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/osdep.h"
25 #include "qapi/error.h"
26 #include "qemu/cutils.h"
27 #include "qemu/error-report.h"
28 #include "qemu/timer.h"
30 #include "block/block_int.h"
31 #include "qemu/module.h"
33 #include "block/thread-pool.h"
35 #include "block/raw-aio.h"
36 #include "qapi/util.h"
37 #include "qapi/qmp/qstring.h"
39 #if defined(__APPLE__) && (__MACH__)
41 #include <sys/param.h>
42 #include <IOKit/IOKitLib.h>
43 #include <IOKit/IOBSD.h>
44 #include <IOKit/storage/IOMediaBSDClient.h>
45 #include <IOKit/storage/IOMedia.h>
46 #include <IOKit/storage/IOCDMedia.h>
47 //#include <IOKit/storage/IOCDTypes.h>
48 #include <IOKit/storage/IODVDMedia.h>
49 #include <CoreFoundation/CoreFoundation.h>
53 #define _POSIX_PTHREAD_SEMANTICS 1
57 #include <sys/ioctl.h>
58 #include <sys/param.h>
59 #include <linux/cdrom.h>
62 #include <linux/hdreg.h>
68 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
71 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
72 #include <linux/falloc.h>
74 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
80 #include <sys/ioctl.h>
81 #include <sys/disklabel.h>
86 #include <sys/ioctl.h>
87 #include <sys/disklabel.h>
93 #include <sys/ioctl.h>
94 #include <sys/diskslice.h>
101 //#define DEBUG_BLOCK
104 # define DEBUG_BLOCK_PRINT 1
106 # define DEBUG_BLOCK_PRINT 0
108 #define DPRINTF(fmt, ...) \
110 if (DEBUG_BLOCK_PRINT) { \
111 printf(fmt, ## __VA_ARGS__); \
115 /* OS X does not have O_DSYNC */
118 #define O_DSYNC O_SYNC
119 #elif defined(O_FSYNC)
120 #define O_DSYNC O_FSYNC
124 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
126 #define O_DIRECT O_DSYNC
132 #define MAX_BLOCKSIZE 4096
134 typedef struct BDRVRawState {
144 bool has_write_zeroes:1;
145 bool discard_zeroes:1;
147 bool needs_alignment;
150 typedef struct BDRVRawReopenState {
153 } BDRVRawReopenState;
155 static int fd_open(BlockDriverState *bs);
156 static int64_t raw_getlength(BlockDriverState *bs);
158 typedef struct RawPosixAIOData {
159 BlockDriverState *bs;
162 struct iovec *aio_iov;
167 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
172 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
173 static int cdrom_reopen(BlockDriverState *bs);
176 #if defined(__NetBSD__)
177 static int raw_normalize_devicepath(const char **filename)
179 static char namebuf[PATH_MAX];
180 const char *dp, *fname;
184 dp = strrchr(fname, '/');
185 if (lstat(fname, &sb) < 0) {
186 fprintf(stderr, "%s: stat failed: %s\n",
187 fname, strerror(errno));
191 if (!S_ISBLK(sb.st_mode)) {
196 snprintf(namebuf, PATH_MAX, "r%s", fname);
198 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
199 (int)(dp - fname), fname, dp + 1);
201 fprintf(stderr, "%s is a block device", fname);
203 fprintf(stderr, ", using %s\n", *filename);
208 static int raw_normalize_devicepath(const char **filename)
215 * Get logical block size via ioctl. On success store it in @sector_size_p.
217 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
219 unsigned int sector_size;
220 bool success = false;
224 /* Try a few ioctls to get the right size */
226 if (ioctl(fd, BLKSSZGET, §or_size) >= 0) {
227 *sector_size_p = sector_size;
231 #ifdef DKIOCGETBLOCKSIZE
232 if (ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) >= 0) {
233 *sector_size_p = sector_size;
237 #ifdef DIOCGSECTORSIZE
238 if (ioctl(fd, DIOCGSECTORSIZE, §or_size) >= 0) {
239 *sector_size_p = sector_size;
244 return success ? 0 : -errno;
248 * Get physical block size of @fd.
249 * On success, store it in @blk_size and return 0.
250 * On failure, return -errno.
252 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
255 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
264 /* Check if read is allowed with given memory buffer and length.
266 * This function is used to check O_DIRECT memory buffer and request alignment.
268 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
270 ssize_t ret = pread(fd, buf, len, 0);
277 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
278 * other errors (e.g. real I/O error), which could happen on a failed
279 * drive, since we only care about probing alignment.
281 if (errno != EINVAL) {
289 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
291 BDRVRawState *s = bs->opaque;
293 size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize());
295 /* For SCSI generic devices the alignment is not really used.
296 With buffered I/O, we don't have any restrictions. */
297 if (bdrv_is_sg(bs) || !s->needs_alignment) {
298 bs->bl.request_alignment = 1;
303 bs->bl.request_alignment = 0;
305 /* Let's try to use the logical blocksize for the alignment. */
306 if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) {
307 bs->bl.request_alignment = 0;
312 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
313 bs->bl.request_alignment = da.d_miniosz;
314 /* The kernel returns wrong information for d_mem */
315 /* s->buf_align = da.d_mem; */
320 /* If we could not get the sizes so far, we can only guess them */
323 buf = qemu_memalign(max_align, 2 * max_align);
324 for (align = 512; align <= max_align; align <<= 1) {
325 if (raw_is_io_aligned(fd, buf + align, max_align)) {
326 s->buf_align = align;
333 if (!bs->bl.request_alignment) {
335 buf = qemu_memalign(s->buf_align, max_align);
336 for (align = 512; align <= max_align; align <<= 1) {
337 if (raw_is_io_aligned(fd, buf, align)) {
338 bs->bl.request_alignment = align;
345 if (!s->buf_align || !bs->bl.request_alignment) {
346 error_setg(errp, "Could not find working O_DIRECT alignment");
347 error_append_hint(errp, "Try cache.direct=off\n");
351 static void raw_parse_flags(int bdrv_flags, int *open_flags)
353 assert(open_flags != NULL);
355 *open_flags |= O_BINARY;
356 *open_flags &= ~O_ACCMODE;
357 if (bdrv_flags & BDRV_O_RDWR) {
358 *open_flags |= O_RDWR;
360 *open_flags |= O_RDONLY;
363 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
364 * and O_DIRECT for no caching. */
365 if ((bdrv_flags & BDRV_O_NOCACHE)) {
366 *open_flags |= O_DIRECT;
370 #ifdef CONFIG_LINUX_AIO
371 static bool raw_use_aio(int bdrv_flags)
374 * Currently Linux do AIO only for files opened with O_DIRECT
375 * specified so check NOCACHE flag too
377 return (bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
