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 "qemu-common.h"
26 #include "qemu/error-report.h"
27 #include "qemu/timer.h"
29 #include "block/block_int.h"
30 #include "qemu/module.h"
32 #include "block/thread-pool.h"
35 #include "qapi/util.h"
36 #include "qapi/qmp/qstring.h"
38 #if defined(__APPLE__) && (__MACH__)
40 #include <sys/param.h>
41 #include <IOKit/IOKitLib.h>
42 #include <IOKit/IOBSD.h>
43 #include <IOKit/storage/IOMediaBSDClient.h>
44 #include <IOKit/storage/IOMedia.h>
45 #include <IOKit/storage/IOCDMedia.h>
46 //#include <IOKit/storage/IOCDTypes.h>
47 #include <CoreFoundation/CoreFoundation.h>
51 #define _POSIX_PTHREAD_SEMANTICS 1
55 #include <sys/ioctl.h>
56 #include <sys/param.h>
57 #include <linux/cdrom.h>
60 #include <linux/hdreg.h>
66 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
69 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
70 #include <linux/falloc.h>
72 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
78 #include <sys/ioctl.h>
79 #include <sys/disklabel.h>
84 #include <sys/ioctl.h>
85 #include <sys/disklabel.h>
91 #include <sys/ioctl.h>
92 #include <sys/diskslice.h>
102 # define DEBUG_BLOCK_PRINT 1
104 # define DEBUG_BLOCK_PRINT 0
106 #define DPRINTF(fmt, ...) \
108 if (DEBUG_BLOCK_PRINT) { \
109 printf(fmt, ## __VA_ARGS__); \
113 /* OS X does not have O_DSYNC */
116 #define O_DSYNC O_SYNC
117 #elif defined(O_FSYNC)
118 #define O_DSYNC O_FSYNC
122 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
124 #define O_DIRECT O_DSYNC
130 #define MAX_BLOCKSIZE 4096
132 typedef struct BDRVRawState {
138 #ifdef CONFIG_LINUX_AIO
146 bool has_write_zeroes:1;
147 bool discard_zeroes:1;
149 bool needs_alignment;
152 typedef struct BDRVRawReopenState {
155 #ifdef CONFIG_LINUX_AIO
158 } BDRVRawReopenState;
160 static int fd_open(BlockDriverState *bs);
161 static int64_t raw_getlength(BlockDriverState *bs);
163 typedef struct RawPosixAIOData {
164 BlockDriverState *bs;
167 struct iovec *aio_iov;
172 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
177 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
178 static int cdrom_reopen(BlockDriverState *bs);
181 #if defined(__NetBSD__)
182 static int raw_normalize_devicepath(const char **filename)
184 static char namebuf[PATH_MAX];
185 const char *dp, *fname;
189 dp = strrchr(fname, '/');
190 if (lstat(fname, &sb) < 0) {
191 fprintf(stderr, "%s: stat failed: %s\n",
192 fname, strerror(errno));
196 if (!S_ISBLK(sb.st_mode)) {
201 snprintf(namebuf, PATH_MAX, "r%s", fname);
203 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
204 (int)(dp - fname), fname, dp + 1);
206 fprintf(stderr, "%s is a block device", fname);
208 fprintf(stderr, ", using %s\n", *filename);
213 static int raw_normalize_devicepath(const char **filename)
220 * Get logical block size via ioctl. On success store it in @sector_size_p.
222 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
224 unsigned int sector_size;
225 bool success = false;
229 /* Try a few ioctls to get the right size */
231 if (ioctl(fd, BLKSSZGET, §or_size) >= 0) {
232 *sector_size_p = sector_size;
236 #ifdef DKIOCGETBLOCKSIZE
237 if (ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) >= 0) {
238 *sector_size_p = sector_size;
242 #ifdef DIOCGSECTORSIZE
243 if (ioctl(fd, DIOCGSECTORSIZE, §or_size) >= 0) {
244 *sector_size_p = sector_size;
249 return success ? 0 : -errno;
253 * Get physical block size of @fd.
254 * On success, store it in @blk_size and return 0.
255 * On failure, return -errno.
257 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
260 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
269 /* Check if read is allowed with given memory buffer and length.
271 * This function is used to check O_DIRECT memory buffer and request alignment.
273 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
275 ssize_t ret = pread(fd, buf, len, 0);
282 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
283 * other errors (e.g. real I/O error), which could happen on a failed
284 * drive, since we only care about probing alignment.
286 if (errno != EINVAL) {
294 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
296 BDRVRawState *s = bs->opaque;
298 size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize());
300 /* For SCSI generic devices the alignment is not really used.
