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 char *getname_flags(const char __user
*filename
, int flags
, int *empty
)
121 char *result
= __getname(), *err
;
124 if (unlikely(!result
))
125 return ERR_PTR(-ENOMEM
);
127 len
= strncpy_from_user(result
, filename
, PATH_MAX
);
129 if (unlikely(len
< 0))
132 /* The empty path is special. */
133 if (unlikely(!len
)) {
136 err
= ERR_PTR(-ENOENT
);
137 if (!(flags
& LOOKUP_EMPTY
))
141 err
= ERR_PTR(-ENAMETOOLONG
);
142 if (likely(len
< PATH_MAX
)) {
143 audit_getname(result
);
152 char *getname(const char __user
* filename
)
154 return getname_flags(filename
, 0, NULL
);
157 #ifdef CONFIG_AUDITSYSCALL
158 void putname(const char *name
)
160 if (unlikely(!audit_dummy_context()))
165 EXPORT_SYMBOL(putname
);
168 static int check_acl(struct inode
*inode
, int mask
)
170 #ifdef CONFIG_FS_POSIX_ACL
171 struct posix_acl
*acl
;
173 if (mask
& MAY_NOT_BLOCK
) {
174 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
177 /* no ->get_acl() calls in RCU mode... */
178 if (acl
== ACL_NOT_CACHED
)
180 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
183 acl
= get_cached_acl(inode
, ACL_TYPE_ACCESS
);
186 * A filesystem can force a ACL callback by just never filling the
187 * ACL cache. But normally you'd fill the cache either at inode
188 * instantiation time, or on the first ->get_acl call.
190 * If the filesystem doesn't have a get_acl() function at all, we'll
191 * just create the negative cache entry.
193 if (acl
== ACL_NOT_CACHED
) {
194 if (inode
->i_op
->get_acl
) {
195 acl
= inode
->i_op
->get_acl(inode
, ACL_TYPE_ACCESS
);
199 set_cached_acl(inode
, ACL_TYPE_ACCESS
, NULL
);
205 int error
= posix_acl_permission(inode
, acl
, mask
);
206 posix_acl_release(acl
);
215 * This does the basic permission checking
217 static int acl_permission_check(struct inode
*inode
, int mask
)
219 unsigned int mode
= inode
->i_mode
;
221 if (current_user_ns() != inode_userns(inode
))
224 if (likely(current_fsuid() == inode
->i_uid
))
227 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
228 int error
= check_acl(inode
, mask
);
229 if (error
!= -EAGAIN
)
233 if (in_group_p(inode
->i_gid
))
239 * If the DACs are ok we don't need any capability check.
241 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
247 * generic_permission - check for access rights on a Posix-like filesystem
248 * @inode: inode to check access rights for
249 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
251 * Used to check for read/write/execute permissions on a file.
252 * We use "fsuid" for this, letting us set arbitrary permissions
253 * for filesystem access without changing the "normal" uids which
254 * are used for other things.
256 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
257 * request cannot be satisfied (eg. requires blocking or too much complexity).
258 * It would then be called again in ref-walk mode.
260 int generic_permission(struct inode
*inode
, int mask
)
265 * Do the basic permission checks.
267 ret
= acl_permission_check(inode
, mask
);
271 if (S_ISDIR(inode
->i_mode
)) {
272 /* DACs are overridable for directories */
273 if (ns_capable(inode_userns(inode
), CAP_DAC_OVERRIDE
))
275 if (!(mask
& MAY_WRITE
))
276 if (ns_capable(inode_userns(inode
), CAP_DAC_READ_SEARCH
))
281 * Read/write DACs are always overridable.
282 * Executable DACs are overridable when there is
283 * at least one exec bit set.
285 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
286 if (ns_capable(inode_userns(inode
), CAP_DAC_OVERRIDE
))
290 * Searching includes executable on directories, else just read.
292 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
293 if (mask
== MAY_READ
)
294 if (ns_capable(inode_userns(inode
), CAP_DAC_READ_SEARCH
))
301 * We _really_ want to just do "generic_permission()" without
302 * even looking at the inode->i_op values. So we keep a cache
303 * flag in inode->i_opflags, that says "this has not special
304 * permission function, use the fast case".
306 static inline int do_inode_permission(struct inode
*inode
, int mask
)
308 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
309 if (likely(inode
->i_op
->permission
))
310 return inode
->i_op
->permission(inode
, mask
);
312 /* This gets set once for the inode lifetime */
313 spin_lock(&inode
->i_lock
);
314 inode
->i_opflags
|= IOP_FASTPERM
;
315 spin_unlock(&inode
->i_lock
);
317 return generic_permission(inode
, mask
);
321 * inode_permission - check for access rights to a given inode
322 * @inode: inode to check permission on
323 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
325 * Used to check for read/write/execute permissions on an inode.
326 * We use "fsuid" for this, letting us set arbitrary permissions
327 * for filesystem access without changing the "normal" uids which
328 * are used for other things.
330 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
332 int inode_permission(struct inode
*inode
, int mask
)
336 if (unlikely(mask
& MAY_WRITE
)) {
337 umode_t mode
= inode
->i_mode
;
340 * Nobody gets write access to a read-only fs.
342 if (IS_RDONLY(inode
) &&
343 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
347 * Nobody gets write access to an immutable file.
349 if (IS_IMMUTABLE(inode
))
353 retval
= do_inode_permission(inode
, mask
);
357 retval
= devcgroup_inode_permission(inode
, mask
);
361 return security_inode_permission(inode
, mask
);
365 * path_get - get a reference to a path
366 * @path: path to get the reference to
368 * Given a path increment the reference count to the dentry and the vfsmount.
370 void path_get(struct path
*path
)
375 EXPORT_SYMBOL(path_get
);
378 * path_put - put a reference to a path
379 * @path: path to put the reference to
381 * Given a path decrement the reference count to the dentry and the vfsmount.
383 void path_put(struct path
*path
)
388 EXPORT_SYMBOL(path_put
);
391 * Path walking has 2 modes, rcu-walk and ref-walk (see
392 * Documentation/filesystems/path-lookup.txt). In situations when we can't
393 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
394 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
395 * mode. Refcounts are grabbed at the last known good point before rcu-walk
396 * got stuck, so ref-walk may continue from there. If this is not successful
397 * (eg. a seqcount has changed), then failure is returned and it's up to caller
398 * to restart the path walk from the beginning in ref-walk mode.
402 * unlazy_walk - try to switch to ref-walk mode.
403 * @nd: nameidata pathwalk data
404 * @dentry: child of nd->path.dentry or NULL
405 * Returns: 0 on success, -ECHILD on failure
407 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
408 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
409 * @nd or NULL. Must be called from rcu-walk context.
411 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
413 struct fs_struct
*fs
= current
->fs
;
414 struct dentry
*parent
= nd
->path
.dentry
;
417 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
418 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
420 spin_lock(&fs
->lock
);
421 if (nd
->root
.mnt
!= fs
->root
.mnt
||
422 nd
->root
.dentry
!= fs
->root
.dentry
)
425 spin_lock(&parent
->d_lock
);
427 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
429 BUG_ON(nd
->inode
!= parent
->d_inode
);
431 if (dentry
->d_parent
!= parent
)
433 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
434 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
437 * If the sequence check on the child dentry passed, then
438 * the child has not been removed from its parent. This
439 * means the parent dentry must be valid and able to take
440 * a reference at this point.
442 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
443 BUG_ON(!parent
->d_count
);
445 spin_unlock(&dentry
->d_lock
);
447 spin_unlock(&parent
->d_lock
);
450 spin_unlock(&fs
->lock
);
452 mntget(nd
->path
.mnt
);
455 br_read_unlock(vfsmount_lock
);
456 nd
->flags
&= ~LOOKUP_RCU
;
460 spin_unlock(&dentry
->d_lock
);
462 spin_unlock(&parent
->d_lock
);
465 spin_unlock(&fs
->lock
);
470 * release_open_intent - free up open intent resources
471 * @nd: pointer to nameidata
473 void release_open_intent(struct nameidata
*nd
)
475 struct file
*file
= nd
->intent
.open
.file
;
477 if (file
&& !IS_ERR(file
)) {
478 if (file
->f_path
.dentry
== NULL
)
485 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
487 return dentry
->d_op
->d_revalidate(dentry
, nd
);
491 * complete_walk - successful completion of path walk
492 * @nd: pointer nameidata
494 * If we had been in RCU mode, drop out of it and legitimize nd->path.
