2 * FUSE: Filesystem in Userspace
3 * Copyright (C) 2001-2007 Miklos Szeredi <miklos@szeredi.hu>
5 * This program can be distributed under the terms of the GNU GPLv2.
6 * See the file COPYING.
11 * This file system mirrors the existing file system hierarchy of the
12 * system, starting at the root file system. This is implemented by
13 * just "passing through" all requests to the corresponding user-space
14 * libc functions. In contrast to passthrough.c and passthrough_fh.c,
15 * this implementation uses the low-level API. Its performance should
16 * be the least bad among the three, but many operations are not
17 * implemented. In particular, it is not possible to remove files (or
18 * directories) because the code necessary to defer actual removal
19 * until the file is not opened anymore would make the example much
22 * When writeback caching is enabled (-o writeback mount option), it
23 * is only possible to write to files for which the mounting user has
24 * read permissions. This is because the writeback cache requires the
25 * kernel to be able to issue read requests for all files (which the
26 * passthrough filesystem cannot satisfy if it can't read the file in
27 * the underlying filesystem).
31 * gcc -Wall passthrough_ll.c `pkg-config fuse3 --cflags --libs` -o
35 * \include passthrough_ll.c
38 #include "qemu/osdep.h"
39 #include "qemu/timer.h"
40 #include "fuse_virtio.h"
42 #include "fuse_lowlevel.h"
57 #include <sys/mount.h>
58 #include <sys/prctl.h>
59 #include <sys/resource.h>
60 #include <sys/syscall.h>
61 #include <sys/types.h>
63 #include <sys/xattr.h>
67 #include "passthrough_helpers.h"
68 #include "passthrough_seccomp.h"
70 /* Keep track of inode posix locks for each owner. */
71 struct lo_inode_plock
{
73 int fd
; /* fd for OFD locks */
78 struct lo_inode
*inode
;
86 /* Maps FUSE fh or ino values to internal objects */
88 struct lo_map_elem
*elems
;
102 * Atomic reference count for this object. The nlookup field holds a
103 * reference and release it when nlookup reaches 0.
110 * This counter keeps the inode alive during the FUSE session.
111 * Incremented when the FUSE inode number is sent in a reply
112 * (FUSE_LOOKUP, FUSE_READDIRPLUS, etc). Decremented when an inode is
113 * released by requests like FUSE_FORGET, FUSE_RMDIR, FUSE_RENAME, etc.
115 * Note that this value is untrusted because the client can manipulate
116 * it arbitrarily using FUSE_FORGET requests.
118 * Protected by lo->mutex.
123 pthread_mutex_t plock_mutex
;
124 GHashTable
*posix_locks
; /* protected by lo_inode->plock_mutex */
141 pthread_mutex_t mutex
;
153 int readdirplus_clear
;
154 struct lo_inode root
;
155 GHashTable
*inodes
; /* protected by lo->mutex */
156 struct lo_map ino_map
; /* protected by lo->mutex */
157 struct lo_map dirp_map
; /* protected by lo->mutex */
158 struct lo_map fd_map
; /* protected by lo->mutex */
160 /* An O_PATH file descriptor to /proc/self/fd/ */
164 static const struct fuse_opt lo_opts
[] = {
165 { "writeback", offsetof(struct lo_data
, writeback
), 1 },
166 { "no_writeback", offsetof(struct lo_data
, writeback
), 0 },
167 { "source=%s", offsetof(struct lo_data
, source
), 0 },
168 { "flock", offsetof(struct lo_data
, flock
), 1 },
169 { "no_flock", offsetof(struct lo_data
, flock
), 0 },
170 { "posix_lock", offsetof(struct lo_data
, posix_lock
), 1 },
171 { "no_posix_lock", offsetof(struct lo_data
, posix_lock
), 0 },
172 { "xattr", offsetof(struct lo_data
, xattr
), 1 },
173 { "no_xattr", offsetof(struct lo_data
, xattr
), 0 },
174 { "modcaps=%s", offsetof(struct lo_data
, modcaps
), 0 },
175 { "timeout=%lf", offsetof(struct lo_data
, timeout
), 0 },
176 { "timeout=", offsetof(struct lo_data
, timeout_set
), 1 },
177 { "cache=none", offsetof(struct lo_data
, cache
), CACHE_NONE
},
178 { "cache=auto", offsetof(struct lo_data
, cache
), CACHE_AUTO
},
179 { "cache=always", offsetof(struct lo_data
, cache
), CACHE_ALWAYS
},
180 { "readdirplus", offsetof(struct lo_data
, readdirplus_set
), 1 },
181 { "no_readdirplus", offsetof(struct lo_data
, readdirplus_clear
), 1 },
184 static bool use_syslog
= false;
185 static int current_log_level
;
186 static void unref_inode_lolocked(struct lo_data
*lo
, struct lo_inode
*inode
,
190 pthread_mutex_t mutex
;
193 /* That we loaded cap-ng in the current thread from the saved */
194 static __thread
bool cap_loaded
= 0;
196 static struct lo_inode
*lo_find(struct lo_data
*lo
, struct stat
*st
);
198 static int is_dot_or_dotdot(const char *name
)
200 return name
[0] == '.' &&
201 (name
[1] == '\0' || (name
[1] == '.' && name
[2] == '\0'));
204 /* Is `path` a single path component that is not "." or ".."? */
205 static int is_safe_path_component(const char *path
)
207 if (strchr(path
, '/')) {
211 return !is_dot_or_dotdot(path
);
214 static struct lo_data
*lo_data(fuse_req_t req
)
216 return (struct lo_data
*)fuse_req_userdata(req
);
220 * Load capng's state from our saved state if the current thread
221 * hadn't previously been loaded.
222 * returns 0 on success
224 static int load_capng(void)
227 pthread_mutex_lock(&cap
.mutex
);
228 capng_restore_state(&cap
.saved
);
230 * restore_state free's the saved copy
233 cap
.saved
= capng_save_state();
235 pthread_mutex_unlock(&cap
.mutex
);
236 fuse_log(FUSE_LOG_ERR
, "capng_save_state (thread)\n");
239 pthread_mutex_unlock(&cap
.mutex
);
242 * We want to use the loaded state for our pid,
245 capng_setpid(syscall(SYS_gettid
));
252 * Helpers for dropping and regaining effective capabilities. Returns 0
253 * on success, error otherwise
255 static int drop_effective_cap(const char *cap_name
, bool *cap_dropped
)
259 cap
= capng_name_to_capability(cap_name
);
262 fuse_log(FUSE_LOG_ERR
, "capng_name_to_capability(%s) failed:%s\n",
263 cap_name
, strerror(errno
));
269 fuse_log(FUSE_LOG_ERR
, "load_capng() failed\n");
273 /* We dont have this capability in effective set already. */
274 if (!capng_have_capability(CAPNG_EFFECTIVE
, cap
)) {
279 if (capng_update(CAPNG_DROP
, CAPNG_EFFECTIVE
, cap
)) {
281 fuse_log(FUSE_LOG_ERR
, "capng_update(DROP,) failed\n");
285 if (capng_apply(CAPNG_SELECT_CAPS
)) {
287 fuse_log(FUSE_LOG_ERR
, "drop:capng_apply() failed\n");
300 static int gain_effective_cap(const char *cap_name
)
305 cap
= capng_name_to_capability(cap_name
);
308 fuse_log(FUSE_LOG_ERR
, "capng_name_to_capability(%s) failed:%s\n",
309 cap_name
, strerror(errno
));
315 fuse_log(FUSE_LOG_ERR
, "load_capng() failed\n");
319 if (capng_update(CAPNG_ADD
, CAPNG_EFFECTIVE
, cap
)) {
321 fuse_log(FUSE_LOG_ERR
, "capng_update(ADD,) failed\n");
325 if (capng_apply(CAPNG_SELECT_CAPS
)) {
327 fuse_log(FUSE_LOG_ERR
, "gain:capng_apply() failed\n");
336 static void lo_map_init(struct lo_map
*map
)
343 static void lo_map_destroy(struct lo_map
*map
)
348 static int lo_map_grow(struct lo_map
*map
, size_t new_nelems
)
350 struct lo_map_elem
*new_elems
;
353 if (new_nelems
<= map
->nelems
) {
357 new_elems
= realloc(map
->elems
, sizeof(map
->elems
[0]) * new_nelems
