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)
217 * Used to check for read/write/execute permissions on a file.
218 * We use "fsuid" for this, letting us set arbitrary permissions
219 * for filesystem access without changing the "normal" uids which
220 * are used for other things.
222 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
223 * request cannot be satisfied (eg. requires blocking or too much complexity).
224 * It would then be called again in ref-walk mode.
226 int generic_permission(struct inode
*inode
, int mask
)
231 * Do the basic POSIX ACL permission checks.
233 ret
= acl_permission_check(inode
, mask
);
237 if (S_ISDIR(inode
->i_mode
)) {
238 /* DACs are overridable for directories */
239 if (ns_capable(inode_userns(inode
), CAP_DAC_OVERRIDE
))
241 if (!(mask
& MAY_WRITE
))
242 if (ns_capable(inode_userns(inode
), CAP_DAC_READ_SEARCH
))
247 * Read/write DACs are always overridable.
248 * Executable DACs are overridable when there is
249 * at least one exec bit set.
251 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
252 if (ns_capable(inode_userns(inode
), CAP_DAC_OVERRIDE
))
256 * Searching includes executable on directories, else just read.
258 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
259 if (mask
== MAY_READ
)
260 if (ns_capable(inode_userns(inode
), CAP_DAC_READ_SEARCH
))
267 * inode_permission - check for access rights to a given inode
268 * @inode: inode to check permission on
269 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
271 * Used to check for read/write/execute permissions on an inode.
272 * We use "fsuid" for this, letting us set arbitrary permissions
273 * for filesystem access without changing the "normal" uids which
274 * are used for other things.
276 int inode_permission(struct inode
*inode
, int mask
)
280 if (mask
& MAY_WRITE
) {
281 umode_t mode
= inode
->i_mode
;
284 * Nobody gets write access to a read-only fs.
286 if (IS_RDONLY(inode
) &&
287 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
291 * Nobody gets write access to an immutable file.
293 if (IS_IMMUTABLE(inode
))
297 if (inode
->i_op
->permission
)
298 retval
= inode
->i_op
->permission(inode
, mask
);
300 retval
= generic_permission(inode
, mask
);
305 retval
= devcgroup_inode_permission(inode
, mask
);
309 return security_inode_permission(inode
, mask
);
313 * path_get - get a reference to a path
314 * @path: path to get the reference to
316 * Given a path increment the reference count to the dentry and the vfsmount.
318 void path_get(struct path
*path
)
323 EXPORT_SYMBOL(path_get
);
326 * path_put - put a reference to a path
327 * @path: path to put the reference to
329 * Given a path decrement the reference count to the dentry and the vfsmount.
331 void path_put(struct path
*path
)
336 EXPORT_SYMBOL(path_put
);
339 * Path walking has 2 modes, rcu-walk and ref-walk (see
340 * Documentation/filesystems/path-lookup.txt). In situations when we can't
341 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
342 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
343 * mode. Refcounts are grabbed at the last known good point before rcu-walk
344 * got stuck, so ref-walk may continue from there. If this is not successful
345 * (eg. a seqcount has changed), then failure is returned and it's up to caller
346 * to restart the path walk from the beginning in ref-walk mode.
350 * unlazy_walk - try to switch to ref-walk mode.
351 * @nd: nameidata pathwalk data
352 * @dentry: child of nd->path.dentry or NULL
353 * Returns: 0 on success, -ECHILD on failure
355 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
356 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
357 * @nd or NULL. Must be called from rcu-walk context.
359 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
361 struct fs_struct
*fs
= current
->fs
;
362 struct dentry
*parent
= nd
->path
.dentry
;
365 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
366 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
368 spin_lock(&fs
->lock
);
369 if (nd
->root
.mnt
!= fs
->root
.mnt
||
370 nd
->root
.dentry
!= fs
->root
.dentry
)
373 spin_lock(&parent
->d_lock
);
375 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
377 BUG_ON(nd
->inode
!= parent
->d_inode
);
379 if (dentry
->d_parent
!= parent
)
381 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
382 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
385 * If the sequence check on the child dentry passed, then
386 * the child has not been removed from its parent. This
387 * means the parent dentry must be valid and able to take
388 * a reference at this point.
390 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
391 BUG_ON(!parent
->d_count
);
393 spin_unlock(&dentry
->d_lock
);
395 spin_unlock(&parent
->d_lock
);
398 spin_unlock(&fs
->lock
);
400 mntget(nd
->path
.mnt
);
403 br_read_unlock(vfsmount_lock
);
404 nd
->flags
&= ~LOOKUP_RCU
;
408 spin_unlock(&dentry
->d_lock
);
410 spin_unlock(&parent
->d_lock
);
413 spin_unlock(&fs
->lock
);
418 * release_open_intent - free up open intent resources
419 * @nd: pointer to nameidata
421 void release_open_intent(struct nameidata
*nd
)
423 struct file
*file
= nd
->intent
.open
.file
;
425 if (file
&& !IS_ERR(file
)) {
426 if (file
->f_path
.dentry
== NULL
)
433 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
435 return dentry
->d_op
->d_revalidate(dentry
, nd
);
439 * complete_walk - successful completion of path walk
440 * @nd: pointer nameidata
442 * If we had been in RCU mode, drop out of it and legitimize nd->path.
443 * Revalidate the final result, unless we'd already done that during
444 * the path walk or the filesystem doesn't ask for it. Return 0 on
445 * success, -error on failure. In case of failure caller does not
446 * need to drop nd->path.
