2 * Block driver for RAW files (posix)
4 * Copyright (c) 2006 Fabrice Bellard
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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
24 #include "qemu-common.h"
25 #include "qemu/error-report.h"
26 #include "qemu/timer.h"
28 #include "block/block_int.h"
29 #include "qemu/module.h"
31 #include "block/thread-pool.h"
34 #include "qapi/util.h"
35 #include "qapi/qmp/qstring.h"
37 #if defined(__APPLE__) && (__MACH__)
39 #include <sys/param.h>
40 #include <IOKit/IOKitLib.h>
41 #include <IOKit/IOBSD.h>
42 #include <IOKit/storage/IOMediaBSDClient.h>
43 #include <IOKit/storage/IOMedia.h>
44 #include <IOKit/storage/IOCDMedia.h>
45 //#include <IOKit/storage/IOCDTypes.h>
46 #include <CoreFoundation/CoreFoundation.h>
50 #define _POSIX_PTHREAD_SEMANTICS 1
54 #include <sys/types.h>
56 #include <sys/ioctl.h>
57 #include <sys/param.h>
58 #include <linux/cdrom.h>
61 #include <linux/hdreg.h>
67 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
70 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
71 #include <linux/falloc.h>
73 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
79 #include <sys/ioctl.h>
80 #include <sys/disklabel.h>
85 #include <sys/ioctl.h>
86 #include <sys/disklabel.h>
92 #include <sys/ioctl.h>
93 #include <sys/diskslice.h>
100 //#define DEBUG_BLOCK
103 # define DEBUG_BLOCK_PRINT 1
105 # define DEBUG_BLOCK_PRINT 0
107 #define DPRINTF(fmt, ...) \
109 if (DEBUG_BLOCK_PRINT) { \
110 printf(fmt, ## __VA_ARGS__); \
114 /* OS X does not have O_DSYNC */
117 #define O_DSYNC O_SYNC
118 #elif defined(O_FSYNC)
119 #define O_DSYNC O_FSYNC
123 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
125 #define O_DIRECT O_DSYNC
132 /* if the FD is not accessed during that time (in ns), we try to
133 reopen it to see if the disk has been changed */
134 #define FD_OPEN_TIMEOUT (1000000000)
136 #define MAX_BLOCKSIZE 4096
138 typedef struct BDRVRawState {
144 #if defined(__linux__)
145 /* linux floppy specific */
146 int64_t fd_open_time;
147 int64_t fd_error_time;
149 int fd_media_changed;
151 #ifdef CONFIG_LINUX_AIO
159 bool has_write_zeroes:1;
160 bool discard_zeroes:1;
162 bool needs_alignment;
165 typedef struct BDRVRawReopenState {
168 #ifdef CONFIG_LINUX_AIO
171 } BDRVRawReopenState;
173 static int fd_open(BlockDriverState *bs);
174 static int64_t raw_getlength(BlockDriverState *bs);
176 typedef struct RawPosixAIOData {
177 BlockDriverState *bs;
180 struct iovec *aio_iov;
185 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
190 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
191 static int cdrom_reopen(BlockDriverState *bs);
194 #if defined(__NetBSD__)
195 static int raw_normalize_devicepath(const char **filename)
197 static char namebuf[PATH_MAX];
198 const char *dp, *fname;
202 dp = strrchr(fname, '/');
203 if (lstat(fname, &sb) < 0) {
204 fprintf(stderr, "%s: stat failed: %s\n",
205 fname, strerror(errno));
209 if (!S_ISBLK(sb.st_mode)) {
214 snprintf(namebuf, PATH_MAX, "r%s", fname);
216 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
217 (int)(dp - fname), fname, dp + 1);
219 fprintf(stderr, "%s is a block device", fname);
221 fprintf(stderr, ", using %s\n", *filename);
226 static int raw_normalize_devicepath(const char **filename)
233 * Get logical block size via ioctl. On success store it in @sector_size_p.
235 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
237 unsigned int sector_size;
238 bool success = false;
242 /* Try a few ioctls to get the right size */
244 if (ioctl(fd, BLKSSZGET, §or_size) >= 0) {
245 *sector_size_p = sector_size;
249 #ifdef DKIOCGETBLOCKSIZE
250 if (ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) >= 0) {
251 *sector_size_p = sector_size;
255 #ifdef DIOCGSECTORSIZE
256 if (ioctl(fd, DIOCGSECTORSIZE, §or_size) >= 0) {
257 *sector_size_p = sector_size;
262 return success ? 0 : -errno;
266 * Get physical block size of @fd.
267 * On success, store it in @blk_size and return 0.
268 * On failure, return -errno.
270 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
273 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
282 /* Check if read is allowed with given memory buffer and length.
284 * This function is used to check O_DIRECT memory buffer and request alignment.
286 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
288 ssize_t ret = pread(fd, buf, len, 0);
295 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
296 * other errors (e.g. real I/O error), which could happen on a failed
297 * drive, since we only care about probing alignment.
299 if (errno != EINVAL) {
307 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
309 BDRVRawState *s = bs->opaque;
311 size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize());
313 /* For SCSI generic devices the alignment is not really used.