301 With buffered I/O, we don't have any restrictions. */
302 if (bdrv_is_sg(bs) || !s->needs_alignment) {
303 bs->request_alignment = 1;
308 bs->request_alignment = 0;
310 /* Let's try to use the logical blocksize for the alignment. */
311 if (probe_logical_blocksize(fd, &bs->request_alignment) < 0) {
312 bs->request_alignment = 0;
317 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
318 bs->request_alignment = da.d_miniosz;
319 /* The kernel returns wrong information for d_mem */
320 /* s->buf_align = da.d_mem; */
325 /* If we could not get the sizes so far, we can only guess them */
328 buf = qemu_memalign(max_align, 2 * max_align);
329 for (align = 512; align <= max_align; align <<= 1) {
330 if (raw_is_io_aligned(fd, buf + align, max_align)) {
331 s->buf_align = align;
338 if (!bs->request_alignment) {
340 buf = qemu_memalign(s->buf_align, max_align);
341 for (align = 512; align <= max_align; align <<= 1) {
342 if (raw_is_io_aligned(fd, buf, align)) {
343 bs->request_alignment = align;
350 if (!s->buf_align || !bs->request_alignment) {
351 error_setg(errp, "Could not find working O_DIRECT alignment. "
352 "Try cache.direct=off.");
356 static void raw_parse_flags(int bdrv_flags, int *open_flags)
358 assert(open_flags != NULL);
360 *open_flags |= O_BINARY;
361 *open_flags &= ~O_ACCMODE;
362 if (bdrv_flags & BDRV_O_RDWR) {
363 *open_flags |= O_RDWR;
365 *open_flags |= O_RDONLY;
368 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
369 * and O_DIRECT for no caching. */
370 if ((bdrv_flags & BDRV_O_NOCACHE)) {
371 *open_flags |= O_DIRECT;
375 static void raw_detach_aio_context(BlockDriverState *bs)
377 #ifdef CONFIG_LINUX_AIO
378 BDRVRawState *s = bs->opaque;
381 laio_detach_aio_context(s->aio_ctx, bdrv_get_aio_context(bs));
386 static void raw_attach_aio_context(BlockDriverState *bs,
387 AioContext *new_context)
389 #ifdef CONFIG_LINUX_AIO
390 BDRVRawState *s = bs->opaque;
393 laio_attach_aio_context(s->aio_ctx, new_context);
398 #ifdef CONFIG_LINUX_AIO
399 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
402 assert(aio_ctx != NULL);
403 assert(use_aio != NULL);
405 * Currently Linux do AIO only for files opened with O_DIRECT
406 * specified so check NOCACHE flag too
408 if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
409 (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
411 /* if non-NULL, laio_init() has already been run */
412 if (*aio_ctx == NULL) {
413 *aio_ctx = laio_init();
430 static void raw_parse_filename(const char *filename, QDict *options,
433 /* The filename does not have to be prefixed by the protocol name, since
434 * "file" is the default protocol; therefore, the return value of this
435 * function call can be ignored. */
436 strstart(filename, "file:", &filename);
438 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
441 static QemuOptsList raw_runtime_opts = {
443 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
447 .type = QEMU_OPT_STRING,
448 .help = "File name of the image",
450 { /* end of list */ }
454 static int raw_open_common(BlockDriverState *bs, QDict *options,
455 int bdrv_flags, int open_flags, Error **errp)
457 BDRVRawState *s = bs->opaque;
459 Error *local_err = NULL;
460 const char *filename = NULL;
464 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
465 qemu_opts_absorb_qdict(opts, options, &local_err);
467 error_propagate(errp, local_err);
472 filename = qemu_opt_get(opts, "filename");
474 ret = raw_normalize_devicepath(&filename);
476 error_setg_errno(errp, -ret, "Could not normalize device path");
480 s->open_flags = open_flags;
481 raw_parse_flags(bdrv_flags, &s->open_flags);
484 fd = qemu_open(filename, s->open_flags, 0644);
494 #ifdef CONFIG_LINUX_AIO
495 if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
498 error_setg_errno(errp, -ret, "Could not set AIO state");
501 if (!s->use_aio && (bdrv_flags & BDRV_O_NATIVE_AIO)) {
502 error_setg(errp, "aio=native was specified, but it requires "
503 "cache.direct=on, which was not specified.");
508 if (bdrv_flags & BDRV_O_NATIVE_AIO) {
509 error_setg(errp, "aio=native was specified, but is not supported "
514 #endif /* !defined(CONFIG_LINUX_AIO) */
516 s->has_discard = true;
517 s->has_write_zeroes = true;
518 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
519 s->needs_alignment = true;
522 if (fstat(s->fd, &st) < 0) {
524 error_setg_errno(errp, errno, "Could not stat file");
527 if (S_ISREG(st.st_mode)) {
528 s->discard_zeroes = true;
529 s->has_fallocate = true;
531 if (S_ISBLK(st.st_mode)) {
532 #ifdef BLKDISCARDZEROES
534 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
535 s->discard_zeroes = true;
539 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
540 * not rely on the contents of discarded blocks unless using O_DIRECT.
541 * Same for BLKZEROOUT.
543 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
544 s->discard_zeroes = false;
545 s->has_write_zeroes = false;
550 if (S_ISCHR(st.st_mode)) {
552 * The file is a char device (disk), which on FreeBSD isn't behind
553 * a pager, so force all requests to be aligned. This is needed
554 * so QEMU makes sure all IO operations on the device are aligned
555 * to sector size, or else FreeBSD will reject them with EINVAL.
557 s->needs_alignment = true;
562 if (platform_test_xfs_fd(s->fd)) {
567 raw_attach_aio_context(bs, bdrv_get_aio_context(bs));
571 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
578 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
581 BDRVRawState *s = bs->opaque;
582 Error *local_err = NULL;
585 s->type = FTYPE_FILE;
586 ret = raw_open_common(bs, options, flags, 0, &local_err);
588 error_propagate(errp, local_err);
593 static int raw_reopen_prepare(BDRVReopenState *state,
594 BlockReopenQueue *queue, Error **errp)
597 BDRVRawReopenState *raw_s;
599 Error *local_err = NULL;
601 assert(state != NULL);
602 assert(state->bs != NULL);
604 s = state->bs->opaque;
606 state->opaque = g_new0(BDRVRawReopenState, 1);
607 raw_s = state->opaque;
609 #ifdef CONFIG_LINUX_AIO
610 raw_s->use_aio = s->use_aio;
612 /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
613 * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
614 * won't override aio_ctx if aio_ctx is non-NULL */
615 if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
616 error_setg(errp, "Could not set AIO state");
621 if (s->type == FTYPE_CD) {
622 raw_s->open_flags |= O_NONBLOCK;
625 raw_parse_flags(state->flags, &raw_s->open_flags);
629 int fcntl_flags = O_APPEND | O_NONBLOCK;
631 fcntl_flags |= O_NOATIME;
635 /* Not all operating systems have O_ASYNC, and those that don't
636 * will not let us track the state into raw_s->open_flags (typically
637 * you achieve the same effect with an ioctl, for example I_SETSIG
638 * on Solaris). But we do not use O_ASYNC, so that's fine.
640 assert((s->open_flags & O_ASYNC) == 0);
643 if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
644 /* dup the original fd */
645 /* TODO: use qemu fcntl wrapper */
646 #ifdef F_DUPFD_CLOEXEC
647 raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
649 raw_s->fd = dup(s->fd);
650 if (raw_s->fd != -1) {
651 qemu_set_cloexec(raw_s->fd);
654 if (raw_s->fd >= 0) {
655 ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
657 qemu_close(raw_s->fd);
663 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
664 if (raw_s->fd == -1) {
665 const char *normalized_filename = state->bs->filename;
666 ret = raw_normalize_devicepath(&normalized_filename);
668 error_setg_errno(errp, -ret, "Could not normalize device path");
670 assert(!(raw_s->open_flags & O_CREAT));
671 raw_s->fd = qemu_open(normalized_filename, raw_s->open_flags);
672 if (raw_s->fd == -1) {
673 error_setg_errno(errp, errno, "Could not reopen file");
679 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
680 * alignment with the new fd. */
681 if (raw_s->fd != -1) {
682 raw_probe_alignment(state->bs, raw_s->fd, &local_err);
684 qemu_close(raw_s->fd);
686 error_propagate(errp, local_err);
694 static void raw_reopen_commit(BDRVReopenState *state)
696 BDRVRawReopenState *raw_s = state->opaque;
697 BDRVRawState *s = state->bs->opaque;
699 s->open_flags = raw_s->open_flags;
703 #ifdef CONFIG_LINUX_AIO
704 s->use_aio = raw_s->use_aio;
707 g_free(state->opaque);
708 state->opaque = NULL;
712 static void raw_reopen_abort(BDRVReopenState *state)
714 BDRVRawReopenState *raw_s = state->opaque;
716 /* nothing to do if NULL, we didn't get far enough */
721 if (raw_s->fd >= 0) {
722 qemu_close(raw_s->fd);
725 g_free(state->opaque);
726 state->opaque = NULL;
729 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
731 BDRVRawState *s = bs->opaque;
733 raw_probe_alignment(bs, s->fd, errp);
734 bs->bl.min_mem_alignment = s->buf_align;
735 bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
738 static int check_for_dasd(int fd)
741 struct dasd_information2_t info = {0};
743 return ioctl(fd, BIODASDINFO2, &info);
750 * Try to get @bs's logical and physical block size.