448 static int complete_walk(struct nameidata
*nd
)
450 struct dentry
*dentry
= nd
->path
.dentry
;
453 if (nd
->flags
& LOOKUP_RCU
) {
454 nd
->flags
&= ~LOOKUP_RCU
;
455 if (!(nd
->flags
& LOOKUP_ROOT
))
457 spin_lock(&dentry
->d_lock
);
458 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
459 spin_unlock(&dentry
->d_lock
);
461 br_read_unlock(vfsmount_lock
);
464 BUG_ON(nd
->inode
!= dentry
->d_inode
);
465 spin_unlock(&dentry
->d_lock
);
466 mntget(nd
->path
.mnt
);
468 br_read_unlock(vfsmount_lock
);
471 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
474 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
477 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
480 /* Note: we do not d_invalidate() */
481 status
= d_revalidate(dentry
, nd
);
492 static __always_inline
void set_root(struct nameidata
*nd
)
495 get_fs_root(current
->fs
, &nd
->root
);
498 static int link_path_walk(const char *, struct nameidata
*);
500 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
503 struct fs_struct
*fs
= current
->fs
;
507 seq
= read_seqcount_begin(&fs
->seq
);
509 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
510 } while (read_seqcount_retry(&fs
->seq
, seq
));
514 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
526 nd
->flags
|= LOOKUP_JUMPED
;
528 nd
->inode
= nd
->path
.dentry
->d_inode
;
530 ret
= link_path_walk(link
, nd
);
534 return PTR_ERR(link
);
537 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
540 if (path
->mnt
!= nd
->path
.mnt
)
544 static inline void path_to_nameidata(const struct path
*path
,
545 struct nameidata
*nd
)
547 if (!(nd
->flags
& LOOKUP_RCU
)) {
548 dput(nd
->path
.dentry
);
549 if (nd
->path
.mnt
!= path
->mnt
)
550 mntput(nd
->path
.mnt
);
552 nd
->path
.mnt
= path
->mnt
;
553 nd
->path
.dentry
= path
->dentry
;
556 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
558 struct inode
*inode
= link
->dentry
->d_inode
;
559 if (!IS_ERR(cookie
) && inode
->i_op
->put_link
)
560 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
564 static __always_inline
int
565 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
568 struct dentry
*dentry
= link
->dentry
;
570 BUG_ON(nd
->flags
& LOOKUP_RCU
);
572 if (link
->mnt
== nd
->path
.mnt
)
575 if (unlikely(current
->total_link_count
>= 40)) {
576 *p
= ERR_PTR(-ELOOP
); /* no ->put_link(), please */
581 current
->total_link_count
++;
583 touch_atime(link
->mnt
, dentry
);
584 nd_set_link(nd
, NULL
);
586 error
= security_inode_follow_link(link
->dentry
, nd
);
588 *p
= ERR_PTR(error
); /* no ->put_link(), please */
593 nd
->last_type
= LAST_BIND
;
594 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
597 char *s
= nd_get_link(nd
);
600 error
= __vfs_follow_link(nd
, s
);
601 else if (nd
->last_type
== LAST_BIND
) {
602 nd
->flags
|= LOOKUP_JUMPED
;
603 nd
->inode
= nd
->path
.dentry
->d_inode
;
604 if (nd
->inode
->i_op
->follow_link
) {
605 /* stepped on a _really_ weird one */
614 static int follow_up_rcu(struct path
*path
)
616 struct vfsmount
*parent
;
617 struct dentry
*mountpoint
;
619 parent
= path
->mnt
->mnt_parent
;
620 if (parent
== path
->mnt
)
622 mountpoint
= path
->mnt
->mnt_mountpoint
;
623 path
->dentry
= mountpoint
;
628 int follow_up(struct path
*path
)
630 struct vfsmount
*parent
;
631 struct dentry
*mountpoint
;
633 br_read_lock(vfsmount_lock
);
634 parent
= path
->mnt
->mnt_parent
;
635 if (parent
== path
->mnt
) {
636 br_read_unlock(vfsmount_lock
);
640 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
641 br_read_unlock(vfsmount_lock
);
643 path
->dentry
= mountpoint
;
650 * Perform an automount
651 * - return -EISDIR to tell follow_managed() to stop and return the path we
654 static int follow_automount(struct path
*path
, unsigned flags
,
657 struct vfsmount
*mnt
;
660 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
663 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
664 * and this is the terminal part of the path.
666 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_PARENT
))
667 return -EISDIR
; /* we actually want to stop here */
669 /* We want to mount if someone is trying to open/create a file of any
670 * type under the mountpoint, wants to traverse through the mountpoint
671 * or wants to open the mounted directory.
673 * We don't want to mount if someone's just doing a stat and they've
674 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
675 * appended a '/' to the name.
677 if (!(flags
& LOOKUP_FOLLOW
) &&
678 !(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
679 LOOKUP_OPEN
| LOOKUP_CREATE
)))
682 current
->total_link_count
++;
683 if (current
->total_link_count
>= 40)
686 mnt
= path
->dentry
->d_op
->d_automount(path
);
689 * The filesystem is allowed to return -EISDIR here to indicate
690 * it doesn't want to automount. For instance, autofs would do
691 * this so that its userspace daemon can mount on this dentry.
693 * However, we can only permit this if it's a terminal point in
694 * the path being looked up; if it wasn't then the remainder of
695 * the path is inaccessible and we should say so.
697 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
702 if (!mnt
) /* mount collision */
706 /* lock_mount() may release path->mnt on error */
710 err
= finish_automount(mnt
, path
);
714 /* Someone else made a mount here whilst we were busy */
719 path
->dentry
= dget(mnt
->mnt_root
);
728 * Handle a dentry that is managed in some way.
729 * - Flagged for transit management (autofs)
730 * - Flagged as mountpoint
731 * - Flagged as automount point
733 * This may only be called in refwalk mode.
735 * Serialization is taken care of in namespace.c
737 static int follow_managed(struct path
*path
, unsigned flags
)
739 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
741 bool need_mntput
= false;
744 /* Given that we're not holding a lock here, we retain the value in a
745 * local variable for each dentry as we look at it so that we don't see
746 * the components of that value change under us */
747 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
748 managed
&= DCACHE_MANAGED_DENTRY
,
749 unlikely(managed
!= 0)) {
750 /* Allow the filesystem to manage the transit without i_mutex
752 if (managed
& DCACHE_MANAGE_TRANSIT
) {
753 BUG_ON(!path
->dentry
->d_op
);
754 BUG_ON(!path
->dentry
->d_op
->d_manage
);
755 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
760 /* Transit to a mounted filesystem. */
761 if (managed
& DCACHE_MOUNTED
) {
762 struct vfsmount
*mounted
= lookup_mnt(path
);
768 path
->dentry
= dget(mounted
->mnt_root
);
773 /* Something is mounted on this dentry in another
774 * namespace and/or whatever was mounted there in this
775 * namespace got unmounted before we managed to get the
779 /* Handle an automount point */
780 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
781 ret
= follow_automount(path
, flags
, &need_mntput
);
787 /* We didn't change the current path point */
791 if (need_mntput
&& path
->mnt
== mnt
)
798 int follow_down_one(struct path
*path
)
800 struct vfsmount
*mounted
;
802 mounted
= lookup_mnt(path
);
807 path
->dentry
= dget(mounted
->mnt_root
);
813 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
815 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
816 dentry
->d_op
->d_manage(dentry
, true) < 0);
820 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
821 * we meet a managed dentry that would need blocking.
823 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
824 struct inode
**inode
)
827 struct vfsmount
*mounted
;
829 * Don't forget we might have a non-mountpoint managed dentry
830 * that wants to block transit.
832 if (unlikely(managed_dentry_might_block(path
->dentry
)))
835 if (!d_mountpoint(path
->dentry
))
838 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
842 path
->dentry
= mounted
->mnt_root
;
843 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
845 * Update the inode too. We don't need to re-check the
846 * dentry sequence number here after this d_inode read,
847 * because a mount-point is always pinned.
