs->opaque;
1817     Error *local_err = NULL;
1818     int ret;
1819     const char *filename = qdict_get_str(options, "filename");
1820
1821 #if defined(__APPLE__) && defined(__MACH__)
1822     if (strstart(filename, "/dev/cdrom", NULL)) {
1823         kern_return_t kernResult;
1824         io_iterator_t mediaIterator;
1825         char bsdPath[ MAXPATHLEN ];
1826         int fd;
1827
1828         kernResult = FindEjectableCDMedia( &mediaIterator );
1829         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1830
1831         if ( bsdPath[ 0 ] != '\0' ) {
1832             strcat(bsdPath,"s0");
1833             /* some CDs don't have a partition 0 */
1834             fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1835             if (fd < 0) {
1836                 bsdPath[strlen(bsdPath)-1] = '1';
1837             } else {
1838                 qemu_close(fd);
1839             }
1840             filename = bsdPath;
1841             qdict_put(options, "filename", qstring_from_str(filename));
1842         }
1843
1844         if ( mediaIterator )
1845             IOObjectRelease( mediaIterator );
1846     }
1847 #endif
1848
1849     s->type = FTYPE_FILE;
1850 #if defined(__linux__)
1851     {
1852         char resolved_path[ MAXPATHLEN ], *temp;
1853
1854         temp = realpath(filename, resolved_path);
1855         if (temp && strstart(temp, "/dev/sg", NULL)) {
1856             bs->sg = 1;
1857         }
1858     }
1859 #endif
1860
1861     ret = raw_open_common(bs, options, flags, 0, &local_err);
1862     if (ret < 0) {
1863         if (local_err) {
1864             error_propagate(errp, local_err);
1865         }
1866         return ret;
1867     }
1868
1869     if (flags & BDRV_O_RDWR) {
1870         ret = check_hdev_writable(s);
1871         if (ret < 0) {
1872             raw_close(bs);
1873             error_setg_errno(errp, -ret, "The device is not writable");
1874             return ret;
1875         }
1876     }
1877
1878     return ret;
1879 }
1880
1881 #if defined(__linux__)
1882 /* Note: we do not have a reliable method to detect if the floppy is
1883    present. The current method is to try to open the floppy at every
1884    I/O and to keep it opened during a few hundreds of ms. */
1885 static int fd_open(BlockDriverState *bs)
1886 {
1887     BDRVRawState *s = bs->opaque;
1888     int last_media_present;
1889
1890     if (s->type != FTYPE_FD)
1891         return 0;
1892     last_media_present = (s->fd >= 0);
1893     if (s->fd >= 0 &&
1894         (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1895         qemu_close(s->fd);
1896         s->fd = -1;
1897 #ifdef DEBUG_FLOPPY
1898         printf("Floppy closed\n");
1899 #endif
1900     }
1901     if (s->fd < 0) {
1902         if (s->fd_got_error &&
1903             (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1904 #ifdef DEBUG_FLOPPY
1905             printf("No floppy (open delayed)\n");
1906 #endif
1907             return -EIO;
1908         }
1909         s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
1910         if (s->fd < 0) {
1911             s->fd_error_time = get_clock();
1912             s->fd_got_error = 1;
1913             if (last_media_present)
1914                 s->fd_media_changed = 1;
1915 #ifdef DEBUG_FLOPPY
1916             printf("No floppy\n");
1917 #endif
1918             return -EIO;
1919         }
1920 #ifdef DEBUG_FLOPPY
1921         printf("Floppy opened\n");
1922 #endif
1923     }
1924     if (!last_media_present)
1925         s->fd_media_changed = 1;
1926     s->fd_open_time = get_clock();
1927     s->fd_got_error = 0;
1928     return 0;
1929 }
1930
1931 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1932 {
1933     BDRVRawState *s = bs->opaque;
1934
1935     return ioctl(s->fd, req, buf);
1936 }
1937
1938 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1939         unsigned long int req, void *buf,
1940         BlockDriverCompletionFunc *cb, void *opaque)
