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/module.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 <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existent name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user
*filename
, char *page
)
120 unsigned long len
= PATH_MAX
;
122 if (!segment_eq(get_fs(), KERNEL_DS
)) {
123 if ((unsigned long) filename
>= TASK_SIZE
)
125 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
126 len
= TASK_SIZE
- (unsigned long) filename
;
129 retval
= strncpy_from_user(page
, filename
, len
);
133 return -ENAMETOOLONG
;
139 static char *getname_flags(const char __user
* filename
, int flags
)
143 result
= ERR_PTR(-ENOMEM
);
146 int retval
= do_getname(filename
, tmp
);
150 if (retval
!= -ENOENT
|| !(flags
& LOOKUP_EMPTY
)) {
152 result
= ERR_PTR(retval
);
156 audit_getname(result
);
160 char *getname(const char __user
* filename
)
162 return getname_flags(filename
, 0);
165 #ifdef CONFIG_AUDITSYSCALL
166 void putname(const char *name
)
168 if (unlikely(!audit_dummy_context()))
173 EXPORT_SYMBOL(putname
);
177 * This does basic POSIX ACL permission checking
179 static int acl_permission_check(struct inode
*inode
, int mask
)
181 int (*check_acl
)(struct inode
*inode
, int mask
);
182 unsigned int mode
= inode
->i_mode
;
184 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
| MAY_NOT_BLOCK
;
186 if (current_user_ns() != inode_userns(inode
))
189 if (current_fsuid() == inode
->i_uid
)
192 check_acl
= inode
->i_op
->check_acl
;
193 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
194 int error
= check_acl(inode
, mask
);
195 if (error
!= -EAGAIN
)
199 if (in_group_p(inode
->i_gid
))
205 * If the DACs are ok we don't need any capability check.
207 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
213 * generic_permission - check for access rights on a Posix-like filesystem
214 * @inode: inode to check access rights for
215 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
216 * @flags: IPERM_FLAG_ flags.
218 * Used to check for read/write/execute permissions on a file.
219 * We use "fsuid" for this, letting us set arbitrary permissions
220 * for filesystem access without changing the "normal" uids which
221 * are used for other things.
223 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
224 * request cannot be satisfied (eg. requires blocking or too much complexity).
225 * It would then be called again in ref-walk mode.
227 int generic_permission(struct inode
*inode
, int mask
)
232 * Do the basic POSIX ACL permission checks.
234 ret
= acl_permission_check(inode
, mask
);
239 * Read/write DACs are always overridable.
240 * Executable DACs are overridable for all directories and
241 * for non-directories that have least one exec bit set.
243 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
244 if (ns_capable(inode_userns(inode
), CAP_DAC_OVERRIDE
))
248 * Searching includes executable on directories, else just read.
250 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
251 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
252 if (ns_capable(inode_userns(inode
), CAP_DAC_READ_SEARCH
))
259 * inode_permission - check for access rights to a given inode
260 * @inode: inode to check permission on
261 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
263 * Used to check for read/write/execute permissions on an inode.
264 * We use "fsuid" for this, letting us set arbitrary permissions
265 * for filesystem access without changing the "normal" uids which
266 * are used for other things.
268 int inode_permission(struct inode
*inode
, int mask
)
272 if (mask
& MAY_WRITE
) {
273 umode_t mode
= inode
->i_mode
;
276 * Nobody gets write access to a read-only fs.
278 if (IS_RDONLY(inode
) &&
279 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
283 * Nobody gets write access to an immutable file.
285 if (IS_IMMUTABLE(inode
))
289 if (inode
->i_op
->permission
)
290 retval
= inode
->i_op
->permission(inode
, mask
);
292 retval
= generic_permission(inode
, mask
);
297 retval
= devcgroup_inode_permission(inode
, mask
);
301 return security_inode_permission(inode
, mask
);
305 * exec_permission - check for right to do lookups in a given directory
306 * @inode: inode to check permission on
307 * @mask: MAY_EXEC and possibly MAY_NOT_BLOCK flags.
309 * Short-cut version of inode_permission(), for calling on directories
310 * during pathname resolution. Combines parts of inode_permission()
311 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
313 * If appropriate, check DAC only. If not appropriate, or
314 * short-cut DAC fails, then call ->permission() to do more
315 * complete permission check.
317 static inline int exec_permission(struct inode
*inode
, int mask
)
320 struct user_namespace
*ns
= inode_userns(inode
);
322 if (inode
->i_op
->permission
) {
323 ret
= inode
->i_op
->permission(inode
, mask
);
327 ret
= acl_permission_check(inode
, mask
);
332 if (ns_capable(ns
, CAP_DAC_OVERRIDE
) ||
333 ns_capable(ns
, CAP_DAC_READ_SEARCH
))
338 return security_inode_permission(inode
, mask
);
342 * path_get - get a reference to a path
343 * @path: path to get the reference to
345 * Given a path increment the reference count to the dentry and the vfsmount.
347 void path_get(struct path
*path
)
352 EXPORT_SYMBOL(path_get
);
355 * path_put - put a reference to a path
356 * @path: path to put the reference to
358 * Given a path decrement the reference count to the dentry and the vfsmount.
360 void path_put(struct path
*path
)
365 EXPORT_SYMBOL(path_put
);
368 * Path walking has 2 modes, rcu-walk and ref-walk (see
369 * Documentation/filesystems/path-lookup.txt). In situations when we can't
370 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
371 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
372 * mode. Refcounts are grabbed at the last known good point before rcu-walk
373 * got stuck, so ref-walk may continue from there. If this is not successful
374 * (eg. a seqcount has changed), then failure is returned and it's up to caller
375 * to restart the path walk from the beginning in ref-walk mode.
379 * unlazy_walk - try to switch to ref-walk mode.
380 * @nd: nameidata pathwalk data
381 * @dentry: child of nd->path.dentry or NULL
382 * Returns: 0 on success, -ECHILD on failure
384 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
385 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
386 * @nd or NULL. Must be called from rcu-walk context.
388 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
390 struct fs_struct
*fs
= current
->fs
;
391 struct dentry
*parent
= nd
->path
.dentry
;
394 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
395 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
397 spin_lock(&fs
->lock
);
398 if (nd
->root
.mnt
!= fs
->root
.mnt
||
399 nd
->root
.dentry
!= fs
->root
.dentry
)
402 spin_lock(&parent
->d_lock
);
404 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
406 BUG_ON(nd
->inode
!= parent
->d_inode
);
408 if (dentry
->d_parent
!= parent
)
410 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
411 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
414 * If the sequence check on the child dentry passed, then
415 * the child has not been removed from its parent. This
416 * means the parent dentry must be valid and able to take
417 * a reference at this point.
419 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
420 BUG_ON(!parent
->d_count
);
422 spin_unlock(&dentry
->d_lock
);
424 spin_unlock(&parent
->d_lock
);
427 spin_unlock(&fs
->lock
);
429 mntget(nd
->path
.mnt
);
432 br_read_unlock(vfsmount_lock
);
433 nd
->flags
&= ~LOOKUP_RCU
;
437 spin_unlock(&dentry
->d_lock
);
439 spin_unlock(&parent
->d_lock
);
442 spin_unlock(&fs
->lock
);
447 * release_open_intent - free up open intent resources
448 * @nd: pointer to nameidata
450 void release_open_intent(struct nameidata
*nd
)
452 struct file
*file
= nd
->intent
.open
.file
;
454 if (file
&& !IS_ERR(file
)) {
455 if (file
->f_path
.dentry
== NULL
)
462 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
464 return dentry
->d_op
->d_revalidate(dentry
, nd
);
467 static struct dentry
*
468 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
470 int status
= d_revalidate(dentry
, nd
);
471 if (unlikely(status
<= 0)) {
473 * The dentry failed validation.