849 *inode
= path
->dentry
->d_inode
;
854 static void follow_mount_rcu(struct nameidata
*nd
)
856 while (d_mountpoint(nd
->path
.dentry
)) {
857 struct vfsmount
*mounted
;
858 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
861 nd
->path
.mnt
= mounted
;
862 nd
->path
.dentry
= mounted
->mnt_root
;
863 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
867 static int follow_dotdot_rcu(struct nameidata
*nd
)
872 if (nd
->path
.dentry
== nd
->root
.dentry
&&
873 nd
->path
.mnt
== nd
->root
.mnt
) {
876 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
877 struct dentry
*old
= nd
->path
.dentry
;
878 struct dentry
*parent
= old
->d_parent
;
881 seq
= read_seqcount_begin(&parent
->d_seq
);
882 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
884 nd
->path
.dentry
= parent
;
888 if (!follow_up_rcu(&nd
->path
))
890 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
892 follow_mount_rcu(nd
);
893 nd
->inode
= nd
->path
.dentry
->d_inode
;
897 nd
->flags
&= ~LOOKUP_RCU
;
898 if (!(nd
->flags
& LOOKUP_ROOT
))
901 br_read_unlock(vfsmount_lock
);
906 * Follow down to the covering mount currently visible to userspace. At each
907 * point, the filesystem owning that dentry may be queried as to whether the
908 * caller is permitted to proceed or not.
910 int follow_down(struct path
*path
)
915 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
916 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
917 /* Allow the filesystem to manage the transit without i_mutex
920 * We indicate to the filesystem if someone is trying to mount
921 * something here. This gives autofs the chance to deny anyone
922 * other than its daemon the right to mount on its
925 * The filesystem may sleep at this point.
927 if (managed
& DCACHE_MANAGE_TRANSIT
) {
928 BUG_ON(!path
->dentry
->d_op
);
929 BUG_ON(!path
->dentry
->d_op
->d_manage
);
930 ret
= path
->dentry
->d_op
->d_manage(
931 path
->dentry
, false);
933 return ret
== -EISDIR
? 0 : ret
;
936 /* Transit to a mounted filesystem. */
937 if (managed
& DCACHE_MOUNTED
) {
938 struct vfsmount
*mounted
= lookup_mnt(path
);
944 path
->dentry
= dget(mounted
->mnt_root
);
948 /* Don't handle automount points here */
955 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
957 static void follow_mount(struct path
*path
)
959 while (d_mountpoint(path
->dentry
)) {
960 struct vfsmount
*mounted
= lookup_mnt(path
);
966 path
->dentry
= dget(mounted
->mnt_root
);
970 static void follow_dotdot(struct nameidata
*nd
)
975 struct dentry
*old
= nd
->path
.dentry
;
977 if (nd
->path
.dentry
== nd
->root
.dentry
&&
978 nd
->path
.mnt
== nd
->root
.mnt
) {
981 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
982 /* rare case of legitimate dget_parent()... */
983 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
987 if (!follow_up(&nd
->path
))
990 follow_mount(&nd
->path
);
991 nd
->inode
= nd
->path
.dentry
->d_inode
;
995 * Allocate a dentry with name and parent, and perform a parent
996 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
997 * on error. parent->d_inode->i_mutex must be held. d_lookup must
998 * have verified that no child exists while under i_mutex.
1000 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1001 struct qstr
*name
, struct nameidata
*nd
)
1003 struct inode
*inode
= parent
->d_inode
;
1004 struct dentry
*dentry
;
1007 /* Don't create child dentry for a dead directory. */
1008 if (unlikely(IS_DEADDIR(inode
)))
1009 return ERR_PTR(-ENOENT
);
1011 dentry
= d_alloc(parent
, name
);
1012 if (unlikely(!dentry
))
1013 return ERR_PTR(-ENOMEM
);
1015 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1016 if (unlikely(old
)) {
1024 * We already have a dentry, but require a lookup to be performed on the parent
1025 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1026 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1027 * child exists while under i_mutex.
1029 static struct dentry
*d_inode_lookup(struct dentry
*parent
, struct dentry
*dentry
,
1030 struct nameidata
*nd
)
1032 struct inode
*inode
= parent
->d_inode
;
1035 /* Don't create child dentry for a dead directory. */
1036 if (unlikely(IS_DEADDIR(inode
)))
1037 return ERR_PTR(-ENOENT
);
1039 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1040 if (unlikely(old
)) {
1048 * It's more convoluted than I'd like it to be, but... it's still fairly
1049 * small and for now I'd prefer to have fast path as straight as possible.
1050 * It _is_ time-critical.
1052 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1053 struct path
*path
, struct inode
**inode
)
1055 struct vfsmount
*mnt
= nd
->path
.mnt
;
1056 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1062 * Rename seqlock is not required here because in the off chance
1063 * of a false negative due to a concurrent rename, we're going to
1064 * do the non-racy lookup, below.
1066 if (nd
->flags
& LOOKUP_RCU
) {
1069 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1073 /* Memory barrier in read_seqcount_begin of child is enough */
1074 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1078 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1079 status
= d_revalidate(dentry
, nd
);
1080 if (unlikely(status
<= 0)) {
1081 if (status
!= -ECHILD
)
1086 if (unlikely(d_need_lookup(dentry
)))
1089 path
->dentry
= dentry
;
1090 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1092 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1096 if (unlazy_walk(nd
, dentry
))
1099 dentry
= __d_lookup(parent
, name
);
1102 if (dentry
&& unlikely(d_need_lookup(dentry
))) {
1107 if (unlikely(!dentry
)) {
1108 struct inode
*dir
= parent
->d_inode
;
1109 BUG_ON(nd
->inode
!= dir
);
1111 mutex_lock(&dir
->i_mutex
);
1112 dentry
= d_lookup(parent
, name
);
1113 if (likely(!dentry
)) {
1114 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1115 if (IS_ERR(dentry
)) {
1116 mutex_unlock(&dir
->i_mutex
);
1117 return PTR_ERR(dentry
);
1122 } else if (unlikely(d_need_lookup(dentry
))) {
1123 dentry
= d_inode_lookup(parent
, dentry
, nd
);
1124 if (IS_ERR(dentry
)) {
1125 mutex_unlock(&dir
->i_mutex
);
1126 return PTR_ERR(dentry
);
1132 mutex_unlock(&dir
->i_mutex
);
1134 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1135 status
= d_revalidate(dentry
, nd
);
1136 if (unlikely(status
<= 0)) {
1141 if (!d_invalidate(dentry
)) {
1150 path
->dentry
= dentry
;
1151 err
= follow_managed(path
, nd
->flags
);
1152 if (unlikely(err
< 0)) {
1153 path_put_conditional(path
, nd
);
1156 *inode
= path
->dentry
->d_inode
;
1160 static inline int may_lookup(struct nameidata
*nd
)
1162 if (nd
->flags
& LOOKUP_RCU
) {
1163 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1166 if (unlazy_walk(nd
, NULL
))
1169 return inode_permission(nd
->inode
, MAY_EXEC
);
1172 static inline int handle_dots(struct nameidata
*nd
, int type
)
1174 if (type
== LAST_DOTDOT
) {
1175 if (nd
->flags
& LOOKUP_RCU
) {
1176 if (follow_dotdot_rcu(nd
))
1184 static void terminate_walk(struct nameidata
*nd
)
1186 if (!(nd
->flags
& LOOKUP_RCU
)) {
1187 path_put(&nd
->path
);
1189 nd
->flags
&= ~LOOKUP_RCU
;
1190 if (!(nd
->flags
& LOOKUP_ROOT
))
1191 nd
->root
.mnt
= NULL
;
1193 br_read_unlock(vfsmount_lock
);
1197 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1198 struct qstr
*name
, int type
, int follow
)
1200 struct inode
*inode
;
1203 * "." and ".." are special - ".." especially so because it has
1204 * to be able to know about the current root directory and
1205 * parent relationships.