495 * Revalidate the final result, unless we'd already done that during
496 * the path walk or the filesystem doesn't ask for it. Return 0 on
497 * success, -error on failure. In case of failure caller does not
498 * need to drop nd->path.
500 static int complete_walk(struct nameidata
*nd
)
502 struct dentry
*dentry
= nd
->path
.dentry
;
505 if (nd
->flags
& LOOKUP_RCU
) {
506 nd
->flags
&= ~LOOKUP_RCU
;
507 if (!(nd
->flags
& LOOKUP_ROOT
))
509 spin_lock(&dentry
->d_lock
);
510 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
511 spin_unlock(&dentry
->d_lock
);
513 br_read_unlock(vfsmount_lock
);
516 BUG_ON(nd
->inode
!= dentry
->d_inode
);
517 spin_unlock(&dentry
->d_lock
);
518 mntget(nd
->path
.mnt
);
520 br_read_unlock(vfsmount_lock
);
523 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
526 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
529 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
532 /* Note: we do not d_invalidate() */
533 status
= d_revalidate(dentry
, nd
);
544 static __always_inline
void set_root(struct nameidata
*nd
)
547 get_fs_root(current
->fs
, &nd
->root
);
550 static int link_path_walk(const char *, struct nameidata
*);
552 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
555 struct fs_struct
*fs
= current
->fs
;
559 seq
= read_seqcount_begin(&fs
->seq
);
561 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
562 } while (read_seqcount_retry(&fs
->seq
, seq
));
566 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
578 nd
->flags
|= LOOKUP_JUMPED
;
580 nd
->inode
= nd
->path
.dentry
->d_inode
;
582 ret
= link_path_walk(link
, nd
);
586 return PTR_ERR(link
);
589 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
592 if (path
->mnt
!= nd
->path
.mnt
)
596 static inline void path_to_nameidata(const struct path
*path
,
597 struct nameidata
*nd
)
599 if (!(nd
->flags
& LOOKUP_RCU
)) {
600 dput(nd
->path
.dentry
);
601 if (nd
->path
.mnt
!= path
->mnt
)
602 mntput(nd
->path
.mnt
);
604 nd
->path
.mnt
= path
->mnt
;
605 nd
->path
.dentry
= path
->dentry
;
608 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
610 struct inode
*inode
= link
->dentry
->d_inode
;
611 if (!IS_ERR(cookie
) && inode
->i_op
->put_link
)
612 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
616 static __always_inline
int
617 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
620 struct dentry
*dentry
= link
->dentry
;
622 BUG_ON(nd
->flags
& LOOKUP_RCU
);
624 if (link
->mnt
== nd
->path
.mnt
)
627 if (unlikely(current
->total_link_count
>= 40)) {
628 *p
= ERR_PTR(-ELOOP
); /* no ->put_link(), please */
633 current
->total_link_count
++;
636 nd_set_link(nd
, NULL
);
638 error
= security_inode_follow_link(link
->dentry
, nd
);
640 *p
= ERR_PTR(error
); /* no ->put_link(), please */
645 nd
->last_type
= LAST_BIND
;
646 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
649 char *s
= nd_get_link(nd
);
652 error
= __vfs_follow_link(nd
, s
);
653 else if (nd
->last_type
== LAST_BIND
) {
654 nd
->flags
|= LOOKUP_JUMPED
;
655 nd
->inode
= nd
->path
.dentry
->d_inode
;
656 if (nd
->inode
->i_op
->follow_link
) {
657 /* stepped on a _really_ weird one */
666 static int follow_up_rcu(struct path
*path
)
668 struct mount
*mnt
= real_mount(path
->mnt
);
669 struct mount
*parent
;
670 struct dentry
*mountpoint
;
672 parent
= mnt
->mnt_parent
;
673 if (&parent
->mnt
== path
->mnt
)
675 mountpoint
= mnt
->mnt_mountpoint
;
676 path
->dentry
= mountpoint
;
677 path
->mnt
= &parent
->mnt
;
681 int follow_up(struct path
*path
)
683 struct mount
*mnt
= real_mount(path
->mnt
);
684 struct mount
*parent
;
685 struct dentry
*mountpoint
;
687 br_read_lock(vfsmount_lock
);
688 parent
= mnt
->mnt_parent
;
689 if (&parent
->mnt
== path
->mnt
) {
690 br_read_unlock(vfsmount_lock
);
693 mntget(&parent
->mnt
);
694 mountpoint
= dget(mnt
->mnt_mountpoint
);
695 br_read_unlock(vfsmount_lock
);
697 path
->dentry
= mountpoint
;
699 path
->mnt
= &parent
->mnt
;
704 * Perform an automount
705 * - return -EISDIR to tell follow_managed() to stop and return the path we
708 static int follow_automount(struct path
*path
, unsigned flags
,
711 struct vfsmount
*mnt
;
714 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
717 /* We don't want to mount if someone's just doing a stat -
718 * unless they're stat'ing a directory and appended a '/' to
721 * We do, however, want to mount if someone wants to open or
722 * create a file of any type under the mountpoint, wants to
723 * traverse through the mountpoint or wants to open the
724 * mounted directory. Also, autofs may mark negative dentries
725 * as being automount points. These will need the attentions
726 * of the daemon to instantiate them before they can be used.
728 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
729 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
730 path
->dentry
->d_inode
)
733 current
->total_link_count
++;
734 if (current
->total_link_count
>= 40)
737 mnt
= path
->dentry
->d_op
->d_automount(path
);
740 * The filesystem is allowed to return -EISDIR here to indicate
741 * it doesn't want to automount. For instance, autofs would do
742 * this so that its userspace daemon can mount on this dentry.
744 * However, we can only permit this if it's a terminal point in
745 * the path being looked up; if it wasn't then the remainder of
746 * the path is inaccessible and we should say so.
748 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
753 if (!mnt
) /* mount collision */
757 /* lock_mount() may release path->mnt on error */
761 err
= finish_automount(mnt
, path
);
765 /* Someone else made a mount here whilst we were busy */
770 path
->dentry
= dget(mnt
->mnt_root
);
779 * Handle a dentry that is managed in some way.
780 * - Flagged for transit management (autofs)
781 * - Flagged as mountpoint
782 * - Flagged as automount point
784 * This may only be called in refwalk mode.
786 * Serialization is taken care of in namespace.c
788 static int follow_managed(struct path
*path
, unsigned flags
)
790 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
792 bool need_mntput
= false;
795 /* Given that we're not holding a lock here, we retain the value in a
796 * local variable for each dentry as we look at it so that we don't see
797 * the components of that value change under us */
798 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
799 managed
&= DCACHE_MANAGED_DENTRY
,
800 unlikely(managed
!= 0)) {
801 /* Allow the filesystem to manage the transit without i_mutex
803 if (managed
& DCACHE_MANAGE_TRANSIT
) {
804 BUG_ON(!path
->dentry
->d_op
);
805 BUG_ON(!path
->dentry
->d_op
->d_manage
);
806 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
811 /* Transit to a mounted filesystem. */
812 if (managed
& DCACHE_MOUNTED
) {
813 struct vfsmount
*mounted
= lookup_mnt(path
);
819 path
->dentry
= dget(mounted
->mnt_root
);
824 /* Something is mounted on this dentry in another
825 * namespace and/or whatever was mounted there in this
826 * namespace got unmounted before we managed to get the
830 /* Handle an automount point */
831 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
832 ret
= follow_automount(path
, flags
, &need_mntput
);
838 /* We didn't change the current path point */
842 if (need_mntput
&& path
->mnt
== mnt
)
846 return ret
< 0 ? ret
: need_mntput
;
849 int follow_down_one(struct path
*path
)
851 struct vfsmount
*mounted
;
853 mounted
= lookup_mnt(path
);
858 path
->dentry
= dget(mounted
->mnt_root
);
864 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
866 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
867 dentry
->d_op
->d_manage(dentry
, true) < 0);
871 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
872 * we meet a managed dentry that would need blocking.
874 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
875 struct inode
**inode
)
878 struct mount
*mounted
;
880 * Don't forget we might have a non-mountpoint managed dentry
881 * that wants to block transit.
883 if (unlikely(managed_dentry_might_block(path
->dentry
)))
886 if (!d_mountpoint(path
->dentry
))
889 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
892 path
->mnt
= &mounted
->mnt
;
893 path
->dentry
= mounted
->mnt
.mnt_root
;
894 nd
->flags
|= LOOKUP_JUMPED
;
895 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
897 * Update the inode too. We don't need to re-check the
898 * dentry sequence number here after this d_inode read,
899 * because a mount-point is always pinned.