474 * If d_revalidate returned 0 attempt to invalidate
475 * the dentry otherwise d_revalidate is asking us
476 * to return a fail status.
480 dentry
= ERR_PTR(status
);
481 } else if (!d_invalidate(dentry
)) {
490 * complete_walk - successful completion of path walk
491 * @nd: pointer nameidata
493 * If we had been in RCU mode, drop out of it and legitimize nd->path.
494 * Revalidate the final result, unless we'd already done that during
495 * the path walk or the filesystem doesn't ask for it. Return 0 on
496 * success, -error on failure. In case of failure caller does not
497 * need to drop nd->path.
499 static int complete_walk(struct nameidata
*nd
)
501 struct dentry
*dentry
= nd
->path
.dentry
;
504 if (nd
->flags
& LOOKUP_RCU
) {
505 nd
->flags
&= ~LOOKUP_RCU
;
506 if (!(nd
->flags
& LOOKUP_ROOT
))
508 spin_lock(&dentry
->d_lock
);
509 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
510 spin_unlock(&dentry
->d_lock
);
512 br_read_unlock(vfsmount_lock
);
515 BUG_ON(nd
->inode
!= dentry
->d_inode
);
516 spin_unlock(&dentry
->d_lock
);
517 mntget(nd
->path
.mnt
);
519 br_read_unlock(vfsmount_lock
);
522 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
525 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
528 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
531 /* Note: we do not d_invalidate() */
532 status
= d_revalidate(dentry
, nd
);
543 static __always_inline
void set_root(struct nameidata
*nd
)
546 get_fs_root(current
->fs
, &nd
->root
);
549 static int link_path_walk(const char *, struct nameidata
*);
551 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
554 struct fs_struct
*fs
= current
->fs
;
558 seq
= read_seqcount_begin(&fs
->seq
);
560 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
561 } while (read_seqcount_retry(&fs
->seq
, seq
));
565 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
577 nd
->flags
|= LOOKUP_JUMPED
;
579 nd
->inode
= nd
->path
.dentry
->d_inode
;
581 ret
= link_path_walk(link
, nd
);
585 return PTR_ERR(link
);
588 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
591 if (path
->mnt
!= nd
->path
.mnt
)
595 static inline void path_to_nameidata(const struct path
*path
,
596 struct nameidata
*nd
)
598 if (!(nd
->flags
& LOOKUP_RCU
)) {
599 dput(nd
->path
.dentry
);
600 if (nd
->path
.mnt
!= path
->mnt
)
601 mntput(nd
->path
.mnt
);
603 nd
->path
.mnt
= path
->mnt
;
604 nd
->path
.dentry
= path
->dentry
;
607 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
609 struct inode
*inode
= link
->dentry
->d_inode
;
610 if (!IS_ERR(cookie
) && inode
->i_op
->put_link
)
611 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
615 static __always_inline
int
616 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
619 struct dentry
*dentry
= link
->dentry
;
621 BUG_ON(nd
->flags
& LOOKUP_RCU
);
623 if (link
->mnt
== nd
->path
.mnt
)
626 if (unlikely(current
->total_link_count
>= 40)) {
627 *p
= ERR_PTR(-ELOOP
); /* no ->put_link(), please */
632 current
->total_link_count
++;
634 touch_atime(link
->mnt
, dentry
);
635 nd_set_link(nd
, NULL
);
637 error
= security_inode_follow_link(link
->dentry
, nd
);
639 *p
= ERR_PTR(error
); /* no ->put_link(), please */
644 nd
->last_type
= LAST_BIND
;
645 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
648 char *s
= nd_get_link(nd
);
651 error
= __vfs_follow_link(nd
, s
);
652 else if (nd
->last_type
== LAST_BIND
) {
653 nd
->flags
|= LOOKUP_JUMPED
;
654 nd
->inode
= nd
->path
.dentry
->d_inode
;
655 if (nd
->inode
->i_op
->follow_link
) {
656 /* stepped on a _really_ weird one */
665 static int follow_up_rcu(struct path
*path
)
667 struct vfsmount
*parent
;
668 struct dentry
*mountpoint
;
670 parent
= path
->mnt
->mnt_parent
;
671 if (parent
== path
->mnt
)
673 mountpoint
= path
->mnt
->mnt_mountpoint
;
674 path
->dentry
= mountpoint
;
679 int follow_up(struct path
*path
)
681 struct vfsmount
*parent
;
682 struct dentry
*mountpoint
;
684 br_read_lock(vfsmount_lock
);
685 parent
= path
->mnt
->mnt_parent
;
686 if (parent
== path
->mnt
) {
687 br_read_unlock(vfsmount_lock
);
691 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
692 br_read_unlock(vfsmount_lock
);
694 path
->dentry
= mountpoint
;
701 * Perform an automount
702 * - return -EISDIR to tell follow_managed() to stop and return the path we
705 static int follow_automount(struct path
*path
, unsigned flags
,
708 struct vfsmount
*mnt
;
711 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
714 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
715 * and this is the terminal part of the path.
717 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
718 return -EISDIR
; /* we actually want to stop here */
720 /* We want to mount if someone is trying to open/create a file of any
721 * type under the mountpoint, wants to traverse through the mountpoint
722 * or wants to open the mounted directory.
724 * We don't want to mount if someone's just doing a stat and they've
725 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
726 * appended a '/' to the name.
728 if (!(flags
& LOOKUP_FOLLOW
) &&
729 !(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
730 LOOKUP_OPEN
| LOOKUP_CREATE
)))
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_CONTINUE
))
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
)
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 vfsmount
*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);
893 path
->dentry
= mounted
->mnt_root
;
894 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
896 * Update the inode too. We don't need to re-check the
897 * dentry sequence number here after this d_inode read,
898 * because a mount-point is always pinned.
900 *inode
= path
->dentry
->d_inode
;
905 static void follow_mount_rcu(struct nameidata
*nd
)
907 while (d_mountpoint(nd
->path
.dentry
)) {
908 struct vfsmount
*mounted
;
909 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
912 nd
->path
.mnt
= mounted
;
913 nd
->path
.dentry
= mounted
->mnt_root
;
914 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
918 static int follow_dotdot_rcu(struct nameidata
*nd
)
923 if (nd
->path
.dentry
== nd
->root
.dentry
&&
924 nd
->path
.mnt
== nd
->root
.mnt
) {
927 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
928 struct dentry
*old
= nd
->path
.dentry
;
929 struct dentry
*parent
= old
->d_parent
;
932 seq
= read_seqcount_begin(&parent
->d_seq
);
933 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
935 nd
->path
.dentry
= parent
;
939 if (!follow_up_rcu(&nd
->path
))
941 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
943 follow_mount_rcu(nd
);
944 nd
->inode
= nd
->path
.dentry
->d_inode
;
948 nd
->flags
&= ~LOOKUP_RCU
;
949 if (!(nd
->flags
& LOOKUP_ROOT
))
952 br_read_unlock(vfsmount_lock
);
957 * Follow down to the covering mount currently visible to userspace. At each
958 * point, the filesystem owning that dentry may be queried as to whether the
959 * caller is permitted to proceed or not.
961 int follow_down(struct path
*path
)
966 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
967 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
968 /* Allow the filesystem to manage the transit without i_mutex
971 * We indicate to the filesystem if someone is trying to mount
972 * something here. This gives autofs the chance to deny anyone
973 * other than its daemon the right to mount on its
976 * The filesystem may sleep at this point.
