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
131 #define MAX_BLOCKSIZE 4096
133 typedef struct BDRVRawState {
139 #ifdef CONFIG_LINUX_AIO
147 bool has_write_zeroes:1;
148 bool discard_zeroes:1;
150 bool needs_alignment;
153 typedef struct BDRVRawReopenState {
156 #ifdef CONFIG_LINUX_AIO
159 } BDRVRawReopenState;
161 static int fd_open(BlockDriverState *bs);
162 static int64_t raw_getlength(BlockDriverState *bs);
164 typedef struct RawPosixAIOData {
165 BlockDriverState *bs;
168 struct iovec *aio_iov;
173 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
178 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
179 static int cdrom_reopen(BlockDriverState *bs);
182 #if defined(__NetBSD__)
183 static int raw_normalize_devicepath(const char **filename)
185 static char namebuf[PATH_MAX];
186 const char *dp, *fname;
190 dp = strrchr(fname, '/');
191 if (lstat(fname, &sb) < 0) {
192 fprintf(stderr, "%s: stat failed: %s\n",
193 fname, strerror(errno));
197 if (!S_ISBLK(sb.st_mode)) {
202 snprintf(namebuf, PATH_MAX, "r%s", fname);
204 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
205 (int)(dp - fname), fname, dp + 1);
207 fprintf(stderr, "%s is a block device", fname);
209 fprintf(stderr, ", using %s\n", *filename);
214 static int raw_normalize_devicepath(const char **filename)
221 * Get logical block size via ioctl. On success store it in @sector_size_p.
223 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
225 unsigned int sector_size;
226 bool success = false;
230 /* Try a few ioctls to get the right size */
232 if (ioctl(fd, BLKSSZGET, §or_size) >= 0) {
233 *sector_size_p = sector_size;
237 #ifdef DKIOCGETBLOCKSIZE
238 if (ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) >= 0) {
239 *sector_size_p = sector_size;
243 #ifdef DIOCGSECTORSIZE
244 if (ioctl(fd, DIOCGSECTORSIZE, §or_size) >= 0) {
245 *sector_size_p = sector_size;
250 return success ? 0 : -errno;
254 * Get physical block size of @fd.
255 * On success, store it in @blk_size and return 0.
256 * On failure, return -errno.
258 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
261 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
270 /* Check if read is allowed with given memory buffer and length.
272 * This function is used to check O_DIRECT memory buffer and request alignment.
274 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
276 ssize_t ret = pread(fd, buf, len, 0);
283 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
284 * other errors (e.g. real I/O error), which could happen on a failed
285 * drive, since we only care about probing alignment.
287 if (errno != EINVAL) {
295 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
297 BDRVRawState *s = bs->opaque;
299 size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize());
301 /* For SCSI generic devices the alignment is not really used.
302 With buffered I/O, we don't have any restrictions. */
303 if (bdrv_is_sg(bs) || !s->needs_alignment) {
304 bs->request_alignment = 1;
309 bs->request_alignment = 0;
311 /* Let's try to use the logical blocksize for the alignment. */
312 if (probe_logical_blocksize(fd, &bs->request_alignment) < 0) {
313 bs->request_alignment = 0;
318 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
319 bs->request_alignment = da.d_miniosz;
320 /* The kernel returns wrong information for d_mem */
321 /* s->buf_align = da.d_mem; */
326 /* If we could not get the sizes so far, we can only guess them */
329 buf = qemu_memalign(max_align, 2 * max_align);
330 for (align = 512; align <= max_align; align <<= 1) {
331 if (raw_is_io_aligned(fd, buf + align, max_align)) {
332 s->buf_align = align;
339 if (!bs->request_alignment) {
341 buf = qemu_memalign(s->buf_align, max_align);
342 for (align = 512; align <= max_align; align <<= 1) {
343 if (raw_is_io_aligned(fd, buf, align)) {
344 bs->request_alignment = align;
351 if (!s->buf_align || !bs->request_alignment) {
352 error_setg(errp, "Could not find working O_DIRECT alignment. "
353 "Try cache.direct=off.");
357 static void raw_parse_flags(int bdrv_flags, int *open_flags)
359 assert(open_flags != NULL);
361 *open_flags |= O_BINARY;
362 *open_flags &= ~O_ACCMODE;
363 if (bdrv_flags & BDRV_O_RDWR) {
364 *open_flags |= O_RDWR;
366 *open_flags |= O_RDONLY;
369 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
370 * and O_DIRECT for no caching. */
371 if ((bdrv_flags & BDRV_O_NOCACHE)) {
372 *open_flags |= O_DIRECT;
376 static void raw_detach_aio_context(BlockDriverState *bs)
378 #ifdef CONFIG_LINUX_AIO
379 BDRVRawState *s = bs->opaque;
382 laio_detach_aio_context(s->aio_ctx, bdrv_get_aio_context(bs));
387 static void raw_attach_aio_context(BlockDriverState *bs,
388 AioContext *new_context)
390 #ifdef CONFIG_LINUX_AIO
391 BDRVRawState *s = bs->opaque;
394 laio_attach_aio_context(s->aio_ctx, new_context);
399 #ifdef CONFIG_LINUX_AIO
400 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
403 assert(aio_ctx != NULL);
404 assert(use_aio != NULL);
406 * Currently Linux do AIO only for files opened with O_DIRECT
407 * specified so check NOCACHE flag too
409 if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
410 (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
412 /* if non-NULL, laio_init() has already been run */
413 if (*aio_ctx == NULL) {
414 *aio_ctx = laio_init();
431 static void raw_parse_filename(const char *filename, QDict *options,
434 /* The filename does not have to be prefixed by the protocol name, since
435 * "file" is the default protocol; therefore, the return value of this
436 * function call can be ignored. */
437 strstart(filename, "file:", &filename);
439 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
442 static QemuOptsList raw_runtime_opts = {
444 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
448 .type = QEMU_OPT_STRING,
449 .help = "File name of the image",
451 { /* end of list */ }
455 static int raw_open_common(BlockDriverState *bs, QDict *options,
456 int bdrv_flags, int open_flags, Error **errp)
458 BDRVRawState *s = bs->opaque;
460 Error *local_err = NULL;
461 const char *filename = NULL;
465 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
466 qemu_opts_absorb_qdict(opts, options, &local_err);
468 error_propagate(errp, local_err);
473 filename = qemu_opt_get(opts, "filename");
475 ret = raw_normalize_devicepath(&filename);
477 error_setg_errno(errp, -ret, "Could not normalize device path");
481 s->open_flags = open_flags;
482 raw_parse_flags(bdrv_flags, &s->open_flags);
485 fd = qemu_open(filename, s->open_flags, 0644);
495 #ifdef CONFIG_LINUX_AIO
496 if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
499 error_setg_errno(errp, -ret, "Could not set AIO state");
502 if (!s->use_aio && (bdrv_flags & BDRV_O_NATIVE_AIO)) {
503 error_setg(errp, "aio=native was specified, but it requires "
504 "cache.direct=on, which was not specified.");
509 if (bdrv_flags & BDRV_O_NATIVE_AIO) {
510 error_setg(errp, "aio=native was specified, but is not supported "
515 #endif /* !defined(CONFIG_LINUX_AIO) */
517 s->has_discard = true;
518 s->has_write_zeroes = true;
519 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
520 s->needs_alignment = true;
523 if (fstat(s->fd, &st) < 0) {
525 error_setg_errno(errp, errno, "Could not stat file");
528 if (S_ISREG(st.st_mode)) {
529 s->discard_zeroes = true;
530 s->has_fallocate = true;
532 if (S_ISBLK(st.st_mode)) {
533 #ifdef BLKDISCARDZEROES
535 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
536 s->discard_zeroes = true;
540 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
541 * not rely on the contents of discarded blocks unless using O_DIRECT.