751 * On success, store them in @bsz and return zero.
752 * On failure, return negative errno.
754 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
756 BDRVRawState *s = bs->opaque;
759 /* If DASD, get blocksizes */
760 if (check_for_dasd(s->fd) < 0) {
763 ret = probe_logical_blocksize(s->fd, &bsz->log);
767 return probe_physical_blocksize(s->fd, &bsz->phys);
771 * Try to get @bs's geometry: cyls, heads, sectors.
772 * On success, store them in @geo and return 0.
773 * On failure return -errno.
774 * (Allows block driver to assign default geometry values that guest sees)
777 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
779 BDRVRawState *s = bs->opaque;
780 struct hd_geometry ioctl_geo = {0};
782 /* If DASD, get its geometry */
783 if (check_for_dasd(s->fd) < 0) {
786 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
789 /* HDIO_GETGEO may return success even though geo contains zeros
790 (e.g. certain multipath setups) */
791 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
794 /* Do not return a geometry for partition */
795 if (ioctl_geo.start != 0) {
798 geo->heads = ioctl_geo.heads;
799 geo->sectors = ioctl_geo.sectors;
800 geo->cylinders = ioctl_geo.cylinders;
804 #else /* __linux__ */
805 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
811 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
815 ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
823 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
827 ret = qemu_fdatasync(aiocb->aio_fildes);
836 static bool preadv_present = true;
839 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
841 return preadv(fd, iov, nr_iov, offset);
845 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
847 return pwritev(fd, iov, nr_iov, offset);
852 static bool preadv_present = false;
855 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
861 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
868 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
873 if (aiocb->aio_type & QEMU_AIO_WRITE)
874 len = qemu_pwritev(aiocb->aio_fildes,
879 len = qemu_preadv(aiocb->aio_fildes,
883 } while (len == -1 && errno == EINTR);
892 * Read/writes the data to/from a given linear buffer.
894 * Returns the number of bytes handles or -errno in case of an error. Short
895 * reads are only returned if the end of the file is reached.
897 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
902 while (offset < aiocb->aio_nbytes) {
903 if (aiocb->aio_type & QEMU_AIO_WRITE) {
904 len = pwrite(aiocb->aio_fildes,
905 (const char *)buf + offset,
906 aiocb->aio_nbytes - offset,
907 aiocb->aio_offset + offset);
909 len = pread(aiocb->aio_fildes,
911 aiocb->aio_nbytes - offset,
912 aiocb->aio_offset + offset);
914 if (len == -1 && errno == EINTR) {
916 } else if (len == -1 && errno == EINVAL &&
917 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
918 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
920 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
921 * after a short read. Assume that O_DIRECT short reads only occur
922 * at EOF. Therefore this is a short read, not an I/O error.
925 } else if (len == -1) {
928 } else if (len == 0) {
937 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
942 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
944 * If there is just a single buffer, and it is properly aligned
945 * we can just use plain pread/pwrite without any problems.
947 if (aiocb->aio_niov == 1) {
948 return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
951 * We have more than one iovec, and all are properly aligned.
953 * Try preadv/pwritev first and fall back to linearizing the
954 * buffer if it's not supported.
956 if (preadv_present) {
957 nbytes = handle_aiocb_rw_vector(aiocb);
958 if (nbytes == aiocb->aio_nbytes ||
959 (nbytes < 0 && nbytes != -ENOSYS)) {
962 preadv_present = false;
966 * XXX(hch): short read/write. no easy way to handle the reminder
967 * using these interfaces. For now retry using plain
973 * Ok, we have to do it the hard way, copy all segments into
974 * a single aligned buffer.