978 if (managed
& DCACHE_MANAGE_TRANSIT
) {
979 BUG_ON(!path
->dentry
->d_op
);
980 BUG_ON(!path
->dentry
->d_op
->d_manage
);
981 ret
= path
->dentry
->d_op
->d_manage(
982 path
->dentry
, false);
984 return ret
== -EISDIR
? 0 : ret
;
987 /* Transit to a mounted filesystem. */
988 if (managed
& DCACHE_MOUNTED
) {
989 struct vfsmount
*mounted
= lookup_mnt(path
);
995 path
->dentry
= dget(mounted
->mnt_root
);
999 /* Don't handle automount points here */
1006 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1008 static void follow_mount(struct path
*path
)
1010 while (d_mountpoint(path
->dentry
)) {
1011 struct vfsmount
*mounted
= lookup_mnt(path
);
1016 path
->mnt
= mounted
;
1017 path
->dentry
= dget(mounted
->mnt_root
);
1021 static void follow_dotdot(struct nameidata
*nd
)
1026 struct dentry
*old
= nd
->path
.dentry
;
1028 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1029 nd
->path
.mnt
== nd
->root
.mnt
) {
1032 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1033 /* rare case of legitimate dget_parent()... */
1034 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1038 if (!follow_up(&nd
->path
))
1041 follow_mount(&nd
->path
);
1042 nd
->inode
= nd
->path
.dentry
->d_inode
;
1046 * Allocate a dentry with name and parent, and perform a parent
1047 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1048 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1049 * have verified that no child exists while under i_mutex.
1051 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1052 struct qstr
*name
, struct nameidata
*nd
)
1054 struct inode
*inode
= parent
->d_inode
;
1055 struct dentry
*dentry
;
1058 /* Don't create child dentry for a dead directory. */
1059 if (unlikely(IS_DEADDIR(inode
)))
1060 return ERR_PTR(-ENOENT
);
1062 dentry
= d_alloc(parent
, name
);
1063 if (unlikely(!dentry
))
1064 return ERR_PTR(-ENOMEM
);
1066 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1067 if (unlikely(old
)) {
1075 * We already have a dentry, but require a lookup to be performed on the parent
1076 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1077 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1078 * child exists while under i_mutex.
1080 static struct dentry
*d_inode_lookup(struct dentry
*parent
, struct dentry
*dentry
,
1081 struct nameidata
*nd
)
1083 struct inode
*inode
= parent
->d_inode
;
1086 /* Don't create child dentry for a dead directory. */
1087 if (unlikely(IS_DEADDIR(inode
)))
1088 return ERR_PTR(-ENOENT
);
1090 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1091 if (unlikely(old
)) {
1099 * It's more convoluted than I'd like it to be, but... it's still fairly
1100 * small and for now I'd prefer to have fast path as straight as possible.
1101 * It _is_ time-critical.
1103 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1104 struct path
*path
, struct inode
**inode
)
1106 struct vfsmount
*mnt
= nd
->path
.mnt
;
1107 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1113 * Rename seqlock is not required here because in the off chance
1114 * of a false negative due to a concurrent rename, we're going to
1115 * do the non-racy lookup, below.
1117 if (nd
->flags
& LOOKUP_RCU
) {
1120 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1124 /* Memory barrier in read_seqcount_begin of child is enough */
1125 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1129 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1130 status
= d_revalidate(dentry
, nd
);
1131 if (unlikely(status
<= 0)) {
1132 if (status
!= -ECHILD
)
1137 if (unlikely(d_need_lookup(dentry
)))
1140 path
->dentry
= dentry
;
1141 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1143 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1147 if (unlazy_walk(nd
, dentry
))
1150 dentry
= __d_lookup(parent
, name
);
1153 if (dentry
&& unlikely(d_need_lookup(dentry
))) {
1158 if (unlikely(!dentry
)) {
1159 struct inode
*dir
= parent
->d_inode
;
1160 BUG_ON(nd
->inode
!= dir
);
1162 mutex_lock(&dir
->i_mutex
);
1163 dentry
= d_lookup(parent
, name
);
1164 if (likely(!dentry
)) {
1165 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1166 if (IS_ERR(dentry
)) {
1167 mutex_unlock(&dir
->i_mutex
);
1168 return PTR_ERR(dentry
);
1173 } else if (unlikely(d_need_lookup(dentry
))) {
1174 dentry
= d_inode_lookup(parent
, dentry
, nd
);
1175 if (IS_ERR(dentry
)) {
1176 mutex_unlock(&dir
->i_mutex
);
1177 return PTR_ERR(dentry
);
1183 mutex_unlock(&dir
->i_mutex
);
1185 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1186 status
= d_revalidate(dentry
, nd
);
1187 if (unlikely(status
<= 0)) {
1192 if (!d_invalidate(dentry
)) {
1201 path
->dentry
= dentry
;
1202 err
= follow_managed(path
, nd
->flags
);
1203 if (unlikely(err
< 0)) {
1204 path_put_conditional(path
, nd
);
1207 *inode
= path
->dentry
->d_inode
;
1211 static inline int may_lookup(struct nameidata
*nd
)
1213 if (nd
->flags
& LOOKUP_RCU
) {
1214 int err
= exec_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1217 if (unlazy_walk(nd
, NULL
))
1220 return exec_permission(nd
->inode
, MAY_EXEC
);
1223 static inline int handle_dots(struct nameidata
*nd
, int type
)
1225 if (type
== LAST_DOTDOT
) {
1226 if (nd
->flags
& LOOKUP_RCU
) {
1227 if (follow_dotdot_rcu(nd
))
1235 static void terminate_walk(struct nameidata
*nd
)
1237 if (!(nd
->flags
& LOOKUP_RCU
)) {
1238 path_put(&nd
->path
);
1240 nd
->flags
&= ~LOOKUP_RCU
;
1241 if (!(nd
->flags
& LOOKUP_ROOT
))
1242 nd
->root
.mnt
= NULL
;
1244 br_read_unlock(vfsmount_lock
);
1248 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1249 struct qstr
*name
, int type
, int follow
)
1251 struct inode
*inode
;
1254 * "." and ".." are special - ".." especially so because it has
1255 * to be able to know about the current root directory and
1256 * parent relationships.
1258 if (unlikely(type
!= LAST_NORM
))
1259 return handle_dots(nd
, type
);
1260 err
= do_lookup(nd
, name
, path
, &inode
);
1261 if (unlikely(err
)) {
1266 path_to_nameidata(path
, nd
);
1270 if (unlikely(inode
->i_op
->follow_link
) && follow
) {
1271 if (nd
->flags
& LOOKUP_RCU
) {
1272 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1277 BUG_ON(inode
!= path
->dentry
->d_inode
);
1280 path_to_nameidata(path
, nd
);
1286 * This limits recursive symlink follows to 8, while
1287 * limiting consecutive symlinks to 40.
1289 * Without that kind of total limit, nasty chains of consecutive
1290 * symlinks can cause almost arbitrarily long lookups.
1292 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1296 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1297 path_put_conditional(path
, nd
);
1298 path_put(&nd
->path
);
1301 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1304 current
->link_count
++;
1307 struct path link
= *path
;
1310 res
= follow_link(&link
, nd
, &cookie
);
1312 res
= walk_component(nd
, path
, &nd
->last
,
1313 nd
->last_type
, LOOKUP_FOLLOW
);
1314 put_link(nd
, &link
, cookie
);
1317 current
->link_count
--;
1324 * This is the basic name resolution function, turning a pathname into
1325 * the final dentry. We expect 'base' to be positive and a directory.