1941 {
1942     BDRVRawState *s = bs->opaque;
1943     RawPosixAIOData *acb;
1944     ThreadPool *pool;
1945
1946     if (fd_open(bs) < 0)
1947         return NULL;
1948
1949     acb = g_slice_new(RawPosixAIOData);
1950     acb->bs = bs;
1951     acb->aio_type = QEMU_AIO_IOCTL;
1952     acb->aio_fildes = s->fd;
1953     acb->aio_offset = 0;
1954     acb->aio_ioctl_buf = buf;
1955     acb->aio_ioctl_cmd = req;
1956     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1957     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1958 }
1959
1960 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1961 static int fd_open(BlockDriverState *bs)
1962 {
1963     BDRVRawState *s = bs->opaque;
1964
1965     /* this is just to ensure s->fd is sane (its called by io ops) */
1966     if (s->fd >= 0)
1967         return 0;
1968     return -EIO;
1969 }
1970 #else /* !linux && !FreeBSD */
1971
1972 static int fd_open(BlockDriverState *bs)
1973 {
1974     return 0;
1975 }
1976
1977 #endif /* !linux && !FreeBSD */
1978
1979 static coroutine_fn BlockDriverAIOCB *hdev_aio_discard(BlockDriverState *bs,
1980     int64_t sector_num, int nb_sectors,
1981     BlockDriverCompletionFunc *cb, void *opaque)
1982 {
1983     BDRVRawState *s = bs->opaque;
1984
1985     if (fd_open(bs) < 0) {
1986         return NULL;
1987     }
1988     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1989                        cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1990 }
1991
1992 static coroutine_fn int hdev_co_write_zeroes(BlockDriverState *bs,
1993     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
1994 {
1995     BDRVRawState *s = bs->opaque;
1996     int rc;
1997
1998     rc = fd_open(bs);
1999     if (rc < 0) {
2000         return rc;
2001     }
2002     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2003         return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2004                               QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2005     } else if (s->discard_zeroes) {
2006         return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2007                               QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2008     }
2009     return -ENOTSUP;
2010 }
2011
2012 static int hdev_create(const char *filename, QemuOpts *opts,
2013                        Error **errp)
2014 {
2015     int fd;
2016     int ret = 0;
2017     struct stat stat_buf;
2018     int64_t total_size = 0;
2019     bool has_prefix;
2020
2021     /* This function is used by all three protocol block drivers and therefore
2022      * any of these three prefixes may be given.
2023      * The return value has to be stored somewhere, otherwise this is an error
2024      * due to -Werror=unused-value. */
2025     has_prefix =
2026         strstart(filename, "host_device:", &filename) ||
2027         strstart(filename, "host_cdrom:" , &filename) ||
2028         strstart(filename, "host_floppy:", &filename);
2029
2030     (void)has_prefix;
2031
2032     /* Read out options */
2033     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2034                           BDRV_SECTOR_SIZE);
2035
2036     fd = qemu_open(filename, O_WRONLY | O_BINARY);
2037     if (fd < 0) {
2038         ret = -errno;
2039         error_setg_errno(errp, -ret, "Could not open device");
2040         return ret;
2041     }
2042
2043     if (fstat(fd, &stat_buf) < 0) {
2044         ret = -errno;
2045         error_setg_errno(errp, -ret, "Could not stat device");
2046     } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2047         error_setg(errp,
2048                    "The given file is neither a block nor a character device");
2049         ret = -ENODEV;
2050     } else if (lseek(fd, 0, SEEK_END) < total_size) {
2051         error_setg(errp, "Device is too small");
2052         ret = -ENOSPC;
2053     }
2054
2055     qemu_close(fd);
2056     return ret;
