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"
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 {
140 #ifdef CONFIG_LINUX_AIO
148 bool has_write_zeroes:1;
149 bool discard_zeroes:1;
151 bool needs_alignment;
154 typedef struct BDRVRawReopenState {
157 #ifdef CONFIG_LINUX_AIO
160 } BDRVRawReopenState;
162 static int fd_open(BlockDriverState *bs);
163 static int64_t raw_getlength(BlockDriverState *bs);
165 typedef struct RawPosixAIOData {
166 BlockDriverState *bs;
169 struct iovec *aio_iov;
174 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
179 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
180 static int cdrom_reopen(BlockDriverState *bs);
183 #if defined(__NetBSD__)
184 static int raw_normalize_devicepath(const char **filename)
186 static char namebuf[PATH_MAX];
187 const char *dp, *fname;
191 dp = strrchr(fname, '/');
192 if (lstat(fname, &sb) < 0) {
193 fprintf(stderr, "%s: stat failed: %s\n",
194 fname, strerror(errno));
198 if (!S_ISBLK(sb.st_mode)) {
203 snprintf(namebuf, PATH_MAX, "r%s", fname);
205 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
206 (int)(dp - fname), fname, dp + 1);
208 fprintf(stderr, "%s is a block device", fname);
210 fprintf(stderr, ", using %s\n", *filename);
215 static int raw_normalize_devicepath(const char **filename)
222 * Get logical block size via ioctl. On success store it in @sector_size_p.
224 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
226 unsigned int sector_size;
227 bool success = false;
231 /* Try a few ioctls to get the right size */
233 if (ioctl(fd, BLKSSZGET, §or_size) >= 0) {
234 *sector_size_p = sector_size;
238 #ifdef DKIOCGETBLOCKSIZE
239 if (ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) >= 0) {
240 *sector_size_p = sector_size;
244 #ifdef DIOCGSECTORSIZE
245 if (ioctl(fd, DIOCGSECTORSIZE, §or_size) >= 0) {
246 *sector_size_p = sector_size;
251 return success ? 0 : -errno;
255 * Get physical block size of @fd.
256 * On success, store it in @blk_size and return 0.
257 * On failure, return -errno.
259 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
262 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
271 /* Check if read is allowed with given memory buffer and length.
273 * This function is used to check O_DIRECT memory buffer and request alignment.
275 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
277 ssize_t ret = pread(fd, buf, len, 0);
284 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
285 * other errors (e.g. real I/O error), which could happen on a failed
286 * drive, since we only care about probing alignment.
288 if (errno != EINVAL) {
296 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
298 BDRVRawState *s = bs->opaque;
300 size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize());
302 /* For SCSI generic devices the alignment is not really used.
303 With buffered I/O, we don't have any restrictions. */
304 if (bdrv_is_sg(bs) || !s->needs_alignment) {
305 bs->request_alignment = 1;
310 bs->request_alignment = 0;
312 /* Let's try to use the logical blocksize for the alignment. */
313 if (probe_logical_blocksize(fd, &bs->request_alignment) < 0) {
314 bs->request_alignment = 0;
319 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
320 bs->request_alignment = da.d_miniosz;
321 /* The kernel returns wrong information for d_mem */
322 /* s->buf_align = da.d_mem; */
327 /* If we could not get the sizes so far, we can only guess them */
330 buf = qemu_memalign(max_align, 2 * max_align);
331 for (align = 512; align <= max_align; align <<= 1) {
332 if (raw_is_io_aligned(fd, buf + align, max_align)) {
333 s->buf_align = align;
340 if (!bs->request_alignment) {
342 buf = qemu_memalign(s->buf_align, max_align);
343 for (align = 512; align <= max_align; align <<= 1) {
344 if (raw_is_io_aligned(fd, buf, align)) {
345 bs->request_alignment = align;
352 if (!s->buf_align || !bs->request_alignment) {
353 error_setg(errp, "Could not find working O_DIRECT alignment. "
354 "Try cache.direct=off.");
358 static void raw_parse_flags(int bdrv_flags, int *open_flags)
360 assert(open_flags != NULL);
362 *open_flags |= O_BINARY;
363 *open_flags &= ~O_ACCMODE;
364 if (bdrv_flags & BDRV_O_RDWR) {
365 *open_flags |= O_RDWR;
367 *open_flags |= O_RDONLY;
370 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
371 * and O_DIRECT for no caching. */
372 if ((bdrv_flags & BDRV_O_NOCACHE)) {
373 *open_flags |= O_DIRECT;
377 static void raw_detach_aio_context(BlockDriverState *bs)
379 #ifdef CONFIG_LINUX_AIO
380 BDRVRawState *s = bs->opaque;
383 laio_detach_aio_context(s->aio_ctx, bdrv_get_aio_context(bs));
388 static void raw_attach_aio_context(BlockDriverState *bs,
389 AioContext *new_context)
391 #ifdef CONFIG_LINUX_AIO
392 BDRVRawState *s = bs->opaque;
395 laio_attach_aio_context(s->aio_ctx, new_context);
400 #ifdef CONFIG_LINUX_AIO
401 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
404 assert(aio_ctx != NULL);
405 assert(use_aio != NULL);
407 * Currently Linux do AIO only for files opened with O_DIRECT
408 * specified so check NOCACHE flag too
410 if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
411 (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
413 /* if non-NULL, laio_init() has already been run */
414 if (*aio_ctx == NULL) {
415 *aio_ctx = laio_init();
432 static void raw_parse_filename(const char *filename, QDict *options,
435 /* The filename does not have to be prefixed by the protocol name, since
436 * "file" is the default protocol; therefore, the return value of this
437 * function call can be ignored. */
438 strstart(filename, "file:", &filename);
440 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
443 static QemuOptsList raw_runtime_opts = {
445 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
449 .type = QEMU_OPT_STRING,
450 .help = "File name of the image",
452 { /* end of list */ }
456 static int raw_open_common(BlockDriverState *bs, QDict *options,
457 int bdrv_flags, int open_flags, Error **errp)
459 BDRVRawState *s = bs->opaque;
461 Error *local_err = NULL;
462 const char *filename = NULL;
466 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
467 qemu_opts_absorb_qdict(opts, options, &local_err);
469 error_propagate(errp, local_err);
474 filename = qemu_opt_get(opts, "filename");
476 ret = raw_normalize_devicepath(&filename);
478 error_setg_errno(errp, -ret, "Could not normalize device path");
482 s->open_flags = open_flags;
483 raw_parse_flags(bdrv_flags, &s->open_flags);
486 fd = qemu_open(filename, s->open_flags, 0644);
496 #ifdef CONFIG_LINUX_AIO
497 if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
500 error_setg_errno(errp, -ret, "Could not set AIO state");
503 if (!s->use_aio && (bdrv_flags & BDRV_O_NATIVE_AIO)) {
504 error_setg(errp, "aio=native was specified, but it requires "
505 "cache.direct=on, which was not specified.");
510 if (bdrv_flags & BDRV_O_NATIVE_AIO) {
511 error_setg(errp, "aio=native was specified, but is not supported "
516 #endif /* !defined(CONFIG_LINUX_AIO) */
518 s->has_discard = true;
519 s->has_write_zeroes = true;
520 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
521 s->needs_alignment = true;
524 if (fstat(s->fd, &st) < 0) {
526 error_setg_errno(errp, errno, "Could not stat file");
529 if (S_ISREG(st.st_mode)) {
530 s->discard_zeroes = true;
531 s->has_fallocate = true;
533 if (S_ISBLK(st.st_mode)) {
534 #ifdef BLKDISCARDZEROES
536 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
537 s->discard_zeroes = true;
541 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
542 * not rely on the contents of discarded blocks unless using O_DIRECT.