378 (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO);
382 static void raw_parse_filename(const char *filename, QDict *options,
385 /* The filename does not have to be prefixed by the protocol name, since
386 * "file" is the default protocol; therefore, the return value of this
387 * function call can be ignored. */
388 strstart(filename, "file:", &filename);
390 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
393 static QemuOptsList raw_runtime_opts = {
395 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
399 .type = QEMU_OPT_STRING,
400 .help = "File name of the image",
402 { /* end of list */ }
406 static int raw_open_common(BlockDriverState *bs, QDict *options,
407 int bdrv_flags, int open_flags, Error **errp)
409 BDRVRawState *s = bs->opaque;
411 Error *local_err = NULL;
412 const char *filename = NULL;
416 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
417 qemu_opts_absorb_qdict(opts, options, &local_err);
419 error_propagate(errp, local_err);
424 filename = qemu_opt_get(opts, "filename");
426 ret = raw_normalize_devicepath(&filename);
428 error_setg_errno(errp, -ret, "Could not normalize device path");
432 s->open_flags = open_flags;
433 raw_parse_flags(bdrv_flags, &s->open_flags);
436 fd = qemu_open(filename, s->open_flags, 0644);
446 #ifdef CONFIG_LINUX_AIO
447 if (!raw_use_aio(bdrv_flags) && (bdrv_flags & BDRV_O_NATIVE_AIO)) {
448 error_setg(errp, "aio=native was specified, but it requires "
449 "cache.direct=on, which was not specified.");
454 if (bdrv_flags & BDRV_O_NATIVE_AIO) {
455 error_setg(errp, "aio=native was specified, but is not supported "
460 #endif /* !defined(CONFIG_LINUX_AIO) */
462 s->has_discard = true;
463 s->has_write_zeroes = true;
464 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
465 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
466 s->needs_alignment = true;
469 if (fstat(s->fd, &st) < 0) {
471 error_setg_errno(errp, errno, "Could not stat file");
474 if (S_ISREG(st.st_mode)) {
475 s->discard_zeroes = true;
476 s->has_fallocate = true;
478 if (S_ISBLK(st.st_mode)) {
479 #ifdef BLKDISCARDZEROES
481 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
482 s->discard_zeroes = true;
486 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
487 * not rely on the contents of discarded blocks unless using O_DIRECT.
488 * Same for BLKZEROOUT.
490 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
491 s->discard_zeroes = false;
492 s->has_write_zeroes = false;
497 if (S_ISCHR(st.st_mode)) {
499 * The file is a char device (disk), which on FreeBSD isn't behind
500 * a pager, so force all requests to be aligned. This is needed
501 * so QEMU makes sure all IO operations on the device are aligned
502 * to sector size, or else FreeBSD will reject them with EINVAL.
504 s->needs_alignment = true;
509 if (platform_test_xfs_fd(s->fd)) {
516 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
523 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
526 BDRVRawState *s = bs->opaque;
528 s->type = FTYPE_FILE;
529 return raw_open_common(bs, options, flags, 0, errp);
532 static int raw_reopen_prepare(BDRVReopenState *state,
533 BlockReopenQueue *queue, Error **errp)
536 BDRVRawReopenState *raw_s;
538 Error *local_err = NULL;
540 assert(state != NULL);
541 assert(state->bs != NULL);
543 s = state->bs->opaque;
545 state->opaque = g_new0(BDRVRawReopenState, 1);
546 raw_s = state->opaque;
548 if (s->type == FTYPE_CD) {
549 raw_s->open_flags |= O_NONBLOCK;
552 raw_parse_flags(state->flags, &raw_s->open_flags);
556 int fcntl_flags = O_APPEND | O_NONBLOCK;
558 fcntl_flags |= O_NOATIME;
562 /* Not all operating systems have O_ASYNC, and those that don't
563 * will not let us track the state into raw_s->open_flags (typically
564 * you achieve the same effect with an ioctl, for example I_SETSIG
565 * on Solaris). But we do not use O_ASYNC, so that's fine.
567 assert((s->open_flags & O_ASYNC) == 0);
570 if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
571 /* dup the original fd */
572 raw_s->fd = qemu_dup(s->fd);
573 if (raw_s->fd >= 0) {
574 ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
576 qemu_close(raw_s->fd);
582 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
583 if (raw_s->fd == -1) {
584 const char *normalized_filename = state->bs->filename;
585 ret = raw_normalize_devicepath(&normalized_filename);
587 error_setg_errno(errp, -ret, "Could not normalize device path");
589 assert(!(raw_s->open_flags & O_CREAT));
590 raw_s->fd = qemu_open(normalized_filename, raw_s->open_flags);
591 if (raw_s->fd == -1) {
592 error_setg_errno(errp, errno, "Could not reopen file");
598 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
599 * alignment with the new fd. */
600 if (raw_s->fd != -1) {
601 raw_probe_alignment(state->bs, raw_s->fd, &local_err);
603 qemu_close(raw_s->fd);
605 error_propagate(errp, local_err);
613 static void raw_reopen_commit(BDRVReopenState *state)
615 BDRVRawReopenState *raw_s = state->opaque;
616 BDRVRawState *s = state->bs->opaque;
618 s->open_flags = raw_s->open_flags;
623 g_free(state->opaque);
624 state->opaque = NULL;
628 static void raw_reopen_abort(BDRVReopenState *state)
630 BDRVRawReopenState *raw_s = state->opaque;
632 /* nothing to do if NULL, we didn't get far enough */
637 if (raw_s->fd >= 0) {
638 qemu_close(raw_s->fd);
641 g_free(state->opaque);
642 state->opaque = NULL;
645 static int hdev_get_max_transfer_length(int fd)
649 if (ioctl(fd, BLKSECTGET, &max_sectors) == 0) {
659 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
661 BDRVRawState *s = bs->opaque;
664 if (!fstat(s->fd, &st)) {
665 if (S_ISBLK(st.st_mode)) {
666 int ret = hdev_get_max_transfer_length(s->fd);
667 if (ret > 0 && ret <= BDRV_REQUEST_MAX_SECTORS) {
668 bs->bl.max_transfer = pow2floor(ret << BDRV_SECTOR_BITS);
673 raw_probe_alignment(bs, s->fd, errp);
674 bs->bl.min_mem_alignment = s->buf_align;
675 bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
678 static int check_for_dasd(int fd)
681 struct dasd_information2_t info = {0};
683 return ioctl(fd, BIODASDINFO2, &info);
690 * Try to get @bs's logical and physical block size.
691 * On success, store them in @bsz and return zero.
692 * On failure, return negative errno.