1207 if (unlikely(type
!= LAST_NORM
))
1208 return handle_dots(nd
, type
);
1209 err
= do_lookup(nd
, name
, path
, &inode
);
1210 if (unlikely(err
)) {
1215 path_to_nameidata(path
, nd
);
1219 if (unlikely(inode
->i_op
->follow_link
) && follow
) {
1220 if (nd
->flags
& LOOKUP_RCU
) {
1221 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1226 BUG_ON(inode
!= path
->dentry
->d_inode
);
1229 path_to_nameidata(path
, nd
);
1235 * This limits recursive symlink follows to 8, while
1236 * limiting consecutive symlinks to 40.
1238 * Without that kind of total limit, nasty chains of consecutive
1239 * symlinks can cause almost arbitrarily long lookups.
1241 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1245 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1246 path_put_conditional(path
, nd
);
1247 path_put(&nd
->path
);
1250 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1253 current
->link_count
++;
1256 struct path link
= *path
;
1259 res
= follow_link(&link
, nd
, &cookie
);
1261 res
= walk_component(nd
, path
, &nd
->last
,
1262 nd
->last_type
, LOOKUP_FOLLOW
);
1263 put_link(nd
, &link
, cookie
);
1266 current
->link_count
--;
1273 * This is the basic name resolution function, turning a pathname into
1274 * the final dentry. We expect 'base' to be positive and a directory.
1276 * Returns 0 and nd will have valid dentry and mnt on success.
1277 * Returns error and drops reference to input namei data on failure.
1279 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1289 /* At this point we know we have a real path component. */
1296 err
= may_lookup(nd
);
1301 c
= *(const unsigned char *)name
;
1303 hash
= init_name_hash();
1306 hash
= partial_name_hash(c
, hash
);
1307 c
= *(const unsigned char *)name
;
1308 } while (c
&& (c
!= '/'));
1309 this.len
= name
- (const char *) this.name
;
1310 this.hash
= end_name_hash(hash
);
1313 if (this.name
[0] == '.') switch (this.len
) {
1315 if (this.name
[1] == '.') {
1317 nd
->flags
|= LOOKUP_JUMPED
;
1323 if (likely(type
== LAST_NORM
)) {
1324 struct dentry
*parent
= nd
->path
.dentry
;
1325 nd
->flags
&= ~LOOKUP_JUMPED
;
1326 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1327 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1334 /* remove trailing slashes? */
1336 goto last_component
;
1337 while (*++name
== '/');
1339 goto last_component
;
1341 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1346 err
= nested_symlink(&next
, nd
);
1351 if (!nd
->inode
->i_op
->lookup
)
1354 /* here ends the main loop */
1358 nd
->last_type
= type
;
1365 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1366 struct nameidata
*nd
, struct file
**fp
)
1372 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1373 nd
->flags
= flags
| LOOKUP_JUMPED
;
1375 if (flags
& LOOKUP_ROOT
) {
1376 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1378 if (!inode
->i_op
->lookup
)
1380 retval
= inode_permission(inode
, MAY_EXEC
);
1384 nd
->path
= nd
->root
;
1386 if (flags
& LOOKUP_RCU
) {
1387 br_read_lock(vfsmount_lock
);
1389 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1391 path_get(&nd
->path
);
1396 nd
->root
.mnt
= NULL
;
1399 if (flags
& LOOKUP_RCU
) {
1400 br_read_lock(vfsmount_lock
);
1405 path_get(&nd
->root
);
1407 nd
->path
= nd
->root
;
1408 } else if (dfd
== AT_FDCWD
) {
1409 if (flags
& LOOKUP_RCU
) {
1410 struct fs_struct
*fs
= current
->fs
;
1413 br_read_lock(vfsmount_lock
);
1417 seq
= read_seqcount_begin(&fs
->seq
);
1419 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1420 } while (read_seqcount_retry(&fs
->seq
, seq
));
1422 get_fs_pwd(current
->fs
, &nd
->path
);
1425 struct dentry
*dentry
;
1427 file
= fget_raw_light(dfd
, &fput_needed
);
1432 dentry
= file
->f_path
.dentry
;
1436 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1439 retval
= inode_permission(dentry
->d_inode
, MAY_EXEC
);
1444 nd
->path
= file
->f_path
;
1445 if (flags
& LOOKUP_RCU
) {
1448 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1449 br_read_lock(vfsmount_lock
);
1452 path_get(&file
->f_path
);
1453 fput_light(file
, fput_needed
);
1457 nd
->inode
= nd
->path
.dentry
->d_inode
;
1461 fput_light(file
, fput_needed
);
1466 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1468 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1469 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1471 nd
->flags
&= ~LOOKUP_PARENT
;
1472 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1473 nd
->flags
& LOOKUP_FOLLOW
);
1476 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1477 static int path_lookupat(int dfd
, const char *name
,
1478 unsigned int flags
, struct nameidata
*nd
)
1480 struct file
*base
= NULL
;
1485 * Path walking is largely split up into 2 different synchronisation
1486 * schemes, rcu-walk and ref-walk (explained in
1487 * Documentation/filesystems/path-lookup.txt). These share much of the
1488 * path walk code, but some things particularly setup, cleanup, and
1489 * following mounts are sufficiently divergent that functions are
1490 * duplicated. Typically there is a function foo(), and its RCU
1491 * analogue, foo_rcu().
1493 * -ECHILD is the error number of choice (just to avoid clashes) that
1494 * is returned if some aspect of an rcu-walk fails. Such an error must
1495 * be handled by restarting a traditional ref-walk (which will always
1496 * be able to complete).
