2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
11 #include <linux/pagemap.h>
12 #include <linux/file.h>
13 #include <linux/fs_context.h>
14 #include <linux/moduleparam.h>
15 #include <linux/sched.h>
16 #include <linux/namei.h>
17 #include <linux/slab.h>
18 #include <linux/xattr.h>
19 #include <linux/iversion.h>
20 #include <linux/posix_acl.h>
21 #include <linux/security.h>
22 #include <linux/types.h>
23 #include <linux/kernel.h>
25 static bool __read_mostly allow_sys_admin_access
;
26 module_param(allow_sys_admin_access
, bool, 0644);
27 MODULE_PARM_DESC(allow_sys_admin_access
,
28 "Allow users with CAP_SYS_ADMIN in initial userns to bypass allow_other access check");
30 static void fuse_advise_use_readdirplus(struct inode
*dir
)
32 struct fuse_inode
*fi
= get_fuse_inode(dir
);
34 set_bit(FUSE_I_ADVISE_RDPLUS
, &fi
->state
);
37 #if BITS_PER_LONG >= 64
38 static inline void __fuse_dentry_settime(struct dentry
*entry
, u64 time
)
40 entry
->d_fsdata
= (void *) time
;
43 static inline u64
fuse_dentry_time(const struct dentry
*entry
)
45 return (u64
)entry
->d_fsdata
;
54 static inline void __fuse_dentry_settime(struct dentry
*dentry
, u64 time
)
56 ((union fuse_dentry
*) dentry
->d_fsdata
)->time
= time
;
59 static inline u64
fuse_dentry_time(const struct dentry
*entry
)
61 return ((union fuse_dentry
*) entry
->d_fsdata
)->time
;
65 static void fuse_dentry_settime(struct dentry
*dentry
, u64 time
)
67 struct fuse_conn
*fc
= get_fuse_conn_super(dentry
->d_sb
);
68 bool delete = !time
&& fc
->delete_stale
;
70 * Mess with DCACHE_OP_DELETE because dput() will be faster without it.
71 * Don't care about races, either way it's just an optimization
73 if ((!delete && (dentry
->d_flags
& DCACHE_OP_DELETE
)) ||
74 (delete && !(dentry
->d_flags
& DCACHE_OP_DELETE
))) {
75 spin_lock(&dentry
->d_lock
);
77 dentry
->d_flags
&= ~DCACHE_OP_DELETE
;
79 dentry
->d_flags
|= DCACHE_OP_DELETE
;
80 spin_unlock(&dentry
->d_lock
);
83 __fuse_dentry_settime(dentry
, time
);
87 * FUSE caches dentries and attributes with separate timeout. The
88 * time in jiffies until the dentry/attributes are valid is stored in
89 * dentry->d_fsdata and fuse_inode->i_time respectively.
93 * Calculate the time in jiffies until a dentry/attributes are valid
95 u64
fuse_time_to_jiffies(u64 sec
, u32 nsec
)
98 struct timespec64 ts
= {
100 min_t(u32
, nsec
, NSEC_PER_SEC
- 1)
103 return get_jiffies_64() + timespec64_to_jiffies(&ts
);
109 * Set dentry and possibly attribute timeouts from the lookup/mk*
112 void fuse_change_entry_timeout(struct dentry
*entry
, struct fuse_entry_out
*o
)
114 fuse_dentry_settime(entry
,
115 fuse_time_to_jiffies(o
->entry_valid
, o
->entry_valid_nsec
));
118 void fuse_invalidate_attr_mask(struct inode
*inode
, u32 mask
)
120 set_mask_bits(&get_fuse_inode(inode
)->inval_mask
, 0, mask
);
124 * Mark the attributes as stale, so that at the next call to
125 * ->getattr() they will be fetched from userspace
127 void fuse_invalidate_attr(struct inode
*inode
)
129 fuse_invalidate_attr_mask(inode
, STATX_BASIC_STATS
);
132 static void fuse_dir_changed(struct inode
*dir
)
134 fuse_invalidate_attr(dir
);
135 inode_maybe_inc_iversion(dir
, false);
139 * Mark the attributes as stale due to an atime change. Avoid the invalidate if
142 void fuse_invalidate_atime(struct inode
*inode
)
144 if (!IS_RDONLY(inode
))
145 fuse_invalidate_attr_mask(inode
, STATX_ATIME
);
149 * Just mark the entry as stale, so that a next attempt to look it up
150 * will result in a new lookup call to userspace
152 * This is called when a dentry is about to become negative and the
153 * timeout is unknown (unlink, rmdir, rename and in some cases
156 void fuse_invalidate_entry_cache(struct dentry
*entry
)
158 fuse_dentry_settime(entry
, 0);
162 * Same as fuse_invalidate_entry_cache(), but also try to remove the
163 * dentry from the hash
165 static void fuse_invalidate_entry(struct dentry
*entry
)
168 fuse_invalidate_entry_cache(entry
);
171 static void fuse_lookup_init(struct fuse_conn
*fc
, struct fuse_args
*args
,
172 u64 nodeid
, const struct qstr
*name
,
173 struct fuse_entry_out
*outarg
)
175 memset(outarg
, 0, sizeof(struct fuse_entry_out
));
176 args
->opcode
= FUSE_LOOKUP
;
177 args
->nodeid
= nodeid
;
178 args
->in_numargs
= 1;
179 args
->in_args
[0].size
= name
->len
+ 1;
180 args
->in_args
[0].value
= name
->name
;
181 args
->out_numargs
= 1;
182 args
->out_args
[0].size
= sizeof(struct fuse_entry_out
);
183 args
->out_args
[0].value
= outarg
;
187 * Check whether the dentry is still valid
189 * If the entry validity timeout has expired and the dentry is
190 * positive, try to redo the lookup. If the lookup results in a
191 * different inode, then let the VFS invalidate the dentry and redo
192 * the lookup once more. If the lookup results in the same inode,
193 * then refresh the attributes, timeouts and mark the dentry valid.
195 static int fuse_dentry_revalidate(struct dentry
*entry
, unsigned int flags
)
198 struct dentry
*parent
;
199 struct fuse_mount
*fm
;
200 struct fuse_inode
*fi
;
203 inode
= d_inode_rcu(entry
);
204 if (inode
&& fuse_is_bad(inode
))
206 else if (time_before64(fuse_dentry_time(entry
), get_jiffies_64()) ||
207 (flags
& (LOOKUP_EXCL
| LOOKUP_REVAL
| LOOKUP_RENAME_TARGET
))) {
208 struct fuse_entry_out outarg
;
210 struct fuse_forget_link
*forget
;
213 /* For negative dentries, always do a fresh lookup */
218 if (flags
& LOOKUP_RCU
)
221 fm
= get_fuse_mount(inode
);
223 forget
= fuse_alloc_forget();
228 attr_version
= fuse_get_attr_version(fm
->fc
);
230 parent
= dget_parent(entry
);
231 fuse_lookup_init(fm
->fc
, &args
, get_node_id(d_inode(parent
)),
232 &entry
->d_name
, &outarg
);
233 ret
= fuse_simple_request(fm
, &args
);
235 /* Zero nodeid is same as -ENOENT */
236 if (!ret
&& !outarg
.nodeid
)
239 fi
= get_fuse_inode(inode
);
240 if (outarg
.nodeid
!= get_node_id(inode
) ||
241 (bool) IS_AUTOMOUNT(inode
) != (bool) (outarg
.attr
.flags
& FUSE_ATTR_SUBMOUNT
)) {
242 fuse_queue_forget(fm
->fc
, forget
,
246 spin_lock(&fi
->lock
);
248 spin_unlock(&fi
->lock
);
251 if (ret
== -ENOMEM
|| ret
== -EINTR
)
253 if (ret
|| fuse_invalid_attr(&outarg
.attr
) ||
254 fuse_stale_inode(inode
, outarg
.generation
, &outarg
.attr
))
257 forget_all_cached_acls(inode
);
258 fuse_change_attributes(inode
, &outarg
.attr
, NULL
,
259 ATTR_TIMEOUT(&outarg
),
261 fuse_change_entry_timeout(entry
, &outarg
);
263 fi
= get_fuse_inode(inode
);
264 if (flags
& LOOKUP_RCU
) {
265 if (test_bit(FUSE_I_INIT_RDPLUS
, &fi
->state
))