694 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
696 BDRVRawState *s = bs->opaque;
699 /* If DASD, get blocksizes */
700 if (check_for_dasd(s->fd) < 0) {
703 ret = probe_logical_blocksize(s->fd, &bsz->log);
707 return probe_physical_blocksize(s->fd, &bsz->phys);
711 * Try to get @bs's geometry: cyls, heads, sectors.
712 * On success, store them in @geo and return 0.
713 * On failure return -errno.
714 * (Allows block driver to assign default geometry values that guest sees)
717 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
719 BDRVRawState *s = bs->opaque;
720 struct hd_geometry ioctl_geo = {0};
722 /* If DASD, get its geometry */
723 if (check_for_dasd(s->fd) < 0) {
726 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
729 /* HDIO_GETGEO may return success even though geo contains zeros
730 (e.g. certain multipath setups) */
731 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
734 /* Do not return a geometry for partition */
735 if (ioctl_geo.start != 0) {
738 geo->heads = ioctl_geo.heads;
739 geo->sectors = ioctl_geo.sectors;
740 geo->cylinders = ioctl_geo.cylinders;
744 #else /* __linux__ */
745 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
751 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
755 ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
763 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
767 ret = qemu_fdatasync(aiocb->aio_fildes);
776 static bool preadv_present = true;
779 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
781 return preadv(fd, iov, nr_iov, offset);
785 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
787 return pwritev(fd, iov, nr_iov, offset);
792 static bool preadv_present = false;
795 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
801 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
808 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
813 if (aiocb->aio_type & QEMU_AIO_WRITE)
814 len = qemu_pwritev(aiocb->aio_fildes,
819 len = qemu_preadv(aiocb->aio_fildes,
823 } while (len == -1 && errno == EINTR);
832 * Read/writes the data to/from a given linear buffer.
834 * Returns the number of bytes handles or -errno in case of an error. Short
835 * reads are only returned if the end of the file is reached.
837 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
842 while (offset < aiocb->aio_nbytes) {
843 if (aiocb->aio_type & QEMU_AIO_WRITE) {
844 len = pwrite(aiocb->aio_fildes,
845 (const char *)buf + offset,
846 aiocb->aio_nbytes - offset,
847 aiocb->aio_offset + offset);
849 len = pread(aiocb->aio_fildes,
851 aiocb->aio_nbytes - offset,
852 aiocb->aio_offset + offset);
854 if (len == -1 && errno == EINTR) {
856 } else if (len == -1 && errno == EINVAL &&
857 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
858 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
860 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
861 * after a short read. Assume that O_DIRECT short reads only occur
862 * at EOF. Therefore this is a short read, not an I/O error.
865 } else if (len == -1) {
868 } else if (len == 0) {
877 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
882 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
884 * If there is just a single buffer, and it is properly aligned
885 * we can just use plain pread/pwrite without any problems.
887 if (aiocb->aio_niov == 1) {
888 return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
891 * We have more than one iovec, and all are properly aligned.
893 * Try preadv/pwritev first and fall back to linearizing the
894 * buffer if it's not supported.
896 if (preadv_present) {
897 nbytes = handle_aiocb_rw_vector(aiocb);
898 if (nbytes == aiocb->aio_nbytes ||
899 (nbytes < 0 && nbytes != -ENOSYS)) {
902 preadv_present = false;
906 * XXX(hch): short read/write. no easy way to handle the reminder
907 * using these interfaces. For now retry using plain
913 * Ok, we have to do it the hard way, copy all segments into
914 * a single aligned buffer.
916 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
921 if (aiocb->aio_type & QEMU_AIO_WRITE) {
925 for (i = 0; i < aiocb->aio_niov; ++i) {
926 memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
927 p += aiocb->aio_iov[i].iov_len;
929 assert(p - buf == aiocb->aio_nbytes);
932 nbytes = handle_aiocb_rw_linear(aiocb, buf);
933 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
935 size_t count = aiocb->aio_nbytes, copy;
938 for (i = 0; i < aiocb->aio_niov && count; ++i) {
940 if (copy > aiocb->aio_iov[i].iov_len) {
941 copy = aiocb->aio_iov[i].iov_len;
943 memcpy(aiocb->aio_iov[i].iov_base, p, copy);
944 assert(count >= copy);
956 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
958 struct xfs_flock64 fl;
961 memset(&fl, 0, sizeof(fl));
962 fl.l_whence = SEEK_SET;
966 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
968 DPRINTF("cannot write zero range (%s)\n", strerror(errno));
975 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
977 struct xfs_flock64 fl;
980 memset(&fl, 0, sizeof(fl));
981 fl.l_whence = SEEK_SET;
985 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
987 DPRINTF("cannot punch hole (%s)\n", strerror(errno));
995 static int translate_err(int err)
997 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1004 #ifdef CONFIG_FALLOCATE
1005 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1008 if (fallocate(fd, mode, offset, len) == 0) {
1011 } while (errno == EINTR);
1012 return translate_err(-errno);
1016 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1019 BDRVRawState *s = aiocb->bs->opaque;
1021 if (!s->has_write_zeroes) {
1027 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1028 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1031 } while (errno == EINTR);
1033 ret = translate_err(-errno);
1036 if (ret == -ENOTSUP) {
1037 s->has_write_zeroes = false;
1042 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
1044 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1045 BDRVRawState *s = aiocb->bs->opaque;
1048 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1049 return handle_aiocb_write_zeroes_block(aiocb);
1054 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
1058 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1059 if (s->has_write_zeroes) {
1060 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1061 aiocb->aio_offset, aiocb->aio_nbytes);
1062 if (ret == 0 || ret != -ENOTSUP) {
1065 s->has_write_zeroes = false;
1069 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1070 if (s->has_discard && s->has_fallocate) {
1071 int ret = do_fallocate(s->fd,
1072 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1073 aiocb->aio_offset, aiocb->aio_nbytes);
1075 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1076 if (ret == 0 || ret != -ENOTSUP) {
1079 s->has_fallocate = false;
1080 } else if (ret != -ENOTSUP) {
1083 s->has_discard = false;
1088 #ifdef CONFIG_FALLOCATE
1089 if (s->has_fallocate && aiocb->aio_offset >= bdrv_getlength(aiocb->bs)) {
1090 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1091 if (ret == 0 || ret != -ENOTSUP) {
1094 s->has_fallocate = false;
1101 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
1103 int ret = -EOPNOTSUPP;
1104 BDRVRawState *s = aiocb->bs->opaque;
1106 if (!s->has_discard) {
1110 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1113 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1114 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1117 } while (errno == EINTR);
1124 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1128 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1129 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1130 aiocb->aio_offset, aiocb->aio_nbytes);
1134 ret = translate_err(ret);
1135 if (ret == -ENOTSUP) {
1136 s->has_discard = false;
1141 static int aio_worker(void *arg)
1143 RawPosixAIOData *aiocb = arg;
1146 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