1327 * Returns 0 and nd will have valid dentry and mnt on success.
1328 * Returns error and drops reference to input namei data on failure.
1330 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1334 unsigned int lookup_flags
= nd
->flags
;
1341 /* At this point we know we have a real path component. */
1348 nd
->flags
|= LOOKUP_CONTINUE
;
1350 err
= may_lookup(nd
);
1355 c
= *(const unsigned char *)name
;
1357 hash
= init_name_hash();
1360 hash
= partial_name_hash(c
, hash
);
1361 c
= *(const unsigned char *)name
;
1362 } while (c
&& (c
!= '/'));
1363 this.len
= name
- (const char *) this.name
;
1364 this.hash
= end_name_hash(hash
);
1367 if (this.name
[0] == '.') switch (this.len
) {
1369 if (this.name
[1] == '.') {
1371 nd
->flags
|= LOOKUP_JUMPED
;
1377 if (likely(type
== LAST_NORM
)) {
1378 struct dentry
*parent
= nd
->path
.dentry
;
1379 nd
->flags
&= ~LOOKUP_JUMPED
;
1380 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1381 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1388 /* remove trailing slashes? */
1390 goto last_component
;
1391 while (*++name
== '/');
1393 goto last_component
;
1395 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1400 err
= nested_symlink(&next
, nd
);
1405 if (!nd
->inode
->i_op
->lookup
)
1408 /* here ends the main loop */
1411 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1412 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1414 nd
->last_type
= type
;
1421 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1422 struct nameidata
*nd
, struct file
**fp
)
1428 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1429 nd
->flags
= flags
| LOOKUP_JUMPED
;
1431 if (flags
& LOOKUP_ROOT
) {
1432 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1434 if (!inode
->i_op
->lookup
)
1436 retval
= inode_permission(inode
, MAY_EXEC
);
1440 nd
->path
= nd
->root
;
1442 if (flags
& LOOKUP_RCU
) {
1443 br_read_lock(vfsmount_lock
);
1445 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1447 path_get(&nd
->path
);
1452 nd
->root
.mnt
= NULL
;
1455 if (flags
& LOOKUP_RCU
) {
1456 br_read_lock(vfsmount_lock
);
1461 path_get(&nd
->root
);
1463 nd
->path
= nd
->root
;
1464 } else if (dfd
== AT_FDCWD
) {
1465 if (flags
& LOOKUP_RCU
) {
1466 struct fs_struct
*fs
= current
->fs
;
1469 br_read_lock(vfsmount_lock
);
1473 seq
= read_seqcount_begin(&fs
->seq
);
1475 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1476 } while (read_seqcount_retry(&fs
->seq
, seq
));
1478 get_fs_pwd(current
->fs
, &nd
->path
);
1481 struct dentry
*dentry
;
1483 file
= fget_raw_light(dfd
, &fput_needed
);
1488 dentry
= file
->f_path
.dentry
;
1492 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1495 retval
= exec_permission(dentry
->d_inode
, MAY_EXEC
);
1500 nd
->path
= file
->f_path
;
1501 if (flags
& LOOKUP_RCU
) {
1504 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1505 br_read_lock(vfsmount_lock
);
1508 path_get(&file
->f_path
);
1509 fput_light(file
, fput_needed
);
1513 nd
->inode
= nd
->path
.dentry
->d_inode
;
1517 fput_light(file
, fput_needed
);
1522 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1524 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1525 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1527 nd
->flags
&= ~LOOKUP_PARENT
;
1528 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1529 nd
->flags
& LOOKUP_FOLLOW
);
1532 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1533 static int path_lookupat(int dfd
, const char *name
,
1534 unsigned int flags
, struct nameidata
*nd
)
1536 struct file
*base
= NULL
;
1541 * Path walking is largely split up into 2 different synchronisation
1542 * schemes, rcu-walk and ref-walk (explained in
1543 * Documentation/filesystems/path-lookup.txt). These share much of the
1544 * path walk code, but some things particularly setup, cleanup, and
1545 * following mounts are sufficiently divergent that functions are
1546 * duplicated. Typically there is a function foo(), and its RCU
1547 * analogue, foo_rcu().
1549 * -ECHILD is the error number of choice (just to avoid clashes) that
1550 * is returned if some aspect of an rcu-walk fails. Such an error must
1551 * be handled by restarting a traditional ref-walk (which will always
1552 * be able to complete).
1554 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1559 current
->total_link_count
= 0;
1560 err
= link_path_walk(name
, nd
);
1562 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1563 err
= lookup_last(nd
, &path
);
1566 struct path link
= path
;
1567 nd
->flags
|= LOOKUP_PARENT
;
1568 err
= follow_link(&link
, nd
, &cookie
);
1570 err
= lookup_last(nd
, &path
);
1571 put_link(nd
, &link
, cookie
);
1576 err
= complete_walk(nd
);
1578 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1579 if (!nd
->inode
->i_op
->lookup
) {
1580 path_put(&nd
->path
);
1588 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1589 path_put(&nd
->root
);
1590 nd
->root
.mnt
= NULL
;
1595 static int do_path_lookup(int dfd
, const char *name
,
1596 unsigned int flags
, struct nameidata
*nd
)
1598 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1599 if (unlikely(retval
== -ECHILD
))
1600 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1601 if (unlikely(retval
== -ESTALE
))
1602 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1604 if (likely(!retval
)) {
1605 if (unlikely(!audit_dummy_context())) {
1606 if (nd
->path
.dentry
&& nd
->inode
)
1607 audit_inode(name
, nd
->path
.dentry
);
1613 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1615 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1618 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1620 struct nameidata nd
;
1621 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1628 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1629 * @dentry: pointer to dentry of the base directory
1630 * @mnt: pointer to vfs mount of the base directory
1631 * @name: pointer to file name
1632 * @flags: lookup flags
1633 * @nd: pointer to nameidata
1635 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1636 const char *name
, unsigned int flags
,
1637 struct nameidata
*nd
)
1639 nd
->root
.dentry
= dentry
;
1641 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1642 return do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, nd
);
1645 static struct dentry
*__lookup_hash(struct qstr
*name
,
1646 struct dentry
*base
, struct nameidata
*nd
)
1648 struct inode
*inode
= base
->d_inode
;
1649 struct dentry
*dentry
;
1652 err
= exec_permission(inode
, MAY_EXEC
);
1654 return ERR_PTR(err
);
1657 * Don't bother with __d_lookup: callers are for creat as
1658 * well as unlink, so a lot of the time it would cost
1661 dentry
= d_lookup(base
, name
);
1663 if (dentry
&& d_need_lookup(dentry
)) {
1665 * __lookup_hash is called with the parent dir's i_mutex already
1666 * held, so we are good to go here.
1668 dentry
= d_inode_lookup(base
, dentry
, nd
);
1673 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1674 dentry
= do_revalidate(dentry
, nd
);
1677 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1683 * Restricted form of lookup. Doesn't follow links, single-component only,
1684 * needs parent already locked. Doesn't follow mounts.
1687 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1689 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1693 * lookup_one_len - filesystem helper to lookup single pathname component
1694 * @name: pathname component to lookup
1695 * @base: base directory to lookup from
1696 * @len: maximum length @len should be interpreted to
1698 * Note that this routine is purely a helper for filesystem usage and should
1699 * not be called by generic code. Also note that by using this function the
1700 * nameidata argument is passed to the filesystem methods and a filesystem
1701 * using this helper needs to be prepared for that.