267 } else if (test_and_clear_bit(FUSE_I_INIT_RDPLUS
, &fi
->state
)) {
268 parent
= dget_parent(entry
);
269 fuse_advise_use_readdirplus(d_inode(parent
));
282 #if BITS_PER_LONG < 64
283 static int fuse_dentry_init(struct dentry
*dentry
)
285 dentry
->d_fsdata
= kzalloc(sizeof(union fuse_dentry
),
286 GFP_KERNEL_ACCOUNT
| __GFP_RECLAIMABLE
);
288 return dentry
->d_fsdata
? 0 : -ENOMEM
;
290 static void fuse_dentry_release(struct dentry
*dentry
)
292 union fuse_dentry
*fd
= dentry
->d_fsdata
;
298 static int fuse_dentry_delete(const struct dentry
*dentry
)
300 return time_before64(fuse_dentry_time(dentry
), get_jiffies_64());
304 * Create a fuse_mount object with a new superblock (with path->dentry
305 * as the root), and return that mount so it can be auto-mounted on
308 static struct vfsmount
*fuse_dentry_automount(struct path
*path
)
310 struct fs_context
*fsc
;
311 struct vfsmount
*mnt
;
312 struct fuse_inode
*mp_fi
= get_fuse_inode(d_inode(path
->dentry
));
314 fsc
= fs_context_for_submount(path
->mnt
->mnt_sb
->s_type
, path
->dentry
);
316 return ERR_CAST(fsc
);
318 /* Pass the FUSE inode of the mount for fuse_get_tree_submount() */
319 fsc
->fs_private
= mp_fi
;
321 /* Create the submount */
330 const struct dentry_operations fuse_dentry_operations
= {
331 .d_revalidate
= fuse_dentry_revalidate
,
332 .d_delete
= fuse_dentry_delete
,
333 #if BITS_PER_LONG < 64
334 .d_init
= fuse_dentry_init
,
335 .d_release
= fuse_dentry_release
,
337 .d_automount
= fuse_dentry_automount
,
340 const struct dentry_operations fuse_root_dentry_operations
= {
341 #if BITS_PER_LONG < 64
342 .d_init
= fuse_dentry_init
,
343 .d_release
= fuse_dentry_release
,
347 int fuse_valid_type(int m
)
349 return S_ISREG(m
) || S_ISDIR(m
) || S_ISLNK(m
) || S_ISCHR(m
) ||
350 S_ISBLK(m
) || S_ISFIFO(m
) || S_ISSOCK(m
);
353 static bool fuse_valid_size(u64 size
)
355 return size
<= LLONG_MAX
;
358 bool fuse_invalid_attr(struct fuse_attr
*attr
)
360 return !fuse_valid_type(attr
->mode
) || !fuse_valid_size(attr
->size
);
363 int fuse_lookup_name(struct super_block
*sb
, u64 nodeid
, const struct qstr
*name
,
364 struct fuse_entry_out
*outarg
, struct inode
**inode
)
366 struct fuse_mount
*fm
= get_fuse_mount_super(sb
);
368 struct fuse_forget_link
*forget
;
374 if (name
->len
> FUSE_NAME_MAX
)
378 forget
= fuse_alloc_forget();
383 attr_version
= fuse_get_attr_version(fm
->fc
);
385 fuse_lookup_init(fm
->fc
, &args
, nodeid
, name
, outarg
);
386 err
= fuse_simple_request(fm
, &args
);
387 /* Zero nodeid is same as -ENOENT, but with valid timeout */
388 if (err
|| !outarg
->nodeid
)
392 if (fuse_invalid_attr(&outarg
->attr
))
394 if (outarg
->nodeid
== FUSE_ROOT_ID
&& outarg
->generation
!= 0) {
395 pr_warn_once("root generation should be zero\n");
396 outarg
->generation
= 0;
399 *inode
= fuse_iget(sb
, outarg
->nodeid
, outarg
->generation
,
400 &outarg
->attr
, ATTR_TIMEOUT(outarg
),
404 fuse_queue_forget(fm
->fc
, forget
, outarg
->nodeid
, 1);
415 static struct dentry
*fuse_lookup(struct inode
*dir
, struct dentry
*entry
,
419 struct fuse_entry_out outarg
;
421 struct dentry
*newent
;
422 bool outarg_valid
= true;
425 if (fuse_is_bad(dir
))
426 return ERR_PTR(-EIO
);
428 locked
= fuse_lock_inode(dir
);
429 err
= fuse_lookup_name(dir
->i_sb
, get_node_id(dir
), &entry
->d_name
,
431 fuse_unlock_inode(dir
, locked
);
432 if (err
== -ENOENT
) {
433 outarg_valid
= false;
440 if (inode
&& get_node_id(inode
) == FUSE_ROOT_ID
)
443 newent
= d_splice_alias(inode
, entry
);
444 err
= PTR_ERR(newent
);
448 entry
= newent
? newent
: entry
;
450 fuse_change_entry_timeout(entry
, &outarg
);
452 fuse_invalidate_entry_cache(entry
);
455 fuse_advise_use_readdirplus(dir
);
464 static int get_security_context(struct dentry
*entry
, umode_t mode
,
465 struct fuse_in_arg
*ext
)
467 struct fuse_secctx
*fctx
;
468 struct fuse_secctx_header
*header
;
469 void *ctx
= NULL
, *ptr
;
470 u32 ctxlen
, total_len
= sizeof(*header
);
475 err
= security_dentry_init_security(entry
, mode
, &entry
->d_name
,
476 &name
, &ctx
, &ctxlen
);
478 if (err
!= -EOPNOTSUPP
)
480 /* No LSM is supporting this security hook. Ignore error */
487 namelen
= strlen(name
) + 1;
489 if (WARN_ON(namelen
> XATTR_NAME_MAX
+ 1 || ctxlen
> S32_MAX
))
491 total_len
+= FUSE_REC_ALIGN(sizeof(*fctx
) + namelen
+ ctxlen
);
495 header
= ptr
= kzalloc(total_len
, GFP_KERNEL
);
499 header
->nr_secctx
= nr_ctx
;
500 header
->size
= total_len
;
501 ptr
+= sizeof(*header
);
505 ptr
+= sizeof(*fctx
);
510 memcpy(ptr
, ctx
, ctxlen
);
512 ext
->size
= total_len
;
520 static void *extend_arg(struct fuse_in_arg
*buf
, u32 bytes
)
523 u32 newlen
= buf
->size
+ bytes
;
525 p
= krealloc(buf
->value
, newlen
, GFP_KERNEL
);
533 memset(p
+ buf
->size
, 0, bytes
);
537 return p
+ newlen
- bytes
;
540 static u32
fuse_ext_size(size_t size
)
542 return FUSE_REC_ALIGN(sizeof(struct fuse_ext_header
) + size
);
546 * This adds just a single supplementary group that matches the parent's group.
548 static int get_create_supp_group(struct mnt_idmap
*idmap
,
550 struct fuse_in_arg
*ext
)
552 struct fuse_conn
*fc
= get_fuse_conn(dir
);
553 struct fuse_ext_header
*xh
;
554 struct fuse_supp_groups
*sg
;
555 kgid_t kgid
= dir
->i_gid
;
556 vfsgid_t vfsgid
= make_vfsgid(idmap
, fc
->user_ns
, kgid
);
557 gid_t parent_gid
= from_kgid(fc
->user_ns
, kgid
);
559 u32 sg_len
= fuse_ext_size(sizeof(*sg
) + sizeof(sg
->groups
[0]));
561 if (parent_gid
== (gid_t
) -1 || vfsgid_eq_kgid(vfsgid
, current_fsgid()) ||
562 !vfsgid_in_group_p(vfsgid
))
565 xh
= extend_arg(ext
, sg_len
);
570 xh
->type
= FUSE_EXT_GROUPS
;
572 sg
= (struct fuse_supp_groups
*) &xh
[1];
574 sg
->groups
[0] = parent_gid
;
579 static int get_create_ext(struct mnt_idmap
*idmap
,
580 struct fuse_args
*args
,
581 struct inode
*dir
, struct dentry
*dentry
,
584 struct fuse_conn
*fc
= get_fuse_conn_super(dentry
->d_sb
);
585 struct fuse_in_arg ext
= { .size
= 0, .value
= NULL
};
588 if (fc
->init_security
)
589 err
= get_security_context(dentry
, mode
, &ext
);
590 if (!err
&& fc
->create_supp_group
)
591 err
= get_create_supp_group(idmap
, dir
, &ext
);
593 if (!err
&& ext
.size
) {
594 WARN_ON(args
->in_numargs
>= ARRAY_SIZE(args
->in_args
));
596 args
->ext_idx
= args
->in_numargs
++;
597 args
->in_args
[args
->ext_idx
] = ext
;
605 static void free_ext_value(struct fuse_args
*args
)
608 kfree(args
->in_args
[args
->ext_idx
].value
);
612 * Atomic create+open operation
614 * If the filesystem doesn't support this, then fall back to separate
615 * 'mknod' + 'open' requests.