542 * Same for BLKZEROOUT.
544 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
545 s->discard_zeroes = false;
546 s->has_write_zeroes = false;
551 if (S_ISCHR(st.st_mode)) {
553 * The file is a char device (disk), which on FreeBSD isn't behind
554 * a pager, so force all requests to be aligned. This is needed
555 * so QEMU makes sure all IO operations on the device are aligned
556 * to sector size, or else FreeBSD will reject them with EINVAL.
558 s->needs_alignment = true;
563 if (platform_test_xfs_fd(s->fd)) {
568 raw_attach_aio_context(bs, bdrv_get_aio_context(bs));
572 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
579 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
582 BDRVRawState *s = bs->opaque;
583 Error *local_err = NULL;
586 s->type = FTYPE_FILE;
587 ret = raw_open_common(bs, options, flags, 0, &local_err);
589 error_propagate(errp, local_err);
594 static int raw_reopen_prepare(BDRVReopenState *state,
595 BlockReopenQueue *queue, Error **errp)
598 BDRVRawReopenState *raw_s;
600 Error *local_err = NULL;
602 assert(state != NULL);
603 assert(state->bs != NULL);
605 s = state->bs->opaque;
607 state->opaque = g_new0(BDRVRawReopenState, 1);
608 raw_s = state->opaque;
610 #ifdef CONFIG_LINUX_AIO
611 raw_s->use_aio = s->use_aio;
613 /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
614 * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
615 * won't override aio_ctx if aio_ctx is non-NULL */
616 if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
617 error_setg(errp, "Could not set AIO state");
622 if (s->type == FTYPE_CD) {
623 raw_s->open_flags |= O_NONBLOCK;
626 raw_parse_flags(state->flags, &raw_s->open_flags);
630 int fcntl_flags = O_APPEND | O_NONBLOCK;
632 fcntl_flags |= O_NOATIME;
636 /* Not all operating systems have O_ASYNC, and those that don't
637 * will not let us track the state into raw_s->open_flags (typically
638 * you achieve the same effect with an ioctl, for example I_SETSIG
639 * on Solaris). But we do not use O_ASYNC, so that's fine.
641 assert((s->open_flags & O_ASYNC) == 0);
644 if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
645 /* dup the original fd */
646 /* TODO: use qemu fcntl wrapper */
647 #ifdef F_DUPFD_CLOEXEC
648 raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
650 raw_s->fd = dup(s->fd);
651 if (raw_s->fd != -1) {
652 qemu_set_cloexec(raw_s->fd);
655 if (raw_s->fd >= 0) {
656 ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
658 qemu_close(raw_s->fd);
664 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
665 if (raw_s->fd == -1) {
666 const char *normalized_filename = state->bs->filename;
667 ret = raw_normalize_devicepath(&normalized_filename);
669 error_setg_errno(errp, -ret, "Could not normalize device path");
671 assert(!(raw_s->open_flags & O_CREAT));
672 raw_s->fd = qemu_open(normalized_filename, raw_s->open_flags);
673 if (raw_s->fd == -1) {
674 error_setg_errno(errp, errno, "Could not reopen file");
680 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
681 * alignment with the new fd. */
682 if (raw_s->fd != -1) {
683 raw_probe_alignment(state->bs, raw_s->fd, &local_err);
685 qemu_close(raw_s->fd);
687 error_propagate(errp, local_err);
695 static void raw_reopen_commit(BDRVReopenState *state)
697 BDRVRawReopenState *raw_s = state->opaque;
698 BDRVRawState *s = state->bs->opaque;
700 s->open_flags = raw_s->open_flags;
704 #ifdef CONFIG_LINUX_AIO
705 s->use_aio = raw_s->use_aio;
708 g_free(state->opaque);
709 state->opaque = NULL;
713 static void raw_reopen_abort(BDRVReopenState *state)
715 BDRVRawReopenState *raw_s = state->opaque;
717 /* nothing to do if NULL, we didn't get far enough */
722 if (raw_s->fd >= 0) {
723 qemu_close(raw_s->fd);
726 g_free(state->opaque);
727 state->opaque = NULL;
730 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
732 BDRVRawState *s = bs->opaque;
734 raw_probe_alignment(bs, s->fd, errp);
735 bs->bl.min_mem_alignment = s->buf_align;
736 bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
739 static int check_for_dasd(int fd)
742 struct dasd_information2_t info = {0};
744 return ioctl(fd, BIODASDINFO2, &info);
751 * Try to get @bs's logical and physical block size.
752 * On success, store them in @bsz and return zero.
753 * On failure, return negative errno.
755 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
757 BDRVRawState *s = bs->opaque;
760 /* If DASD, get blocksizes */
761 if (check_for_dasd(s->fd) < 0) {
764 ret = probe_logical_blocksize(s->fd, &bsz->log);
768 return probe_physical_blocksize(s->fd, &bsz->phys);
772 * Try to get @bs's geometry: cyls, heads, sectors.