);
362 for (i
= map
->nelems
; i
< new_nelems
; i
++) {
363 new_elems
[i
].freelist
= i
+ 1;
364 new_elems
[i
].in_use
= false;
366 new_elems
[new_nelems
- 1].freelist
= -1;
368 map
->elems
= new_elems
;
369 map
->freelist
= map
->nelems
;
370 map
->nelems
= new_nelems
;
374 static struct lo_map_elem
*lo_map_alloc_elem(struct lo_map
*map
)
376 struct lo_map_elem
*elem
;
378 if (map
->freelist
== -1 && !lo_map_grow(map
, map
->nelems
+ 256)) {
382 elem
= &map
->elems
[map
->freelist
];
383 map
->freelist
= elem
->freelist
;
390 static struct lo_map_elem
*lo_map_reserve(struct lo_map
*map
, size_t key
)
394 if (!lo_map_grow(map
, key
+ 1)) {
398 for (prev
= &map
->freelist
; *prev
!= -1;
399 prev
= &map
->elems
[*prev
].freelist
) {
401 struct lo_map_elem
*elem
= &map
->elems
[key
];
403 *prev
= elem
->freelist
;
411 static struct lo_map_elem
*lo_map_get(struct lo_map
*map
, size_t key
)
413 if (key
>= map
->nelems
) {
416 if (!map
->elems
[key
].in_use
) {
419 return &map
->elems
[key
];
422 static void lo_map_remove(struct lo_map
*map
, size_t key
)
424 struct lo_map_elem
*elem
;
426 if (key
>= map
->nelems
) {
430 elem
= &map
->elems
[key
];
435 elem
->in_use
= false;
437 elem
->freelist
= map
->freelist
;
441 /* Assumes lo->mutex is held */
442 static ssize_t
lo_add_fd_mapping(fuse_req_t req
, int fd
)
444 struct lo_map_elem
*elem
;
446 elem
= lo_map_alloc_elem(&lo_data(req
)->fd_map
);
452 return elem
- lo_data(req
)->fd_map
.elems
;
455 /* Assumes lo->mutex is held */
456 static ssize_t
lo_add_dirp_mapping(fuse_req_t req
, struct lo_dirp
*dirp
)
458 struct lo_map_elem
*elem
;
460 elem
= lo_map_alloc_elem(&lo_data(req
)->dirp_map
);
466 return elem
- lo_data(req
)->dirp_map
.elems
;
469 /* Assumes lo->mutex is held */
470 static ssize_t
lo_add_inode_mapping(fuse_req_t req
, struct lo_inode
*inode
)
472 struct lo_map_elem
*elem
;
474 elem
= lo_map_alloc_elem(&lo_data(req
)->ino_map
);
480 return elem
- lo_data(req
)->ino_map
.elems
;
483 static void lo_inode_put(struct lo_data
*lo
, struct lo_inode
**inodep
)
485 struct lo_inode
*inode
= *inodep
;
493 if (g_atomic_int_dec_and_test(&inode
->refcount
)) {
499 /* Caller must release refcount using lo_inode_put() */
500 static struct lo_inode
*lo_inode(fuse_req_t req
, fuse_ino_t ino
)
502 struct lo_data
*lo
= lo_data(req
);
503 struct lo_map_elem
*elem
;
505 pthread_mutex_lock(&lo
->mutex
);
506 elem
= lo_map_get(&lo
->ino_map
, ino
);
508 g_atomic_int_inc(&elem
->inode
->refcount
);
510 pthread_mutex_unlock(&lo
->mutex
);
520 * TODO Remove this helper and force callers to hold an inode refcount until
521 * they are done with the fd. This will be done in a later patch to make
524 static int lo_fd(fuse_req_t req
, fuse_ino_t ino
)
526 struct lo_inode
*inode
= lo_inode(req
, ino
);
534 lo_inode_put(lo_data(req
), &inode
);
538 static void lo_init(void *userdata
, struct fuse_conn_info
*conn
)
540 struct lo_data
*lo
= (struct lo_data
*)userdata
;
542 if (conn
->capable
& FUSE_CAP_EXPORT_SUPPORT
) {
543 conn
->want
|= FUSE_CAP_EXPORT_SUPPORT
;
546 if (lo
->writeback
&& conn
->capable
& FUSE_CAP_WRITEBACK_CACHE
) {
547 fuse_log(FUSE_LOG_DEBUG
, "lo_init: activating writeback\n");
548 conn
->want
|= FUSE_CAP_WRITEBACK_CACHE
;
550 if (conn
->capable
& FUSE_CAP_FLOCK_LOCKS
) {
552 fuse_log(FUSE_LOG_DEBUG
, "lo_init: activating flock locks\n");
553 conn
->want
|= FUSE_CAP_FLOCK_LOCKS
;
555 fuse_log(FUSE_LOG_DEBUG
, "lo_init: disabling flock locks\n");
556 conn
->want
&= ~FUSE_CAP_FLOCK_LOCKS
;
560 if (conn
->capable
& FUSE_CAP_POSIX_LOCKS
) {
561 if (lo
->posix_lock
) {
562 fuse_log(FUSE_LOG_DEBUG
, "lo_init: activating posix locks\n");
563 conn
->want
|= FUSE_CAP_POSIX_LOCKS
;
565 fuse_log(FUSE_LOG_DEBUG
, "lo_init: disabling posix locks\n");
566 conn
->want
&= ~FUSE_CAP_POSIX_LOCKS
;
570 if ((lo
->cache
== CACHE_NONE
&& !lo
->readdirplus_set
) ||
571 lo
->readdirplus_clear
) {
572 fuse_log(FUSE_LOG_DEBUG
, "lo_init: disabling readdirplus\n");
573 conn
->want
&= ~FUSE_CAP_READDIRPLUS
;
577 static void lo_getattr(fuse_req_t req
, fuse_ino_t ino
,
578 struct fuse_file_info
*fi
)
582 struct lo_data
*lo
= lo_data(req
);
587 fstatat(lo_fd(req
, ino
), "", &buf
, AT_EMPTY_PATH
| AT_SYMLINK_NOFOLLOW
);
589 return (void)fuse_reply_err(req
, errno
);
592 fuse_reply_attr(req
, &buf
, lo
->timeout
);
595 static int lo_fi_fd(fuse_req_t req
, struct fuse_file_info
*fi
)
597 struct lo_data
*lo
= lo_data(req
);
598 struct lo_map_elem
*elem
;
600 pthread_mutex_lock(&lo
->mutex
);
601 elem
= lo_map_get(&lo
->fd_map
, fi
->fh
);
602 pthread_mutex_unlock(&lo
->mutex
);
611 static void lo_setattr(fuse_req_t req
, fuse_ino_t ino
, struct stat
*attr
,
612 int valid
, struct fuse_file_info
*fi
)
616 struct lo_data
*lo
= lo_data(req
);
617 struct lo_inode
*inode
;
622 inode
= lo_inode(req
, ino
);
624 fuse_reply_err(req
, EBADF
);
630 /* If fi->fh is invalid we'll report EBADF later */
632 fd
= lo_fi_fd(req
, fi
);
635 if (valid
& FUSE_SET_ATTR_MODE
) {
637 res
= fchmod(fd
, attr
->st_mode
);
639 sprintf(procname
, "%i", ifd
);
640 res
= fchmodat(lo
->proc_self_fd
, procname
, attr
->st_mode
, 0);
646 if (valid
& (FUSE_SET_ATTR_UID
| FUSE_SET_ATTR_GID
)) {
647 uid_t uid
= (valid
& FUSE_SET_ATTR_UID
) ? attr
->st_uid
: (uid_t
)-1;
648 gid_t gid
= (valid
& FUSE_SET_ATTR_GID
) ? attr
->st_gid
: (gid_t
)-1;
650 res
= fchownat(ifd
, "", uid
, gid
, AT_EMPTY_PATH
| AT_SYMLINK_NOFOLLOW
);
655 if (valid
& FUSE_SET_ATTR_SIZE
) {
661 sprintf(procname
, "%i", ifd
);
662 truncfd
= openat(lo
->proc_self_fd
, procname
, O_RDWR
);
668 res
= ftruncate(truncfd
, attr
->st_size
);
678 if (valid
& (FUSE_SET_ATTR_ATIME
| FUSE_SET_ATTR_MTIME
)) {
679 struct timespec tv
[2];
683 tv
[0].tv_nsec
= UTIME_OMIT
;
684 tv
[1].tv_nsec
= UTIME_OMIT
;
686 if (valid
& FUSE_SET_ATTR_ATIME_NOW
) {
687 tv
[0].tv_nsec
= UTIME_NOW
;
688 } else if (valid
& FUSE_SET_ATTR_ATIME
) {
689 tv
[0] = attr
->st_atim
;
692 if (valid
& FUSE_SET_ATTR_MTIME_NOW
) {
693 tv
[1].tv_nsec
= UTIME_NOW
;
694 } else if (valid
& FUSE_SET_ATTR_MTIME
) {
695 tv
[1] = attr
->st_mtim
;
699 res
= futimens(fd
, tv
);
701 sprintf(procname
, "%i", inode
->fd
);
702 res
= utimensat(lo
->proc_self_fd
, procname
, tv
, 0);
708 lo_inode_put(lo
, &inode
);
710 return lo_getattr(req
, ino
, fi
);
714 lo_inode_put(lo
, &inode
);
715 fuse_reply_err(req
, saverr
);
718 static struct lo_inode
*lo_find(struct lo_data
*lo
, struct stat
*st
)
721 struct lo_key key
= {
726 pthread_mutex_lock(&lo
->mutex
);
727 p
= g_hash_table_lookup(lo
->inodes
, &key
);
729 assert(p
->nlookup
> 0);
731 g_atomic_int_inc(&p
->refcount
);
733 pthread_mutex_unlock(&lo
->mutex
);
738 /* value_destroy_func for posix_locks GHashTable */
739 static void posix_locks_value_destroy(gpointer data
)
741 struct lo_inode_plock
*plock
= data
;
744 * We had used open() for locks and had only one fd. So
745 * closing this fd should release all OFD locks.
752 * Increments nlookup and caller must release refcount using
753 * lo_inode_put(&parent).