543 * Same for BLKZEROOUT.
545 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
546 s->discard_zeroes = false;
547 s->has_write_zeroes = false;
552 if (S_ISCHR(st.st_mode)) {
554 * The file is a char device (disk), which on FreeBSD isn't behind
555 * a pager, so force all requests to be aligned. This is needed
556 * so QEMU makes sure all IO operations on the device are aligned
557 * to sector size, or else FreeBSD will reject them with EINVAL.
559 s->needs_alignment = true;
564 if (platform_test_xfs_fd(s->fd)) {
569 raw_attach_aio_context(bs, bdrv_get_aio_context(bs));
573 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
580 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
583 BDRVRawState *s = bs->opaque;
584 Error *local_err = NULL;
587 s->type = FTYPE_FILE;
588 ret = raw_open_common(bs, options, flags, 0, &local_err);
590 error_propagate(errp, local_err);
595 static int raw_reopen_prepare(BDRVReopenState *state,
596 BlockReopenQueue *queue, Error **errp)
599 BDRVRawReopenState *raw_s;
601 Error *local_err = NULL;
603 assert(state != NULL);
604 assert(state->bs != NULL);
606 s = state->bs->opaque;
608 state->opaque = g_new0(BDRVRawReopenState, 1);
609 raw_s = state->opaque;
611 #ifdef CONFIG_LINUX_AIO
612 raw_s->use_aio = s->use_aio;
614 /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
615 * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
616 * won't override aio_ctx if aio_ctx is non-NULL */
617 if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
618 error_setg(errp, "Could not set AIO state");
623 if (s->type == FTYPE_CD) {
624 raw_s->open_flags |= O_NONBLOCK;
627 raw_parse_flags(state->flags, &raw_s->open_flags);
631 int fcntl_flags = O_APPEND | O_NONBLOCK;
633 fcntl_flags |= O_NOATIME;
637 /* Not all operating systems have O_ASYNC, and those that don't
638 * will not let us track the state into raw_s->open_flags (typically
639 * you achieve the same effect with an ioctl, for example I_SETSIG
640 * on Solaris). But we do not use O_ASYNC, so that's fine.
642 assert((s->open_flags & O_ASYNC) == 0);
645 if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
646 /* dup the original fd */
647 /* TODO: use qemu fcntl wrapper */
648 #ifdef F_DUPFD_CLOEXEC
649 raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
651 raw_s->fd = dup(s->fd);
652 if (raw_s->fd != -1) {
653 qemu_set_cloexec(raw_s->fd);
656 if (raw_s->fd >= 0) {
657 ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
659 qemu_close(raw_s->fd);
665 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
666 if (raw_s->fd == -1) {
667 const char *normalized_filename = state->bs->filename;
668 ret = raw_normalize_devicepath(&normalized_filename);
670 error_setg_errno(errp, -ret, "Could not normalize device path");
672 assert(!(raw_s->open_flags & O_CREAT));
673 raw_s->fd = qemu_open(normalized_filename, raw_s->open_flags);
674 if (raw_s->fd == -1) {
675 error_setg_errno(errp, errno, "Could not reopen file");
681 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
682 * alignment with the new fd. */
683 if (raw_s->fd != -1) {
684 raw_probe_alignment(state->bs, raw_s->fd, &local_err);
686 qemu_close(raw_s->fd);
688 error_propagate(errp, local_err);
696 static void raw_reopen_commit(BDRVReopenState *state)
698 BDRVRawReopenState *raw_s = state->opaque;
699 BDRVRawState *s = state->bs->opaque;
701 s->open_flags = raw_s->open_flags;
705 #ifdef CONFIG_LINUX_AIO
706 s->use_aio = raw_s->use_aio;
709 g_free(state->opaque);
710 state->opaque = NULL;
714 static void raw_reopen_abort(BDRVReopenState *state)
716 BDRVRawReopenState *raw_s = state->opaque;
718 /* nothing to do if NULL, we didn't get far enough */
723 if (raw_s->fd >= 0) {
724 qemu_close(raw_s->fd);
727 g_free(state->opaque);
728 state->opaque = NULL;
731 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
733 BDRVRawState *s = bs->opaque;
735 raw_probe_alignment(bs, s->fd, errp);
736 bs->bl.min_mem_alignment = s->buf_align;
737 bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
740 static int check_for_dasd(int fd)
743 struct dasd_information2_t info = {0};
745 return ioctl(fd, BIODASDINFO2, &info);
752 * Try to get @bs's logical and physical block size.
753 * On success, store them in @bsz and return zero.
754 * On failure, return negative errno.
756 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
758 BDRVRawState *s = bs->opaque;
761 /* If DASD, get blocksizes */
762 if (check_for_dasd(s->fd) < 0) {
765 ret = probe_logical_blocksize(s->fd, &bsz->log);
769 return probe_physical_blocksize(s->fd, &bsz->phys);
773 * Try to get @bs's geometry: cyls, heads, sectors.
774 * On success, store them in @geo and return 0.
775 * On failure return -errno.
776 * (Allows block driver to assign default geometry values that guest sees)
779 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
781 BDRVRawState *s = bs->opaque;
782 struct hd_geometry ioctl_geo = {0};
784 /* If DASD, get its geometry */
785 if (check_for_dasd(s->fd) < 0) {
788 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
791 /* HDIO_GETGEO may return success even though geo contains zeros
792 (e.g. certain multipath setups) */
793 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
796 /* Do not return a geometry for partition */
797 if (ioctl_geo.start != 0) {
800 geo->heads = ioctl_geo.heads;
801 geo->sectors = ioctl_geo.sectors;
802 geo->cylinders = ioctl_geo.cylinders;
806 #else /* __linux__ */
807 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
813 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
817 ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
825 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
829 ret = qemu_fdatasync(aiocb->aio_fildes);
838 static bool preadv_present = true;
841 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
843 return preadv(fd, iov, nr_iov, offset);
847 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
849 return pwritev(fd, iov, nr_iov, offset);
854 static bool preadv_present = false;
857 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
863 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
870 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
875 if (aiocb->aio_type & QEMU_AIO_WRITE)
876 len = qemu_pwritev(aiocb->aio_fildes,
881 len = qemu_preadv(aiocb->aio_fildes,
885 } while (len == -1 && errno == EINTR);
894 * Read/writes the data to/from a given linear buffer.