901 *inode
= path
->dentry
->d_inode
;
906 static void follow_mount_rcu(struct nameidata
*nd
)
908 while (d_mountpoint(nd
->path
.dentry
)) {
909 struct mount
*mounted
;
910 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
913 nd
->path
.mnt
= &mounted
->mnt
;
914 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
915 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
919 static int follow_dotdot_rcu(struct nameidata
*nd
)
924 if (nd
->path
.dentry
== nd
->root
.dentry
&&
925 nd
->path
.mnt
== nd
->root
.mnt
) {
928 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
929 struct dentry
*old
= nd
->path
.dentry
;
930 struct dentry
*parent
= old
->d_parent
;
933 seq
= read_seqcount_begin(&parent
->d_seq
);
934 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
936 nd
->path
.dentry
= parent
;
940 if (!follow_up_rcu(&nd
->path
))
942 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
944 follow_mount_rcu(nd
);
945 nd
->inode
= nd
->path
.dentry
->d_inode
;
949 nd
->flags
&= ~LOOKUP_RCU
;
950 if (!(nd
->flags
& LOOKUP_ROOT
))
953 br_read_unlock(vfsmount_lock
);
958 * Follow down to the covering mount currently visible to userspace. At each
959 * point, the filesystem owning that dentry may be queried as to whether the
960 * caller is permitted to proceed or not.
962 int follow_down(struct path
*path
)
967 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
968 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
969 /* Allow the filesystem to manage the transit without i_mutex
972 * We indicate to the filesystem if someone is trying to mount
973 * something here. This gives autofs the chance to deny anyone
974 * other than its daemon the right to mount on its
977 * The filesystem may sleep at this point.
979 if (managed
& DCACHE_MANAGE_TRANSIT
) {
980 BUG_ON(!path
->dentry
->d_op
);
981 BUG_ON(!path
->dentry
->d_op
->d_manage
);
982 ret
= path
->dentry
->d_op
->d_manage(
983 path
->dentry
, false);
985 return ret
== -EISDIR
? 0 : ret
;
988 /* Transit to a mounted filesystem. */
989 if (managed
& DCACHE_MOUNTED
) {
990 struct vfsmount
*mounted
= lookup_mnt(path
);
996 path
->dentry
= dget(mounted
->mnt_root
);
1000 /* Don't handle automount points here */
1007 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1009 static void follow_mount(struct path
*path
)
1011 while (d_mountpoint(path
->dentry
)) {
1012 struct vfsmount
*mounted
= lookup_mnt(path
);
1017 path
->mnt
= mounted
;
1018 path
->dentry
= dget(mounted
->mnt_root
);
1022 static void follow_dotdot(struct nameidata
*nd
)
1027 struct dentry
*old
= nd
->path
.dentry
;
1029 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1030 nd
->path
.mnt
== nd
->root
.mnt
) {
1033 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1034 /* rare case of legitimate dget_parent()... */
1035 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1039 if (!follow_up(&nd
->path
))
1042 follow_mount(&nd
->path
);
1043 nd
->inode
= nd
->path
.dentry
->d_inode
;
1047 * This looks up the name in dcache, possibly revalidates the old dentry and
1048 * allocates a new one if not found or not valid. In the need_lookup argument
1049 * returns whether i_op->lookup is necessary.
1051 * dir->d_inode->i_mutex must be held
1053 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1054 struct nameidata
*nd
, bool *need_lookup
)
1056 struct dentry
*dentry
;
1059 *need_lookup
= false;
1060 dentry
= d_lookup(dir
, name
);
1062 if (d_need_lookup(dentry
)) {
1063 *need_lookup
= true;
1064 } else if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1065 error
= d_revalidate(dentry
, nd
);
1066 if (unlikely(error
<= 0)) {
1069 return ERR_PTR(error
);
1070 } else if (!d_invalidate(dentry
)) {
1079 dentry
= d_alloc(dir
, name
);
1080 if (unlikely(!dentry
))
1081 return ERR_PTR(-ENOMEM
);
1083 *need_lookup
= true;
1089 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1090 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1092 * dir->d_inode->i_mutex must be held
1094 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1095 struct nameidata
*nd
)
1099 /* Don't create child dentry for a dead directory. */
1100 if (unlikely(IS_DEADDIR(dir
))) {
1102 return ERR_PTR(-ENOENT
);
1105 old
= dir
->i_op
->lookup(dir
, dentry
, nd
);
1106 if (unlikely(old
)) {
1113 static struct dentry
*__lookup_hash(struct qstr
*name
,
1114 struct dentry
*base
, struct nameidata
*nd
)
1117 struct dentry
*dentry
;
1119 dentry
= lookup_dcache(name
, base
, nd
, &need_lookup
);
1123 return lookup_real(base
->d_inode
, dentry
, nd
);
1127 * It's more convoluted than I'd like it to be, but... it's still fairly
1128 * small and for now I'd prefer to have fast path as straight as possible.
1129 * It _is_ time-critical.
1131 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1132 struct path
*path
, struct inode
**inode
)
1134 struct vfsmount
*mnt
= nd
->path
.mnt
;
1135 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1141 * Rename seqlock is not required here because in the off chance
1142 * of a false negative due to a concurrent rename, we're going to
1143 * do the non-racy lookup, below.
1145 if (nd
->flags
& LOOKUP_RCU
) {
1147 dentry
= __d_lookup_rcu(parent
, name
, &seq
, nd
->inode
);
1152 * This sequence count validates that the inode matches
1153 * the dentry name information from lookup.
1155 *inode
= dentry
->d_inode
;
1156 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1160 * This sequence count validates that the parent had no
1161 * changes while we did the lookup of the dentry above.
1163 * The memory barrier in read_seqcount_begin of child is
1164 * enough, we can use __read_seqcount_retry here.
1166 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1170 if (unlikely(d_need_lookup(dentry
)))
1172 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1173 status
= d_revalidate(dentry
, nd
);
1174 if (unlikely(status
<= 0)) {
1175 if (status
!= -ECHILD
)
1181 path
->dentry
= dentry
;
1182 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1184 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1188 if (unlazy_walk(nd
, dentry
))
1191 dentry
= __d_lookup(parent
, name
);
1194 if (unlikely(!dentry
))
1197 if (unlikely(d_need_lookup(dentry
))) {
1202 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1203 status
= d_revalidate(dentry
, nd
);
1204 if (unlikely(status
<= 0)) {
1209 if (!d_invalidate(dentry
)) {
1216 path
->dentry
= dentry
;
1217 err
= follow_managed(path
, nd
->flags
);
1218 if (unlikely(err
< 0)) {
1219 path_put_conditional(path
, nd
);
1223 nd
->flags
|= LOOKUP_JUMPED
;
1224 *inode
= path
->dentry
->d_inode
;
1228 BUG_ON(nd
->inode
!= parent
->d_inode
);
1230 mutex_lock(&parent
->d_inode
->i_mutex
);
1231 dentry
= __lookup_hash(name
, parent
, nd
);
1232 mutex_unlock(&parent
->d_inode
->i_mutex
);
1234 return PTR_ERR(dentry
);
1238 static inline int may_lookup(struct nameidata
*nd
)
1240 if (nd
->flags
& LOOKUP_RCU
) {
1241 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1244 if (unlazy_walk(nd
, NULL
))
1247 return inode_permission(nd
->inode
, MAY_EXEC
);
1250 static inline int handle_dots(struct nameidata
*nd
, int type
)
1252 if (type
== LAST_DOTDOT
) {
1253 if (nd
->flags
& LOOKUP_RCU
) {
1254 if (follow_dotdot_rcu(nd
))
1262 static void terminate_walk(struct nameidata
*nd
)
1264 if (!(nd
->flags
& LOOKUP_RCU
)) {
1265 path_put(&nd
->path
);
1267 nd
->flags
&= ~LOOKUP_RCU
;
1268 if (!(nd
->flags
& LOOKUP_ROOT
))
1269 nd
->root
.mnt
= NULL
;
1271 br_read_unlock(vfsmount_lock
);
1276 * Do we need to follow links? We _really_ want to be able
1277 * to do this check without having to look at inode->i_op,
1278 * so we keep a cache of "no, this doesn't need follow_link"
1279 * for the common case.
