4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <linux/posix_acl.h>
36 #include <asm/uaccess.h>
41 /* [Feb-1997 T. Schoebel-Theuer]
42 * Fundamental changes in the pathname lookup mechanisms (namei)
43 * were necessary because of omirr. The reason is that omirr needs
44 * to know the _real_ pathname, not the user-supplied one, in case
45 * of symlinks (and also when transname replacements occur).
47 * The new code replaces the old recursive symlink resolution with
48 * an iterative one (in case of non-nested symlink chains). It does
49 * this with calls to <fs>_follow_link().
50 * As a side effect, dir_namei(), _namei() and follow_link() are now
51 * replaced with a single function lookup_dentry() that can handle all
52 * the special cases of the former code.
54 * With the new dcache, the pathname is stored at each inode, at least as
55 * long as the refcount of the inode is positive. As a side effect, the
56 * size of the dcache depends on the inode cache and thus is dynamic.
58 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
59 * resolution to correspond with current state of the code.
61 * Note that the symlink resolution is not *completely* iterative.
62 * There is still a significant amount of tail- and mid- recursion in
63 * the algorithm. Also, note that <fs>_readlink() is not used in
64 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
65 * may return different results than <fs>_follow_link(). Many virtual
66 * filesystems (including /proc) exhibit this behavior.
69 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
70 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
71 * and the name already exists in form of a symlink, try to create the new
72 * name indicated by the symlink. The old code always complained that the
73 * name already exists, due to not following the symlink even if its target
74 * is nonexistent. The new semantics affects also mknod() and link() when
75 * the name is a symlink pointing to a non-existent name.
77 * I don't know which semantics is the right one, since I have no access
78 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
79 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
80 * "old" one. Personally, I think the new semantics is much more logical.
81 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
82 * file does succeed in both HP-UX and SunOs, but not in Solaris
83 * and in the old Linux semantics.
86 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
87 * semantics. See the comments in "open_namei" and "do_link" below.
89 * [10-Sep-98 Alan Modra] Another symlink change.
92 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
93 * inside the path - always follow.
94 * in the last component in creation/removal/renaming - never follow.
95 * if LOOKUP_FOLLOW passed - follow.
96 * if the pathname has trailing slashes - follow.
97 * otherwise - don't follow.
98 * (applied in that order).
100 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
101 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
102 * During the 2.4 we need to fix the userland stuff depending on it -
103 * hopefully we will be able to get rid of that wart in 2.5. So far only
104 * XEmacs seems to be relying on it...
107 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
108 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
109 * any extra contention...
112 /* In order to reduce some races, while at the same time doing additional
113 * checking and hopefully speeding things up, we copy filenames to the
114 * kernel data space before using them..
116 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
117 * PATH_MAX includes the nul terminator --RR.
119 static int do_getname(const char __user
*filename
, char *page
)
122 unsigned long len
= PATH_MAX
;
124 if (!segment_eq(get_fs(), KERNEL_DS
)) {
125 if ((unsigned long) filename
>= TASK_SIZE
)
127 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
128 len
= TASK_SIZE
- (unsigned long) filename
;
131 retval
= strncpy_from_user(page
, filename
, len
);
135 return -ENAMETOOLONG
;
141 static char *getname_flags(const char __user
*filename
, int flags
, int *empty
)
143 char *result
= __getname();
147 return ERR_PTR(-ENOMEM
);
149 retval
= do_getname(filename
, result
);
151 if (retval
== -ENOENT
&& empty
)
153 if (retval
!= -ENOENT
|| !(flags
& LOOKUP_EMPTY
)) {
155 return ERR_PTR(retval
);
158 audit_getname(result
);
162 char *getname(const char __user
* filename
)
164 return getname_flags(filename
, 0, NULL
);
167 #ifdef CONFIG_AUDITSYSCALL
168 void putname(const char *name
)
170 if (unlikely(!audit_dummy_context()))
175 EXPORT_SYMBOL(putname
);
178 static int check_acl(struct inode
*inode
, int mask
)
180 #ifdef CONFIG_FS_POSIX_ACL
181 struct posix_acl
*acl
;
183 if (mask
& MAY_NOT_BLOCK
) {
184 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
187 /* no ->get_acl() calls in RCU mode... */
188 if (acl
== ACL_NOT_CACHED
)
190 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
193 acl
= get_cached_acl(inode
, ACL_TYPE_ACCESS
);
196 * A filesystem can force a ACL callback by just never filling the
197 * ACL cache. But normally you'd fill the cache either at inode
198 * instantiation time, or on the first ->get_acl call.
200 * If the filesystem doesn't have a get_acl() function at all, we'll
201 * just create the negative cache entry.
203 if (acl
== ACL_NOT_CACHED
) {
204 if (inode
->i_op
->get_acl
) {
205 acl
= inode
->i_op
->get_acl(inode
, ACL_TYPE_ACCESS
);
209 set_cached_acl(inode
, ACL_TYPE_ACCESS
, NULL
);
215 int error
= posix_acl_permission(inode
, acl
, mask
);
216 posix_acl_release(acl
);
225 * This does the basic permission checking
227 static int acl_permission_check(struct inode
*inode
, int mask
)
229 unsigned int mode
= inode
->i_mode
;
231 if (current_user_ns() != inode_userns(inode
))
234 if (likely(current_fsuid() == inode
->i_uid
))
237 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
238 int error
= check_acl(inode
, mask
);
239 if (error
!= -EAGAIN
)
243 if (in_group_p(inode
->i_gid
))
249 * If the DACs are ok we don't need any capability check.
251 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
257 * generic_permission - check for access rights on a Posix-like filesystem
258 * @inode: inode to check access rights for
259 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
261 * Used to check for read/write/execute permissions on a file.
262 * We use "fsuid" for this, letting us set arbitrary permissions
263 * for filesystem access without changing the "normal" uids which
264 * are used for other things.
266 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
267 * request cannot be satisfied (eg. requires blocking or too much complexity).
268 * It would then be called again in ref-walk mode.
270 int generic_permission(struct inode
*inode
, int mask
)
275 * Do the basic permission checks.
277 ret
= acl_permission_check(inode
, mask
);
281 if (S_ISDIR(inode
->i_mode
)) {
282 /* DACs are overridable for directories */
283 if (ns_capable(inode_userns(inode
), CAP_DAC_OVERRIDE
))
285 if (!(mask
& MAY_WRITE
))
286 if (ns_capable(inode_userns(inode
), CAP_DAC_READ_SEARCH
))
291 * Read/write DACs are always overridable.
292 * Executable DACs are overridable when there is
293 * at least one exec bit set.
295 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
296 if (ns_capable(inode_userns(inode
), CAP_DAC_OVERRIDE
))
300 * Searching includes executable on directories, else just read.
302 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
303 if (mask
== MAY_READ
)
304 if (ns_capable(inode_userns(inode
), CAP_DAC_READ_SEARCH
))
311 * We _really_ want to just do "generic_permission()" without
312 * even looking at the inode->i_op values. So we keep a cache
313 * flag in inode->i_opflags, that says "this has not special
314 * permission function, use the fast case".
316 static inline int do_inode_permission(struct inode
*inode
, int mask
)
318 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
319 if (likely(inode
->i_op
->permission
))
320 return inode
->i_op
->permission(inode
, mask
);
322 /* This gets set once for the inode lifetime */
323 spin_lock(&inode
->i_lock
);
324 inode
->i_opflags
|= IOP_FASTPERM
;
325 spin_unlock(&inode
->i_lock
);
327 return generic_permission(inode
, mask
);
331 * inode_permission - check for access rights to a given inode
332 * @inode: inode to check permission on
333 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
335 * Used to check for read/write/execute permissions on an inode.
336 * We use "fsuid" for this, letting us set arbitrary permissions
337 * for filesystem access without changing the "normal" uids which
338 * are used for other things.
340 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
342 int inode_permission(struct inode
*inode
, int mask
)
346 if (unlikely(mask
& MAY_WRITE
)) {
347 umode_t mode
= inode
->i_mode
;
350 * Nobody gets write access to a read-only fs.
352 if (IS_RDONLY(inode
) &&
353 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
357 * Nobody gets write access to an immutable file.
359 if (IS_IMMUTABLE(inode
))
363 retval
= do_inode_permission(inode
, mask
);
367 retval
= devcgroup_inode_permission(inode
, mask
);
371 return security_inode_permission(inode
, mask
);
375 * path_get - get a reference to a path
376 * @path: path to get the reference to
378 * Given a path increment the reference count to the dentry and the vfsmount.
380 void path_get(struct path
*path
)
385 EXPORT_SYMBOL(path_get
);
388 * path_put - put a reference to a path
389 * @path: path to put the reference to
391 * Given a path decrement the reference count to the dentry and the vfsmount.