1498 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1503 current
->total_link_count
= 0;
1504 err
= link_path_walk(name
, nd
);
1506 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1507 err
= lookup_last(nd
, &path
);
1510 struct path link
= path
;
1511 nd
->flags
|= LOOKUP_PARENT
;
1512 err
= follow_link(&link
, nd
, &cookie
);
1514 err
= lookup_last(nd
, &path
);
1515 put_link(nd
, &link
, cookie
);
1520 err
= complete_walk(nd
);
1522 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1523 if (!nd
->inode
->i_op
->lookup
) {
1524 path_put(&nd
->path
);
1532 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1533 path_put(&nd
->root
);
1534 nd
->root
.mnt
= NULL
;
1539 static int do_path_lookup(int dfd
, const char *name
,
1540 unsigned int flags
, struct nameidata
*nd
)
1542 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1543 if (unlikely(retval
== -ECHILD
))
1544 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1545 if (unlikely(retval
== -ESTALE
))
1546 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1548 if (likely(!retval
)) {
1549 if (unlikely(!audit_dummy_context())) {
1550 if (nd
->path
.dentry
&& nd
->inode
)
1551 audit_inode(name
, nd
->path
.dentry
);
1557 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1559 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1562 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1564 struct nameidata nd
;
1565 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1572 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1573 * @dentry: pointer to dentry of the base directory
1574 * @mnt: pointer to vfs mount of the base directory
1575 * @name: pointer to file name
1576 * @flags: lookup flags
1577 * @path: pointer to struct path to fill
1579 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1580 const char *name
, unsigned int flags
,
1583 struct nameidata nd
;
1585 nd
.root
.dentry
= dentry
;
1587 BUG_ON(flags
& LOOKUP_PARENT
);
1588 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1589 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
1595 static struct dentry
*__lookup_hash(struct qstr
*name
,
1596 struct dentry
*base
, struct nameidata
*nd
)
1598 struct inode
*inode
= base
->d_inode
;
1599 struct dentry
*dentry
;
1602 err
= inode_permission(inode
, MAY_EXEC
);
1604 return ERR_PTR(err
);
1607 * Don't bother with __d_lookup: callers are for creat as
1608 * well as unlink, so a lot of the time it would cost
1611 dentry
= d_lookup(base
, name
);
1613 if (dentry
&& d_need_lookup(dentry
)) {
1615 * __lookup_hash is called with the parent dir's i_mutex already
1616 * held, so we are good to go here.
1618 dentry
= d_inode_lookup(base
, dentry
, nd
);
1623 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1624 int status
= d_revalidate(dentry
, nd
);
1625 if (unlikely(status
<= 0)) {
1627 * The dentry failed validation.
1628 * If d_revalidate returned 0 attempt to invalidate
1629 * the dentry otherwise d_revalidate is asking us
1630 * to return a fail status.
1634 return ERR_PTR(status
);
1635 } else if (!d_invalidate(dentry
)) {
1643 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1649 * Restricted form of lookup. Doesn't follow links, single-component only,
1650 * needs parent already locked. Doesn't follow mounts.
1653 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1655 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1659 * lookup_one_len - filesystem helper to lookup single pathname component
1660 * @name: pathname component to lookup
1661 * @base: base directory to lookup from
1662 * @len: maximum length @len should be interpreted to
1664 * Note that this routine is purely a helper for filesystem usage and should
1665 * not be called by generic code. Also note that by using this function the
1666 * nameidata argument is passed to the filesystem methods and a filesystem
1667 * using this helper needs to be prepared for that.
1669 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1675 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1680 return ERR_PTR(-EACCES
);
1682 hash
= init_name_hash();
1684 c
= *(const unsigned char *)name
++;
1685 if (c
== '/' || c
== '\0')
1686 return ERR_PTR(-EACCES
);
1687 hash
= partial_name_hash(c
, hash
);
1689 this.hash
= end_name_hash(hash
);
1691 * See if the low-level filesystem might want
1692 * to use its own hash..
1694 if (base
->d_flags
& DCACHE_OP_HASH
) {
1695 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1697 return ERR_PTR(err
);
1700 return __lookup_hash(&this, base
, NULL
);
1703 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1706 struct nameidata nd
;
1707 char *tmp
= getname_flags(name
, flags
);
1708 int err
= PTR_ERR(tmp
);
1711 BUG_ON(flags
& LOOKUP_PARENT
);
1713 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1721 static int user_path_parent(int dfd
, const char __user
*path
,
1722 struct nameidata
*nd
, char **name
)
1724 char *s
= getname(path
);
1730 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1740 * It's inline, so penalty for filesystems that don't use sticky bit is
1743 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1745 uid_t fsuid
= current_fsuid();
1747 if (!(dir
->i_mode
& S_ISVTX
))
1749 if (current_user_ns() != inode_userns(inode
))
1751 if (inode
->i_uid
== fsuid
)
1753 if (dir
->i_uid
== fsuid
)
1757 return !ns_capable(inode_userns(inode
), CAP_FOWNER
);
1761 * Check whether we can remove a link victim from directory dir, check
1762 * whether the type of victim is right.
1763 * 1. We can't do it if dir is read-only (done in permission())
1764 * 2. We should have write and exec permissions on dir
1765 * 3. We can't remove anything from append-only dir
1766 * 4. We can't do anything with immutable dir (done in permission())
1767 * 5. If the sticky bit on dir is set we should either
1768 * a. be owner of dir, or
1769 * b. be owner of victim, or
1770 * c. have CAP_FOWNER capability
1771 * 6. If the victim is append-only or immutable we can't do antyhing with
1772 * links pointing to it.
1773 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1774 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1775 * 9. We can't remove a root or mountpoint.
1776 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1777 * nfs_async_unlink().
1779 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1783 if (!victim
->d_inode
)
1786 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1787 audit_inode_child(victim
, dir
);
1789 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1794 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1795 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1798 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1800 if (IS_ROOT(victim
))
1802 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1804 if (IS_DEADDIR(dir
))
1806 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1811 /* Check whether we can create an object with dentry child in directory
1813 * 1. We can't do it if child already exists (open has special treatment for
1814 * this case, but since we are inlined it's OK)
1815 * 2. We can't do it if dir is read-only (done in permission())
1816 * 3. We should have write and exec permissions on dir
1817 * 4. We can't do it if dir is immutable (done in permission())
1819 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1823 if (IS_DEADDIR(dir
))
1825 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1829 * p1 and p2 should be directories on the same fs.
1831 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1836 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1840 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1842 p
= d_ancestor(p2
, p1
);
1844 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1845 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1849 p
= d_ancestor(p1
, p2
);
1851 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1852 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1856 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1857 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1861 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1863 mutex_unlock(&p1
->d_inode
->i_mutex
);
1865 mutex_unlock(&p2
->d_inode
->i_mutex
);
1866 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1870 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1871 struct nameidata
*nd
)
1873 int error
= may_create(dir
, dentry
);
1878 if (!dir
->i_op
->create
)
1879 return -EACCES
; /* shouldn't it be ENOSYS? */
1882 error
= security_inode_create(dir
, dentry
, mode
);
1885 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1887 fsnotify_create(dir
, dentry
);
1891 static int may_open(struct path
*path
, int acc_mode
, int flag
)
1893 struct dentry
*dentry
= path
->dentry
;
1894 struct inode
*inode
= dentry
->d_inode
;
1904 switch (inode
->i_mode
& S_IFMT
) {
1908 if (acc_mode
& MAY_WRITE
)
1913 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
1922 error
= inode_permission(inode
, acc_mode
);
1927 * An append-only file must be opened in append mode for writing.
1929 if (IS_APPEND(inode
)) {
1930 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
1936 /* O_NOATIME can only be set by the owner or superuser */
1937 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
1941 * Ensure there are no outstanding leases on the file.