314 With buffered I/O, we don't have any restrictions. */
315 if (bdrv_is_sg(bs) || !s->needs_alignment) {
316 bs->request_alignment = 1;
321 bs->request_alignment = 0;
323 /* Let's try to use the logical blocksize for the alignment. */
324 if (probe_logical_blocksize(fd, &bs->request_alignment) < 0) {
325 bs->request_alignment = 0;
330 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
331 bs->request_alignment = da.d_miniosz;
332 /* The kernel returns wrong information for d_mem */
333 /* s->buf_align = da.d_mem; */
338 /* If we could not get the sizes so far, we can only guess them */
341 buf = qemu_memalign(max_align, 2 * max_align);
342 for (align = 512; align <= max_align; align <<= 1) {
343 if (raw_is_io_aligned(fd, buf + align, max_align)) {
344 s->buf_align = align;
351 if (!bs->request_alignment) {
353 buf = qemu_memalign(s->buf_align, max_align);
354 for (align = 512; align <= max_align; align <<= 1) {
355 if (raw_is_io_aligned(fd, buf, align)) {
356 bs->request_alignment = align;
363 if (!s->buf_align || !bs->request_alignment) {
364 error_setg(errp, "Could not find working O_DIRECT alignment. "
365 "Try cache.direct=off.");
369 static void raw_parse_flags(int bdrv_flags, int *open_flags)
371 assert(open_flags != NULL);
373 *open_flags |= O_BINARY;
374 *open_flags &= ~O_ACCMODE;
375 if (bdrv_flags & BDRV_O_RDWR) {
376 *open_flags |= O_RDWR;
378 *open_flags |= O_RDONLY;
381 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
382 * and O_DIRECT for no caching. */
383 if ((bdrv_flags & BDRV_O_NOCACHE)) {
384 *open_flags |= O_DIRECT;
388 static void raw_detach_aio_context(BlockDriverState *bs)
390 #ifdef CONFIG_LINUX_AIO
391 BDRVRawState *s = bs->opaque;
394 laio_detach_aio_context(s->aio_ctx, bdrv_get_aio_context(bs));
399 static void raw_attach_aio_context(BlockDriverState *bs,
400 AioContext *new_context)
402 #ifdef CONFIG_LINUX_AIO
403 BDRVRawState *s = bs->opaque;
406 laio_attach_aio_context(s->aio_ctx, new_context);
411 #ifdef CONFIG_LINUX_AIO
412 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
415 assert(aio_ctx != NULL);
416 assert(use_aio != NULL);
418 * Currently Linux do AIO only for files opened with O_DIRECT
419 * specified so check NOCACHE flag too
421 if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
422 (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
424 /* if non-NULL, laio_init() has already been run */
425 if (*aio_ctx == NULL) {
426 *aio_ctx = laio_init();
443 static void raw_parse_filename(const char *filename, QDict *options,
446 /* The filename does not have to be prefixed by the protocol name, since
447 * "file" is the default protocol; therefore, the return value of this
448 * function call can be ignored. */
449 strstart(filename, "file:", &filename);
451 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
454 static QemuOptsList raw_runtime_opts = {
456 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
460 .type = QEMU_OPT_STRING,
461 .help = "File name of the image",
463 { /* end of list */ }
467 static int raw_open_common(BlockDriverState *bs, QDict *options,
468 int bdrv_flags, int open_flags, Error **errp)
470 BDRVRawState *s = bs->opaque;
472 Error *local_err = NULL;
473 const char *filename = NULL;
477 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
478 qemu_opts_absorb_qdict(opts, options, &local_err);
480 error_propagate(errp, local_err);
485 filename = qemu_opt_get(opts, "filename");
487 ret = raw_normalize_devicepath(&filename);
489 error_setg_errno(errp, -ret, "Could not normalize device path");
493 s->open_flags = open_flags;
494 raw_parse_flags(bdrv_flags, &s->open_flags);
497 fd = qemu_open(filename, s->open_flags, 0644);
507 #ifdef CONFIG_LINUX_AIO
508 if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
511 error_setg_errno(errp, -ret, "Could not set AIO state");
514 if (!s->use_aio && (bdrv_flags & BDRV_O_NATIVE_AIO)) {
515 error_printf("WARNING: aio=native was specified for '%s', but "
516 "it requires cache.direct=on, which was not "
517 "specified. Falling back to aio=threads.\n"
518 " This will become an error condition in "
519 "future QEMU versions.\n",
524 s->has_discard = true;
525 s->has_write_zeroes = true;
526 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
527 s->needs_alignment = true;
530 if (fstat(s->fd, &st) < 0) {
532 error_setg_errno(errp, errno, "Could not stat file");
535 if (S_ISREG(st.st_mode)) {
536 s->discard_zeroes = true;
537 s->has_fallocate = true;
539 if (S_ISBLK(st.st_mode)) {
540 #ifdef BLKDISCARDZEROES
542 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
543 s->discard_zeroes = true;
547 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
548 * not rely on the contents of discarded blocks unless using O_DIRECT.
549 * Same for BLKZEROOUT.
551 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
552 s->discard_zeroes = false;
553 s->has_write_zeroes = false;
558 if (S_ISCHR(st.st_mode)) {
560 * The file is a char device (disk), which on FreeBSD isn't behind
561 * a pager, so force all requests to be aligned. This is needed
562 * so QEMU makes sure all IO operations on the device are aligned
563 * to sector size, or else FreeBSD will reject them with EINVAL.
565 s->needs_alignment = true;
570 if (platform_test_xfs_fd(s->fd)) {
575 raw_attach_aio_context(bs, bdrv_get_aio_context(bs));
579 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
586 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
589 BDRVRawState *s = bs->opaque;
590 Error *local_err = NULL;
593 s->type = FTYPE_FILE;
594 ret = raw_open_common(bs, options, flags, 0, &local_err);
596 error_propagate(errp, local_err);
601 static int raw_reopen_prepare(BDRVReopenState *state,
602 BlockReopenQueue *queue, Error **errp)
605 BDRVRawReopenState *raw_s;
607 Error *local_err = NULL;
609 assert(state != NULL);
610 assert(state->bs != NULL);
612 s = state->bs->opaque;
614 state->opaque = g_new0(BDRVRawReopenState, 1);
615 raw_s = state->opaque;
617 #ifdef CONFIG_LINUX_AIO
618 raw_s->use_aio = s->use_aio;
620 /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
621 * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
622 * won't override aio_ctx if aio_ctx is non-NULL */
623 if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
624 error_setg(errp, "Could not set AIO state");
629 if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
630 raw_s->open_flags |= O_NONBLOCK;
633 raw_parse_flags(state->flags, &raw_s->open_flags);
637 int fcntl_flags = O_APPEND | O_NONBLOCK;
639 fcntl_flags |= O_NOATIME;
643 /* Not all operating systems have O_ASYNC, and those that don't
644 * will not let us track the state into raw_s->open_flags (typically
645 * you achieve the same effect with an ioctl, for example I_SETSIG
646 * on Solaris). But we do not use O_ASYNC, so that's fine.
648 assert((s->open_flags & O_ASYNC) == 0);
651 if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
652 /* dup the original fd */
653 /* TODO: use qemu fcntl wrapper */
654 #ifdef F_DUPFD_CLOEXEC
655 raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
657 raw_s->fd = dup(s->fd);
658 if (raw_s->fd != -1) {
659 qemu_set_cloexec(raw_s->fd);
662 if (raw_s->fd >= 0) {
663 ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
665 qemu_close(raw_s->fd);
671 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
672 if (raw_s->fd == -1) {
673 assert(!(raw_s->open_flags & O_CREAT));
674 raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
675 if (raw_s->fd == -1) {
676 error_setg_errno(errp, errno, "Could not reopen file");
681 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
682 * alignment with the new fd. */
683 if (raw_s->fd != -1) {
684 raw_probe_alignment(state->bs, raw_s->fd, &local_err);
686 qemu_close(raw_s->fd);
688 error_propagate(errp, local_err);
696 static void raw_reopen_commit(BDRVReopenState *state)
698 BDRVRawReopenState *raw_s = state->opaque;
699 BDRVRawState *s = state->bs->opaque;
701 s->open_flags = raw_s->open_flags;
705 #ifdef CONFIG_LINUX_AIO
706 s->use_aio = raw_s->use_aio;
709 g_free(state->opaque);
710 state->opaque = NULL;
714 static void raw_reopen_abort(BDRVReopenState *state)
716 BDRVRawReopenState *raw_s = state->opaque;
718 /* nothing to do if NULL, we didn't get far enough */
723 if (raw_s->fd >= 0) {
724 qemu_close(raw_s->fd);
727 g_free(state->opaque);
728 state->opaque = NULL;
731 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
733 BDRVRawState *s = bs->opaque;
735 raw_probe_alignment(bs, s->fd, errp);
736 bs->bl.min_mem_alignment = s->buf_align;
737 bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
740 static int check_for_dasd(int fd)
743 struct dasd_information2_t info = {0};
745 return ioctl(fd, BIODASDINFO2, &info);
752 * Try to get @bs's logical and physical block size.
753 * On success, store them in @bsz and return zero.
754 * On failure, return negative errno.