393 void path_put(struct path
*path
)
398 EXPORT_SYMBOL(path_put
);
401 * Path walking has 2 modes, rcu-walk and ref-walk (see
402 * Documentation/filesystems/path-lookup.txt). In situations when we can't
403 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
404 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
405 * mode. Refcounts are grabbed at the last known good point before rcu-walk
406 * got stuck, so ref-walk may continue from there. If this is not successful
407 * (eg. a seqcount has changed), then failure is returned and it's up to caller
408 * to restart the path walk from the beginning in ref-walk mode.
412 * unlazy_walk - try to switch to ref-walk mode.
413 * @nd: nameidata pathwalk data
414 * @dentry: child of nd->path.dentry or NULL
415 * Returns: 0 on success, -ECHILD on failure
417 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
418 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
419 * @nd or NULL. Must be called from rcu-walk context.
421 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
423 struct fs_struct
*fs
= current
->fs
;
424 struct dentry
*parent
= nd
->path
.dentry
;
427 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
428 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
430 spin_lock(&fs
->lock
);
431 if (nd
->root
.mnt
!= fs
->root
.mnt
||
432 nd
->root
.dentry
!= fs
->root
.dentry
)
435 spin_lock(&parent
->d_lock
);
437 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
439 BUG_ON(nd
->inode
!= parent
->d_inode
);
441 if (dentry
->d_parent
!= parent
)
443 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
444 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
447 * If the sequence check on the child dentry passed, then
448 * the child has not been removed from its parent. This
449 * means the parent dentry must be valid and able to take
450 * a reference at this point.
452 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
453 BUG_ON(!parent
->d_count
);
455 spin_unlock(&dentry
->d_lock
);
457 spin_unlock(&parent
->d_lock
);
460 spin_unlock(&fs
->lock
);
462 mntget(nd
->path
.mnt
);
465 br_read_unlock(vfsmount_lock
);
466 nd
->flags
&= ~LOOKUP_RCU
;
470 spin_unlock(&dentry
->d_lock
);
472 spin_unlock(&parent
->d_lock
);
475 spin_unlock(&fs
->lock
);
480 * release_open_intent - free up open intent resources
481 * @nd: pointer to nameidata
483 void release_open_intent(struct nameidata
*nd
)
485 struct file
*file
= nd
->intent
.open
.file
;
487 if (file
&& !IS_ERR(file
)) {
488 if (file
->f_path
.dentry
== NULL
)
495 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
497 return dentry
->d_op
->d_revalidate(dentry
, nd
);
501 * complete_walk - successful completion of path walk
502 * @nd: pointer nameidata
504 * If we had been in RCU mode, drop out of it and legitimize nd->path.
505 * Revalidate the final result, unless we'd already done that during
506 * the path walk or the filesystem doesn't ask for it. Return 0 on
507 * success, -error on failure. In case of failure caller does not
508 * need to drop nd->path.
510 static int complete_walk(struct nameidata
*nd
)
512 struct dentry
*dentry
= nd
->path
.dentry
;
515 if (nd
->flags
& LOOKUP_RCU
) {
516 nd
->flags
&= ~LOOKUP_RCU
;
517 if (!(nd
->flags
& LOOKUP_ROOT
))
519 spin_lock(&dentry
->d_lock
);
520 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
521 spin_unlock(&dentry
->d_lock
);
523 br_read_unlock(vfsmount_lock
);
526 BUG_ON(nd
->inode
!= dentry
->d_inode
);
527 spin_unlock(&dentry
->d_lock
);
528 mntget(nd
->path
.mnt
);
530 br_read_unlock(vfsmount_lock
);
533 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
536 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
539 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
542 /* Note: we do not d_invalidate() */
543 status
= d_revalidate(dentry
, nd
);
554 static __always_inline
void set_root(struct nameidata
*nd
)
557 get_fs_root(current
->fs
, &nd
->root
);
560 static int link_path_walk(const char *, struct nameidata
*);
562 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
565 struct fs_struct
*fs
= current
->fs
;
569 seq
= read_seqcount_begin(&fs
->seq
);
571 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
572 } while (read_seqcount_retry(&fs
->seq
, seq
));
576 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
588 nd
->flags
|= LOOKUP_JUMPED
;
590 nd
->inode
= nd
->path
.dentry
->d_inode
;
592 ret
= link_path_walk(link
, nd
);
596 return PTR_ERR(link
);
599 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
602 if (path
->mnt
!= nd
->path
.mnt
)
606 static inline void path_to_nameidata(const struct path
*path
,
607 struct nameidata
*nd
)
609 if (!(nd
->flags
& LOOKUP_RCU
)) {
610 dput(nd
->path
.dentry
);
611 if (nd
->path
.mnt
!= path
->mnt
)
612 mntput(nd
->path
.mnt
);
614 nd
->path
.mnt
= path
->mnt
;
615 nd
->path
.dentry
= path
->dentry
;
618 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
620 struct inode
*inode
= link
->dentry
->d_inode
;
621 if (!IS_ERR(cookie
) && inode
->i_op
->put_link
)
622 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
626 static __always_inline
int
627 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
630 struct dentry
*dentry
= link
->dentry
;
632 BUG_ON(nd
->flags
& LOOKUP_RCU
);
634 if (link
->mnt
== nd
->path
.mnt
)
637 if (unlikely(current
->total_link_count
>= 40)) {
638 *p
= ERR_PTR(-ELOOP
); /* no ->put_link(), please */
643 current
->total_link_count
++;
646 nd_set_link(nd
, NULL
);
648 error
= security_inode_follow_link(link
->dentry
, nd
);
650 *p
= ERR_PTR(error
); /* no ->put_link(), please */
655 nd
->last_type
= LAST_BIND
;
656 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
659 char *s
= nd_get_link(nd
);
662 error
= __vfs_follow_link(nd
, s
);
663 else if (nd
->last_type
== LAST_BIND
) {
664 nd
->flags
|= LOOKUP_JUMPED
;
665 nd
->inode
= nd
->path
.dentry
->d_inode
;
666 if (nd
->inode
->i_op
->follow_link
) {
667 /* stepped on a _really_ weird one */
676 static int follow_up_rcu(struct path
*path
)
678 struct mount
*mnt
= real_mount(path
->mnt
);
679 struct mount
*parent
;
680 struct dentry
*mountpoint
;
682 parent
= mnt
->mnt_parent
;
683 if (&parent
->mnt
== path
->mnt
)
685 mountpoint
= mnt
->mnt_mountpoint
;
686 path
->dentry
= mountpoint
;
687 path
->mnt
= &parent
->mnt
;
691 int follow_up(struct path
*path
)
693 struct mount
*mnt
= real_mount(path
->mnt
);
694 struct mount
*parent
;
695 struct dentry
*mountpoint
;
697 br_read_lock(vfsmount_lock
);
698 parent
= mnt
->mnt_parent
;
699 if (&parent
->mnt
== path
->mnt
) {
700 br_read_unlock(vfsmount_lock
);
703 mntget(&parent
->mnt
);
704 mountpoint
= dget(mnt
->mnt_mountpoint
);
705 br_read_unlock(vfsmount_lock
);
707 path
->dentry
= mountpoint
;
709 path
->mnt
= &parent
->mnt
;
714 * Perform an automount
715 * - return -EISDIR to tell follow_managed() to stop and return the path we
718 static int follow_automount(struct path
*path
, unsigned flags
,
721 struct vfsmount
*mnt
;
724 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
727 /* We don't want to mount if someone's just doing a stat -
728 * unless they're stat'ing a directory and appended a '/' to
731 * We do, however, want to mount if someone wants to open or
732 * create a file of any type under the mountpoint, wants to
733 * traverse through the mountpoint or wants to open the
734 * mounted directory. Also, autofs may mark negative dentries
735 * as being automount points. These will need the attentions
736 * of the daemon to instantiate them before they can be used.
738 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
739 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
740 path
->dentry
->d_inode
)
743 current
->total_link_count
++;
744 if (current
->total_link_count
>= 40)
747 mnt
= path
->dentry
->d_op
->d_automount(path
);
750 * The filesystem is allowed to return -EISDIR here to indicate
751 * it doesn't want to automount. For instance, autofs would do
752 * this so that its userspace daemon can mount on this dentry.
754 * However, we can only permit this if it's a terminal point in
755 * the path being looked up; if it wasn't then the remainder of
756 * the path is inaccessible and we should say so.
758 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
763 if (!mnt
) /* mount collision */
767 /* lock_mount() may release path->mnt on error */
771 err
= finish_automount(mnt
, path
);
775 /* Someone else made a mount here whilst we were busy */
780 path
->dentry
= dget(mnt
->mnt_root
);
789 * Handle a dentry that is managed in some way.
790 * - Flagged for transit management (autofs)
791 * - Flagged as mountpoint
792 * - Flagged as automount point
794 * This may only be called in refwalk mode.