617 static int fuse_create_open(struct mnt_idmap
*idmap
, struct inode
*dir
,
618 struct dentry
*entry
, struct file
*file
,
619 unsigned int flags
, umode_t mode
, u32 opcode
)
623 struct fuse_mount
*fm
= get_fuse_mount(dir
);
625 struct fuse_forget_link
*forget
;
626 struct fuse_create_in inarg
;
627 struct fuse_open_out
*outopenp
;
628 struct fuse_entry_out outentry
;
629 struct fuse_inode
*fi
;
630 struct fuse_file
*ff
;
631 bool trunc
= flags
& O_TRUNC
;
633 /* Userspace expects S_IFREG in create mode */
634 BUG_ON((mode
& S_IFMT
) != S_IFREG
);
636 forget
= fuse_alloc_forget();
642 ff
= fuse_file_alloc(fm
, true);
644 goto out_put_forget_req
;
646 if (!fm
->fc
->dont_mask
)
647 mode
&= ~current_umask();
650 memset(&inarg
, 0, sizeof(inarg
));
651 memset(&outentry
, 0, sizeof(outentry
));
654 inarg
.umask
= current_umask();
656 if (fm
->fc
->handle_killpriv_v2
&& trunc
&&
657 !(flags
& O_EXCL
) && !capable(CAP_FSETID
)) {
658 inarg
.open_flags
|= FUSE_OPEN_KILL_SUIDGID
;
661 args
.opcode
= opcode
;
662 args
.nodeid
= get_node_id(dir
);
664 args
.in_args
[0].size
= sizeof(inarg
);
665 args
.in_args
[0].value
= &inarg
;
666 args
.in_args
[1].size
= entry
->d_name
.len
+ 1;
667 args
.in_args
[1].value
= entry
->d_name
.name
;
668 args
.out_numargs
= 2;
669 args
.out_args
[0].size
= sizeof(outentry
);
670 args
.out_args
[0].value
= &outentry
;
671 /* Store outarg for fuse_finish_open() */
672 outopenp
= &ff
->args
->open_outarg
;
673 args
.out_args
[1].size
= sizeof(*outopenp
);
674 args
.out_args
[1].value
= outopenp
;
676 err
= get_create_ext(idmap
, &args
, dir
, entry
, mode
);
680 err
= fuse_simple_idmap_request(idmap
, fm
, &args
);
681 free_ext_value(&args
);
686 if (!S_ISREG(outentry
.attr
.mode
) || invalid_nodeid(outentry
.nodeid
) ||
687 fuse_invalid_attr(&outentry
.attr
))
690 ff
->fh
= outopenp
->fh
;
691 ff
->nodeid
= outentry
.nodeid
;
692 ff
->open_flags
= outopenp
->open_flags
;
693 inode
= fuse_iget(dir
->i_sb
, outentry
.nodeid
, outentry
.generation
,
694 &outentry
.attr
, ATTR_TIMEOUT(&outentry
), 0);
696 flags
&= ~(O_CREAT
| O_EXCL
| O_TRUNC
);
697 fuse_sync_release(NULL
, ff
, flags
);
698 fuse_queue_forget(fm
->fc
, forget
, outentry
.nodeid
, 1);
703 d_instantiate(entry
, inode
);
704 fuse_change_entry_timeout(entry
, &outentry
);
705 fuse_dir_changed(dir
);
706 err
= generic_file_open(inode
, file
);
708 file
->private_data
= ff
;
709 err
= finish_open(file
, entry
, fuse_finish_open
);
712 fi
= get_fuse_inode(inode
);
713 fuse_sync_release(fi
, ff
, flags
);
715 if (fm
->fc
->atomic_o_trunc
&& trunc
)
716 truncate_pagecache(inode
, 0);
717 else if (!(ff
->open_flags
& FOPEN_KEEP_CACHE
))
718 invalidate_inode_pages2(inode
->i_mapping
);
730 static int fuse_mknod(struct mnt_idmap
*, struct inode
*, struct dentry
*,
732 static int fuse_atomic_open(struct inode
*dir
, struct dentry
*entry
,
733 struct file
*file
, unsigned flags
,
737 struct mnt_idmap
*idmap
= file_mnt_idmap(file
);
738 struct fuse_conn
*fc
= get_fuse_conn(dir
);
739 struct dentry
*res
= NULL
;
741 if (fuse_is_bad(dir
))
744 if (d_in_lookup(entry
)) {
745 res
= fuse_lookup(dir
, entry
, 0);
753 if (!(flags
& O_CREAT
) || d_really_is_positive(entry
))
757 file
->f_mode
|= FMODE_CREATED
;
762 err
= fuse_create_open(idmap
, dir
, entry
, file
, flags
, mode
, FUSE_CREATE
);
763 if (err
== -ENOSYS
) {
766 } else if (err
== -EEXIST
)
767 fuse_invalidate_entry(entry
);
773 err
= fuse_mknod(idmap
, dir
, entry
, mode
, 0);
777 return finish_no_open(file
, res
);
781 * Code shared between mknod, mkdir, symlink and link
783 static int create_new_entry(struct mnt_idmap
*idmap
, struct fuse_mount
*fm
,
784 struct fuse_args
*args
, struct inode
*dir
,
785 struct dentry
*entry
, umode_t mode
)
787 struct fuse_entry_out outarg
;
791 struct fuse_forget_link
*forget
;
793 if (fuse_is_bad(dir
))
796 forget
= fuse_alloc_forget();
800 memset(&outarg
, 0, sizeof(outarg
));
801 args
->nodeid
= get_node_id(dir
);
802 args
->out_numargs
= 1;
803 args
->out_args
[0].size
= sizeof(outarg
);
804 args
->out_args
[0].value
= &outarg
;
806 if (args
->opcode
!= FUSE_LINK
) {
807 err
= get_create_ext(idmap
, args
, dir
, entry
, mode
);
809 goto out_put_forget_req
;
812 err
= fuse_simple_idmap_request(idmap
, fm
, args
);
813 free_ext_value(args
);
815 goto out_put_forget_req
;
818 if (invalid_nodeid(outarg
.nodeid
) || fuse_invalid_attr(&outarg
.attr
))
819 goto out_put_forget_req
;
821 if ((outarg
.attr
.mode
^ mode
) & S_IFMT
)
822 goto out_put_forget_req
;
824 inode
= fuse_iget(dir
->i_sb
, outarg
.nodeid
, outarg
.generation
,
825 &outarg
.attr
, ATTR_TIMEOUT(&outarg
), 0);
827 fuse_queue_forget(fm
->fc
, forget
, outarg
.nodeid
, 1);
833 d
= d_splice_alias(inode
, entry
);
838 fuse_change_entry_timeout(d
, &outarg
);
841 fuse_change_entry_timeout(entry
, &outarg
);
843 fuse_dir_changed(dir
);
848 fuse_invalidate_entry(entry
);
853 static int fuse_mknod(struct mnt_idmap
*idmap
, struct inode
*dir
,
854 struct dentry
*entry
, umode_t mode
, dev_t rdev
)
856 struct fuse_mknod_in inarg
;
857 struct fuse_mount
*fm
= get_fuse_mount(dir
);
860 if (!fm
->fc
->dont_mask
)
861 mode
&= ~current_umask();
863 memset(&inarg
, 0, sizeof(inarg
));
865 inarg
.rdev
= new_encode_dev(rdev
);
866 inarg
.umask
= current_umask();
867 args
.opcode
= FUSE_MKNOD
;
869 args
.in_args
[0].size
= sizeof(inarg
);
870 args
.in_args
[0].value
= &inarg
;
871 args
.in_args
[1].size
= entry
->d_name
.len
+ 1;
872 args
.in_args
[1].value
= entry
->d_name
.name
;
873 return create_new_entry(idmap
, fm
, &args
, dir
, entry
, mode
);
876 static int fuse_create(struct mnt_idmap
*idmap
, struct inode
*dir
,
877 struct dentry
*entry
, umode_t mode
, bool excl
)
879 return fuse_mknod(idmap
, dir
, entry
, mode
, 0);
882 static int fuse_tmpfile(struct mnt_idmap
*idmap
, struct inode
*dir
,
883 struct file
*file
, umode_t mode
)
885 struct fuse_conn
*fc
= get_fuse_conn(dir
);
891 err
= fuse_create_open(idmap
, dir
, file
->f_path
.dentry
, file
,
892 file
->f_flags
, mode
, FUSE_TMPFILE
);
893 if (err
== -ENOSYS
) {
900 static int fuse_mkdir(struct mnt_idmap
*idmap
, struct inode
*dir
,
901 struct dentry
*entry
, umode_t mode
)
903 struct fuse_mkdir_in inarg
;
904 struct fuse_mount
*fm
= get_fuse_mount(dir
);
907 if (!fm
->fc
->dont_mask
)
908 mode
&= ~current_umask();
910 memset(&inarg
, 0, sizeof(inarg
));
912 inarg
.umask
= current_umask();
913 args
.opcode
= FUSE_MKDIR
;
915 args
.in_args
[0].size
= sizeof(inarg
);
916 args
.in_args
[0].value
= &inarg
;
917 args
.in_args
[1].size
= entry
->d_name
.len
+ 1;
918 args
.in_args
[1].value
= entry
->d_name
.name
;
919 return create_new_entry(idmap
, fm
, &args
, dir
, entry
, S_IFDIR
);
922 static int fuse_symlink(struct mnt_idmap
*idmap
, struct inode
*dir
,
923 struct dentry
*entry
, const char *link
)
925 struct fuse_mount
*fm
= get_fuse_mount(dir
);
926 unsigned len
= strlen(link
) + 1;
929 args
.opcode
= FUSE_SYMLINK
;
931 args
.in_args
[0].size
= entry
->d_name
.len
