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
25 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu/cutils.h"
28 #include "qemu/error-report.h"
29 #include "block/block_int.h"
30 #include "qemu/module.h"
31 #include "qemu/option.h"
33 #include "block/thread-pool.h"
35 #include "block/raw-aio.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qstring.h"
39 #include "scsi/pr-manager.h"
40 #include "scsi/constants.h"
42 #if defined(__APPLE__) && (__MACH__)
44 #include <sys/param.h>
45 #include <IOKit/IOKitLib.h>
46 #include <IOKit/IOBSD.h>
47 #include <IOKit/storage/IOMediaBSDClient.h>
48 #include <IOKit/storage/IOMedia.h>
49 #include <IOKit/storage/IOCDMedia.h>
50 //#include <IOKit/storage/IOCDTypes.h>
51 #include <IOKit/storage/IODVDMedia.h>
52 #include <CoreFoundation/CoreFoundation.h>
56 #define _POSIX_PTHREAD_SEMANTICS 1
60 #include <sys/ioctl.h>
61 #include <sys/param.h>
62 #include <linux/cdrom.h>
65 #include <linux/hdreg.h>
71 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
74 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
75 #include <linux/falloc.h>
77 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
83 #include <sys/ioctl.h>
84 #include <sys/disklabel.h>
89 #include <sys/ioctl.h>
90 #include <sys/disklabel.h>
96 #include <sys/ioctl.h>
97 #include <sys/diskslice.h>
104 //#define DEBUG_BLOCK
107 # define DEBUG_BLOCK_PRINT 1
109 # define DEBUG_BLOCK_PRINT 0
111 #define DPRINTF(fmt, ...) \
113 if (DEBUG_BLOCK_PRINT) { \
114 printf(fmt, ## __VA_ARGS__); \
118 /* OS X does not have O_DSYNC */
121 #define O_DSYNC O_SYNC
122 #elif defined(O_FSYNC)
123 #define O_DSYNC O_FSYNC
127 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
129 #define O_DIRECT O_DSYNC
135 #define MAX_BLOCKSIZE 4096
137 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
138 * leaving a few more bytes for its future use. */
139 #define RAW_LOCK_PERM_BASE 100
140 #define RAW_LOCK_SHARED_BASE 200
142 typedef struct BDRVRawState {
150 /* The current permissions. */
152 uint64_t shared_perm;
158 bool has_write_zeroes:1;
159 bool discard_zeroes:1;
160 bool use_linux_aio:1;
161 bool page_cache_inconsistent:1;
163 bool needs_alignment;
168 typedef struct BDRVRawReopenState {
171 } BDRVRawReopenState;
173 static int fd_open(BlockDriverState *bs);
174 static int64_t raw_getlength(BlockDriverState *bs);
176 typedef struct RawPosixAIOData {
177 BlockDriverState *bs;
180 struct iovec *aio_iov;
185 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
190 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
191 static int cdrom_reopen(BlockDriverState *bs);
194 #if defined(__NetBSD__)
195 static int raw_normalize_devicepath(const char **filename)
197 static char namebuf[PATH_MAX];
198 const char *dp, *fname;
202 dp = strrchr(fname, '/');
203 if (lstat(fname, &sb) < 0) {
204 fprintf(stderr, "%s: stat failed: %s\n",
205 fname, strerror(errno));
209 if (!S_ISBLK(sb.st_mode)) {
214 snprintf(namebuf, PATH_MAX, "r%s", fname);
216 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
217 (int)(dp - fname), fname, dp + 1);
219 fprintf(stderr, "%s is a block device", fname);
221 fprintf(stderr, ", using %s\n", *filename);
226 static int raw_normalize_devicepath(const char **filename)
233 * Get logical block size via ioctl. On success store it in @sector_size_p.
235 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
237 unsigned int sector_size;
238 bool success = false;
242 static const unsigned long ioctl_list[] = {
246 #ifdef DKIOCGETBLOCKSIZE
249 #ifdef DIOCGSECTORSIZE
254 /* Try a few ioctls to get the right size */
255 for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) {
256 if (ioctl(fd, ioctl_list[i], §or_size) >= 0) {
257 *sector_size_p = sector_size;
262 return success ? 0 : -errno;
266 * Get physical block size of @fd.
267 * On success, store it in @blk_size and return 0.
268 * On failure, return -errno.
270 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
273 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
282 /* Check if read is allowed with given memory buffer and length.
284 * This function is used to check O_DIRECT memory buffer and request alignment.
286 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
288 ssize_t ret = pread(fd, buf, len, 0);
295 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
296 * other errors (e.g. real I/O error), which could happen on a failed
297 * drive, since we only care about probing alignment.
299 if (errno != EINVAL) {
307 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
309 BDRVRawState *s = bs->opaque;
311 size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize());
313 /* For SCSI generic devices the alignment is not really used.
314 With buffered I/O, we don't have any restrictions. */
315 if (bdrv_is_sg(bs) || !s->needs_alignment) {
316 bs->bl.request_alignment = 1;
321 bs->bl.request_alignment = 0;
323 /* Let's try to use the logical blocksize for the alignment. */
324 if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) {
325 bs->bl.request_alignment = 0;
330 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
331 bs->bl.request_alignment = da.d_miniosz;
332 /* The kernel returns wrong information for d_mem */
333 /* s->buf_align = da.d_mem; */
338 /* If we could not get the sizes so far, we can only guess them */
341 buf = qemu_memalign(max_align, 2 * max_align);
342 for (align = 512; align <= max_align; align <<= 1) {
343 if (raw_is_io_aligned(fd, buf + align, max_align)) {
344 s->buf_align = align;
351 if (!bs->bl.request_alignment) {
353 buf = qemu_memalign(s->buf_align, max_align);
354 for (align = 512; align <= max_align; align <<= 1) {
355 if (raw_is_io_aligned(fd, buf, align)) {
356 bs->bl.request_alignment = align;
363 if (!s->buf_align || !bs->bl.request_alignment) {
364 error_setg(errp, "Could not find working O_DIRECT alignment");
365 error_append_hint(errp, "Try cache.direct=off\n");
369 static void raw_parse_flags(int bdrv_flags, int *open_flags)
371 assert(open_flags != NULL);
373 *open_flags |= O_BINARY;
374 *open_flags &= ~O_ACCMODE;
375 if (bdrv_flags & BDRV_O_RDWR) {
376 *open_flags |= O_RDWR;
378 *open_flags |= O_RDONLY;
381 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
382 * and O_DIRECT for no caching. */
383 if ((bdrv_flags & BDRV_O_NOCACHE)) {
384 *open_flags |= O_DIRECT;
388 static void raw_parse_filename(const char *filename, QDict *options,
391 bdrv_parse_filename_strip_prefix(filename, "file:", options);
394 static QemuOptsList raw_runtime_opts = {
396 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
400 .type = QEMU_OPT_STRING,
401 .help = "File name of the image",
405 .type = QEMU_OPT_STRING,
406 .help = "host AIO implementation (threads, native)",
410 .type = QEMU_OPT_STRING,
411 .help = "file locking mode (on/off/auto, default: auto)",
414 .name = "pr-manager",
415 .type = QEMU_OPT_STRING,
416 .help = "id of persistent reservation manager object (default: none)",
418 { /* end of list */ }
422 static int raw_open_common(BlockDriverState *bs, QDict *options,