796 * Serialization is taken care of in namespace.c
798 static int follow_managed(struct path
*path
, unsigned flags
)
800 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
802 bool need_mntput
= false;
805 /* Given that we're not holding a lock here, we retain the value in a
806 * local variable for each dentry as we look at it so that we don't see
807 * the components of that value change under us */
808 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
809 managed
&= DCACHE_MANAGED_DENTRY
,
810 unlikely(managed
!= 0)) {
811 /* Allow the filesystem to manage the transit without i_mutex
813 if (managed
& DCACHE_MANAGE_TRANSIT
) {
814 BUG_ON(!path
->dentry
->d_op
);
815 BUG_ON(!path
->dentry
->d_op
->d_manage
);
816 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
821 /* Transit to a mounted filesystem. */
822 if (managed
& DCACHE_MOUNTED
) {
823 struct vfsmount
*mounted
= lookup_mnt(path
);
829 path
->dentry
= dget(mounted
->mnt_root
);
834 /* Something is mounted on this dentry in another
835 * namespace and/or whatever was mounted there in this
836 * namespace got unmounted before we managed to get the
840 /* Handle an automount point */
841 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
842 ret
= follow_automount(path
, flags
, &need_mntput
);
848 /* We didn't change the current path point */
852 if (need_mntput
&& path
->mnt
== mnt
)
856 return ret
< 0 ? ret
: need_mntput
;
859 int follow_down_one(struct path
*path
)
861 struct vfsmount
*mounted
;
863 mounted
= lookup_mnt(path
);
868 path
->dentry
= dget(mounted
->mnt_root
);
874 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
876 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
877 dentry
->d_op
->d_manage(dentry
, true) < 0);
881 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
882 * we meet a managed dentry that would need blocking.
884 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
885 struct inode
**inode
)
888 struct mount
*mounted
;
890 * Don't forget we might have a non-mountpoint managed dentry
891 * that wants to block transit.
893 if (unlikely(managed_dentry_might_block(path
->dentry
)))
896 if (!d_mountpoint(path
->dentry
))
899 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
902 path
->mnt
= &mounted
->mnt
;
903 path
->dentry
= mounted
->mnt
.mnt_root
;
904 nd
->flags
|= LOOKUP_JUMPED
;
905 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
907 * Update the inode too. We don't need to re-check the
908 * dentry sequence number here after this d_inode read,
909 * because a mount-point is always pinned.
911 *inode
= path
->dentry
->d_inode
;
916 static void follow_mount_rcu(struct nameidata
*nd
)
918 while (d_mountpoint(nd
->path
.dentry
)) {
919 struct mount
*mounted
;
920 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
923 nd
->path
.mnt
= &mounted
->mnt
;
924 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
925 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
929 static int follow_dotdot_rcu(struct nameidata
*nd
)
934 if (nd
->path
.dentry
== nd
->root
.dentry
&&
935 nd
->path
.mnt
== nd
->root
.mnt
) {
938 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
939 struct dentry
*old
= nd
->path
.dentry
;
940 struct dentry
*parent
= old
->d_parent
;
943 seq
= read_seqcount_begin(&parent
->d_seq
);
944 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
946 nd
->path
.dentry
= parent
;
950 if (!follow_up_rcu(&nd
->path
))
952 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
954 follow_mount_rcu(nd
);
955 nd
->inode
= nd
->path
.dentry
->d_inode
;
959 nd
->flags
&= ~LOOKUP_RCU
;
960 if (!(nd
->flags
& LOOKUP_ROOT
))
963 br_read_unlock(vfsmount_lock
);
968 * Follow down to the covering mount currently visible to userspace. At each
969 * point, the filesystem owning that dentry may be queried as to whether the
970 * caller is permitted to proceed or not.
972 int follow_down(struct path
*path
)
977 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
978 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
979 /* Allow the filesystem to manage the transit without i_mutex
982 * We indicate to the filesystem if someone is trying to mount
983 * something here. This gives autofs the chance to deny anyone
984 * other than its daemon the right to mount on its
987 * The filesystem may sleep at this point.
989 if (managed
& DCACHE_MANAGE_TRANSIT
) {
990 BUG_ON(!path
->dentry
->d_op
);
991 BUG_ON(!path
->dentry
->d_op
->d_manage
);
992 ret
= path
->dentry
->d_op
->d_manage(
993 path
->dentry
, false);
995 return ret
== -EISDIR
? 0 : ret
;
998 /* Transit to a mounted filesystem. */
999 if (managed
& DCACHE_MOUNTED
) {
1000 struct vfsmount
*mounted
= lookup_mnt(path
);
1005 path
->mnt
= mounted
;
1006 path
->dentry
= dget(mounted
->mnt_root
);
1010 /* Don't handle automount points here */
1017 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1019 static void follow_mount(struct path
*path
)
1021 while (d_mountpoint(path
->dentry
)) {
1022 struct vfsmount
*mounted
= lookup_mnt(path
);
1027 path
->mnt
= mounted
;
1028 path
->dentry
= dget(mounted
->mnt_root
);
1032 static void follow_dotdot(struct nameidata
*nd
)
1037 struct dentry
*old
= nd
->path
.dentry
;
1039 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1040 nd
->path
.mnt
== nd
->root
.mnt
) {
1043 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1044 /* rare case of legitimate dget_parent()... */
1045 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1049 if (!follow_up(&nd
->path
))
1052 follow_mount(&nd
->path
);
1053 nd
->inode
= nd
->path
.dentry
->d_inode
;
1057 * Allocate a dentry with name and parent, and perform a parent
1058 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1059 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1060 * have verified that no child exists while under i_mutex.
1062 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1063 struct qstr
*name
, struct nameidata
*nd
)
1065 struct inode
*inode
= parent
->d_inode
;
1066 struct dentry
*dentry
;
1069 /* Don't create child dentry for a dead directory. */
1070 if (unlikely(IS_DEADDIR(inode
)))
1071 return ERR_PTR(-ENOENT
);
1073 dentry
= d_alloc(parent
, name
);
1074 if (unlikely(!dentry
))
1075 return ERR_PTR(-ENOMEM
);
1077 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1078 if (unlikely(old
)) {
1086 * We already have a dentry, but require a lookup to be performed on the parent
1087 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1088 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1089 * child exists while under i_mutex.
1091 static struct dentry
*d_inode_lookup(struct dentry
*parent
, struct dentry
*dentry
,
1092 struct nameidata
*nd
)
1094 struct inode
*inode
= parent
->d_inode
;
1097 /* Don't create child dentry for a dead directory. */
1098 if (unlikely(IS_DEADDIR(inode
))) {
1100 return ERR_PTR(-ENOENT
);
1103 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1104 if (unlikely(old
)) {
1111 static struct dentry
*__lookup_hash(struct qstr
*name
,
1112 struct dentry
*base
, struct nameidata
*nd
)
1114 struct dentry
*dentry
;
1117 * Don't bother with __d_lookup: callers are for creat as
1118 * well as unlink, so a lot of the time it would cost
1121 dentry
= d_lookup(base
, name
);
1123 if (dentry
&& d_need_lookup(dentry
)) {
1125 * __lookup_hash is called with the parent dir's i_mutex already
1126 * held, so we are good to go here.
1128 return d_inode_lookup(base
, dentry
, nd
);
1131 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1132 int status
= d_revalidate(dentry
, nd
);
1133 if (unlikely(status
<= 0)) {
1135 * The dentry failed validation.
1136 * If d_revalidate returned 0 attempt to invalidate
1137 * the dentry otherwise d_revalidate is asking us
1138 * to return a fail status.
1142 return ERR_PTR(status
);
1143 } else if (!d_invalidate(dentry
)) {
1151 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1157 * It's more convoluted than I'd like it to be, but... it's still fairly
1158 * small and for now I'd prefer to have fast path as straight as possible.
1159 * It _is_ time-critical.
1161 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1162 struct path
*path
, struct inode
**inode
)
1164 struct vfsmount
*mnt
= nd
->path
.mnt
;
1165 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1171 * Rename seqlock is not required here because in the off chance
1172 * of a false negative due to a concurrent rename, we're going to
1173 * do the non-racy lookup, below.