+ 1;
932 args
.in_args
[0].value
= entry
->d_name
.name
;
933 args
.in_args
[1].size
= len
;
934 args
.in_args
[1].value
= link
;
935 return create_new_entry(idmap
, fm
, &args
, dir
, entry
, S_IFLNK
);
938 void fuse_flush_time_update(struct inode
*inode
)
940 int err
= sync_inode_metadata(inode
, 1);
942 mapping_set_error(inode
->i_mapping
, err
);
945 static void fuse_update_ctime_in_cache(struct inode
*inode
)
947 if (!IS_NOCMTIME(inode
)) {
948 inode_set_ctime_current(inode
);
949 mark_inode_dirty_sync(inode
);
950 fuse_flush_time_update(inode
);
954 void fuse_update_ctime(struct inode
*inode
)
956 fuse_invalidate_attr_mask(inode
, STATX_CTIME
);
957 fuse_update_ctime_in_cache(inode
);
960 static void fuse_entry_unlinked(struct dentry
*entry
)
962 struct inode
*inode
= d_inode(entry
);
963 struct fuse_conn
*fc
= get_fuse_conn(inode
);
964 struct fuse_inode
*fi
= get_fuse_inode(inode
);
966 spin_lock(&fi
->lock
);
967 fi
->attr_version
= atomic64_inc_return(&fc
->attr_version
);
969 * If i_nlink == 0 then unlink doesn't make sense, yet this can
970 * happen if userspace filesystem is careless. It would be
971 * difficult to enforce correct nlink usage so just ignore this
974 if (S_ISDIR(inode
->i_mode
))
976 else if (inode
->i_nlink
> 0)
978 spin_unlock(&fi
->lock
);
979 fuse_invalidate_entry_cache(entry
);
980 fuse_update_ctime(inode
);
983 static int fuse_unlink(struct inode
*dir
, struct dentry
*entry
)
986 struct fuse_mount
*fm
= get_fuse_mount(dir
);
989 if (fuse_is_bad(dir
))
992 args
.opcode
= FUSE_UNLINK
;
993 args
.nodeid
= get_node_id(dir
);
995 args
.in_args
[0].size
= entry
->d_name
.len
+ 1;
996 args
.in_args
[0].value
= entry
->d_name
.name
;
997 err
= fuse_simple_request(fm
, &args
);
999 fuse_dir_changed(dir
);
1000 fuse_entry_unlinked(entry
);
1001 } else if (err
== -EINTR
|| err
== -ENOENT
)
1002 fuse_invalidate_entry(entry
);
1006 static int fuse_rmdir(struct inode
*dir
, struct dentry
*entry
)
1009 struct fuse_mount
*fm
= get_fuse_mount(dir
);
1012 if (fuse_is_bad(dir
))
1015 args
.opcode
= FUSE_RMDIR
;
1016 args
.nodeid
= get_node_id(dir
);
1017 args
.in_numargs
= 1;
1018 args
.in_args
[0].size
= entry
->d_name
.len
+ 1;
1019 args
.in_args
[0].value
= entry
->d_name
.name
;
1020 err
= fuse_simple_request(fm
, &args
);
1022 fuse_dir_changed(dir
);
1023 fuse_entry_unlinked(entry
);
1024 } else if (err
== -EINTR
|| err
== -ENOENT
)
1025 fuse_invalidate_entry(entry
);
1029 static int fuse_rename_common(struct mnt_idmap
*idmap
, struct inode
*olddir
, struct dentry
*oldent
,
1030 struct inode
*newdir
, struct dentry
*newent
,
1031 unsigned int flags
, int opcode
, size_t argsize
)
1034 struct fuse_rename2_in inarg
;
1035 struct fuse_mount
*fm
= get_fuse_mount(olddir
);
1038 memset(&inarg
, 0, argsize
);
1039 inarg
.newdir
= get_node_id(newdir
);
1040 inarg
.flags
= flags
;
1041 args
.opcode
= opcode
;
1042 args
.nodeid
= get_node_id(olddir
);
1043 args
.in_numargs
= 3;
1044 args
.in_args
[0].size
= argsize
;
1045 args
.in_args
[0].value
= &inarg
;
1046 args
.in_args
[1].size
= oldent
->d_name
.len
+ 1;
1047 args
.in_args
[1].value
= oldent
->d_name
.name
;
1048 args
.in_args
[2].size
= newent
->d_name
.len
+ 1;
1049 args
.in_args
[2].value
= newent
->d_name
.name
;
1050 err
= fuse_simple_idmap_request(idmap
, fm
, &args
);
1053 fuse_update_ctime(d_inode(oldent
));
1055 if (flags
& RENAME_EXCHANGE
)
1056 fuse_update_ctime(d_inode(newent
));
1058 fuse_dir_changed(olddir
);
1059 if (olddir
!= newdir
)
1060 fuse_dir_changed(newdir
);
1062 /* newent will end up negative */
1063 if (!(flags
& RENAME_EXCHANGE
) && d_really_is_positive(newent
))
1064 fuse_entry_unlinked(newent
);
1065 } else if (err
== -EINTR
|| err
== -ENOENT
) {
1066 /* If request was interrupted, DEITY only knows if the
1067 rename actually took place. If the invalidation
1068 fails (e.g. some process has CWD under the renamed
1069 directory), then there can be inconsistency between
1070 the dcache and the real filesystem. Tough luck. */
1071 fuse_invalidate_entry(oldent
);
1072 if (d_really_is_positive(newent
))
1073 fuse_invalidate_entry(newent
);
1079 static int fuse_rename2(struct mnt_idmap
*idmap
, struct inode
*olddir
,
1080 struct dentry
*oldent
, struct inode
*newdir
,
1081 struct dentry
*newent
, unsigned int flags
)
1083 struct fuse_conn
*fc
= get_fuse_conn(olddir
);
1086 if (fuse_is_bad(olddir
))
1089 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
1093 if (fc
->no_rename2
|| fc
->minor
< 23)
1096 err
= fuse_rename_common((flags
& RENAME_WHITEOUT
) ? idmap
: &invalid_mnt_idmap
,
1097 olddir
, oldent
, newdir
, newent
, flags
,
1099 sizeof(struct fuse_rename2_in
));
1100 if (err
== -ENOSYS
) {
1105 err
= fuse_rename_common(&invalid_mnt_idmap
, olddir
, oldent
, newdir
, newent
, 0,
1107 sizeof(struct fuse_rename_in
));
1113 static int fuse_link(struct dentry
*entry
, struct inode
*newdir
,
1114 struct dentry
*newent
)
1117 struct fuse_link_in inarg
;
1118 struct inode
*inode
= d_inode(entry
);
1119 struct fuse_mount
*fm
= get_fuse_mount(inode
);
1122 memset(&inarg
, 0, sizeof(inarg
));
1123 inarg
.oldnodeid
= get_node_id(inode
);
1124 args
.opcode
= FUSE_LINK
;
1125 args
.in_numargs
= 2;
1126 args
.in_args
[0].size
= sizeof(inarg
);
1127 args
.in_args
[0].value
= &inarg
;
1128 args
.in_args
[1].size
= newent
->d_name
.len
+ 1;
1129 args
.in_args
[1].value
= newent
->d_name
.name
;
1130 err
= create_new_entry(&invalid_mnt_idmap
, fm
, &args
, newdir
, newent
, inode
->i_mode
);
1132 fuse_update_ctime_in_cache(inode
);
1133 else if (err
== -EINTR
)
1134 fuse_invalidate_attr(inode
);
1139 static void fuse_fillattr(struct mnt_idmap
*idmap
, struct inode
*inode
,
1140 struct fuse_attr
*attr
, struct kstat
*stat
)
1142 unsigned int blkbits
;
1143 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1144 vfsuid_t vfsuid
= make_vfsuid(idmap
, fc
->user_ns
,
1145 make_kuid(fc
->user_ns
, attr
->uid
));
1146 vfsgid_t vfsgid
= make_vfsgid(idmap
, fc
->user_ns
,
1147 make_kgid(fc
->user_ns
, attr
->gid
));
1149 stat
->dev
= inode
->i_sb
->s_dev
;
1150 stat
->ino
= attr
->ino
;
1151 stat
->mode
= (inode
->i_mode
& S_IFMT
) | (attr
->mode
& 07777);
1152 stat
->nlink
= attr
->nlink
;
1153 stat
->uid
= vfsuid_into_kuid(vfsuid
);
1154 stat
->gid
= vfsgid_into_kgid(vfsgid
);
1155 stat
->rdev
= inode
->i_rdev
;
1156 stat
->atime
.tv_sec
= attr
->atime
;
1157 stat
->atime
.tv_nsec
= attr
->atimensec
;
1158 stat
->mtime
.tv_sec
= attr
->mtime
;
1159 stat
->mtime
.tv_nsec
= attr
->mtimensec
;
1160 stat
->ctime
.tv_sec
= attr
->ctime
;
1161 stat
->ctime
.tv_nsec
= attr
->ctimensec
;
1162 stat
->size
= attr
->size
;
1163 stat
->blocks
= attr
->blocks
;
1165 if (attr
->blksize
!= 0)
1166 blkbits
= ilog2(attr
->blksize
);
1168 blkbits
= inode
->i_sb
->s_blocksize_bits
;
1170 stat
->blksize
= 1 << blkbits
;
1173 static void fuse_statx_to_attr(struct fuse_statx
*sx
, struct fuse_attr
*attr
)
1175 memset(attr
, 0, sizeof(*attr
));
1176 attr
->ino
= sx
->ino
;
1177 attr