1943 return break_lease(inode
, flag
);
1946 static int handle_truncate(struct file
*filp
)
1948 struct path
*path
= &filp
->f_path
;
1949 struct inode
*inode
= path
->dentry
->d_inode
;
1950 int error
= get_write_access(inode
);
1954 * Refuse to truncate files with mandatory locks held on them.
1956 error
= locks_verify_locked(inode
);
1958 error
= security_path_truncate(path
);
1960 error
= do_truncate(path
->dentry
, 0,
1961 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
1964 put_write_access(inode
);
1968 static inline int open_to_namei_flags(int flag
)
1970 if ((flag
& O_ACCMODE
) == 3)
1976 * Handle the last step of open()
1978 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
1979 const struct open_flags
*op
, const char *pathname
)
1981 struct dentry
*dir
= nd
->path
.dentry
;
1982 struct dentry
*dentry
;
1983 int open_flag
= op
->open_flag
;
1984 int will_truncate
= open_flag
& O_TRUNC
;
1986 int acc_mode
= op
->acc_mode
;
1990 nd
->flags
&= ~LOOKUP_PARENT
;
1991 nd
->flags
|= op
->intent
;
1993 switch (nd
->last_type
) {
1996 error
= handle_dots(nd
, nd
->last_type
);
1998 return ERR_PTR(error
);
2001 error
= complete_walk(nd
);
2003 return ERR_PTR(error
);
2004 audit_inode(pathname
, nd
->path
.dentry
);
2005 if (open_flag
& O_CREAT
) {
2011 error
= complete_walk(nd
);
2013 return ERR_PTR(error
);
2014 audit_inode(pathname
, dir
);
2018 if (!(open_flag
& O_CREAT
)) {
2020 if (nd
->last
.name
[nd
->last
.len
])
2021 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2022 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2024 /* we _can_ be in RCU mode here */
2025 error
= walk_component(nd
, path
, &nd
->last
, LAST_NORM
,
2028 return ERR_PTR(error
);
2029 if (error
) /* symlink */
2032 error
= complete_walk(nd
);
2034 return ERR_PTR(-ECHILD
);
2037 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2038 if (!nd
->inode
->i_op
->lookup
)
2041 audit_inode(pathname
, nd
->path
.dentry
);
2045 /* create side of things */
2046 error
= complete_walk(nd
);
2048 return ERR_PTR(error
);
2050 audit_inode(pathname
, dir
);
2052 /* trailing slashes? */
2053 if (nd
->last
.name
[nd
->last
.len
])
2056 mutex_lock(&dir
->d_inode
->i_mutex
);
2058 dentry
= lookup_hash(nd
);
2059 error
= PTR_ERR(dentry
);
2060 if (IS_ERR(dentry
)) {
2061 mutex_unlock(&dir
->d_inode
->i_mutex
);
2065 path
->dentry
= dentry
;
2066 path
->mnt
= nd
->path
.mnt
;
2068 /* Negative dentry, just create the file */
2069 if (!dentry
->d_inode
) {
2070 int mode
= op
->mode
;
2071 if (!IS_POSIXACL(dir
->d_inode
))
2072 mode
&= ~current_umask();
2074 * This write is needed to ensure that a
2075 * rw->ro transition does not occur between
2076 * the time when the file is created and when
2077 * a permanent write count is taken through
2078 * the 'struct file' in nameidata_to_filp().
2080 error
= mnt_want_write(nd
->path
.mnt
);
2082 goto exit_mutex_unlock
;
2084 /* Don't check for write permission, don't truncate */
2085 open_flag
&= ~O_TRUNC
;
2087 acc_mode
= MAY_OPEN
;
2088 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2090 goto exit_mutex_unlock
;
2091 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2093 goto exit_mutex_unlock
;
2094 mutex_unlock(&dir
->d_inode
->i_mutex
);
2095 dput(nd
->path
.dentry
);
2096 nd
->path
.dentry
= dentry
;
2101 * It already exists.
2103 mutex_unlock(&dir
->d_inode
->i_mutex
);
2104 audit_inode(pathname
, path
->dentry
);
2107 if (open_flag
& O_EXCL
)
2110 error
= follow_managed(path
, nd
->flags
);
2115 if (!path
->dentry
->d_inode
)
2118 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2121 path_to_nameidata(path
, nd
);
2122 nd
->inode
= path
->dentry
->d_inode
;
2124 if (S_ISDIR(nd
->inode
->i_mode
))
2127 if (!S_ISREG(nd
->inode
->i_mode
))
2130 if (will_truncate
) {
2131 error
= mnt_want_write(nd
->path
.mnt
);
2137 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2140 filp
= nameidata_to_filp(nd
);
2141 if (!IS_ERR(filp
)) {
2142 error
= ima_file_check(filp
, op
->acc_mode
);
2145 filp
= ERR_PTR(error
);
2148 if (!IS_ERR(filp
)) {
2149 if (will_truncate
) {
2150 error
= handle_truncate(filp
);
2153 filp
= ERR_PTR(error
);
2159 mnt_drop_write(nd
->path
.mnt
);
2160 path_put(&nd
->path
);
2164 mutex_unlock(&dir
->d_inode
->i_mutex
);
2166 path_put_conditional(path
, nd
);
2168 filp
= ERR_PTR(error
);
2172 static struct file
*path_openat(int dfd
, const char *pathname
,
2173 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2175 struct file
*base
= NULL
;
2180 filp
= get_empty_filp();
2182 return ERR_PTR(-ENFILE
);
2184 filp
->f_flags
= op
->open_flag
;
2185 nd
->intent
.open
.file
= filp
;
2186 nd
->intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2187 nd
->intent
.open
.create_mode
= op
->mode
;
2189 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2190 if (unlikely(error
))
2193 current
->total_link_count
= 0;
2194 error
= link_path_walk(pathname
, nd
);
2195 if (unlikely(error
))
2198 filp
= do_last(nd
, &path
, op
, pathname
);
2199 while (unlikely(!filp
)) { /* trailing symlink */
2200 struct path link
= path
;
2202 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2203 path_put_conditional(&path
, nd
);
2204 path_put(&nd
->path
);
2205 filp
= ERR_PTR(-ELOOP
);
2208 nd
->flags
|= LOOKUP_PARENT
;
2209 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2210 error
= follow_link(&link
, nd
, &cookie
);
2211 if (unlikely(error
))
2212 filp
= ERR_PTR(error
);
2214 filp
= do_last(nd
, &path
, op
, pathname
);
2215 put_link(nd
, &link
, cookie
);
2218 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2219 path_put(&nd
->root
);
2222 release_open_intent(nd
);
2226 filp
= ERR_PTR(error
);
2230 struct file
*do_filp_open(int dfd
, const char *pathname
,
2231 const struct open_flags
*op
, int flags
)
2233 struct nameidata nd
;
2236 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2237 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2238 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2239 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2240 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2244 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2245 const char *name
, const struct open_flags
*op
, int flags
)
2247 struct nameidata nd
;
2251 nd
.root
.dentry
= dentry
;
2253 flags
|= LOOKUP_ROOT
;
2255 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2256 return ERR_PTR(-ELOOP
);
2258 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2259 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2260 file
= path_openat(-1, name
, &nd
, op
, flags
);
2261 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2262 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2266 struct dentry
*kern_path_create(int dfd
, const char *pathname
, struct path
*path
, int is_dir
)
2268 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2269 struct nameidata nd
;
2270 int error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
, &nd
);
2272 return ERR_PTR(error
);
2275 * Yucky last component or no last component at all?