755 static int lo_do_lookup(fuse_req_t req
, fuse_ino_t parent
, const char *name
,
756 struct fuse_entry_param
*e
)
761 struct lo_data
*lo
= lo_data(req
);
762 struct lo_inode
*inode
= NULL
;
763 struct lo_inode
*dir
= lo_inode(req
, parent
);
766 * name_to_handle_at() and open_by_handle_at() can reach here with fuse
767 * mount point in guest, but we don't have its inode info in the
774 memset(e
, 0, sizeof(*e
));
775 e
->attr_timeout
= lo
->timeout
;
776 e
->entry_timeout
= lo
->timeout
;
778 /* Do not allow escaping root directory */
779 if (dir
== &lo
->root
&& strcmp(name
, "..") == 0) {
783 newfd
= openat(dir
->fd
, name
, O_PATH
| O_NOFOLLOW
);
788 res
= fstatat(newfd
, "", &e
->attr
, AT_EMPTY_PATH
| AT_SYMLINK_NOFOLLOW
);
793 inode
= lo_find(lo
, &e
->attr
);
797 inode
= calloc(1, sizeof(struct lo_inode
));
802 /* cache only filetype */
803 inode
->filetype
= (e
->attr
.st_mode
& S_IFMT
);
806 * One for the caller and one for nlookup (released in
807 * unref_inode_lolocked())
809 g_atomic_int_set(&inode
->refcount
, 2);
813 inode
->key
.ino
= e
->attr
.st_ino
;
814 inode
->key
.dev
= e
->attr
.st_dev
;
815 pthread_mutex_init(&inode
->plock_mutex
, NULL
);
816 inode
->posix_locks
= g_hash_table_new_full(
817 g_direct_hash
, g_direct_equal
, NULL
, posix_locks_value_destroy
);
819 pthread_mutex_lock(&lo
->mutex
);
820 inode
->fuse_ino
= lo_add_inode_mapping(req
, inode
);
821 g_hash_table_insert(lo
->inodes
, &inode
->key
, inode
);
822 pthread_mutex_unlock(&lo
->mutex
);
824 e
->ino
= inode
->fuse_ino
;
825 lo_inode_put(lo
, &inode
);
826 lo_inode_put(lo
, &dir
);
828 fuse_log(FUSE_LOG_DEBUG
, " %lli/%s -> %lli\n", (unsigned long long)parent
,
829 name
, (unsigned long long)e
->ino
);
838 lo_inode_put(lo
, &inode
);
839 lo_inode_put(lo
, &dir
);
843 static void lo_lookup(fuse_req_t req
, fuse_ino_t parent
, const char *name
)
845 struct fuse_entry_param e
;
848 fuse_log(FUSE_LOG_DEBUG
, "lo_lookup(parent=%" PRIu64
", name=%s)\n", parent
,
852 * Don't use is_safe_path_component(), allow "." and ".." for NFS export
855 if (strchr(name
, '/')) {
856 fuse_reply_err(req
, EINVAL
);
860 err
= lo_do_lookup(req
, parent
, name
, &e
);
862 fuse_reply_err(req
, err
);
864 fuse_reply_entry(req
, &e
);
869 * On some archs, setres*id is limited to 2^16 but they
870 * provide setres*id32 variants that allow 2^32.
871 * Others just let setres*id do 2^32 anyway.
873 #ifdef SYS_setresgid32
874 #define OURSYS_setresgid SYS_setresgid32
876 #define OURSYS_setresgid SYS_setresgid
879 #ifdef SYS_setresuid32
880 #define OURSYS_setresuid SYS_setresuid32
882 #define OURSYS_setresuid SYS_setresuid
886 * Change to uid/gid of caller so that file is created with
887 * ownership of caller.
888 * TODO: What about selinux context?
890 static int lo_change_cred(fuse_req_t req
, struct lo_cred
*old
)
894 old
->euid
= geteuid();
895 old
->egid
= getegid();
897 res
= syscall(OURSYS_setresgid
, -1, fuse_req_ctx(req
)->gid
, -1);
902 res
= syscall(OURSYS_setresuid
, -1, fuse_req_ctx(req
)->uid
, -1);
904 int errno_save
= errno
;
906 syscall(OURSYS_setresgid
, -1, old
->egid
, -1);
913 /* Regain Privileges */
914 static void lo_restore_cred(struct lo_cred
*old
)
918 res
= syscall(OURSYS_setresuid
, -1, old
->euid
, -1);
920 fuse_log(FUSE_LOG_ERR
, "seteuid(%u): %m\n", old
->euid
);
924 res
= syscall(OURSYS_setresgid
, -1, old
->egid
, -1);
926 fuse_log(FUSE_LOG_ERR
, "setegid(%u): %m\n", old
->egid
);
931 static void lo_mknod_symlink(fuse_req_t req
, fuse_ino_t parent
,
932 const char *name
, mode_t mode
, dev_t rdev
,
937 struct lo_data
*lo
= lo_data(req
);
938 struct lo_inode
*dir
;
939 struct fuse_entry_param e
;
940 struct lo_cred old
= {};
942 if (!is_safe_path_component(name
)) {
943 fuse_reply_err(req
, EINVAL
);
947 dir
= lo_inode(req
, parent
);
949 fuse_reply_err(req
, EBADF
);
953 saverr
= lo_change_cred(req
, &old
);
958 res
= mknod_wrapper(dir
->fd
, name
, link
, mode
, rdev
);
962 lo_restore_cred(&old
);
968 saverr
= lo_do_lookup(req
, parent
, name
, &e
);
973 fuse_log(FUSE_LOG_DEBUG
, " %lli/%s -> %lli\n", (unsigned long long)parent
,
974 name
, (unsigned long long)e
.ino
);
976 fuse_reply_entry(req
, &e
);
977 lo_inode_put(lo
, &dir
);
981 lo_inode_put(lo
, &dir
);
982 fuse_reply_err(req
, saverr
);
985 static void lo_mknod(fuse_req_t req
, fuse_ino_t parent
, const char *name
,
986 mode_t mode
, dev_t rdev
)
988 lo_mknod_symlink(req
, parent
, name
, mode
, rdev
, NULL
);
991 static void lo_mkdir(fuse_req_t req
, fuse_ino_t parent
, const char *name
,
994 lo_mknod_symlink(req
, parent
, name
, S_IFDIR
| mode
, 0, NULL
);
997 static void lo_symlink(fuse_req_t req
, const char *link
, fuse_ino_t parent
,
1000 lo_mknod_symlink(req
, parent
, name
, S_IFLNK
, 0, link
);
1003 static void lo_link(fuse_req_t req
, fuse_ino_t ino
, fuse_ino_t parent
,
1007 struct lo_data
*lo
= lo_data(req
);
1008 struct lo_inode
*parent_inode
;
1009 struct lo_inode
*inode
;
1010 struct fuse_entry_param e
;
1014 if (!is_safe_path_component(name
)) {
1015 fuse_reply_err(req
, EINVAL
);
1019 parent_inode
= lo_inode(req
, parent
);
1020 inode
= lo_inode(req
, ino
);
1021 if (!parent_inode
|| !inode
) {
1026 memset(&e
, 0, sizeof(struct fuse_entry_param
));
1027 e
.attr_timeout
= lo
->timeout
;
1028 e
.entry_timeout
= lo
->timeout
;
1030 sprintf(procname
, "%i", inode
->fd
);
1031 res
= linkat(lo
->proc_self_fd
, procname
, parent_inode
->fd
, name
,
1037 res
= fstatat(inode
->fd
, "", &e
.attr
, AT_EMPTY_PATH
| AT_SYMLINK_NOFOLLOW
);
1042 pthread_mutex_lock(&lo
->mutex
);
1044 pthread_mutex_unlock(&lo
->mutex
);
1045 e
.ino
= inode
->fuse_ino
;
1047 fuse_log(FUSE_LOG_DEBUG
, " %lli/%s -> %lli\n", (unsigned long long)parent
,
1048 name
, (unsigned long long)e
.ino
);
1050 fuse_reply_entry(req
, &e
);
1051 lo_inode_put(lo
, &parent_inode
);
1052 lo_inode_put(lo
, &inode
);
1057 lo_inode_put(lo
, &parent_inode
);
1058 lo_inode_put(lo
, &inode
);
1059 fuse_reply_err(req
, saverr
);
1062 /* Increments nlookup and caller must release refcount using lo_inode_put() */
1063 static struct lo_inode
*lookup_name(fuse_req_t req
, fuse_ino_t parent
,
1069 res
= fstatat(lo_fd(req
, parent
), name
, &attr
,
1070 AT_EMPTY_PATH
| AT_SYMLINK_NOFOLLOW
);
1075 return lo_find(lo_data(req
), &attr
);
1078 static void lo_rmdir(fuse_req_t req
, fuse_ino_t parent
, const char *name
)
1081 struct lo_inode
*inode
;
1082 struct lo_data
*lo
= lo_data(req
);
1084 if (!is_safe_path_component(name
)) {
1085 fuse_reply_err(req
, EINVAL
);
1089 inode
= lookup_name(req
, parent
, name
);
1091 fuse_reply_err(req
, EIO
);
1095 res
= unlinkat(lo_fd(req
, parent
), name
, AT_REMOVEDIR
);
1097 fuse_reply_err(req
, res
== -1 ? errno
: 0);
1098 unref_inode_lolocked(lo
, inode
, 1);
1099 lo_inode_put(lo
, &inode
);
1102 static void lo_rename(fuse_req_t req
, fuse_ino_t parent
, const char *name
,
1103 fuse_ino_t newparent
, const char *newname
,
1107 struct lo_inode
*parent_inode
;
1108 struct lo_inode
*newparent_inode
;
1109 struct lo_inode
*oldinode
= NULL
;
1110 struct lo_inode
*newinode
= NULL
;
1111 struct lo_data
*lo
= lo_data(req
);
1113 if (!is_safe_path_component(name
) || !is_safe_path_component(newname
)) {
1114 fuse_reply_err(req
, EINVAL
);
1118 parent_inode
= lo_inode(req
, parent
);
1119 newparent_inode
= lo_inode(req
, newparent
);
1120 if (!parent_inode
|| !newparent_inode
) {
1121 fuse_reply_err(req
, EBADF
);
1125 oldinode
= lookup_name(req
, parent