1281 static inline int should_follow_link(struct inode
*inode
, int follow
)
1283 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1284 if (likely(inode
->i_op
->follow_link
))
1287 /* This gets set once for the inode lifetime */
1288 spin_lock(&inode
->i_lock
);
1289 inode
->i_opflags
|= IOP_NOFOLLOW
;
1290 spin_unlock(&inode
->i_lock
);
1295 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1296 struct qstr
*name
, int type
, int follow
)
1298 struct inode
*inode
;
1301 * "." and ".." are special - ".." especially so because it has
1302 * to be able to know about the current root directory and
1303 * parent relationships.
1305 if (unlikely(type
!= LAST_NORM
))
1306 return handle_dots(nd
, type
);
1307 err
= do_lookup(nd
, name
, path
, &inode
);
1308 if (unlikely(err
)) {
1313 path_to_nameidata(path
, nd
);
1317 if (should_follow_link(inode
, follow
)) {
1318 if (nd
->flags
& LOOKUP_RCU
) {
1319 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1324 BUG_ON(inode
!= path
->dentry
->d_inode
);
1327 path_to_nameidata(path
, nd
);
1333 * This limits recursive symlink follows to 8, while
1334 * limiting consecutive symlinks to 40.
1336 * Without that kind of total limit, nasty chains of consecutive
1337 * symlinks can cause almost arbitrarily long lookups.
1339 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1343 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1344 path_put_conditional(path
, nd
);
1345 path_put(&nd
->path
);
1348 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1351 current
->link_count
++;
1354 struct path link
= *path
;
1357 res
= follow_link(&link
, nd
, &cookie
);
1359 res
= walk_component(nd
, path
, &nd
->last
,
1360 nd
->last_type
, LOOKUP_FOLLOW
);
1361 put_link(nd
, &link
, cookie
);
1364 current
->link_count
--;
1370 * We really don't want to look at inode->i_op->lookup
1371 * when we don't have to. So we keep a cache bit in
1372 * the inode ->i_opflags field that says "yes, we can
1373 * do lookup on this inode".
1375 static inline int can_lookup(struct inode
*inode
)
1377 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1379 if (likely(!inode
->i_op
->lookup
))
1382 /* We do this once for the lifetime of the inode */
1383 spin_lock(&inode
->i_lock
);
1384 inode
->i_opflags
|= IOP_LOOKUP
;
1385 spin_unlock(&inode
->i_lock
);
1390 * We can do the critical dentry name comparison and hashing
1391 * operations one word at a time, but we are limited to:
1393 * - Architectures with fast unaligned word accesses. We could
1394 * do a "get_unaligned()" if this helps and is sufficiently
1397 * - Little-endian machines (so that we can generate the mask
1398 * of low bytes efficiently). Again, we *could* do a byte
1399 * swapping load on big-endian architectures if that is not
1400 * expensive enough to make the optimization worthless.
1402 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1403 * do not trap on the (extremely unlikely) case of a page
1404 * crossing operation.
1406 * - Furthermore, we need an efficient 64-bit compile for the
1407 * 64-bit case in order to generate the "number of bytes in
1408 * the final mask". Again, that could be replaced with a
1409 * efficient population count instruction or similar.
1411 #ifdef CONFIG_DCACHE_WORD_ACCESS
1413 #include <asm/word-at-a-time.h>
1417 static inline unsigned int fold_hash(unsigned long hash
)
1419 hash
+= hash
>> (8*sizeof(int));
1423 #else /* 32-bit case */
1425 #define fold_hash(x) (x)
1429 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1431 unsigned long a
, mask
;
1432 unsigned long hash
= 0;
1435 a
= load_unaligned_zeropad(name
);
1436 if (len
< sizeof(unsigned long))
1440 name
+= sizeof(unsigned long);
1441 len
-= sizeof(unsigned long);
1445 mask
= ~(~0ul << len
*8);
1448 return fold_hash(hash
);
1450 EXPORT_SYMBOL(full_name_hash
);
1453 * Calculate the length and hash of the path component, and
1454 * return the length of the component;
1456 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1458 unsigned long a
, mask
, hash
, len
;
1461 len
= -sizeof(unsigned long);
1463 hash
= (hash
+ a
) * 9;
1464 len
+= sizeof(unsigned long);
1465 a
= load_unaligned_zeropad(name
+len
);
1466 /* Do we have any NUL or '/' bytes in this word? */
1467 mask
= has_zero(a
) | has_zero(a
^ REPEAT_BYTE('/'));
1470 /* The mask *below* the first high bit set */
1471 mask
= (mask
- 1) & ~mask
;
1474 *hashp
= fold_hash(hash
);
1476 return len
+ count_masked_bytes(mask
);
1481 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1483 unsigned long hash
= init_name_hash();
1485 hash
= partial_name_hash(*name
++, hash
);
1486 return end_name_hash(hash
);
1488 EXPORT_SYMBOL(full_name_hash
);
1491 * We know there's a real path component here of at least
1494 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1496 unsigned long hash
= init_name_hash();
1497 unsigned long len
= 0, c
;
1499 c
= (unsigned char)*name
;
1502 hash
= partial_name_hash(c
, hash
);
1503 c
= (unsigned char)name
[len
];
1504 } while (c
&& c
!= '/');
1505 *hashp
= end_name_hash(hash
);
1513 * This is the basic name resolution function, turning a pathname into
1514 * the final dentry. We expect 'base' to be positive and a directory.
1516 * Returns 0 and nd will have valid dentry and mnt on success.
1517 * Returns error and drops reference to input namei data on failure.
1519 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1529 /* At this point we know we have a real path component. */
1535 err
= may_lookup(nd
);
1539 len
= hash_name(name
, &this.hash
);
1544 if (name
[0] == '.') switch (len
) {
1546 if (name
[1] == '.') {
1548 nd
->flags
|= LOOKUP_JUMPED
;
1554 if (likely(type
== LAST_NORM
)) {
1555 struct dentry
*parent
= nd
->path
.dentry
;
1556 nd
->flags
&= ~LOOKUP_JUMPED
;
1557 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1558 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1566 goto last_component
;
1568 * If it wasn't NUL, we know it was '/'. Skip that
1569 * slash, and continue until no more slashes.
1573 } while (unlikely(name
[len
] == '/'));
1575 goto last_component
;
1578 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1583 err
= nested_symlink(&next
, nd
);
1587 if (can_lookup(nd
->inode
))
1591 /* here ends the main loop */
1595 nd
->last_type
= type
;
1602 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1603 struct nameidata
*nd
, struct file
**fp
)
1609 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1610 nd
->flags
= flags
| LOOKUP_JUMPED
;
1612 if (flags
& LOOKUP_ROOT
) {
1613 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1615 if (!inode
->i_op
->lookup
)
1617 retval
= inode_permission(inode
, MAY_EXEC
);
1621 nd
->path
= nd
->root
;
1623 if (flags
& LOOKUP_RCU
) {
1624 br_read_lock(vfsmount_lock
);
1626 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1628 path_get(&nd
->path
);
1633 nd
->root
.mnt
= NULL
;
1636 if (flags
& LOOKUP_RCU
) {
1637 br_read_lock(vfsmount_lock
);
1642 path_get(&nd
->root
);
1644 nd
->path
= nd
->root
;
1645 } else if (dfd
== AT_FDCWD
) {
1646 if (flags
& LOOKUP_RCU
) {
1647 struct fs_struct
*fs
= current
->fs
;
1650 br_read_lock(vfsmount_lock
);
1654 seq
= read_seqcount_begin(&fs
->seq
);
1656 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1657 } while (read_seqcount_retry(&fs
->seq
, seq
));
1659 get_fs_pwd(current
->fs
, &nd
->path
);
1662 struct dentry
*dentry
;
1664 file
= fget_raw_light(dfd
, &fput_needed
);
1669 dentry
= file
->f_path
.dentry
;
1673 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1676 retval
= inode_permission(dentry
->d_inode
, MAY_EXEC
);
1681 nd
->path
= file
->f_path
;
1682 if (flags
& LOOKUP_RCU
) {
1685 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1686 br_read_lock(vfsmount_lock
);
1689 path_get(&file
->f_path
);
1690 fput_light(file
, fput_needed
);
1694 nd
->inode
= nd
->path
.dentry
->d_inode
;
1698 fput_light(file
, fput_needed
);
1703 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1705 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1706 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1708 nd
->flags
&= ~LOOKUP_PARENT
;
1709 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1710 nd
->flags
& LOOKUP_FOLLOW
);
1713 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1714 static int path_lookupat(int dfd
, const char *name
,
1715 unsigned int flags
, struct nameidata
*nd
)
1717 struct file
*base
= NULL
;
1722 * Path walking is largely split up into 2 different synchronisation
1723 * schemes, rcu-walk and ref-walk (explained in
1724 * Documentation/filesystems/path-lookup.txt). These share much of the
1725 * path walk code, but some things particularly setup, cleanup, and
1726 * following mounts are sufficiently divergent that functions are
1727 * duplicated. Typically there is a function foo(), and its RCU
1728 * analogue, foo_rcu().