423 int bdrv_flags, int open_flags, Error **errp)
425 BDRVRawState *s = bs->opaque;
427 Error *local_err = NULL;
428 const char *filename = NULL;
430 BlockdevAioOptions aio, aio_default;
435 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
436 qemu_opts_absorb_qdict(opts, options, &local_err);
438 error_propagate(errp, local_err);
443 filename = qemu_opt_get(opts, "filename");
445 ret = raw_normalize_devicepath(&filename);
447 error_setg_errno(errp, -ret, "Could not normalize device path");
451 aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO)
452 ? BLOCKDEV_AIO_OPTIONS_NATIVE
453 : BLOCKDEV_AIO_OPTIONS_THREADS;
454 aio = qapi_enum_parse(&BlockdevAioOptions_lookup,
455 qemu_opt_get(opts, "aio"),
456 aio_default, &local_err);
458 error_propagate(errp, local_err);
462 s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE);
464 locking = qapi_enum_parse(&OnOffAuto_lookup,
465 qemu_opt_get(opts, "locking"),
466 ON_OFF_AUTO_AUTO, &local_err);
468 error_propagate(errp, local_err);
475 if (!qemu_has_ofd_lock()) {
477 "File lock requested but OFD locking syscall is "
478 "unavailable, falling back to POSIX file locks.\n"
479 "Due to the implementation, locks can be lost "
483 case ON_OFF_AUTO_OFF:
486 case ON_OFF_AUTO_AUTO:
487 s->use_lock = qemu_has_ofd_lock();
493 str = qemu_opt_get(opts, "pr-manager");
495 s->pr_mgr = pr_manager_lookup(str, &local_err);
497 error_propagate(errp, local_err);
503 s->open_flags = open_flags;
504 raw_parse_flags(bdrv_flags, &s->open_flags);
507 fd = qemu_open(filename, s->open_flags, 0644);
510 error_setg_errno(errp, errno, "Could not open '%s'", filename);
520 fd = qemu_open(filename, s->open_flags);
523 error_setg_errno(errp, errno, "Could not open '%s' for locking",
531 s->shared_perm = BLK_PERM_ALL;
533 #ifdef CONFIG_LINUX_AIO
534 /* Currently Linux does AIO only for files opened with O_DIRECT */
535 if (s->use_linux_aio && !(s->open_flags & O_DIRECT)) {
536 error_setg(errp, "aio=native was specified, but it requires "
537 "cache.direct=on, which was not specified.");
542 if (s->use_linux_aio) {
543 error_setg(errp, "aio=native was specified, but is not supported "
548 #endif /* !defined(CONFIG_LINUX_AIO) */
550 s->has_discard = true;
551 s->has_write_zeroes = true;
552 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
553 s->needs_alignment = true;
556 if (fstat(s->fd, &st) < 0) {
558 error_setg_errno(errp, errno, "Could not stat file");
561 if (S_ISREG(st.st_mode)) {
562 s->discard_zeroes = true;
563 s->has_fallocate = true;
565 if (S_ISBLK(st.st_mode)) {
566 #ifdef BLKDISCARDZEROES
568 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
569 s->discard_zeroes = true;
573 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
574 * not rely on the contents of discarded blocks unless using O_DIRECT.
575 * Same for BLKZEROOUT.
577 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
578 s->discard_zeroes = false;
579 s->has_write_zeroes = false;
584 if (S_ISCHR(st.st_mode)) {
586 * The file is a char device (disk), which on FreeBSD isn't behind
587 * a pager, so force all requests to be aligned. This is needed
588 * so QEMU makes sure all IO operations on the device are aligned
589 * to sector size, or else FreeBSD will reject them with EINVAL.
591 s->needs_alignment = true;
596 if (platform_test_xfs_fd(s->fd)) {
601 bs->supported_zero_flags = s->discard_zeroes ? BDRV_REQ_MAY_UNMAP : 0;
604 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
611 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
614 BDRVRawState *s = bs->opaque;
616 s->type = FTYPE_FILE;
617 return raw_open_common(bs, options, flags, 0, errp);
626 #define PERM_FOREACH(i) \
627 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
629 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
630 * file; if @unlock == true, also unlock the unneeded bytes.
631 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
633 static int raw_apply_lock_bytes(BDRVRawState *s,
634 uint64_t perm_lock_bits,
635 uint64_t shared_perm_lock_bits,
636 bool unlock, Error **errp)
642 int off = RAW_LOCK_PERM_BASE + i;
643 if (perm_lock_bits & (1ULL << i)) {
644 ret = qemu_lock_fd(s->lock_fd, off, 1, false);
646 error_setg(errp, "Failed to lock byte %d", off);
650 ret = qemu_unlock_fd(s->lock_fd, off, 1);
652 error_setg(errp, "Failed to unlock byte %d", off);
658 int off = RAW_LOCK_SHARED_BASE + i;
659 if (shared_perm_lock_bits & (1ULL << i)) {
660 ret = qemu_lock_fd(s->lock_fd, off, 1, false);
662 error_setg(errp, "Failed to lock byte %d", off);
666 ret = qemu_unlock_fd(s->lock_fd, off, 1);
668 error_setg(errp, "Failed to unlock byte %d", off);
676 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
677 static int raw_check_lock_bytes(BDRVRawState *s,
678 uint64_t perm, uint64_t shared_perm,
685 int off = RAW_LOCK_SHARED_BASE + i;
686 uint64_t p = 1ULL << i;
688 ret = qemu_lock_fd_test(s->lock_fd, off, 1, true);
690 char *perm_name = bdrv_perm_names(p);
692 "Failed to get \"%s\" lock",
695 error_append_hint(errp,
696 "Is another process using the image?\n");
702 int off = RAW_LOCK_PERM_BASE + i;
703 uint64_t p = 1ULL << i;
704 if (!(shared_perm & p)) {
705 ret = qemu_lock_fd_test(s->lock_fd, off, 1, true);
707 char *perm_name = bdrv_perm_names(p);
709 "Failed to get shared \"%s\" lock",
712 error_append_hint(errp,
713 "Is another process using the image?\n");
721 static int raw_handle_perm_lock(BlockDriverState *bs,
723 uint64_t new_perm, uint64_t new_shared,
726 BDRVRawState *s = bs->opaque;
728 Error *local_err = NULL;
734 if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) {
738 assert(s->lock_fd > 0);
742 ret = raw_apply_lock_bytes(s, s->perm | new_perm,
743 ~s->shared_perm | ~new_shared,
746 ret = raw_check_lock_bytes(s, new_perm, new_shared, errp);
752 /* fall through to unlock bytes. */
754 raw_apply_lock_bytes(s, s->perm, ~s->shared_perm, true, &local_err);
756 /* Theoretically the above call only unlocks bytes and it cannot
757 * fail. Something weird happened, report it.
759 error_report_err(local_err);
763 raw_apply_lock_bytes(s, new_perm, ~new_shared, true, &local_err);
765 /* Theoretically the above call only unlocks bytes and it cannot
766 * fail. Something weird happened, report it.
768 error_report_err(local_err);
775 static int raw_reopen_prepare(BDRVReopenState *state,
776 BlockReopenQueue *queue, Error **errp)
779 BDRVRawReopenState *rs;
781 Error *local_err = NULL;
783 assert(state != NULL);
784 assert(state->bs != NULL);
786 s = state->bs->opaque;
788 state->opaque = g_new0(BDRVRawReopenState, 1);
791 if (s->type == FTYPE_CD) {
792 rs->open_flags |= O_NONBLOCK;
795 raw_parse_flags(state->flags, &rs->open_flags);
799 int fcntl_flags = O_APPEND | O_NONBLOCK;
801 fcntl_flags |= O_NOATIME;
805 /* Not all operating systems have O_ASYNC, and those that don't
806 * will not let us track the state into rs->open_flags (typically
807 * you achieve the same effect with an ioctl, for example I_SETSIG
808 * on Solaris). But we do not use O_ASYNC, so that's fine.