->size
= sx
->size
;
1178 attr
->blocks
= sx
->blocks
;
1179 attr
->atime
= sx
->atime
.tv_sec
;
1180 attr
->mtime
= sx
->mtime
.tv_sec
;
1181 attr
->ctime
= sx
->ctime
.tv_sec
;
1182 attr
->atimensec
= sx
->atime
.tv_nsec
;
1183 attr
->mtimensec
= sx
->mtime
.tv_nsec
;
1184 attr
->ctimensec
= sx
->ctime
.tv_nsec
;
1185 attr
->mode
= sx
->mode
;
1186 attr
->nlink
= sx
->nlink
;
1187 attr
->uid
= sx
->uid
;
1188 attr
->gid
= sx
->gid
;
1189 attr
->rdev
= new_encode_dev(MKDEV(sx
->rdev_major
, sx
->rdev_minor
));
1190 attr
->blksize
= sx
->blksize
;
1193 static int fuse_do_statx(struct mnt_idmap
*idmap
, struct inode
*inode
,
1194 struct file
*file
, struct kstat
*stat
)
1197 struct fuse_attr attr
;
1198 struct fuse_statx
*sx
;
1199 struct fuse_statx_in inarg
;
1200 struct fuse_statx_out outarg
;
1201 struct fuse_mount
*fm
= get_fuse_mount(inode
);
1202 u64 attr_version
= fuse_get_attr_version(fm
->fc
);
1205 memset(&inarg
, 0, sizeof(inarg
));
1206 memset(&outarg
, 0, sizeof(outarg
));
1207 /* Directories have separate file-handle space */
1208 if (file
&& S_ISREG(inode
->i_mode
)) {
1209 struct fuse_file
*ff
= file
->private_data
;
1211 inarg
.getattr_flags
|= FUSE_GETATTR_FH
;
1214 /* For now leave sync hints as the default, request all stats. */
1216 inarg
.sx_mask
= STATX_BASIC_STATS
| STATX_BTIME
;
1217 args
.opcode
= FUSE_STATX
;
1218 args
.nodeid
= get_node_id(inode
);
1219 args
.in_numargs
= 1;
1220 args
.in_args
[0].size
= sizeof(inarg
);
1221 args
.in_args
[0].value
= &inarg
;
1222 args
.out_numargs
= 1;
1223 args
.out_args
[0].size
= sizeof(outarg
);
1224 args
.out_args
[0].value
= &outarg
;
1225 err
= fuse_simple_request(fm
, &args
);
1230 if (((sx
->mask
& STATX_SIZE
) && !fuse_valid_size(sx
->size
)) ||
1231 ((sx
->mask
& STATX_TYPE
) && (!fuse_valid_type(sx
->mode
) ||
1232 inode_wrong_type(inode
, sx
->mode
)))) {
1233 fuse_make_bad(inode
);
1237 fuse_statx_to_attr(&outarg
.stat
, &attr
);
1238 if ((sx
->mask
& STATX_BASIC_STATS
) == STATX_BASIC_STATS
) {
1239 fuse_change_attributes(inode
, &attr
, &outarg
.stat
,
1240 ATTR_TIMEOUT(&outarg
), attr_version
);
1244 stat
->result_mask
= sx
->mask
& (STATX_BASIC_STATS
| STATX_BTIME
);
1245 stat
->btime
.tv_sec
= sx
->btime
.tv_sec
;
1246 stat
->btime
.tv_nsec
= min_t(u32
, sx
->btime
.tv_nsec
, NSEC_PER_SEC
- 1);
1247 fuse_fillattr(idmap
, inode
, &attr
, stat
);
1248 stat
->result_mask
|= STATX_TYPE
;
1254 static int fuse_do_getattr(struct mnt_idmap
*idmap
, struct inode
*inode
,
1255 struct kstat
*stat
, struct file
*file
)
1258 struct fuse_getattr_in inarg
;
1259 struct fuse_attr_out outarg
;
1260 struct fuse_mount
*fm
= get_fuse_mount(inode
);
1264 attr_version
= fuse_get_attr_version(fm
->fc
);
1266 memset(&inarg
, 0, sizeof(inarg
));
1267 memset(&outarg
, 0, sizeof(outarg
));
1268 /* Directories have separate file-handle space */
1269 if (file
&& S_ISREG(inode
->i_mode
)) {
1270 struct fuse_file
*ff
= file
->private_data
;
1272 inarg
.getattr_flags
|= FUSE_GETATTR_FH
;
1275 args
.opcode
= FUSE_GETATTR
;
1276 args
.nodeid
= get_node_id(inode
);
1277 args
.in_numargs
= 1;
1278 args
.in_args
[0].size
= sizeof(inarg
);
1279 args
.in_args
[0].value
= &inarg
;
1280 args
.out_numargs
= 1;
1281 args
.out_args
[0].size
= sizeof(outarg
);
1282 args
.out_args
[0].value
= &outarg
;
1283 err
= fuse_simple_request(fm
, &args
);
1285 if (fuse_invalid_attr(&outarg
.attr
) ||
1286 inode_wrong_type(inode
, outarg
.attr
.mode
)) {
1287 fuse_make_bad(inode
);
1290 fuse_change_attributes(inode
, &outarg
.attr
, NULL
,
1291 ATTR_TIMEOUT(&outarg
),
1294 fuse_fillattr(idmap
, inode
, &outarg
.attr
, stat
);
1300 static int fuse_update_get_attr(struct mnt_idmap
*idmap
, struct inode
*inode
,
1301 struct file
*file
, struct kstat
*stat
,
1302 u32 request_mask
, unsigned int flags
)
1304 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1305 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1308 u32 inval_mask
= READ_ONCE(fi
->inval_mask
);
1309 u32 cache_mask
= fuse_get_cache_mask(inode
);
1312 /* FUSE only supports basic stats and possibly btime */
1313 request_mask
&= STATX_BASIC_STATS
| STATX_BTIME
;
1316 request_mask
&= STATX_BASIC_STATS
;
1320 else if (flags
& AT_STATX_FORCE_SYNC
)
1322 else if (flags
& AT_STATX_DONT_SYNC
)
1324 else if (request_mask
& inval_mask
& ~cache_mask
)
1327 sync
= time_before64(fi
->i_time
, get_jiffies_64());
1330 forget_all_cached_acls(inode
);
1331 /* Try statx if BTIME is requested */
1332 if (!fc
->no_statx
&& (request_mask
& ~STATX_BASIC_STATS
)) {
1333 err
= fuse_do_statx(idmap
, inode
, file
, stat
);
1334 if (err
== -ENOSYS
) {
1340 err
= fuse_do_getattr(idmap
, inode
, stat
, file
);
1343 generic_fillattr(idmap
, request_mask
, inode
, stat
);
1344 stat
->mode
= fi
->orig_i_mode
;
1345 stat
->ino
= fi
->orig_ino
;
1346 if (test_bit(FUSE_I_BTIME
, &fi
->state
)) {
1347 stat
->btime
= fi
->i_btime
;
1348 stat
->result_mask
|= STATX_BTIME
;
1355 int fuse_update_attributes(struct inode
*inode
, struct file
*file
, u32 mask
)
1357 return fuse_update_get_attr(&nop_mnt_idmap
, inode
, file
, NULL
, mask
, 0);
1360 int fuse_reverse_inval_entry(struct fuse_conn
*fc
, u64 parent_nodeid
,
1361 u64 child_nodeid
, struct qstr
*name
, u32 flags
)
1364 struct inode
*parent
;
1366 struct dentry
*entry
;
1368 parent
= fuse_ilookup(fc
, parent_nodeid
, NULL
);
1372 inode_lock_nested(parent
, I_MUTEX_PARENT
);
1373 if (!S_ISDIR(parent
->i_mode
))
1377 dir
= d_find_alias(parent
);
1381 name
->hash
= full_name_hash(dir
, name
->name
, name
->len
);
1382 entry
= d_lookup(dir
, name
);
1387 fuse_dir_changed(parent
);
1388 if (!(flags
& FUSE_EXPIRE_ONLY
))
1389 d_invalidate(entry
);
1390 fuse_invalidate_entry_cache(entry
);
1392 if (child_nodeid
!= 0 && d_really_is_positive(entry
)) {
1393 inode_lock(d_inode(entry
));
1394 if (get_node_id(d_inode(entry
)) != child_nodeid
) {
1398 if (d_mountpoint(entry
)) {
1402 if (d_is_dir(entry
)) {
1403 shrink_dcache_parent(entry
);
1404 if (!simple_empty(entry
)) {
1408 d_inode(entry
)->i_flags
|= S_DEAD
;
1411 clear_nlink(d_inode(entry
));
1414 inode_unlock(d_inode(entry
));
1423 inode_unlock(parent
);
1428 static inline bool fuse_permissible_uidgid(struct fuse_conn
*fc
)
1430 const struct cred
*cred
= current_cred();
1432 return (uid_eq(cred
->euid
, fc
->user_id
) &&
1433 uid_eq(cred
->suid
, fc
->user_id
) &&
1434 uid_eq(cred
->uid
, fc
->user_id
) &&
1435 gid_eq(cred
->egid
, fc
->group_id
) &&
1436 gid_eq(cred
->sgid
, fc
->group_id
) &&
1437 gid_eq(cred
->gid
, fc
->group_id
));
1441 * Calling into a user-controlled filesystem gives the filesystem
1442 * daemon ptrace-like capabilities over the current process. This
1443 * means, that the filesystem daemon is able to record the exact
1444 * filesystem operations performed, and can also control the behavior
1445 * of the requester process in otherwise impossible ways. For example
1446 * it can delay the operation for arbitrary length of time allowing
1447 * DoS against the requester.