1730 * -ECHILD is the error number of choice (just to avoid clashes) that
1731 * is returned if some aspect of an rcu-walk fails. Such an error must
1732 * be handled by restarting a traditional ref-walk (which will always
1733 * be able to complete).
1735 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1740 current
->total_link_count
= 0;
1741 err
= link_path_walk(name
, nd
);
1743 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1744 err
= lookup_last(nd
, &path
);
1747 struct path link
= path
;
1748 nd
->flags
|= LOOKUP_PARENT
;
1749 err
= follow_link(&link
, nd
, &cookie
);
1751 err
= lookup_last(nd
, &path
);
1752 put_link(nd
, &link
, cookie
);
1757 err
= complete_walk(nd
);
1759 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1760 if (!nd
->inode
->i_op
->lookup
) {
1761 path_put(&nd
->path
);
1769 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1770 path_put(&nd
->root
);
1771 nd
->root
.mnt
= NULL
;
1776 static int do_path_lookup(int dfd
, const char *name
,
1777 unsigned int flags
, struct nameidata
*nd
)
1779 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1780 if (unlikely(retval
== -ECHILD
))
1781 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1782 if (unlikely(retval
== -ESTALE
))
1783 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1785 if (likely(!retval
)) {
1786 if (unlikely(!audit_dummy_context())) {
1787 if (nd
->path
.dentry
&& nd
->inode
)
1788 audit_inode(name
, nd
->path
.dentry
);
1794 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1796 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1799 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1801 struct nameidata nd
;
1802 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1809 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1810 * @dentry: pointer to dentry of the base directory
1811 * @mnt: pointer to vfs mount of the base directory
1812 * @name: pointer to file name
1813 * @flags: lookup flags
1814 * @path: pointer to struct path to fill
1816 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1817 const char *name
, unsigned int flags
,
1820 struct nameidata nd
;
1822 nd
.root
.dentry
= dentry
;
1824 BUG_ON(flags
& LOOKUP_PARENT
);
1825 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1826 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
1833 * Restricted form of lookup. Doesn't follow links, single-component only,
1834 * needs parent already locked. Doesn't follow mounts.
1837 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1839 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1843 * lookup_one_len - filesystem helper to lookup single pathname component
1844 * @name: pathname component to lookup
1845 * @base: base directory to lookup from
1846 * @len: maximum length @len should be interpreted to
1848 * Note that this routine is purely a helper for filesystem usage and should
1849 * not be called by generic code. Also note that by using this function the
1850 * nameidata argument is passed to the filesystem methods and a filesystem
1851 * using this helper needs to be prepared for that.
1853 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1859 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1863 this.hash
= full_name_hash(name
, len
);
1865 return ERR_PTR(-EACCES
);
1868 c
= *(const unsigned char *)name
++;
1869 if (c
== '/' || c
== '\0')
1870 return ERR_PTR(-EACCES
);
1873 * See if the low-level filesystem might want
1874 * to use its own hash..
1876 if (base
->d_flags
& DCACHE_OP_HASH
) {
1877 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1879 return ERR_PTR(err
);
1882 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
1884 return ERR_PTR(err
);
1886 return __lookup_hash(&this, base
, NULL
);
1889 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
1890 struct path
*path
, int *empty
)
1892 struct nameidata nd
;
1893 char *tmp
= getname_flags(name
, flags
, empty
);
1894 int err
= PTR_ERR(tmp
);
1897 BUG_ON(flags
& LOOKUP_PARENT
);
1899 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1907 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1910 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
1913 static int user_path_parent(int dfd
, const char __user
*path
,
1914 struct nameidata
*nd
, char **name
)
1916 char *s
= getname(path
);
1922 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1932 * It's inline, so penalty for filesystems that don't use sticky bit is
1935 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1937 uid_t fsuid
= current_fsuid();
1939 if (!(dir
->i_mode
& S_ISVTX
))
1941 if (current_user_ns() != inode_userns(inode
))
1943 if (inode
->i_uid
== fsuid
)
1945 if (dir
->i_uid
== fsuid
)
1949 return !ns_capable(inode_userns(inode
), CAP_FOWNER
);
1953 * Check whether we can remove a link victim from directory dir, check
1954 * whether the type of victim is right.
1955 * 1. We can't do it if dir is read-only (done in permission())
1956 * 2. We should have write and exec permissions on dir
1957 * 3. We can't remove anything from append-only dir
1958 * 4. We can't do anything with immutable dir (done in permission())
1959 * 5. If the sticky bit on dir is set we should either
1960 * a. be owner of dir, or
1961 * b. be owner of victim, or
1962 * c. have CAP_FOWNER capability
1963 * 6. If the victim is append-only or immutable we can't do antyhing with
1964 * links pointing to it.
1965 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1966 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1967 * 9. We can't remove a root or mountpoint.
1968 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1969 * nfs_async_unlink().
1971 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1975 if (!victim
->d_inode
)
1978 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1979 audit_inode_child(victim
, dir
);
1981 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1986 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1987 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1990 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1992 if (IS_ROOT(victim
))
1994 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1996 if (IS_DEADDIR(dir
))
1998 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2003 /* Check whether we can create an object with dentry child in directory
2005 * 1. We can't do it if child already exists (open has special treatment for
2006 * this case, but since we are inlined it's OK)
2007 * 2. We can't do it if dir is read-only (done in permission())
2008 * 3. We should have write and exec permissions on dir
2009 * 4. We can't do it if dir is immutable (done in permission())
2011 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2015 if (IS_DEADDIR(dir
))
2017 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2021 * p1 and p2 should be directories on the same fs.
2023 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2028 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2032 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2034 p
= d_ancestor(p2
, p1
);
2036 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2037 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2041 p
= d_ancestor(p1
, p2
);
2043 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2044 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2048 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2049 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2053 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2055 mutex_unlock(&p1
->d_inode
->i_mutex
);
2057 mutex_unlock(&p2
->d_inode
->i_mutex
);
2058 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2062 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2063 struct nameidata
*nd
)
2065 int error
= may_create(dir
, dentry
);
2070 if (!dir
->i_op
->create
)
2071 return -EACCES
; /* shouldn't it be ENOSYS? */
2074 error
= security_inode_create(dir
, dentry
, mode
);
2077 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
2079 fsnotify_create(dir
, dentry
);
2083 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2085 struct dentry
*dentry
= path
->dentry
;
2086 struct inode
*inode
= dentry
->d_inode
;
2096 switch (inode
->i_mode
& S_IFMT
) {
2100 if (acc_mode
& MAY_WRITE
)
2105 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2114 error
= inode_permission(inode
, acc_mode
);
2119 * An append-only file must be opened in append mode for writing.
2121 if (IS_APPEND(inode
)) {
2122 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2128 /* O_NOATIME can only be set by the owner or superuser */
2129 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2135 static int handle_truncate(struct file
*filp
)
2137 struct path
*path
= &filp
->f_path
;
2138 struct inode
*inode
= path
->dentry
->d_inode
;
2139 int error
= get_write_access(inode
);
2143 * Refuse to truncate files with mandatory locks held on them.