756 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
758 BDRVRawState *s = bs->opaque;
761 /* If DASD, get blocksizes */
762 if (check_for_dasd(s->fd) < 0) {
765 ret = probe_logical_blocksize(s->fd, &bsz->log);
769 return probe_physical_blocksize(s->fd, &bsz->phys);
773 * Try to get @bs's geometry: cyls, heads, sectors.
774 * On success, store them in @geo and return 0.
775 * On failure return -errno.
776 * (Allows block driver to assign default geometry values that guest sees)
779 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
781 BDRVRawState *s = bs->opaque;
782 struct hd_geometry ioctl_geo = {0};
785 /* If DASD, get its geometry */
786 if (check_for_dasd(s->fd) < 0) {
789 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
792 /* HDIO_GETGEO may return success even though geo contains zeros
793 (e.g. certain multipath setups) */
794 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
797 /* Do not return a geometry for partition */
798 if (ioctl_geo.start != 0) {
801 geo->heads = ioctl_geo.heads;
802 geo->sectors = ioctl_geo.sectors;
803 if (!probe_physical_blocksize(s->fd, &blksize)) {
804 /* overwrite cyls: HDIO_GETGEO result is incorrect for big drives */
805 geo->cylinders = bdrv_nb_sectors(bs) / (blksize / BDRV_SECTOR_SIZE)
806 / (geo->heads * geo->sectors);
809 geo->cylinders = ioctl_geo.cylinders;
813 #else /* __linux__ */
814 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
820 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
824 ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
832 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
836 ret = qemu_fdatasync(aiocb->aio_fildes);
845 static bool preadv_present = true;
848 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
850 return preadv(fd, iov, nr_iov, offset);
854 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
856 return pwritev(fd, iov, nr_iov, offset);
861 static bool preadv_present = false;
864 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
870 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
877 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
882 if (aiocb->aio_type & QEMU_AIO_WRITE)
883 len = qemu_pwritev(aiocb->aio_fildes,
888 len = qemu_preadv(aiocb->aio_fildes,
892 } while (len == -1 && errno == EINTR);
901 * Read/writes the data to/from a given linear buffer.
903 * Returns the number of bytes handles or -errno in case of an error. Short
904 * reads are only returned if the end of the file is reached.
906 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
911 while (offset < aiocb->aio_nbytes) {
912 if (aiocb->aio_type & QEMU_AIO_WRITE) {
913 len = pwrite(aiocb->aio_fildes,
914 (const char *)buf + offset,
915 aiocb->aio_nbytes - offset,
916 aiocb->aio_offset + offset);
918 len = pread(aiocb->aio_fildes,
920 aiocb->aio_nbytes - offset,
921 aiocb->aio_offset + offset);
923 if (len == -1 && errno == EINTR) {
925 } else if (len == -1 && errno == EINVAL &&
926 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
927 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
929 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
930 * after a short read. Assume that O_DIRECT short reads only occur
931 * at EOF. Therefore this is a short read, not an I/O error.
934 } else if (len == -1) {
937 } else if (len == 0) {
946 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
951 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
953 * If there is just a single buffer, and it is properly aligned
954 * we can just use plain pread/pwrite without any problems.
956 if (aiocb->aio_niov == 1) {
957 return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
960 * We have more than one iovec, and all are properly aligned.
962 * Try preadv/pwritev first and fall back to linearizing the
963 * buffer if it's not supported.
965 if (preadv_present) {
966 nbytes = handle_aiocb_rw_vector(aiocb);
967 if (nbytes == aiocb->aio_nbytes ||
968 (nbytes < 0 && nbytes != -ENOSYS)) {
971 preadv_present = false;
975 * XXX(hch): short read/write. no easy way to handle the reminder
976 * using these interfaces. For now retry using plain
982 * Ok, we have to do it the hard way, copy all segments into
983 * a single aligned buffer.
985 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
990 if (aiocb->aio_type & QEMU_AIO_WRITE) {
994 for (i = 0; i < aiocb->aio_niov; ++i) {
995 memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
996 p += aiocb->aio_iov[i].iov_len;
998 assert(p - buf == aiocb->aio_nbytes);
1001 nbytes = handle_aiocb_rw_linear(aiocb, buf);
1002 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
1004 size_t count = aiocb->aio_nbytes, copy;
1007 for (i = 0; i < aiocb->aio_niov && count; ++i) {
1009 if (copy > aiocb->aio_iov[i].iov_len) {
1010 copy = aiocb->aio_iov[i].iov_len;
1012 memcpy(aiocb->aio_iov[i].iov_base, p, copy);
1013 assert(count >= copy);
1025 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
1027 struct xfs_flock64 fl;
1030 memset(&fl, 0, sizeof(fl));
1031 fl.l_whence = SEEK_SET;
1032 fl.l_start = offset;
1035 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
1037 DPRINTF("cannot write zero range (%s)\n", strerror(errno));
1044 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
1046 struct xfs_flock64 fl;
1049 memset(&fl, 0, sizeof(fl));
1050 fl.l_whence = SEEK_SET;
1051 fl.l_start = offset;
1054 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
1056 DPRINTF("cannot punch hole (%s)\n", strerror(errno));
1064 static int translate_err(int err)
1066 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1073 #ifdef CONFIG_FALLOCATE
1074 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1077 if (fallocate(fd, mode, offset, len) == 0) {
1080 } while (errno == EINTR);
1081 return translate_err(-errno);
1085 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1088 BDRVRawState *s = aiocb->bs->opaque;
1090 if (!s->has_write_zeroes) {
1096 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1097 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1100 } while (errno == EINTR);
1102 ret = translate_err(-errno);
1105 if (ret == -ENOTSUP) {
1106 s->has_write_zeroes = false;
1111 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
1113 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1114 BDRVRawState *s = aiocb->bs->opaque;
1117 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1118 return handle_aiocb_write_zeroes_block(aiocb);
1123 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
1127 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1128 if (s->has_write_zeroes) {
1129 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1130 aiocb->aio_offset, aiocb->aio_nbytes);
1131 if (ret == 0 || ret != -ENOTSUP) {
1134 s->has_write_zeroes = false;
1138 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1139 if (s->has_discard && s->has_fallocate) {
1140 int ret = do_fallocate(s->fd,
1141 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1142 aiocb->aio_offset, aiocb->aio_nbytes);
1144 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1145 if (ret == 0 || ret != -ENOTSUP) {
1148 s->has_fallocate = false;
1149 } else if (ret != -ENOTSUP) {
1152 s->has_discard = false;
1157 #ifdef CONFIG_FALLOCATE
1158 if (s->has_fallocate && aiocb->aio_offset >= bdrv_getlength(aiocb->bs)) {
1159 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1160 if (ret == 0 || ret != -ENOTSUP) {
1163 s->has_fallocate = false;
1170 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
1172 int ret = -EOPNOTSUPP;
1173 BDRVRawState *s = aiocb->bs->opaque;
1175 if (!s->has_discard) {
1179 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1182 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1183 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1186 } while (errno == EINTR);
1193 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1197 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1198 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1199 aiocb->aio_offset, aiocb->aio_nbytes);
1203 ret = translate_err(ret);
1204 if (ret == -ENOTSUP) {
1205 s->has_discard = false;
1210 static int aio_worker(void *arg)
1212 RawPosixAIOData *aiocb = arg;
1215 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