896 * Returns the number of bytes handles or -errno in case of an error. Short
897 * reads are only returned if the end of the file is reached.
899 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
904 while (offset < aiocb->aio_nbytes) {
905 if (aiocb->aio_type & QEMU_AIO_WRITE) {
906 len = pwrite(aiocb->aio_fildes,
907 (const char *)buf + offset,
908 aiocb->aio_nbytes - offset,
909 aiocb->aio_offset + offset);
911 len = pread(aiocb->aio_fildes,
913 aiocb->aio_nbytes - offset,
914 aiocb->aio_offset + offset);
916 if (len == -1 && errno == EINTR) {
918 } else if (len == -1 && errno == EINVAL &&
919 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
920 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
922 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
923 * after a short read. Assume that O_DIRECT short reads only occur
924 * at EOF. Therefore this is a short read, not an I/O error.
927 } else if (len == -1) {
930 } else if (len == 0) {
939 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
944 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
946 * If there is just a single buffer, and it is properly aligned
947 * we can just use plain pread/pwrite without any problems.
949 if (aiocb->aio_niov == 1) {
950 return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
953 * We have more than one iovec, and all are properly aligned.
955 * Try preadv/pwritev first and fall back to linearizing the
956 * buffer if it's not supported.
958 if (preadv_present) {
959 nbytes = handle_aiocb_rw_vector(aiocb);
960 if (nbytes == aiocb->aio_nbytes ||
961 (nbytes < 0 && nbytes != -ENOSYS)) {
964 preadv_present = false;
968 * XXX(hch): short read/write. no easy way to handle the reminder
969 * using these interfaces. For now retry using plain
975 * Ok, we have to do it the hard way, copy all segments into
976 * a single aligned buffer.
978 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
983 if (aiocb->aio_type & QEMU_AIO_WRITE) {
987 for (i = 0; i < aiocb->aio_niov; ++i) {
988 memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
989 p += aiocb->aio_iov[i].iov_len;
991 assert(p - buf == aiocb->aio_nbytes);
994 nbytes = handle_aiocb_rw_linear(aiocb, buf);
995 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
997 size_t count = aiocb->aio_nbytes, copy;
1000 for (i = 0; i < aiocb->aio_niov && count; ++i) {
1002 if (copy > aiocb->aio_iov[i].iov_len) {
1003 copy = aiocb->aio_iov[i].iov_len;
1005 memcpy(aiocb->aio_iov[i].iov_base, p, copy);
1006 assert(count >= copy);
1018 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
1020 struct xfs_flock64 fl;
1023 memset(&fl, 0, sizeof(fl));
1024 fl.l_whence = SEEK_SET;
1025 fl.l_start = offset;
1028 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
1030 DPRINTF("cannot write zero range (%s)\n", strerror(errno));
1037 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
1039 struct xfs_flock64 fl;
1042 memset(&fl, 0, sizeof(fl));
1043 fl.l_whence = SEEK_SET;
1044 fl.l_start = offset;
1047 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
1049 DPRINTF("cannot punch hole (%s)\n", strerror(errno));
1057 static int translate_err(int err)
1059 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1066 #ifdef CONFIG_FALLOCATE
1067 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1070 if (fallocate(fd, mode, offset, len) == 0) {
1073 } while (errno == EINTR);
1074 return translate_err(-errno);
1078 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1081 BDRVRawState *s = aiocb->bs->opaque;
1083 if (!s->has_write_zeroes) {
1089 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1090 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1093 } while (errno == EINTR);
1095 ret = translate_err(-errno);
1098 if (ret == -ENOTSUP) {
1099 s->has_write_zeroes = false;
1104 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
1106 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1107 BDRVRawState *s = aiocb->bs->opaque;
1110 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1111 return handle_aiocb_write_zeroes_block(aiocb);
1116 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
1120 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1121 if (s->has_write_zeroes) {
1122 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1123 aiocb->aio_offset, aiocb->aio_nbytes);
1124 if (ret == 0 || ret != -ENOTSUP) {
1127 s->has_write_zeroes = false;
1131 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1132 if (s->has_discard && s->has_fallocate) {
1133 int ret = do_fallocate(s->fd,
1134 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1135 aiocb->aio_offset, aiocb->aio_nbytes);
1137 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1138 if (ret == 0 || ret != -ENOTSUP) {
1141 s->has_fallocate = false;
1142 } else if (ret != -ENOTSUP) {
1145 s->has_discard = false;
1150 #ifdef CONFIG_FALLOCATE
1151 if (s->has_fallocate && aiocb->aio_offset >= bdrv_getlength(aiocb->bs)) {
1152 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1153 if (ret == 0 || ret != -ENOTSUP) {
1156 s->has_fallocate = false;
1163 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
1165 int ret = -EOPNOTSUPP;
1166 BDRVRawState *s = aiocb->bs->opaque;
1168 if (!s->has_discard) {
1172 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1175 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1176 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1179 } while (errno == EINTR);
1186 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1190 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1191 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1192 aiocb->aio_offset, aiocb->aio_nbytes);
1196 ret = translate_err(ret);
1197 if (ret == -ENOTSUP) {
1198 s->has_discard = false;
1203 static int aio_worker(void *arg)
1205 RawPosixAIOData *aiocb = arg;
1208 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
1210 ret = handle_aiocb_rw(aiocb);
1211 if (ret >= 0 && ret < aiocb->aio_nbytes) {
1212 iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
1213 0, aiocb->aio_nbytes - ret);
1215 ret = aiocb->aio_nbytes;
1217 if (ret == aiocb->aio_nbytes) {
1219 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1223 case QEMU_AIO_WRITE:
1224 ret = handle_aiocb_rw(aiocb);
1225 if (ret == aiocb->aio_nbytes) {
1227 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1231 case QEMU_AIO_FLUSH:
1232 ret = handle_aiocb_flush(aiocb);
1234 case QEMU_AIO_IOCTL:
1235 ret = handle_aiocb_ioctl(aiocb);
1237 case QEMU_AIO_DISCARD:
1238 ret = handle_aiocb_discard(aiocb);
1240 case QEMU_AIO_WRITE_ZEROES:
1241 ret = handle_aiocb_write_zeroes(aiocb);
1244 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1253 static int paio_submit_co(BlockDriverState *bs, int fd,
1254 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1257 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1261 acb->aio_type = type;
1262 acb->aio_fildes = fd;
1264 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1265 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1268 acb->aio_iov = qiov->iov;
1269 acb->aio_niov = qiov->niov;
1270 assert(qiov->size == acb->aio_nbytes);
1273 trace_paio_submit_co(sector_num, nb_sectors, type);
1274 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1275 return thread_pool_submit_co(pool, aio_worker, acb);
1278 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
1279 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1280 BlockCompletionFunc *cb, void *opaque, int type)
1282 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1286 acb->aio_type = type;
1287 acb->aio_fildes = fd;
1289 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1290 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1293 acb->aio_iov = qiov->iov;
1294 acb->aio_niov = qiov->niov;
1295 assert(qiov->size == acb->aio_nbytes);
1298 trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
1299 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1300 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1303 static BlockAIOCB *raw_aio_submit(BlockDriverState *bs,
1304 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1305 BlockCompletionFunc *cb, void *opaque, int type)
1307 BDRVRawState *s = bs->opaque;
1309 if (fd_open(bs) < 0)
1313 * Check if the underlying device requires requests to be aligned,
1314 * and if the request we are trying to submit is aligned or not.