, name
);
1126 newinode
= lookup_name(req
, newparent
, newname
);
1129 fuse_reply_err(req
, EIO
);
1134 #ifndef SYS_renameat2
1135 fuse_reply_err(req
, EINVAL
);
1137 res
= syscall(SYS_renameat2
, parent_inode
->fd
, name
,
1138 newparent_inode
->fd
, newname
, flags
);
1139 if (res
== -1 && errno
== ENOSYS
) {
1140 fuse_reply_err(req
, EINVAL
);
1142 fuse_reply_err(req
, res
== -1 ? errno
: 0);
1148 res
= renameat(parent_inode
->fd
, name
, newparent_inode
->fd
, newname
);
1150 fuse_reply_err(req
, res
== -1 ? errno
: 0);
1152 unref_inode_lolocked(lo
, oldinode
, 1);
1153 unref_inode_lolocked(lo
, newinode
, 1);
1154 lo_inode_put(lo
, &oldinode
);
1155 lo_inode_put(lo
, &newinode
);
1156 lo_inode_put(lo
, &parent_inode
);
1157 lo_inode_put(lo
, &newparent_inode
);
1160 static void lo_unlink(fuse_req_t req
, fuse_ino_t parent
, const char *name
)
1163 struct lo_inode
*inode
;
1164 struct lo_data
*lo
= lo_data(req
);
1166 if (!is_safe_path_component(name
)) {
1167 fuse_reply_err(req
, EINVAL
);
1171 inode
= lookup_name(req
, parent
, name
);
1173 fuse_reply_err(req
, EIO
);
1177 res
= unlinkat(lo_fd(req
, parent
), name
, 0);
1179 fuse_reply_err(req
, res
== -1 ? errno
: 0);
1180 unref_inode_lolocked(lo
, inode
, 1);
1181 lo_inode_put(lo
, &inode
);
1184 /* To be called with lo->mutex held */
1185 static void unref_inode(struct lo_data
*lo
, struct lo_inode
*inode
, uint64_t n
)
1191 assert(inode
->nlookup
>= n
);
1192 inode
->nlookup
-= n
;
1193 if (!inode
->nlookup
) {
1194 lo_map_remove(&lo
->ino_map
, inode
->fuse_ino
);
1195 g_hash_table_remove(lo
->inodes
, &inode
->key
);
1196 if (g_hash_table_size(inode
->posix_locks
)) {
1197 fuse_log(FUSE_LOG_WARNING
, "Hash table is not empty\n");
1199 g_hash_table_destroy(inode
->posix_locks
);
1200 pthread_mutex_destroy(&inode
->plock_mutex
);
1202 /* Drop our refcount from lo_do_lookup() */
1203 lo_inode_put(lo
, &inode
);
1207 static void unref_inode_lolocked(struct lo_data
*lo
, struct lo_inode
*inode
,
1214 pthread_mutex_lock(&lo
->mutex
);
1215 unref_inode(lo
, inode
, n
);
1216 pthread_mutex_unlock(&lo
->mutex
);
1219 static void lo_forget_one(fuse_req_t req
, fuse_ino_t ino
, uint64_t nlookup
)
1221 struct lo_data
*lo
= lo_data(req
);
1222 struct lo_inode
*inode
;
1224 inode
= lo_inode(req
, ino
);
1229 fuse_log(FUSE_LOG_DEBUG
, " forget %lli %lli -%lli\n",
1230 (unsigned long long)ino
, (unsigned long long)inode
->nlookup
,
1231 (unsigned long long)nlookup
);
1233 unref_inode_lolocked(lo
, inode
, nlookup
);
1234 lo_inode_put(lo
, &inode
);
1237 static void lo_forget(fuse_req_t req
, fuse_ino_t ino
, uint64_t nlookup
)
1239 lo_forget_one(req
, ino
, nlookup
);
1240 fuse_reply_none(req
);
1243 static void lo_forget_multi(fuse_req_t req
, size_t count
,
1244 struct fuse_forget_data
*forgets
)
1248 for (i
= 0; i
< count
; i
++) {
1249 lo_forget_one(req
, forgets
[i
].ino
, forgets
[i
].nlookup
);
1251 fuse_reply_none(req
);
1254 static void lo_readlink(fuse_req_t req
, fuse_ino_t ino
)
1256 char buf
[PATH_MAX
+ 1];
1259 res
= readlinkat(lo_fd(req
, ino
), "", buf
, sizeof(buf
));
1261 return (void)fuse_reply_err(req
, errno
);
1264 if (res
== sizeof(buf
)) {
1265 return (void)fuse_reply_err(req
, ENAMETOOLONG
);
1270 fuse_reply_readlink(req
, buf
);
1276 struct dirent
*entry
;
1280 static void lo_dirp_put(struct lo_dirp
**dp
)
1282 struct lo_dirp
*d
= *dp
;
1289 if (g_atomic_int_dec_and_test(&d
->refcount
)) {
1295 /* Call lo_dirp_put() on the return value when no longer needed */
1296 static struct lo_dirp
*lo_dirp(fuse_req_t req
, struct fuse_file_info
*fi
)
1298 struct lo_data
*lo
= lo_data(req
);
1299 struct lo_map_elem
*elem
;
1301 pthread_mutex_lock(&lo
->mutex
);
1302 elem
= lo_map_get(&lo
->dirp_map
, fi
->fh
);
1304 g_atomic_int_inc(&elem
->dirp
->refcount
);
1306 pthread_mutex_unlock(&lo
->mutex
);
1314 static void lo_opendir(fuse_req_t req
, fuse_ino_t ino
,
1315 struct fuse_file_info
*fi
)
1318 struct lo_data
*lo
= lo_data(req
);
1323 d
= calloc(1, sizeof(struct lo_dirp
));
1328 fd
= openat(lo_fd(req
, ino
), ".", O_RDONLY
);
1333 d
->dp
= fdopendir(fd
);
1334 if (d
->dp
== NULL
) {
1341 g_atomic_int_set(&d
->refcount
, 1); /* paired with lo_releasedir() */
1342 pthread_mutex_lock(&lo
->mutex
);
1343 fh
= lo_add_dirp_mapping(req
, d
);
1344 pthread_mutex_unlock(&lo
->mutex
);
1350 if (lo
->cache
== CACHE_ALWAYS
) {
1351 fi
->cache_readdir
= 1;
1353 fuse_reply_open(req
, fi
);
1362 } else if (fd
!= -1) {
1367 fuse_reply_err(req
, error
);
1370 static void lo_do_readdir(fuse_req_t req
, fuse_ino_t ino
, size_t size
,
1371 off_t offset
, struct fuse_file_info
*fi
, int plus
)
1373 struct lo_data
*lo
= lo_data(req
);
1374 struct lo_dirp
*d
= NULL
;
1375 struct lo_inode
*dinode
;
1381 dinode
= lo_inode(req
, ino
);
1386 d
= lo_dirp(req
, fi
);
1392 buf
= calloc(1, size
);
1398 if (offset
!= d
->offset
) {
1399 seekdir(d
->dp
, offset
);
1410 d
->entry
= readdir(d
->dp
);
1412 if (errno
) { /* Error */
1415 } else { /* End of stream */
1420 nextoff
= d
->entry
->d_off
;
1421 name
= d
->entry
->d_name
;
1423 fuse_ino_t entry_ino
= 0;
1424 struct fuse_entry_param e
= (struct fuse_entry_param
){
1425 .attr
.st_ino
= d
->entry
->d_ino
,
1426 .attr
.st_mode
= d
->entry
->d_type
<< 12,
1429 /* Hide root's parent directory */
1430 if (dinode
== &lo
->root
&& strcmp(name
, "..") == 0) {
1431 e
.attr
.st_ino
= lo
->root
.key
.ino
;
1432 e
.attr
.st_mode
= DT_DIR
<< 12;
1436 if (!is_dot_or_dotdot(name
)) {
1437 err
= lo_do_lookup(req
, ino
, name
, &e
);
1444 entsize
= fuse_add_direntry_plus(req
, p
, rem
, name
, &e
, nextoff
);
1446 entsize
= fuse_add_direntry(req
, p
, rem
, name
, &e
.attr
, nextoff
);
1448 if (entsize
> rem
) {
1449 if (entry_ino
!= 0) {
1450 lo_forget_one(req
, entry_ino
, 1);
1459 d
->offset
= nextoff
;
1465 lo_inode_put(lo
, &dinode
);
1468 * If there's an error, we can only signal it if we haven't stored
1469 * any entries yet - otherwise we'd end up with wrong lookup
1470 * counts for the entries that are already in the buffer. So we
1471 * return what we've collected until that point.
1473 if (err
&& rem
== size
) {
1474 fuse_reply_err(req
, err
);
1476 fuse_reply_buf(req
, buf
, size
- rem
);
1481 static void lo_readdir(fuse_req_t req
, fuse_ino_t ino
, size_t size
,
1482 off_t offset
, struct fuse_file_info
*fi
)
1484 lo_do_readdir(req
, ino
, size
, offset
, fi
, 0);
1487 static void lo_readdirplus(fuse_req_t req
, fuse_ino_t ino
, size_t size
,
1488 off_t offset
, struct fuse_file_info
*fi
)
1490 lo_do_readdir(req
, ino
, size
, offset
, fi
, 1);
1493 static void lo_releasedir(fuse_req_t req
, fuse_ino_t ino
,
1494 struct fuse_file_info
*fi
)
1496 struct lo_data
*lo
= lo_data(req
);
1497 struct lo_map_elem
*elem
;
1502 pthread_mutex_lock(&lo
->mutex
);
1503 elem
= lo_map_get(&lo
->dirp_map
, fi
->fh
);
1505 pthread_mutex_unlock(&lo
->mutex
);
1506 fuse_reply_err(req
, EBADF
);
1511 lo_map_remove(&lo
->dirp_map
, fi
->fh
);
1512 pthread_mutex_unlock(&lo
->mutex
);
1514 lo_dirp_put(&d
); /* paired with lo_opendir() */
1516 fuse_reply_err(req
, 0);
1519 static void update_open_flags(int writeback
, struct fuse_file_info
*fi
)
1522 * With writeback cache, kernel may send read requests even
1523 * when userspace opened write-only
1525 if (writeback
&& (fi
->flags
& O_ACCMODE
) == O_WRONLY
) {
1526 fi
->flags
&= ~O_ACCMODE
;
1527 fi
->flags
|= O_RDWR
;
1531 * With writeback cache, O_APPEND is handled by the kernel.