976 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
981 if (aiocb->aio_type & QEMU_AIO_WRITE) {
985 for (i = 0; i < aiocb->aio_niov; ++i) {
986 memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
987 p += aiocb->aio_iov[i].iov_len;
989 assert(p - buf == aiocb->aio_nbytes);
992 nbytes = handle_aiocb_rw_linear(aiocb, buf);
993 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
995 size_t count = aiocb->aio_nbytes, copy;
998 for (i = 0; i < aiocb->aio_niov && count; ++i) {
1000 if (copy > aiocb->aio_iov[i].iov_len) {
1001 copy = aiocb->aio_iov[i].iov_len;
1003 memcpy(aiocb->aio_iov[i].iov_base, p, copy);
1004 assert(count >= copy);
1016 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
1018 struct xfs_flock64 fl;
1021 memset(&fl, 0, sizeof(fl));
1022 fl.l_whence = SEEK_SET;
1023 fl.l_start = offset;
1026 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
1028 DPRINTF("cannot write zero range (%s)\n", strerror(errno));
1035 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
1037 struct xfs_flock64 fl;
1040 memset(&fl, 0, sizeof(fl));
1041 fl.l_whence = SEEK_SET;
1042 fl.l_start = offset;
1045 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
1047 DPRINTF("cannot punch hole (%s)\n", strerror(errno));
1055 static int translate_err(int err)
1057 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1064 #ifdef CONFIG_FALLOCATE
1065 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1068 if (fallocate(fd, mode, offset, len) == 0) {
1071 } while (errno == EINTR);
1072 return translate_err(-errno);
1076 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1079 BDRVRawState *s = aiocb->bs->opaque;
1081 if (!s->has_write_zeroes) {
1087 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1088 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1091 } while (errno == EINTR);
1093 ret = translate_err(-errno);
1096 if (ret == -ENOTSUP) {
1097 s->has_write_zeroes = false;
1102 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
1104 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1105 BDRVRawState *s = aiocb->bs->opaque;
1108 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1109 return handle_aiocb_write_zeroes_block(aiocb);
1114 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
1118 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1119 if (s->has_write_zeroes) {
1120 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1121 aiocb->aio_offset, aiocb->aio_nbytes);
1122 if (ret == 0 || ret != -ENOTSUP) {
1125 s->has_write_zeroes = false;
1129 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1130 if (s->has_discard && s->has_fallocate) {
1131 int ret = do_fallocate(s->fd,
1132 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1133 aiocb->aio_offset, aiocb->aio_nbytes);
1135 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1136 if (ret == 0 || ret != -ENOTSUP) {
1139 s->has_fallocate = false;
1140 } else if (ret != -ENOTSUP) {
1143 s->has_discard = false;
1148 #ifdef CONFIG_FALLOCATE
1149 if (s->has_fallocate && aiocb->aio_offset >= bdrv_getlength(aiocb->bs)) {
1150 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1151 if (ret == 0 || ret != -ENOTSUP) {
1154 s->has_fallocate = false;
1161 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
1163 int ret = -EOPNOTSUPP;
1164 BDRVRawState *s = aiocb->bs->opaque;
1166 if (!s->has_discard) {
1170 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1173 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1174 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1177 } while (errno == EINTR);
1184 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1188 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1189 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1190 aiocb->aio_offset, aiocb->aio_nbytes);
1194 ret = translate_err(ret);
1195 if (ret == -ENOTSUP) {
1196 s->has_discard = false;
1201 static int aio_worker(void *arg)
1203 RawPosixAIOData *aiocb = arg;
1206 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
1208 ret = handle_aiocb_rw(aiocb);
1209 if (ret >= 0 && ret < aiocb->aio_nbytes) {
1210 iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
1211 0, aiocb->aio_nbytes - ret);
1213 ret = aiocb->aio_nbytes;
1215 if (ret == aiocb->aio_nbytes) {
1217 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1221 case QEMU_AIO_WRITE:
1222 ret = handle_aiocb_rw(aiocb);
1223 if (ret == aiocb->aio_nbytes) {
1225 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1229 case QEMU_AIO_FLUSH:
1230 ret = handle_aiocb_flush(aiocb);
1232 case QEMU_AIO_IOCTL:
1233 ret = handle_aiocb_ioctl(aiocb);
1235 case QEMU_AIO_DISCARD:
1236 ret = handle_aiocb_discard(aiocb);
1238 case QEMU_AIO_WRITE_ZEROES:
1239 ret = handle_aiocb_write_zeroes(aiocb);
1242 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1251 static int paio_submit_co(BlockDriverState *bs, int fd,
1252 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1255 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1259 acb->aio_type = type;
1260 acb->aio_fildes = fd;
1262 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1263 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1266 acb->aio_iov = qiov->iov;
1267 acb->aio_niov = qiov->niov;
1268 assert(qiov->size == acb->aio_nbytes);
1271 trace_paio_submit_co(sector_num, nb_sectors, type);
1272 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1273 return thread_pool_submit_co(pool, aio_worker, acb);
1276 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
1277 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1278 BlockCompletionFunc *cb, void *opaque, int type)
1280 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1284 acb->aio_type = type;
1285 acb->aio_fildes = fd;
1287 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1288 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1291 acb->aio_iov = qiov->iov;
1292 acb->aio_niov = qiov->niov;
1293 assert(qiov->size == acb->aio_nbytes);
1296 trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
1297 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1298 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1301 static BlockAIOCB *raw_aio_submit(BlockDriverState *bs,
1302 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1303 BlockCompletionFunc *cb, void *opaque, int type)
1305 BDRVRawState *s = bs->opaque;
1307 if (fd_open(bs) < 0)
1311 * Check if the underlying device requires requests to be aligned,
1312 * and if the request we are trying to submit is aligned or not.
1313 * If this is the case tell the low-level driver that it needs
1314 * to copy the buffer.
1316 if (s->needs_alignment) {
1317 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1318 type |= QEMU_AIO_MISALIGNED;
1319 #ifdef CONFIG_LINUX_AIO
1320 } else if (s->use_aio) {
1321 return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
1322 nb_sectors, cb, opaque, type);
1327 return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
1331 static void raw_aio_plug(BlockDriverState *bs)
1333 #ifdef CONFIG_LINUX_AIO
1334 BDRVRawState *s = bs->opaque;
1336 laio_io_plug(bs, s->aio_ctx);
1341 static void raw_aio_unplug(BlockDriverState *bs)
1343 #ifdef CONFIG_LINUX_AIO
1344 BDRVRawState *s = bs->opaque;
1346 laio_io_unplug(bs, s->aio_ctx, true);
1351 static void raw_aio_flush_io_queue(BlockDriverState *bs)
1353 #ifdef CONFIG_LINUX_AIO
1354 BDRVRawState *s = bs->opaque;
1356 laio_io_unplug(bs, s->aio_ctx, false);
1361 static BlockAIOCB *raw_aio_readv(BlockDriverState *bs,
1362 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1363 BlockCompletionFunc *cb, void *opaque)
1365 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1366 cb, opaque, QEMU_AIO_READ);
1369 static BlockAIOCB *raw_aio_writev(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_WRITE);
1377 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
1378 BlockCompletionFunc *cb, void *opaque)
1380 BDRVRawState *s = bs->opaque;
1382 if (fd_open(bs) < 0)
1385 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1388 static void raw_close(BlockDriverState *bs)
1390 BDRVRawState *s = bs->opaque;
1392 raw_detach_aio_context(bs);
1394 #ifdef CONFIG_LINUX_AIO
1396 laio_cleanup(s->aio_ctx);
1405 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1407 BDRVRawState *s = bs->opaque;
1410 if (fstat(s->fd, &st)) {
1414 if (S_ISREG(st.st_mode)) {
1415 if (ftruncate(s->fd, offset) < 0) {
1418 } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1419 if (offset > raw_getlength(bs)) {
1430 static int64_t raw_getlength(BlockDriverState *bs)
1432 BDRVRawState *s = bs->opaque;
1438 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1439 struct disklabel dl;
1441 if (ioctl(fd, DIOCGDINFO, &dl))
1443 return (uint64_t)dl.d_secsize *
1444 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1448 #elif defined(__NetBSD__)
1449 static int64_t raw_getlength(BlockDriverState *bs)
1451 BDRVRawState *s = bs->opaque;
1457 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1458 struct dkwedge_info dkw;
1460 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1461 return dkw.dkw_size * 512;
1463 struct disklabel dl;
1465 if (ioctl(fd, DIOCGDINFO, &dl))
1467 return (uint64_t)dl.d_secsize *
1468 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1473 #elif defined(__sun__)
1474 static int64_t raw_getlength(BlockDriverState *bs)
1476 BDRVRawState *s = bs->opaque;
1477 struct dk_minfo minfo;
1487 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1489 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1491 return minfo.dki_lbsize * minfo.dki_capacity;
1495 * There are reports that lseek on some devices fails, but
1496 * irc discussion said that contingency on contingency was overkill.