1175 if (nd
->flags
& LOOKUP_RCU
) {
1178 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1182 /* Memory barrier in read_seqcount_begin of child is enough */
1183 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1187 if (unlikely(d_need_lookup(dentry
)))
1189 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1190 status
= d_revalidate(dentry
, nd
);
1191 if (unlikely(status
<= 0)) {
1192 if (status
!= -ECHILD
)
1198 path
->dentry
= dentry
;
1199 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1201 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1205 if (unlazy_walk(nd
, dentry
))
1208 dentry
= __d_lookup(parent
, name
);
1211 if (dentry
&& unlikely(d_need_lookup(dentry
))) {
1216 if (unlikely(!dentry
)) {
1217 struct inode
*dir
= parent
->d_inode
;
1218 BUG_ON(nd
->inode
!= dir
);
1220 mutex_lock(&dir
->i_mutex
);
1221 dentry
= __lookup_hash(name
, parent
, nd
);
1222 mutex_unlock(&dir
->i_mutex
);
1224 return PTR_ERR(dentry
);
1227 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1228 status
= d_revalidate(dentry
, nd
);
1229 if (unlikely(status
<= 0)) {
1234 if (!d_invalidate(dentry
)) {
1242 path
->dentry
= dentry
;
1243 err
= follow_managed(path
, nd
->flags
);
1244 if (unlikely(err
< 0)) {
1245 path_put_conditional(path
, nd
);
1249 nd
->flags
|= LOOKUP_JUMPED
;
1250 *inode
= path
->dentry
->d_inode
;
1254 static inline int may_lookup(struct nameidata
*nd
)
1256 if (nd
->flags
& LOOKUP_RCU
) {
1257 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1260 if (unlazy_walk(nd
, NULL
))
1263 return inode_permission(nd
->inode
, MAY_EXEC
);
1266 static inline int handle_dots(struct nameidata
*nd
, int type
)
1268 if (type
== LAST_DOTDOT
) {
1269 if (nd
->flags
& LOOKUP_RCU
) {
1270 if (follow_dotdot_rcu(nd
))
1278 static void terminate_walk(struct nameidata
*nd
)
1280 if (!(nd
->flags
& LOOKUP_RCU
)) {
1281 path_put(&nd
->path
);
1283 nd
->flags
&= ~LOOKUP_RCU
;
1284 if (!(nd
->flags
& LOOKUP_ROOT
))
1285 nd
->root
.mnt
= NULL
;
1287 br_read_unlock(vfsmount_lock
);
1292 * Do we need to follow links? We _really_ want to be able
1293 * to do this check without having to look at inode->i_op,
1294 * so we keep a cache of "no, this doesn't need follow_link"
1295 * for the common case.
1297 static inline int should_follow_link(struct inode
*inode
, int follow
)
1299 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1300 if (likely(inode
->i_op
->follow_link
))
1303 /* This gets set once for the inode lifetime */
1304 spin_lock(&inode
->i_lock
);
1305 inode
->i_opflags
|= IOP_NOFOLLOW
;
1306 spin_unlock(&inode
->i_lock
);
1311 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1312 struct qstr
*name
, int type
, int follow
)
1314 struct inode
*inode
;
1317 * "." and ".." are special - ".." especially so because it has
1318 * to be able to know about the current root directory and
1319 * parent relationships.
1321 if (unlikely(type
!= LAST_NORM
))
1322 return handle_dots(nd
, type
);
1323 err
= do_lookup(nd
, name
, path
, &inode
);
1324 if (unlikely(err
)) {
1329 path_to_nameidata(path
, nd
);
1333 if (should_follow_link(inode
, follow
)) {
1334 if (nd
->flags
& LOOKUP_RCU
) {
1335 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1340 BUG_ON(inode
!= path
->dentry
->d_inode
);
1343 path_to_nameidata(path
, nd
);
1349 * This limits recursive symlink follows to 8, while
1350 * limiting consecutive symlinks to 40.
1352 * Without that kind of total limit, nasty chains of consecutive
1353 * symlinks can cause almost arbitrarily long lookups.
1355 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1359 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1360 path_put_conditional(path
, nd
);
1361 path_put(&nd
->path
);
1364 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1367 current
->link_count
++;
1370 struct path link
= *path
;
1373 res
= follow_link(&link
, nd
, &cookie
);
1375 res
= walk_component(nd
, path
, &nd
->last
,
1376 nd
->last_type
, LOOKUP_FOLLOW
);
1377 put_link(nd
, &link
, cookie
);
1380 current
->link_count
--;
1386 * We really don't want to look at inode->i_op->lookup
1387 * when we don't have to. So we keep a cache bit in
1388 * the inode ->i_opflags field that says "yes, we can
1389 * do lookup on this inode".
1391 static inline int can_lookup(struct inode
*inode
)
1393 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1395 if (likely(!inode
->i_op
->lookup
))
1398 /* We do this once for the lifetime of the inode */
1399 spin_lock(&inode
->i_lock
);
1400 inode
->i_opflags
|= IOP_LOOKUP
;
1401 spin_unlock(&inode
->i_lock
);
1406 * We can do the critical dentry name comparison and hashing
1407 * operations one word at a time, but we are limited to:
1409 * - Architectures with fast unaligned word accesses. We could
1410 * do a "get_unaligned()" if this helps and is sufficiently
1413 * - Little-endian machines (so that we can generate the mask
1414 * of low bytes efficiently). Again, we *could* do a byte
1415 * swapping load on big-endian architectures if that is not
1416 * expensive enough to make the optimization worthless.
1418 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1419 * do not trap on the (extremely unlikely) case of a page
1420 * crossing operation.
1422 * - Furthermore, we need an efficient 64-bit compile for the
1423 * 64-bit case in order to generate the "number of bytes in
1424 * the final mask". Again, that could be replaced with a
1425 * efficient population count instruction or similar.
1427 #ifdef CONFIG_DCACHE_WORD_ACCESS
1432 * Jan Achrenius on G+: microoptimized version of
1433 * the simpler "(mask & ONEBYTES) * ONEBYTES >> 56"
1434 * that works for the bytemasks without having to
1437 static inline long count_masked_bytes(unsigned long mask
)
1439 return mask
*0x0001020304050608ul
>> 56;
1442 static inline unsigned int fold_hash(unsigned long hash
)
1444 hash
+= hash
>> (8*sizeof(int));
1448 #else /* 32-bit case */
1450 /* Carl Chatfield / Jan Achrenius G+ version for 32-bit */
1451 static inline long count_masked_bytes(long mask
)
1453 /* (000000 0000ff 00ffff ffffff) -> ( 1 1 2 3 ) */
1454 long a
= (0x0ff0001+mask
) >> 23;
1455 /* Fix the 1 for 00 case */
1459 #define fold_hash(x) (x)
1463 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1465 unsigned long a
, mask
;
1466 unsigned long hash
= 0;
1469 a
= *(unsigned long *)name
;
1470 if (len
< sizeof(unsigned long))
1474 name
+= sizeof(unsigned long);
1475 len
-= sizeof(unsigned long);
1479 mask
= ~(~0ul << len
*8);
1482 return fold_hash(hash
);
1484 EXPORT_SYMBOL(full_name_hash
);
1486 #define REPEAT_BYTE(x) ((~0ul / 0xff) * (x))
1487 #define ONEBYTES REPEAT_BYTE(0x01)
1488 #define SLASHBYTES REPEAT_BYTE('/')
1489 #define HIGHBITS REPEAT_BYTE(0x80)
1491 /* Return the high bit set in the first byte that is a zero */
1492 static inline unsigned long has_zero(unsigned long a
)
1494 return ((a
- ONEBYTES
) & ~a
) & HIGHBITS
;
1498 * Calculate the length and hash of the path component, and
1499 * return the length of the component;
1501 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1503 unsigned long a
, mask
, hash
, len
;
1506 len
= -sizeof(unsigned long);
1508 hash
= (hash
+ a
) * 9;
1509 len
+= sizeof(unsigned long);
1510 a
= *(unsigned long *)(name
+len
);
1511 /* Do we have any NUL or '/' bytes in this word? */
1512 mask
= has_zero(a
) | has_zero(a
^ SLASHBYTES
);
1515 /* The mask *below* the first high bit set */
1516 mask
= (mask
- 1) & ~mask
;
1519 *hashp
= fold_hash(hash
);
1521 return len
+ count_masked_bytes(mask
);
1526 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1528 unsigned long hash
= init_name_hash();
1530 hash
= partial_name_hash(*name
++, hash
);
1531 return end_name_hash(hash
);
1533 EXPORT_SYMBOL(full_name_hash
);
1536 * We know there's a real path component here of at least
1539 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1541 unsigned long hash
= init_name_hash();
1542 unsigned long len
= 0, c
;
1544 c
= (unsigned char)*name
;
1547 hash
= partial_name_hash(c
, hash
);
1548 c
= (unsigned char)name
[len
];
1549 } while (c
&& c
!= '/');
1550 *hashp
= end_name_hash(hash
);
1558 * This is the basic name resolution function, turning a pathname into
1559 * the final dentry. We expect 'base' to be positive and a directory.
1561 * Returns 0 and nd will have valid dentry and mnt on success.
1562 * Returns error and drops reference to input namei data on failure.
1564 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1574 /* At this point we know we have a real path component. */
1580 err
= may_lookup(nd
);
1584 len
= hash_name(name
, &this.hash
);
1589 if (name
[0] == '.') switch (len
) {
1591 if (name
[1] == '.') {
1593 nd
->flags
|= LOOKUP_JUMPED
;
1599 if (likely(type
== LAST_NORM
)) {
1600 struct dentry
*parent
= nd
->path
.dentry
;
1601 nd
->flags
&= ~LOOKUP_JUMPED
;
1602 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1603 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1611 goto last_component
;
1613 * If it wasn't NUL, we know it was '/'. Skip that
1614 * slash, and continue until no more slashes.