810 assert((s->open_flags & O_ASYNC) == 0);
813 if ((rs->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
814 /* dup the original fd */
815 rs->fd = qemu_dup(s->fd);
817 ret = fcntl_setfl(rs->fd, rs->open_flags);
825 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
827 const char *normalized_filename = state->bs->filename;
828 ret = raw_normalize_devicepath(&normalized_filename);
830 error_setg_errno(errp, -ret, "Could not normalize device path");
832 assert(!(rs->open_flags & O_CREAT));
833 rs->fd = qemu_open(normalized_filename, rs->open_flags);
835 error_setg_errno(errp, errno, "Could not reopen file");
841 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
842 * alignment with the new fd. */
844 raw_probe_alignment(state->bs, rs->fd, &local_err);
848 error_propagate(errp, local_err);
856 static void raw_reopen_commit(BDRVReopenState *state)
858 BDRVRawReopenState *rs = state->opaque;
859 BDRVRawState *s = state->bs->opaque;
861 s->open_flags = rs->open_flags;
866 g_free(state->opaque);
867 state->opaque = NULL;
871 static void raw_reopen_abort(BDRVReopenState *state)
873 BDRVRawReopenState *rs = state->opaque;
875 /* nothing to do if NULL, we didn't get far enough */
884 g_free(state->opaque);
885 state->opaque = NULL;
888 static int hdev_get_max_transfer_length(BlockDriverState *bs, int fd)
892 short max_sectors = 0;
893 if (bs->sg && ioctl(fd, BLKSECTGET, &max_bytes) == 0) {
895 } else if (!bs->sg && ioctl(fd, BLKSECTGET, &max_sectors) == 0) {
896 return max_sectors << BDRV_SECTOR_BITS;
905 static int hdev_get_max_segments(const struct stat *st)
915 sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
916 major(st->st_rdev), minor(st->st_rdev));
917 fd = open(sysfspath, O_RDONLY);
923 ret = read(fd, buf, sizeof(buf) - 1);
924 } while (ret == -1 && errno == EINTR);
928 } else if (ret == 0) {
933 /* The file is ended with '\n', pass 'end' to accept that. */
934 ret = qemu_strtol(buf, &end, 10, &max_segments);
935 if (ret == 0 && end && *end == '\n') {
950 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
952 BDRVRawState *s = bs->opaque;
955 if (!fstat(s->fd, &st)) {
956 if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode)) {
957 int ret = hdev_get_max_transfer_length(bs, s->fd);
958 if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) {
959 bs->bl.max_transfer = pow2floor(ret);
961 ret = hdev_get_max_segments(&st);
963 bs->bl.max_transfer = MIN(bs->bl.max_transfer,
964 ret * getpagesize());
969 raw_probe_alignment(bs, s->fd, errp);
970 bs->bl.min_mem_alignment = s->buf_align;
971 bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
974 static int check_for_dasd(int fd)
977 struct dasd_information2_t info = {0};
979 return ioctl(fd, BIODASDINFO2, &info);
986 * Try to get @bs's logical and physical block size.
987 * On success, store them in @bsz and return zero.
988 * On failure, return negative errno.
990 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
992 BDRVRawState *s = bs->opaque;
995 /* If DASD, get blocksizes */
996 if (check_for_dasd(s->fd) < 0) {
999 ret = probe_logical_blocksize(s->fd, &bsz->log);
1003 return probe_physical_blocksize(s->fd, &bsz->phys);
1007 * Try to get @bs's geometry: cyls, heads, sectors.
1008 * On success, store them in @geo and return 0.
1009 * On failure return -errno.
1010 * (Allows block driver to assign default geometry values that guest sees)
1013 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1015 BDRVRawState *s = bs->opaque;
1016 struct hd_geometry ioctl_geo = {0};
1018 /* If DASD, get its geometry */
1019 if (check_for_dasd(s->fd) < 0) {
1022 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
1025 /* HDIO_GETGEO may return success even though geo contains zeros
1026 (e.g. certain multipath setups) */
1027 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
1030 /* Do not return a geometry for partition */
1031 if (ioctl_geo.start != 0) {
1034 geo->heads = ioctl_geo.heads;
1035 geo->sectors = ioctl_geo.sectors;
1036 geo->cylinders = ioctl_geo.cylinders;
1040 #else /* __linux__ */
1041 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1047 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
1051 ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
1059 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
1061 BDRVRawState *s = aiocb->bs->opaque;
1064 if (s->page_cache_inconsistent) {
1068 ret = qemu_fdatasync(aiocb->aio_fildes);
1070 /* There is no clear definition of the semantics of a failing fsync(),
1071 * so we may have to assume the worst. The sad truth is that this
1072 * assumption is correct for Linux. Some pages are now probably marked
1073 * clean in the page cache even though they are inconsistent with the
1074 * on-disk contents. The next fdatasync() call would succeed, but no
1075 * further writeback attempt will be made. We can't get back to a state
1076 * in which we know what is on disk (we would have to rewrite
1077 * everything that was touched since the last fdatasync() at least), so
1078 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1079 * really defined, I have little hope that other OSes are doing better.
1081 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1083 if ((s->open_flags & O_DIRECT) == 0) {
1084 s->page_cache_inconsistent = true;
1091 #ifdef CONFIG_PREADV
1093 static bool preadv_present = true;
1096 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1098 return preadv(fd, iov, nr_iov, offset);
1102 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1104 return pwritev(fd, iov, nr_iov, offset);
1109 static bool preadv_present = false;
1112 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1118 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1125 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
1130 if (aiocb->aio_type & QEMU_AIO_WRITE)
1131 len = qemu_pwritev(aiocb->aio_fildes,
1136 len = qemu_preadv(aiocb->aio_fildes,
1140 } while (len == -1 && errno == EINTR);
1149 * Read/writes the data to/from a given linear buffer.
1151 * Returns the number of bytes handles or -errno in case of an error. Short
1152 * reads are only returned if the end of the file is reached.
1154 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
1159 while (offset < aiocb->aio_nbytes) {
1160 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1161 len = pwrite(aiocb->aio_fildes,
1162 (const char *)buf + offset,
1163 aiocb->aio_nbytes - offset,
1164 aiocb->aio_offset + offset);
1166 len = pread(aiocb->aio_fildes,
1168 aiocb->aio_nbytes - offset,
1169 aiocb->aio_offset + offset);
1171 if (len == -1 && errno == EINTR) {
1173 } else if (len == -1 && errno == EINVAL &&
1174 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
1175 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
1177 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1178 * after a short read. Assume that O_DIRECT short reads only occur
1179 * at EOF. Therefore this is a short read, not an I/O error.
1182 } else if (len == -1) {
1185 } else if (len == 0) {
1194 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
1199 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
1201 * If there is just a single buffer, and it is properly aligned
1202 * we can just use plain pread/pwrite without any problems.
1204 if (aiocb->aio_niov == 1) {
1205 return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
1208 * We have more than one iovec, and all are properly aligned.
1210 * Try preadv/pwritev first and fall back to linearizing the
1211 * buffer if it's not supported.
1213 if (preadv_present) {
1214 nbytes = handle_aiocb_rw_vector(aiocb);
1215 if (nbytes == aiocb->aio_nbytes ||
1216 (nbytes < 0 && nbytes != -ENOSYS)) {
1219 preadv_present = false;
1223 * XXX(hch): short read/write. no easy way to handle the reminder
1224 * using these interfaces. For now retry using plain
1230 * Ok, we have to do it the hard way, copy all segments into
1231 * a single aligned buffer.