1498 size = lseek(s->fd, 0, SEEK_END);
1504 #elif defined(CONFIG_BSD)
1505 static int64_t raw_getlength(BlockDriverState *bs)
1507 BDRVRawState *s = bs->opaque;
1511 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1520 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1523 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1524 #ifdef DIOCGMEDIASIZE
1525 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1526 #elif defined(DIOCGPART)
1529 if (ioctl(fd, DIOCGPART, &pi) == 0)
1530 size = pi.media_size;
1536 #if defined(__APPLE__) && defined(__MACH__)
1538 uint64_t sectors = 0;
1539 uint32_t sector_size = 0;
1541 if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0
1542 && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) {
1543 size = sectors * sector_size;
1545 size = lseek(fd, 0LL, SEEK_END);
1552 size = lseek(fd, 0LL, SEEK_END);
1557 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1560 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1561 if (size == 2048LL * (unsigned)-1)
1563 /* XXX no disc? maybe we need to reopen... */
1564 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1571 size = lseek(fd, 0, SEEK_END);
1579 static int64_t raw_getlength(BlockDriverState *bs)
1581 BDRVRawState *s = bs->opaque;
1590 size = lseek(s->fd, 0, SEEK_END);
1598 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1601 BDRVRawState *s = bs->opaque;
1603 if (fstat(s->fd, &st) < 0) {
1606 return (int64_t)st.st_blocks * 512;
1609 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
1613 int64_t total_size = 0;
1615 PreallocMode prealloc;
1617 Error *local_err = NULL;
1619 strstart(filename, "file:", &filename);
1621 /* Read out options */
1622 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1624 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
1625 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1626 prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
1627 PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
1631 error_propagate(errp, local_err);
1636 fd = qemu_open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY,
1640 error_setg_errno(errp, -result, "Could not create file");
1646 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1647 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1648 * will be ignored since any failure of this operation should not
1649 * block the left work.
1652 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
1653 attr |= FS_NOCOW_FL;
1654 ioctl(fd, FS_IOC_SETFLAGS, &attr);
1659 if (ftruncate(fd, total_size) != 0) {
1661 error_setg_errno(errp, -result, "Could not resize file");
1666 #ifdef CONFIG_POSIX_FALLOCATE
1667 case PREALLOC_MODE_FALLOC:
1668 /* posix_fallocate() doesn't set errno. */
1669 result = -posix_fallocate(fd, 0, total_size);
1671 error_setg_errno(errp, -result,
1672 "Could not preallocate data for the new file");
1676 case PREALLOC_MODE_FULL:
1678 int64_t num = 0, left = total_size;
1679 buf = g_malloc0(65536);
1682 num = MIN(left, 65536);
1683 result = write(fd, buf, num);
1686 error_setg_errno(errp, -result,
1687 "Could not write to the new file");
1696 error_setg_errno(errp, -result,
1697 "Could not flush new file to disk");
1703 case PREALLOC_MODE_OFF:
1707 error_setg(errp, "Unsupported preallocation mode: %s",
1708 PreallocMode_lookup[prealloc]);
1713 if (qemu_close(fd) != 0 && result == 0) {
1715 error_setg_errno(errp, -result, "Could not close the new file");
1722 * Find allocation range in @bs around offset @start.
1723 * May change underlying file descriptor's file offset.
1724 * If @start is not in a hole, store @start in @data, and the
1725 * beginning of the next hole in @hole, and return 0.
1726 * If @start is in a non-trailing hole, store @start in @hole and the
1727 * beginning of the next non-hole in @data, and return 0.
1728 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1729 * If we can't find out, return a negative errno other than -ENXIO.
1731 static int find_allocation(BlockDriverState *bs, off_t start,
1732 off_t *data, off_t *hole)
1734 #if defined SEEK_HOLE && defined SEEK_DATA
1735 BDRVRawState *s = bs->opaque;
1740 * D1. offs == start: start is in data
1741 * D2. offs > start: start is in a hole, next data at offs
1742 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1743 * or start is beyond EOF
1744 * If the latter happens, the file has been truncated behind
1745 * our back since we opened it. All bets are off then.
1746 * Treating like a trailing hole is simplest.
1747 * D4. offs < 0, errno != ENXIO: we learned nothing
1749 offs = lseek(s->fd, start, SEEK_DATA);
1751 return -errno; /* D3 or D4 */
1753 assert(offs >= start);
1756 /* D2: in hole, next data at offs */
1762 /* D1: in data, end not yet known */
1766 * H1. offs == start: start is in a hole
1767 * If this happens here, a hole has been dug behind our back
1768 * since the previous lseek().
1769 * H2. offs > start: either start is in data, next hole at offs,
1770 * or start is in trailing hole, EOF at offs
1771 * Linux treats trailing holes like any other hole: offs ==
1772 * start. Solaris seeks to EOF instead: offs > start (blech).
1773 * If that happens here, a hole has been dug behind our back
1774 * since the previous lseek().
1775 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1776 * If this happens, the file has been truncated behind our
1777 * back since we opened it. Treat it like a trailing hole.
1778 * H4. offs < 0, errno != ENXIO: we learned nothing
1779 * Pretend we know nothing at all, i.e. "forget" about D1.
1781 offs = lseek(s->fd, start, SEEK_HOLE);
1783 return -errno; /* D1 and (H3 or H4) */
1785 assert(offs >= start);
1789 * D1 and H2: either in data, next hole at offs, or it was in
1790 * data but is now in a trailing hole. In the latter case,
1791 * all bets are off. Treating it as if it there was data all
1792 * the way to EOF is safe, so simply do that.
1807 * Returns the allocation status of the specified sectors.
1809 * If 'sector_num' is beyond the end of the disk image the return value is 0
1810 * and 'pnum' is set to 0.
1812 * 'pnum' is set to the number of sectors (including and immediately following
1813 * the specified sector) that are known to be in the same
1814 * allocated/unallocated state.
1816 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1817 * beyond the end of the disk image it will be clamped.