1703 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1709 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1714 return ERR_PTR(-EACCES
);
1716 hash
= init_name_hash();
1718 c
= *(const unsigned char *)name
++;
1719 if (c
== '/' || c
== '\0')
1720 return ERR_PTR(-EACCES
);
1721 hash
= partial_name_hash(c
, hash
);
1723 this.hash
= end_name_hash(hash
);
1725 * See if the low-level filesystem might want
1726 * to use its own hash..
1728 if (base
->d_flags
& DCACHE_OP_HASH
) {
1729 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1731 return ERR_PTR(err
);
1734 return __lookup_hash(&this, base
, NULL
);
1737 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1740 struct nameidata nd
;
1741 char *tmp
= getname_flags(name
, flags
);
1742 int err
= PTR_ERR(tmp
);
1745 BUG_ON(flags
& LOOKUP_PARENT
);
1747 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1755 static int user_path_parent(int dfd
, const char __user
*path
,
1756 struct nameidata
*nd
, char **name
)
1758 char *s
= getname(path
);
1764 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1774 * It's inline, so penalty for filesystems that don't use sticky bit is
1777 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1779 uid_t fsuid
= current_fsuid();
1781 if (!(dir
->i_mode
& S_ISVTX
))
1783 if (current_user_ns() != inode_userns(inode
))
1785 if (inode
->i_uid
== fsuid
)
1787 if (dir
->i_uid
== fsuid
)
1791 return !ns_capable(inode_userns(inode
), CAP_FOWNER
);
1795 * Check whether we can remove a link victim from directory dir, check
1796 * whether the type of victim is right.
1797 * 1. We can't do it if dir is read-only (done in permission())
1798 * 2. We should have write and exec permissions on dir
1799 * 3. We can't remove anything from append-only dir
1800 * 4. We can't do anything with immutable dir (done in permission())
1801 * 5. If the sticky bit on dir is set we should either
1802 * a. be owner of dir, or
1803 * b. be owner of victim, or
1804 * c. have CAP_FOWNER capability
1805 * 6. If the victim is append-only or immutable we can't do antyhing with
1806 * links pointing to it.
1807 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1808 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1809 * 9. We can't remove a root or mountpoint.
1810 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1811 * nfs_async_unlink().
1813 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1817 if (!victim
->d_inode
)
1820 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1821 audit_inode_child(victim
, dir
);
1823 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1828 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1829 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1832 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1834 if (IS_ROOT(victim
))
1836 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1838 if (IS_DEADDIR(dir
))
1840 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1845 /* Check whether we can create an object with dentry child in directory
1847 * 1. We can't do it if child already exists (open has special treatment for
1848 * this case, but since we are inlined it's OK)
1849 * 2. We can't do it if dir is read-only (done in permission())
1850 * 3. We should have write and exec permissions on dir
1851 * 4. We can't do it if dir is immutable (done in permission())
1853 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1857 if (IS_DEADDIR(dir
))
1859 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1863 * p1 and p2 should be directories on the same fs.
1865 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1870 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1874 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1876 p
= d_ancestor(p2
, p1
);
1878 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1879 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1883 p
= d_ancestor(p1
, p2
);
1885 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1886 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1890 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1891 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1895 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1897 mutex_unlock(&p1
->d_inode
->i_mutex
);
1899 mutex_unlock(&p2
->d_inode
->i_mutex
);
1900 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1904 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1905 struct nameidata
*nd
)
1907 int error
= may_create(dir
, dentry
);
1912 if (!dir
->i_op
->create
)
1913 return -EACCES
; /* shouldn't it be ENOSYS? */
1916 error
= security_inode_create(dir
, dentry
, mode
);
1919 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1921 fsnotify_create(dir
, dentry
);
1925 static int may_open(struct path
*path
, int acc_mode
, int flag
)
1927 struct dentry
*dentry
= path
->dentry
;
1928 struct inode
*inode
= dentry
->d_inode
;
1938 switch (inode
->i_mode
& S_IFMT
) {
1942 if (acc_mode
& MAY_WRITE
)
1947 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
1956 error
= inode_permission(inode
, acc_mode
);
1961 * An append-only file must be opened in append mode for writing.
1963 if (IS_APPEND(inode
)) {
1964 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
1970 /* O_NOATIME can only be set by the owner or superuser */
1971 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
1975 * Ensure there are no outstanding leases on the file.
1977 return break_lease(inode
, flag
);
1980 static int handle_truncate(struct file
*filp
)
1982 struct path
*path
= &filp
->f_path
;
1983 struct inode
*inode
= path
->dentry
->d_inode
;
1984 int error
= get_write_access(inode
);
1988 * Refuse to truncate files with mandatory locks held on them.
1990 error
= locks_verify_locked(inode
);
1992 error
= security_path_truncate(path
);
1994 error
= do_truncate(path
->dentry
, 0,
1995 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
1998 put_write_access(inode
);
2003 * Note that while the flag value (low two bits) for sys_open means:
2008 * it is changed into
2009 * 00 - no permissions needed
2010 * 01 - read-permission
2011 * 10 - write-permission
2013 * for the internal routines (ie open_namei()/follow_link() etc)
2014 * This is more logical, and also allows the 00 "no perm needed"
2015 * to be used for symlinks (where the permissions are checked
2019 static inline int open_to_namei_flags(int flag
)
2021 if ((flag
+1) & O_ACCMODE
)
2027 * Handle the last step of open()
2029 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2030 const struct open_flags
*op
, const char *pathname
)
2032 struct dentry
*dir
= nd
->path
.dentry
;
2033 struct dentry
*dentry
;
2034 int open_flag
= op
->open_flag
;
2035 int will_truncate
= open_flag
& O_TRUNC
;
2037 int acc_mode
= op
->acc_mode
;
2041 nd
->flags
&= ~LOOKUP_PARENT
;
2042 nd
->flags
|= op
->intent
;
2044 switch (nd
->last_type
) {
2047 error
= handle_dots(nd
, nd
->last_type
);
2049 return ERR_PTR(error
);
2052 error
= complete_walk(nd
);
2054 return ERR_PTR(error
);
2055 audit_inode(pathname
, nd
->path
.dentry
);
2056 if (open_flag
& O_CREAT
) {
2062 error
= complete_walk(nd
);
2064 return ERR_PTR(error
);
2065 audit_inode(pathname
, dir
);
2069 if (!(open_flag
& O_CREAT
)) {
2071 if (nd
->last
.name
[nd
->last
.len
])
2072 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2073 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2075 /* we _can_ be in RCU mode here */
2076 error
= walk_component(nd
, path
, &nd
->last
, LAST_NORM
,
2079 return ERR_PTR(error
);
2080 if (error
) /* symlink */
2083 error
= complete_walk(nd
);
2085 return ERR_PTR(-ECHILD
);
2088 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2089 if (!nd
->inode
->i_op
->lookup
)
2092 audit_inode(pathname
, nd
->path
.dentry
);
2096 /* create side of things */
2097 error
= complete_walk(nd
);
2099 return ERR_PTR(error
);
2101 audit_inode(pathname
, dir
);
2103 /* trailing slashes? */
2104 if (nd
->last
.name
[nd
->last
.len
])
2107 mutex_lock(&dir
->d_inode
->i_mutex
);
2109 dentry
= lookup_hash(nd
);
2110 error
= PTR_ERR(dentry
);
2111 if (IS_ERR(dentry
)) {
2112 mutex_unlock(&dir
->d_inode
->i_mutex
);
2116 path
->dentry
= dentry
;
2117 path
->mnt
= nd
->path
.mnt
;
2119 /* Negative dentry, just create the file */
2120 if (!dentry
->d_inode
) {
2121 int mode
= op
->mode
;
2122 if (!IS_POSIXACL(dir
->d_inode
))
2123 mode
&= ~current_umask();
2125 * This write is needed to ensure that a
2126 * rw->ro transition does not occur between
2127 * the time when the file is created and when
2128 * a permanent write count is taken through
2129 * the 'struct file' in nameidata_to_filp().