1233 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
1238 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1242 for (i = 0; i < aiocb->aio_niov; ++i) {
1243 memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
1244 p += aiocb->aio_iov[i].iov_len;
1246 assert(p - buf == aiocb->aio_nbytes);
1249 nbytes = handle_aiocb_rw_linear(aiocb, buf);
1250 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
1252 size_t count = aiocb->aio_nbytes, copy;
1255 for (i = 0; i < aiocb->aio_niov && count; ++i) {
1257 if (copy > aiocb->aio_iov[i].iov_len) {
1258 copy = aiocb->aio_iov[i].iov_len;
1260 memcpy(aiocb->aio_iov[i].iov_base, p, copy);
1261 assert(count >= copy);
1273 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
1275 struct xfs_flock64 fl;
1278 memset(&fl, 0, sizeof(fl));
1279 fl.l_whence = SEEK_SET;
1280 fl.l_start = offset;
1283 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
1285 DPRINTF("cannot write zero range (%s)\n", strerror(errno));
1292 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
1294 struct xfs_flock64 fl;
1297 memset(&fl, 0, sizeof(fl));
1298 fl.l_whence = SEEK_SET;
1299 fl.l_start = offset;
1302 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
1304 DPRINTF("cannot punch hole (%s)\n", strerror(errno));
1312 static int translate_err(int err)
1314 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1321 #ifdef CONFIG_FALLOCATE
1322 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1325 if (fallocate(fd, mode, offset, len) == 0) {
1328 } while (errno == EINTR);
1329 return translate_err(-errno);
1333 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1336 BDRVRawState *s = aiocb->bs->opaque;
1338 if (!s->has_write_zeroes) {
1344 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1345 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1348 } while (errno == EINTR);
1350 ret = translate_err(-errno);
1353 if (ret == -ENOTSUP) {
1354 s->has_write_zeroes = false;
1359 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
1361 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1362 BDRVRawState *s = aiocb->bs->opaque;
1364 #ifdef CONFIG_FALLOCATE
1368 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1369 return handle_aiocb_write_zeroes_block(aiocb);
1374 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
1378 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1379 if (s->has_write_zeroes) {
1380 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1381 aiocb->aio_offset, aiocb->aio_nbytes);
1382 if (ret == 0 || ret != -ENOTSUP) {
1385 s->has_write_zeroes = false;
1389 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1390 if (s->has_discard && s->has_fallocate) {
1391 int ret = do_fallocate(s->fd,
1392 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1393 aiocb->aio_offset, aiocb->aio_nbytes);
1395 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1396 if (ret == 0 || ret != -ENOTSUP) {
1399 s->has_fallocate = false;
1400 } else if (ret != -ENOTSUP) {
1403 s->has_discard = false;
1408 #ifdef CONFIG_FALLOCATE
1409 /* Last resort: we are trying to extend the file with zeroed data. This
1410 * can be done via fallocate(fd, 0) */
1411 len = bdrv_getlength(aiocb->bs);
1412 if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) {
1413 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1414 if (ret == 0 || ret != -ENOTSUP) {
1417 s->has_fallocate = false;
1424 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
1426 int ret = -EOPNOTSUPP;
1427 BDRVRawState *s = aiocb->bs->opaque;
1429 if (!s->has_discard) {
1433 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1436 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1437 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1440 } while (errno == EINTR);
1447 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1451 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1452 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1453 aiocb->aio_offset, aiocb->aio_nbytes);
1457 ret = translate_err(ret);
1458 if (ret == -ENOTSUP) {
1459 s->has_discard = false;
1464 static int aio_worker(void *arg)
1466 RawPosixAIOData *aiocb = arg;
1469 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
1471 ret = handle_aiocb_rw(aiocb);
1472 if (ret >= 0 && ret < aiocb->aio_nbytes) {
1473 iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
1474 0, aiocb->aio_nbytes - ret);
1476 ret = aiocb->aio_nbytes;
1478 if (ret == aiocb->aio_nbytes) {
1480 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1484 case QEMU_AIO_WRITE:
1485 ret = handle_aiocb_rw(aiocb);
1486 if (ret == aiocb->aio_nbytes) {
1488 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1492 case QEMU_AIO_FLUSH:
1493 ret = handle_aiocb_flush(aiocb);
1495 case QEMU_AIO_IOCTL:
1496 ret = handle_aiocb_ioctl(aiocb);
1498 case QEMU_AIO_DISCARD:
1499 ret = handle_aiocb_discard(aiocb);
1501 case QEMU_AIO_WRITE_ZEROES:
1502 ret = handle_aiocb_write_zeroes(aiocb);
1505 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1514 static int paio_submit_co(BlockDriverState *bs, int fd,
1515 int64_t offset, QEMUIOVector *qiov,
1516 int bytes, int type)
1518 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1522 acb->aio_type = type;
1523 acb->aio_fildes = fd;
1525 acb->aio_nbytes = bytes;
1526 acb->aio_offset = offset;
1529 acb->aio_iov = qiov->iov;
1530 acb->aio_niov = qiov->niov;
1531 assert(qiov->size == bytes);
1534 trace_paio_submit_co(offset, bytes, type);
1535 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1536 return thread_pool_submit_co(pool, aio_worker, acb);
1539 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
1540 int64_t offset, QEMUIOVector *qiov, int bytes,
1541 BlockCompletionFunc *cb, void *opaque, int type)
1543 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1547 acb->aio_type = type;
1548 acb->aio_fildes = fd;
1550 acb->aio_nbytes = bytes;
1551 acb->aio_offset = offset;
1554 acb->aio_iov = qiov->iov;
1555 acb->aio_niov = qiov->niov;
1556 assert(qiov->size == acb->aio_nbytes);
1559 trace_paio_submit(acb, opaque, offset, bytes, type);
1560 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1561 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1564 static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset,
1565 uint64_t bytes, QEMUIOVector *qiov, int type)
1567 BDRVRawState *s = bs->opaque;
1569 if (fd_open(bs) < 0)
1573 * Check if the underlying device requires requests to be aligned,
1574 * and if the request we are trying to submit is aligned or not.
1575 * If this is the case tell the low-level driver that it needs
1576 * to copy the buffer.
1578 if (s->needs_alignment) {
1579 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1580 type |= QEMU_AIO_MISALIGNED;
1581 #ifdef CONFIG_LINUX_AIO
1582 } else if (s->use_linux_aio) {
1583 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1584 assert(qiov->size == bytes);
1585 return laio_co_submit(bs, aio, s->fd, offset, qiov, type);
1590 return paio_submit_co(bs, s->fd, offset, qiov, bytes, type);
1593 static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset,
1594 uint64_t bytes, QEMUIOVector *qiov,
1597 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ);
1600 static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
1601 uint64_t bytes, QEMUIOVector *qiov,
1605 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE);
1608 static void raw_aio_plug(BlockDriverState *bs)
1610 #ifdef CONFIG_LINUX_AIO
1611 BDRVRawState *s = bs->opaque;
1612 if (s->use_linux_aio) {
1613 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1614 laio_io_plug(bs, aio);
1619 static void raw_aio_unplug(BlockDriverState *bs)
1621 #ifdef CONFIG_LINUX_AIO
1622 BDRVRawState *s = bs->opaque;
1623 if (s->use_linux_aio) {
1624 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1625 laio_io_unplug(bs, aio);
1630 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
1631 BlockCompletionFunc *cb, void *opaque)
1633 BDRVRawState *s = bs->opaque;
1635 if (fd_open(bs) < 0)
1638 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1641 static void raw_close(BlockDriverState *bs)
1643 BDRVRawState *s = bs->opaque;
1649 if (s->lock_fd >= 0) {
1650 qemu_close(s->lock_fd);
1656 * Truncates the given regular file @fd to @offset and, when growing, fills the
1657 * new space according to @prealloc.
1659 * Returns: 0 on success, -errno on failure.
1661 static int raw_regular_truncate(int fd, int64_t offset, PreallocMode prealloc,
1665 int64_t current_length = 0;
1669 if (fstat(fd, &st) < 0) {
1671 error_setg_errno(errp, -result, "Could not stat file");
1675 current_length = st.st_size;
1676 if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1677 error_setg(errp, "Cannot use preallocation for shrinking files");
1682 #ifdef CONFIG_POSIX_FALLOCATE
1683 case PREALLOC_MODE_FALLOC:
1685 * Truncating before posix_fallocate() makes it about twice slower on
1686 * file systems that do not support fallocate(), trying to check if a
1687 * block is allocated before allocating it, so don't do that here.
1689 if (offset != current_length) {
1690 result = -posix_fallocate(fd, current_length, offset - current_length);
1692 /* posix_fallocate() doesn't set errno. */
1693 error_setg_errno(errp, -result,
1694 "Could not preallocate new data");
1701 case PREALLOC_MODE_FULL:
1703 int64_t num = 0, left = offset - current_length;
1706 * Knowing the final size from the beginning could allow the file
1707 * system driver to do less allocations and possibly avoid
1708 * fragmentation of the file.