2145 error
= locks_verify_locked(inode
);
2147 error
= security_path_truncate(path
);
2149 error
= do_truncate(path
->dentry
, 0,
2150 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2153 put_write_access(inode
);
2157 static inline int open_to_namei_flags(int flag
)
2159 if ((flag
& O_ACCMODE
) == 3)
2165 * Handle the last step of open()
2167 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2168 const struct open_flags
*op
, const char *pathname
)
2170 struct dentry
*dir
= nd
->path
.dentry
;
2171 struct dentry
*dentry
;
2172 int open_flag
= op
->open_flag
;
2173 int will_truncate
= open_flag
& O_TRUNC
;
2175 int acc_mode
= op
->acc_mode
;
2179 nd
->flags
&= ~LOOKUP_PARENT
;
2180 nd
->flags
|= op
->intent
;
2182 switch (nd
->last_type
) {
2185 error
= handle_dots(nd
, nd
->last_type
);
2187 return ERR_PTR(error
);
2190 error
= complete_walk(nd
);
2192 return ERR_PTR(error
);
2193 audit_inode(pathname
, nd
->path
.dentry
);
2194 if (open_flag
& O_CREAT
) {
2200 error
= complete_walk(nd
);
2202 return ERR_PTR(error
);
2203 audit_inode(pathname
, dir
);
2207 if (!(open_flag
& O_CREAT
)) {
2209 if (nd
->last
.name
[nd
->last
.len
])
2210 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2211 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2213 /* we _can_ be in RCU mode here */
2214 error
= walk_component(nd
, path
, &nd
->last
, LAST_NORM
,
2217 return ERR_PTR(error
);
2218 if (error
) /* symlink */
2221 error
= complete_walk(nd
);
2223 return ERR_PTR(error
);
2226 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2227 if (!nd
->inode
->i_op
->lookup
)
2230 audit_inode(pathname
, nd
->path
.dentry
);
2234 /* create side of things */
2236 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2237 * cleared when we got to the last component we are about to look up
2239 error
= complete_walk(nd
);
2241 return ERR_PTR(error
);
2243 audit_inode(pathname
, dir
);
2245 /* trailing slashes? */
2246 if (nd
->last
.name
[nd
->last
.len
])
2249 mutex_lock(&dir
->d_inode
->i_mutex
);
2251 dentry
= lookup_hash(nd
);
2252 error
= PTR_ERR(dentry
);
2253 if (IS_ERR(dentry
)) {
2254 mutex_unlock(&dir
->d_inode
->i_mutex
);
2258 path
->dentry
= dentry
;
2259 path
->mnt
= nd
->path
.mnt
;
2261 /* Negative dentry, just create the file */
2262 if (!dentry
->d_inode
) {
2263 umode_t mode
= op
->mode
;
2264 if (!IS_POSIXACL(dir
->d_inode
))
2265 mode
&= ~current_umask();
2267 * This write is needed to ensure that a
2268 * rw->ro transition does not occur between
2269 * the time when the file is created and when
2270 * a permanent write count is taken through
2271 * the 'struct file' in nameidata_to_filp().
2273 error
= mnt_want_write(nd
->path
.mnt
);
2275 goto exit_mutex_unlock
;
2277 /* Don't check for write permission, don't truncate */
2278 open_flag
&= ~O_TRUNC
;
2280 acc_mode
= MAY_OPEN
;
2281 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2283 goto exit_mutex_unlock
;
2284 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2286 goto exit_mutex_unlock
;
2287 mutex_unlock(&dir
->d_inode
->i_mutex
);
2288 dput(nd
->path
.dentry
);
2289 nd
->path
.dentry
= dentry
;
2294 * It already exists.
2296 mutex_unlock(&dir
->d_inode
->i_mutex
);
2297 audit_inode(pathname
, path
->dentry
);
2300 if (open_flag
& O_EXCL
)
2303 error
= follow_managed(path
, nd
->flags
);
2308 nd
->flags
|= LOOKUP_JUMPED
;
2311 if (!path
->dentry
->d_inode
)
2314 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2317 path_to_nameidata(path
, nd
);
2318 nd
->inode
= path
->dentry
->d_inode
;
2319 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2320 error
= complete_walk(nd
);
2322 return ERR_PTR(error
);
2324 if (S_ISDIR(nd
->inode
->i_mode
))
2327 if (!S_ISREG(nd
->inode
->i_mode
))
2330 if (will_truncate
) {
2331 error
= mnt_want_write(nd
->path
.mnt
);
2337 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2340 filp
= nameidata_to_filp(nd
);
2341 if (!IS_ERR(filp
)) {
2342 error
= ima_file_check(filp
, op
->acc_mode
);
2345 filp
= ERR_PTR(error
);
2348 if (!IS_ERR(filp
)) {
2349 if (will_truncate
) {
2350 error
= handle_truncate(filp
);
2353 filp
= ERR_PTR(error
);
2359 mnt_drop_write(nd
->path
.mnt
);
2360 path_put(&nd
->path
);
2364 mutex_unlock(&dir
->d_inode
->i_mutex
);
2366 path_put_conditional(path
, nd
);
2368 filp
= ERR_PTR(error
);
2372 static struct file
*path_openat(int dfd
, const char *pathname
,
2373 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2375 struct file
*base
= NULL
;
2380 filp
= get_empty_filp();
2382 return ERR_PTR(-ENFILE
);
2384 filp
->f_flags
= op
->open_flag
;
2385 nd
->intent
.open
.file
= filp
;
2386 nd
->intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2387 nd
->intent
.open
.create_mode
= op
->mode
;
2389 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2390 if (unlikely(error
))
2393 current
->total_link_count
= 0;
2394 error
= link_path_walk(pathname
, nd
);
2395 if (unlikely(error
))
2398 filp
= do_last(nd
, &path
, op
, pathname
);
2399 while (unlikely(!filp
)) { /* trailing symlink */
2400 struct path link
= path
;
2402 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2403 path_put_conditional(&path
, nd
);
2404 path_put(&nd
->path
);
2405 filp
= ERR_PTR(-ELOOP
);
2408 nd
->flags
|= LOOKUP_PARENT
;
2409 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2410 error
= follow_link(&link
, nd
, &cookie
);
2411 if (unlikely(error
))
2412 filp
= ERR_PTR(error
);
2414 filp
= do_last(nd
, &path
, op
, pathname
);
2415 put_link(nd
, &link
, cookie
);
2418 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2419 path_put(&nd
->root
);
2422 release_open_intent(nd
);
2426 filp
= ERR_PTR(error
);
2430 struct file
*do_filp_open(int dfd
, const char *pathname
,
2431 const struct open_flags
*op
, int flags
)
2433 struct nameidata nd
;
2436 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2437 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2438 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2439 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2440 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2444 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2445 const char *name
, const struct open_flags
*op
, int flags
)
2447 struct nameidata nd
;
2451 nd
.root
.dentry
= dentry
;
2453 flags
|= LOOKUP_ROOT
;
2455 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2456 return ERR_PTR(-ELOOP
);
2458 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2459 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2460 file
= path_openat(-1, name
, &nd
, op
, flags
);
2461 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2462 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2466 struct dentry
*kern_path_create(int dfd
, const char *pathname
, struct path
*path
, int is_dir
)
2468 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2469 struct nameidata nd
;
2470 int error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
, &nd
);
2472 return ERR_PTR(error
);
2475 * Yucky last component or no last component at all?
2476 * (foo/., foo/.., /////)
2478 if (nd
.last_type
!= LAST_NORM
)
2480 nd
.flags
&= ~LOOKUP_PARENT
;
2481 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2482 nd
.intent
.open
.flags
= O_EXCL
;
2485 * Do the final lookup.
2487 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2488 dentry
= lookup_hash(&nd
);
2492 if (dentry
->d_inode
)
2495 * Special case - lookup gave negative, but... we had foo/bar/
2496 * From the vfs_mknod() POV we just have a negative dentry -
2497 * all is fine. Let's be bastards - you had / on the end, you've
2498 * been asking for (non-existent) directory. -ENOENT for you.