1315 * If this is the case tell the low-level driver that it needs
1316 * to copy the buffer.
1318 if (s->needs_alignment) {
1319 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1320 type |= QEMU_AIO_MISALIGNED;
1321 #ifdef CONFIG_LINUX_AIO
1322 } else if (s->use_aio) {
1323 return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
1324 nb_sectors, cb, opaque, type);
1329 return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
1333 static void raw_aio_plug(BlockDriverState *bs)
1335 #ifdef CONFIG_LINUX_AIO
1336 BDRVRawState *s = bs->opaque;
1338 laio_io_plug(bs, s->aio_ctx);
1343 static void raw_aio_unplug(BlockDriverState *bs)
1345 #ifdef CONFIG_LINUX_AIO
1346 BDRVRawState *s = bs->opaque;
1348 laio_io_unplug(bs, s->aio_ctx, true);
1353 static void raw_aio_flush_io_queue(BlockDriverState *bs)
1355 #ifdef CONFIG_LINUX_AIO
1356 BDRVRawState *s = bs->opaque;
1358 laio_io_unplug(bs, s->aio_ctx, false);
1363 static BlockAIOCB *raw_aio_readv(BlockDriverState *bs,
1364 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1365 BlockCompletionFunc *cb, void *opaque)
1367 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1368 cb, opaque, QEMU_AIO_READ);
1371 static BlockAIOCB *raw_aio_writev(BlockDriverState *bs,
1372 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1373 BlockCompletionFunc *cb, void *opaque)
1375 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1376 cb, opaque, QEMU_AIO_WRITE);
1379 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
1380 BlockCompletionFunc *cb, void *opaque)
1382 BDRVRawState *s = bs->opaque;
1384 if (fd_open(bs) < 0)
1387 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1390 static void raw_close(BlockDriverState *bs)
1392 BDRVRawState *s = bs->opaque;
1394 raw_detach_aio_context(bs);
1396 #ifdef CONFIG_LINUX_AIO
1398 laio_cleanup(s->aio_ctx);
1407 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1409 BDRVRawState *s = bs->opaque;
1412 if (fstat(s->fd, &st)) {
1416 if (S_ISREG(st.st_mode)) {
1417 if (ftruncate(s->fd, offset) < 0) {
1420 } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1421 if (offset > raw_getlength(bs)) {
1432 static int64_t raw_getlength(BlockDriverState *bs)
1434 BDRVRawState *s = bs->opaque;
1440 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1441 struct disklabel dl;
1443 if (ioctl(fd, DIOCGDINFO, &dl))
1445 return (uint64_t)dl.d_secsize *
1446 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1450 #elif defined(__NetBSD__)
1451 static int64_t raw_getlength(BlockDriverState *bs)
1453 BDRVRawState *s = bs->opaque;
1459 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1460 struct dkwedge_info dkw;
1462 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1463 return dkw.dkw_size * 512;
1465 struct disklabel dl;
1467 if (ioctl(fd, DIOCGDINFO, &dl))
1469 return (uint64_t)dl.d_secsize *
1470 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1475 #elif defined(__sun__)
1476 static int64_t raw_getlength(BlockDriverState *bs)
1478 BDRVRawState *s = bs->opaque;
1479 struct dk_minfo minfo;
1489 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1491 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1493 return minfo.dki_lbsize * minfo.dki_capacity;
1497 * There are reports that lseek on some devices fails, but
1498 * irc discussion said that contingency on contingency was overkill.
1500 size = lseek(s->fd, 0, SEEK_END);
1506 #elif defined(CONFIG_BSD)
1507 static int64_t raw_getlength(BlockDriverState *bs)
1509 BDRVRawState *s = bs->opaque;
1513 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1522 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1525 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1526 #ifdef DIOCGMEDIASIZE
1527 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1528 #elif defined(DIOCGPART)
1531 if (ioctl(fd, DIOCGPART, &pi) == 0)
1532 size = pi.media_size;
1538 #if defined(__APPLE__) && defined(__MACH__)
1540 uint64_t sectors = 0;
1541 uint32_t sector_size = 0;
1543 if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0
1544 && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) {
1545 size = sectors * sector_size;
1547 size = lseek(fd, 0LL, SEEK_END);
1554 size = lseek(fd, 0LL, SEEK_END);
1559 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1562 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1563 if (size == 2048LL * (unsigned)-1)
1565 /* XXX no disc? maybe we need to reopen... */
1566 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1573 size = lseek(fd, 0, SEEK_END);
1581 static int64_t raw_getlength(BlockDriverState *bs)
1583 BDRVRawState *s = bs->opaque;
1592 size = lseek(s->fd, 0, SEEK_END);
1600 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1603 BDRVRawState *s = bs->opaque;
1605 if (fstat(s->fd, &st) < 0) {
1608 return (int64_t)st.st_blocks * 512;
1611 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
1615 int64_t total_size = 0;
1617 PreallocMode prealloc;
1619 Error *local_err = NULL;
1621 strstart(filename, "file:", &filename);
1623 /* Read out options */
1624 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1626 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
1627 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1628 prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
1629 PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
1633 error_propagate(errp, local_err);
1638 fd = qemu_open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY,
1642 error_setg_errno(errp, -result, "Could not create file");
1648 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1649 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1650 * will be ignored since any failure of this operation should not
1651 * block the left work.