1618 } while (unlikely(name
[len
] == '/'));
1620 goto last_component
;
1623 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1628 err
= nested_symlink(&next
, nd
);
1632 if (can_lookup(nd
->inode
))
1636 /* here ends the main loop */
1640 nd
->last_type
= type
;
1647 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1648 struct nameidata
*nd
, struct file
**fp
)
1654 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1655 nd
->flags
= flags
| LOOKUP_JUMPED
;
1657 if (flags
& LOOKUP_ROOT
) {
1658 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1660 if (!inode
->i_op
->lookup
)
1662 retval
= inode_permission(inode
, MAY_EXEC
);
1666 nd
->path
= nd
->root
;
1668 if (flags
& LOOKUP_RCU
) {
1669 br_read_lock(vfsmount_lock
);
1671 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1673 path_get(&nd
->path
);
1678 nd
->root
.mnt
= NULL
;
1681 if (flags
& LOOKUP_RCU
) {
1682 br_read_lock(vfsmount_lock
);
1687 path_get(&nd
->root
);
1689 nd
->path
= nd
->root
;
1690 } else if (dfd
== AT_FDCWD
) {
1691 if (flags
& LOOKUP_RCU
) {
1692 struct fs_struct
*fs
= current
->fs
;
1695 br_read_lock(vfsmount_lock
);
1699 seq
= read_seqcount_begin(&fs
->seq
);
1701 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1702 } while (read_seqcount_retry(&fs
->seq
, seq
));
1704 get_fs_pwd(current
->fs
, &nd
->path
);
1707 struct dentry
*dentry
;
1709 file
= fget_raw_light(dfd
, &fput_needed
);
1714 dentry
= file
->f_path
.dentry
;
1718 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1721 retval
= inode_permission(dentry
->d_inode
, MAY_EXEC
);
1726 nd
->path
= file
->f_path
;
1727 if (flags
& LOOKUP_RCU
) {
1730 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1731 br_read_lock(vfsmount_lock
);
1734 path_get(&file
->f_path
);
1735 fput_light(file
, fput_needed
);
1739 nd
->inode
= nd
->path
.dentry
->d_inode
;
1743 fput_light(file
, fput_needed
);
1748 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1750 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1751 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1753 nd
->flags
&= ~LOOKUP_PARENT
;
1754 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1755 nd
->flags
& LOOKUP_FOLLOW
);
1758 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1759 static int path_lookupat(int dfd
, const char *name
,
1760 unsigned int flags
, struct nameidata
*nd
)
1762 struct file
*base
= NULL
;
1767 * Path walking is largely split up into 2 different synchronisation
1768 * schemes, rcu-walk and ref-walk (explained in
1769 * Documentation/filesystems/path-lookup.txt). These share much of the
1770 * path walk code, but some things particularly setup, cleanup, and
1771 * following mounts are sufficiently divergent that functions are
1772 * duplicated. Typically there is a function foo(), and its RCU
1773 * analogue, foo_rcu().
1775 * -ECHILD is the error number of choice (just to avoid clashes) that
1776 * is returned if some aspect of an rcu-walk fails. Such an error must
1777 * be handled by restarting a traditional ref-walk (which will always
1778 * be able to complete).
1780 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1785 current
->total_link_count
= 0;
1786 err
= link_path_walk(name
, nd
);
1788 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1789 err
= lookup_last(nd
, &path
);
1792 struct path link
= path
;
1793 nd
->flags
|= LOOKUP_PARENT
;
1794 err
= follow_link(&link
, nd
, &cookie
);
1796 err
= lookup_last(nd
, &path
);
1797 put_link(nd
, &link
, cookie
);
1802 err
= complete_walk(nd
);
1804 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1805 if (!nd
->inode
->i_op
->lookup
) {
1806 path_put(&nd
->path
);
1814 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1815 path_put(&nd
->root
);
1816 nd
->root
.mnt
= NULL
;
1821 static int do_path_lookup(int dfd
, const char *name
,
1822 unsigned int flags
, struct nameidata
*nd
)
1824 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1825 if (unlikely(retval
== -ECHILD
))
1826 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1827 if (unlikely(retval
== -ESTALE
))
1828 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1830 if (likely(!retval
)) {
1831 if (unlikely(!audit_dummy_context())) {
1832 if (nd
->path
.dentry
&& nd
->inode
)
1833 audit_inode(name
, nd
->path
.dentry
);
1839 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1841 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1844 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1846 struct nameidata nd
;
1847 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1854 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1855 * @dentry: pointer to dentry of the base directory
1856 * @mnt: pointer to vfs mount of the base directory
1857 * @name: pointer to file name
1858 * @flags: lookup flags
1859 * @path: pointer to struct path to fill
1861 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1862 const char *name
, unsigned int flags
,
1865 struct nameidata nd
;
1867 nd
.root
.dentry
= dentry
;
1869 BUG_ON(flags
& LOOKUP_PARENT
);
1870 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1871 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
1878 * Restricted form of lookup. Doesn't follow links, single-component only,
1879 * needs parent already locked. Doesn't follow mounts.
1882 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1884 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1888 * lookup_one_len - filesystem helper to lookup single pathname component
1889 * @name: pathname component to lookup
1890 * @base: base directory to lookup from
1891 * @len: maximum length @len should be interpreted to
1893 * Note that this routine is purely a helper for filesystem usage and should
1894 * not be called by generic code. Also note that by using this function the
1895 * nameidata argument is passed to the filesystem methods and a filesystem
1896 * using this helper needs to be prepared for that.
1898 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1904 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1908 this.hash
= full_name_hash(name
, len
);
1910 return ERR_PTR(-EACCES
);
1913 c
= *(const unsigned char *)name
++;
1914 if (c
== '/' || c
== '\0')
1915 return ERR_PTR(-EACCES
);
1918 * See if the low-level filesystem might want
1919 * to use its own hash..
1921 if (base
->d_flags
& DCACHE_OP_HASH
) {
1922 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1924 return ERR_PTR(err
);
1927 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
1929 return ERR_PTR(err
);
1931 return __lookup_hash(&this, base
, NULL
);
1934 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
1935 struct path
*path
, int *empty
)
1937 struct nameidata nd
;
1938 char *tmp
= getname_flags(name
, flags
, empty
);
1939 int err
= PTR_ERR(tmp
);
1942 BUG_ON(flags
& LOOKUP_PARENT
);
1944 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1952 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1955 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
1958 static int user_path_parent(int dfd
, const char __user
*path
,
1959 struct nameidata
*nd
, char **name
)
1961 char *s
= getname(path
);
1967 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1977 * It's inline, so penalty for filesystems that don't use sticky bit is
1980 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1982 uid_t fsuid
= current_fsuid();
1984 if (!(dir
->i_mode
& S_ISVTX
))
1986 if (current_user_ns() != inode_userns(inode
))
1988 if (inode
->i_uid
== fsuid
)
1990 if (dir
->i_uid
== fsuid
)
1994 return !ns_capable(inode_userns(inode
), CAP_FOWNER
);
1998 * Check whether we can remove a link victim from directory dir, check
1999 * whether the type of victim is right.
2000 * 1. We can't do it if dir is read-only (done in permission())
2001 * 2. We should have write and exec permissions on dir
2002 * 3. We can't remove anything from append-only dir
2003 * 4. We can't do anything with immutable dir (done in permission())
2004 * 5. If the sticky bit on dir is set we should either
2005 * a. be owner of dir, or
2006 * b. be owner of victim, or
2007 * c. have CAP_FOWNER capability
2008 * 6. If the victim is append-only or immutable we can't do antyhing with
2009 * links pointing to it.
2010 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2011 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2012 * 9. We can't remove a root or mountpoint.
2013 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2014 * nfs_async_unlink().
2016 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
2020 if (!victim
->d_inode
)
2023 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2024 audit_inode_child(victim
, dir
);
2026 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2031 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
2032 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
2035 if (!S_ISDIR(victim
->d_inode
->i_mode
))
2037 if (IS_ROOT(victim
))
2039 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
2041 if (IS_DEADDIR(dir
))
2043 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2048 /* Check whether we can create an object with dentry child in directory
2050 * 1. We can't do it if child already exists (open has special treatment for
2051 * this case, but since we are inlined it's OK)
2052 * 2. We can't do it if dir is read-only (done in permission())
2053 * 3. We should have write and exec permissions on dir
2054 * 4. We can't do it if dir is immutable (done in permission())
2056 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2060 if (IS_DEADDIR(dir
))
2062 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2066 * p1 and p2 should be directories on the same fs.
2068 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2073 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2077 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2079 p
= d_ancestor(p2
, p1
);
2081 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2082 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2086 p
= d_ancestor(p1
, p2
);
2088 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2089 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2093 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2094 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2098 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2100 mutex_unlock(&p1
->d_inode
->i_mutex
);
2102 mutex_unlock(&p2
->d_inode
->i_mutex
);
2103 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2107 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2108 struct nameidata
*nd
)
2110 int error
= may_create(dir
, dentry
);
2115 if (!dir
->i_op
->create
)
2116 return -EACCES
; /* shouldn't it be ENOSYS? */
2119 error
= security_inode_create(dir
, dentry
, mode
);
2122 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
2124 fsnotify_create(dir
, dentry
);
2128 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2130 struct dentry
*dentry
= path
->dentry
;
2131 struct inode
*inode
= dentry
->d_inode
;
2141 switch (inode
->i_mode
& S_IFMT
) {
2145 if (acc_mode
& MAY_WRITE
)
2150 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2159 error
= inode_permission(inode
, acc_mode
);
2164 * An append-only file must be opened in append mode for writing.