1217 ret = handle_aiocb_rw(aiocb);
1218 if (ret >= 0 && ret < aiocb->aio_nbytes) {
1219 iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
1220 0, aiocb->aio_nbytes - ret);
1222 ret = aiocb->aio_nbytes;
1224 if (ret == aiocb->aio_nbytes) {
1226 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1230 case QEMU_AIO_WRITE:
1231 ret = handle_aiocb_rw(aiocb);
1232 if (ret == aiocb->aio_nbytes) {
1234 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1238 case QEMU_AIO_FLUSH:
1239 ret = handle_aiocb_flush(aiocb);
1241 case QEMU_AIO_IOCTL:
1242 ret = handle_aiocb_ioctl(aiocb);
1244 case QEMU_AIO_DISCARD:
1245 ret = handle_aiocb_discard(aiocb);
1247 case QEMU_AIO_WRITE_ZEROES:
1248 ret = handle_aiocb_write_zeroes(aiocb);
1251 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1256 g_slice_free(RawPosixAIOData, aiocb);
1260 static int paio_submit_co(BlockDriverState *bs, int fd,
1261 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1264 RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
1268 acb->aio_type = type;
1269 acb->aio_fildes = fd;
1271 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1272 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1275 acb->aio_iov = qiov->iov;
1276 acb->aio_niov = qiov->niov;
1277 assert(qiov->size == acb->aio_nbytes);
1280 trace_paio_submit_co(sector_num, nb_sectors, type);
1281 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1282 return thread_pool_submit_co(pool, aio_worker, acb);
1285 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
1286 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1287 BlockCompletionFunc *cb, void *opaque, int type)
1289 RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
1293 acb->aio_type = type;
1294 acb->aio_fildes = fd;
1296 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1297 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1300 acb->aio_iov = qiov->iov;
1301 acb->aio_niov = qiov->niov;
1302 assert(qiov->size == acb->aio_nbytes);
1305 trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
1306 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1307 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1310 static BlockAIOCB *raw_aio_submit(BlockDriverState *bs,
1311 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1312 BlockCompletionFunc *cb, void *opaque, int type)
1314 BDRVRawState *s = bs->opaque;
1316 if (fd_open(bs) < 0)
1320 * Check if the underlying device requires requests to be aligned,
1321 * and if the request we are trying to submit is aligned or not.
1322 * If this is the case tell the low-level driver that it needs
1323 * to copy the buffer.
1325 if (s->needs_alignment) {
1326 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1327 type |= QEMU_AIO_MISALIGNED;
1328 #ifdef CONFIG_LINUX_AIO
1329 } else if (s->use_aio) {
1330 return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
1331 nb_sectors, cb, opaque, type);
1336 return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
1340 static void raw_aio_plug(BlockDriverState *bs)
1342 #ifdef CONFIG_LINUX_AIO
1343 BDRVRawState *s = bs->opaque;
1345 laio_io_plug(bs, s->aio_ctx);
1350 static void raw_aio_unplug(BlockDriverState *bs)
1352 #ifdef CONFIG_LINUX_AIO
1353 BDRVRawState *s = bs->opaque;
1355 laio_io_unplug(bs, s->aio_ctx, true);
1360 static void raw_aio_flush_io_queue(BlockDriverState *bs)
1362 #ifdef CONFIG_LINUX_AIO
1363 BDRVRawState *s = bs->opaque;
1365 laio_io_unplug(bs, s->aio_ctx, false);
1370 static BlockAIOCB *raw_aio_readv(BlockDriverState *bs,
1371 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1372 BlockCompletionFunc *cb, void *opaque)
1374 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1375 cb, opaque, QEMU_AIO_READ);
1378 static BlockAIOCB *raw_aio_writev(BlockDriverState *bs,
1379 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1380 BlockCompletionFunc *cb, void *opaque)
1382 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1383 cb, opaque, QEMU_AIO_WRITE);
1386 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
1387 BlockCompletionFunc *cb, void *opaque)
1389 BDRVRawState *s = bs->opaque;
1391 if (fd_open(bs) < 0)
1394 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1397 static void raw_close(BlockDriverState *bs)
1399 BDRVRawState *s = bs->opaque;
1401 raw_detach_aio_context(bs);
1403 #ifdef CONFIG_LINUX_AIO
1405 laio_cleanup(s->aio_ctx);
1414 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1416 BDRVRawState *s = bs->opaque;
1419 if (fstat(s->fd, &st)) {
1423 if (S_ISREG(st.st_mode)) {
1424 if (ftruncate(s->fd, offset) < 0) {
1427 } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1428 if (offset > raw_getlength(bs)) {
1439 static int64_t raw_getlength(BlockDriverState *bs)
1441 BDRVRawState *s = bs->opaque;
1447 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1448 struct disklabel dl;
1450 if (ioctl(fd, DIOCGDINFO, &dl))
1452 return (uint64_t)dl.d_secsize *
1453 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1457 #elif defined(__NetBSD__)
1458 static int64_t raw_getlength(BlockDriverState *bs)
1460 BDRVRawState *s = bs->opaque;
1466 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1467 struct dkwedge_info dkw;
1469 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1470 return dkw.dkw_size * 512;
1472 struct disklabel dl;
1474 if (ioctl(fd, DIOCGDINFO, &dl))
1476 return (uint64_t)dl.d_secsize *
1477 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1482 #elif defined(__sun__)
1483 static int64_t raw_getlength(BlockDriverState *bs)
1485 BDRVRawState *s = bs->opaque;
1486 struct dk_minfo minfo;
1496 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1498 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1500 return minfo.dki_lbsize * minfo.dki_capacity;
1504 * There are reports that lseek on some devices fails, but
1505 * irc discussion said that contingency on contingency was overkill.
1507 size = lseek(s->fd, 0, SEEK_END);
1513 #elif defined(CONFIG_BSD)
1514 static int64_t raw_getlength(BlockDriverState *bs)
1516 BDRVRawState *s = bs->opaque;
1520 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1529 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1532 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1533 #ifdef DIOCGMEDIASIZE
1534 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1535 #elif defined(DIOCGPART)
1538 if (ioctl(fd, DIOCGPART, &pi) == 0)
1539 size = pi.media_size;
1545 #if defined(__APPLE__) && defined(__MACH__)
1547 uint64_t sectors = 0;
1548 uint32_t sector_size = 0;
1550 if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0
1551 && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) {
1552 size = sectors * sector_size;
1554 size = lseek(fd, 0LL, SEEK_END);
1561 size = lseek(fd, 0LL, SEEK_END);
1566 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1569 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1570 if (size == 2048LL * (unsigned)-1)
1572 /* XXX no disc? maybe we need to reopen... */
1573 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1580 size = lseek(fd, 0, SEEK_END);
1588 static int64_t raw_getlength(BlockDriverState *bs)
1590 BDRVRawState *s = bs->opaque;
1599 size = lseek(s->fd, 0, SEEK_END);
1607 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1610 BDRVRawState *s = bs->opaque;
1612 if (fstat(s->fd, &st) < 0) {
1615 return (int64_t)st.st_blocks * 512;
1618 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
1622 int64_t total_size = 0;
1624 PreallocMode prealloc;
1626 Error *local_err = NULL;
1628 strstart(filename, "file:", &filename);
1630 /* Read out options */
1631 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1633 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
1634 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1635 prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
1636 PREALLOC_MODE_MAX, PREALLOC_MODE_OFF,
1640 error_propagate(errp, local_err);
1645 fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
1649 error_setg_errno(errp, -result, "Could not create file");
1655 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1656 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1657 * will be ignored since any failure of this operation should not
1658 * block the left work.