2057 }
2058
2059 static BlockDriver bdrv_host_device = {
2060     .format_name        = "host_device",
2061     .protocol_name        = "host_device",
2062     .instance_size      = sizeof(BDRVRawState),
2063     .bdrv_needs_filename = true,
2064     .bdrv_probe_device  = hdev_probe_device,
2065     .bdrv_parse_filename = hdev_parse_filename,
2066     .bdrv_file_open     = hdev_open,
2067     .bdrv_close         = raw_close,
2068     .bdrv_reopen_prepare = raw_reopen_prepare,
2069     .bdrv_reopen_commit  = raw_reopen_commit,
2070     .bdrv_reopen_abort   = raw_reopen_abort,
2071     .bdrv_create         = hdev_create,
2072     .create_opts         = &raw_create_opts,
2073     .bdrv_co_write_zeroes = hdev_co_write_zeroes,
2074
2075     .bdrv_aio_readv     = raw_aio_readv,
2076     .bdrv_aio_writev    = raw_aio_writev,
2077     .bdrv_aio_flush     = raw_aio_flush,
2078     .bdrv_aio_discard   = hdev_aio_discard,
2079     .bdrv_refresh_limits = raw_refresh_limits,
2080     .bdrv_io_plug = raw_aio_plug,
2081     .bdrv_io_unplug = raw_aio_unplug,
2082     .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2083
2084     .bdrv_truncate      = raw_truncate,
2085     .bdrv_getlength     = raw_getlength,
2086     .bdrv_get_info = raw_get_info,
2087     .bdrv_get_allocated_file_size
2088                         = raw_get_allocated_file_size,
2089
2090     .bdrv_detach_aio_context = raw_detach_aio_context,
2091     .bdrv_attach_aio_context = raw_attach_aio_context,
2092
2093     /* generic scsi device */
2094 #ifdef __linux__
2095     .bdrv_ioctl         = hdev_ioctl,
2096     .bdrv_aio_ioctl     = hdev_aio_ioctl,
2097 #endif
2098 };
2099
2100 #ifdef __linux__
2101 static void floppy_parse_filename(const char *filename, QDict *options,
2102                                   Error **errp)
2103 {
2104     /* The prefix is optional, just as for "file". */
2105     strstart(filename, "host_floppy:", &filename);
2106
2107     qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2108 }
2109
2110 static int floppy_open(BlockDriverState *bs, QDict *options, int flags,
2111                        Error **errp)
2112 {
2113     BDRVRawState *s = bs->opaque;
2114     Error *local_err = NULL;
2115     int ret;
2116
2117     s->type = FTYPE_FD;
2118
2119     /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
2120     ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2121     if (ret) {
2122         if (local_err) {
2123             error_propagate(errp, local_err);
2124         }
2125         return ret;
2126     }
2127
2128     /* close fd so that we can reopen it as needed */
2129     qemu_close(s->fd);
2130     s->fd = -1;
2131     s->fd_media_changed = 1;
2132
2133     return 0;
2134 }
2135
2136 static int floppy_probe_device(const char *filename)
2137 {
2138     int fd, ret;
2139     int prio = 0;
2140     struct floppy_struct fdparam;
2141     struct stat st;
2142
2143     if (strstart(filename, "/dev/fd", NULL) &&
2144         !strstart(filename, "/dev/fdset/", NULL)) {
2145         prio = 50;
2146     }
2147
2148     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2149     if (fd < 0) {
2150         goto out;
2151     }
2152     ret = fstat(fd, &st);
2153     if (ret == -1 || !S_ISBLK(st.st_mode)) {
2154         goto outc;
2155     }
2156
2157     /* Attempt to detect via a floppy specific ioctl */
2158     ret = ioctl(fd, FDGETPRM, &fdparam);
2159     if (ret >= 0)
2160         prio = 100;
2161
2162 outc:
2163     qemu_close(fd);
2164 out:
2165     return prio;
2166 }
2167
2168
2169 static int floppy_is_inserted(BlockDriverState *bs)
2170 {
2171     return fd_open(bs) >= 0;
2172 }
2173
2174 static int floppy_media_changed(BlockDriverState *bs)
2175 {
2176     BDRVRawState *s = bs->opaque;
2177     int ret;
2178
2179     /*
2180      * XXX: we do not have a true media changed indication.
2181      * It does not work if the floppy is changed without trying to read it.