2131 error
= mnt_want_write(nd
->path
.mnt
);
2133 goto exit_mutex_unlock
;
2135 /* Don't check for write permission, don't truncate */
2136 open_flag
&= ~O_TRUNC
;
2138 acc_mode
= MAY_OPEN
;
2139 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2141 goto exit_mutex_unlock
;
2142 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2144 goto exit_mutex_unlock
;
2145 mutex_unlock(&dir
->d_inode
->i_mutex
);
2146 dput(nd
->path
.dentry
);
2147 nd
->path
.dentry
= dentry
;
2152 * It already exists.
2154 mutex_unlock(&dir
->d_inode
->i_mutex
);
2155 audit_inode(pathname
, path
->dentry
);
2158 if (open_flag
& O_EXCL
)
2161 error
= follow_managed(path
, nd
->flags
);
2166 if (!path
->dentry
->d_inode
)
2169 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2172 path_to_nameidata(path
, nd
);
2173 nd
->inode
= path
->dentry
->d_inode
;
2175 if (S_ISDIR(nd
->inode
->i_mode
))
2178 if (!S_ISREG(nd
->inode
->i_mode
))
2181 if (will_truncate
) {
2182 error
= mnt_want_write(nd
->path
.mnt
);
2188 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2191 filp
= nameidata_to_filp(nd
);
2192 if (!IS_ERR(filp
)) {
2193 error
= ima_file_check(filp
, op
->acc_mode
);
2196 filp
= ERR_PTR(error
);
2199 if (!IS_ERR(filp
)) {
2200 if (will_truncate
) {
2201 error
= handle_truncate(filp
);
2204 filp
= ERR_PTR(error
);
2210 mnt_drop_write(nd
->path
.mnt
);
2211 path_put(&nd
->path
);
2215 mutex_unlock(&dir
->d_inode
->i_mutex
);
2217 path_put_conditional(path
, nd
);
2219 filp
= ERR_PTR(error
);
2223 static struct file
*path_openat(int dfd
, const char *pathname
,
2224 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2226 struct file
*base
= NULL
;
2231 filp
= get_empty_filp();
2233 return ERR_PTR(-ENFILE
);
2235 filp
->f_flags
= op
->open_flag
;
2236 nd
->intent
.open
.file
= filp
;
2237 nd
->intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2238 nd
->intent
.open
.create_mode
= op
->mode
;
2240 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2241 if (unlikely(error
))
2244 current
->total_link_count
= 0;
2245 error
= link_path_walk(pathname
, nd
);
2246 if (unlikely(error
))
2249 filp
= do_last(nd
, &path
, op
, pathname
);
2250 while (unlikely(!filp
)) { /* trailing symlink */
2251 struct path link
= path
;
2253 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2254 path_put_conditional(&path
, nd
);
2255 path_put(&nd
->path
);
2256 filp
= ERR_PTR(-ELOOP
);
2259 nd
->flags
|= LOOKUP_PARENT
;
2260 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2261 error
= follow_link(&link
, nd
, &cookie
);
2262 if (unlikely(error
))
2263 filp
= ERR_PTR(error
);
2265 filp
= do_last(nd
, &path
, op
, pathname
);
2266 put_link(nd
, &link
, cookie
);
2269 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2270 path_put(&nd
->root
);
2273 release_open_intent(nd
);
2277 filp
= ERR_PTR(error
);
2281 struct file
*do_filp_open(int dfd
, const char *pathname
,
2282 const struct open_flags
*op
, int flags
)
2284 struct nameidata nd
;
2287 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2288 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2289 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2290 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2291 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2295 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2296 const char *name
, const struct open_flags
*op
, int flags
)
2298 struct nameidata nd
;
2302 nd
.root
.dentry
= dentry
;
2304 flags
|= LOOKUP_ROOT
;
2306 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2307 return ERR_PTR(-ELOOP
);
2309 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2310 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2311 file
= path_openat(-1, name
, &nd
, op
, flags
);
2312 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2313 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2318 * lookup_create - lookup a dentry, creating it if it doesn't exist
2319 * @nd: nameidata info
2320 * @is_dir: directory flag
2322 * Simple function to lookup and return a dentry and create it
2323 * if it doesn't exist. Is SMP-safe.
2325 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2327 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2329 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2331 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2333 * Yucky last component or no last component at all?
2334 * (foo/., foo/.., /////)
2336 if (nd
->last_type
!= LAST_NORM
)
2338 nd
->flags
&= ~LOOKUP_PARENT
;
2339 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2340 nd
->intent
.open
.flags
= O_EXCL
;
2343 * Do the final lookup.
2345 dentry
= lookup_hash(nd
);
2349 if (dentry
->d_inode
)
2352 * Special case - lookup gave negative, but... we had foo/bar/
2353 * From the vfs_mknod() POV we just have a negative dentry -
2354 * all is fine. Let's be bastards - you had / on the end, you've
2355 * been asking for (non-existent) directory. -ENOENT for you.
2357 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2359 dentry
= ERR_PTR(-ENOENT
);
2364 dentry
= ERR_PTR(-EEXIST
);
2368 EXPORT_SYMBOL_GPL(lookup_create
);
2370 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2372 int error
= may_create(dir
, dentry
);
2377 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) &&
2378 !ns_capable(inode_userns(dir
), CAP_MKNOD
))
2381 if (!dir
->i_op
->mknod
)
2384 error
= devcgroup_inode_mknod(mode
, dev
);
2388 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2392 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2394 fsnotify_create(dir
, dentry
);
2398 static int may_mknod(mode_t mode
)
2400 switch (mode
& S_IFMT
) {
2406 case 0: /* zero mode translates to S_IFREG */
2415 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2420 struct dentry
*dentry
;
2421 struct nameidata nd
;
2426 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2430 dentry
= lookup_create(&nd
, 0);
2431 if (IS_ERR(dentry
)) {
2432 error
= PTR_ERR(dentry
);
2435 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2436 mode
&= ~current_umask();
2437 error
= may_mknod(mode
);
2440 error
= mnt_want_write(nd
.path
.mnt
);
2443 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2445 goto out_drop_write
;
2446 switch (mode
& S_IFMT
) {
2447 case 0: case S_IFREG
:
2448 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2450 case S_IFCHR
: case S_IFBLK
:
2451 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2452 new_decode_dev(dev
));
2454 case S_IFIFO
: case S_IFSOCK
:
2455 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2459 mnt_drop_write(nd
.path
.mnt
);
2463 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2470 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2472 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2475 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2477 int error
= may_create(dir
, dentry
);
2482 if (!dir
->i_op
->mkdir
)
2485 mode
&= (S_IRWXUGO
|S_ISVTX
);
2486 error
= security_inode_mkdir(dir
, dentry
, mode
);
2490 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2492 fsnotify_mkdir(dir
, dentry
);
2496 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2500 struct dentry
*dentry
;
2501 struct nameidata nd
;
2503 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2507 dentry
= lookup_create(&nd
, 1);
2508 error
= PTR_ERR(dentry
);
2512 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2513 mode
&= ~current_umask();
2514 error
= mnt_want_write(nd
.path
.mnt
);
2517 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2519 goto out_drop_write
;
2520 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2522 mnt_drop_write(nd
.path
.mnt
);
2526 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2533 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2535 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2539 * The dentry_unhash() helper will try to drop the dentry early: we
2540 * should have a usage count of 2 if we're the only user of this
2541 * dentry, and if that is true (possibly after pruning the dcache),
2542 * then we drop the dentry now.