1654 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
1655 attr |= FS_NOCOW_FL;
1656 ioctl(fd, FS_IOC_SETFLAGS, &attr);
1661 if (ftruncate(fd, total_size) != 0) {
1663 error_setg_errno(errp, -result, "Could not resize file");
1668 #ifdef CONFIG_POSIX_FALLOCATE
1669 case PREALLOC_MODE_FALLOC:
1670 /* posix_fallocate() doesn't set errno. */
1671 result = -posix_fallocate(fd, 0, total_size);
1673 error_setg_errno(errp, -result,
1674 "Could not preallocate data for the new file");
1678 case PREALLOC_MODE_FULL:
1680 int64_t num = 0, left = total_size;
1681 buf = g_malloc0(65536);
1684 num = MIN(left, 65536);
1685 result = write(fd, buf, num);
1688 error_setg_errno(errp, -result,
1689 "Could not write to the new file");
1698 error_setg_errno(errp, -result,
1699 "Could not flush new file to disk");
1705 case PREALLOC_MODE_OFF:
1709 error_setg(errp, "Unsupported preallocation mode: %s",
1710 PreallocMode_lookup[prealloc]);
1715 if (qemu_close(fd) != 0 && result == 0) {
1717 error_setg_errno(errp, -result, "Could not close the new file");
1724 * Find allocation range in @bs around offset @start.
1725 * May change underlying file descriptor's file offset.
1726 * If @start is not in a hole, store @start in @data, and the
1727 * beginning of the next hole in @hole, and return 0.
1728 * If @start is in a non-trailing hole, store @start in @hole and the
1729 * beginning of the next non-hole in @data, and return 0.
1730 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1731 * If we can't find out, return a negative errno other than -ENXIO.
1733 static int find_allocation(BlockDriverState *bs, off_t start,
1734 off_t *data, off_t *hole)
1736 #if defined SEEK_HOLE && defined SEEK_DATA
1737 BDRVRawState *s = bs->opaque;
1742 * D1. offs == start: start is in data
1743 * D2. offs > start: start is in a hole, next data at offs
1744 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1745 * or start is beyond EOF
1746 * If the latter happens, the file has been truncated behind
1747 * our back since we opened it. All bets are off then.
1748 * Treating like a trailing hole is simplest.
1749 * D4. offs < 0, errno != ENXIO: we learned nothing
1751 offs = lseek(s->fd, start, SEEK_DATA);
1753 return -errno; /* D3 or D4 */
1755 assert(offs >= start);
1758 /* D2: in hole, next data at offs */
1764 /* D1: in data, end not yet known */
1768 * H1. offs == start: start is in a hole
1769 * If this happens here, a hole has been dug behind our back
1770 * since the previous lseek().
1771 * H2. offs > start: either start is in data, next hole at offs,
1772 * or start is in trailing hole, EOF at offs
1773 * Linux treats trailing holes like any other hole: offs ==
1774 * start. Solaris seeks to EOF instead: offs > start (blech).
1775 * If that happens here, a hole has been dug behind our back
1776 * since the previous lseek().
1777 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1778 * If this happens, the file has been truncated behind our
1779 * back since we opened it. Treat it like a trailing hole.
1780 * H4. offs < 0, errno != ENXIO: we learned nothing
1781 * Pretend we know nothing at all, i.e. "forget" about D1.
1783 offs = lseek(s->fd, start, SEEK_HOLE);
1785 return -errno; /* D1 and (H3 or H4) */
1787 assert(offs >= start);
1791 * D1 and H2: either in data, next hole at offs, or it was in
1792 * data but is now in a trailing hole. In the latter case,
1793 * all bets are off. Treating it as if it there was data all
1794 * the way to EOF is safe, so simply do that.
1809 * Returns the allocation status of the specified sectors.
1811 * If 'sector_num' is beyond the end of the disk image the return value is 0
1812 * and 'pnum' is set to 0.
1814 * 'pnum' is set to the number of sectors (including and immediately following
1815 * the specified sector) that are known to be in the same
1816 * allocated/unallocated state.
1818 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1819 * beyond the end of the disk image it will be clamped.
1821 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1823 int nb_sectors, int *pnum,
1824 BlockDriverState **file)
1826 off_t start, data = 0, hole = 0;
1835 start = sector_num * BDRV_SECTOR_SIZE;
1836 total_size = bdrv_getlength(bs);
1837 if (total_size < 0) {
1839 } else if (start >= total_size) {
1842 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1843 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1846 ret = find_allocation(bs, start, &data, &hole);
1847 if (ret == -ENXIO) {
1850 ret = BDRV_BLOCK_ZERO;
1851 } else if (ret < 0) {
1852 /* No info available, so pretend there are no holes */
1854 ret = BDRV_BLOCK_DATA;
1855 } else if (data == start) {
1856 /* On a data extent, compute sectors to the end of the extent,
1857 * possibly including a partial sector at EOF. */
1858 *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1859 ret = BDRV_BLOCK_DATA;
1861 /* On a hole, compute sectors to the beginning of the next extent. */
1862 assert(hole == start);
1863 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1864 ret = BDRV_BLOCK_ZERO;
1867 return ret | BDRV_BLOCK_OFFSET_VALID | start;
1870 static coroutine_fn BlockAIOCB *raw_aio_discard(BlockDriverState *bs,
1871 int64_t sector_num, int nb_sectors,
1872 BlockCompletionFunc *cb, void *opaque)
1874 BDRVRawState *s = bs->opaque;
1876 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1877 cb, opaque, QEMU_AIO_DISCARD);
1880 static int coroutine_fn raw_co_write_zeroes(
1881 BlockDriverState *bs, int64_t sector_num,
1882 int nb_sectors, BdrvRequestFlags flags)
1884 BDRVRawState *s = bs->opaque;
1886 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1887 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1888 QEMU_AIO_WRITE_ZEROES);
1889 } else if (s->discard_zeroes) {
1890 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1896 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1898 BDRVRawState *s = bs->opaque;