2166 if (IS_APPEND(inode
)) {
2167 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2173 /* O_NOATIME can only be set by the owner or superuser */
2174 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2180 static int handle_truncate(struct file
*filp
)
2182 struct path
*path
= &filp
->f_path
;
2183 struct inode
*inode
= path
->dentry
->d_inode
;
2184 int error
= get_write_access(inode
);
2188 * Refuse to truncate files with mandatory locks held on them.
2190 error
= locks_verify_locked(inode
);
2192 error
= security_path_truncate(path
);
2194 error
= do_truncate(path
->dentry
, 0,
2195 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2198 put_write_access(inode
);
2202 static inline int open_to_namei_flags(int flag
)
2204 if ((flag
& O_ACCMODE
) == 3)
2210 * Handle the last step of open()
2212 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2213 const struct open_flags
*op
, const char *pathname
)
2215 struct dentry
*dir
= nd
->path
.dentry
;
2216 struct dentry
*dentry
;
2217 int open_flag
= op
->open_flag
;
2218 int will_truncate
= open_flag
& O_TRUNC
;
2220 int acc_mode
= op
->acc_mode
;
2224 nd
->flags
&= ~LOOKUP_PARENT
;
2225 nd
->flags
|= op
->intent
;
2227 switch (nd
->last_type
) {
2230 error
= handle_dots(nd
, nd
->last_type
);
2232 return ERR_PTR(error
);
2235 error
= complete_walk(nd
);
2237 return ERR_PTR(error
);
2238 audit_inode(pathname
, nd
->path
.dentry
);
2239 if (open_flag
& O_CREAT
) {
2245 error
= complete_walk(nd
);
2247 return ERR_PTR(error
);
2248 audit_inode(pathname
, dir
);
2252 if (!(open_flag
& O_CREAT
)) {
2254 if (nd
->last
.name
[nd
->last
.len
])
2255 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2256 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2258 /* we _can_ be in RCU mode here */
2259 error
= walk_component(nd
, path
, &nd
->last
, LAST_NORM
,
2262 return ERR_PTR(error
);
2263 if (error
) /* symlink */
2266 error
= complete_walk(nd
);
2268 return ERR_PTR(error
);
2271 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2272 if (!nd
->inode
->i_op
->lookup
)
2275 audit_inode(pathname
, nd
->path
.dentry
);
2279 /* create side of things */
2281 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2282 * cleared when we got to the last component we are about to look up
2284 error
= complete_walk(nd
);
2286 return ERR_PTR(error
);
2288 audit_inode(pathname
, dir
);
2290 /* trailing slashes? */
2291 if (nd
->last
.name
[nd
->last
.len
])
2294 mutex_lock(&dir
->d_inode
->i_mutex
);
2296 dentry
= lookup_hash(nd
);
2297 error
= PTR_ERR(dentry
);
2298 if (IS_ERR(dentry
)) {
2299 mutex_unlock(&dir
->d_inode
->i_mutex
);
2303 path
->dentry
= dentry
;
2304 path
->mnt
= nd
->path
.mnt
;
2306 /* Negative dentry, just create the file */
2307 if (!dentry
->d_inode
) {
2308 umode_t mode
= op
->mode
;
2309 if (!IS_POSIXACL(dir
->d_inode
))
2310 mode
&= ~current_umask();
2312 * This write is needed to ensure that a
2313 * rw->ro transition does not occur between
2314 * the time when the file is created and when
2315 * a permanent write count is taken through
2316 * the 'struct file' in nameidata_to_filp().
2318 error
= mnt_want_write(nd
->path
.mnt
);
2320 goto exit_mutex_unlock
;
2322 /* Don't check for write permission, don't truncate */
2323 open_flag
&= ~O_TRUNC
;
2325 acc_mode
= MAY_OPEN
;
2326 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2328 goto exit_mutex_unlock
;
2329 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2331 goto exit_mutex_unlock
;
2332 mutex_unlock(&dir
->d_inode
->i_mutex
);
2333 dput(nd
->path
.dentry
);
2334 nd
->path
.dentry
= dentry
;
2339 * It already exists.
2341 mutex_unlock(&dir
->d_inode
->i_mutex
);
2342 audit_inode(pathname
, path
->dentry
);
2345 if (open_flag
& O_EXCL
)
2348 error
= follow_managed(path
, nd
->flags
);
2353 nd
->flags
|= LOOKUP_JUMPED
;
2356 if (!path
->dentry
->d_inode
)
2359 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2362 path_to_nameidata(path
, nd
);
2363 nd
->inode
= path
->dentry
->d_inode
;
2364 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2365 error
= complete_walk(nd
);
2367 return ERR_PTR(error
);
2369 if (S_ISDIR(nd
->inode
->i_mode
))
2372 if (!S_ISREG(nd
->inode
->i_mode
))
2375 if (will_truncate
) {
2376 error
= mnt_want_write(nd
->path
.mnt
);
2382 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2385 filp
= nameidata_to_filp(nd
);
2386 if (!IS_ERR(filp
)) {
2387 error
= ima_file_check(filp
, op
->acc_mode
);
2390 filp
= ERR_PTR(error
);
2393 if (!IS_ERR(filp
)) {
2394 if (will_truncate
) {
2395 error
= handle_truncate(filp
);
2398 filp
= ERR_PTR(error
);
2404 mnt_drop_write(nd
->path
.mnt
);
2405 path_put(&nd
->path
);
2409 mutex_unlock(&dir
->d_inode
->i_mutex
);
2411 path_put_conditional(path
, nd
);
2413 filp
= ERR_PTR(error
);
2417 static struct file
*path_openat(int dfd
, const char *pathname
,
2418 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2420 struct file
*base
= NULL
;
2425 filp
= get_empty_filp();
2427 return ERR_PTR(-ENFILE
);
2429 filp
->f_flags
= op
->open_flag
;
2430 nd
->intent
.open
.file
= filp
;
2431 nd
->intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2432 nd
->intent
.open
.create_mode
= op
->mode
;
2434 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2435 if (unlikely(error
))
2438 current
->total_link_count
= 0;
2439 error
= link_path_walk(pathname
, nd
);
2440 if (unlikely(error
))
2443 filp
= do_last(nd
, &path
, op
, pathname
);
2444 while (unlikely(!filp
)) { /* trailing symlink */
2445 struct path link
= path
;
2447 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2448 path_put_conditional(&path
, nd
);
2449 path_put(&nd
->path
);
2450 filp
= ERR_PTR(-ELOOP
);
2453 nd
->flags
|= LOOKUP_PARENT
;
2454 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2455 error
= follow_link(&link
, nd
, &cookie
);
2456 if (unlikely(error
))
2457 filp
= ERR_PTR(error
);
2459 filp
= do_last(nd
, &path
, op
, pathname
);
2460 put_link(nd
, &link
, cookie
);
2463 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2464 path_put(&nd
->root
);
2467 release_open_intent(nd
);
2471 filp
= ERR_PTR(error
);
2475 struct file
*do_filp_open(int dfd
, const char *pathname
,
2476 const struct open_flags
*op
, int flags
)
2478 struct nameidata nd
;
2481 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2482 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2483 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2484 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2485 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2489 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2490 const char *name
, const struct open_flags
*op
, int flags
)
2492 struct nameidata nd
;
2496 nd
.root
.dentry
= dentry
;
2498 flags
|= LOOKUP_ROOT
;
2500 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2501 return ERR_PTR(-ELOOP
);
2503 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2504 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2505 file
= path_openat(-1, name
, &nd
, op
, flags
);
2506 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2507 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2511 struct dentry
*kern_path_create(int dfd
, const char *pathname
, struct path
*path
, int is_dir
)
2513 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2514 struct nameidata nd
;
2515 int error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
, &nd
);
2517 return ERR_PTR(error
);
2520 * Yucky last component or no last component at all?
2521 * (foo/., foo/.., /////)
2523 if (nd
.last_type
!= LAST_NORM
)
2525 nd
.flags
&= ~LOOKUP_PARENT
;
2526 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2527 nd
.intent
.open
.flags
= O_EXCL
;
2530 * Do the final lookup.
2532 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2533 dentry
= lookup_hash(&nd
);
2537 if (dentry
->d_inode
)
2540 * Special case - lookup gave negative, but... we had foo/bar/
2541 * From the vfs_mknod() POV we just have a negative dentry -
2542 * all is fine. Let's be bastards - you had / on the end, you've
2543 * been asking for (non-existent) directory. -ENOENT for you.