773 * On success, store them in @geo and return 0.
774 * On failure return -errno.
775 * (Allows block driver to assign default geometry values that guest sees)
778 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
780 BDRVRawState *s = bs->opaque;
781 struct hd_geometry ioctl_geo = {0};
784 /* If DASD, get its geometry */
785 if (check_for_dasd(s->fd) < 0) {
788 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
791 /* HDIO_GETGEO may return success even though geo contains zeros
792 (e.g. certain multipath setups) */
793 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
796 /* Do not return a geometry for partition */
797 if (ioctl_geo.start != 0) {
800 geo->heads = ioctl_geo.heads;
801 geo->sectors = ioctl_geo.sectors;
802 if (!probe_physical_blocksize(s->fd, &blksize)) {
803 /* overwrite cyls: HDIO_GETGEO result is incorrect for big drives */
804 geo->cylinders = bdrv_nb_sectors(bs) / (blksize / BDRV_SECTOR_SIZE)
805 / (geo->heads * geo->sectors);
808 geo->cylinders = ioctl_geo.cylinders;
812 #else /* __linux__ */
813 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
819 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
823 ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
831 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
835 ret = qemu_fdatasync(aiocb->aio_fildes);
844 static bool preadv_present = true;
847 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
849 return preadv(fd, iov, nr_iov, offset);
853 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
855 return pwritev(fd, iov, nr_iov, offset);
860 static bool preadv_present = false;
863 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
869 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
876 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
881 if (aiocb->aio_type & QEMU_AIO_WRITE)
882 len = qemu_pwritev(aiocb->aio_fildes,
887 len = qemu_preadv(aiocb->aio_fildes,
891 } while (len == -1 && errno == EINTR);
900 * Read/writes the data to/from a given linear buffer.
902 * Returns the number of bytes handles or -errno in case of an error. Short
903 * reads are only returned if the end of the file is reached.
905 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
910 while (offset < aiocb->aio_nbytes) {
911 if (aiocb->aio_type & QEMU_AIO_WRITE) {
912 len = pwrite(aiocb->aio_fildes,
913 (const char *)buf + offset,
914 aiocb->aio_nbytes - offset,
915 aiocb->aio_offset + offset);
917 len = pread(aiocb->aio_fildes,
919 aiocb->aio_nbytes - offset,
920 aiocb->aio_offset + offset);
922 if (len == -1 && errno == EINTR) {
924 } else if (len == -1 && errno == EINVAL &&
925 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
926 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
928 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
929 * after a short read. Assume that O_DIRECT short reads only occur
930 * at EOF. Therefore this is a short read, not an I/O error.
933 } else if (len == -1) {
936 } else if (len == 0) {
945 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
950 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
952 * If there is just a single buffer, and it is properly aligned
953 * we can just use plain pread/pwrite without any problems.
955 if (aiocb->aio_niov == 1) {
956 return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
959 * We have more than one iovec, and all are properly aligned.
961 * Try preadv/pwritev first and fall back to linearizing the
962 * buffer if it's not supported.
964 if (preadv_present) {
965 nbytes = handle_aiocb_rw_vector(aiocb);
966 if (nbytes == aiocb->aio_nbytes ||
967 (nbytes < 0 && nbytes != -ENOSYS)) {
970 preadv_present = false;
974 * XXX(hch): short read/write. no easy way to handle the reminder
975 * using these interfaces. For now retry using plain
981 * Ok, we have to do it the hard way, copy all segments into
982 * a single aligned buffer.
984 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
989 if (aiocb->aio_type & QEMU_AIO_WRITE) {
993 for (i = 0; i < aiocb->aio_niov; ++i) {
994 memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
995 p += aiocb->aio_iov[i].iov_len;
997 assert(p - buf == aiocb->aio_nbytes);
1000 nbytes = handle_aiocb_rw_linear(aiocb, buf);
1001 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
1003 size_t count = aiocb->aio_nbytes, copy;
1006 for (i = 0; i < aiocb->aio_niov && count; ++i) {
1008 if (copy > aiocb->aio_iov[i].iov_len) {
1009 copy = aiocb->aio_iov[i].iov_len;
1011 memcpy(aiocb->aio_iov[i].iov_base, p, copy);
1012 assert(count >= copy);
1024 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
1026 struct xfs_flock64 fl;
1029 memset(&fl, 0, sizeof(fl));
1030 fl.l_whence = SEEK_SET;
1031 fl.l_start = offset;
1034 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
1036 DPRINTF("cannot write zero range (%s)\n", strerror(errno));
1043 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
1045 struct xfs_flock64 fl;
1048 memset(&fl, 0, sizeof(fl));
1049 fl.l_whence = SEEK_SET;
1050 fl.l_start = offset;
1053 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
1055 DPRINTF("cannot punch hole (%s)\n", strerror(errno));
1063 static int translate_err(int err)
1065 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1072 #ifdef CONFIG_FALLOCATE
1073 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1076 if (fallocate(fd, mode, offset, len) == 0) {
1079 } while (errno == EINTR);
1080 return translate_err(-errno);
1084 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1087 BDRVRawState *s = aiocb->bs->opaque;
1089 if (!s->has_write_zeroes) {
1095 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1096 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1099 } while (errno == EINTR);
1101 ret = translate_err(-errno);
1104 if (ret == -ENOTSUP) {
1105 s->has_write_zeroes = false;
1110 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
1112 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1113 BDRVRawState *s = aiocb->bs->opaque;
1116 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1117 return handle_aiocb_write_zeroes_block(aiocb);
1122 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
1126 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1127 if (s->has_write_zeroes) {
1128 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1129 aiocb->aio_offset, aiocb->aio_nbytes);
1130 if (ret == 0 || ret != -ENOTSUP) {
1133 s->has_write_zeroes = false;
1137 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1138 if (s->has_discard && s->has_fallocate) {
1139 int ret = do_fallocate(s->fd,
1140 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1141 aiocb->aio_offset, aiocb->aio_nbytes);
1143 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1144 if (ret == 0 || ret != -ENOTSUP) {
1147 s->has_fallocate = false;
1148 } else if (ret != -ENOTSUP) {
1151 s->has_discard = false;