1900 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1901 bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1905 static QemuOptsList raw_create_opts = {
1906 .name = "raw-create-opts",
1907 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
1910 .name = BLOCK_OPT_SIZE,
1911 .type = QEMU_OPT_SIZE,
1912 .help = "Virtual disk size"
1915 .name = BLOCK_OPT_NOCOW,
1916 .type = QEMU_OPT_BOOL,
1917 .help = "Turn off copy-on-write (valid only on btrfs)"
1920 .name = BLOCK_OPT_PREALLOC,
1921 .type = QEMU_OPT_STRING,
1922 .help = "Preallocation mode (allowed values: off, falloc, full)"
1924 { /* end of list */ }
1928 BlockDriver bdrv_file = {
1929 .format_name = "file",
1930 .protocol_name = "file",
1931 .instance_size = sizeof(BDRVRawState),
1932 .bdrv_needs_filename = true,
1933 .bdrv_probe = NULL, /* no probe for protocols */
1934 .bdrv_parse_filename = raw_parse_filename,
1935 .bdrv_file_open = raw_open,
1936 .bdrv_reopen_prepare = raw_reopen_prepare,
1937 .bdrv_reopen_commit = raw_reopen_commit,
1938 .bdrv_reopen_abort = raw_reopen_abort,
1939 .bdrv_close = raw_close,
1940 .bdrv_create = raw_create,
1941 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1942 .bdrv_co_get_block_status = raw_co_get_block_status,
1943 .bdrv_co_write_zeroes = raw_co_write_zeroes,
1945 .bdrv_aio_readv = raw_aio_readv,
1946 .bdrv_aio_writev = raw_aio_writev,
1947 .bdrv_aio_flush = raw_aio_flush,
1948 .bdrv_aio_discard = raw_aio_discard,
1949 .bdrv_refresh_limits = raw_refresh_limits,
1950 .bdrv_io_plug = raw_aio_plug,
1951 .bdrv_io_unplug = raw_aio_unplug,
1952 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
1954 .bdrv_truncate = raw_truncate,
1955 .bdrv_getlength = raw_getlength,
1956 .bdrv_get_info = raw_get_info,
1957 .bdrv_get_allocated_file_size
1958 = raw_get_allocated_file_size,
1960 .bdrv_detach_aio_context = raw_detach_aio_context,
1961 .bdrv_attach_aio_context = raw_attach_aio_context,
1963 .create_opts = &raw_create_opts,
1966 /***********************************************/
1969 #if defined(__APPLE__) && defined(__MACH__)
1970 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
1971 CFIndex maxPathSize, int flags);
1972 static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
1974 kern_return_t kernResult = KERN_FAILURE;
1975 mach_port_t masterPort;
1976 CFMutableDictionaryRef classesToMatch;
1977 const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
1978 char *mediaType = NULL;
1980 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1981 if ( KERN_SUCCESS != kernResult ) {
1982 printf( "IOMasterPort returned %d\n", kernResult );
1986 for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
1987 classesToMatch = IOServiceMatching(matching_array[index]);
1988 if (classesToMatch == NULL) {
1989 error_report("IOServiceMatching returned NULL for %s",
1990 matching_array[index]);
1993 CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
1995 kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch,
1997 if (kernResult != KERN_SUCCESS) {
1998 error_report("Note: IOServiceGetMatchingServices returned %d",
2003 /* If a match was found, leave the loop */
2004 if (*mediaIterator != 0) {
2005 DPRINTF("Matching using %s\n", matching_array[index]);
2006 mediaType = g_strdup(matching_array[index]);
2013 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
2014 CFIndex maxPathSize, int flags)
2016 io_object_t nextMedia;
2017 kern_return_t kernResult = KERN_FAILURE;
2019 nextMedia = IOIteratorNext( mediaIterator );
2022 CFTypeRef bsdPathAsCFString;
2023 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
2024 if ( bsdPathAsCFString ) {
2025 size_t devPathLength;
2026 strcpy( bsdPath, _PATH_DEV );
2027 if (flags & BDRV_O_NOCACHE) {
2028 strcat(bsdPath, "r");
2030 devPathLength = strlen( bsdPath );
2031 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
2032 kernResult = KERN_SUCCESS;
2034 CFRelease( bsdPathAsCFString );
2036 IOObjectRelease( nextMedia );
2042 /* Sets up a real cdrom for use in QEMU */
2043 static bool setup_cdrom(char *bsd_path, Error **errp)
2045 int index, num_of_test_partitions = 2, fd;
2046 char test_partition[MAXPATHLEN];
2047 bool partition_found = false;
2049 /* look for a working partition */
2050 for (index = 0; index < num_of_test_partitions; index++) {
2051 snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
2053 fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE);
2055 partition_found = true;
2061 /* if a working partition on the device was not found */
2062 if (partition_found == false) {
2063 error_setg(errp, "Failed to find a working partition on disc");
2065 DPRINTF("Using %s as optical disc\n", test_partition);
2066 pstrcpy(bsd_path, MAXPATHLEN, test_partition);
2068 return partition_found;
2071 /* Prints directions on mounting and unmounting a device */
2072 static void print_unmounting_directions(const char *file_name)
2074 error_report("If device %s is mounted on the desktop, unmount"
2075 " it first before using it in QEMU", file_name);
2076 error_report("Command to unmount device: diskutil unmountDisk %s",
2078 error_report("Command to mount device: diskutil mountDisk %s", file_name);
2081 #endif /* defined(__APPLE__) && defined(__MACH__) */
2083 static int hdev_probe_device(const char *filename)
2087 /* allow a dedicated CD-ROM driver to match with a higher priority */
2088 if (strstart(filename, "/dev/cdrom", NULL))
2091 if (stat(filename, &st) >= 0 &&
2092 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
2099 static int check_hdev_writable(BDRVRawState *s)
2101 #if defined(BLKROGET)
2102 /* Linux block devices can be configured "read-only" using blockdev(8).
2103 * This is independent of device node permissions and therefore open(2)
2104 * with O_RDWR succeeds. Actual writes fail with EPERM.