1148 ret = handle_aiocb_rw(aiocb);
1149 if (ret >= 0 && ret < aiocb->aio_nbytes) {
1150 iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
1151 0, aiocb->aio_nbytes - ret);
1153 ret = aiocb->aio_nbytes;
1155 if (ret == aiocb->aio_nbytes) {
1157 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1161 case QEMU_AIO_WRITE:
1162 ret = handle_aiocb_rw(aiocb);
1163 if (ret == aiocb->aio_nbytes) {
1165 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1169 case QEMU_AIO_FLUSH:
1170 ret = handle_aiocb_flush(aiocb);
1172 case QEMU_AIO_IOCTL:
1173 ret = handle_aiocb_ioctl(aiocb);
1175 case QEMU_AIO_DISCARD:
1176 ret = handle_aiocb_discard(aiocb);
1178 case QEMU_AIO_WRITE_ZEROES:
1179 ret = handle_aiocb_write_zeroes(aiocb);
1182 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1191 static int paio_submit_co(BlockDriverState *bs, int fd,
1192 int64_t offset, QEMUIOVector *qiov,
1193 int count, int type)
1195 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1199 acb->aio_type = type;
1200 acb->aio_fildes = fd;
1202 acb->aio_nbytes = count;
1203 acb->aio_offset = offset;
1206 acb->aio_iov = qiov->iov;
1207 acb->aio_niov = qiov->niov;
1208 assert(qiov->size == count);
1211 trace_paio_submit_co(offset, count, type);
1212 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1213 return thread_pool_submit_co(pool, aio_worker, acb);
1216 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
1217 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1218 BlockCompletionFunc *cb, void *opaque, int type)
1220 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1224 acb->aio_type = type;
1225 acb->aio_fildes = fd;
1227 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1228 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1231 acb->aio_iov = qiov->iov;
1232 acb->aio_niov = qiov->niov;
1233 assert(qiov->size == acb->aio_nbytes);
1236 trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
1237 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1238 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1241 static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset,
1242 uint64_t bytes, QEMUIOVector *qiov, int type)
1244 BDRVRawState *s = bs->opaque;
1246 if (fd_open(bs) < 0)
1250 * Check if the underlying device requires requests to be aligned,
1251 * and if the request we are trying to submit is aligned or not.
1252 * If this is the case tell the low-level driver that it needs
1253 * to copy the buffer.
1255 if (s->needs_alignment) {
1256 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1257 type |= QEMU_AIO_MISALIGNED;
1258 #ifdef CONFIG_LINUX_AIO
1259 } else if (bs->open_flags & BDRV_O_NATIVE_AIO) {
1260 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1261 assert(qiov->size == bytes);
1262 return laio_co_submit(bs, aio, s->fd, offset, qiov, type);
1267 return paio_submit_co(bs, s->fd, offset, qiov, bytes, type);
1270 static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset,
1271 uint64_t bytes, QEMUIOVector *qiov,
1274 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ);
1277 static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
1278 uint64_t bytes, QEMUIOVector *qiov,
1282 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE);
1285 static void raw_aio_plug(BlockDriverState *bs)
1287 #ifdef CONFIG_LINUX_AIO
1288 if (bs->open_flags & BDRV_O_NATIVE_AIO) {
1289 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1290 laio_io_plug(bs, aio);
1295 static void raw_aio_unplug(BlockDriverState *bs)
1297 #ifdef CONFIG_LINUX_AIO
1298 if (bs->open_flags & BDRV_O_NATIVE_AIO) {
1299 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1300 laio_io_unplug(bs, aio);
1305 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
1306 BlockCompletionFunc *cb, void *opaque)
1308 BDRVRawState *s = bs->opaque;
1310 if (fd_open(bs) < 0)
1313 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1316 static void raw_close(BlockDriverState *bs)
1318 BDRVRawState *s = bs->opaque;
1326 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1328 BDRVRawState *s = bs->opaque;
1331 if (fstat(s->fd, &st)) {
1335 if (S_ISREG(st.st_mode)) {
1336 if (ftruncate(s->fd, offset) < 0) {
1339 } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1340 if (offset > raw_getlength(bs)) {
1351 static int64_t raw_getlength(BlockDriverState *bs)
1353 BDRVRawState *s = bs->opaque;
1359 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1360 struct disklabel dl;
1362 if (ioctl(fd, DIOCGDINFO, &dl))
1364 return (uint64_t)dl.d_secsize *
1365 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1369 #elif defined(__NetBSD__)
1370 static int64_t raw_getlength(BlockDriverState *bs)
1372 BDRVRawState *s = bs->opaque;
1378 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1379 struct dkwedge_info dkw;
1381 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1382 return dkw.dkw_size * 512;
1384 struct disklabel dl;
1386 if (ioctl(fd, DIOCGDINFO, &dl))
1388 return (uint64_t)dl.d_secsize *
1389 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1394 #elif defined(__sun__)
1395 static int64_t raw_getlength(BlockDriverState *bs)
1397 BDRVRawState *s = bs->opaque;
1398 struct dk_minfo minfo;
1408 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1410 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1412 return minfo.dki_lbsize * minfo.dki_capacity;
1416 * There are reports that lseek on some devices fails, but
1417 * irc discussion said that contingency on contingency was overkill.
1419 size = lseek(s->fd, 0, SEEK_END);
1425 #elif defined(CONFIG_BSD)
1426 static int64_t raw_getlength(BlockDriverState *bs)
1428 BDRVRawState *s = bs->opaque;
1432 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1441 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1444 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1445 #ifdef DIOCGMEDIASIZE
1446 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1447 #elif defined(DIOCGPART)
1450 if (ioctl(fd, DIOCGPART, &pi) == 0)
1451 size = pi.media_size;
1457 #if defined(__APPLE__) && defined(__MACH__)
1459 uint64_t sectors = 0;
1460 uint32_t sector_size = 0;
1462 if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0
1463 && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) {
1464 size = sectors * sector_size;
1466 size = lseek(fd, 0LL, SEEK_END);
1473 size = lseek(fd, 0LL, SEEK_END);
1478 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1481 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1482 if (size == 2048LL * (unsigned)-1)
1484 /* XXX no disc? maybe we need to reopen... */
1485 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1492 size = lseek(fd, 0, SEEK_END);
1500 static int64_t raw_getlength(BlockDriverState *bs)
1502 BDRVRawState *s = bs->opaque;
1511 size = lseek(s->fd, 0, SEEK_END);
1519 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1522 BDRVRawState *s = bs->opaque;
1524 if (fstat(s->fd, &st) < 0) {
1527 return (int64_t)st.st_blocks * 512;
1530 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
1534 int64_t total_size = 0;
1536 PreallocMode prealloc;
1538 Error *local_err = NULL;
1540 strstart(filename, "file:", &filename);
1542 /* Read out options */
1543 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1545 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
1546 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1547 prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
1548 PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
1552 error_propagate(errp, local_err);
1557 fd = qemu_open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY,
1561 error_setg_errno(errp, -result, "Could not create file");
1567 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1568 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1569 * will be ignored since any failure of this operation should not
1570 * block the left work.