1661 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
1662 attr |= FS_NOCOW_FL;
1663 ioctl(fd, FS_IOC_SETFLAGS, &attr);
1668 if (ftruncate(fd, total_size) != 0) {
1670 error_setg_errno(errp, -result, "Could not resize file");
1675 #ifdef CONFIG_POSIX_FALLOCATE
1676 case PREALLOC_MODE_FALLOC:
1677 /* posix_fallocate() doesn't set errno. */
1678 result = -posix_fallocate(fd, 0, total_size);
1680 error_setg_errno(errp, -result,
1681 "Could not preallocate data for the new file");
1685 case PREALLOC_MODE_FULL:
1687 int64_t num = 0, left = total_size;
1688 buf = g_malloc0(65536);
1691 num = MIN(left, 65536);
1692 result = write(fd, buf, num);
1695 error_setg_errno(errp, -result,
1696 "Could not write to the new file");
1705 error_setg_errno(errp, -result,
1706 "Could not flush new file to disk");
1712 case PREALLOC_MODE_OFF:
1716 error_setg(errp, "Unsupported preallocation mode: %s",
1717 PreallocMode_lookup[prealloc]);
1722 if (qemu_close(fd) != 0 && result == 0) {
1724 error_setg_errno(errp, -result, "Could not close the new file");
1731 * Find allocation range in @bs around offset @start.
1732 * May change underlying file descriptor's file offset.
1733 * If @start is not in a hole, store @start in @data, and the
1734 * beginning of the next hole in @hole, and return 0.
1735 * If @start is in a non-trailing hole, store @start in @hole and the
1736 * beginning of the next non-hole in @data, and return 0.
1737 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1738 * If we can't find out, return a negative errno other than -ENXIO.
1740 static int find_allocation(BlockDriverState *bs, off_t start,
1741 off_t *data, off_t *hole)
1743 #if defined SEEK_HOLE && defined SEEK_DATA
1744 BDRVRawState *s = bs->opaque;
1749 * D1. offs == start: start is in data
1750 * D2. offs > start: start is in a hole, next data at offs
1751 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1752 * or start is beyond EOF
1753 * If the latter happens, the file has been truncated behind
1754 * our back since we opened it. All bets are off then.
1755 * Treating like a trailing hole is simplest.
1756 * D4. offs < 0, errno != ENXIO: we learned nothing
1758 offs = lseek(s->fd, start, SEEK_DATA);
1760 return -errno; /* D3 or D4 */
1762 assert(offs >= start);
1765 /* D2: in hole, next data at offs */
1771 /* D1: in data, end not yet known */
1775 * H1. offs == start: start is in a hole
1776 * If this happens here, a hole has been dug behind our back
1777 * since the previous lseek().
1778 * H2. offs > start: either start is in data, next hole at offs,
1779 * or start is in trailing hole, EOF at offs
1780 * Linux treats trailing holes like any other hole: offs ==
1781 * start. Solaris seeks to EOF instead: offs > start (blech).
1782 * If that happens here, a hole has been dug behind our back
1783 * since the previous lseek().
1784 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1785 * If this happens, the file has been truncated behind our
1786 * back since we opened it. Treat it like a trailing hole.
1787 * H4. offs < 0, errno != ENXIO: we learned nothing
1788 * Pretend we know nothing at all, i.e. "forget" about D1.
1790 offs = lseek(s->fd, start, SEEK_HOLE);
1792 return -errno; /* D1 and (H3 or H4) */
1794 assert(offs >= start);
1798 * D1 and H2: either in data, next hole at offs, or it was in
1799 * data but is now in a trailing hole. In the latter case,
1800 * all bets are off. Treating it as if it there was data all
1801 * the way to EOF is safe, so simply do that.
1816 * Returns the allocation status of the specified sectors.
1818 * If 'sector_num' is beyond the end of the disk image the return value is 0
1819 * and 'pnum' is set to 0.
1821 * 'pnum' is set to the number of sectors (including and immediately following
1822 * the specified sector) that are known to be in the same
1823 * allocated/unallocated state.
1825 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1826 * beyond the end of the disk image it will be clamped.
1828 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1830 int nb_sectors, int *pnum)
1832 off_t start, data = 0, hole = 0;
1841 start = sector_num * BDRV_SECTOR_SIZE;
1842 total_size = bdrv_getlength(bs);
1843 if (total_size < 0) {
1845 } else if (start >= total_size) {
1848 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1849 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1852 ret = find_allocation(bs, start, &data, &hole);
1853 if (ret == -ENXIO) {
1856 ret = BDRV_BLOCK_ZERO;
1857 } else if (ret < 0) {
1858 /* No info available, so pretend there are no holes */
1860 ret = BDRV_BLOCK_DATA;
1861 } else if (data == start) {
1862 /* On a data extent, compute sectors to the end of the extent,
1863 * possibly including a partial sector at EOF. */
1864 *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1865 ret = BDRV_BLOCK_DATA;
1867 /* On a hole, compute sectors to the beginning of the next extent. */
1868 assert(hole == start);
1869 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1870 ret = BDRV_BLOCK_ZERO;
1872 return ret | BDRV_BLOCK_OFFSET_VALID | start;
1875 static coroutine_fn BlockAIOCB *raw_aio_discard(BlockDriverState *bs,
1876 int64_t sector_num, int nb_sectors,
1877 BlockCompletionFunc *cb, void *opaque)
1879 BDRVRawState *s = bs->opaque;
1881 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1882 cb, opaque, QEMU_AIO_DISCARD);
1885 static int coroutine_fn raw_co_write_zeroes(
1886 BlockDriverState *bs, int64_t sector_num,
1887 int nb_sectors, BdrvRequestFlags flags)
1889 BDRVRawState *s = bs->opaque;
1891 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1892 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1893 QEMU_AIO_WRITE_ZEROES);
1894 } else if (s->discard_zeroes) {
1895 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1901 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1903 BDRVRawState *s = bs->opaque;
1905 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1906 bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1910 static QemuOptsList raw_create_opts = {
1911 .name = "raw-create-opts",
1912 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
1915 .name = BLOCK_OPT_SIZE,
1916 .type = QEMU_OPT_SIZE,
1917 .help = "Virtual disk size"
1920 .name = BLOCK_OPT_NOCOW,
1921 .type = QEMU_OPT_BOOL,
1922 .help = "Turn off copy-on-write (valid only on btrfs)"
1925 .name = BLOCK_OPT_PREALLOC,
1926 .type = QEMU_OPT_STRING,
1927 .help = "Preallocation mode (allowed values: off, falloc, full)"
1929 { /* end of list */ }
1933 BlockDriver bdrv_file = {
1934 .format_name = "file",
1935 .protocol_name = "file",
1936 .instance_size = sizeof(BDRVRawState),
1937 .bdrv_needs_filename = true,
1938 .bdrv_probe = NULL, /* no probe for protocols */
1939 .bdrv_parse_filename = raw_parse_filename,
1940 .bdrv_file_open = raw_open,
1941 .bdrv_reopen_prepare = raw_reopen_prepare,
1942 .bdrv_reopen_commit = raw_reopen_commit,
1943 .bdrv_reopen_abort = raw_reopen_abort,
1944 .bdrv_close = raw_close,
1945 .bdrv_create = raw_create,
1946 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1947 .bdrv_co_get_block_status = raw_co_get_block_status,
1948 .bdrv_co_write_zeroes = raw_co_write_zeroes,
1950 .bdrv_aio_readv = raw_aio_readv,
1951 .bdrv_aio_writev = raw_aio_writev,
1952 .bdrv_aio_flush = raw_aio_flush,
1953 .bdrv_aio_discard = raw_aio_discard,
1954 .bdrv_refresh_limits = raw_refresh_limits,
1955 .bdrv_io_plug = raw_aio_plug,
1956 .bdrv_io_unplug = raw_aio_unplug,
1957 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
1959 .bdrv_truncate = raw_truncate,
1960 .bdrv_getlength = raw_getlength,
1961 .bdrv_get_info = raw_get_info,