2182      */
2183     fd_open(bs);
2184     ret = s->fd_media_changed;
2185     s->fd_media_changed = 0;
2186 #ifdef DEBUG_FLOPPY
2187     printf("Floppy changed=%d\n", ret);
2188 #endif
2189     return ret;
2190 }
2191
2192 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
2193 {
2194     BDRVRawState *s = bs->opaque;
2195     int fd;
2196
2197     if (s->fd >= 0) {
2198         qemu_close(s->fd);
2199         s->fd = -1;
2200     }
2201     fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
2202     if (fd >= 0) {
2203         if (ioctl(fd, FDEJECT, 0) < 0)
2204             perror("FDEJECT");
2205         qemu_close(fd);
2206     }
2207 }
2208
2209 static BlockDriver bdrv_host_floppy = {
2210     .format_name        = "host_floppy",
2211     .protocol_name      = "host_floppy",
2212     .instance_size      = sizeof(BDRVRawState),
2213     .bdrv_needs_filename = true,
2214     .bdrv_probe_device  = floppy_probe_device,
2215     .bdrv_parse_filename = floppy_parse_filename,
2216     .bdrv_file_open     = floppy_open,
2217     .bdrv_close         = raw_close,
2218     .bdrv_reopen_prepare = raw_reopen_prepare,
2219     .bdrv_reopen_commit  = raw_reopen_commit,
2220     .bdrv_reopen_abort   = raw_reopen_abort,
2221     .bdrv_create         = hdev_create,
2222     .create_opts         = &raw_create_opts,
2223
2224     .bdrv_aio_readv     = raw_aio_readv,
2225     .bdrv_aio_writev    = raw_aio_writev,
2226     .bdrv_aio_flush     = raw_aio_flush,
2227     .bdrv_refresh_limits = raw_refresh_limits,
2228     .bdrv_io_plug = raw_aio_plug,
2229     .bdrv_io_unplug = raw_aio_unplug,
2230     .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2231
2232     .bdrv_truncate      = raw_truncate,
2233     .bdrv_getlength      = raw_getlength,
2234     .has_variable_length = true,
2235     .bdrv_get_allocated_file_size
2236                         = raw_get_allocated_file_size,
2237
2238     .bdrv_detach_aio_context = raw_detach_aio_context,
2239     .bdrv_attach_aio_context = raw_attach_aio_context,
2240
2241     /* removable device support */
2242     .bdrv_is_inserted   = floppy_is_inserted,
2243     .bdrv_media_changed = floppy_media_changed,
2244     .bdrv_eject         = floppy_eject,
2245 };
2246 #endif
2247
2248 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2249 static void cdrom_parse_filename(const char *filename, QDict *options,
2250                                  Error **errp)
2251 {
2252     /* The prefix is optional, just as for "file". */
2253     strstart(filename, "host_cdrom:", &filename);
2254
2255     qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2256 }
2257 #endif
2258
2259 #ifdef __linux__
2260 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2261                       Error **errp)
2262 {
2263     BDRVRawState *s = bs->opaque;
2264     Error *local_err = NULL;
2265     int ret;
2266
2267     s->type = FTYPE_CD;
2268
2269     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2270     ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2271     if (local_err) {
2272         error_propagate(errp, local_err);
2273     }
2274     return ret;
2275 }
2276
2277 static int cdrom_probe_device(const char *filename)
2278 {
2279     int fd, ret;
2280     int prio = 0;
2281     struct stat st;
2282
2283     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2284     if (fd < 0) {
2285         goto out;
2286     }
2287     ret = fstat(fd, &st);
2288     if (ret == -1 || !S_ISBLK(st.st_mode)) {
2289         goto outc;
2290     }
2291
2292     /* Attempt to detect via a CDROM specific ioctl */
2293     ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2294     if (ret >= 0)
2295         prio = 100;
2296
2297 outc:
2298     qemu_close(fd);
2299 out:
2300     return prio;
2301 }
2302
2303 static int cdrom_is_inserted(BlockDriverState *bs)