1573 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
1574 attr |= FS_NOCOW_FL;
1575 ioctl(fd, FS_IOC_SETFLAGS, &attr);
1580 if (ftruncate(fd, total_size) != 0) {
1582 error_setg_errno(errp, -result, "Could not resize file");
1587 #ifdef CONFIG_POSIX_FALLOCATE
1588 case PREALLOC_MODE_FALLOC:
1589 /* posix_fallocate() doesn't set errno. */
1590 result = -posix_fallocate(fd, 0, total_size);
1592 error_setg_errno(errp, -result,
1593 "Could not preallocate data for the new file");
1597 case PREALLOC_MODE_FULL:
1599 int64_t num = 0, left = total_size;
1600 buf = g_malloc0(65536);
1603 num = MIN(left, 65536);
1604 result = write(fd, buf, num);
1607 error_setg_errno(errp, -result,
1608 "Could not write to the new file");
1617 error_setg_errno(errp, -result,
1618 "Could not flush new file to disk");
1624 case PREALLOC_MODE_OFF:
1628 error_setg(errp, "Unsupported preallocation mode: %s",
1629 PreallocMode_lookup[prealloc]);
1634 if (qemu_close(fd) != 0 && result == 0) {
1636 error_setg_errno(errp, -result, "Could not close the new file");
1643 * Find allocation range in @bs around offset @start.
1644 * May change underlying file descriptor's file offset.
1645 * If @start is not in a hole, store @start in @data, and the
1646 * beginning of the next hole in @hole, and return 0.
1647 * If @start is in a non-trailing hole, store @start in @hole and the
1648 * beginning of the next non-hole in @data, and return 0.
1649 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1650 * If we can't find out, return a negative errno other than -ENXIO.
1652 static int find_allocation(BlockDriverState *bs, off_t start,
1653 off_t *data, off_t *hole)
1655 #if defined SEEK_HOLE && defined SEEK_DATA
1656 BDRVRawState *s = bs->opaque;
1661 * D1. offs == start: start is in data
1662 * D2. offs > start: start is in a hole, next data at offs
1663 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1664 * or start is beyond EOF
1665 * If the latter happens, the file has been truncated behind
1666 * our back since we opened it. All bets are off then.
1667 * Treating like a trailing hole is simplest.
1668 * D4. offs < 0, errno != ENXIO: we learned nothing
1670 offs = lseek(s->fd, start, SEEK_DATA);
1672 return -errno; /* D3 or D4 */
1674 assert(offs >= start);
1677 /* D2: in hole, next data at offs */
1683 /* D1: in data, end not yet known */
1687 * H1. offs == start: start is in a hole
1688 * If this happens here, a hole has been dug behind our back
1689 * since the previous lseek().
1690 * H2. offs > start: either start is in data, next hole at offs,
1691 * or start is in trailing hole, EOF at offs
1692 * Linux treats trailing holes like any other hole: offs ==
1693 * start. Solaris seeks to EOF instead: offs > start (blech).
1694 * If that happens here, a hole has been dug behind our back
1695 * since the previous lseek().
1696 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1697 * If this happens, the file has been truncated behind our
1698 * back since we opened it. Treat it like a trailing hole.
1699 * H4. offs < 0, errno != ENXIO: we learned nothing
1700 * Pretend we know nothing at all, i.e. "forget" about D1.
1702 offs = lseek(s->fd, start, SEEK_HOLE);
1704 return -errno; /* D1 and (H3 or H4) */
1706 assert(offs >= start);
1710 * D1 and H2: either in data, next hole at offs, or it was in
1711 * data but is now in a trailing hole. In the latter case,
1712 * all bets are off. Treating it as if it there was data all
1713 * the way to EOF is safe, so simply do that.
1728 * Returns the allocation status of the specified sectors.
1730 * If 'sector_num' is beyond the end of the disk image the return value is 0
1731 * and 'pnum' is set to 0.
1733 * 'pnum' is set to the number of sectors (including and immediately following
1734 * the specified sector) that are known to be in the same
1735 * allocated/unallocated state.
1737 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1738 * beyond the end of the disk image it will be clamped.
1740 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1742 int nb_sectors, int *pnum,
1743 BlockDriverState **file)
1745 off_t start, data = 0, hole = 0;
1754 start = sector_num * BDRV_SECTOR_SIZE;
1755 total_size = bdrv_getlength(bs);
1756 if (total_size < 0) {
1758 } else if (start >= total_size) {
1761 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1762 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1765 ret = find_allocation(bs, start, &data, &hole);
1766 if (ret == -ENXIO) {
1769 ret = BDRV_BLOCK_ZERO;
1770 } else if (ret < 0) {
1771 /* No info available, so pretend there are no holes */
1773 ret = BDRV_BLOCK_DATA;
1774 } else if (data == start) {
1775 /* On a data extent, compute sectors to the end of the extent,
1776 * possibly including a partial sector at EOF. */
1777 *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1778 ret = BDRV_BLOCK_DATA;
1780 /* On a hole, compute sectors to the beginning of the next extent. */
1781 assert(hole == start);
1782 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1783 ret = BDRV_BLOCK_ZERO;
1786 return ret | BDRV_BLOCK_OFFSET_VALID | start;
1789 static coroutine_fn BlockAIOCB *raw_aio_discard(BlockDriverState *bs,
1790 int64_t sector_num, int nb_sectors,
1791 BlockCompletionFunc *cb, void *opaque)
1793 BDRVRawState *s = bs->opaque;
1795 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1796 cb, opaque, QEMU_AIO_DISCARD);
1799 static int coroutine_fn raw_co_pwrite_zeroes(
1800 BlockDriverState *bs, int64_t offset,
1801 int count, BdrvRequestFlags flags)
1803 BDRVRawState *s = bs->opaque;
1805 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1806 return paio_submit_co(bs, s->fd, offset, NULL, count,
1807 QEMU_AIO_WRITE_ZEROES);
1808 } else if (s->discard_zeroes) {