2544 * A low-level filesystem can, if it choses, legally
2547 * if (!d_unhashed(dentry))
2550 * if it cannot handle the case of removing a directory
2551 * that is still in use by something else..
2553 void dentry_unhash(struct dentry
*dentry
)
2555 shrink_dcache_parent(dentry
);
2556 spin_lock(&dentry
->d_lock
);
2557 if (dentry
->d_count
== 1)
2559 spin_unlock(&dentry
->d_lock
);
2562 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2564 int error
= may_delete(dir
, dentry
, 1);
2569 if (!dir
->i_op
->rmdir
)
2572 mutex_lock(&dentry
->d_inode
->i_mutex
);
2575 if (d_mountpoint(dentry
))
2578 error
= security_inode_rmdir(dir
, dentry
);
2582 shrink_dcache_parent(dentry
);
2583 error
= dir
->i_op
->rmdir(dir
, dentry
);
2587 dentry
->d_inode
->i_flags
|= S_DEAD
;
2591 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2597 static long do_rmdir(int dfd
, const char __user
*pathname
)
2601 struct dentry
*dentry
;
2602 struct nameidata nd
;
2604 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2608 switch(nd
.last_type
) {
2620 nd
.flags
&= ~LOOKUP_PARENT
;
2622 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2623 dentry
= lookup_hash(&nd
);
2624 error
= PTR_ERR(dentry
);
2627 if (!dentry
->d_inode
) {
2631 error
= mnt_want_write(nd
.path
.mnt
);
2634 error
= security_path_rmdir(&nd
.path
, dentry
);
2637 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2639 mnt_drop_write(nd
.path
.mnt
);
2643 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2650 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2652 return do_rmdir(AT_FDCWD
, pathname
);
2655 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2657 int error
= may_delete(dir
, dentry
, 0);
2662 if (!dir
->i_op
->unlink
)
2665 mutex_lock(&dentry
->d_inode
->i_mutex
);
2666 if (d_mountpoint(dentry
))
2669 error
= security_inode_unlink(dir
, dentry
);
2671 error
= dir
->i_op
->unlink(dir
, dentry
);
2676 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2678 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2679 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2680 fsnotify_link_count(dentry
->d_inode
);
2688 * Make sure that the actual truncation of the file will occur outside its
2689 * directory's i_mutex. Truncate can take a long time if there is a lot of
2690 * writeout happening, and we don't want to prevent access to the directory
2691 * while waiting on the I/O.
2693 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2697 struct dentry
*dentry
;
2698 struct nameidata nd
;
2699 struct inode
*inode
= NULL
;
2701 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2706 if (nd
.last_type
!= LAST_NORM
)
2709 nd
.flags
&= ~LOOKUP_PARENT
;
2711 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2712 dentry
= lookup_hash(&nd
);
2713 error
= PTR_ERR(dentry
);
2714 if (!IS_ERR(dentry
)) {
2715 /* Why not before? Because we want correct error value */
2716 if (nd
.last
.name
[nd
.last
.len
])
2718 inode
= dentry
->d_inode
;
2722 error
= mnt_want_write(nd
.path
.mnt
);
2725 error
= security_path_unlink(&nd
.path
, dentry
);
2728 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2730 mnt_drop_write(nd
.path
.mnt
);
2734 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2736 iput(inode
); /* truncate the inode here */
2743 error
= !dentry
->d_inode
? -ENOENT
:
2744 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2748 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2750 if ((flag
& ~AT_REMOVEDIR
) != 0)
2753 if (flag
& AT_REMOVEDIR
)
2754 return do_rmdir(dfd
, pathname
);
2756 return do_unlinkat(dfd
, pathname
);
2759 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2761 return do_unlinkat(AT_FDCWD
, pathname
);
2764 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2766 int error
= may_create(dir
, dentry
);
2771 if (!dir
->i_op
->symlink
)
2774 error
= security_inode_symlink(dir
, dentry
, oldname
);
2778 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2780 fsnotify_create(dir
, dentry
);
2784 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2785 int, newdfd
, const char __user
*, newname
)
2790 struct dentry
*dentry
;
2791 struct nameidata nd
;
2793 from
= getname(oldname
);
2795 return PTR_ERR(from
);
2797 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2801 dentry
= lookup_create(&nd
, 0);
2802 error
= PTR_ERR(dentry
);
2806 error
= mnt_want_write(nd
.path
.mnt
);
2809 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2811 goto out_drop_write
;
2812 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2814 mnt_drop_write(nd
.path
.mnt
);
2818 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2826 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2828 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2831 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2833 struct inode
*inode
= old_dentry
->d_inode
;
2839 error
= may_create(dir
, new_dentry
);
2843 if (dir
->i_sb
!= inode
->i_sb
)
2847 * A link to an append-only or immutable file cannot be created.
2849 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2851 if (!dir
->i_op
->link
)
2853 if (S_ISDIR(inode
->i_mode
))
2856 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2860 mutex_lock(&inode
->i_mutex
);
2861 /* Make sure we don't allow creating hardlink to an unlinked file */
2862 if (inode
->i_nlink
== 0)
2865 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2866 mutex_unlock(&inode
->i_mutex
);
2868 fsnotify_link(dir
, inode
, new_dentry
);
2873 * Hardlinks are often used in delicate situations. We avoid
2874 * security-related surprises by not following symlinks on the
2877 * We don't follow them on the oldname either to be compatible
2878 * with linux 2.0, and to avoid hard-linking to directories
2879 * and other special files. --ADM
2881 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2882 int, newdfd
, const char __user
*, newname
, int, flags
)
2884 struct dentry
*new_dentry
;
2885 struct nameidata nd
;
2886 struct path old_path
;
2891 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
2894 * To use null names we require CAP_DAC_READ_SEARCH
2895 * This ensures that not everyone will be able to create
2896 * handlink using the passed filedescriptor.
2898 if (flags
& AT_EMPTY_PATH
) {
2899 if (!capable(CAP_DAC_READ_SEARCH
))
2904 if (flags
& AT_SYMLINK_FOLLOW
)
2905 how
|= LOOKUP_FOLLOW
;
2907 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
2911 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2915 if (old_path
.mnt
!= nd
.path
.mnt
)
2917 new_dentry
= lookup_create(&nd
, 0);
2918 error
= PTR_ERR(new_dentry
);
2919 if (IS_ERR(new_dentry
))
2921 error
= mnt_want_write(nd
.path
.mnt
);
2924 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2926 goto out_drop_write
;
2927 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2929 mnt_drop_write(nd
.path
.mnt
);
2933 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2938 path_put(&old_path
);
2943 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
2945 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2949 * The worst of all namespace operations - renaming directory. "Perverted"
2950 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2952 * a) we can get into loop creation. Check is done in is_subdir().
2953 * b) race potential - two innocent renames can create a loop together.
2954 * That's where 4.4 screws up. Current fix: serialization on
2955 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2957 * c) we have to lock _three_ objects - parents and victim (if it exists).
2958 * And that - after we got ->i_mutex on parents (until then we don't know
2959 * whether the target exists). Solution: try to be smart with locking
2960 * order for inodes. We rely on the fact that tree topology may change
2961 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2962 * move will be locked. Thus we can rank directories by the tree
2963 * (ancestors first) and rank all non-directories after them.
2964 * That works since everybody except rename does "lock parent, lookup,
2965 * lock child" and rename is under ->s_vfs_rename_mutex.