1532 * This breaks atomicity (since the file may change in the
1533 * underlying filesystem, so that the kernel's idea of the
1534 * end of the file isn't accurate anymore). In this example,
1535 * we just accept that. A more rigorous filesystem may want
1536 * to return an error here
1538 if (writeback
&& (fi
->flags
& O_APPEND
)) {
1539 fi
->flags
&= ~O_APPEND
;
1543 * O_DIRECT in guest should not necessarily mean bypassing page
1544 * cache on host as well. If somebody needs that behavior, it
1545 * probably should be a configuration knob in daemon.
1547 fi
->flags
&= ~O_DIRECT
;
1550 static void lo_create(fuse_req_t req
, fuse_ino_t parent
, const char *name
,
1551 mode_t mode
, struct fuse_file_info
*fi
)
1554 struct lo_data
*lo
= lo_data(req
);
1555 struct lo_inode
*parent_inode
;
1556 struct fuse_entry_param e
;
1558 struct lo_cred old
= {};
1560 fuse_log(FUSE_LOG_DEBUG
, "lo_create(parent=%" PRIu64
", name=%s)\n", parent
,
1563 if (!is_safe_path_component(name
)) {
1564 fuse_reply_err(req
, EINVAL
);
1568 parent_inode
= lo_inode(req
, parent
);
1569 if (!parent_inode
) {
1570 fuse_reply_err(req
, EBADF
);
1574 err
= lo_change_cred(req
, &old
);
1579 update_open_flags(lo
->writeback
, fi
);
1581 fd
= openat(parent_inode
->fd
, name
, (fi
->flags
| O_CREAT
) & ~O_NOFOLLOW
,
1583 err
= fd
== -1 ? errno
: 0;
1584 lo_restore_cred(&old
);
1589 pthread_mutex_lock(&lo
->mutex
);
1590 fh
= lo_add_fd_mapping(req
, fd
);
1591 pthread_mutex_unlock(&lo
->mutex
);
1599 err
= lo_do_lookup(req
, parent
, name
, &e
);
1601 if (lo
->cache
== CACHE_NONE
) {
1603 } else if (lo
->cache
== CACHE_ALWAYS
) {
1608 lo_inode_put(lo
, &parent_inode
);
1611 fuse_reply_err(req
, err
);
1613 fuse_reply_create(req
, &e
, fi
);
1617 /* Should be called with inode->plock_mutex held */
1618 static struct lo_inode_plock
*lookup_create_plock_ctx(struct lo_data
*lo
,
1619 struct lo_inode
*inode
,
1620 uint64_t lock_owner
,
1621 pid_t pid
, int *err
)
1623 struct lo_inode_plock
*plock
;
1628 g_hash_table_lookup(inode
->posix_locks
, GUINT_TO_POINTER(lock_owner
));
1634 plock
= malloc(sizeof(struct lo_inode_plock
));
1640 /* Open another instance of file which can be used for ofd locks. */
1641 sprintf(procname
, "%i", inode
->fd
);
1643 /* TODO: What if file is not writable? */
1644 fd
= openat(lo
->proc_self_fd
, procname
, O_RDWR
);
1651 plock
->lock_owner
= lock_owner
;
1653 g_hash_table_insert(inode
->posix_locks
, GUINT_TO_POINTER(plock
->lock_owner
),
1658 static void lo_getlk(fuse_req_t req
, fuse_ino_t ino
, struct fuse_file_info
*fi
,
1661 struct lo_data
*lo
= lo_data(req
);
1662 struct lo_inode
*inode
;
1663 struct lo_inode_plock
*plock
;
1664 int ret
, saverr
= 0;
1666 fuse_log(FUSE_LOG_DEBUG
,
1667 "lo_getlk(ino=%" PRIu64
", flags=%d)"
1668 " owner=0x%lx, l_type=%d l_start=0x%lx"
1670 ino
, fi
->flags
, fi
->lock_owner
, lock
->l_type
, lock
->l_start
,
1673 inode
= lo_inode(req
, ino
);
1675 fuse_reply_err(req
, EBADF
);
1679 pthread_mutex_lock(&inode
->plock_mutex
);
1681 lookup_create_plock_ctx(lo
, inode
, fi
->lock_owner
, lock
->l_pid
, &ret
);
1687 ret
= fcntl(plock
->fd
, F_OFD_GETLK
, lock
);
1693 pthread_mutex_unlock(&inode
->plock_mutex
);
1694 lo_inode_put(lo
, &inode
);
1697 fuse_reply_err(req
, saverr
);
1699 fuse_reply_lock(req
, lock
);
1703 static void lo_setlk(fuse_req_t req
, fuse_ino_t ino
, struct fuse_file_info
*fi
,
1704 struct flock
*lock
, int sleep
)
1706 struct lo_data
*lo
= lo_data(req
);
1707 struct lo_inode
*inode
;
1708 struct lo_inode_plock
*plock
;
1709 int ret
, saverr
= 0;
1711 fuse_log(FUSE_LOG_DEBUG
,
1712 "lo_setlk(ino=%" PRIu64
", flags=%d)"
1713 " cmd=%d pid=%d owner=0x%lx sleep=%d l_whence=%d"
1714 " l_start=0x%lx l_len=0x%lx\n",
1715 ino
, fi
->flags
, lock
->l_type
, lock
->l_pid
, fi
->lock_owner
, sleep
,
1716 lock
->l_whence
, lock
->l_start
, lock
->l_len
);
1719 fuse_reply_err(req
, EOPNOTSUPP
);
1723 inode
= lo_inode(req
, ino
);
1725 fuse_reply_err(req
, EBADF
);
1729 pthread_mutex_lock(&inode
->plock_mutex
);
1731 lookup_create_plock_ctx(lo
, inode
, fi
->lock_owner
, lock
->l_pid
, &ret
);
1738 /* TODO: Is it alright to modify flock? */
1740 ret
= fcntl(plock
->fd
, F_OFD_SETLK
, lock
);
1746 pthread_mutex_unlock(&inode
->plock_mutex
);
1747 lo_inode_put(lo
, &inode
);
1749 fuse_reply_err(req
, saverr
);
1752 static void lo_fsyncdir(fuse_req_t req
, fuse_ino_t ino
, int datasync
,
1753 struct fuse_file_info
*fi
)
1761 d
= lo_dirp(req
, fi
);
1763 fuse_reply_err(req
, EBADF
);
1769 res
= fdatasync(fd
);
1776 fuse_reply_err(req
, res
== -1 ? errno
: 0);
1779 static void lo_open(fuse_req_t req
, fuse_ino_t ino
, struct fuse_file_info
*fi
)
1784 struct lo_data
*lo
= lo_data(req
);
1786 fuse_log(FUSE_LOG_DEBUG
, "lo_open(ino=%" PRIu64
", flags=%d)\n", ino
,
1789 update_open_flags(lo
->writeback
, fi
);
1791 sprintf(buf
, "%i", lo_fd(req
, ino
));
1792 fd
= openat(lo
->proc_self_fd
, buf
, fi
->flags
& ~O_NOFOLLOW
);
1794 return (void)fuse_reply_err(req
, errno
);
1797 pthread_mutex_lock(&lo
->mutex
);
1798 fh
= lo_add_fd_mapping(req
, fd
);
1799 pthread_mutex_unlock(&lo
->mutex
);
1802 fuse_reply_err(req
, ENOMEM
);
1807 if (lo
->cache
== CACHE_NONE
) {
1809 } else if (lo
->cache
== CACHE_ALWAYS
) {
1812 fuse_reply_open(req
, fi
);
1815 static void lo_release(fuse_req_t req
, fuse_ino_t ino
,
1816 struct fuse_file_info
*fi
)
1818 struct lo_data
*lo
= lo_data(req
);
1819 struct lo_map_elem
*elem
;
1824 pthread_mutex_lock(&lo
->mutex
);
1825 elem
= lo_map_get(&lo
->fd_map
, fi
->fh
);
1829 lo_map_remove(&lo
->fd_map
, fi
->fh
);
1831 pthread_mutex_unlock(&lo
->mutex
);
1834 fuse_reply_err(req
, 0);
1837 static void lo_flush(fuse_req_t req
, fuse_ino_t ino
, struct fuse_file_info
*fi
)
1841 struct lo_inode
*inode
;
1843 inode
= lo_inode(req
, ino
);
1845 fuse_reply_err(req
, EBADF
);
1849 /* An fd is going away. Cleanup associated posix locks */
1850 pthread_mutex_lock(&inode
->plock_mutex
);
1851 g_hash_table_remove(inode
->posix_locks
, GUINT_TO_POINTER(fi
->lock_owner
));
1852 pthread_mutex_unlock(&inode
->plock_mutex
);
1854 res
= close(dup(lo_fi_fd(req
, fi
)));
1855 lo_inode_put(lo_data(req
), &inode
);
1856 fuse_reply_err(req
, res
== -1 ? errno
: 0);
1859 static void lo_fsync(fuse_req_t req
, fuse_ino_t ino
, int datasync
,
1860 struct fuse_file_info
*fi
)
1866 fuse_log(FUSE_LOG_DEBUG
, "lo_fsync(ino=%" PRIu64
", fi=0x%p)\n", ino
,
1870 struct lo_data
*lo
= lo_data(req
);
1872 res
= asprintf(&buf
, "%i", lo_fd(req
, ino
));
1874 return (void)fuse_reply_err(req
, errno
);
1877 fd
= openat(lo
->proc_self_fd
, buf
, O_RDWR
);
1880 return (void)fuse_reply_err(req
, errno
);
1883 fd
= lo_fi_fd(req
, fi
);
1887 res
= fdatasync(fd
);
1894 fuse_reply_err(req
, res
== -1 ? errno
: 0);
1897 static void lo_read(fuse_req_t req
, fuse_ino_t ino
, size_t size
, off_t offset
,
1898 struct fuse_file_info
*fi
)
1900 struct fuse_bufvec buf
= FUSE_BUFVEC_INIT(size
);
1902 fuse_log(FUSE_LOG_DEBUG
,
1903 "lo_read(ino=%" PRIu64
", size=%zd, "
1905 ino
, size
, (unsigned long)offset
);
1907 buf
.buf
[0].flags
= FUSE_BUF_IS_FD
| FUSE_BUF_FD_SEEK
;
1908 buf
.buf
[0].fd
= lo_fi_fd(req
, fi
);
1909 buf
.buf
[0].pos
= offset
;
1911 fuse_reply_data(req
, &buf
);
1914 static void lo_write_buf(fuse_req_t req
, fuse_ino_t ino
,
1915 struct fuse_bufvec
*in_buf
, off_t off
,
1916 struct fuse_file_info
*fi
)
1920 struct fuse_bufvec out_buf
= FUSE_BUFVEC_INIT(fuse_buf_size(in_buf
));
1921 bool cap_fsetid_dropped
= false;
1923 out_buf
.buf
[0].flags
= FUSE_BUF_IS_FD
| FUSE_BUF_FD_SEEK
;
1924 out_buf
.buf
[0].fd
= lo_fi_fd(req
, fi
);
1925 out_buf
.buf
[0].pos
= off
;
1927 fuse_log(FUSE_LOG_DEBUG
,
1928 "lo_write_buf(ino=%" PRIu64
", size=%zd, off=%lu)\n", ino
,
1929 out_buf
.buf
[0].size
, (unsigned long)off
);
1932 * If kill_priv is set, drop CAP_FSETID which should lead to kernel
1933 * clearing setuid/setgid on file.