2276 * (foo/., foo/.., /////)
2278 if (nd
.last_type
!= LAST_NORM
)
2280 nd
.flags
&= ~LOOKUP_PARENT
;
2281 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2282 nd
.intent
.open
.flags
= O_EXCL
;
2285 * Do the final lookup.
2287 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2288 dentry
= lookup_hash(&nd
);
2292 if (dentry
->d_inode
)
2295 * Special case - lookup gave negative, but... we had foo/bar/
2296 * From the vfs_mknod() POV we just have a negative dentry -
2297 * all is fine. Let's be bastards - you had / on the end, you've
2298 * been asking for (non-existent) directory. -ENOENT for you.
2300 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
2302 dentry
= ERR_PTR(-ENOENT
);
2309 dentry
= ERR_PTR(-EEXIST
);
2311 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2316 EXPORT_SYMBOL(kern_path_create
);
2318 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
, struct path
*path
, int is_dir
)
2320 char *tmp
= getname(pathname
);
2323 return ERR_CAST(tmp
);
2324 res
= kern_path_create(dfd
, tmp
, path
, is_dir
);
2328 EXPORT_SYMBOL(user_path_create
);
2330 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2332 int error
= may_create(dir
, dentry
);
2337 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) &&
2338 !ns_capable(inode_userns(dir
), CAP_MKNOD
))
2341 if (!dir
->i_op
->mknod
)
2344 error
= devcgroup_inode_mknod(mode
, dev
);
2348 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2352 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2354 fsnotify_create(dir
, dentry
);
2358 static int may_mknod(mode_t mode
)
2360 switch (mode
& S_IFMT
) {
2366 case 0: /* zero mode translates to S_IFREG */
2375 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2378 struct dentry
*dentry
;
2385 dentry
= user_path_create(dfd
, filename
, &path
, 0);
2387 return PTR_ERR(dentry
);
2389 if (!IS_POSIXACL(path
.dentry
->d_inode
))
2390 mode
&= ~current_umask();
2391 error
= may_mknod(mode
);
2394 error
= mnt_want_write(path
.mnt
);
2397 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
2399 goto out_drop_write
;
2400 switch (mode
& S_IFMT
) {
2401 case 0: case S_IFREG
:
2402 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,NULL
);
2404 case S_IFCHR
: case S_IFBLK
:
2405 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
2406 new_decode_dev(dev
));
2408 case S_IFIFO
: case S_IFSOCK
:
2409 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
2413 mnt_drop_write(path
.mnt
);
2416 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2422 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2424 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2427 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2429 int error
= may_create(dir
, dentry
);
2434 if (!dir
->i_op
->mkdir
)
2437 mode
&= (S_IRWXUGO
|S_ISVTX
);
2438 error
= security_inode_mkdir(dir
, dentry
, mode
);
2442 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2444 fsnotify_mkdir(dir
, dentry
);
2448 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2450 struct dentry
*dentry
;
2454 dentry
= user_path_create(dfd
, pathname
, &path
, 1);
2456 return PTR_ERR(dentry
);
2458 if (!IS_POSIXACL(path
.dentry
->d_inode
))
2459 mode
&= ~current_umask();
2460 error
= mnt_want_write(path
.mnt
);
2463 error
= security_path_mkdir(&path
, dentry
, mode
);
2465 goto out_drop_write
;
2466 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
2468 mnt_drop_write(path
.mnt
);
2471 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2476 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2478 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2482 * The dentry_unhash() helper will try to drop the dentry early: we
2483 * should have a usage count of 2 if we're the only user of this
2484 * dentry, and if that is true (possibly after pruning the dcache),
2485 * then we drop the dentry now.
2487 * A low-level filesystem can, if it choses, legally
2490 * if (!d_unhashed(dentry))
2493 * if it cannot handle the case of removing a directory
2494 * that is still in use by something else..
2496 void dentry_unhash(struct dentry
*dentry
)
2498 shrink_dcache_parent(dentry
);
2499 spin_lock(&dentry
->d_lock
);
2500 if (dentry
->d_count
== 1)
2502 spin_unlock(&dentry
->d_lock
);
2505 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2507 int error
= may_delete(dir
, dentry
, 1);
2512 if (!dir
->i_op
->rmdir
)
2515 mutex_lock(&dentry
->d_inode
->i_mutex
);
2518 if (d_mountpoint(dentry
))
2521 error
= security_inode_rmdir(dir
, dentry
);
2525 shrink_dcache_parent(dentry
);
2526 error
= dir
->i_op
->rmdir(dir
, dentry
);
2530 dentry
->d_inode
->i_flags
|= S_DEAD
;
2534 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2540 static long do_rmdir(int dfd
, const char __user
*pathname
)
2544 struct dentry
*dentry
;
2545 struct nameidata nd
;
2547 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2551 switch(nd
.last_type
) {
2563 nd
.flags
&= ~LOOKUP_PARENT
;
2565 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2566 dentry
= lookup_hash(&nd
);
2567 error
= PTR_ERR(dentry
);
2570 if (!dentry
->d_inode
) {
2574 error
= mnt_want_write(nd
.path
.mnt
);
2577 error
= security_path_rmdir(&nd
.path
, dentry
);
2580 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2582 mnt_drop_write(nd
.path
.mnt
);
2586 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2593 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2595 return do_rmdir(AT_FDCWD
, pathname
);
2598 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2600 int error
= may_delete(dir
, dentry
, 0);
2605 if (!dir
->i_op
->unlink
)
2608 mutex_lock(&dentry
->d_inode
->i_mutex
);
2609 if (d_mountpoint(dentry
))
2612 error
= security_inode_unlink(dir
, dentry
);
2614 error
= dir
->i_op
->unlink(dir
, dentry
);
2619 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2621 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2622 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2623 fsnotify_link_count(dentry
->d_inode
);
2631 * Make sure that the actual truncation of the file will occur outside its
2632 * directory's i_mutex. Truncate can take a long time if there is a lot of
2633 * writeout happening, and we don't want to prevent access to the directory
2634 * while waiting on the I/O.