1710 if (ftruncate(fd, offset) != 0) {
1712 error_setg_errno(errp, -result, "Could not resize file");
1716 buf = g_malloc0(65536);
1718 result = lseek(fd, current_length, SEEK_SET);
1721 error_setg_errno(errp, -result,
1722 "Failed to seek to the old end of file");
1727 num = MIN(left, 65536);
1728 result = write(fd, buf, num);
1731 error_setg_errno(errp, -result,
1732 "Could not write zeros for preallocation");
1741 error_setg_errno(errp, -result,
1742 "Could not flush file to disk");
1748 case PREALLOC_MODE_OFF:
1749 if (ftruncate(fd, offset) != 0) {
1751 error_setg_errno(errp, -result, "Could not resize file");
1756 error_setg(errp, "Unsupported preallocation mode: %s",
1757 PreallocMode_str(prealloc));
1763 if (ftruncate(fd, current_length) < 0) {
1764 error_report("Failed to restore old file length: %s",
1773 static int raw_truncate(BlockDriverState *bs, int64_t offset,
1774 PreallocMode prealloc, Error **errp)
1776 BDRVRawState *s = bs->opaque;
1780 if (fstat(s->fd, &st)) {
1782 error_setg_errno(errp, -ret, "Failed to fstat() the file");
1786 if (S_ISREG(st.st_mode)) {
1787 return raw_regular_truncate(s->fd, offset, prealloc, errp);
1790 if (prealloc != PREALLOC_MODE_OFF) {
1791 error_setg(errp, "Preallocation mode '%s' unsupported for this "
1792 "non-regular file", PreallocMode_str(prealloc));
1796 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1797 if (offset > raw_getlength(bs)) {
1798 error_setg(errp, "Cannot grow device files");
1802 error_setg(errp, "Resizing this file is not supported");
1810 static int64_t raw_getlength(BlockDriverState *bs)
1812 BDRVRawState *s = bs->opaque;
1818 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1819 struct disklabel dl;
1821 if (ioctl(fd, DIOCGDINFO, &dl))
1823 return (uint64_t)dl.d_secsize *
1824 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1828 #elif defined(__NetBSD__)
1829 static int64_t raw_getlength(BlockDriverState *bs)
1831 BDRVRawState *s = bs->opaque;
1837 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1838 struct dkwedge_info dkw;
1840 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1841 return dkw.dkw_size * 512;
1843 struct disklabel dl;
1845 if (ioctl(fd, DIOCGDINFO, &dl))
1847 return (uint64_t)dl.d_secsize *
1848 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1853 #elif defined(__sun__)
1854 static int64_t raw_getlength(BlockDriverState *bs)
1856 BDRVRawState *s = bs->opaque;
1857 struct dk_minfo minfo;
1867 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1869 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1871 return minfo.dki_lbsize * minfo.dki_capacity;
1875 * There are reports that lseek on some devices fails, but
1876 * irc discussion said that contingency on contingency was overkill.
1878 size = lseek(s->fd, 0, SEEK_END);
1884 #elif defined(CONFIG_BSD)
1885 static int64_t raw_getlength(BlockDriverState *bs)
1887 BDRVRawState *s = bs->opaque;
1891 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1900 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1903 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1904 #ifdef DIOCGMEDIASIZE
1905 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1906 #elif defined(DIOCGPART)
1909 if (ioctl(fd, DIOCGPART, &pi) == 0)
1910 size = pi.media_size;
1916 #if defined(__APPLE__) && defined(__MACH__)
1918 uint64_t sectors = 0;
1919 uint32_t sector_size = 0;
1921 if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0
1922 && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) {
1923 size = sectors * sector_size;
1925 size = lseek(fd, 0LL, SEEK_END);
1932 size = lseek(fd, 0LL, SEEK_END);
1937 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1940 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1941 if (size == 2048LL * (unsigned)-1)
1943 /* XXX no disc? maybe we need to reopen... */
1944 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1951 size = lseek(fd, 0, SEEK_END);
1959 static int64_t raw_getlength(BlockDriverState *bs)
1961 BDRVRawState *s = bs->opaque;
1970 size = lseek(s->fd, 0, SEEK_END);
1978 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1981 BDRVRawState *s = bs->opaque;
1983 if (fstat(s->fd, &st) < 0) {
1986 return (int64_t)st.st_blocks * 512;
1989 static int raw_co_create(BlockdevCreateOptions *options, Error **errp)
1991 BlockdevCreateOptionsFile *file_opts;
1995 /* Validate options and set default values */
1996 assert(options->driver == BLOCKDEV_DRIVER_FILE);
1997 file_opts = &options->u.file;
1999 if (!file_opts->has_nocow) {
2000 file_opts->nocow = false;
2002 if (!file_opts->has_preallocation) {
2003 file_opts->preallocation = PREALLOC_MODE_OFF;
2007 fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY,
2011 error_setg_errno(errp, -result, "Could not create file");
2015 if (file_opts->nocow) {
2017 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2018 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2019 * will be ignored since any failure of this operation should not
2020 * block the left work.
2023 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
2024 attr |= FS_NOCOW_FL;
2025 ioctl(fd, FS_IOC_SETFLAGS, &attr);
2030 result = raw_regular_truncate(fd, file_opts->size, file_opts->preallocation,
2037 if (qemu_close(fd) != 0 && result == 0) {
2039 error_setg_errno(errp, -result, "Could not close the new file");
2045 static int coroutine_fn raw_co_create_opts(const char *filename, QemuOpts *opts,
2048 BlockdevCreateOptions options;
2049 int64_t total_size = 0;
2051 PreallocMode prealloc;
2053 Error *local_err = NULL;
2055 /* Skip file: protocol prefix */
2056 strstart(filename, "file:", &filename);
2058 /* Read out options */
2059 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2061 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
2062 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2063 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
2064 PREALLOC_MODE_OFF, &local_err);
2067 error_propagate(errp, local_err);
2071 options = (BlockdevCreateOptions) {
2072 .driver = BLOCKDEV_DRIVER_FILE,
2074 .filename = (char *) filename,
2076 .has_preallocation = true,
2077 .preallocation = prealloc,
2082 return raw_co_create(&options, errp);
2086 * Find allocation range in @bs around offset @start.
2087 * May change underlying file descriptor's file offset.
2088 * If @start is not in a hole, store @start in @data, and the
2089 * beginning of the next hole in @hole, and return 0.
2090 * If @start is in a non-trailing hole, store @start in @hole and the
2091 * beginning of the next non-hole in @data, and return 0.
2092 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2093 * If we can't find out, return a negative errno other than -ENXIO.
2095 static int find_allocation(BlockDriverState *bs, off_t start,
2096 off_t *data, off_t *hole)
2098 #if defined SEEK_HOLE && defined SEEK_DATA
2099 BDRVRawState *s = bs->opaque;
2104 * D1. offs == start: start is in data
2105 * D2. offs > start: start is in a hole, next data at offs
2106 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2107 * or start is beyond EOF
2108 * If the latter happens, the file has been truncated behind
2109 * our back since we opened it. All bets are off then.
2110 * Treating like a trailing hole is simplest.
2111 * D4. offs < 0, errno != ENXIO: we learned nothing
2113 offs = lseek(s->fd, start, SEEK_DATA);
2115 return -errno; /* D3 or D4 */
2117 assert(offs >= start);
2120 /* D2: in hole, next data at offs */
2126 /* D1: in data, end not yet known */
2130 * H1. offs == start: start is in a hole
2131 * If this happens here, a hole has been dug behind our back
2132 * since the previous lseek().
2133 * H2. offs > start: either start is in data, next hole at offs,
2134 * or start is in trailing hole, EOF at offs
2135 * Linux treats trailing holes like any other hole: offs ==
2136 * start. Solaris seeks to EOF instead: offs > start (blech).
2137 * If that happens here, a hole has been dug behind our back
2138 * since the previous lseek().
2139 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2140 * If this happens, the file has been truncated behind our
2141 * back since we opened it. Treat it like a trailing hole.
2142 * H4. offs < 0, errno != ENXIO: we learned nothing
2143 * Pretend we know nothing at all, i.e. "forget" about D1.
2145 offs = lseek(s->fd, start, SEEK_HOLE);
2147 return -errno; /* D1 and (H3 or H4) */
2149 assert(offs >= start);
2153 * D1 and H2: either in data, next hole at offs, or it was in
2154 * data but is now in a trailing hole. In the latter case,
2155 * all bets are off. Treating it as if it there was data all
2156 * the way to EOF is safe, so simply do that.