1819 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1821 int nb_sectors, int *pnum,
1822 BlockDriverState **file)
1824 off_t start, data = 0, hole = 0;
1833 start = sector_num * BDRV_SECTOR_SIZE;
1834 total_size = bdrv_getlength(bs);
1835 if (total_size < 0) {
1837 } else if (start >= total_size) {
1840 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1841 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1844 ret = find_allocation(bs, start, &data, &hole);
1845 if (ret == -ENXIO) {
1848 ret = BDRV_BLOCK_ZERO;
1849 } else if (ret < 0) {
1850 /* No info available, so pretend there are no holes */
1852 ret = BDRV_BLOCK_DATA;
1853 } else if (data == start) {
1854 /* On a data extent, compute sectors to the end of the extent,
1855 * possibly including a partial sector at EOF. */
1856 *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1857 ret = BDRV_BLOCK_DATA;
1859 /* On a hole, compute sectors to the beginning of the next extent. */
1860 assert(hole == start);
1861 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1862 ret = BDRV_BLOCK_ZERO;
1865 return ret | BDRV_BLOCK_OFFSET_VALID | start;
1868 static coroutine_fn BlockAIOCB *raw_aio_discard(BlockDriverState *bs,
1869 int64_t sector_num, int nb_sectors,
1870 BlockCompletionFunc *cb, void *opaque)
1872 BDRVRawState *s = bs->opaque;
1874 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1875 cb, opaque, QEMU_AIO_DISCARD);
1878 static int coroutine_fn raw_co_write_zeroes(
1879 BlockDriverState *bs, int64_t sector_num,
1880 int nb_sectors, BdrvRequestFlags flags)
1882 BDRVRawState *s = bs->opaque;
1884 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1885 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1886 QEMU_AIO_WRITE_ZEROES);
1887 } else if (s->discard_zeroes) {
1888 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1894 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1896 BDRVRawState *s = bs->opaque;
1898 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1899 bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1903 static QemuOptsList raw_create_opts = {
1904 .name = "raw-create-opts",
1905 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
1908 .name = BLOCK_OPT_SIZE,
1909 .type = QEMU_OPT_SIZE,
1910 .help = "Virtual disk size"
1913 .name = BLOCK_OPT_NOCOW,
1914 .type = QEMU_OPT_BOOL,
1915 .help = "Turn off copy-on-write (valid only on btrfs)"
1918 .name = BLOCK_OPT_PREALLOC,
1919 .type = QEMU_OPT_STRING,
1920 .help = "Preallocation mode (allowed values: off, falloc, full)"
1922 { /* end of list */ }
1926 BlockDriver bdrv_file = {
1927 .format_name = "file",
1928 .protocol_name = "file",
1929 .instance_size = sizeof(BDRVRawState),
1930 .bdrv_needs_filename = true,
1931 .bdrv_probe = NULL, /* no probe for protocols */
1932 .bdrv_parse_filename = raw_parse_filename,
1933 .bdrv_file_open = raw_open,
1934 .bdrv_reopen_prepare = raw_reopen_prepare,
1935 .bdrv_reopen_commit = raw_reopen_commit,
1936 .bdrv_reopen_abort = raw_reopen_abort,
1937 .bdrv_close = raw_close,
1938 .bdrv_create = raw_create,
1939 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1940 .bdrv_co_get_block_status = raw_co_get_block_status,
1941 .bdrv_co_write_zeroes = raw_co_write_zeroes,
1943 .bdrv_aio_readv = raw_aio_readv,
1944 .bdrv_aio_writev = raw_aio_writev,
1945 .bdrv_aio_flush = raw_aio_flush,
1946 .bdrv_aio_discard = raw_aio_discard,
1947 .bdrv_refresh_limits = raw_refresh_limits,
1948 .bdrv_io_plug = raw_aio_plug,
1949 .bdrv_io_unplug = raw_aio_unplug,
1950 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
1952 .bdrv_truncate = raw_truncate,
1953 .bdrv_getlength = raw_getlength,
1954 .bdrv_get_info = raw_get_info,
1955 .bdrv_get_allocated_file_size
1956 = raw_get_allocated_file_size,
1958 .bdrv_detach_aio_context = raw_detach_aio_context,
1959 .bdrv_attach_aio_context = raw_attach_aio_context,
1961 .create_opts = &raw_create_opts,
1964 /***********************************************/
1967 #if defined(__APPLE__) && defined(__MACH__)
1968 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1969 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
1970 CFIndex maxPathSize, int flags);
1971 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1973 kern_return_t kernResult;
1974 mach_port_t masterPort;
1975 CFMutableDictionaryRef classesToMatch;
1977 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1978 if ( KERN_SUCCESS != kernResult ) {
1979 printf( "IOMasterPort returned %d\n", kernResult );
1982 classesToMatch = IOServiceMatching( kIOCDMediaClass );
1983 if ( classesToMatch == NULL ) {
1984 printf( "IOServiceMatching returned a NULL dictionary.\n" );
1986 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1988 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1989 if ( KERN_SUCCESS != kernResult )
1991 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1997 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
1998 CFIndex maxPathSize, int flags)
2000 io_object_t nextMedia;
2001 kern_return_t kernResult = KERN_FAILURE;
2003 nextMedia = IOIteratorNext( mediaIterator );
2006 CFTypeRef bsdPathAsCFString;
2007 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
2008 if ( bsdPathAsCFString ) {
2009 size_t devPathLength;
2010 strcpy( bsdPath, _PATH_DEV );
2011 if (flags & BDRV_O_NOCACHE) {
2012 strcat(bsdPath, "r");
2014 devPathLength = strlen( bsdPath );
2015 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
2016 kernResult = KERN_SUCCESS;
2018 CFRelease( bsdPathAsCFString );
2020 IOObjectRelease( nextMedia );
2028 static int hdev_probe_device(const char *filename)
2032 /* allow a dedicated CD-ROM driver to match with a higher priority */
2033 if (strstart(filename, "/dev/cdrom", NULL))
2036 if (stat(filename, &st) >= 0 &&
2037 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
2044 static int check_hdev_writable(BDRVRawState *s)
2046 #if defined(BLKROGET)
2047 /* Linux block devices can be configured "read-only" using blockdev(8).
2048 * This is independent of device node permissions and therefore open(2)
2049 * with O_RDWR succeeds. Actual writes fail with EPERM.