1962 .bdrv_get_allocated_file_size
1963 = raw_get_allocated_file_size,
1965 .bdrv_detach_aio_context = raw_detach_aio_context,
1966 .bdrv_attach_aio_context = raw_attach_aio_context,
1968 .create_opts = &raw_create_opts,
1971 /***********************************************/
1974 #if defined(__APPLE__) && defined(__MACH__)
1975 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1976 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1978 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1980 kern_return_t kernResult;
1981 mach_port_t masterPort;
1982 CFMutableDictionaryRef classesToMatch;
1984 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1985 if ( KERN_SUCCESS != kernResult ) {
1986 printf( "IOMasterPort returned %d\n", kernResult );
1989 classesToMatch = IOServiceMatching( kIOCDMediaClass );
1990 if ( classesToMatch == NULL ) {
1991 printf( "IOServiceMatching returned a NULL dictionary.\n" );
1993 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1995 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1996 if ( KERN_SUCCESS != kernResult )
1998 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
2004 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
2006 io_object_t nextMedia;
2007 kern_return_t kernResult = KERN_FAILURE;
2009 nextMedia = IOIteratorNext( mediaIterator );
2012 CFTypeRef bsdPathAsCFString;
2013 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
2014 if ( bsdPathAsCFString ) {
2015 size_t devPathLength;
2016 strcpy( bsdPath, _PATH_DEV );
2017 strcat( bsdPath, "r" );
2018 devPathLength = strlen( bsdPath );
2019 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
2020 kernResult = KERN_SUCCESS;
2022 CFRelease( bsdPathAsCFString );
2024 IOObjectRelease( nextMedia );
2032 static int hdev_probe_device(const char *filename)
2036 /* allow a dedicated CD-ROM driver to match with a higher priority */
2037 if (strstart(filename, "/dev/cdrom", NULL))
2040 if (stat(filename, &st) >= 0 &&
2041 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
2048 static int check_hdev_writable(BDRVRawState *s)
2050 #if defined(BLKROGET)
2051 /* Linux block devices can be configured "read-only" using blockdev(8).
2052 * This is independent of device node permissions and therefore open(2)
2053 * with O_RDWR succeeds. Actual writes fail with EPERM.
2055 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2056 * check for read-only block devices so that Linux block devices behave
2062 if (fstat(s->fd, &st)) {
2066 if (!S_ISBLK(st.st_mode)) {
2070 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
2077 #endif /* defined(BLKROGET) */
2081 static void hdev_parse_filename(const char *filename, QDict *options,
2084 /* The prefix is optional, just as for "file". */
2085 strstart(filename, "host_device:", &filename);
2087 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2090 static bool hdev_is_sg(BlockDriverState *bs)
2093 #if defined(__linux__)
2096 struct sg_scsi_id scsiid;
2099 if (stat(bs->filename, &st) >= 0 && S_ISCHR(st.st_mode) &&
2100 !bdrv_ioctl(bs, SG_GET_VERSION_NUM, &sg_version) &&
2101 !bdrv_ioctl(bs, SG_GET_SCSI_ID, &scsiid)) {
2102 DPRINTF("SG device found: type=%d, version=%d\n",
2103 scsiid.scsi_type, sg_version);
2112 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
2115 BDRVRawState *s = bs->opaque;
2116 Error *local_err = NULL;
2119 #if defined(__APPLE__) && defined(__MACH__)
2120 const char *filename = qdict_get_str(options, "filename");
2122 if (strstart(filename, "/dev/cdrom", NULL)) {
2123 kern_return_t kernResult;
2124 io_iterator_t mediaIterator;
2125 char bsdPath[ MAXPATHLEN ];
2128 kernResult = FindEjectableCDMedia( &mediaIterator );
2129 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
2131 if ( bsdPath[ 0 ] != '\0' ) {
2132 strcat(bsdPath,"s0");
2133 /* some CDs don't have a partition 0 */
2134 fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
2136 bsdPath[strlen(bsdPath)-1] = '1';
2141 qdict_put(options, "filename", qstring_from_str(filename));
2144 if ( mediaIterator )
2145 IOObjectRelease( mediaIterator );
2149 s->type = FTYPE_FILE;
2151 ret = raw_open_common(bs, options, flags, 0, &local_err);
2154 error_propagate(errp, local_err);
2159 /* Since this does ioctl the device must be already opened */
2160 bs->sg = hdev_is_sg(bs);
2162 if (flags & BDRV_O_RDWR) {
2163 ret = check_hdev_writable(s);
2166 error_setg_errno(errp, -ret, "The device is not writable");
2174 #if defined(__linux__)
2175 /* Note: we do not have a reliable method to detect if the floppy is
2176 present. The current method is to try to open the floppy at every
2177 I/O and to keep it opened during a few hundreds of ms. */
2178 static int fd_open(BlockDriverState *bs)
2180 BDRVRawState *s = bs->opaque;
2181 int last_media_present;
2183 if (s->type != FTYPE_FD)
2185 last_media_present = (s->fd >= 0);
2187 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
2190 DPRINTF("Floppy closed\n");
2193 if (s->fd_got_error &&
2194 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
2195 DPRINTF("No floppy (open delayed)\n");
2198 s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
2200 s->fd_error_time = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2201 s->fd_got_error = 1;
2202 if (last_media_present)
2203 s->fd_media_changed = 1;
2204 DPRINTF("No floppy\n");
2207 DPRINTF("Floppy opened\n");
2209 if (!last_media_present)
2210 s->fd_media_changed = 1;
2211 s->fd_open_time = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2212 s->fd_got_error = 0;
2216 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2218 BDRVRawState *s = bs->opaque;
2220 return ioctl(s->fd, req, buf);
2223 static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
2224 unsigned long int req, void *buf,
2225 BlockCompletionFunc *cb, void *opaque)
2227 BDRVRawState *s = bs->opaque;
2228 RawPosixAIOData *acb;
2231 if (fd_open(bs) < 0)
2234 acb = g_slice_new(RawPosixAIOData);
2236 acb->aio_type = QEMU_AIO_IOCTL;
2237 acb->aio_fildes = s->fd;
2238 acb->aio_offset = 0;
2239 acb->aio_ioctl_buf = buf;
2240 acb->aio_ioctl_cmd = req;
2241 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
2242 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
2245 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2246 static int fd_open(BlockDriverState *bs)
2248 BDRVRawState *s = bs->opaque;
2250 /* this is just to ensure s->fd is sane (its called by io ops) */
2255 #else /* !linux && !FreeBSD */
2257 static int fd_open(BlockDriverState *bs)
2262 #endif /* !linux && !FreeBSD */
2264 static coroutine_fn BlockAIOCB *hdev_aio_discard(BlockDriverState *bs,
2265 int64_t sector_num, int nb_sectors,
2266 BlockCompletionFunc *cb, void *opaque)
2268 BDRVRawState *s = bs->opaque;
2270 if (fd_open(bs) < 0) {
2273 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
2274 cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2277 static coroutine_fn int hdev_co_write_zeroes(BlockDriverState *bs,
2278 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2280 BDRVRawState *s = bs->opaque;
2287 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2288 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2289 QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2290 } else if (s->discard_zeroes) {
2291 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2292 QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2297 static int hdev_create(const char *filename, QemuOpts *opts,
2302 struct stat stat_buf;
2303 int64_t total_size = 0;
2306 /* This function is used by all three protocol block drivers and therefore
2307 * any of these three prefixes may be given.