1156 #ifdef CONFIG_FALLOCATE
1157 if (s->has_fallocate && aiocb->aio_offset >= bdrv_getlength(aiocb->bs)) {
1158 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1159 if (ret == 0 || ret != -ENOTSUP) {
1162 s->has_fallocate = false;
1169 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
1171 int ret = -EOPNOTSUPP;
1172 BDRVRawState *s = aiocb->bs->opaque;
1174 if (!s->has_discard) {
1178 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1181 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1182 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1185 } while (errno == EINTR);
1192 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1196 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1197 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1198 aiocb->aio_offset, aiocb->aio_nbytes);
1202 ret = translate_err(ret);
1203 if (ret == -ENOTSUP) {
1204 s->has_discard = false;
1209 static int aio_worker(void *arg)
1211 RawPosixAIOData *aiocb = arg;
1214 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
1216 ret = handle_aiocb_rw(aiocb);
1217 if (ret >= 0 && ret < aiocb->aio_nbytes) {
1218 iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
1219 0, aiocb->aio_nbytes - ret);
1221 ret = aiocb->aio_nbytes;
1223 if (ret == aiocb->aio_nbytes) {
1225 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1229 case QEMU_AIO_WRITE:
1230 ret = handle_aiocb_rw(aiocb);
1231 if (ret == aiocb->aio_nbytes) {
1233 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1237 case QEMU_AIO_FLUSH:
1238 ret = handle_aiocb_flush(aiocb);
1240 case QEMU_AIO_IOCTL:
1241 ret = handle_aiocb_ioctl(aiocb);
1243 case QEMU_AIO_DISCARD:
1244 ret = handle_aiocb_discard(aiocb);
1246 case QEMU_AIO_WRITE_ZEROES:
1247 ret = handle_aiocb_write_zeroes(aiocb);
1250 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1259 static int paio_submit_co(BlockDriverState *bs, int fd,
1260 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1263 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1267 acb->aio_type = type;
1268 acb->aio_fildes = fd;
1270 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1271 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1274 acb->aio_iov = qiov->iov;
1275 acb->aio_niov = qiov->niov;
1276 assert(qiov->size == acb->aio_nbytes);
1279 trace_paio_submit_co(sector_num, nb_sectors, type);
1280 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1281 return thread_pool_submit_co(pool, aio_worker, acb);
1284 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
1285 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1286 BlockCompletionFunc *cb, void *opaque, int type)
1288 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1292 acb->aio_type = type;
1293 acb->aio_fildes = fd;
1295 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1296 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1299 acb->aio_iov = qiov->iov;
1300 acb->aio_niov = qiov->niov;
1301 assert(qiov->size == acb->aio_nbytes);
1304 trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
1305 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1306 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1309 static BlockAIOCB *raw_aio_submit(BlockDriverState *bs,
1310 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1311 BlockCompletionFunc *cb, void *opaque, int type)
1313 BDRVRawState *s = bs->opaque;
1315 if (fd_open(bs) < 0)
1319 * Check if the underlying device requires requests to be aligned,
1320 * and if the request we are trying to submit is aligned or not.
1321 * If this is the case tell the low-level driver that it needs
1322 * to copy the buffer.
1324 if (s->needs_alignment) {
1325 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1326 type |= QEMU_AIO_MISALIGNED;
1327 #ifdef CONFIG_LINUX_AIO
1328 } else if (s->use_aio) {
1329 return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
1330 nb_sectors, cb, opaque, type);
1335 return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
1339 static void raw_aio_plug(BlockDriverState *bs)
1341 #ifdef CONFIG_LINUX_AIO
1342 BDRVRawState *s = bs->opaque;
1344 laio_io_plug(bs, s->aio_ctx);
1349 static void raw_aio_unplug(BlockDriverState *bs)
1351 #ifdef CONFIG_LINUX_AIO
1352 BDRVRawState *s = bs->opaque;
1354 laio_io_unplug(bs, s->aio_ctx, true);
1359 static void raw_aio_flush_io_queue(BlockDriverState *bs)
1361 #ifdef CONFIG_LINUX_AIO
1362 BDRVRawState *s = bs->opaque;
1364 laio_io_unplug(bs, s->aio_ctx, false);
1369 static BlockAIOCB *raw_aio_readv(BlockDriverState *bs,
1370 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1371 BlockCompletionFunc *cb, void *opaque)
1373 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1374 cb, opaque, QEMU_AIO_READ);
1377 static BlockAIOCB *raw_aio_writev(BlockDriverState *bs,
1378 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1379 BlockCompletionFunc *cb, void *opaque)
1381 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1382 cb, opaque, QEMU_AIO_WRITE);
1385 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
1386 BlockCompletionFunc *cb, void *opaque)
1388 BDRVRawState *s = bs->opaque;
1390 if (fd_open(bs) < 0)
1393 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1396 static void raw_close(BlockDriverState *bs)
1398 BDRVRawState *s = bs->opaque;
1400 raw_detach_aio_context(bs);
1402 #ifdef CONFIG_LINUX_AIO
1404 laio_cleanup(s->aio_ctx);
1413 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1415 BDRVRawState *s = bs->opaque;
1418 if (fstat(s->fd, &st)) {
1422 if (S_ISREG(st.st_mode)) {
1423 if (ftruncate(s->fd, offset) < 0) {
1426 } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1427 if (offset > raw_getlength(bs)) {
1438 static int64_t raw_getlength(BlockDriverState *bs)
1440 BDRVRawState *s = bs->opaque;
1446 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1447 struct disklabel dl;
1449 if (ioctl(fd, DIOCGDINFO, &dl))
1451 return (uint64_t)dl.d_secsize *
1452 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1456 #elif defined(__NetBSD__)
1457 static int64_t raw_getlength(BlockDriverState *bs)
1459 BDRVRawState *s = bs->opaque;
1465 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1466 struct dkwedge_info dkw;
1468 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1469 return dkw.dkw_size * 512;
1471 struct disklabel dl;
1473 if (ioctl(fd, DIOCGDINFO, &dl))
1475 return (uint64_t)dl.d_secsize *
1476 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1481 #elif defined(__sun__)
1482 static int64_t raw_getlength(BlockDriverState *bs)
1484 BDRVRawState *s = bs->opaque;
1485 struct dk_minfo minfo;
1495 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1497 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1499 return minfo.dki_lbsize * minfo.dki_capacity;
1503 * There are reports that lseek on some devices fails, but
1504 * irc discussion said that contingency on contingency was overkill.