1449 * For this reason only those processes can call into the filesystem,
1450 * for which the owner of the mount has ptrace privilege. This
1451 * excludes processes started by other users, suid or sgid processes.
1453 bool fuse_allow_current_process(struct fuse_conn
*fc
)
1457 if (fc
->allow_other
)
1458 allow
= current_in_userns(fc
->user_ns
);
1460 allow
= fuse_permissible_uidgid(fc
);
1462 if (!allow
&& allow_sys_admin_access
&& capable(CAP_SYS_ADMIN
))
1468 static int fuse_access(struct inode
*inode
, int mask
)
1470 struct fuse_mount
*fm
= get_fuse_mount(inode
);
1472 struct fuse_access_in inarg
;
1475 BUG_ON(mask
& MAY_NOT_BLOCK
);
1478 * We should not send FUSE_ACCESS to the userspace
1479 * when idmapped mounts are enabled as for this case
1480 * we have fc->default_permissions = 1 and access
1481 * permission checks are done on the kernel side.
1483 WARN_ON_ONCE(!(fm
->sb
->s_iflags
& SB_I_NOIDMAP
));
1485 if (fm
->fc
->no_access
)
1488 memset(&inarg
, 0, sizeof(inarg
));
1489 inarg
.mask
= mask
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
);
1490 args
.opcode
= FUSE_ACCESS
;
1491 args
.nodeid
= get_node_id(inode
);
1492 args
.in_numargs
= 1;
1493 args
.in_args
[0].size
= sizeof(inarg
);
1494 args
.in_args
[0].value
= &inarg
;
1495 err
= fuse_simple_request(fm
, &args
);
1496 if (err
== -ENOSYS
) {
1497 fm
->fc
->no_access
= 1;
1503 static int fuse_perm_getattr(struct inode
*inode
, int mask
)
1505 if (mask
& MAY_NOT_BLOCK
)
1508 forget_all_cached_acls(inode
);
1509 return fuse_do_getattr(&nop_mnt_idmap
, inode
, NULL
, NULL
);
1513 * Check permission. The two basic access models of FUSE are:
1515 * 1) Local access checking ('default_permissions' mount option) based
1516 * on file mode. This is the plain old disk filesystem permission
1519 * 2) "Remote" access checking, where server is responsible for
1520 * checking permission in each inode operation. An exception to this
1521 * is if ->permission() was invoked from sys_access() in which case an
1522 * access request is sent. Execute permission is still checked
1523 * locally based on file mode.
1525 static int fuse_permission(struct mnt_idmap
*idmap
,
1526 struct inode
*inode
, int mask
)
1528 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1529 bool refreshed
= false;
1532 if (fuse_is_bad(inode
))
1535 if (!fuse_allow_current_process(fc
))
1539 * If attributes are needed, refresh them before proceeding
1541 if (fc
->default_permissions
||
1542 ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
))) {
1543 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1544 u32 perm_mask
= STATX_MODE
| STATX_UID
| STATX_GID
;
1546 if (perm_mask
& READ_ONCE(fi
->inval_mask
) ||
1547 time_before64(fi
->i_time
, get_jiffies_64())) {
1550 err
= fuse_perm_getattr(inode
, mask
);
1556 if (fc
->default_permissions
) {
1557 err
= generic_permission(idmap
, inode
, mask
);
1559 /* If permission is denied, try to refresh file
1560 attributes. This is also needed, because the root
1561 node will at first have no permissions */
1562 if (err
== -EACCES
&& !refreshed
) {
1563 err
= fuse_perm_getattr(inode
, mask
);
1565 err
= generic_permission(idmap
,
1569 /* Note: the opposite of the above test does not
1570 exist. So if permissions are revoked this won't be
1571 noticed immediately, only after the attribute
1572 timeout has expired */
1573 } else if (mask
& (MAY_ACCESS
| MAY_CHDIR
)) {
1574 err
= fuse_access(inode
, mask
);
1575 } else if ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
)) {
1576 if (!(inode
->i_mode
& S_IXUGO
)) {
1580 err
= fuse_perm_getattr(inode
, mask
);
1581 if (!err
&& !(inode
->i_mode
& S_IXUGO
))
1588 static int fuse_readlink_page(struct inode
*inode
, struct page
*page
)
1590 struct fuse_mount
*fm
= get_fuse_mount(inode
);
1591 struct fuse_page_desc desc
= { .length
= PAGE_SIZE
- 1 };
1592 struct fuse_args_pages ap
= {
1600 ap
.args
.opcode
= FUSE_READLINK
;
1601 ap
.args
.nodeid
= get_node_id(inode
);
1602 ap
.args
.out_pages
= true;
1603 ap
.args
.out_argvar
= true;
1604 ap
.args
.page_zeroing
= true;
1605 ap
.args
.out_numargs
= 1;
1606 ap
.args
.out_args
[0].size
= desc
.length
;
1607 res
= fuse_simple_request(fm
, &ap
.args
);
1609 fuse_invalidate_atime(inode
);
1614 if (WARN_ON(res
>= PAGE_SIZE
))
1617 link
= page_address(page
);
1623 static const char *fuse_get_link(struct dentry
*dentry
, struct inode
*inode
,
1624 struct delayed_call
*callback
)
1626 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1631 if (fuse_is_bad(inode
))
1634 if (fc
->cache_symlinks
)
1635 return page_get_link(dentry
, inode
, callback
);
1641 page
= alloc_page(GFP_KERNEL
);
1646 err
= fuse_readlink_page(inode
, page
);
1652 set_delayed_call(callback
, page_put_link
, page
);
1654 return page_address(page
);
1657 return ERR_PTR(err
);
1660 static int fuse_dir_open(struct inode
*inode
, struct file
*file
)
1662 struct fuse_mount
*fm
= get_fuse_mount(inode
);
1665 if (fuse_is_bad(inode
))
1668 err
= generic_file_open(inode
, file
);
1672 err
= fuse_do_open(fm
, get_node_id(inode
), file
, true);
1674 struct fuse_file
*ff
= file
->private_data
;
1677 * Keep handling FOPEN_STREAM and FOPEN_NONSEEKABLE for
1678 * directories for backward compatibility, though it's unlikely
1681 if (ff
->open_flags
& (FOPEN_STREAM
| FOPEN_NONSEEKABLE
))
1682 nonseekable_open(inode
, file
);
1688 static int fuse_dir_release(struct inode
*inode
, struct file
*file
)
1690 fuse_release_common(file
, true);
1695 static int fuse_dir_fsync(struct file
*file
, loff_t start
, loff_t end
,
1698 struct inode
*inode
= file
->f_mapping
->host
;
1699 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1702 if (fuse_is_bad(inode
))
1705 if (fc
->no_fsyncdir
)
1709 err
= fuse_fsync_common(file
, start
, end
, datasync
, FUSE_FSYNCDIR
);
1710 if (err
== -ENOSYS
) {
1711 fc
->no_fsyncdir
= 1;
1714 inode_unlock(inode
);
1719 static long fuse_dir_ioctl(struct file
*file
, unsigned int cmd
,
1722 struct fuse_conn
*fc
= get_fuse_conn(file
->f_mapping
->host
);
1724 /* FUSE_IOCTL_DIR only supported for API version >= 7.18 */
1728 return fuse_ioctl_common(file
, cmd
, arg
, FUSE_IOCTL_DIR
);