1809 return paio_submit_co(bs, s->fd, offset, NULL, count,
1815 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1817 BDRVRawState *s = bs->opaque;
1819 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1820 bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1824 static QemuOptsList raw_create_opts = {
1825 .name = "raw-create-opts",
1826 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
1829 .name = BLOCK_OPT_SIZE,
1830 .type = QEMU_OPT_SIZE,
1831 .help = "Virtual disk size"
1834 .name = BLOCK_OPT_NOCOW,
1835 .type = QEMU_OPT_BOOL,
1836 .help = "Turn off copy-on-write (valid only on btrfs)"
1839 .name = BLOCK_OPT_PREALLOC,
1840 .type = QEMU_OPT_STRING,
1841 .help = "Preallocation mode (allowed values: off, falloc, full)"
1843 { /* end of list */ }
1847 BlockDriver bdrv_file = {
1848 .format_name = "file",
1849 .protocol_name = "file",
1850 .instance_size = sizeof(BDRVRawState),
1851 .bdrv_needs_filename = true,
1852 .bdrv_probe = NULL, /* no probe for protocols */
1853 .bdrv_parse_filename = raw_parse_filename,
1854 .bdrv_file_open = raw_open,
1855 .bdrv_reopen_prepare = raw_reopen_prepare,
1856 .bdrv_reopen_commit = raw_reopen_commit,
1857 .bdrv_reopen_abort = raw_reopen_abort,
1858 .bdrv_close = raw_close,
1859 .bdrv_create = raw_create,
1860 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1861 .bdrv_co_get_block_status = raw_co_get_block_status,
1862 .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes,
1864 .bdrv_co_preadv = raw_co_preadv,
1865 .bdrv_co_pwritev = raw_co_pwritev,
1866 .bdrv_aio_flush = raw_aio_flush,
1867 .bdrv_aio_discard = raw_aio_discard,
1868 .bdrv_refresh_limits = raw_refresh_limits,
1869 .bdrv_io_plug = raw_aio_plug,
1870 .bdrv_io_unplug = raw_aio_unplug,
1872 .bdrv_truncate = raw_truncate,
1873 .bdrv_getlength = raw_getlength,
1874 .bdrv_get_info = raw_get_info,
1875 .bdrv_get_allocated_file_size
1876 = raw_get_allocated_file_size,
1878 .create_opts = &raw_create_opts,
1881 /***********************************************/
1884 #if defined(__APPLE__) && defined(__MACH__)
1885 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
1886 CFIndex maxPathSize, int flags);
1887 static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
1889 kern_return_t kernResult = KERN_FAILURE;
1890 mach_port_t masterPort;
1891 CFMutableDictionaryRef classesToMatch;
1892 const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
1893 char *mediaType = NULL;
1895 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1896 if ( KERN_SUCCESS != kernResult ) {
1897 printf( "IOMasterPort returned %d\n", kernResult );
1901 for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
1902 classesToMatch = IOServiceMatching(matching_array[index]);
1903 if (classesToMatch == NULL) {
1904 error_report("IOServiceMatching returned NULL for %s",
1905 matching_array[index]);
1908 CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
1910 kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch,
1912 if (kernResult != KERN_SUCCESS) {
1913 error_report("Note: IOServiceGetMatchingServices returned %d",
1918 /* If a match was found, leave the loop */
1919 if (*mediaIterator != 0) {
1920 DPRINTF("Matching using %s\n", matching_array[index]);
1921 mediaType = g_strdup(matching_array[index]);
1928 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
1929 CFIndex maxPathSize, int flags)
1931 io_object_t nextMedia;
1932 kern_return_t kernResult = KERN_FAILURE;
1934 nextMedia = IOIteratorNext( mediaIterator );
1937 CFTypeRef bsdPathAsCFString;
1938 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1939 if ( bsdPathAsCFString ) {
1940 size_t devPathLength;
1941 strcpy( bsdPath, _PATH_DEV );
1942 if (flags & BDRV_O_NOCACHE) {
1943 strcat(bsdPath, "r");
1945 devPathLength = strlen( bsdPath );
1946 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1947 kernResult = KERN_SUCCESS;
1949 CFRelease( bsdPathAsCFString );
1951 IOObjectRelease( nextMedia );
1957 /* Sets up a real cdrom for use in QEMU */
1958 static bool setup_cdrom(char *bsd_path, Error **errp)
1960 int index, num_of_test_partitions = 2, fd;
1961 char test_partition[MAXPATHLEN];
1962 bool partition_found = false;
1964 /* look for a working partition */
1965 for (index = 0; index < num_of_test_partitions; index++) {
1966 snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
1968 fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE);
1970 partition_found = true;
1976 /* if a working partition on the device was not found */
1977 if (partition_found == false) {
1978 error_setg(errp, "Failed to find a working partition on disc");
1980 DPRINTF("Using %s as optical disc\n", test_partition);
1981 pstrcpy(bsd_path, MAXPATHLEN, test_partition);
1983 return partition_found;
1986 /* Prints directions on mounting and unmounting a device */
1987 static void print_unmounting_directions(const char *file_name)
1989 error_report("If device %s is mounted on the desktop, unmount"
1990 " it first before using it in QEMU", file_name);
1991 error_report("Command to unmount device: diskutil unmountDisk %s",
1993 error_report("Command to mount device: diskutil mountDisk %s", file_name);
1996 #endif /* defined(__APPLE__) && defined(__MACH__) */
1998 static int hdev_probe_device(const char *filename)
2002 /* allow a dedicated CD-ROM driver to match with a higher priority */
2003 if (strstart(filename, "/dev/cdrom", NULL))
2006 if (stat(filename, &st) >= 0 &&
2007 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
2014 static int check_hdev_writable(BDRVRawState *s)
2016 #if defined(BLKROGET)
2017 /* Linux block devices can be configured "read-only" using blockdev(8).
2018 * This is independent of device node permissions and therefore open(2)
2019 * with O_RDWR succeeds. Actual writes fail with EPERM.