2171 * Returns the allocation status of the specified offset.
2173 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2175 * 'pnum' is set to the number of bytes (including and immediately following
2176 * the specified offset) that are known to be in the same
2177 * allocated/unallocated state.
2179 * 'bytes' is the max value 'pnum' should be set to.
2181 static int coroutine_fn raw_co_block_status(BlockDriverState *bs,
2184 int64_t bytes, int64_t *pnum,
2186 BlockDriverState **file)
2188 off_t data = 0, hole = 0;
2200 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
2203 ret = find_allocation(bs, offset, &data, &hole);
2204 if (ret == -ENXIO) {
2207 ret = BDRV_BLOCK_ZERO;
2208 } else if (ret < 0) {
2209 /* No info available, so pretend there are no holes */
2211 ret = BDRV_BLOCK_DATA;
2212 } else if (data == offset) {
2213 /* On a data extent, compute bytes to the end of the extent,
2214 * possibly including a partial sector at EOF. */
2215 *pnum = MIN(bytes, hole - offset);
2216 ret = BDRV_BLOCK_DATA;
2218 /* On a hole, compute bytes to the beginning of the next extent. */
2219 assert(hole == offset);
2220 *pnum = MIN(bytes, data - offset);
2221 ret = BDRV_BLOCK_ZERO;
2225 return ret | BDRV_BLOCK_OFFSET_VALID;
2228 static coroutine_fn BlockAIOCB *raw_aio_pdiscard(BlockDriverState *bs,
2229 int64_t offset, int bytes,
2230 BlockCompletionFunc *cb, void *opaque)
2232 BDRVRawState *s = bs->opaque;
2234 return paio_submit(bs, s->fd, offset, NULL, bytes,
2235 cb, opaque, QEMU_AIO_DISCARD);
2238 static int coroutine_fn raw_co_pwrite_zeroes(
2239 BlockDriverState *bs, int64_t offset,
2240 int bytes, BdrvRequestFlags flags)
2242 BDRVRawState *s = bs->opaque;
2244 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2245 return paio_submit_co(bs, s->fd, offset, NULL, bytes,
2246 QEMU_AIO_WRITE_ZEROES);
2247 } else if (s->discard_zeroes) {
2248 return paio_submit_co(bs, s->fd, offset, NULL, bytes,
2254 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2256 BDRVRawState *s = bs->opaque;
2258 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
2262 static QemuOptsList raw_create_opts = {
2263 .name = "raw-create-opts",
2264 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
2267 .name = BLOCK_OPT_SIZE,
2268 .type = QEMU_OPT_SIZE,
2269 .help = "Virtual disk size"
2272 .name = BLOCK_OPT_NOCOW,
2273 .type = QEMU_OPT_BOOL,
2274 .help = "Turn off copy-on-write (valid only on btrfs)"
2277 .name = BLOCK_OPT_PREALLOC,
2278 .type = QEMU_OPT_STRING,
2279 .help = "Preallocation mode (allowed values: off, falloc, full)"
2281 { /* end of list */ }
2285 static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared,
2288 return raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp);
2291 static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared)
2293 BDRVRawState *s = bs->opaque;
2294 raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL);
2296 s->shared_perm = shared;
2299 static void raw_abort_perm_update(BlockDriverState *bs)
2301 raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
2304 BlockDriver bdrv_file = {
2305 .format_name = "file",
2306 .protocol_name = "file",
2307 .instance_size = sizeof(BDRVRawState),
2308 .bdrv_needs_filename = true,
2309 .bdrv_probe = NULL, /* no probe for protocols */
2310 .bdrv_parse_filename = raw_parse_filename,
2311 .bdrv_file_open = raw_open,
2312 .bdrv_reopen_prepare = raw_reopen_prepare,
2313 .bdrv_reopen_commit = raw_reopen_commit,
2314 .bdrv_reopen_abort = raw_reopen_abort,
2315 .bdrv_close = raw_close,
2316 .bdrv_co_create = raw_co_create,
2317 .bdrv_co_create_opts = raw_co_create_opts,
2318 .bdrv_has_zero_init = bdrv_has_zero_init_1,
2319 .bdrv_co_block_status = raw_co_block_status,
2320 .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes,
2322 .bdrv_co_preadv = raw_co_preadv,
2323 .bdrv_co_pwritev = raw_co_pwritev,
2324 .bdrv_aio_flush = raw_aio_flush,
2325 .bdrv_aio_pdiscard = raw_aio_pdiscard,
2326 .bdrv_refresh_limits = raw_refresh_limits,
2327 .bdrv_io_plug = raw_aio_plug,
2328 .bdrv_io_unplug = raw_aio_unplug,
2330 .bdrv_truncate = raw_truncate,
2331 .bdrv_getlength = raw_getlength,
2332 .bdrv_get_info = raw_get_info,
2333 .bdrv_get_allocated_file_size
2334 = raw_get_allocated_file_size,
2335 .bdrv_check_perm = raw_check_perm,
2336 .bdrv_set_perm = raw_set_perm,
2337 .bdrv_abort_perm_update = raw_abort_perm_update,
2338 .create_opts = &raw_create_opts,
2341 /***********************************************/
2344 #if defined(__APPLE__) && defined(__MACH__)
2345 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
2346 CFIndex maxPathSize, int flags);
2347 static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
2349 kern_return_t kernResult = KERN_FAILURE;
2350 mach_port_t masterPort;
2351 CFMutableDictionaryRef classesToMatch;
2352 const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
2353 char *mediaType = NULL;
2355 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
2356 if ( KERN_SUCCESS != kernResult ) {
2357 printf( "IOMasterPort returned %d\n", kernResult );
2361 for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
2362 classesToMatch = IOServiceMatching(matching_array[index]);
2363 if (classesToMatch == NULL) {
2364 error_report("IOServiceMatching returned NULL for %s",
2365 matching_array[index]);
2368 CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
2370 kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch,
2372 if (kernResult != KERN_SUCCESS) {
2373 error_report("Note: IOServiceGetMatchingServices returned %d",
2378 /* If a match was found, leave the loop */
2379 if (*mediaIterator != 0) {
2380 DPRINTF("Matching using %s\n", matching_array[index]);
2381 mediaType = g_strdup(matching_array[index]);
2388 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
2389 CFIndex maxPathSize, int flags)
2391 io_object_t nextMedia;
2392 kern_return_t kernResult = KERN_FAILURE;
2394 nextMedia = IOIteratorNext( mediaIterator );
2397 CFTypeRef bsdPathAsCFString;
2398 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
2399 if ( bsdPathAsCFString ) {
2400 size_t devPathLength;
2401 strcpy( bsdPath, _PATH_DEV );
2402 if (flags & BDRV_O_NOCACHE) {
2403 strcat(bsdPath, "r");
2405 devPathLength = strlen( bsdPath );
2406 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
2407 kernResult = KERN_SUCCESS;
2409 CFRelease( bsdPathAsCFString );
2411 IOObjectRelease( nextMedia );
2417 /* Sets up a real cdrom for use in QEMU */
2418 static bool setup_cdrom(char *bsd_path, Error **errp)
2420 int index, num_of_test_partitions = 2, fd;
2421 char test_partition[MAXPATHLEN];
2422 bool partition_found = false;
2424 /* look for a working partition */
2425 for (index = 0; index < num_of_test_partitions; index++) {
2426 snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
2428 fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE);
2430 partition_found = true;
2436 /* if a working partition on the device was not found */
2437 if (partition_found == false) {
2438 error_setg(errp, "Failed to find a working partition on disc");
2440 DPRINTF("Using %s as optical disc\n", test_partition);
2441 pstrcpy(bsd_path, MAXPATHLEN, test_partition);
2443 return partition_found;
2446 /* Prints directions on mounting and unmounting a device */
2447 static void print_unmounting_directions(const char *file_name)
2449 error_report("If device %s is mounted on the desktop, unmount"
2450 " it first before using it in QEMU", file_name);
2451 error_report("Command to unmount device: diskutil unmountDisk %s",
2453 error_report("Command to mount device: diskutil mountDisk %s", file_name);
2456 #endif /* defined(__APPLE__) && defined(__MACH__) */
2458 static int hdev_probe_device(const char *filename)
2462 /* allow a dedicated CD-ROM driver to match with a higher priority */
2463 if (strstart(filename, "/dev/cdrom", NULL))
2466 if (stat(filename, &st) >= 0 &&
2467 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
2474 static int check_hdev_writable(BDRVRawState *s)
2476 #if defined(BLKROGET)
2477 /* Linux block devices can be configured "read-only" using blockdev(8).