1731 static long fuse_dir_compat_ioctl(struct file
*file
, unsigned int cmd
,
1734 struct fuse_conn
*fc
= get_fuse_conn(file
->f_mapping
->host
);
1739 return fuse_ioctl_common(file
, cmd
, arg
,
1740 FUSE_IOCTL_COMPAT
| FUSE_IOCTL_DIR
);
1743 static bool update_mtime(unsigned ivalid
, bool trust_local_mtime
)
1745 /* Always update if mtime is explicitly set */
1746 if (ivalid
& ATTR_MTIME_SET
)
1749 /* Or if kernel i_mtime is the official one */
1750 if (trust_local_mtime
)
1753 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1754 if ((ivalid
& ATTR_SIZE
) && (ivalid
& (ATTR_OPEN
| ATTR_FILE
)))
1757 /* In all other cases update */
1761 static void iattr_to_fattr(struct mnt_idmap
*idmap
, struct fuse_conn
*fc
,
1762 struct iattr
*iattr
, struct fuse_setattr_in
*arg
,
1763 bool trust_local_cmtime
)
1765 unsigned ivalid
= iattr
->ia_valid
;
1767 if (ivalid
& ATTR_MODE
)
1768 arg
->valid
|= FATTR_MODE
, arg
->mode
= iattr
->ia_mode
;
1770 if (ivalid
& ATTR_UID
) {
1771 kuid_t fsuid
= from_vfsuid(idmap
, fc
->user_ns
, iattr
->ia_vfsuid
);
1773 arg
->valid
|= FATTR_UID
;
1774 arg
->uid
= from_kuid(fc
->user_ns
, fsuid
);
1777 if (ivalid
& ATTR_GID
) {
1778 kgid_t fsgid
= from_vfsgid(idmap
, fc
->user_ns
, iattr
->ia_vfsgid
);
1780 arg
->valid
|= FATTR_GID
;
1781 arg
->gid
= from_kgid(fc
->user_ns
, fsgid
);
1784 if (ivalid
& ATTR_SIZE
)
1785 arg
->valid
|= FATTR_SIZE
, arg
->size
= iattr
->ia_size
;
1786 if (ivalid
& ATTR_ATIME
) {
1787 arg
->valid
|= FATTR_ATIME
;
1788 arg
->atime
= iattr
->ia_atime
.tv_sec
;
1789 arg
->atimensec
= iattr
->ia_atime
.tv_nsec
;
1790 if (!(ivalid
& ATTR_ATIME_SET
))
1791 arg
->valid
|= FATTR_ATIME_NOW
;
1793 if ((ivalid
& ATTR_MTIME
) && update_mtime(ivalid
, trust_local_cmtime
)) {
1794 arg
->valid
|= FATTR_MTIME
;
1795 arg
->mtime
= iattr
->ia_mtime
.tv_sec
;
1796 arg
->mtimensec
= iattr
->ia_mtime
.tv_nsec
;
1797 if (!(ivalid
& ATTR_MTIME_SET
) && !trust_local_cmtime
)
1798 arg
->valid
|= FATTR_MTIME_NOW
;
1800 if ((ivalid
& ATTR_CTIME
) && trust_local_cmtime
) {
1801 arg
->valid
|= FATTR_CTIME
;
1802 arg
->ctime
= iattr
->ia_ctime
.tv_sec
;
1803 arg
->ctimensec
= iattr
->ia_ctime
.tv_nsec
;
1808 * Prevent concurrent writepages on inode
1810 * This is done by adding a negative bias to the inode write counter
1811 * and waiting for all pending writes to finish.
1813 void fuse_set_nowrite(struct inode
*inode
)
1815 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1817 BUG_ON(!inode_is_locked(inode
));
1819 spin_lock(&fi
->lock
);
1820 BUG_ON(fi
->writectr
< 0);
1821 fi
->writectr
+= FUSE_NOWRITE
;
1822 spin_unlock(&fi
->lock
);
1823 wait_event(fi
->page_waitq
, fi
->writectr
== FUSE_NOWRITE
);
1827 * Allow writepages on inode
1829 * Remove the bias from the writecounter and send any queued
1832 static void __fuse_release_nowrite(struct inode
*inode
)
1834 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1836 BUG_ON(fi
->writectr
!= FUSE_NOWRITE
);
1838 fuse_flush_writepages(inode
);
1841 void fuse_release_nowrite(struct inode
*inode
)
1843 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1845 spin_lock(&fi
->lock
);
1846 __fuse_release_nowrite(inode
);
1847 spin_unlock(&fi
->lock
);
1850 static void fuse_setattr_fill(struct fuse_conn
*fc
, struct fuse_args
*args
,
1851 struct inode
*inode
,
1852 struct fuse_setattr_in
*inarg_p
,
1853 struct fuse_attr_out
*outarg_p
)
1855 args
->opcode
= FUSE_SETATTR
;
1856 args
->nodeid
= get_node_id(inode
);
1857 args
->in_numargs
= 1;
1858 args
->in_args
[0].size
= sizeof(*inarg_p
);
1859 args
->in_args
[0].value
= inarg_p
;
1860 args
->out_numargs
= 1;
1861 args
->out_args
[0].size
= sizeof(*outarg_p
);
1862 args
->out_args
[0].value
= outarg_p
;
1866 * Flush inode->i_mtime to the server
1868 int fuse_flush_times(struct inode
*inode
, struct fuse_file
*ff
)
1870 struct fuse_mount
*fm
= get_fuse_mount(inode
);
1872 struct fuse_setattr_in inarg
;
1873 struct fuse_attr_out outarg
;
1875 memset(&inarg
, 0, sizeof(inarg
));
1876 memset(&outarg
, 0, sizeof(outarg
));
1878 inarg
.valid
= FATTR_MTIME
;
1879 inarg
.mtime
= inode_get_mtime_sec(inode
);
1880 inarg
.mtimensec
= inode_get_mtime_nsec(inode
);
1881 if (fm
->fc
->minor
>= 23) {
1882 inarg
.valid
|= FATTR_CTIME
;
1883 inarg
.ctime
= inode_get_ctime_sec(inode
);
1884 inarg
.ctimensec
= inode_get_ctime_nsec(inode
);
1887 inarg
.valid
|= FATTR_FH
;
1890 fuse_setattr_fill(fm
->fc
, &args
, inode
, &inarg
, &outarg
);
1892 return fuse_simple_request(fm
, &args
);
1896 * Set attributes, and at the same time refresh them.
1898 * Truncation is slightly complicated, because the 'truncate' request
1899 * may fail, in which case we don't want to touch the mapping.
1900 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1901 * and the actual truncation by hand.
1903 int fuse_do_setattr(struct mnt_idmap
*idmap
, struct dentry
*dentry
,
1904 struct iattr
*attr
, struct file
*file
)
1906 struct inode
*inode
= d_inode(dentry
);
1907 struct fuse_mount
*fm
= get_fuse_mount(inode
);
1908 struct fuse_conn
*fc
= fm
->fc
;
1909 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1910 struct address_space
*mapping
= inode
->i_mapping
;
1912 struct fuse_setattr_in inarg
;
1913 struct fuse_attr_out outarg
;
1914 bool is_truncate
= false;
1915 bool is_wb
= fc
->writeback_cache
&& S_ISREG(inode
->i_mode
);
1918 bool trust_local_cmtime
= is_wb
;
1919 bool fault_blocked
= false;
1921 if (!fc
->default_permissions
)
1922 attr
->ia_valid
|= ATTR_FORCE
;
1924 err
= setattr_prepare(idmap
, dentry
, attr
);
1928 if (attr
->ia_valid
& ATTR_SIZE
) {
1929 if (WARN_ON(!S_ISREG(inode
->i_mode
)))
1934 if (FUSE_IS_DAX(inode
) && is_truncate
) {
1935 filemap_invalidate_lock(mapping
);
1936 fault_blocked
= true;
1937 err
= fuse_dax_break_layouts(inode
, 0, 0);
1939 filemap_invalidate_unlock(mapping
);
1944 if (attr
->ia_valid
& ATTR_OPEN
) {
1945 /* This is coming from open(..., ... | O_TRUNC); */
1946 WARN_ON(!(attr
->ia_valid
& ATTR_SIZE
));
1947 WARN_ON(attr
->ia_size
!= 0);
1948 if (fc
->atomic_o_trunc
) {
1950 * No need to send request to userspace, since actual
1951 * truncation has already been done by OPEN. But still
1952 * need to truncate page cache.