2021 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2022 * check for read-only block devices so that Linux block devices behave
2028 if (fstat(s->fd, &st)) {
2032 if (!S_ISBLK(st.st_mode)) {
2036 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
2043 #endif /* defined(BLKROGET) */
2047 static void hdev_parse_filename(const char *filename, QDict *options,
2050 /* The prefix is optional, just as for "file". */
2051 strstart(filename, "host_device:", &filename);
2053 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2056 static bool hdev_is_sg(BlockDriverState *bs)
2059 #if defined(__linux__)
2062 struct sg_scsi_id scsiid;
2065 if (stat(bs->filename, &st) >= 0 && S_ISCHR(st.st_mode) &&
2066 !bdrv_ioctl(bs, SG_GET_VERSION_NUM, &sg_version) &&
2067 !bdrv_ioctl(bs, SG_GET_SCSI_ID, &scsiid)) {
2068 DPRINTF("SG device found: type=%d, version=%d\n",
2069 scsiid.scsi_type, sg_version);
2078 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
2081 BDRVRawState *s = bs->opaque;
2082 Error *local_err = NULL;
2085 #if defined(__APPLE__) && defined(__MACH__)
2086 const char *filename = qdict_get_str(options, "filename");
2087 char bsd_path[MAXPATHLEN] = "";
2088 bool error_occurred = false;
2090 /* If using a real cdrom */
2091 if (strcmp(filename, "/dev/cdrom") == 0) {
2092 char *mediaType = NULL;
2093 kern_return_t ret_val;
2094 io_iterator_t mediaIterator = 0;
2096 mediaType = FindEjectableOpticalMedia(&mediaIterator);
2097 if (mediaType == NULL) {
2098 error_setg(errp, "Please make sure your CD/DVD is in the optical"
2100 error_occurred = true;
2101 goto hdev_open_Mac_error;
2104 ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
2105 if (ret_val != KERN_SUCCESS) {
2106 error_setg(errp, "Could not get BSD path for optical drive");
2107 error_occurred = true;
2108 goto hdev_open_Mac_error;
2111 /* If a real optical drive was not found */
2112 if (bsd_path[0] == '\0') {
2113 error_setg(errp, "Failed to obtain bsd path for optical drive");
2114 error_occurred = true;
2115 goto hdev_open_Mac_error;
2118 /* If using a cdrom disc and finding a partition on the disc failed */
2119 if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
2120 setup_cdrom(bsd_path, errp) == false) {
2121 print_unmounting_directions(bsd_path);
2122 error_occurred = true;
2123 goto hdev_open_Mac_error;
2126 qdict_put(options, "filename", qstring_from_str(bsd_path));
2128 hdev_open_Mac_error:
2130 if (mediaIterator) {
2131 IOObjectRelease(mediaIterator);
2133 if (error_occurred) {
2137 #endif /* defined(__APPLE__) && defined(__MACH__) */
2139 s->type = FTYPE_FILE;
2141 ret = raw_open_common(bs, options, flags, 0, &local_err);
2143 error_propagate(errp, local_err);
2144 #if defined(__APPLE__) && defined(__MACH__)
2146 filename = bsd_path;
2148 /* if a physical device experienced an error while being opened */
2149 if (strncmp(filename, "/dev/", 5) == 0) {
2150 print_unmounting_directions(filename);
2152 #endif /* defined(__APPLE__) && defined(__MACH__) */
2156 /* Since this does ioctl the device must be already opened */
2157 bs->sg = hdev_is_sg(bs);
2159 if (flags & BDRV_O_RDWR) {
2160 ret = check_hdev_writable(s);
2163 error_setg_errno(errp, -ret, "The device is not writable");
2171 #if defined(__linux__)
2173 static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
2174 unsigned long int req, void *buf,
2175 BlockCompletionFunc *cb, void *opaque)
2177 BDRVRawState *s = bs->opaque;
2178 RawPosixAIOData *acb;
2181 if (fd_open(bs) < 0)
2184 acb = g_new(RawPosixAIOData, 1);
2186 acb->aio_type = QEMU_AIO_IOCTL;
2187 acb->aio_fildes = s->fd;
2188 acb->aio_offset = 0;
2189 acb->aio_ioctl_buf = buf;
2190 acb->aio_ioctl_cmd = req;
2191 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
2192 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
2196 static int fd_open(BlockDriverState *bs)
2198 BDRVRawState *s = bs->opaque;
2200 /* this is just to ensure s->fd is sane (its called by io ops) */
2206 static coroutine_fn BlockAIOCB *hdev_aio_discard(BlockDriverState *bs,
2207 int64_t sector_num, int nb_sectors,
2208 BlockCompletionFunc *cb, void *opaque)
2210 BDRVRawState *s = bs->opaque;
2212 if (fd_open(bs) < 0) {
2215 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
2216 cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2219 static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs,
2220 int64_t offset, int count, BdrvRequestFlags flags)
2222 BDRVRawState *s = bs->opaque;
2229 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2230 return paio_submit_co(bs, s->fd, offset, NULL, count,
2231 QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2232 } else if (s->discard_zeroes) {
2233 return paio_submit_co(bs, s->fd, offset, NULL, count,
2234 QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2239 static int hdev_create(const char *filename, QemuOpts *opts,
2244 struct stat stat_buf;
2245 int64_t total_size = 0;
2248 /* This function is used by both protocol block drivers and therefore either
2249 * of these prefixes may be given.
2250 * The return value has to be stored somewhere, otherwise this is an error
2251 * due to -Werror=unused-value. */
2253 strstart(filename, "host_device:", &filename) ||
2254 strstart(filename, "host_cdrom:" , &filename);
2258 ret = raw_normalize_devicepath(&filename);
2260 error_setg_errno(errp, -ret, "Could not normalize device path");
2264 /* Read out options */
2265 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2268 fd = qemu_open(filename, O_WRONLY | O_BINARY);
2271 error_setg_errno(errp, -ret, "Could not open device");
2275 if (fstat(fd, &stat_buf) < 0) {
2277 error_setg_errno(errp, -ret, "Could not stat device");
2278 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2280 "The given file is neither a block nor a character device");
2282 } else if (lseek(fd, 0, SEEK_END) < total_size) {
2283 error_setg(errp, "Device is too small");
2291 static BlockDriver bdrv_host_device = {
2292 .format_name = "host_device",
2293 .protocol_name = "host_device",
2294 .instance_size = sizeof(BDRVRawState),
2295 .bdrv_needs_filename = true,
2296 .bdrv_probe_device = hdev_probe_device,
2297 .bdrv_parse_filename = hdev_parse_filename,
2298 .bdrv_file_open = hdev_open,
2299 .bdrv_close = raw_close,
2300 .bdrv_reopen_prepare = raw_reopen_prepare,
2301 .bdrv_reopen_commit = raw_reopen_commit,
2302 .bdrv_reopen_abort = raw_reopen_abort,
2303 .bdrv_create = hdev_create,
2304 .create_opts = &raw_create_opts,