2051 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2052 * check for read-only block devices so that Linux block devices behave
2058 if (fstat(s->fd, &st)) {
2062 if (!S_ISBLK(st.st_mode)) {
2066 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
2073 #endif /* defined(BLKROGET) */
2077 static void hdev_parse_filename(const char *filename, QDict *options,
2080 /* The prefix is optional, just as for "file". */
2081 strstart(filename, "host_device:", &filename);
2083 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2086 static bool hdev_is_sg(BlockDriverState *bs)
2089 #if defined(__linux__)
2092 struct sg_scsi_id scsiid;
2095 if (stat(bs->filename, &st) >= 0 && S_ISCHR(st.st_mode) &&
2096 !bdrv_ioctl(bs, SG_GET_VERSION_NUM, &sg_version) &&
2097 !bdrv_ioctl(bs, SG_GET_SCSI_ID, &scsiid)) {
2098 DPRINTF("SG device found: type=%d, version=%d\n",
2099 scsiid.scsi_type, sg_version);
2108 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
2111 BDRVRawState *s = bs->opaque;
2112 Error *local_err = NULL;
2115 #if defined(__APPLE__) && defined(__MACH__)
2116 const char *filename = qdict_get_str(options, "filename");
2118 if (strstart(filename, "/dev/cdrom", NULL)) {
2119 kern_return_t kernResult;
2120 io_iterator_t mediaIterator;
2121 char bsdPath[ MAXPATHLEN ];
2124 kernResult = FindEjectableCDMedia( &mediaIterator );
2125 kernResult = GetBSDPath(mediaIterator, bsdPath, sizeof(bsdPath),
2127 if ( bsdPath[ 0 ] != '\0' ) {
2128 strcat(bsdPath,"s0");
2129 /* some CDs don't have a partition 0 */
2130 fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
2132 bsdPath[strlen(bsdPath)-1] = '1';
2137 qdict_put(options, "filename", qstring_from_str(filename));
2140 if ( mediaIterator )
2141 IOObjectRelease( mediaIterator );
2145 s->type = FTYPE_FILE;
2147 ret = raw_open_common(bs, options, flags, 0, &local_err);
2150 error_propagate(errp, local_err);
2155 /* Since this does ioctl the device must be already opened */
2156 bs->sg = hdev_is_sg(bs);
2158 if (flags & BDRV_O_RDWR) {
2159 ret = check_hdev_writable(s);
2162 error_setg_errno(errp, -ret, "The device is not writable");
2170 #if defined(__linux__)
2172 static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
2173 unsigned long int req, void *buf,
2174 BlockCompletionFunc *cb, void *opaque)
2176 BDRVRawState *s = bs->opaque;
2177 RawPosixAIOData *acb;
2180 if (fd_open(bs) < 0)
2183 acb = g_new(RawPosixAIOData, 1);
2185 acb->aio_type = QEMU_AIO_IOCTL;
2186 acb->aio_fildes = s->fd;
2187 acb->aio_offset = 0;
2188 acb->aio_ioctl_buf = buf;
2189 acb->aio_ioctl_cmd = req;
2190 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
2191 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
2195 static int fd_open(BlockDriverState *bs)
2197 BDRVRawState *s = bs->opaque;
2199 /* this is just to ensure s->fd is sane (its called by io ops) */
2205 static coroutine_fn BlockAIOCB *hdev_aio_discard(BlockDriverState *bs,
2206 int64_t sector_num, int nb_sectors,
2207 BlockCompletionFunc *cb, void *opaque)
2209 BDRVRawState *s = bs->opaque;
2211 if (fd_open(bs) < 0) {
2214 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
2215 cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2218 static coroutine_fn int hdev_co_write_zeroes(BlockDriverState *bs,
2219 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2221 BDRVRawState *s = bs->opaque;
2228 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2229 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2230 QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2231 } else if (s->discard_zeroes) {
2232 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2233 QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2238 static int hdev_create(const char *filename, QemuOpts *opts,
2243 struct stat stat_buf;
2244 int64_t total_size = 0;
2247 /* This function is used by both protocol block drivers and therefore either
2248 * of these prefixes may be given.
2249 * The return value has to be stored somewhere, otherwise this is an error
2250 * due to -Werror=unused-value. */
2252 strstart(filename, "host_device:", &filename) ||
2253 strstart(filename, "host_cdrom:" , &filename);
2257 ret = raw_normalize_devicepath(&filename);
2259 error_setg_errno(errp, -ret, "Could not normalize device path");
2263 /* Read out options */
2264 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2267 fd = qemu_open(filename, O_WRONLY | O_BINARY);
2270 error_setg_errno(errp, -ret, "Could not open device");
2274 if (fstat(fd, &stat_buf) < 0) {
2276 error_setg_errno(errp, -ret, "Could not stat device");
2277 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2279 "The given file is neither a block nor a character device");
2281 } else if (lseek(fd, 0, SEEK_END) < total_size) {
2282 error_setg(errp, "Device is too small");
2290 static BlockDriver bdrv_host_device = {
2291 .format_name = "host_device",
2292 .protocol_name = "host_device",
2293 .instance_size = sizeof(BDRVRawState),
2294 .bdrv_needs_filename = true,
2295 .bdrv_probe_device = hdev_probe_device,
2296 .bdrv_parse_filename = hdev_parse_filename,
2297 .bdrv_file_open = hdev_open,
2298 .bdrv_close = raw_close,
2299 .bdrv_reopen_prepare = raw_reopen_prepare,
2300 .bdrv_reopen_commit = raw_reopen_commit,
2301 .bdrv_reopen_abort = raw_reopen_abort,
2302 .bdrv_create = hdev_create,
2303 .create_opts = &raw_create_opts,
2304 .bdrv_co_write_zeroes = hdev_co_write_zeroes,
2306 .bdrv_aio_readv = raw_aio_readv,
2307 .bdrv_aio_writev = raw_aio_writev,
2308 .bdrv_aio_flush = raw_aio_flush,
2309 .bdrv_aio_discard = hdev_aio_discard,
2310 .bdrv_refresh_limits = raw_refresh_limits,
2311 .bdrv_io_plug = raw_aio_plug,
2312 .bdrv_io_unplug = raw_aio_unplug,
2313 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
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 .bdrv_detach_aio_context = raw_detach_aio_context,
2324 .bdrv_attach_aio_context = raw_attach_aio_context,
2326 /* generic scsi device */
2328 .bdrv_aio_ioctl = hdev_aio_ioctl,
2332 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2333 static void cdrom_parse_filename(const char *filename, QDict *options,