2106 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2107 * check for read-only block devices so that Linux block devices behave
2113 if (fstat(s->fd, &st)) {
2117 if (!S_ISBLK(st.st_mode)) {
2121 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
2128 #endif /* defined(BLKROGET) */
2132 static void hdev_parse_filename(const char *filename, QDict *options,
2135 /* The prefix is optional, just as for "file". */
2136 strstart(filename, "host_device:", &filename);
2138 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2141 static bool hdev_is_sg(BlockDriverState *bs)
2144 #if defined(__linux__)
2147 struct sg_scsi_id scsiid;
2150 if (stat(bs->filename, &st) >= 0 && S_ISCHR(st.st_mode) &&
2151 !bdrv_ioctl(bs, SG_GET_VERSION_NUM, &sg_version) &&
2152 !bdrv_ioctl(bs, SG_GET_SCSI_ID, &scsiid)) {
2153 DPRINTF("SG device found: type=%d, version=%d\n",
2154 scsiid.scsi_type, sg_version);
2163 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
2166 BDRVRawState *s = bs->opaque;
2167 Error *local_err = NULL;
2170 #if defined(__APPLE__) && defined(__MACH__)
2171 const char *filename = qdict_get_str(options, "filename");
2172 char bsd_path[MAXPATHLEN] = "";
2173 bool error_occurred = false;
2175 /* If using a real cdrom */
2176 if (strcmp(filename, "/dev/cdrom") == 0) {
2177 char *mediaType = NULL;
2178 kern_return_t ret_val;
2179 io_iterator_t mediaIterator = 0;
2181 mediaType = FindEjectableOpticalMedia(&mediaIterator);
2182 if (mediaType == NULL) {
2183 error_setg(errp, "Please make sure your CD/DVD is in the optical"
2185 error_occurred = true;
2186 goto hdev_open_Mac_error;
2189 ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
2190 if (ret_val != KERN_SUCCESS) {
2191 error_setg(errp, "Could not get BSD path for optical drive");
2192 error_occurred = true;
2193 goto hdev_open_Mac_error;
2196 /* If a real optical drive was not found */
2197 if (bsd_path[0] == '\0') {
2198 error_setg(errp, "Failed to obtain bsd path for optical drive");
2199 error_occurred = true;
2200 goto hdev_open_Mac_error;
2203 /* If using a cdrom disc and finding a partition on the disc failed */
2204 if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
2205 setup_cdrom(bsd_path, errp) == false) {
2206 print_unmounting_directions(bsd_path);
2207 error_occurred = true;
2208 goto hdev_open_Mac_error;
2211 qdict_put(options, "filename", qstring_from_str(bsd_path));
2213 hdev_open_Mac_error:
2215 if (mediaIterator) {
2216 IOObjectRelease(mediaIterator);
2218 if (error_occurred) {
2222 #endif /* defined(__APPLE__) && defined(__MACH__) */
2224 s->type = FTYPE_FILE;
2226 ret = raw_open_common(bs, options, flags, 0, &local_err);
2229 error_propagate(errp, local_err);
2231 #if defined(__APPLE__) && defined(__MACH__)
2233 filename = bsd_path;
2235 /* if a physical device experienced an error while being opened */
2236 if (strncmp(filename, "/dev/", 5) == 0) {
2237 print_unmounting_directions(filename);
2239 #endif /* defined(__APPLE__) && defined(__MACH__) */
2243 /* Since this does ioctl the device must be already opened */
2244 bs->sg = hdev_is_sg(bs);
2246 if (flags & BDRV_O_RDWR) {
2247 ret = check_hdev_writable(s);
2250 error_setg_errno(errp, -ret, "The device is not writable");
2258 #if defined(__linux__)
2260 static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
2261 unsigned long int req, void *buf,
2262 BlockCompletionFunc *cb, void *opaque)
2264 BDRVRawState *s = bs->opaque;
2265 RawPosixAIOData *acb;
2268 if (fd_open(bs) < 0)
2271 acb = g_new(RawPosixAIOData, 1);
2273 acb->aio_type = QEMU_AIO_IOCTL;
2274 acb->aio_fildes = s->fd;
2275 acb->aio_offset = 0;
2276 acb->aio_ioctl_buf = buf;
2277 acb->aio_ioctl_cmd = req;
2278 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
2279 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
2283 static int fd_open(BlockDriverState *bs)
2285 BDRVRawState *s = bs->opaque;
2287 /* this is just to ensure s->fd is sane (its called by io ops) */
2293 static coroutine_fn BlockAIOCB *hdev_aio_discard(BlockDriverState *bs,
2294 int64_t sector_num, int nb_sectors,
2295 BlockCompletionFunc *cb, void *opaque)
2297 BDRVRawState *s = bs->opaque;
2299 if (fd_open(bs) < 0) {
2302 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
2303 cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2306 static coroutine_fn int hdev_co_write_zeroes(BlockDriverState *bs,
2307 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2309 BDRVRawState *s = bs->opaque;
2316 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2317 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2318 QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2319 } else if (s->discard_zeroes) {
2320 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2321 QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2326 static int hdev_create(const char *filename, QemuOpts *opts,
2331 struct stat stat_buf;
2332 int64_t total_size = 0;
2335 /* This function is used by both protocol block drivers and therefore either
2336 * of these prefixes may be given.
2337 * The return value has to be stored somewhere, otherwise this is an error
2338 * due to -Werror=unused-value. */
2340 strstart(filename, "host_device:", &filename) ||
2341 strstart(filename, "host_cdrom:" , &filename);
2345 ret = raw_normalize_devicepath(&filename);
2347 error_setg_errno(errp, -ret, "Could not normalize device path");
2351 /* Read out options */
2352 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2355 fd = qemu_open(filename, O_WRONLY | O_BINARY);
2358 error_setg_errno(errp, -ret, "Could not open device");
2362 if (fstat(fd, &stat_buf) < 0) {
2364 error_setg_errno(errp, -ret, "Could not stat device");
2365 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2367 "The given file is neither a block nor a character device");
2369 } else if (lseek(fd, 0, SEEK_END) < total_size) {
2370 error_setg(errp, "Device is too small");
2378 static BlockDriver bdrv_host_device = {
2379 .format_name = "host_device",
2380 .protocol_name = "host_device",
2381 .instance_size = sizeof(BDRVRawState),
2382 .bdrv_needs_filename = true,
2383 .bdrv_probe_device = hdev_probe_device,
2384 .bdrv_parse_filename = hdev_parse_filename,
2385 .bdrv_file_open = hdev_open,
2386 .bdrv_close = raw_close,
2387 .bdrv_reopen_prepare = raw_reopen_prepare,
2388 .bdrv_reopen_commit = raw_reopen_commit,
2389 .bdrv_reopen_abort = raw_reopen_abort,
2390 .bdrv_create = hdev_create,
2391 .create_opts = &raw_create_opts,
2392 .bdrv_co_write_zeroes = hdev_co_write_zeroes,
2394 .bdrv_aio_readv = raw_aio_readv,
2395 .bdrv_aio_writev = raw_aio_writev,
2396 .bdrv_aio_flush = raw_aio_flush,
2397 .bdrv_aio_discard = hdev_aio_discard,
2398 .bdrv_refresh_limits = raw_refresh_limits,
2399 .bdrv_io_plug = raw_aio_plug,