2500 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
2502 dentry
= ERR_PTR(-ENOENT
);
2509 dentry
= ERR_PTR(-EEXIST
);
2511 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2516 EXPORT_SYMBOL(kern_path_create
);
2518 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
, struct path
*path
, int is_dir
)
2520 char *tmp
= getname(pathname
);
2523 return ERR_CAST(tmp
);
2524 res
= kern_path_create(dfd
, tmp
, path
, is_dir
);
2528 EXPORT_SYMBOL(user_path_create
);
2530 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2532 int error
= may_create(dir
, dentry
);
2537 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) &&
2538 !ns_capable(inode_userns(dir
), CAP_MKNOD
))
2541 if (!dir
->i_op
->mknod
)
2544 error
= devcgroup_inode_mknod(mode
, dev
);
2548 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2552 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2554 fsnotify_create(dir
, dentry
);
2558 static int may_mknod(umode_t mode
)
2560 switch (mode
& S_IFMT
) {
2566 case 0: /* zero mode translates to S_IFREG */
2575 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
2578 struct dentry
*dentry
;
2585 dentry
= user_path_create(dfd
, filename
, &path
, 0);
2587 return PTR_ERR(dentry
);
2589 if (!IS_POSIXACL(path
.dentry
->d_inode
))
2590 mode
&= ~current_umask();
2591 error
= may_mknod(mode
);
2594 error
= mnt_want_write(path
.mnt
);
2597 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
2599 goto out_drop_write
;
2600 switch (mode
& S_IFMT
) {
2601 case 0: case S_IFREG
:
2602 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,NULL
);
2604 case S_IFCHR
: case S_IFBLK
:
2605 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
2606 new_decode_dev(dev
));
2608 case S_IFIFO
: case S_IFSOCK
:
2609 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
2613 mnt_drop_write(path
.mnt
);
2616 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2622 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
2624 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2627 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2629 int error
= may_create(dir
, dentry
);
2630 unsigned max_links
= dir
->i_sb
->s_max_links
;
2635 if (!dir
->i_op
->mkdir
)
2638 mode
&= (S_IRWXUGO
|S_ISVTX
);
2639 error
= security_inode_mkdir(dir
, dentry
, mode
);
2643 if (max_links
&& dir
->i_nlink
>= max_links
)
2646 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2648 fsnotify_mkdir(dir
, dentry
);
2652 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
2654 struct dentry
*dentry
;
2658 dentry
= user_path_create(dfd
, pathname
, &path
, 1);
2660 return PTR_ERR(dentry
);
2662 if (!IS_POSIXACL(path
.dentry
->d_inode
))
2663 mode
&= ~current_umask();
2664 error
= mnt_want_write(path
.mnt
);
2667 error
= security_path_mkdir(&path
, dentry
, mode
);
2669 goto out_drop_write
;
2670 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
2672 mnt_drop_write(path
.mnt
);
2675 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2680 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
2682 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2686 * The dentry_unhash() helper will try to drop the dentry early: we
2687 * should have a usage count of 1 if we're the only user of this
2688 * dentry, and if that is true (possibly after pruning the dcache),
2689 * then we drop the dentry now.
2691 * A low-level filesystem can, if it choses, legally
2694 * if (!d_unhashed(dentry))
2697 * if it cannot handle the case of removing a directory
2698 * that is still in use by something else..
2700 void dentry_unhash(struct dentry
*dentry
)
2702 shrink_dcache_parent(dentry
);
2703 spin_lock(&dentry
->d_lock
);
2704 if (dentry
->d_count
== 1)
2706 spin_unlock(&dentry
->d_lock
);
2709 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2711 int error
= may_delete(dir
, dentry
, 1);
2716 if (!dir
->i_op
->rmdir
)
2720 mutex_lock(&dentry
->d_inode
->i_mutex
);
2723 if (d_mountpoint(dentry
))
2726 error
= security_inode_rmdir(dir
, dentry
);
2730 shrink_dcache_parent(dentry
);
2731 error
= dir
->i_op
->rmdir(dir
, dentry
);
2735 dentry
->d_inode
->i_flags
|= S_DEAD
;
2739 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2746 static long do_rmdir(int dfd
, const char __user
*pathname
)
2750 struct dentry
*dentry
;
2751 struct nameidata nd
;
2753 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2757 switch(nd
.last_type
) {
2769 nd
.flags
&= ~LOOKUP_PARENT
;
2771 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2772 dentry
= lookup_hash(&nd
);
2773 error
= PTR_ERR(dentry
);
2776 if (!dentry
->d_inode
) {
2780 error
= mnt_want_write(nd
.path
.mnt
);
2783 error
= security_path_rmdir(&nd
.path
, dentry
);
2786 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2788 mnt_drop_write(nd
.path
.mnt
);
2792 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2799 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2801 return do_rmdir(AT_FDCWD
, pathname
);
2804 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2806 int error
= may_delete(dir
, dentry
, 0);
2811 if (!dir
->i_op
->unlink
)
2814 mutex_lock(&dentry
->d_inode
->i_mutex
);
2815 if (d_mountpoint(dentry
))
2818 error
= security_inode_unlink(dir
, dentry
);
2820 error
= dir
->i_op
->unlink(dir
, dentry
);
2825 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2827 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2828 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2829 fsnotify_link_count(dentry
->d_inode
);
2837 * Make sure that the actual truncation of the file will occur outside its
2838 * directory's i_mutex. Truncate can take a long time if there is a lot of
2839 * writeout happening, and we don't want to prevent access to the directory
2840 * while waiting on the I/O.
2842 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2846 struct dentry
*dentry
;
2847 struct nameidata nd
;
2848 struct inode
*inode
= NULL
;
2850 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2855 if (nd
.last_type
!= LAST_NORM
)
2858 nd
.flags
&= ~LOOKUP_PARENT
;
2860 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2861 dentry
= lookup_hash(&nd
);
2862 error
= PTR_ERR(dentry
);
2863 if (!IS_ERR(dentry
)) {
2864 /* Why not before? Because we want correct error value */
2865 if (nd
.last
.name
[nd
.last
.len
])
2867 inode
= dentry
->d_inode
;
2871 error
= mnt_want_write(nd
.path
.mnt
);
2874 error
= security_path_unlink(&nd
.path
, dentry
);
2877 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2879 mnt_drop_write(nd
.path
.mnt
);
2883 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2885 iput(inode
); /* truncate the inode here */
2892 error
= !dentry
->d_inode
? -ENOENT
:
2893 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2897 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2899 if ((flag
& ~AT_REMOVEDIR
) != 0)
2902 if (flag
& AT_REMOVEDIR
)
2903 return do_rmdir(dfd
, pathname
);
2905 return do_unlinkat(dfd
, pathname
);
2908 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2910 return do_unlinkat(AT_FDCWD
, pathname
);
2913 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2915 int error
= may_create(dir
, dentry
);
2920 if (!dir
->i_op
->symlink
)
2923 error
= security_inode_symlink(dir
, dentry
, oldname
);
2927 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2929 fsnotify_create(dir
, dentry
);
2933 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2934 int, newdfd
, const char __user
*, newname
)
2938 struct dentry
*dentry
;
2941 from
= getname(oldname
);
2943 return PTR_ERR(from
);
2945 dentry
= user_path_create(newdfd
, newname
, &path
, 0);
2946 error
= PTR_ERR(dentry
);
2950 error
= mnt_want_write(path
.mnt
);
2953 error
= security_path_symlink(&path
, dentry
, from
);
2955 goto out_drop_write
;
2956 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
);
2958 mnt_drop_write(path
.mnt
);
2961 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2968 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2970 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2973 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2975 struct inode
*inode
= old_dentry
->d_inode
;
2976 unsigned max_links
= dir
->i_sb
->s_max_links
;
2982 error
= may_create(dir
, new_dentry
);
2986 if (dir
->i_sb
!= inode
->i_sb
)
2990 * A link to an append-only or immutable file cannot be created.
2992 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2994 if (!dir
->i_op
->link
)
2996 if (S_ISDIR(inode
->i_mode
))
2999 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3003 mutex_lock(&inode
->i_mutex
);
3004 /* Make sure we don't allow creating hardlink to an unlinked file */
3005 if (inode
->i_nlink
== 0)
3007 else if (max_links
&& inode
->i_nlink
>= max_links
)
3010 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3011 mutex_unlock(&inode
->i_mutex
);
3013 fsnotify_link(dir
, inode
, new_dentry
);
3018 * Hardlinks are often used in delicate situations. We avoid
3019 * security-related surprises by not following symlinks on the
3022 * We don't follow them on the oldname either to be compatible
3023 * with linux 2.0, and to avoid hard-linking to directories
3024 * and other special files. --ADM
3026 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3027 int, newdfd
, const char __user
*, newname
, int, flags
)
3029 struct dentry
*new_dentry
;
3030 struct path old_path
, new_path
;
3034 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3037 * To use null names we require CAP_DAC_READ_SEARCH
3038 * This ensures that not everyone will be able to create
3039 * handlink using the passed filedescriptor.
3041 if (flags
& AT_EMPTY_PATH
) {
3042 if (!capable(CAP_DAC_READ_SEARCH
))
3047 if (flags
& AT_SYMLINK_FOLLOW
)
3048 how
|= LOOKUP_FOLLOW
;
3050 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3054 new_dentry
= user_path_create(newdfd
, newname
, &new_path
, 0);
3055 error
= PTR_ERR(new_dentry
);
3056 if (IS_ERR(new_dentry
))
3060 if (old_path
.mnt
!= new_path
.mnt
)
3062 error
= mnt_want_write(new_path
.mnt
);
3065 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3067 goto out_drop_write
;
3068 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3070 mnt_drop_write(new_path
.mnt
);
3073 mutex_unlock(&new_path
.dentry
->d_inode
->i_mutex
);
3074 path_put(&new_path
);
3076 path_put(&old_path
);
3081 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3083 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3087 * The worst of all namespace operations - renaming directory. "Perverted"
3088 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3090 * a) we can get into loop creation. Check is done in is_subdir().