1506 size = lseek(s->fd, 0, SEEK_END);
1512 #elif defined(CONFIG_BSD)
1513 static int64_t raw_getlength(BlockDriverState *bs)
1515 BDRVRawState *s = bs->opaque;
1519 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1528 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1531 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1532 #ifdef DIOCGMEDIASIZE
1533 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1534 #elif defined(DIOCGPART)
1537 if (ioctl(fd, DIOCGPART, &pi) == 0)
1538 size = pi.media_size;
1544 #if defined(__APPLE__) && defined(__MACH__)
1546 uint64_t sectors = 0;
1547 uint32_t sector_size = 0;
1549 if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0
1550 && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) {
1551 size = sectors * sector_size;
1553 size = lseek(fd, 0LL, SEEK_END);
1560 size = lseek(fd, 0LL, SEEK_END);
1565 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1568 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1569 if (size == 2048LL * (unsigned)-1)
1571 /* XXX no disc? maybe we need to reopen... */
1572 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1579 size = lseek(fd, 0, SEEK_END);
1587 static int64_t raw_getlength(BlockDriverState *bs)
1589 BDRVRawState *s = bs->opaque;
1598 size = lseek(s->fd, 0, SEEK_END);
1606 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1609 BDRVRawState *s = bs->opaque;
1611 if (fstat(s->fd, &st) < 0) {
1614 return (int64_t)st.st_blocks * 512;
1617 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
1621 int64_t total_size = 0;
1623 PreallocMode prealloc;
1625 Error *local_err = NULL;
1627 strstart(filename, "file:", &filename);
1629 /* Read out options */
1630 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1632 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
1633 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1634 prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
1635 PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
1639 error_propagate(errp, local_err);
1644 fd = qemu_open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY,
1648 error_setg_errno(errp, -result, "Could not create file");
1654 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1655 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1656 * will be ignored since any failure of this operation should not
1657 * block the left work.
1660 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
1661 attr |= FS_NOCOW_FL;
1662 ioctl(fd, FS_IOC_SETFLAGS, &attr);
1667 if (ftruncate(fd, total_size) != 0) {
1669 error_setg_errno(errp, -result, "Could not resize file");
1674 #ifdef CONFIG_POSIX_FALLOCATE
1675 case PREALLOC_MODE_FALLOC:
1676 /* posix_fallocate() doesn't set errno. */
1677 result = -posix_fallocate(fd, 0, total_size);
1679 error_setg_errno(errp, -result,
1680 "Could not preallocate data for the new file");
1684 case PREALLOC_MODE_FULL:
1686 int64_t num = 0, left = total_size;
1687 buf = g_malloc0(65536);
1690 num = MIN(left, 65536);
1691 result = write(fd, buf, num);
1694 error_setg_errno(errp, -result,
1695 "Could not write to the new file");
1704 error_setg_errno(errp, -result,
1705 "Could not flush new file to disk");
1711 case PREALLOC_MODE_OFF:
1715 error_setg(errp, "Unsupported preallocation mode: %s",
1716 PreallocMode_lookup[prealloc]);
1721 if (qemu_close(fd) != 0 && result == 0) {
1723 error_setg_errno(errp, -result, "Could not close the new file");
1730 * Find allocation range in @bs around offset @start.
1731 * May change underlying file descriptor's file offset.
1732 * If @start is not in a hole, store @start in @data, and the
1733 * beginning of the next hole in @hole, and return 0.
1734 * If @start is in a non-trailing hole, store @start in @hole and the
1735 * beginning of the next non-hole in @data, and return 0.
1736 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1737 * If we can't find out, return a negative errno other than -ENXIO.
1739 static int find_allocation(BlockDriverState *bs, off_t start,
1740 off_t *data, off_t *hole)
1742 #if defined SEEK_HOLE && defined SEEK_DATA
1743 BDRVRawState *s = bs->opaque;
1748 * D1. offs == start: start is in data
1749 * D2. offs > start: start is in a hole, next data at offs
1750 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1751 * or start is beyond EOF
1752 * If the latter happens, the file has been truncated behind
1753 * our back since we opened it. All bets are off then.
1754 * Treating like a trailing hole is simplest.
1755 * D4. offs < 0, errno != ENXIO: we learned nothing
1757 offs = lseek(s->fd, start, SEEK_DATA);
1759 return -errno; /* D3 or D4 */
1761 assert(offs >= start);
1764 /* D2: in hole, next data at offs */
1770 /* D1: in data, end not yet known */
1774 * H1. offs == start: start is in a hole
1775 * If this happens here, a hole has been dug behind our back
1776 * since the previous lseek().
1777 * H2. offs > start: either start is in data, next hole at offs,
1778 * or start is in trailing hole, EOF at offs
1779 * Linux treats trailing holes like any other hole: offs ==
1780 * start. Solaris seeks to EOF instead: offs > start (blech).
1781 * If that happens here, a hole has been dug behind our back
1782 * since the previous lseek().
1783 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1784 * If this happens, the file has been truncated behind our
1785 * back since we opened it. Treat it like a trailing hole.
1786 * H4. offs < 0, errno != ENXIO: we learned nothing
1787 * Pretend we know nothing at all, i.e. "forget" about D1.
1789 offs = lseek(s->fd, start, SEEK_HOLE);
1791 return -errno; /* D1 and (H3 or H4) */
1793 assert(offs >= start);
1797 * D1 and H2: either in data, next hole at offs, or it was in
1798 * data but is now in a trailing hole. In the latter case,
1799 * all bets are off. Treating it as if it there was data all
1800 * the way to EOF is safe, so simply do that.
1815 * Returns the allocation status of the specified sectors.
1817 * If 'sector_num' is beyond the end of the disk image the return value is 0
1818 * and 'pnum' is set to 0.
1820 * 'pnum' is set to the number of sectors (including and immediately following
1821 * the specified sector) that are known to be in the same
1822 * allocated/unallocated state.
1824 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1825 * beyond the end of the disk image it will be clamped.