2400 .bdrv_io_unplug = raw_aio_unplug,
2401 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2403 .bdrv_truncate = raw_truncate,
2404 .bdrv_getlength = raw_getlength,
2405 .bdrv_get_info = raw_get_info,
2406 .bdrv_get_allocated_file_size
2407 = raw_get_allocated_file_size,
2408 .bdrv_probe_blocksizes = hdev_probe_blocksizes,
2409 .bdrv_probe_geometry = hdev_probe_geometry,
2411 .bdrv_detach_aio_context = raw_detach_aio_context,
2412 .bdrv_attach_aio_context = raw_attach_aio_context,
2414 /* generic scsi device */
2416 .bdrv_aio_ioctl = hdev_aio_ioctl,
2420 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2421 static void cdrom_parse_filename(const char *filename, QDict *options,
2424 /* The prefix is optional, just as for "file". */
2425 strstart(filename, "host_cdrom:", &filename);
2427 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2432 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2435 BDRVRawState *s = bs->opaque;
2436 Error *local_err = NULL;
2441 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2442 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2444 error_propagate(errp, local_err);
2449 static int cdrom_probe_device(const char *filename)
2455 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2459 ret = fstat(fd, &st);
2460 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2464 /* Attempt to detect via a CDROM specific ioctl */
2465 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2475 static bool cdrom_is_inserted(BlockDriverState *bs)
2477 BDRVRawState *s = bs->opaque;
2480 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2481 return ret == CDS_DISC_OK;
2484 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2486 BDRVRawState *s = bs->opaque;
2489 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2490 perror("CDROMEJECT");
2492 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2493 perror("CDROMEJECT");
2497 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2499 BDRVRawState *s = bs->opaque;
2501 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2503 * Note: an error can happen if the distribution automatically
2506 /* perror("CDROM_LOCKDOOR"); */
2510 static BlockDriver bdrv_host_cdrom = {
2511 .format_name = "host_cdrom",
2512 .protocol_name = "host_cdrom",
2513 .instance_size = sizeof(BDRVRawState),
2514 .bdrv_needs_filename = true,
2515 .bdrv_probe_device = cdrom_probe_device,
2516 .bdrv_parse_filename = cdrom_parse_filename,
2517 .bdrv_file_open = cdrom_open,
2518 .bdrv_close = raw_close,
2519 .bdrv_reopen_prepare = raw_reopen_prepare,
2520 .bdrv_reopen_commit = raw_reopen_commit,
2521 .bdrv_reopen_abort = raw_reopen_abort,
2522 .bdrv_create = hdev_create,
2523 .create_opts = &raw_create_opts,
2525 .bdrv_aio_readv = raw_aio_readv,
2526 .bdrv_aio_writev = raw_aio_writev,
2527 .bdrv_aio_flush = raw_aio_flush,
2528 .bdrv_refresh_limits = raw_refresh_limits,
2529 .bdrv_io_plug = raw_aio_plug,
2530 .bdrv_io_unplug = raw_aio_unplug,
2531 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2533 .bdrv_truncate = raw_truncate,
2534 .bdrv_getlength = raw_getlength,
2535 .has_variable_length = true,
2536 .bdrv_get_allocated_file_size
2537 = raw_get_allocated_file_size,
2539 .bdrv_detach_aio_context = raw_detach_aio_context,
2540 .bdrv_attach_aio_context = raw_attach_aio_context,
2542 /* removable device support */
2543 .bdrv_is_inserted = cdrom_is_inserted,
2544 .bdrv_eject = cdrom_eject,
2545 .bdrv_lock_medium = cdrom_lock_medium,
2547 /* generic scsi device */
2548 .bdrv_aio_ioctl = hdev_aio_ioctl,
2550 #endif /* __linux__ */
2552 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2553 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2556 BDRVRawState *s = bs->opaque;
2557 Error *local_err = NULL;
2562 ret = raw_open_common(bs, options, flags, 0, &local_err);
2565 error_propagate(errp, local_err);
2570 /* make sure the door isn't locked at this time */
2571 ioctl(s->fd, CDIOCALLOW);
2575 static int cdrom_probe_device(const char *filename)
2577 if (strstart(filename, "/dev/cd", NULL) ||
2578 strstart(filename, "/dev/acd", NULL))
2583 static int cdrom_reopen(BlockDriverState *bs)
2585 BDRVRawState *s = bs->opaque;
2589 * Force reread of possibly changed/newly loaded disc,
2590 * FreeBSD seems to not notice sometimes...
2594 fd = qemu_open(bs->filename, s->open_flags, 0644);
2601 /* make sure the door isn't locked at this time */
2602 ioctl(s->fd, CDIOCALLOW);
2606 static bool cdrom_is_inserted(BlockDriverState *bs)
2608 return raw_getlength(bs) > 0;
2611 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2613 BDRVRawState *s = bs->opaque;
2618 (void) ioctl(s->fd, CDIOCALLOW);
2621 if (ioctl(s->fd, CDIOCEJECT) < 0)
2622 perror("CDIOCEJECT");
2624 if (ioctl(s->fd, CDIOCCLOSE) < 0)
2625 perror("CDIOCCLOSE");
2631 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2633 BDRVRawState *s = bs->opaque;
2637 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2639 * Note: an error can happen if the distribution automatically
2642 /* perror("CDROM_LOCKDOOR"); */
2646 static BlockDriver bdrv_host_cdrom = {
2647 .format_name = "host_cdrom",
2648 .protocol_name = "host_cdrom",
2649 .instance_size = sizeof(BDRVRawState),
2650 .bdrv_needs_filename = true,
2651 .bdrv_probe_device = cdrom_probe_device,
2652 .bdrv_parse_filename = cdrom_parse_filename,
2653 .bdrv_file_open = cdrom_open,
2654 .bdrv_close = raw_close,
2655 .bdrv_reopen_prepare = raw_reopen_prepare,
2656 .bdrv_reopen_commit = raw_reopen_commit,
2657 .bdrv_reopen_abort = raw_reopen_abort,
2658 .bdrv_create = hdev_create,
2659 .create_opts = &raw_create_opts,
2661 .bdrv_aio_readv = raw_aio_readv,
2662 .bdrv_aio_writev = raw_aio_writev,
2663 .bdrv_aio_flush = raw_aio_flush,
2664 .bdrv_refresh_limits = raw_refresh_limits,
2665 .bdrv_io_plug = raw_aio_plug,
2666 .bdrv_io_unplug = raw_aio_unplug,
2667 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2669 .bdrv_truncate = raw_truncate,
2670 .bdrv_getlength = raw_getlength,
2671 .has_variable_length = true,
2672 .bdrv_get_allocated_file_size
2673 = raw_get_allocated_file_size,
2675 .bdrv_detach_aio_context = raw_detach_aio_context,
2676 .bdrv_attach_aio_context = raw_attach_aio_context,
2678 /* removable device support */
2679 .bdrv_is_inserted = cdrom_is_inserted,
2680 .bdrv_eject = cdrom_eject,
2681 .bdrv_lock_medium = cdrom_lock_medium,
2683 #endif /* __FreeBSD__ */
2685 static void bdrv_file_init(void)
2688 * Register all the drivers. Note that order is important, the driver
2689 * registered last will get probed first.
2691 bdrv_register(&bdrv_file);
2692 bdrv_register(&bdrv_host_device);
2694 bdrv_register(&bdrv_host_cdrom);
2696 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2697 bdrv_register(&bdrv_host_cdrom);
2701 block_init(bdrv_file_init);