2308 * The return value has to be stored somewhere, otherwise this is an error
2309 * due to -Werror=unused-value. */
2311 strstart(filename, "host_device:", &filename) ||
2312 strstart(filename, "host_cdrom:" , &filename) ||
2313 strstart(filename, "host_floppy:", &filename);
2317 /* Read out options */
2318 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2321 fd = qemu_open(filename, O_WRONLY | O_BINARY);
2324 error_setg_errno(errp, -ret, "Could not open device");
2328 if (fstat(fd, &stat_buf) < 0) {
2330 error_setg_errno(errp, -ret, "Could not stat device");
2331 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2333 "The given file is neither a block nor a character device");
2335 } else if (lseek(fd, 0, SEEK_END) < total_size) {
2336 error_setg(errp, "Device is too small");
2344 static BlockDriver bdrv_host_device = {
2345 .format_name = "host_device",
2346 .protocol_name = "host_device",
2347 .instance_size = sizeof(BDRVRawState),
2348 .bdrv_needs_filename = true,
2349 .bdrv_probe_device = hdev_probe_device,
2350 .bdrv_parse_filename = hdev_parse_filename,
2351 .bdrv_file_open = hdev_open,
2352 .bdrv_close = raw_close,
2353 .bdrv_reopen_prepare = raw_reopen_prepare,
2354 .bdrv_reopen_commit = raw_reopen_commit,
2355 .bdrv_reopen_abort = raw_reopen_abort,
2356 .bdrv_create = hdev_create,
2357 .create_opts = &raw_create_opts,
2358 .bdrv_co_write_zeroes = hdev_co_write_zeroes,
2360 .bdrv_aio_readv = raw_aio_readv,
2361 .bdrv_aio_writev = raw_aio_writev,
2362 .bdrv_aio_flush = raw_aio_flush,
2363 .bdrv_aio_discard = hdev_aio_discard,
2364 .bdrv_refresh_limits = raw_refresh_limits,
2365 .bdrv_io_plug = raw_aio_plug,
2366 .bdrv_io_unplug = raw_aio_unplug,
2367 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2369 .bdrv_truncate = raw_truncate,
2370 .bdrv_getlength = raw_getlength,
2371 .bdrv_get_info = raw_get_info,
2372 .bdrv_get_allocated_file_size
2373 = raw_get_allocated_file_size,
2374 .bdrv_probe_blocksizes = hdev_probe_blocksizes,
2375 .bdrv_probe_geometry = hdev_probe_geometry,
2377 .bdrv_detach_aio_context = raw_detach_aio_context,
2378 .bdrv_attach_aio_context = raw_attach_aio_context,
2380 /* generic scsi device */
2382 .bdrv_ioctl = hdev_ioctl,
2383 .bdrv_aio_ioctl = hdev_aio_ioctl,
2388 static void floppy_parse_filename(const char *filename, QDict *options,
2391 /* The prefix is optional, just as for "file". */
2392 strstart(filename, "host_floppy:", &filename);
2394 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2397 static int floppy_open(BlockDriverState *bs, QDict *options, int flags,
2400 BDRVRawState *s = bs->opaque;
2401 Error *local_err = NULL;
2406 /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
2407 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2410 error_propagate(errp, local_err);
2415 /* close fd so that we can reopen it as needed */
2418 s->fd_media_changed = 1;
2420 error_report("Host floppy pass-through is deprecated");
2421 error_printf("Support for it will be removed in a future release.\n");
2425 static int floppy_probe_device(const char *filename)
2429 struct floppy_struct fdparam;
2432 if (strstart(filename, "/dev/fd", NULL) &&
2433 !strstart(filename, "/dev/fdset/", NULL) &&
2434 !strstart(filename, "/dev/fd/", NULL)) {
2438 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2442 ret = fstat(fd, &st);
2443 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2447 /* Attempt to detect via a floppy specific ioctl */
2448 ret = ioctl(fd, FDGETPRM, &fdparam);
2459 static int floppy_is_inserted(BlockDriverState *bs)
2461 return fd_open(bs) >= 0;
2464 static int floppy_media_changed(BlockDriverState *bs)
2466 BDRVRawState *s = bs->opaque;
2470 * XXX: we do not have a true media changed indication.
2471 * It does not work if the floppy is changed without trying to read it.