3091 * b) race potential - two innocent renames can create a loop together.
3092 * That's where 4.4 screws up. Current fix: serialization on
3093 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3095 * c) we have to lock _three_ objects - parents and victim (if it exists).
3096 * And that - after we got ->i_mutex on parents (until then we don't know
3097 * whether the target exists). Solution: try to be smart with locking
3098 * order for inodes. We rely on the fact that tree topology may change
3099 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3100 * move will be locked. Thus we can rank directories by the tree
3101 * (ancestors first) and rank all non-directories after them.
3102 * That works since everybody except rename does "lock parent, lookup,
3103 * lock child" and rename is under ->s_vfs_rename_mutex.
3104 * HOWEVER, it relies on the assumption that any object with ->lookup()
3105 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3106 * we'd better make sure that there's no link(2) for them.
3107 * d) conversion from fhandle to dentry may come in the wrong moment - when
3108 * we are removing the target. Solution: we will have to grab ->i_mutex
3109 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3110 * ->i_mutex on parents, which works but leads to some truly excessive
3113 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3114 struct inode
*new_dir
, struct dentry
*new_dentry
)
3117 struct inode
*target
= new_dentry
->d_inode
;
3118 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3121 * If we are going to change the parent - check write permissions,
3122 * we'll need to flip '..'.
3124 if (new_dir
!= old_dir
) {
3125 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3130 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3136 mutex_lock(&target
->i_mutex
);
3139 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3143 if (max_links
&& !target
&& new_dir
!= old_dir
&&
3144 new_dir
->i_nlink
>= max_links
)
3148 shrink_dcache_parent(new_dentry
);
3149 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3154 target
->i_flags
|= S_DEAD
;
3155 dont_mount(new_dentry
);
3159 mutex_unlock(&target
->i_mutex
);
3162 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3163 d_move(old_dentry
,new_dentry
);
3167 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3168 struct inode
*new_dir
, struct dentry
*new_dentry
)
3170 struct inode
*target
= new_dentry
->d_inode
;
3173 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3179 mutex_lock(&target
->i_mutex
);
3182 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3185 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3190 dont_mount(new_dentry
);
3191 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3192 d_move(old_dentry
, new_dentry
);
3195 mutex_unlock(&target
->i_mutex
);
3200 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3201 struct inode
*new_dir
, struct dentry
*new_dentry
)
3204 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3205 const unsigned char *old_name
;
3207 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3210 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3214 if (!new_dentry
->d_inode
)
3215 error
= may_create(new_dir
, new_dentry
);
3217 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3221 if (!old_dir
->i_op
->rename
)
3224 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3227 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3229 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3231 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3232 new_dentry
->d_inode
, old_dentry
);
3233 fsnotify_oldname_free(old_name
);
3238 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3239 int, newdfd
, const char __user
*, newname
)
3241 struct dentry
*old_dir
, *new_dir
;
3242 struct dentry
*old_dentry
, *new_dentry
;
3243 struct dentry
*trap
;
3244 struct nameidata oldnd
, newnd
;
3249 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3253 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3258 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3261 old_dir
= oldnd
.path
.dentry
;
3263 if (oldnd
.last_type
!= LAST_NORM
)
3266 new_dir
= newnd
.path
.dentry
;
3267 if (newnd
.last_type
!= LAST_NORM
)
3270 oldnd
.flags
&= ~LOOKUP_PARENT
;
3271 newnd
.flags
&= ~LOOKUP_PARENT
;
3272 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3274 trap
= lock_rename(new_dir
, old_dir
);
3276 old_dentry
= lookup_hash(&oldnd
);
3277 error
= PTR_ERR(old_dentry
);
3278 if (IS_ERR(old_dentry
))
3280 /* source must exist */
3282 if (!old_dentry
->d_inode
)
3284 /* unless the source is a directory trailing slashes give -ENOTDIR */
3285 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3287 if (oldnd
.last
.name
[oldnd
.last
.len
])
3289 if (newnd
.last
.name
[newnd
.last
.len
])
3292 /* source should not be ancestor of target */
3294 if (old_dentry
== trap
)
3296 new_dentry
= lookup_hash(&newnd
);
3297 error
= PTR_ERR(new_dentry
);
3298 if (IS_ERR(new_dentry
))
3300 /* target should not be an ancestor of source */
3302 if (new_dentry
== trap
)
3305 error
= mnt_want_write(oldnd
.path
.mnt
);
3308 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3309 &newnd
.path
, new_dentry
);
3312 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3313 new_dir
->d_inode
, new_dentry
);
3315 mnt_drop_write(oldnd
.path
.mnt
);
3321 unlock_rename(new_dir
, old_dir
);
3323 path_put(&newnd
.path
);
3326 path_put(&oldnd
.path
);
3332 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3334 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3337 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3341 len
= PTR_ERR(link
);
3346 if (len
> (unsigned) buflen
)
3348 if (copy_to_user(buffer
, link
, len
))
3355 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3356 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3357 * using) it for any given inode is up to filesystem.
3359 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3361 struct nameidata nd
;
3366 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3368 return PTR_ERR(cookie
);
3370 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3371 if (dentry
->d_inode
->i_op
->put_link
)
3372 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3376 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3378 return __vfs_follow_link(nd
, link
);
3381 /* get the link contents into pagecache */
3382 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3386 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3387 page
= read_mapping_page(mapping
, 0, NULL
);
3392 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3396 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3398 struct page
*page
= NULL
;
3399 char *s
= page_getlink(dentry
, &page
);
3400 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3403 page_cache_release(page
);
3408 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3410 struct page
*page
= NULL
;
3411 nd_set_link(nd
, page_getlink(dentry
, &page
));
3415 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3417 struct page
*page
= cookie
;
3421 page_cache_release(page
);
3426 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3428 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3430 struct address_space
*mapping
= inode
->i_mapping
;
3435 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3437 flags
|= AOP_FLAG_NOFS
;
3440 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3441 flags
, &page
, &fsdata
);
3445 kaddr
= kmap_atomic(page
);
3446 memcpy(kaddr
, symname
, len
-1);
3447 kunmap_atomic(kaddr
);
3449 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3456 mark_inode_dirty(inode
);
3462 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3464 return __page_symlink(inode
, symname
, len
,
3465 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3468 const struct inode_operations page_symlink_inode_operations
= {
3469 .readlink
= generic_readlink
,
3470 .follow_link
= page_follow_link_light
,
3471 .put_link
= page_put_link
,
3474 EXPORT_SYMBOL(user_path_at
);
3475 EXPORT_SYMBOL(follow_down_one
);
3476 EXPORT_SYMBOL(follow_down
);
3477 EXPORT_SYMBOL(follow_up
);
3478 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3479 EXPORT_SYMBOL(getname
);
3480 EXPORT_SYMBOL(lock_rename
);
3481 EXPORT_SYMBOL(lookup_one_len
);
3482 EXPORT_SYMBOL(page_follow_link_light
);
3483 EXPORT_SYMBOL(page_put_link
);
3484 EXPORT_SYMBOL(page_readlink
);
3485 EXPORT_SYMBOL(__page_symlink
);
3486 EXPORT_SYMBOL(page_symlink
);
3487 EXPORT_SYMBOL(page_symlink_inode_operations
);
3488 EXPORT_SYMBOL(kern_path
);
3489 EXPORT_SYMBOL(vfs_path_lookup
);
3490 EXPORT_SYMBOL(inode_permission
);
3491 EXPORT_SYMBOL(unlock_rename
);
3492 EXPORT_SYMBOL(vfs_create
);
3493 EXPORT_SYMBOL(vfs_follow_link
);
3494 EXPORT_SYMBOL(vfs_link
);
3495 EXPORT_SYMBOL(vfs_mkdir
);
3496 EXPORT_SYMBOL(vfs_mknod
);
3497 EXPORT_SYMBOL(generic_permission
);
3498 EXPORT_SYMBOL(vfs_readlink
);
3499 EXPORT_SYMBOL(vfs_rename
);
3500 EXPORT_SYMBOL(vfs_rmdir
);
3501 EXPORT_SYMBOL(vfs_symlink
);
3502 EXPORT_SYMBOL(vfs_unlink
);
3503 EXPORT_SYMBOL(dentry_unhash
);
3504 EXPORT_SYMBOL(generic_readlink
);