1935 if (fi
->kill_priv
) {
1936 res
= drop_effective_cap("FSETID", &cap_fsetid_dropped
);
1938 fuse_reply_err(req
, res
);
1943 res
= fuse_buf_copy(&out_buf
, in_buf
);
1945 fuse_reply_err(req
, -res
);
1947 fuse_reply_write(req
, (size_t)res
);
1950 if (cap_fsetid_dropped
) {
1951 res
= gain_effective_cap("FSETID");
1953 fuse_log(FUSE_LOG_ERR
, "Failed to gain CAP_FSETID\n");
1958 static void lo_statfs(fuse_req_t req
, fuse_ino_t ino
)
1961 struct statvfs stbuf
;
1963 res
= fstatvfs(lo_fd(req
, ino
), &stbuf
);
1965 fuse_reply_err(req
, errno
);
1967 fuse_reply_statfs(req
, &stbuf
);
1971 static void lo_fallocate(fuse_req_t req
, fuse_ino_t ino
, int mode
, off_t offset
,
1972 off_t length
, struct fuse_file_info
*fi
)
1974 int err
= EOPNOTSUPP
;
1977 #ifdef CONFIG_FALLOCATE
1978 err
= fallocate(lo_fi_fd(req
, fi
), mode
, offset
, length
);
1983 #elif defined(CONFIG_POSIX_FALLOCATE)
1985 fuse_reply_err(req
, EOPNOTSUPP
);
1989 err
= posix_fallocate(lo_fi_fd(req
, fi
), offset
, length
);
1992 fuse_reply_err(req
, err
);
1995 static void lo_flock(fuse_req_t req
, fuse_ino_t ino
, struct fuse_file_info
*fi
,
2001 res
= flock(lo_fi_fd(req
, fi
), op
);
2003 fuse_reply_err(req
, res
== -1 ? errno
: 0);
2006 static void lo_getxattr(fuse_req_t req
, fuse_ino_t ino
, const char *name
,
2009 struct lo_data
*lo
= lo_data(req
);
2012 struct lo_inode
*inode
;
2017 inode
= lo_inode(req
, ino
);
2019 fuse_reply_err(req
, EBADF
);
2024 if (!lo_data(req
)->xattr
) {
2028 fuse_log(FUSE_LOG_DEBUG
, "lo_getxattr(ino=%" PRIu64
", name=%s size=%zd)\n",
2032 value
= malloc(size
);
2038 sprintf(procname
, "%i", inode
->fd
);
2040 * It is not safe to open() non-regular/non-dir files in file server
2041 * unless O_PATH is used, so use that method for regular files/dir
2042 * only (as it seems giving less performance overhead).
2043 * Otherwise, call fchdir() to avoid open().
2045 if (S_ISREG(inode
->filetype
) || S_ISDIR(inode
->filetype
)) {
2046 fd
= openat(lo
->proc_self_fd
, procname
, O_RDONLY
);
2050 ret
= fgetxattr(fd
, name
, value
, size
);
2052 /* fchdir should not fail here */
2053 assert(fchdir(lo
->proc_self_fd
) == 0);
2054 ret
= getxattr(procname
, name
, value
, size
);
2055 assert(fchdir(lo
->root
.fd
) == 0);
2066 fuse_reply_buf(req
, value
, ret
);
2068 fuse_reply_xattr(req
, ret
);
2077 lo_inode_put(lo
, &inode
);
2083 fuse_reply_err(req
, saverr
);
2087 static void lo_listxattr(fuse_req_t req
, fuse_ino_t ino
, size_t size
)
2089 struct lo_data
*lo
= lo_data(req
);
2092 struct lo_inode
*inode
;
2097 inode
= lo_inode(req
, ino
);
2099 fuse_reply_err(req
, EBADF
);
2104 if (!lo_data(req
)->xattr
) {
2108 fuse_log(FUSE_LOG_DEBUG
, "lo_listxattr(ino=%" PRIu64
", size=%zd)\n", ino
,
2112 value
= malloc(size
);
2118 sprintf(procname
, "%i", inode
->fd
);
2119 if (S_ISREG(inode
->filetype
) || S_ISDIR(inode
->filetype
)) {
2120 fd
= openat(lo
->proc_self_fd
, procname
, O_RDONLY
);
2124 ret
= flistxattr(fd
, value
, size
);
2126 /* fchdir should not fail here */
2127 assert(fchdir(lo
->proc_self_fd
) == 0);
2128 ret
= listxattr(procname
, value
, size
);
2129 assert(fchdir(lo
->root
.fd
) == 0);
2140 fuse_reply_buf(req
, value
, ret
);
2142 fuse_reply_xattr(req
, ret
);
2151 lo_inode_put(lo
, &inode
);
2157 fuse_reply_err(req
, saverr
);
2161 static void lo_setxattr(fuse_req_t req
, fuse_ino_t ino
, const char *name
,
2162 const char *value
, size_t size
, int flags
)
2165 struct lo_data
*lo
= lo_data(req
);
2166 struct lo_inode
*inode
;
2171 inode
= lo_inode(req
, ino
);
2173 fuse_reply_err(req
, EBADF
);
2178 if (!lo_data(req
)->xattr
) {
2182 fuse_log(FUSE_LOG_DEBUG
, "lo_setxattr(ino=%" PRIu64
2183 ", name=%s value=%s size=%zd)\n", ino
, name
, value
, size
);
2185 sprintf(procname
, "%i", inode
->fd
);
2186 if (S_ISREG(inode
->filetype
) || S_ISDIR(inode
->filetype
)) {
2187 fd
= openat(lo
->proc_self_fd
, procname
, O_RDONLY
);
2192 ret
= fsetxattr(fd
, name
, value
, size
, flags
);
2194 /* fchdir should not fail here */
2195 assert(fchdir(lo
->proc_self_fd
) == 0);
2196 ret
= setxattr(procname
, name
, value
, size
, flags
);
2197 assert(fchdir(lo
->root
.fd
) == 0);
2200 saverr
= ret
== -1 ? errno
: 0;
2207 lo_inode_put(lo
, &inode
);
2208 fuse_reply_err(req
, saverr
);
2211 static void lo_removexattr(fuse_req_t req
, fuse_ino_t ino
, const char *name
)
2214 struct lo_data
*lo
= lo_data(req
);
2215 struct lo_inode
*inode
;
2220 inode
= lo_inode(req
, ino
);
2222 fuse_reply_err(req
, EBADF
);
2227 if (!lo_data(req
)->xattr
) {
2231 fuse_log(FUSE_LOG_DEBUG
, "lo_removexattr(ino=%" PRIu64
", name=%s)\n", ino
,
2234 sprintf(procname
, "%i", inode
->fd
);
2235 if (S_ISREG(inode
->filetype
) || S_ISDIR(inode
->filetype
)) {
2236 fd
= openat(lo
->proc_self_fd
, procname
, O_RDONLY
);
2241 ret
= fremovexattr(fd
, name
);
2243 /* fchdir should not fail here */
2244 assert(fchdir(lo
->proc_self_fd
) == 0);
2245 ret
= removexattr(procname
, name
);
2246 assert(fchdir(lo
->root
.fd
) == 0);
2249 saverr
= ret
== -1 ? errno
: 0;
2256 lo_inode_put(lo
, &inode
);
2257 fuse_reply_err(req
, saverr
);
2260 #ifdef HAVE_COPY_FILE_RANGE
2261 static void lo_copy_file_range(fuse_req_t req
, fuse_ino_t ino_in
, off_t off_in
,
2262 struct fuse_file_info
*fi_in
, fuse_ino_t ino_out
,
2263 off_t off_out
, struct fuse_file_info
*fi_out
,
2264 size_t len
, int flags
)
2269 in_fd
= lo_fi_fd(req
, fi_in
);
2270 out_fd
= lo_fi_fd(req
, fi_out
);
2272 fuse_log(FUSE_LOG_DEBUG
,
2273 "lo_copy_file_range(ino=%" PRIu64
"/fd=%d, "
2274 "off=%lu, ino=%" PRIu64
"/fd=%d, "
2275 "off=%lu, size=%zd, flags=0x%x)\n",
2276 ino_in
, in_fd
, off_in
, ino_out
, out_fd
, off_out
, len
, flags
);
2278 res
= copy_file_range(in_fd
, &off_in
, out_fd
, &off_out
, len
, flags
);
2280 fuse_reply_err(req
, errno
);
2282 fuse_reply_write(req
, res
);
2287 static void lo_lseek(fuse_req_t req
, fuse_ino_t ino
, off_t off
, int whence
,
2288 struct fuse_file_info
*fi
)
2293 res
= lseek(lo_fi_fd(req
, fi
), off
, whence
);
2295 fuse_reply_lseek(req
, res
);
2297 fuse_reply_err(req
, errno
);
2301 static void lo_destroy(void *userdata
)
2303 struct lo_data
*lo
= (struct lo_data
*)userdata
;
2305 pthread_mutex_lock(&lo
->mutex
);
2307 GHashTableIter iter
;
2308 gpointer key
, value
;
2310 g_hash_table_iter_init(&iter
, lo