1827 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1829 int nb_sectors, int *pnum)
1831 off_t start, data = 0, hole = 0;
1840 start = sector_num * BDRV_SECTOR_SIZE;
1841 total_size = bdrv_getlength(bs);
1842 if (total_size < 0) {
1844 } else if (start >= total_size) {
1847 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1848 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1851 ret = find_allocation(bs, start, &data, &hole);
1852 if (ret == -ENXIO) {
1855 ret = BDRV_BLOCK_ZERO;
1856 } else if (ret < 0) {
1857 /* No info available, so pretend there are no holes */
1859 ret = BDRV_BLOCK_DATA;
1860 } else if (data == start) {
1861 /* On a data extent, compute sectors to the end of the extent,
1862 * possibly including a partial sector at EOF. */
1863 *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1864 ret = BDRV_BLOCK_DATA;
1866 /* On a hole, compute sectors to the beginning of the next extent. */
1867 assert(hole == start);
1868 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1869 ret = BDRV_BLOCK_ZERO;
1871 return ret | BDRV_BLOCK_OFFSET_VALID | start;
1874 static coroutine_fn BlockAIOCB *raw_aio_discard(BlockDriverState *bs,
1875 int64_t sector_num, int nb_sectors,
1876 BlockCompletionFunc *cb, void *opaque)
1878 BDRVRawState *s = bs->opaque;
1880 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1881 cb, opaque, QEMU_AIO_DISCARD);
1884 static int coroutine_fn raw_co_write_zeroes(
1885 BlockDriverState *bs, int64_t sector_num,
1886 int nb_sectors, BdrvRequestFlags flags)
1888 BDRVRawState *s = bs->opaque;
1890 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1891 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1892 QEMU_AIO_WRITE_ZEROES);
1893 } else if (s->discard_zeroes) {
1894 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1900 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1902 BDRVRawState *s = bs->opaque;
1904 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1905 bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1909 static QemuOptsList raw_create_opts = {
1910 .name = "raw-create-opts",
1911 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
1914 .name = BLOCK_OPT_SIZE,
1915 .type = QEMU_OPT_SIZE,
1916 .help = "Virtual disk size"
1919 .name = BLOCK_OPT_NOCOW,
1920 .type = QEMU_OPT_BOOL,
1921 .help = "Turn off copy-on-write (valid only on btrfs)"
1924 .name = BLOCK_OPT_PREALLOC,
1925 .type = QEMU_OPT_STRING,
1926 .help = "Preallocation mode (allowed values: off, falloc, full)"
1928 { /* end of list */ }
1932 BlockDriver bdrv_file = {
1933 .format_name = "file",
1934 .protocol_name = "file",
1935 .instance_size = sizeof(BDRVRawState),
1936 .bdrv_needs_filename = true,
1937 .bdrv_probe = NULL, /* no probe for protocols */
1938 .bdrv_parse_filename = raw_parse_filename,
1939 .bdrv_file_open = raw_open,
1940 .bdrv_reopen_prepare = raw_reopen_prepare,
1941 .bdrv_reopen_commit = raw_reopen_commit,
1942 .bdrv_reopen_abort = raw_reopen_abort,
1943 .bdrv_close = raw_close,
1944 .bdrv_create = raw_create,
1945 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1946 .bdrv_co_get_block_status = raw_co_get_block_status,
1947 .bdrv_co_write_zeroes = raw_co_write_zeroes,
1949 .bdrv_aio_readv = raw_aio_readv,
1950 .bdrv_aio_writev = raw_aio_writev,
1951 .bdrv_aio_flush = raw_aio_flush,
1952 .bdrv_aio_discard = raw_aio_discard,
1953 .bdrv_refresh_limits = raw_refresh_limits,
1954 .bdrv_io_plug = raw_aio_plug,
1955 .bdrv_io_unplug = raw_aio_unplug,
1956 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
1958 .bdrv_truncate = raw_truncate,
1959 .bdrv_getlength = raw_getlength,
1960 .bdrv_get_info = raw_get_info,
1961 .bdrv_get_allocated_file_size
1962 = raw_get_allocated_file_size,
1964 .bdrv_detach_aio_context = raw_detach_aio_context,
1965 .bdrv_attach_aio_context = raw_attach_aio_context,
1967 .create_opts = &raw_create_opts,
1970 /***********************************************/
1973 #if defined(__APPLE__) && defined(__MACH__)
1974 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1975 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
1976 CFIndex maxPathSize, int flags);
1977 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1979 kern_return_t kernResult;
1980 mach_port_t masterPort;
1981 CFMutableDictionaryRef classesToMatch;
1983 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1984 if ( KERN_SUCCESS != kernResult ) {
1985 printf( "IOMasterPort returned %d\n", kernResult );
1988 classesToMatch = IOServiceMatching( kIOCDMediaClass );
1989 if ( classesToMatch == NULL ) {
1990 printf( "IOServiceMatching returned a NULL dictionary.\n" );
1992 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1994 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1995 if ( KERN_SUCCESS != kernResult )
1997 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
2003 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
2004 CFIndex maxPathSize, int flags)
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 if (flags & BDRV_O_NOCACHE) {
2018 strcat(bsdPath, "r");
2020 devPathLength = strlen( bsdPath );
2021 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
2022 kernResult = KERN_SUCCESS;
2024 CFRelease( bsdPathAsCFString );
2026 IOObjectRelease( nextMedia );
2034 static int hdev_probe_device(const char *filename)
2038 /* allow a dedicated CD-ROM driver to match with a higher priority */
2039 if (strstart(filename, "/dev/cdrom", NULL))
2042 if (stat(filename, &st) >= 0 &&
2043 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
2050 static int check_hdev_writable(BDRVRawState *s)
2052 #if defined(BLKROGET)
2053 /* Linux block devices can be configured "read-only" using blockdev(8).
2054 * This is independent of device node permissions and therefore open(2)
2055 * with O_RDWR succeeds. Actual writes fail with EPERM.