2478 * This is independent of device node permissions and therefore open(2)
2479 * with O_RDWR succeeds. Actual writes fail with EPERM.
2481 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2482 * check for read-only block devices so that Linux block devices behave
2488 if (fstat(s->fd, &st)) {
2492 if (!S_ISBLK(st.st_mode)) {
2496 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
2503 #endif /* defined(BLKROGET) */
2507 static void hdev_parse_filename(const char *filename, QDict *options,
2510 bdrv_parse_filename_strip_prefix(filename, "host_device:", options);
2513 static bool hdev_is_sg(BlockDriverState *bs)
2516 #if defined(__linux__)
2518 BDRVRawState *s = bs->opaque;
2520 struct sg_scsi_id scsiid;
2524 if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) {
2528 ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version);
2533 ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid);
2535 DPRINTF("SG device found: type=%d, version=%d\n",
2536 scsiid.scsi_type, sg_version);
2545 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
2548 BDRVRawState *s = bs->opaque;
2549 Error *local_err = NULL;
2552 #if defined(__APPLE__) && defined(__MACH__)
2554 * Caution: while qdict_get_str() is fine, getting non-string types
2555 * would require more care. When @options come from -blockdev or
2556 * blockdev_add, its members are typed according to the QAPI
2557 * schema, but when they come from -drive, they're all QString.
2559 const char *filename = qdict_get_str(options, "filename");
2560 char bsd_path[MAXPATHLEN] = "";
2561 bool error_occurred = false;
2563 /* If using a real cdrom */
2564 if (strcmp(filename, "/dev/cdrom") == 0) {
2565 char *mediaType = NULL;
2566 kern_return_t ret_val;
2567 io_iterator_t mediaIterator = 0;
2569 mediaType = FindEjectableOpticalMedia(&mediaIterator);
2570 if (mediaType == NULL) {
2571 error_setg(errp, "Please make sure your CD/DVD is in the optical"
2573 error_occurred = true;
2574 goto hdev_open_Mac_error;
2577 ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
2578 if (ret_val != KERN_SUCCESS) {
2579 error_setg(errp, "Could not get BSD path for optical drive");
2580 error_occurred = true;
2581 goto hdev_open_Mac_error;
2584 /* If a real optical drive was not found */
2585 if (bsd_path[0] == '\0') {
2586 error_setg(errp, "Failed to obtain bsd path for optical drive");
2587 error_occurred = true;
2588 goto hdev_open_Mac_error;
2591 /* If using a cdrom disc and finding a partition on the disc failed */
2592 if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
2593 setup_cdrom(bsd_path, errp) == false) {
2594 print_unmounting_directions(bsd_path);
2595 error_occurred = true;
2596 goto hdev_open_Mac_error;
2599 qdict_put_str(options, "filename", bsd_path);
2601 hdev_open_Mac_error:
2603 if (mediaIterator) {
2604 IOObjectRelease(mediaIterator);
2606 if (error_occurred) {
2610 #endif /* defined(__APPLE__) && defined(__MACH__) */
2612 s->type = FTYPE_FILE;
2614 ret = raw_open_common(bs, options, flags, 0, &local_err);
2616 error_propagate(errp, local_err);
2617 #if defined(__APPLE__) && defined(__MACH__)
2619 filename = bsd_path;
2621 /* if a physical device experienced an error while being opened */
2622 if (strncmp(filename, "/dev/", 5) == 0) {
2623 print_unmounting_directions(filename);
2625 #endif /* defined(__APPLE__) && defined(__MACH__) */
2629 /* Since this does ioctl the device must be already opened */
2630 bs->sg = hdev_is_sg(bs);
2632 if (flags & BDRV_O_RDWR) {
2633 ret = check_hdev_writable(s);
2636 error_setg_errno(errp, -ret, "The device is not writable");
2644 #if defined(__linux__)
2646 static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
2647 unsigned long int req, void *buf,
2648 BlockCompletionFunc *cb, void *opaque)
2650 BDRVRawState *s = bs->opaque;
2651 RawPosixAIOData *acb;
2654 if (fd_open(bs) < 0)
2657 if (req == SG_IO && s->pr_mgr) {
2658 struct sg_io_hdr *io_hdr = buf;
2659 if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT ||
2660 io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) {
2661 return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs),
2662 s->fd, io_hdr, cb, opaque);
2666 acb = g_new(RawPosixAIOData, 1);
2668 acb->aio_type = QEMU_AIO_IOCTL;
2669 acb->aio_fildes = s->fd;
2670 acb->aio_offset = 0;
2671 acb->aio_ioctl_buf = buf;
2672 acb->aio_ioctl_cmd = req;
2673 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
2674 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
2678 static int fd_open(BlockDriverState *bs)
2680 BDRVRawState *s = bs->opaque;
2682 /* this is just to ensure s->fd is sane (its called by io ops) */
2688 static coroutine_fn BlockAIOCB *hdev_aio_pdiscard(BlockDriverState *bs,
2689 int64_t offset, int bytes,
2690 BlockCompletionFunc *cb, void *opaque)
2692 BDRVRawState *s = bs->opaque;
2694 if (fd_open(bs) < 0) {
2697 return paio_submit(bs, s->fd, offset, NULL, bytes,
2698 cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2701 static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs,
2702 int64_t offset, int bytes, BdrvRequestFlags flags)
2704 BDRVRawState *s = bs->opaque;
2711 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2712 return paio_submit_co(bs, s->fd, offset, NULL, bytes,
2713 QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2714 } else if (s->discard_zeroes) {
2715 return paio_submit_co(bs, s->fd, offset, NULL, bytes,
2716 QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2721 static int coroutine_fn hdev_co_create_opts(const char *filename, QemuOpts *opts,
2726 struct stat stat_buf;
2727 int64_t total_size = 0;
2730 /* This function is used by both protocol block drivers and therefore either
2731 * of these prefixes may be given.