1954 i_size_write(inode
, 0);
1955 truncate_pagecache(inode
, 0);
1961 /* Flush dirty data/metadata before non-truncate SETATTR */
1964 (ATTR_MODE
| ATTR_UID
| ATTR_GID
| ATTR_MTIME_SET
|
1966 err
= write_inode_now(inode
, true);
1970 fuse_set_nowrite(inode
);
1971 fuse_release_nowrite(inode
);
1975 fuse_set_nowrite(inode
);
1976 set_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
);
1977 if (trust_local_cmtime
&& attr
->ia_size
!= inode
->i_size
)
1978 attr
->ia_valid
|= ATTR_MTIME
| ATTR_CTIME
;
1981 memset(&inarg
, 0, sizeof(inarg
));
1982 memset(&outarg
, 0, sizeof(outarg
));
1983 iattr_to_fattr(idmap
, fc
, attr
, &inarg
, trust_local_cmtime
);
1985 struct fuse_file
*ff
= file
->private_data
;
1986 inarg
.valid
|= FATTR_FH
;
1990 /* Kill suid/sgid for non-directory chown unconditionally */
1991 if (fc
->handle_killpriv_v2
&& !S_ISDIR(inode
->i_mode
) &&
1992 attr
->ia_valid
& (ATTR_UID
| ATTR_GID
))
1993 inarg
.valid
|= FATTR_KILL_SUIDGID
;
1995 if (attr
->ia_valid
& ATTR_SIZE
) {
1996 /* For mandatory locking in truncate */
1997 inarg
.valid
|= FATTR_LOCKOWNER
;
1998 inarg
.lock_owner
= fuse_lock_owner_id(fc
, current
->files
);
2000 /* Kill suid/sgid for truncate only if no CAP_FSETID */
2001 if (fc
->handle_killpriv_v2
&& !capable(CAP_FSETID
))
2002 inarg
.valid
|= FATTR_KILL_SUIDGID
;
2004 fuse_setattr_fill(fc
, &args
, inode
, &inarg
, &outarg
);
2005 err
= fuse_simple_request(fm
, &args
);
2008 fuse_invalidate_attr(inode
);
2012 if (fuse_invalid_attr(&outarg
.attr
) ||
2013 inode_wrong_type(inode
, outarg
.attr
.mode
)) {
2014 fuse_make_bad(inode
);
2019 spin_lock(&fi
->lock
);
2020 /* the kernel maintains i_mtime locally */
2021 if (trust_local_cmtime
) {
2022 if (attr
->ia_valid
& ATTR_MTIME
)
2023 inode_set_mtime_to_ts(inode
, attr
->ia_mtime
);
2024 if (attr
->ia_valid
& ATTR_CTIME
)
2025 inode_set_ctime_to_ts(inode
, attr
->ia_ctime
);
2026 /* FIXME: clear I_DIRTY_SYNC? */
2029 fuse_change_attributes_common(inode
, &outarg
.attr
, NULL
,
2030 ATTR_TIMEOUT(&outarg
),
2031 fuse_get_cache_mask(inode
));
2032 oldsize
= inode
->i_size
;
2033 /* see the comment in fuse_change_attributes() */
2034 if (!is_wb
|| is_truncate
)
2035 i_size_write(inode
, outarg
.attr
.size
);
2038 /* NOTE: this may release/reacquire fi->lock */
2039 __fuse_release_nowrite(inode
);
2041 spin_unlock(&fi
->lock
);
2044 * Only call invalidate_inode_pages2() after removing
2045 * FUSE_NOWRITE, otherwise fuse_launder_folio() would deadlock.
2047 if ((is_truncate
|| !is_wb
) &&
2048 S_ISREG(inode
->i_mode
) && oldsize
!= outarg
.attr
.size
) {
2049 truncate_pagecache(inode
, outarg
.attr
.size
);
2050 invalidate_inode_pages2(mapping
);
2053 clear_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
);
2056 filemap_invalidate_unlock(mapping
);
2062 fuse_release_nowrite(inode
);
2064 clear_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
);
2067 filemap_invalidate_unlock(mapping
);
2071 static int fuse_setattr(struct mnt_idmap
*idmap
, struct dentry
*entry
,
2074 struct inode
*inode
= d_inode(entry
);
2075 struct fuse_conn
*fc
= get_fuse_conn(inode
);
2076 struct file
*file
= (attr
->ia_valid
& ATTR_FILE
) ? attr
->ia_file
: NULL
;
2079 if (fuse_is_bad(inode
))
2082 if (!fuse_allow_current_process(get_fuse_conn(inode
)))
2085 if (attr
->ia_valid
& (ATTR_KILL_SUID
| ATTR_KILL_SGID
)) {
2086 attr
->ia_valid
&= ~(ATTR_KILL_SUID
| ATTR_KILL_SGID
|
2090 * The only sane way to reliably kill suid/sgid is to do it in
2091 * the userspace filesystem
2093 * This should be done on write(), truncate() and chown().
2095 if (!fc
->handle_killpriv
&& !fc
->handle_killpriv_v2
) {
2097 * ia_mode calculation may have used stale i_mode.
2098 * Refresh and recalculate.
2100 ret
= fuse_do_getattr(idmap
, inode
, NULL
, file
);
2104 attr
->ia_mode
= inode
->i_mode
;
2105 if (inode
->i_mode
& S_ISUID
) {
2106 attr
->ia_valid
|= ATTR_MODE
;
2107 attr
->ia_mode
&= ~S_ISUID
;
2109 if ((inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
2110 attr
->ia_valid
|= ATTR_MODE
;
2111 attr
->ia_mode
&= ~S_ISGID
;
2115 if (!attr
->ia_valid
)
2118 ret
= fuse_do_setattr(idmap
, entry
, attr
, file
);
2121 * If filesystem supports acls it may have updated acl xattrs in
2122 * the filesystem, so forget cached acls for the inode.
2125 forget_all_cached_acls(inode
);
2127 /* Directory mode changed, may need to revalidate access */
2128 if (d_is_dir(entry
) && (attr
->ia_valid
& ATTR_MODE
))
2129 fuse_invalidate_entry_cache(entry
);
2134 static int fuse_getattr(struct mnt_idmap
*idmap
,
2135 const struct path
*path
, struct kstat
*stat
,
2136 u32 request_mask
, unsigned int flags
)
2138 struct inode
*inode
= d_inode(path
->dentry
);
2139 struct fuse_conn
*fc
= get_fuse_conn(inode
);
2141 if (fuse_is_bad(inode
))
2144 if (!fuse_allow_current_process(fc
)) {
2145 if (!request_mask
) {
2147 * If user explicitly requested *nothing* then don't
2148 * error out, but return st_dev only.
2150 stat
->result_mask
= 0;
2151 stat
->dev
= inode
->i_sb
->s_dev
;
2157 return fuse_update_get_attr(idmap
, inode
, NULL
, stat
, request_mask
, flags
);
2160 static const struct inode_operations fuse_dir_inode_operations
= {
2161 .lookup
= fuse_lookup
,
2162 .mkdir
= fuse_mkdir
,
2163 .symlink
= fuse_symlink
,
2164 .unlink
= fuse_unlink
,
2165 .rmdir
= fuse_rmdir
,
2166 .rename
= fuse_rename2
,
2168 .setattr
= fuse_setattr
,
2169 .create
= fuse_create
,
2170 .atomic_open
= fuse_atomic_open
,
2171 .tmpfile
= fuse_tmpfile
,
2172 .mknod
= fuse_mknod
,
2173 .permission
= fuse_permission
,
2174 .getattr
= fuse_getattr
,
2175 .listxattr
= fuse_listxattr
,
2176 .get_inode_acl
= fuse_get_inode_acl
,
2177 .get_acl
= fuse_get_acl
,
2178 .set_acl
= fuse_set_acl
,
2179 .fileattr_get
= fuse_fileattr_get
,
2180 .fileattr_set
= fuse_fileattr_set
,
2183 static const struct file_operations fuse_dir_operations
= {
2184 .llseek
= generic_file_llseek
,
2185 .read
= generic_read_dir
,
2186 .iterate_shared
= fuse_readdir
,
2187 .open
= fuse_dir_open
,
2188 .release
= fuse_dir_release
,
2189 .fsync
= fuse_dir_fsync
,
2190 .unlocked_ioctl
= fuse_dir_ioctl
,
2191 .compat_ioctl
= fuse_dir_compat_ioctl
,
2194 static const struct inode_operations fuse_common_inode_operations
= {
2195 .setattr
= fuse_setattr
,
2196 .permission
= fuse_permission
,
2197 .getattr
= fuse_getattr
,
2198 .listxattr
= fuse_listxattr
,
2199 .get_inode_acl
= fuse_get_inode_acl
,
2200 .get_acl
= fuse_get_acl
,
2201 .set_acl
= fuse_set_acl
,
2202 .fileattr_get
= fuse_fileattr_get
,
2203 .fileattr_set
= fuse_fileattr_set
,
2206 static const struct inode_operations fuse_symlink_inode_operations
= {
2207 .setattr
= fuse_setattr
,
2208 .get_link
= fuse_get_link
,
2209 .getattr
= fuse_getattr
,
2210 .listxattr
= fuse_listxattr
,
2213 void fuse_init_common(struct inode
*inode
)
2215 inode
->i_op
= &fuse_common_inode_operations
;
2218 void fuse_init_dir(struct inode
*inode
)
2220 struct fuse_inode
*fi
= get_fuse_inode(inode
);
2222 inode
->i_op
= &fuse_dir_inode_operations
;
2223 inode
->i_fop
= &fuse_dir_operations
;
2225 spin_lock_init(&fi
->rdc
.lock
);
2226 fi
->rdc
.cached
= false;
2229 fi
->rdc
.version
= 0;
2232 static int fuse_symlink_read_folio(struct file
*null
, struct folio
*folio
)
2234 int err
= fuse_readlink_page(folio
->mapping
->host
, &folio
->page
);
2237 folio_mark_uptodate(folio
);
2239 folio_unlock(folio
);
2244 static const struct address_space_operations fuse_symlink_aops
= {
2245 .read_folio
= fuse_symlink_read_folio
,
2248 void fuse_init_symlink(struct inode
*inode
)
2250 inode
->i_op
= &fuse_symlink_inode_operations
;
2251 inode
->i_data
.a_ops
= &fuse_symlink_aops
;
2252 inode_nohighmem(inode
);