2304 {
2305     BDRVRawState *s = bs->opaque;
2306     int ret;
2307
2308     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2309     if (ret == CDS_DISC_OK)
2310         return 1;
2311     return 0;
2312 }
2313
2314 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2315 {
2316     BDRVRawState *s = bs->opaque;
2317
2318     if (eject_flag) {
2319         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2320             perror("CDROMEJECT");
2321     } else {
2322         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2323             perror("CDROMEJECT");
2324     }
2325 }
2326
2327 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2328 {
2329     BDRVRawState *s = bs->opaque;
2330
2331     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2332         /*
2333          * Note: an error can happen if the distribution automatically
2334          * mounts the CD-ROM
2335          */
2336         /* perror("CDROM_LOCKDOOR"); */
2337     }
2338 }
2339
2340 static BlockDriver bdrv_host_cdrom = {
2341     .format_name        = "host_cdrom",
2342     .protocol_name      = "host_cdrom",
2343     .instance_size      = sizeof(BDRVRawState),
2344     .bdrv_needs_filename = true,
2345     .bdrv_probe_device  = cdrom_probe_device,
2346     .bdrv_parse_filename = cdrom_parse_filename,
2347     .bdrv_file_open     = cdrom_open,
2348     .bdrv_close         = raw_close,
2349     .bdrv_reopen_prepare = raw_reopen_prepare,
2350     .bdrv_reopen_commit  = raw_reopen_commit,
2351     .bdrv_reopen_abort   = raw_reopen_abort,
2352     .bdrv_create         = hdev_create,
2353     .create_opts         = &raw_create_opts,
2354
2355     .bdrv_aio_readv     = raw_aio_readv,
2356     .bdrv_aio_writev    = raw_aio_writev,
2357     .bdrv_aio_flush     = raw_aio_flush,
2358     .bdrv_refresh_limits = raw_refresh_limits,
2359     .bdrv_io_plug = raw_aio_plug,
2360     .bdrv_io_unplug = raw_aio_unplug,
2361     .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2362
2363     .bdrv_truncate      = raw_truncate,
2364     .bdrv_getlength      = raw_getlength,
2365     .has_variable_length = true,
2366     .bdrv_get_allocated_file_size
2367                         = raw_get_allocated_file_size,
2368
2369     .bdrv_detach_aio_context = raw_detach_aio_context,
2370     .bdrv_attach_aio_context = raw_attach_aio_context,
2371
2372     /* removable device support */
2373     .bdrv_is_inserted   = cdrom_is_inserted,
2374     .bdrv_eject         = cdrom_eject,
2375     .bdrv_lock_medium   = cdrom_lock_medium,
2376
2377     /* generic scsi device */
2378     .bdrv_ioctl         = hdev_ioctl,
2379     .bdrv_aio_ioctl     = hdev_aio_ioctl,
2380 };
2381 #endif /* __linux__ */
2382
2383 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2384 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2385                       Error **errp)
2386 {
2387     BDRVRawState *s = bs->opaque;
2388     Error *local_err = NULL;
2389     int ret;
2390
2391     s->type = FTYPE_CD;
2392
2393     ret = raw_open_common(bs, options, flags, 0, &local_err);
2394     if (ret) {
2395         if (local_err) {
2396             error_propagate(errp, local_err);
2397         }
2398         return ret;
2399     }
2400
2401     /* make sure the door isn't locked at this time */
2402     ioctl(s->fd, CDIOCALLOW);
2403     return 0;
2404 }
2405
2406 static int cdrom_probe_device(const char *filename)
2407 {
2408     if (strstart(filename, "/dev/cd", NULL) ||
2409             strstart(filename, "/dev/acd", NULL))
2410         return 100;
2411     return 0;
2412 }
2413
2414 static int cdrom_reopen(BlockDriverState *bs)
2415 {
2416     BDRVRawState *s = bs->opaque;
2417     int fd;
2418
2419     /*
2420      * Force reread of possibly changed/newly loaded disc,
2421      * FreeBSD seems to not notice sometimes...