2732 * The return value has to be stored somewhere, otherwise this is an error
2733 * due to -Werror=unused-value. */
2735 strstart(filename, "host_device:", &filename) ||
2736 strstart(filename, "host_cdrom:" , &filename);
2740 ret = raw_normalize_devicepath(&filename);
2742 error_setg_errno(errp, -ret, "Could not normalize device path");
2746 /* Read out options */
2747 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2750 fd = qemu_open(filename, O_WRONLY | O_BINARY);
2753 error_setg_errno(errp, -ret, "Could not open device");
2757 if (fstat(fd, &stat_buf) < 0) {
2759 error_setg_errno(errp, -ret, "Could not stat device");
2760 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2762 "The given file is neither a block nor a character device");
2764 } else if (lseek(fd, 0, SEEK_END) < total_size) {
2765 error_setg(errp, "Device is too small");
2769 if (!ret && total_size) {
2770 uint8_t buf[BDRV_SECTOR_SIZE] = { 0 };
2771 int64_t zero_size = MIN(BDRV_SECTOR_SIZE, total_size);
2772 if (lseek(fd, 0, SEEK_SET) == -1) {
2775 ret = qemu_write_full(fd, buf, zero_size);
2776 ret = ret == zero_size ? 0 : -errno;
2783 static BlockDriver bdrv_host_device = {
2784 .format_name = "host_device",
2785 .protocol_name = "host_device",
2786 .instance_size = sizeof(BDRVRawState),
2787 .bdrv_needs_filename = true,
2788 .bdrv_probe_device = hdev_probe_device,
2789 .bdrv_parse_filename = hdev_parse_filename,
2790 .bdrv_file_open = hdev_open,
2791 .bdrv_close = raw_close,
2792 .bdrv_reopen_prepare = raw_reopen_prepare,
2793 .bdrv_reopen_commit = raw_reopen_commit,
2794 .bdrv_reopen_abort = raw_reopen_abort,
2795 .bdrv_co_create_opts = hdev_co_create_opts,
2796 .create_opts = &raw_create_opts,
2797 .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes,
2799 .bdrv_co_preadv = raw_co_preadv,
2800 .bdrv_co_pwritev = raw_co_pwritev,
2801 .bdrv_aio_flush = raw_aio_flush,
2802 .bdrv_aio_pdiscard = hdev_aio_pdiscard,
2803 .bdrv_refresh_limits = raw_refresh_limits,
2804 .bdrv_io_plug = raw_aio_plug,
2805 .bdrv_io_unplug = raw_aio_unplug,
2807 .bdrv_truncate = raw_truncate,
2808 .bdrv_getlength = raw_getlength,
2809 .bdrv_get_info = raw_get_info,
2810 .bdrv_get_allocated_file_size
2811 = raw_get_allocated_file_size,
2812 .bdrv_check_perm = raw_check_perm,
2813 .bdrv_set_perm = raw_set_perm,
2814 .bdrv_abort_perm_update = raw_abort_perm_update,
2815 .bdrv_probe_blocksizes = hdev_probe_blocksizes,
2816 .bdrv_probe_geometry = hdev_probe_geometry,
2818 /* generic scsi device */
2820 .bdrv_aio_ioctl = hdev_aio_ioctl,
2824 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2825 static void cdrom_parse_filename(const char *filename, QDict *options,
2828 bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options);
2833 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2836 BDRVRawState *s = bs->opaque;
2840 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2841 return raw_open_common(bs, options, flags, O_NONBLOCK, errp);
2844 static int cdrom_probe_device(const char *filename)
2850 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2854 ret = fstat(fd, &st);
2855 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2859 /* Attempt to detect via a CDROM specific ioctl */
2860 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2870 static bool cdrom_is_inserted(BlockDriverState *bs)
2872 BDRVRawState *s = bs->opaque;
2875 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2876 return ret == CDS_DISC_OK;
2879 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2881 BDRVRawState *s = bs->opaque;
2884 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2885 perror("CDROMEJECT");
2887 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2888 perror("CDROMEJECT");
2892 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2894 BDRVRawState *s = bs->opaque;
2896 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2898 * Note: an error can happen if the distribution automatically
2901 /* perror("CDROM_LOCKDOOR"); */
2905 static BlockDriver bdrv_host_cdrom = {
2906 .format_name = "host_cdrom",
2907 .protocol_name = "host_cdrom",
2908 .instance_size = sizeof(BDRVRawState),
2909 .bdrv_needs_filename = true,
2910 .bdrv_probe_device = cdrom_probe_device,
2911 .bdrv_parse_filename = cdrom_parse_filename,
2912 .bdrv_file_open = cdrom_open,
2913 .bdrv_close = raw_close,
2914 .bdrv_reopen_prepare = raw_reopen_prepare,
2915 .bdrv_reopen_commit = raw_reopen_commit,
2916 .bdrv_reopen_abort = raw_reopen_abort,
2917 .bdrv_co_create_opts = hdev_co_create_opts,
2918 .create_opts = &raw_create_opts,
2921 .bdrv_co_preadv = raw_co_preadv,
2922 .bdrv_co_pwritev = raw_co_pwritev,
2923 .bdrv_aio_flush = raw_aio_flush,
2924 .bdrv_refresh_limits = raw_refresh_limits,
2925 .bdrv_io_plug = raw_aio_plug,
2926 .bdrv_io_unplug = raw_aio_unplug,
2928 .bdrv_truncate = raw_truncate,
2929 .bdrv_getlength = raw_getlength,
2930 .has_variable_length = true,
2931 .bdrv_get_allocated_file_size
2932 = raw_get_allocated_file_size,
2934 /* removable device support */
2935 .bdrv_is_inserted = cdrom_is_inserted,
2936 .bdrv_eject = cdrom_eject,
2937 .bdrv_lock_medium = cdrom_lock_medium,
2939 /* generic scsi device */
2940 .bdrv_aio_ioctl = hdev_aio_ioctl,
2942 #endif /* __linux__ */
2944 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2945 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2948 BDRVRawState *s = bs->opaque;
2949 Error *local_err = NULL;
2954 ret = raw_open_common(bs, options, flags, 0, &local_err);
2956 error_propagate(errp, local_err);
2960 /* make sure the door isn't locked at this time */
2961 ioctl(s->fd, CDIOCALLOW);
2965 static int cdrom_probe_device(const char *filename)
2967 if (strstart(filename, "/dev/cd", NULL) ||
2968 strstart(filename, "/dev/acd", NULL))
2973 static int cdrom_reopen(BlockDriverState *bs)
2975 BDRVRawState *s = bs->opaque;
2979 * Force reread of possibly changed/newly loaded disc,
2980 * FreeBSD seems to not notice sometimes...
2984 fd = qemu_open(bs->filename, s->open_flags, 0644);
2991 /* make sure the door isn't locked at this time */
2992 ioctl(s->fd, CDIOCALLOW);
2996 static bool cdrom_is_inserted(BlockDriverState *bs)
2998 return raw_getlength(bs) > 0;
3001 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
3003 BDRVRawState *s = bs->opaque;
3008 (void) ioctl(s->fd, CDIOCALLOW);
3011 if (ioctl(s->fd, CDIOCEJECT) < 0)
3012 perror("CDIOCEJECT");
3014 if (ioctl(s->fd, CDIOCCLOSE) < 0)
3015 perror("CDIOCCLOSE");
3021 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
3023 BDRVRawState *s = bs->opaque;
3027 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
3029 * Note: an error can happen if the distribution automatically
3032 /* perror("CDROM_LOCKDOOR"); */
3036 static BlockDriver bdrv_host_cdrom = {
3037 .format_name = "host_cdrom",
3038 .protocol_name = "host_cdrom",
3039 .instance_size = sizeof(BDRVRawState),
3040 .bdrv_needs_filename = true,
3041 .bdrv_probe_device = cdrom_probe_device,
3042 .bdrv_parse_filename = cdrom_parse_filename,
3043 .bdrv_file_open = cdrom_open,
3044 .bdrv_close = raw_close,
3045 .bdrv_reopen_prepare = raw_reopen_prepare,
3046 .bdrv_reopen_commit = raw_reopen_commit,
3047 .bdrv_reopen_abort = raw_reopen_abort,
3048 .bdrv_co_create_opts = hdev_co_create_opts,
3049 .create_opts = &raw_create_opts,
3051 .bdrv_co_preadv = raw_co_preadv,
3052 .bdrv_co_pwritev = raw_co_pwritev,
3053 .bdrv_aio_flush = raw_aio_flush,
3054 .bdrv_refresh_limits = raw_refresh_limits,
3055 .bdrv_io_plug = raw_aio_plug,
3056 .bdrv_io_unplug = raw_aio_unplug,
3058 .bdrv_truncate = raw_truncate,
3059 .bdrv_getlength = raw_getlength,
3060 .has_variable_length = true,
3061 .bdrv_get_allocated_file_size
3062 = raw_get_allocated_file_size,
3064 /* removable device support */
3065 .bdrv_is_inserted = cdrom_is_inserted,
3066 .bdrv_eject = cdrom_eject,
3067 .bdrv_lock_medium = cdrom_lock_medium,
3069 #endif /* __FreeBSD__ */
3071 static void bdrv_file_init(void)
3074 * Register all the drivers. Note that order is important, the driver
3075 * registered last will get probed first.
3077 bdrv_register(&bdrv_file);
3078 bdrv_register(&bdrv_host_device);
3080 bdrv_register(&bdrv_host_cdrom);
3082 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3083 bdrv_register(&bdrv_host_cdrom);
3087 block_init(bdrv_file_init);