2545 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
2547 dentry
= ERR_PTR(-ENOENT
);
2554 dentry
= ERR_PTR(-EEXIST
);
2556 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2561 EXPORT_SYMBOL(kern_path_create
);
2563 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
, struct path
*path
, int is_dir
)
2565 char *tmp
= getname(pathname
);
2568 return ERR_CAST(tmp
);
2569 res
= kern_path_create(dfd
, tmp
, path
, is_dir
);
2573 EXPORT_SYMBOL(user_path_create
);
2575 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2577 int error
= may_create(dir
, dentry
);
2582 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) &&
2583 !ns_capable(inode_userns(dir
), CAP_MKNOD
))
2586 if (!dir
->i_op
->mknod
)
2589 error
= devcgroup_inode_mknod(mode
, dev
);
2593 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2597 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2599 fsnotify_create(dir
, dentry
);
2603 static int may_mknod(umode_t mode
)
2605 switch (mode
& S_IFMT
) {
2611 case 0: /* zero mode translates to S_IFREG */
2620 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
2623 struct dentry
*dentry
;
2630 dentry
= user_path_create(dfd
, filename
, &path
, 0);
2632 return PTR_ERR(dentry
);
2634 if (!IS_POSIXACL(path
.dentry
->d_inode
))
2635 mode
&= ~current_umask();
2636 error
= may_mknod(mode
);
2639 error
= mnt_want_write(path
.mnt
);
2642 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
2644 goto out_drop_write
;
2645 switch (mode
& S_IFMT
) {
2646 case 0: case S_IFREG
:
2647 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,NULL
);
2649 case S_IFCHR
: case S_IFBLK
:
2650 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
2651 new_decode_dev(dev
));
2653 case S_IFIFO
: case S_IFSOCK
:
2654 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
2658 mnt_drop_write(path
.mnt
);
2661 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2667 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
2669 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2672 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2674 int error
= may_create(dir
, dentry
);
2675 unsigned max_links
= dir
->i_sb
->s_max_links
;
2680 if (!dir
->i_op
->mkdir
)
2683 mode
&= (S_IRWXUGO
|S_ISVTX
);
2684 error
= security_inode_mkdir(dir
, dentry
, mode
);
2688 if (max_links
&& dir
->i_nlink
>= max_links
)
2691 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2693 fsnotify_mkdir(dir
, dentry
);
2697 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
2699 struct dentry
*dentry
;
2703 dentry
= user_path_create(dfd
, pathname
, &path
, 1);
2705 return PTR_ERR(dentry
);
2707 if (!IS_POSIXACL(path
.dentry
->d_inode
))
2708 mode
&= ~current_umask();
2709 error
= mnt_want_write(path
.mnt
);
2712 error
= security_path_mkdir(&path
, dentry
, mode
);
2714 goto out_drop_write
;
2715 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
2717 mnt_drop_write(path
.mnt
);
2720 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2725 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
2727 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2731 * The dentry_unhash() helper will try to drop the dentry early: we
2732 * should have a usage count of 2 if we're the only user of this
2733 * dentry, and if that is true (possibly after pruning the dcache),
2734 * then we drop the dentry now.
2736 * A low-level filesystem can, if it choses, legally
2739 * if (!d_unhashed(dentry))
2742 * if it cannot handle the case of removing a directory
2743 * that is still in use by something else..
2745 void dentry_unhash(struct dentry
*dentry
)
2747 shrink_dcache_parent(dentry
);
2748 spin_lock(&dentry
->d_lock
);
2749 if (dentry
->d_count
== 1)
2751 spin_unlock(&dentry
->d_lock
);
2754 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2756 int error
= may_delete(dir
, dentry
, 1);
2761 if (!dir
->i_op
->rmdir
)
2765 mutex_lock(&dentry
->d_inode
->i_mutex
);
2768 if (d_mountpoint(dentry
))
2771 error
= security_inode_rmdir(dir
, dentry
);
2775 shrink_dcache_parent(dentry
);
2776 error
= dir
->i_op
->rmdir(dir
, dentry
);
2780 dentry
->d_inode
->i_flags
|= S_DEAD
;
2784 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2791 static long do_rmdir(int dfd
, const char __user
*pathname
)
2795 struct dentry
*dentry
;
2796 struct nameidata nd
;
2798 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2802 switch(nd
.last_type
) {
2814 nd
.flags
&= ~LOOKUP_PARENT
;
2816 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2817 dentry
= lookup_hash(&nd
);
2818 error
= PTR_ERR(dentry
);
2821 if (!dentry
->d_inode
) {
2825 error
= mnt_want_write(nd
.path
.mnt
);
2828 error
= security_path_rmdir(&nd
.path
, dentry
);
2831 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2833 mnt_drop_write(nd
.path
.mnt
);
2837 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2844 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2846 return do_rmdir(AT_FDCWD
, pathname
);
2849 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2851 int error
= may_delete(dir
, dentry
, 0);
2856 if (!dir
->i_op
->unlink
)
2859 mutex_lock(&dentry
->d_inode
->i_mutex
);
2860 if (d_mountpoint(dentry
))
2863 error
= security_inode_unlink(dir
, dentry
);
2865 error
= dir
->i_op
->unlink(dir
, dentry
);
2870 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2872 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2873 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2874 fsnotify_link_count(dentry
->d_inode
);
2882 * Make sure that the actual truncation of the file will occur outside its
2883 * directory's i_mutex. Truncate can take a long time if there is a lot of
2884 * writeout happening, and we don't want to prevent access to the directory
2885 * while waiting on the I/O.
2887 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2891 struct dentry
*dentry
;
2892 struct nameidata nd
;
2893 struct inode
*inode
= NULL
;
2895 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2900 if (nd
.last_type
!= LAST_NORM
)
2903 nd
.flags
&= ~LOOKUP_PARENT
;
2905 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2906 dentry
= lookup_hash(&nd
);
2907 error
= PTR_ERR(dentry
);
2908 if (!IS_ERR(dentry
)) {
2909 /* Why not before? Because we want correct error value */
2910 if (nd
.last
.name
[nd
.last
.len
])
2912 inode
= dentry
->d_inode
;
2916 error
= mnt_want_write(nd
.path
.mnt
);
2919 error
= security_path_unlink(&nd
.path
, dentry
);
2922 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2924 mnt_drop_write(nd
.path
.mnt
);
2928 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2930 iput(inode
); /* truncate the inode here */
2937 error
= !dentry
->d_inode
? -ENOENT
:
2938 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2942 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2944 if ((flag
& ~AT_REMOVEDIR
) != 0)
2947 if (flag
& AT_REMOVEDIR
)
2948 return do_rmdir(dfd
, pathname
);
2950 return do_unlinkat(dfd
, pathname
);
2953 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2955 return do_unlinkat(AT_FDCWD
, pathname
);
2958 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2960 int error
= may_create(dir
, dentry
);
2965 if (!dir
->i_op
->symlink
)
2968 error
= security_inode_symlink(dir
, dentry
, oldname
);
2972 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2974 fsnotify_create(dir
, dentry
);
2978 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2979 int, newdfd
, const char __user
*, newname
)
2983 struct dentry
*dentry
;
2986 from
= getname(oldname
);
2988 return PTR_ERR(from
);
2990 dentry
= user_path_create(newdfd
, newname
, &path
, 0);
2991 error
= PTR_ERR(dentry
);
2995 error
= mnt_want_write(path
.mnt
);
2998 error
= security_path_symlink(&path
, dentry
, from
);
3000 goto out_drop_write
;
3001 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
);
3003 mnt_drop_write(path
.mnt
);
3006 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
3013 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3015 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3018 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3020 struct inode
*inode
= old_dentry
->d_inode
;
3021 unsigned max_links
= dir
->i_sb
->s_max_links
;
3027 error
= may_create(dir
, new_dentry
);
3031 if (dir
->i_sb
!= inode
->i_sb
)
3035 * A link to an append-only or immutable file cannot be created.
3037 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3039 if (!dir
->i_op
->link
)
3041 if (S_ISDIR(inode
->i_mode
))
3044 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3048 mutex_lock(&inode
->i_mutex
);
3049 /* Make sure we don't allow creating hardlink to an unlinked file */
3050 if (inode
->i_nlink
== 0)
3052 else if (max_links
&& inode
->i_nlink
>= max_links
)
3055 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3056 mutex_unlock(&inode
->i_mutex
);
3058 fsnotify_link(dir
, inode
, new_dentry
);
3063 * Hardlinks are often used in delicate situations. We avoid
3064 * security-related surprises by not following symlinks on the
3067 * We don't follow them on the oldname either to be compatible
3068 * with linux 2.0, and to avoid hard-linking to directories
3069 * and other special files. --ADM
3071 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3072 int, newdfd
, const char __user
*, newname
, int, flags
)
3074 struct dentry
*new_dentry
;
3075 struct path old_path
, new_path
;
3079 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3082 * To use null names we require CAP_DAC_READ_SEARCH
3083 * This ensures that not everyone will be able to create
3084 * handlink using the passed filedescriptor.