2966 * HOWEVER, it relies on the assumption that any object with ->lookup()
2967 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2968 * we'd better make sure that there's no link(2) for them.
2969 * d) conversion from fhandle to dentry may come in the wrong moment - when
2970 * we are removing the target. Solution: we will have to grab ->i_mutex
2971 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2972 * ->i_mutex on parents, which works but leads to some truly excessive
2975 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2976 struct inode
*new_dir
, struct dentry
*new_dentry
)
2979 struct inode
*target
= new_dentry
->d_inode
;
2982 * If we are going to change the parent - check write permissions,
2983 * we'll need to flip '..'.
2985 if (new_dir
!= old_dir
) {
2986 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
2991 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2996 mutex_lock(&target
->i_mutex
);
2999 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3003 shrink_dcache_parent(new_dentry
);
3004 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3009 target
->i_flags
|= S_DEAD
;
3010 dont_mount(new_dentry
);
3014 mutex_unlock(&target
->i_mutex
);
3016 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3017 d_move(old_dentry
,new_dentry
);
3021 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3022 struct inode
*new_dir
, struct dentry
*new_dentry
)
3024 struct inode
*target
= new_dentry
->d_inode
;
3027 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3033 mutex_lock(&target
->i_mutex
);
3036 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3039 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3044 dont_mount(new_dentry
);
3045 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3046 d_move(old_dentry
, new_dentry
);
3049 mutex_unlock(&target
->i_mutex
);
3054 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3055 struct inode
*new_dir
, struct dentry
*new_dentry
)
3058 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3059 const unsigned char *old_name
;
3061 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3064 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3068 if (!new_dentry
->d_inode
)
3069 error
= may_create(new_dir
, new_dentry
);
3071 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3075 if (!old_dir
->i_op
->rename
)
3078 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3081 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3083 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3085 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3086 new_dentry
->d_inode
, old_dentry
);
3087 fsnotify_oldname_free(old_name
);
3092 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3093 int, newdfd
, const char __user
*, newname
)
3095 struct dentry
*old_dir
, *new_dir
;
3096 struct dentry
*old_dentry
, *new_dentry
;
3097 struct dentry
*trap
;
3098 struct nameidata oldnd
, newnd
;
3103 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3107 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3112 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3115 old_dir
= oldnd
.path
.dentry
;
3117 if (oldnd
.last_type
!= LAST_NORM
)
3120 new_dir
= newnd
.path
.dentry
;
3121 if (newnd
.last_type
!= LAST_NORM
)
3124 oldnd
.flags
&= ~LOOKUP_PARENT
;
3125 newnd
.flags
&= ~LOOKUP_PARENT
;
3126 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3128 trap
= lock_rename(new_dir
, old_dir
);
3130 old_dentry
= lookup_hash(&oldnd
);
3131 error
= PTR_ERR(old_dentry
);
3132 if (IS_ERR(old_dentry
))
3134 /* source must exist */
3136 if (!old_dentry
->d_inode
)
3138 /* unless the source is a directory trailing slashes give -ENOTDIR */
3139 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3141 if (oldnd
.last
.name
[oldnd
.last
.len
])
3143 if (newnd
.last
.name
[newnd
.last
.len
])
3146 /* source should not be ancestor of target */
3148 if (old_dentry
== trap
)
3150 new_dentry
= lookup_hash(&newnd
);
3151 error
= PTR_ERR(new_dentry
);
3152 if (IS_ERR(new_dentry
))
3154 /* target should not be an ancestor of source */
3156 if (new_dentry
== trap
)
3159 error
= mnt_want_write(oldnd
.path
.mnt
);
3162 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3163 &newnd
.path
, new_dentry
);
3166 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3167 new_dir
->d_inode
, new_dentry
);
3169 mnt_drop_write(oldnd
.path
.mnt
);
3175 unlock_rename(new_dir
, old_dir
);
3177 path_put(&newnd
.path
);
3180 path_put(&oldnd
.path
);
3186 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3188 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3191 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3195 len
= PTR_ERR(link
);
3200 if (len
> (unsigned) buflen
)
3202 if (copy_to_user(buffer
, link
, len
))
3209 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3210 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3211 * using) it for any given inode is up to filesystem.
3213 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3215 struct nameidata nd
;
3220 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3222 return PTR_ERR(cookie
);
3224 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3225 if (dentry
->d_inode
->i_op
->put_link
)
3226 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3230 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3232 return __vfs_follow_link(nd
, link
);
3235 /* get the link contents into pagecache */
3236 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3240 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3241 page
= read_mapping_page(mapping
, 0, NULL
);
3246 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3250 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3252 struct page
*page
= NULL
;
3253 char *s
= page_getlink(dentry
, &page
);
3254 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3257 page_cache_release(page
);
3262 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3264 struct page
*page
= NULL
;
3265 nd_set_link(nd
, page_getlink(dentry
, &page
));
3269 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3271 struct page
*page
= cookie
;
3275 page_cache_release(page
);
3280 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3282 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3284 struct address_space
*mapping
= inode
->i_mapping
;
3289 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3291 flags
|= AOP_FLAG_NOFS
;
3294 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3295 flags
, &page
, &fsdata
);
3299 kaddr
= kmap_atomic(page
, KM_USER0
);
3300 memcpy(kaddr
, symname
, len
-1);
3301 kunmap_atomic(kaddr
, KM_USER0
);
3303 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3310 mark_inode_dirty(inode
);
3316 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3318 return __page_symlink(inode
, symname
, len
,
3319 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3322 const struct inode_operations page_symlink_inode_operations
= {
3323 .readlink
= generic_readlink
,
3324 .follow_link
= page_follow_link_light
,
3325 .put_link
= page_put_link
,
3328 EXPORT_SYMBOL(user_path_at
);
3329 EXPORT_SYMBOL(follow_down_one
);
3330 EXPORT_SYMBOL(follow_down
);
3331 EXPORT_SYMBOL(follow_up
);
3332 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3333 EXPORT_SYMBOL(getname
);
3334 EXPORT_SYMBOL(lock_rename
);
3335 EXPORT_SYMBOL(lookup_one_len
);
3336 EXPORT_SYMBOL(page_follow_link_light
);
3337 EXPORT_SYMBOL(page_put_link
);
3338 EXPORT_SYMBOL(page_readlink
);
3339 EXPORT_SYMBOL(__page_symlink
);
3340 EXPORT_SYMBOL(page_symlink
);
3341 EXPORT_SYMBOL(page_symlink_inode_operations
);
3342 EXPORT_SYMBOL(kern_path_parent
);
3343 EXPORT_SYMBOL(kern_path
);
3344 EXPORT_SYMBOL(vfs_path_lookup
);
3345 EXPORT_SYMBOL(inode_permission
);
3346 EXPORT_SYMBOL(unlock_rename
);
3347 EXPORT_SYMBOL(vfs_create
);
3348 EXPORT_SYMBOL(vfs_follow_link
);
3349 EXPORT_SYMBOL(vfs_link
);
3350 EXPORT_SYMBOL(vfs_mkdir
);
3351 EXPORT_SYMBOL(vfs_mknod
);
3352 EXPORT_SYMBOL(generic_permission
);
3353 EXPORT_SYMBOL(vfs_readlink
);
3354 EXPORT_SYMBOL(vfs_rename
);
3355 EXPORT_SYMBOL(vfs_rmdir
);
3356 EXPORT_SYMBOL(vfs_symlink
);
3357 EXPORT_SYMBOL(vfs_unlink
);
3358 EXPORT_SYMBOL(dentry_unhash
);
3359 EXPORT_SYMBOL(generic_readlink
);