->inodes
);
2311 if (!g_hash_table_iter_next(&iter
, &key
, &value
)) {
2315 struct lo_inode
*inode
= value
;
2316 unref_inode(lo
, inode
, inode
->nlookup
);
2318 pthread_mutex_unlock(&lo
->mutex
);
2321 static struct fuse_lowlevel_ops lo_oper
= {
2323 .lookup
= lo_lookup
,
2326 .symlink
= lo_symlink
,
2328 .unlink
= lo_unlink
,
2330 .rename
= lo_rename
,
2331 .forget
= lo_forget
,
2332 .forget_multi
= lo_forget_multi
,
2333 .getattr
= lo_getattr
,
2334 .setattr
= lo_setattr
,
2335 .readlink
= lo_readlink
,
2336 .opendir
= lo_opendir
,
2337 .readdir
= lo_readdir
,
2338 .readdirplus
= lo_readdirplus
,
2339 .releasedir
= lo_releasedir
,
2340 .fsyncdir
= lo_fsyncdir
,
2341 .create
= lo_create
,
2345 .release
= lo_release
,
2349 .write_buf
= lo_write_buf
,
2350 .statfs
= lo_statfs
,
2351 .fallocate
= lo_fallocate
,
2353 .getxattr
= lo_getxattr
,
2354 .listxattr
= lo_listxattr
,
2355 .setxattr
= lo_setxattr
,
2356 .removexattr
= lo_removexattr
,
2357 #ifdef HAVE_COPY_FILE_RANGE
2358 .copy_file_range
= lo_copy_file_range
,
2361 .destroy
= lo_destroy
,
2364 /* Print vhost-user.json backend program capabilities */
2365 static void print_capabilities(void)
2368 printf(" \"type\": \"fs\"\n");
2373 * Drop all Linux capabilities because the wait parent process only needs to
2374 * sit in waitpid(2) and terminate.
2376 static void setup_wait_parent_capabilities(void)
2378 capng_setpid(syscall(SYS_gettid
));
2379 capng_clear(CAPNG_SELECT_BOTH
);
2380 capng_apply(CAPNG_SELECT_BOTH
);
2384 * Move to a new mount, net, and pid namespaces to isolate this process.
2386 static void setup_namespaces(struct lo_data
*lo
, struct fuse_session
*se
)
2389 char template[] = "virtiofsd-XXXXXX";
2393 * Create a new pid namespace for *child* processes. We'll have to
2394 * fork in order to enter the new pid namespace. A new mount namespace
2395 * is also needed so that we can remount /proc for the new pid
2398 * Our UNIX domain sockets have been created. Now we can move to
2399 * an empty network namespace to prevent TCP/IP and other network
2400 * activity in case this process is compromised.
2402 if (unshare(CLONE_NEWPID
| CLONE_NEWNS
| CLONE_NEWNET
) != 0) {
2403 fuse_log(FUSE_LOG_ERR
, "unshare(CLONE_NEWPID | CLONE_NEWNS): %m\n");
2409 fuse_log(FUSE_LOG_ERR
, "fork() failed: %m\n");
2416 setup_wait_parent_capabilities();
2418 /* The parent waits for the child */
2420 waited
= waitpid(child
, &wstatus
, 0);
2421 } while (waited
< 0 && errno
== EINTR
&& !se
->exited
);
2423 /* We were terminated by a signal, see fuse_signals.c */
2428 if (WIFEXITED(wstatus
)) {
2429 exit(WEXITSTATUS(wstatus
));
2435 /* Send us SIGTERM when the parent thread terminates, see prctl(2) */
2436 prctl(PR_SET_PDEATHSIG
, SIGTERM
);
2439 * If the mounts have shared propagation then we want to opt out so our
2440 * mount changes don't affect the parent mount namespace.
2442 if (mount(NULL
, "/", NULL
, MS_REC
| MS_SLAVE
, NULL
) < 0) {
2443 fuse_log(FUSE_LOG_ERR
, "mount(/, MS_REC|MS_SLAVE): %m\n");
2447 /* The child must remount /proc to use the new pid namespace */
2448 if (mount("proc", "/proc", "proc",
2449 MS_NODEV
| MS_NOEXEC
| MS_NOSUID
| MS_RELATIME
, NULL
) < 0) {
2450 fuse_log(FUSE_LOG_ERR
, "mount(/proc): %m\n");
2454 tmpdir
= mkdtemp(template);
2456 fuse_log(FUSE_LOG_ERR
, "tmpdir(%s): %m\n", template);
2460 if (mount("/proc/self/fd", tmpdir
, NULL
, MS_BIND
, NULL
) < 0) {
2461 fuse_log(FUSE_LOG_ERR
, "mount(/proc/self/fd, %s, MS_BIND): %m\n",
2466 /* Now we can get our /proc/self/fd directory file descriptor */
2467 lo
->proc_self_fd
= open(tmpdir
, O_PATH
);
2468 if (lo
->proc_self_fd
== -1) {
2469 fuse_log(FUSE_LOG_ERR
, "open(%s, O_PATH): %m\n", tmpdir
);
2473 if (umount2(tmpdir
, MNT_DETACH
) < 0) {
2474 fuse_log(FUSE_LOG_ERR
, "umount2(%s, MNT_DETACH): %m\n", tmpdir
);
2478 if (rmdir(tmpdir
) < 0) {
2479 fuse_log(FUSE_LOG_ERR
, "rmdir(%s): %m\n", tmpdir
);
2484 * Capture the capability state, we'll need to restore this for individual
2485 * threads later; see load_capng.
2487 static void setup_capng(void)
2489 /* Note this accesses /proc so has to happen before the sandbox */
2490 if (capng_get_caps_process()) {
2491 fuse_log(FUSE_LOG_ERR
, "capng_get_caps_process\n");
2494 pthread_mutex_init(&cap
.mutex
, NULL
);
2495 pthread_mutex_lock(&cap
.mutex
);
2496 cap
.saved
= capng_save_state();
2498 fuse_log(FUSE_LOG_ERR
, "capng_save_state\n");
2501 pthread_mutex_unlock(&cap
.mutex
);
2504 static void cleanup_capng(void)
2508 pthread_mutex_destroy(&cap
.mutex
);
2513 * Make the source directory our root so symlinks cannot escape and no other
2514 * files are accessible. Assumes unshare(CLONE_NEWNS) was already called.
2516 static void setup_mounts(const char *source
)
2521 if (mount(source
, source
, NULL
, MS_BIND
| MS_REC
, NULL
) < 0) {
2522 fuse_log(FUSE_LOG_ERR
, "mount(%s, %s, MS_BIND): %m\n", source
, source
);
2526 /* This magic is based on lxc's lxc_pivot_root() */
2527 oldroot
= open("/", O_DIRECTORY
| O_RDONLY
| O_CLOEXEC
);
2529 fuse_log(FUSE_LOG_ERR
, "open(/): %m\n");
2533 newroot
= open(source
, O_DIRECTORY
| O_RDONLY
| O_CLOEXEC
);
2535 fuse_log(FUSE_LOG_ERR
, "open(%s): %m\n", source
);
2539 if (fchdir(newroot
) < 0) {
2540 fuse_log(FUSE_LOG_ERR
, "fchdir(newroot): %m\n");
2544 if (syscall(__NR_pivot_root
, ".", ".") < 0) {
2545 fuse_log(FUSE_LOG_ERR
, "pivot_root(., .): %m\n");
2549 if (fchdir(oldroot
) < 0) {
2550 fuse_log(FUSE_LOG_ERR
, "fchdir(oldroot): %m\n");
2554 if (mount("", ".", "", MS_SLAVE
| MS_REC
, NULL
) < 0) {
2555 fuse_log(FUSE_LOG_ERR
, "mount(., MS_SLAVE | MS_REC): %m\n");
2559 if (umount2(".", MNT_DETACH
) < 0) {
2560 fuse_log(FUSE_LOG_ERR
, "umount2(., MNT_DETACH): %m\n");
2564 if (fchdir(newroot
) < 0) {
2565 fuse_log(FUSE_LOG_ERR
, "fchdir(newroot): %m\n");
2574 * Only keep whitelisted capabilities that are needed for file system operation
2575 * The (possibly NULL) modcaps_in string passed in is free'd before exit.
2577 static void setup_capabilities(char *modcaps_in
)
2579 char *modcaps
= modcaps_in
;
2580 pthread_mutex_lock(&cap
.mutex
);
2581 capng_restore_state(&cap
.saved
);
2584 * Whitelist file system-related capabilities that are needed for a file
2585 * server to act like root. Drop everything else like networking and
2586 * sysadmin capabilities.