2057 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2058 * check for read-only block devices so that Linux block devices behave
2064 if (fstat(s->fd, &st)) {
2068 if (!S_ISBLK(st.st_mode)) {
2072 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
2079 #endif /* defined(BLKROGET) */
2083 static void hdev_parse_filename(const char *filename, QDict *options,
2086 /* The prefix is optional, just as for "file". */
2087 strstart(filename, "host_device:", &filename);
2089 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2092 static bool hdev_is_sg(BlockDriverState *bs)
2095 #if defined(__linux__)
2098 struct sg_scsi_id scsiid;
2101 if (stat(bs->filename, &st) >= 0 && S_ISCHR(st.st_mode) &&
2102 !bdrv_ioctl(bs, SG_GET_VERSION_NUM, &sg_version) &&
2103 !bdrv_ioctl(bs, SG_GET_SCSI_ID, &scsiid)) {
2104 DPRINTF("SG device found: type=%d, version=%d\n",
2105 scsiid.scsi_type, sg_version);
2114 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
2117 BDRVRawState *s = bs->opaque;
2118 Error *local_err = NULL;
2121 #if defined(__APPLE__) && defined(__MACH__)
2122 const char *filename = qdict_get_str(options, "filename");
2124 if (strstart(filename, "/dev/cdrom", NULL)) {
2125 kern_return_t kernResult;
2126 io_iterator_t mediaIterator;
2127 char bsdPath[ MAXPATHLEN ];
2130 kernResult = FindEjectableCDMedia( &mediaIterator );
2131 kernResult = GetBSDPath(mediaIterator, bsdPath, sizeof(bsdPath),
2133 if ( bsdPath[ 0 ] != '\0' ) {
2134 strcat(bsdPath,"s0");
2135 /* some CDs don't have a partition 0 */
2136 fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
2138 bsdPath[strlen(bsdPath)-1] = '1';
2143 qdict_put(options, "filename", qstring_from_str(filename));
2146 if ( mediaIterator )
2147 IOObjectRelease( mediaIterator );
2151 s->type = FTYPE_FILE;
2153 ret = raw_open_common(bs, options, flags, 0, &local_err);
2156 error_propagate(errp, local_err);
2161 /* Since this does ioctl the device must be already opened */
2162 bs->sg = hdev_is_sg(bs);
2164 if (flags & BDRV_O_RDWR) {
2165 ret = check_hdev_writable(s);
2168 error_setg_errno(errp, -ret, "The device is not writable");
2176 #if defined(__linux__)
2178 static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
2179 unsigned long int req, void *buf,
2180 BlockCompletionFunc *cb, void *opaque)
2182 BDRVRawState *s = bs->opaque;
2183 RawPosixAIOData *acb;
2186 if (fd_open(bs) < 0)
2189 acb = g_new(RawPosixAIOData, 1);
2191 acb->aio_type = QEMU_AIO_IOCTL;
2192 acb->aio_fildes = s->fd;
2193 acb->aio_offset = 0;
2194 acb->aio_ioctl_buf = buf;
2195 acb->aio_ioctl_cmd = req;
2196 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
2197 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
2201 static int fd_open(BlockDriverState *bs)
2203 BDRVRawState *s = bs->opaque;
2205 /* this is just to ensure s->fd is sane (its called by io ops) */
2211 static coroutine_fn BlockAIOCB *hdev_aio_discard(BlockDriverState *bs,
2212 int64_t sector_num, int nb_sectors,
2213 BlockCompletionFunc *cb, void *opaque)
2215 BDRVRawState *s = bs->opaque;
2217 if (fd_open(bs) < 0) {
2220 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
2221 cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2224 static coroutine_fn int hdev_co_write_zeroes(BlockDriverState *bs,
2225 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2227 BDRVRawState *s = bs->opaque;
2234 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2235 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2236 QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2237 } else if (s->discard_zeroes) {
2238 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2239 QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2244 static int hdev_create(const char *filename, QemuOpts *opts,
2249 struct stat stat_buf;
2250 int64_t total_size = 0;
2253 /* This function is used by both protocol block drivers and therefore either
2254 * of these prefixes may be given.
2255 * The return value has to be stored somewhere, otherwise this is an error
2256 * due to -Werror=unused-value. */
2258 strstart(filename, "host_device:", &filename) ||
2259 strstart(filename, "host_cdrom:" , &filename);
2263 ret = raw_normalize_devicepath(&filename);
2265 error_setg_errno(errp, -ret, "Could not normalize device path");
2269 /* Read out options */
2270 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2273 fd = qemu_open(filename, O_WRONLY | O_BINARY);
2276 error_setg_errno(errp, -ret, "Could not open device");
2280 if (fstat(fd, &stat_buf) < 0) {
2282 error_setg_errno(errp, -ret, "Could not stat device");
2283 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2285 "The given file is neither a block nor a character device");
2287 } else if (lseek(fd, 0, SEEK_END) < total_size) {
2288 error_setg(errp, "Device is too small");
2296 static BlockDriver bdrv_host_device = {
2297 .format_name = "host_device",
2298 .protocol_name = "host_device",
2299 .instance_size = sizeof(BDRVRawState),
2300 .bdrv_needs_filename = true,
2301 .bdrv_probe_device = hdev_probe_device,
2302 .bdrv_parse_filename = hdev_parse_filename,
2303 .bdrv_file_open = hdev_open,
2304 .bdrv_close = raw_close,
2305 .bdrv_reopen_prepare = raw_reopen_prepare,
2306 .bdrv_reopen_commit = raw_reopen_commit,
2307 .bdrv_reopen_abort = raw_reopen_abort,
2308 .bdrv_create = hdev_create,
2309 .create_opts = &raw_create_opts,
2310 .bdrv_co_write_zeroes = hdev_co_write_zeroes,
2312 .bdrv_aio_readv = raw_aio_readv,
2313 .bdrv_aio_writev = raw_aio_writev,
2314 .bdrv_aio_flush = raw_aio_flush,
2315 .bdrv_aio_discard = hdev_aio_discard,
2316 .bdrv_refresh_limits = raw_refresh_limits,
2317 .bdrv_io_plug = raw_aio_plug,
2318 .bdrv_io_unplug = raw_aio_unplug,
2319 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2321 .bdrv_truncate = raw_truncate,
2322 .bdrv_getlength = raw_getlength,
2323 .bdrv_get_info = raw_get_info,
2324 .bdrv_get_allocated_file_size
2325 = raw_get_allocated_file_size,
2326 .bdrv_probe_blocksizes = hdev_probe_blocksizes,
2327 .bdrv_probe_geometry = hdev_probe_geometry,
2329 .bdrv_detach_aio_context = raw_detach_aio_context,
2330 .bdrv_attach_aio_context = raw_attach_aio_context,
2332 /* generic scsi device */
2334 .bdrv_aio_ioctl = hdev_aio_ioctl,