2474 ret = s->fd_media_changed;
2475 s->fd_media_changed = 0;
2476 DPRINTF("Floppy changed=%d\n", ret);
2480 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
2482 BDRVRawState *s = bs->opaque;
2489 fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
2491 if (ioctl(fd, FDEJECT, 0) < 0)
2497 static BlockDriver bdrv_host_floppy = {
2498 .format_name = "host_floppy",
2499 .protocol_name = "host_floppy",
2500 .instance_size = sizeof(BDRVRawState),
2501 .bdrv_needs_filename = true,
2502 .bdrv_probe_device = floppy_probe_device,
2503 .bdrv_parse_filename = floppy_parse_filename,
2504 .bdrv_file_open = floppy_open,
2505 .bdrv_close = raw_close,
2506 .bdrv_reopen_prepare = raw_reopen_prepare,
2507 .bdrv_reopen_commit = raw_reopen_commit,
2508 .bdrv_reopen_abort = raw_reopen_abort,
2509 .bdrv_create = hdev_create,
2510 .create_opts = &raw_create_opts,
2512 .bdrv_aio_readv = raw_aio_readv,
2513 .bdrv_aio_writev = raw_aio_writev,
2514 .bdrv_aio_flush = raw_aio_flush,
2515 .bdrv_refresh_limits = raw_refresh_limits,
2516 .bdrv_io_plug = raw_aio_plug,
2517 .bdrv_io_unplug = raw_aio_unplug,
2518 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2520 .bdrv_truncate = raw_truncate,
2521 .bdrv_getlength = raw_getlength,
2522 .has_variable_length = true,
2523 .bdrv_get_allocated_file_size
2524 = raw_get_allocated_file_size,
2526 .bdrv_detach_aio_context = raw_detach_aio_context,
2527 .bdrv_attach_aio_context = raw_attach_aio_context,
2529 /* removable device support */
2530 .bdrv_is_inserted = floppy_is_inserted,
2531 .bdrv_media_changed = floppy_media_changed,
2532 .bdrv_eject = floppy_eject,
2536 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2537 static void cdrom_parse_filename(const char *filename, QDict *options,
2540 /* The prefix is optional, just as for "file". */
2541 strstart(filename, "host_cdrom:", &filename);
2543 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2548 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2551 BDRVRawState *s = bs->opaque;
2552 Error *local_err = NULL;
2557 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2558 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2560 error_propagate(errp, local_err);
2565 static int cdrom_probe_device(const char *filename)
2571 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2575 ret = fstat(fd, &st);
2576 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2580 /* Attempt to detect via a CDROM specific ioctl */
2581 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2591 static int cdrom_is_inserted(BlockDriverState *bs)
2593 BDRVRawState *s = bs->opaque;
2596 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2597 if (ret == CDS_DISC_OK)
2602 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2604 BDRVRawState *s = bs->opaque;
2607 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2608 perror("CDROMEJECT");
2610 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2611 perror("CDROMEJECT");
2615 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2617 BDRVRawState *s = bs->opaque;
2619 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2621 * Note: an error can happen if the distribution automatically
2624 /* perror("CDROM_LOCKDOOR"); */
2628 static BlockDriver bdrv_host_cdrom = {
2629 .format_name = "host_cdrom",
2630 .protocol_name = "host_cdrom",
2631 .instance_size = sizeof(BDRVRawState),
2632 .bdrv_needs_filename = true,
2633 .bdrv_probe_device = cdrom_probe_device,
2634 .bdrv_parse_filename = cdrom_parse_filename,
2635 .bdrv_file_open = cdrom_open,
2636 .bdrv_close = raw_close,
2637 .bdrv_reopen_prepare = raw_reopen_prepare,
2638 .bdrv_reopen_commit = raw_reopen_commit,
2639 .bdrv_reopen_abort = raw_reopen_abort,
2640 .bdrv_create = hdev_create,
2641 .create_opts = &raw_create_opts,
2643 .bdrv_aio_readv = raw_aio_readv,
2644 .bdrv_aio_writev = raw_aio_writev,
2645 .bdrv_aio_flush = raw_aio_flush,
2646 .bdrv_refresh_limits = raw_refresh_limits,
2647 .bdrv_io_plug = raw_aio_plug,
2648 .bdrv_io_unplug = raw_aio_unplug,
2649 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2651 .bdrv_truncate = raw_truncate,
2652 .bdrv_getlength = raw_getlength,
2653 .has_variable_length = true,
2654 .bdrv_get_allocated_file_size
2655 = raw_get_allocated_file_size,
2657 .bdrv_detach_aio_context = raw_detach_aio_context,
2658 .bdrv_attach_aio_context = raw_attach_aio_context,
2660 /* removable device support */
2661 .bdrv_is_inserted = cdrom_is_inserted,
2662 .bdrv_eject = cdrom_eject,
2663 .bdrv_lock_medium = cdrom_lock_medium,
2665 /* generic scsi device */
2666 .bdrv_ioctl = hdev_ioctl,
2667 .bdrv_aio_ioctl = hdev_aio_ioctl,
2669 #endif /* __linux__ */
2671 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2672 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2675 BDRVRawState *s = bs->opaque;
2676 Error *local_err = NULL;
2681 ret = raw_open_common(bs, options, flags, 0, &local_err);
2684 error_propagate(errp, local_err);
2689 /* make sure the door isn't locked at this time */
2690 ioctl(s->fd, CDIOCALLOW);
2694 static int cdrom_probe_device(const char *filename)
2696 if (strstart(filename, "/dev/cd", NULL) ||
2697 strstart(filename, "/dev/acd", NULL))
2702 static int cdrom_reopen(BlockDriverState *bs)
2704 BDRVRawState *s = bs->opaque;
2708 * Force reread of possibly changed/newly loaded disc,
2709 * FreeBSD seems to not notice sometimes...
2713 fd = qemu_open(bs->filename, s->open_flags, 0644);
2720 /* make sure the door isn't locked at this time */
2721 ioctl(s->fd, CDIOCALLOW);
2725 static int cdrom_is_inserted(BlockDriverState *bs)
2727 return raw_getlength(bs) > 0;
2730 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2732 BDRVRawState *s = bs->opaque;
2737 (void) ioctl(s->fd, CDIOCALLOW);
2740 if (ioctl(s->fd, CDIOCEJECT) < 0)
2741 perror("CDIOCEJECT");
2743 if (ioctl(s->fd, CDIOCCLOSE) < 0)
2744 perror("CDIOCCLOSE");
2750 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2752 BDRVRawState *s = bs->opaque;
2756 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2758 * Note: an error can happen if the distribution automatically
2761 /* perror("CDROM_LOCKDOOR"); */
2765 static BlockDriver bdrv_host_cdrom = {
2766 .format_name = "host_cdrom",
2767 .protocol_name = "host_cdrom",
2768 .instance_size = sizeof(BDRVRawState),
2769 .bdrv_needs_filename = true,
2770 .bdrv_probe_device = cdrom_probe_device,
2771 .bdrv_parse_filename = cdrom_parse_filename,
2772 .bdrv_file_open = cdrom_open,
2773 .bdrv_close = raw_close,
2774 .bdrv_reopen_prepare = raw_reopen_prepare,
2775 .bdrv_reopen_commit = raw_reopen_commit,
2776 .bdrv_reopen_abort = raw_reopen_abort,
2777 .bdrv_create = hdev_create,
2778 .create_opts = &raw_create_opts,
2780 .bdrv_aio_readv = raw_aio_readv,
2781 .bdrv_aio_writev = raw_aio_writev,
2782 .bdrv_aio_flush = raw_aio_flush,
2783 .bdrv_refresh_limits = raw_refresh_limits,
2784 .bdrv_io_plug = raw_aio_plug,
2785 .bdrv_io_unplug = raw_aio_unplug,
2786 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2788 .bdrv_truncate = raw_truncate,
2789 .bdrv_getlength = raw_getlength,
2790 .has_variable_length = true,
2791 .bdrv_get_allocated_file_size
2792 = raw_get_allocated_file_size,
2794 .bdrv_detach_aio_context = raw_detach_aio_context,
2795 .bdrv_attach_aio_context = raw_attach_aio_context,
2797 /* removable device support */
2798 .bdrv_is_inserted = cdrom_is_inserted,
2799 .bdrv_eject = cdrom_eject,
2800 .bdrv_lock_medium = cdrom_lock_medium,
2802 #endif /* __FreeBSD__ */
2804 static void bdrv_file_init(void)
2807 * Register all the drivers. Note that order is important, the driver
2808 * registered last will get probed first.
2810 bdrv_register(&bdrv_file);
2811 bdrv_register(&bdrv_host_device);
2813 bdrv_register(&bdrv_host_floppy);
2814 bdrv_register(&bdrv_host_cdrom);
2816 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2817 bdrv_register(&bdrv_host_cdrom);
2821 block_init(bdrv_file_init);