2422      */
2423     if (s->fd >= 0)
2424         qemu_close(s->fd);
2425     fd = qemu_open(bs->filename, s->open_flags, 0644);
2426     if (fd < 0) {
2427         s->fd = -1;
2428         return -EIO;
2429     }
2430     s->fd = fd;
2431
2432     /* make sure the door isn't locked at this time */
2433     ioctl(s->fd, CDIOCALLOW);
2434     return 0;
2435 }
2436
2437 static int cdrom_is_inserted(BlockDriverState *bs)
2438 {
2439     return raw_getlength(bs) > 0;
2440 }
2441
2442 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2443 {
2444     BDRVRawState *s = bs->opaque;
2445
2446     if (s->fd < 0)
2447         return;
2448
2449     (void) ioctl(s->fd, CDIOCALLOW);
2450
2451     if (eject_flag) {
2452         if (ioctl(s->fd, CDIOCEJECT) < 0)
2453             perror("CDIOCEJECT");
2454     } else {
2455         if (ioctl(s->fd, CDIOCCLOSE) < 0)
2456             perror("CDIOCCLOSE");
2457     }
2458
2459     cdrom_reopen(bs);
2460 }
2461
2462 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2463 {
2464     BDRVRawState *s = bs->opaque;
2465
2466     if (s->fd < 0)
2467         return;
2468     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2469         /*
2470          * Note: an error can happen if the distribution automatically
2471          * mounts the CD-ROM
2472          */
2473         /* perror("CDROM_LOCKDOOR"); */
2474     }
2475 }
2476
2477 static BlockDriver bdrv_host_cdrom = {
2478     .format_name        = "host_cdrom",
2479     .protocol_name      = "host_cdrom",
2480     .instance_size      = sizeof(BDRVRawState),
2481     .bdrv_needs_filename = true,
2482     .bdrv_probe_device  = cdrom_probe_device,
2483     .bdrv_parse_filename = cdrom_parse_filename,
2484     .bdrv_file_open     = cdrom_open,
2485     .bdrv_close         = raw_close,
2486     .bdrv_reopen_prepare = raw_reopen_prepare,
2487     .bdrv_reopen_commit  = raw_reopen_commit,
2488     .bdrv_reopen_abort   = raw_reopen_abort,
2489     .bdrv_create        = hdev_create,
2490     .create_opts        = &raw_create_opts,
2491
2492     .bdrv_aio_readv     = raw_aio_readv,
2493     .bdrv_aio_writev    = raw_aio_writev,
2494     .bdrv_aio_flush     = raw_aio_flush,
2495     .bdrv_refresh_limits = raw_refresh_limits,
2496     .bdrv_io_plug = raw_aio_plug,
2497     .bdrv_io_unplug = raw_aio_unplug,
2498     .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2499
2500     .bdrv_truncate      = raw_truncate,
2501     .bdrv_getlength      = raw_getlength,
2502     .has_variable_length = true,
2503     .bdrv_get_allocated_file_size
2504                         = raw_get_allocated_file_size,
2505
2506     .bdrv_detach_aio_context = raw_detach_aio_context,
2507     .bdrv_attach_aio_context = raw_attach_aio_context,
2508
2509     /* removable device support */
2510     .bdrv_is_inserted   = cdrom_is_inserted,
2511     .bdrv_eject         = cdrom_eject,
2512     .bdrv_lock_medium   = cdrom_lock_medium,
2513 };
2514 #endif /* __FreeBSD__ */
2515
2516 static void bdrv_file_init(void)
2517 {
2518     /*
2519      * Register all the drivers.  Note that order is important, the driver
2520      * registered last will get probed first.
2521      */
2522     bdrv_register(&bdrv_file);
2523     bdrv_register(&bdrv_host_device);
2524 #ifdef __linux__
2525     bdrv_register(&bdrv_host_floppy);
2526     bdrv_register(&bdrv_host_cdrom);
2527 #endif
2528 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2529     bdrv_register(&bdrv_host_cdrom);
2530 #endif
2531 }
2532
2533 block_init(bdrv_file_init);