2305 .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes,
2307 .bdrv_co_preadv = raw_co_preadv,
2308 .bdrv_co_pwritev = raw_co_pwritev,
2309 .bdrv_aio_flush = raw_aio_flush,
2310 .bdrv_aio_discard = hdev_aio_discard,
2311 .bdrv_refresh_limits = raw_refresh_limits,
2312 .bdrv_io_plug = raw_aio_plug,
2313 .bdrv_io_unplug = raw_aio_unplug,
2315 .bdrv_truncate = raw_truncate,
2316 .bdrv_getlength = raw_getlength,
2317 .bdrv_get_info = raw_get_info,
2318 .bdrv_get_allocated_file_size
2319 = raw_get_allocated_file_size,
2320 .bdrv_probe_blocksizes = hdev_probe_blocksizes,
2321 .bdrv_probe_geometry = hdev_probe_geometry,
2323 /* generic scsi device */
2325 .bdrv_aio_ioctl = hdev_aio_ioctl,
2329 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2330 static void cdrom_parse_filename(const char *filename, QDict *options,
2333 /* The prefix is optional, just as for "file". */
2334 strstart(filename, "host_cdrom:", &filename);
2336 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2341 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2344 BDRVRawState *s = bs->opaque;
2348 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2349 return raw_open_common(bs, options, flags, O_NONBLOCK, errp);
2352 static int cdrom_probe_device(const char *filename)
2358 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2362 ret = fstat(fd, &st);
2363 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2367 /* Attempt to detect via a CDROM specific ioctl */
2368 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2378 static bool cdrom_is_inserted(BlockDriverState *bs)
2380 BDRVRawState *s = bs->opaque;
2383 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2384 return ret == CDS_DISC_OK;
2387 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2389 BDRVRawState *s = bs->opaque;
2392 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2393 perror("CDROMEJECT");
2395 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2396 perror("CDROMEJECT");
2400 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2402 BDRVRawState *s = bs->opaque;
2404 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2406 * Note: an error can happen if the distribution automatically
2409 /* perror("CDROM_LOCKDOOR"); */
2413 static BlockDriver bdrv_host_cdrom = {
2414 .format_name = "host_cdrom",
2415 .protocol_name = "host_cdrom",
2416 .instance_size = sizeof(BDRVRawState),
2417 .bdrv_needs_filename = true,
2418 .bdrv_probe_device = cdrom_probe_device,
2419 .bdrv_parse_filename = cdrom_parse_filename,
2420 .bdrv_file_open = cdrom_open,
2421 .bdrv_close = raw_close,
2422 .bdrv_reopen_prepare = raw_reopen_prepare,
2423 .bdrv_reopen_commit = raw_reopen_commit,
2424 .bdrv_reopen_abort = raw_reopen_abort,
2425 .bdrv_create = hdev_create,
2426 .create_opts = &raw_create_opts,
2429 .bdrv_co_preadv = raw_co_preadv,
2430 .bdrv_co_pwritev = raw_co_pwritev,
2431 .bdrv_aio_flush = raw_aio_flush,
2432 .bdrv_refresh_limits = raw_refresh_limits,
2433 .bdrv_io_plug = raw_aio_plug,
2434 .bdrv_io_unplug = raw_aio_unplug,
2436 .bdrv_truncate = raw_truncate,
2437 .bdrv_getlength = raw_getlength,
2438 .has_variable_length = true,
2439 .bdrv_get_allocated_file_size
2440 = raw_get_allocated_file_size,
2442 /* removable device support */
2443 .bdrv_is_inserted = cdrom_is_inserted,
2444 .bdrv_eject = cdrom_eject,
2445 .bdrv_lock_medium = cdrom_lock_medium,
2447 /* generic scsi device */
2448 .bdrv_aio_ioctl = hdev_aio_ioctl,
2450 #endif /* __linux__ */
2452 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2453 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2456 BDRVRawState *s = bs->opaque;
2457 Error *local_err = NULL;
2462 ret = raw_open_common(bs, options, flags, 0, &local_err);
2464 error_propagate(errp, local_err);
2468 /* make sure the door isn't locked at this time */
2469 ioctl(s->fd, CDIOCALLOW);
2473 static int cdrom_probe_device(const char *filename)
2475 if (strstart(filename, "/dev/cd", NULL) ||
2476 strstart(filename, "/dev/acd", NULL))
2481 static int cdrom_reopen(BlockDriverState *bs)
2483 BDRVRawState *s = bs->opaque;
2487 * Force reread of possibly changed/newly loaded disc,
2488 * FreeBSD seems to not notice sometimes...
2492 fd = qemu_open(bs->filename, s->open_flags, 0644);
2499 /* make sure the door isn't locked at this time */
2500 ioctl(s->fd, CDIOCALLOW);
2504 static bool cdrom_is_inserted(BlockDriverState *bs)
2506 return raw_getlength(bs) > 0;
2509 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2511 BDRVRawState *s = bs->opaque;
2516 (void) ioctl(s->fd, CDIOCALLOW);
2519 if (ioctl(s->fd, CDIOCEJECT) < 0)
2520 perror("CDIOCEJECT");
2522 if (ioctl(s->fd, CDIOCCLOSE) < 0)
2523 perror("CDIOCCLOSE");
2529 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2531 BDRVRawState *s = bs->opaque;
2535 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2537 * Note: an error can happen if the distribution automatically
2540 /* perror("CDROM_LOCKDOOR"); */
2544 static BlockDriver bdrv_host_cdrom = {
2545 .format_name = "host_cdrom",
2546 .protocol_name = "host_cdrom",
2547 .instance_size = sizeof(BDRVRawState),
2548 .bdrv_needs_filename = true,
2549 .bdrv_probe_device = cdrom_probe_device,
2550 .bdrv_parse_filename = cdrom_parse_filename,
2551 .bdrv_file_open = cdrom_open,
2552 .bdrv_close = raw_close,
2553 .bdrv_reopen_prepare = raw_reopen_prepare,
2554 .bdrv_reopen_commit = raw_reopen_commit,
2555 .bdrv_reopen_abort = raw_reopen_abort,
2556 .bdrv_create = hdev_create,
2557 .create_opts = &raw_create_opts,
2559 .bdrv_co_preadv = raw_co_preadv,
2560 .bdrv_co_pwritev = raw_co_pwritev,
2561 .bdrv_aio_flush = raw_aio_flush,
2562 .bdrv_refresh_limits = raw_refresh_limits,
2563 .bdrv_io_plug = raw_aio_plug,
2564 .bdrv_io_unplug = raw_aio_unplug,
2566 .bdrv_truncate = raw_truncate,
2567 .bdrv_getlength = raw_getlength,
2568 .has_variable_length = true,
2569 .bdrv_get_allocated_file_size
2570 = raw_get_allocated_file_size,
2572 /* removable device support */
2573 .bdrv_is_inserted = cdrom_is_inserted,
2574 .bdrv_eject = cdrom_eject,
2575 .bdrv_lock_medium = cdrom_lock_medium,
2577 #endif /* __FreeBSD__ */
2579 static void bdrv_file_init(void)
2582 * Register all the drivers. Note that order is important, the driver
2583 * registered last will get probed first.
2585 bdrv_register(&bdrv_file);
2586 bdrv_register(&bdrv_host_device);
2588 bdrv_register(&bdrv_host_cdrom);
2590 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2591 bdrv_register(&bdrv_host_cdrom);
2595 block_init(bdrv_file_init);