3086 if (flags
& AT_EMPTY_PATH
) {
3087 if (!capable(CAP_DAC_READ_SEARCH
))
3092 if (flags
& AT_SYMLINK_FOLLOW
)
3093 how
|= LOOKUP_FOLLOW
;
3095 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3099 new_dentry
= user_path_create(newdfd
, newname
, &new_path
, 0);
3100 error
= PTR_ERR(new_dentry
);
3101 if (IS_ERR(new_dentry
))
3105 if (old_path
.mnt
!= new_path
.mnt
)
3107 error
= mnt_want_write(new_path
.mnt
);
3110 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3112 goto out_drop_write
;
3113 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3115 mnt_drop_write(new_path
.mnt
);
3118 mutex_unlock(&new_path
.dentry
->d_inode
->i_mutex
);
3119 path_put(&new_path
);
3121 path_put(&old_path
);
3126 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3128 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3132 * The worst of all namespace operations - renaming directory. "Perverted"
3133 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3135 * a) we can get into loop creation. Check is done in is_subdir().
3136 * b) race potential - two innocent renames can create a loop together.
3137 * That's where 4.4 screws up. Current fix: serialization on
3138 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3140 * c) we have to lock _three_ objects - parents and victim (if it exists).
3141 * And that - after we got ->i_mutex on parents (until then we don't know
3142 * whether the target exists). Solution: try to be smart with locking
3143 * order for inodes. We rely on the fact that tree topology may change
3144 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3145 * move will be locked. Thus we can rank directories by the tree
3146 * (ancestors first) and rank all non-directories after them.
3147 * That works since everybody except rename does "lock parent, lookup,
3148 * lock child" and rename is under ->s_vfs_rename_mutex.
3149 * HOWEVER, it relies on the assumption that any object with ->lookup()
3150 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3151 * we'd better make sure that there's no link(2) for them.
3152 * d) conversion from fhandle to dentry may come in the wrong moment - when
3153 * we are removing the target. Solution: we will have to grab ->i_mutex
3154 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3155 * ->i_mutex on parents, which works but leads to some truly excessive
3158 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3159 struct inode
*new_dir
, struct dentry
*new_dentry
)
3162 struct inode
*target
= new_dentry
->d_inode
;
3163 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3166 * If we are going to change the parent - check write permissions,
3167 * we'll need to flip '..'.
3169 if (new_dir
!= old_dir
) {
3170 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3175 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3181 mutex_lock(&target
->i_mutex
);
3184 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3188 if (max_links
&& !target
&& new_dir
!= old_dir
&&
3189 new_dir
->i_nlink
>= max_links
)
3193 shrink_dcache_parent(new_dentry
);
3194 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3199 target
->i_flags
|= S_DEAD
;
3200 dont_mount(new_dentry
);
3204 mutex_unlock(&target
->i_mutex
);
3207 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3208 d_move(old_dentry
,new_dentry
);
3212 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3213 struct inode
*new_dir
, struct dentry
*new_dentry
)
3215 struct inode
*target
= new_dentry
->d_inode
;
3218 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3224 mutex_lock(&target
->i_mutex
);
3227 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3230 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3235 dont_mount(new_dentry
);
3236 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3237 d_move(old_dentry
, new_dentry
);
3240 mutex_unlock(&target
->i_mutex
);
3245 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3246 struct inode
*new_dir
, struct dentry
*new_dentry
)
3249 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3250 const unsigned char *old_name
;
3252 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3255 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3259 if (!new_dentry
->d_inode
)
3260 error
= may_create(new_dir
, new_dentry
);
3262 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3266 if (!old_dir
->i_op
->rename
)
3269 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3272 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3274 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3276 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3277 new_dentry
->d_inode
, old_dentry
);
3278 fsnotify_oldname_free(old_name
);
3283 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3284 int, newdfd
, const char __user
*, newname
)
3286 struct dentry
*old_dir
, *new_dir
;
3287 struct dentry
*old_dentry
, *new_dentry
;
3288 struct dentry
*trap
;
3289 struct nameidata oldnd
, newnd
;
3294 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3298 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3303 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3306 old_dir
= oldnd
.path
.dentry
;
3308 if (oldnd
.last_type
!= LAST_NORM
)
3311 new_dir
= newnd
.path
.dentry
;
3312 if (newnd
.last_type
!= LAST_NORM
)
3315 oldnd
.flags
&= ~LOOKUP_PARENT
;
3316 newnd
.flags
&= ~LOOKUP_PARENT
;
3317 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3319 trap
= lock_rename(new_dir
, old_dir
);
3321 old_dentry
= lookup_hash(&oldnd
);
3322 error
= PTR_ERR(old_dentry
);
3323 if (IS_ERR(old_dentry
))
3325 /* source must exist */
3327 if (!old_dentry
->d_inode
)
3329 /* unless the source is a directory trailing slashes give -ENOTDIR */
3330 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3332 if (oldnd
.last
.name
[oldnd
.last
.len
])
3334 if (newnd
.last
.name
[newnd
.last
.len
])
3337 /* source should not be ancestor of target */
3339 if (old_dentry
== trap
)
3341 new_dentry
= lookup_hash(&newnd
);
3342 error
= PTR_ERR(new_dentry
);
3343 if (IS_ERR(new_dentry
))
3345 /* target should not be an ancestor of source */
3347 if (new_dentry
== trap
)
3350 error
= mnt_want_write(oldnd
.path
.mnt
);
3353 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3354 &newnd
.path
, new_dentry
);
3357 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3358 new_dir
->d_inode
, new_dentry
);
3360 mnt_drop_write(oldnd
.path
.mnt
);
3366 unlock_rename(new_dir
, old_dir
);
3368 path_put(&newnd
.path
);
3371 path_put(&oldnd
.path
);
3377 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3379 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3382 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3386 len
= PTR_ERR(link
);
3391 if (len
> (unsigned) buflen
)
3393 if (copy_to_user(buffer
, link
, len
))
3400 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3401 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3402 * using) it for any given inode is up to filesystem.
3404 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3406 struct nameidata nd
;
3411 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3413 return PTR_ERR(cookie
);
3415 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3416 if (dentry
->d_inode
->i_op
->put_link
)
3417 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3421 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3423 return __vfs_follow_link(nd
, link
);
3426 /* get the link contents into pagecache */
3427 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3431 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3432 page
= read_mapping_page(mapping
, 0, NULL
);
3437 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3441 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3443 struct page
*page
= NULL
;
3444 char *s
= page_getlink(dentry
, &page
);
3445 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3448 page_cache_release(page
);
3453 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3455 struct page
*page
= NULL
;
3456 nd_set_link(nd
, page_getlink(dentry
, &page
));
3460 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3462 struct page
*page
= cookie
;
3466 page_cache_release(page
);
3471 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3473 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3475 struct address_space
*mapping
= inode
->i_mapping
;
3480 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3482 flags
|= AOP_FLAG_NOFS
;
3485 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3486 flags
, &page
, &fsdata
);
3490 kaddr
= kmap_atomic(page
);
3491 memcpy(kaddr
, symname
, len
-1);
3492 kunmap_atomic(kaddr
);
3494 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3501 mark_inode_dirty(inode
);
3507 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3509 return __page_symlink(inode
, symname
, len
,
3510 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3513 const struct inode_operations page_symlink_inode_operations
= {
3514 .readlink
= generic_readlink
,
3515 .follow_link
= page_follow_link_light
,
3516 .put_link
= page_put_link
,
3519 EXPORT_SYMBOL(user_path_at
);
3520 EXPORT_SYMBOL(follow_down_one
);
3521 EXPORT_SYMBOL(follow_down
);
3522 EXPORT_SYMBOL(follow_up
);
3523 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3524 EXPORT_SYMBOL(getname
);
3525 EXPORT_SYMBOL(lock_rename
);
3526 EXPORT_SYMBOL(lookup_one_len
);
3527 EXPORT_SYMBOL(page_follow_link_light
);
3528 EXPORT_SYMBOL(page_put_link
);
3529 EXPORT_SYMBOL(page_readlink
);
3530 EXPORT_SYMBOL(__page_symlink
);
3531 EXPORT_SYMBOL(page_symlink
);
3532 EXPORT_SYMBOL(page_symlink_inode_operations
);
3533 EXPORT_SYMBOL(kern_path
);
3534 EXPORT_SYMBOL(vfs_path_lookup
);
3535 EXPORT_SYMBOL(inode_permission
);
3536 EXPORT_SYMBOL(unlock_rename
);
3537 EXPORT_SYMBOL(vfs_create
);
3538 EXPORT_SYMBOL(vfs_follow_link
);
3539 EXPORT_SYMBOL(vfs_link
);
3540 EXPORT_SYMBOL(vfs_mkdir
);
3541 EXPORT_SYMBOL(vfs_mknod
);
3542 EXPORT_SYMBOL(generic_permission
);
3543 EXPORT_SYMBOL(vfs_readlink
);
3544 EXPORT_SYMBOL(vfs_rename
);
3545 EXPORT_SYMBOL(vfs_rmdir
);
3546 EXPORT_SYMBOL(vfs_symlink
);
3547 EXPORT_SYMBOL(vfs_unlink
);
3548 EXPORT_SYMBOL(dentry_unhash
);
3549 EXPORT_SYMBOL(generic_readlink
);