2589 * 1. CAP_LINUX_IMMUTABLE is not included because it's only used via ioctl
2590 * and we don't support that.
2591 * 2. CAP_MAC_OVERRIDE is not included because it only seems to be
2592 * used by the Smack LSM. Omit it until there is demand for it.
2594 capng_setpid(syscall(SYS_gettid
));
2595 capng_clear(CAPNG_SELECT_BOTH
);
2596 if (capng_updatev(CAPNG_ADD
, CAPNG_PERMITTED
| CAPNG_EFFECTIVE
,
2599 CAP_DAC_READ_SEARCH
,
2607 fuse_log(FUSE_LOG_ERR
, "%s: capng_updatev failed\n", __func__
);
2612 * The modcaps option is a colon separated list of caps,
2613 * each preceded by either + or -.
2619 char *next
= strchr(modcaps
, ':');
2625 switch (modcaps
[0]) {
2631 action
= CAPNG_DROP
;
2635 fuse_log(FUSE_LOG_ERR
,
2636 "%s: Expecting '+'/'-' in modcaps but found '%c'\n",
2637 __func__
, modcaps
[0]);
2640 cap
= capng_name_to_capability(modcaps
+ 1);
2642 fuse_log(FUSE_LOG_ERR
, "%s: Unknown capability '%s'\n", __func__
,
2646 if (capng_update(action
, CAPNG_PERMITTED
| CAPNG_EFFECTIVE
, cap
)) {
2647 fuse_log(FUSE_LOG_ERR
, "%s: capng_update failed for '%s'\n",
2656 if (capng_apply(CAPNG_SELECT_BOTH
)) {
2657 fuse_log(FUSE_LOG_ERR
, "%s: capng_apply failed\n", __func__
);
2661 cap
.saved
= capng_save_state();
2663 fuse_log(FUSE_LOG_ERR
, "%s: capng_save_state failed\n", __func__
);
2666 pthread_mutex_unlock(&cap
.mutex
);
2670 * Lock down this process to prevent access to other processes or files outside
2671 * source directory. This reduces the impact of arbitrary code execution bugs.
2673 static void setup_sandbox(struct lo_data
*lo
, struct fuse_session
*se
,
2676 setup_namespaces(lo
, se
);
2677 setup_mounts(lo
->source
);
2678 setup_seccomp(enable_syslog
);
2679 setup_capabilities(g_strdup(lo
->modcaps
));
2682 /* Set the maximum number of open file descriptors */
2683 static void setup_nofile_rlimit(unsigned long rlimit_nofile
)
2685 struct rlimit rlim
= {
2686 .rlim_cur
= rlimit_nofile
,
2687 .rlim_max
= rlimit_nofile
,
2690 if (rlimit_nofile
== 0) {
2691 return; /* nothing to do */
2694 if (setrlimit(RLIMIT_NOFILE
, &rlim
) < 0) {
2695 /* Ignore SELinux denials */
2696 if (errno
== EPERM
) {
2700 fuse_log(FUSE_LOG_ERR
, "setrlimit(RLIMIT_NOFILE): %m\n");
2705 static void GCC_FMT_ATTR(2, 0)
2706 log_func(enum fuse_log_level level
, const char *fmt
, va_list ap
)
2708 g_autofree
char *localfmt
= NULL
;
2710 if (current_log_level
< level
) {
2714 if (current_log_level
== FUSE_LOG_DEBUG
) {
2716 localfmt
= g_strdup_printf("[%" PRId64
"] [ID: %08ld] %s",
2717 get_clock(), syscall(__NR_gettid
), fmt
);
2719 localfmt
= g_strdup_printf("[ID: %08ld] %s", syscall(__NR_gettid
),
2726 int priority
= LOG_ERR
;
2728 case FUSE_LOG_EMERG
:
2729 priority
= LOG_EMERG
;
2731 case FUSE_LOG_ALERT
:
2732 priority
= LOG_ALERT
;
2735 priority
= LOG_CRIT
;
2740 case FUSE_LOG_WARNING
:
2741 priority
= LOG_WARNING
;
2743 case FUSE_LOG_NOTICE
:
2744 priority
= LOG_NOTICE
;
2747 priority
= LOG_INFO
;
2749 case FUSE_LOG_DEBUG
:
2750 priority
= LOG_DEBUG
;
2753 vsyslog(priority
, fmt
, ap
);
2755 vfprintf(stderr
, fmt
, ap
);
2759 static void setup_root(struct lo_data
*lo
, struct lo_inode
*root
)
2764 fd
= open("/", O_PATH
);
2766 fuse_log(FUSE_LOG_ERR
, "open(%s, O_PATH): %m\n", lo
->source
);
2770 res
= fstatat(fd
, "", &stat
, AT_EMPTY_PATH
| AT_SYMLINK_NOFOLLOW
);
2772 fuse_log(FUSE_LOG_ERR
, "fstatat(%s): %m\n", lo
->source
);
2776 root
->filetype
= S_IFDIR
;
2778 root
->key
.ino
= stat
.st_ino
;
2779 root
->key
.dev
= stat
.st_dev
;
2781 g_atomic_int_set(&root
->refcount
, 2);
2784 static guint
lo_key_hash(gconstpointer key
)
2786 const struct lo_key
*lkey
= key
;
2788 return (guint
)lkey
->ino
+ (guint
)lkey
->dev
;
2791 static gboolean
lo_key_equal(gconstpointer a
, gconstpointer b
)
2793 const struct lo_key
*la
= a
;
2794 const struct lo_key
*lb
= b
;
2796 return la
->ino
== lb
->ino
&& la
->dev
== lb
->dev
;
2799 static void fuse_lo_data_cleanup(struct lo_data
*lo
)
2802 g_hash_table_destroy(lo
->inodes
);
2804 lo_map_destroy(&lo
->fd_map
);
2805 lo_map_destroy(&lo
->dirp_map
);
2806 lo_map_destroy(&lo
->ino_map
);
2808 if (lo
->proc_self_fd
>= 0) {
2809 close(lo
->proc_self_fd
);
2812 if (lo
->root
.fd
>= 0) {
2819 int main(int argc
, char *argv
[])
2821 struct fuse_args args
= FUSE_ARGS_INIT(argc
, argv
);
2822 struct fuse_session
*se
;
2823 struct fuse_cmdline_opts opts
;
2824 struct lo_data lo
= {
2830 struct lo_map_elem
*root_elem
;
2833 /* Don't mask creation mode, kernel already did that */
2836 pthread_mutex_init(&lo
.mutex
, NULL
);
2837 lo
.inodes
= g_hash_table_new(lo_key_hash
, lo_key_equal
);
2839 lo
.root
.fuse_ino
= FUSE_ROOT_ID
;
2840 lo
.cache
= CACHE_AUTO
;
2843 * Set up the ino map like this:
2844 * [0] Reserved (will not be used)
2847 lo_map_init(&lo
.ino_map
);
2848 lo_map_reserve(&lo
.ino_map
, 0)->in_use
= false;
2849 root_elem
= lo_map_reserve(&lo
.ino_map
, lo
.root
.fuse_ino
);
2850 root_elem
->inode
= &lo
.root
;
2852 lo_map_init(&lo
.dirp_map
);
2853 lo_map_init(&lo
.fd_map
);
2855 if (fuse_parse_cmdline(&args
, &opts
) != 0) {
2858 fuse_set_log_func(log_func
);
2859 use_syslog
= opts
.syslog
;
2861 openlog("virtiofsd", LOG_PID
, LOG_DAEMON
);
2864 if (opts
.show_help
) {
2865 printf("usage: %s [options]\n\n", argv
[0]);
2866 fuse_cmdline_help();
2867 printf(" -o source=PATH shared directory tree\n");
2868 fuse_lowlevel_help();
2871 } else if (opts
.show_version
) {
2872 fuse_lowlevel_version();
2875 } else if (opts
.print_capabilities
) {
2876 print_capabilities();
2881 if (fuse_opt_parse(&args
, &lo
, lo_opts
, NULL
) == -1) {
2886 * log_level is 0 if not configured via cmd options (0 is LOG_EMERG,
2887 * and we don't use this log level).
2889 if (opts
.log_level
!= 0) {
2890 current_log_level
= opts
.log_level
;
2892 lo
.debug
= opts
.debug
;
2894 current_log_level
= FUSE_LOG_DEBUG
;
2900 res
= lstat(lo
.source
, &stat
);
2902 fuse_log(FUSE_LOG_ERR
, "failed to stat source (\"%s\"): %m\n",
2906 if (!S_ISDIR(stat
.st_mode
)) {
2907 fuse_log(FUSE_LOG_ERR
, "source is not a directory\n");
2911 lo
.source
= strdup("/");
2913 if (!lo
.timeout_set
) {
2924 lo
.timeout
= 86400.0;
2927 } else if (lo
.timeout
< 0) {
2928 fuse_log(FUSE_LOG_ERR
, "timeout is negative (%lf)\n", lo
.timeout
);
2932 se
= fuse_session_new(&args
, &lo_oper
, sizeof(lo_oper
), &lo
);
2937 if (fuse_set_signal_handlers(se
) != 0) {
2941 if (fuse_session_mount(se
) != 0) {
2945 fuse_daemonize(opts
.foreground
);
2947 setup_nofile_rlimit(opts
.rlimit_nofile
);
2949 /* Must be before sandbox since it wants /proc */
2952 setup_sandbox(&lo
, se
, opts
.syslog
);
2954 setup_root(&lo
, &lo
.root
);
2955 /* Block until ctrl+c or fusermount -u */
2956 ret
= virtio_loop(se
);
2958 fuse_session_unmount(se
);
2961 fuse_remove_signal_handlers(se
);
2963 fuse_session_destroy(se
);
2965 fuse_opt_free_args(&args
);
2967 fuse_lo_data_cleanup(&lo
);