2336 /* The prefix is optional, just as for "file". */
2337 strstart(filename, "host_cdrom:", &filename);
2339 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2344 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2347 BDRVRawState *s = bs->opaque;
2348 Error *local_err = NULL;
2353 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2354 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2356 error_propagate(errp, local_err);
2361 static int cdrom_probe_device(const char *filename)
2367 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2371 ret = fstat(fd, &st);
2372 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2376 /* Attempt to detect via a CDROM specific ioctl */
2377 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2387 static bool cdrom_is_inserted(BlockDriverState *bs)
2389 BDRVRawState *s = bs->opaque;
2392 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2393 return ret == CDS_DISC_OK;
2396 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2398 BDRVRawState *s = bs->opaque;
2401 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2402 perror("CDROMEJECT");
2404 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2405 perror("CDROMEJECT");
2409 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2411 BDRVRawState *s = bs->opaque;
2413 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2415 * Note: an error can happen if the distribution automatically
2418 /* perror("CDROM_LOCKDOOR"); */
2422 static BlockDriver bdrv_host_cdrom = {
2423 .format_name = "host_cdrom",
2424 .protocol_name = "host_cdrom",
2425 .instance_size = sizeof(BDRVRawState),
2426 .bdrv_needs_filename = true,
2427 .bdrv_probe_device = cdrom_probe_device,
2428 .bdrv_parse_filename = cdrom_parse_filename,
2429 .bdrv_file_open = cdrom_open,
2430 .bdrv_close = raw_close,
2431 .bdrv_reopen_prepare = raw_reopen_prepare,
2432 .bdrv_reopen_commit = raw_reopen_commit,
2433 .bdrv_reopen_abort = raw_reopen_abort,
2434 .bdrv_create = hdev_create,
2435 .create_opts = &raw_create_opts,
2437 .bdrv_aio_readv = raw_aio_readv,
2438 .bdrv_aio_writev = raw_aio_writev,
2439 .bdrv_aio_flush = raw_aio_flush,
2440 .bdrv_refresh_limits = raw_refresh_limits,
2441 .bdrv_io_plug = raw_aio_plug,
2442 .bdrv_io_unplug = raw_aio_unplug,
2443 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2445 .bdrv_truncate = raw_truncate,
2446 .bdrv_getlength = raw_getlength,
2447 .has_variable_length = true,
2448 .bdrv_get_allocated_file_size
2449 = raw_get_allocated_file_size,
2451 .bdrv_detach_aio_context = raw_detach_aio_context,
2452 .bdrv_attach_aio_context = raw_attach_aio_context,
2454 /* removable device support */
2455 .bdrv_is_inserted = cdrom_is_inserted,
2456 .bdrv_eject = cdrom_eject,
2457 .bdrv_lock_medium = cdrom_lock_medium,
2459 /* generic scsi device */
2460 .bdrv_aio_ioctl = hdev_aio_ioctl,
2462 #endif /* __linux__ */
2464 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2465 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2468 BDRVRawState *s = bs->opaque;
2469 Error *local_err = NULL;
2474 ret = raw_open_common(bs, options, flags, 0, &local_err);
2477 error_propagate(errp, local_err);
2482 /* make sure the door isn't locked at this time */
2483 ioctl(s->fd, CDIOCALLOW);
2487 static int cdrom_probe_device(const char *filename)
2489 if (strstart(filename, "/dev/cd", NULL) ||
2490 strstart(filename, "/dev/acd", NULL))
2495 static int cdrom_reopen(BlockDriverState *bs)
2497 BDRVRawState *s = bs->opaque;
2501 * Force reread of possibly changed/newly loaded disc,
2502 * FreeBSD seems to not notice sometimes...
2506 fd = qemu_open(bs->filename, s->open_flags, 0644);
2513 /* make sure the door isn't locked at this time */
2514 ioctl(s->fd, CDIOCALLOW);
2518 static bool cdrom_is_inserted(BlockDriverState *bs)
2520 return raw_getlength(bs) > 0;
2523 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2525 BDRVRawState *s = bs->opaque;
2530 (void) ioctl(s->fd, CDIOCALLOW);
2533 if (ioctl(s->fd, CDIOCEJECT) < 0)
2534 perror("CDIOCEJECT");
2536 if (ioctl(s->fd, CDIOCCLOSE) < 0)
2537 perror("CDIOCCLOSE");
2543 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2545 BDRVRawState *s = bs->opaque;
2549 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2551 * Note: an error can happen if the distribution automatically
2554 /* perror("CDROM_LOCKDOOR"); */
2558 static BlockDriver bdrv_host_cdrom = {
2559 .format_name = "host_cdrom",
2560 .protocol_name = "host_cdrom",
2561 .instance_size = sizeof(BDRVRawState),
2562 .bdrv_needs_filename = true,
2563 .bdrv_probe_device = cdrom_probe_device,
2564 .bdrv_parse_filename = cdrom_parse_filename,
2565 .bdrv_file_open = cdrom_open,
2566 .bdrv_close = raw_close,
2567 .bdrv_reopen_prepare = raw_reopen_prepare,
2568 .bdrv_reopen_commit = raw_reopen_commit,
2569 .bdrv_reopen_abort = raw_reopen_abort,
2570 .bdrv_create = hdev_create,
2571 .create_opts = &raw_create_opts,
2573 .bdrv_aio_readv = raw_aio_readv,
2574 .bdrv_aio_writev = raw_aio_writev,
2575 .bdrv_aio_flush = raw_aio_flush,
2576 .bdrv_refresh_limits = raw_refresh_limits,
2577 .bdrv_io_plug = raw_aio_plug,
2578 .bdrv_io_unplug = raw_aio_unplug,
2579 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2581 .bdrv_truncate = raw_truncate,
2582 .bdrv_getlength = raw_getlength,
2583 .has_variable_length = true,
2584 .bdrv_get_allocated_file_size
2585 = raw_get_allocated_file_size,
2587 .bdrv_detach_aio_context = raw_detach_aio_context,
2588 .bdrv_attach_aio_context = raw_attach_aio_context,
2590 /* removable device support */
2591 .bdrv_is_inserted = cdrom_is_inserted,
2592 .bdrv_eject = cdrom_eject,
2593 .bdrv_lock_medium = cdrom_lock_medium,
2595 #endif /* __FreeBSD__ */
2597 static void bdrv_file_init(void)
2600 * Register all the drivers. Note that order is important, the driver
2601 * registered last will get probed first.
2603 bdrv_register(&bdrv_file);
2604 bdrv_register(&bdrv_host_device);
2606 bdrv_register(&bdrv_host_cdrom);
2608 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2609 bdrv_register(&bdrv_host_cdrom);
2613 block_init(bdrv_file_init);