2338 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2339 static void cdrom_parse_filename(const char *filename, QDict *options,
2342 /* The prefix is optional, just as for "file". */
2343 strstart(filename, "host_cdrom:", &filename);
2345 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2350 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2353 BDRVRawState *s = bs->opaque;
2354 Error *local_err = NULL;
2359 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2360 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2362 error_propagate(errp, local_err);
2367 static int cdrom_probe_device(const char *filename)
2373 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2377 ret = fstat(fd, &st);
2378 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2382 /* Attempt to detect via a CDROM specific ioctl */
2383 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2393 static bool cdrom_is_inserted(BlockDriverState *bs)
2395 BDRVRawState *s = bs->opaque;
2398 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2399 return ret == CDS_DISC_OK;
2402 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2404 BDRVRawState *s = bs->opaque;
2407 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2408 perror("CDROMEJECT");
2410 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2411 perror("CDROMEJECT");
2415 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2417 BDRVRawState *s = bs->opaque;
2419 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2421 * Note: an error can happen if the distribution automatically
2424 /* perror("CDROM_LOCKDOOR"); */
2428 static BlockDriver bdrv_host_cdrom = {
2429 .format_name = "host_cdrom",
2430 .protocol_name = "host_cdrom",
2431 .instance_size = sizeof(BDRVRawState),
2432 .bdrv_needs_filename = true,
2433 .bdrv_probe_device = cdrom_probe_device,
2434 .bdrv_parse_filename = cdrom_parse_filename,
2435 .bdrv_file_open = cdrom_open,
2436 .bdrv_close = raw_close,
2437 .bdrv_reopen_prepare = raw_reopen_prepare,
2438 .bdrv_reopen_commit = raw_reopen_commit,
2439 .bdrv_reopen_abort = raw_reopen_abort,
2440 .bdrv_create = hdev_create,
2441 .create_opts = &raw_create_opts,
2443 .bdrv_aio_readv = raw_aio_readv,
2444 .bdrv_aio_writev = raw_aio_writev,
2445 .bdrv_aio_flush = raw_aio_flush,
2446 .bdrv_refresh_limits = raw_refresh_limits,
2447 .bdrv_io_plug = raw_aio_plug,
2448 .bdrv_io_unplug = raw_aio_unplug,
2449 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2451 .bdrv_truncate = raw_truncate,
2452 .bdrv_getlength = raw_getlength,
2453 .has_variable_length = true,
2454 .bdrv_get_allocated_file_size
2455 = raw_get_allocated_file_size,
2457 .bdrv_detach_aio_context = raw_detach_aio_context,
2458 .bdrv_attach_aio_context = raw_attach_aio_context,
2460 /* removable device support */
2461 .bdrv_is_inserted = cdrom_is_inserted,
2462 .bdrv_eject = cdrom_eject,
2463 .bdrv_lock_medium = cdrom_lock_medium,
2465 /* generic scsi device */
2466 .bdrv_aio_ioctl = hdev_aio_ioctl,
2468 #endif /* __linux__ */
2470 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2471 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2474 BDRVRawState *s = bs->opaque;
2475 Error *local_err = NULL;
2480 ret = raw_open_common(bs, options, flags, 0, &local_err);
2483 error_propagate(errp, local_err);
2488 /* make sure the door isn't locked at this time */
2489 ioctl(s->fd, CDIOCALLOW);
2493 static int cdrom_probe_device(const char *filename)
2495 if (strstart(filename, "/dev/cd", NULL) ||
2496 strstart(filename, "/dev/acd", NULL))
2501 static int cdrom_reopen(BlockDriverState *bs)
2503 BDRVRawState *s = bs->opaque;
2507 * Force reread of possibly changed/newly loaded disc,
2508 * FreeBSD seems to not notice sometimes...
2512 fd = qemu_open(bs->filename, s->open_flags, 0644);
2519 /* make sure the door isn't locked at this time */
2520 ioctl(s->fd, CDIOCALLOW);
2524 static bool cdrom_is_inserted(BlockDriverState *bs)
2526 return raw_getlength(bs) > 0;
2529 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2531 BDRVRawState *s = bs->opaque;
2536 (void) ioctl(s->fd, CDIOCALLOW);
2539 if (ioctl(s->fd, CDIOCEJECT) < 0)
2540 perror("CDIOCEJECT");
2542 if (ioctl(s->fd, CDIOCCLOSE) < 0)
2543 perror("CDIOCCLOSE");
2549 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2551 BDRVRawState *s = bs->opaque;
2555 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2557 * Note: an error can happen if the distribution automatically
2560 /* perror("CDROM_LOCKDOOR"); */
2564 static BlockDriver bdrv_host_cdrom = {
2565 .format_name = "host_cdrom",
2566 .protocol_name = "host_cdrom",
2567 .instance_size = sizeof(BDRVRawState),
2568 .bdrv_needs_filename = true,
2569 .bdrv_probe_device = cdrom_probe_device,
2570 .bdrv_parse_filename = cdrom_parse_filename,
2571 .bdrv_file_open = cdrom_open,
2572 .bdrv_close = raw_close,
2573 .bdrv_reopen_prepare = raw_reopen_prepare,
2574 .bdrv_reopen_commit = raw_reopen_commit,
2575 .bdrv_reopen_abort = raw_reopen_abort,
2576 .bdrv_create = hdev_create,
2577 .create_opts = &raw_create_opts,
2579 .bdrv_aio_readv = raw_aio_readv,
2580 .bdrv_aio_writev = raw_aio_writev,
2581 .bdrv_aio_flush = raw_aio_flush,
2582 .bdrv_refresh_limits = raw_refresh_limits,
2583 .bdrv_io_plug = raw_aio_plug,
2584 .bdrv_io_unplug = raw_aio_unplug,
2585 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2587 .bdrv_truncate = raw_truncate,
2588 .bdrv_getlength = raw_getlength,
2589 .has_variable_length = true,
2590 .bdrv_get_allocated_file_size
2591 = raw_get_allocated_file_size,
2593 .bdrv_detach_aio_context = raw_detach_aio_context,
2594 .bdrv_attach_aio_context = raw_attach_aio_context,
2596 /* removable device support */
2597 .bdrv_is_inserted = cdrom_is_inserted,
2598 .bdrv_eject = cdrom_eject,
2599 .bdrv_lock_medium = cdrom_lock_medium,
2601 #endif /* __FreeBSD__ */
2603 static void bdrv_file_init(void)
2606 * Register all the drivers. Note that order is important, the driver
2607 * registered last will get probed first.
2609 bdrv_register(&bdrv_file);
2610 bdrv_register(&bdrv_host_device);
2612 bdrv_register(&bdrv_host_cdrom);
2614 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2615 bdrv_register(&bdrv_host_cdrom);
2619 block_init(bdrv_file_init);