2636 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2640 struct dentry
*dentry
;
2641 struct nameidata nd
;
2642 struct inode
*inode
= NULL
;
2644 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2649 if (nd
.last_type
!= LAST_NORM
)
2652 nd
.flags
&= ~LOOKUP_PARENT
;
2654 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2655 dentry
= lookup_hash(&nd
);
2656 error
= PTR_ERR(dentry
);
2657 if (!IS_ERR(dentry
)) {
2658 /* Why not before? Because we want correct error value */
2659 if (nd
.last
.name
[nd
.last
.len
])
2661 inode
= dentry
->d_inode
;
2665 error
= mnt_want_write(nd
.path
.mnt
);
2668 error
= security_path_unlink(&nd
.path
, dentry
);
2671 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2673 mnt_drop_write(nd
.path
.mnt
);
2677 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2679 iput(inode
); /* truncate the inode here */
2686 error
= !dentry
->d_inode
? -ENOENT
:
2687 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2691 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2693 if ((flag
& ~AT_REMOVEDIR
) != 0)
2696 if (flag
& AT_REMOVEDIR
)
2697 return do_rmdir(dfd
, pathname
);
2699 return do_unlinkat(dfd
, pathname
);
2702 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2704 return do_unlinkat(AT_FDCWD
, pathname
);
2707 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2709 int error
= may_create(dir
, dentry
);
2714 if (!dir
->i_op
->symlink
)
2717 error
= security_inode_symlink(dir
, dentry
, oldname
);
2721 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2723 fsnotify_create(dir
, dentry
);
2727 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2728 int, newdfd
, const char __user
*, newname
)
2732 struct dentry
*dentry
;
2735 from
= getname(oldname
);
2737 return PTR_ERR(from
);
2739 dentry
= user_path_create(newdfd
, newname
, &path
, 0);
2740 error
= PTR_ERR(dentry
);
2744 error
= mnt_want_write(path
.mnt
);
2747 error
= security_path_symlink(&path
, dentry
, from
);
2749 goto out_drop_write
;
2750 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
);
2752 mnt_drop_write(path
.mnt
);
2755 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2762 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2764 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2767 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2769 struct inode
*inode
= old_dentry
->d_inode
;
2775 error
= may_create(dir
, new_dentry
);
2779 if (dir
->i_sb
!= inode
->i_sb
)
2783 * A link to an append-only or immutable file cannot be created.
2785 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2787 if (!dir
->i_op
->link
)
2789 if (S_ISDIR(inode
->i_mode
))
2792 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2796 mutex_lock(&inode
->i_mutex
);
2797 /* Make sure we don't allow creating hardlink to an unlinked file */
2798 if (inode
->i_nlink
== 0)
2801 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2802 mutex_unlock(&inode
->i_mutex
);
2804 fsnotify_link(dir
, inode
, new_dentry
);
2809 * Hardlinks are often used in delicate situations. We avoid
2810 * security-related surprises by not following symlinks on the
2813 * We don't follow them on the oldname either to be compatible
2814 * with linux 2.0, and to avoid hard-linking to directories
2815 * and other special files. --ADM
2817 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2818 int, newdfd
, const char __user
*, newname
, int, flags
)
2820 struct dentry
*new_dentry
;
2821 struct path old_path
, new_path
;
2825 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
2828 * To use null names we require CAP_DAC_READ_SEARCH
2829 * This ensures that not everyone will be able to create
2830 * handlink using the passed filedescriptor.
2832 if (flags
& AT_EMPTY_PATH
) {
2833 if (!capable(CAP_DAC_READ_SEARCH
))
2838 if (flags
& AT_SYMLINK_FOLLOW
)
2839 how
|= LOOKUP_FOLLOW
;
2841 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
2845 new_dentry
= user_path_create(newdfd
, newname
, &new_path
, 0);
2846 error
= PTR_ERR(new_dentry
);
2847 if (IS_ERR(new_dentry
))
2851 if (old_path
.mnt
!= new_path
.mnt
)
2853 error
= mnt_want_write(new_path
.mnt
);
2856 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
2858 goto out_drop_write
;
2859 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
2861 mnt_drop_write(new_path
.mnt
);
2864 mutex_unlock(&new_path
.dentry
->d_inode
->i_mutex
);
2865 path_put(&new_path
);
2867 path_put(&old_path
);
2872 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
2874 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2878 * The worst of all namespace operations - renaming directory. "Perverted"
2879 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2881 * a) we can get into loop creation. Check is done in is_subdir().
2882 * b) race potential - two innocent renames can create a loop together.
2883 * That's where 4.4 screws up. Current fix: serialization on
2884 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2886 * c) we have to lock _three_ objects - parents and victim (if it exists).
2887 * And that - after we got ->i_mutex on parents (until then we don't know
2888 * whether the target exists). Solution: try to be smart with locking
2889 * order for inodes. We rely on the fact that tree topology may change
2890 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2891 * move will be locked. Thus we can rank directories by the tree
2892 * (ancestors first) and rank all non-directories after them.
2893 * That works since everybody except rename does "lock parent, lookup,
2894 * lock child" and rename is under ->s_vfs_rename_mutex.
2895 * HOWEVER, it relies on the assumption that any object with ->lookup()
2896 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2897 * we'd better make sure that there's no link(2) for them.
2898 * d) conversion from fhandle to dentry may come in the wrong moment - when
2899 * we are removing the target. Solution: we will have to grab ->i_mutex
2900 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2901 * ->i_mutex on parents, which works but leads to some truly excessive
2904 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2905 struct inode
*new_dir
, struct dentry
*new_dentry
)
2908 struct inode
*target
= new_dentry
->d_inode
;
2911 * If we are going to change the parent - check write permissions,
2912 * we'll need to flip '..'.
2914 if (new_dir
!= old_dir
) {
2915 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
2920 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2925 mutex_lock(&target
->i_mutex
);
2928 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
2932 shrink_dcache_parent(new_dentry
);
2933 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2938 target
->i_flags
|= S_DEAD
;
2939 dont_mount(new_dentry
);
2943 mutex_unlock(&target
->i_mutex
);
2945 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2946 d_move(old_dentry
,new_dentry
);
2950 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
2951 struct inode
*new_dir
, struct dentry
*new_dentry
)
2953 struct inode
*target
= new_dentry
->d_inode
;
2956 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2962 mutex_lock(&target
->i_mutex
);
2965 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2968 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2973 dont_mount(new_dentry
);
2974 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2975 d_move(old_dentry
, new_dentry
);
2978 mutex_unlock(&target
->i_mutex
);
2983 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2984 struct inode
*new_dir
, struct dentry
*new_dentry
)
2987 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2988 const unsigned char *old_name
;
2990 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
2993 error
= may_delete(old_dir
, old_dentry
, is_dir
);
2997 if (!new_dentry
->d_inode
)
2998 error
= may_create(new_dir
, new_dentry
);
3000 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3004 if (!old_dir
->i_op
->rename
)
3007 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3010 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3012 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3014 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3015 new_dentry
->d_inode
, old_dentry
);
3016 fsnotify_oldname_free(old_name
);
3021 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3022 int, newdfd
, const char __user
*, newname
)
3024 struct dentry
*old_dir
, *new_dir
;
3025 struct dentry
*old_dentry
, *new_dentry
;
3026 struct dentry
*trap
;
3027 struct nameidata oldnd
, newnd
;
3032 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3036 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3041 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3044 old_dir
= oldnd
.path
.dentry
;
3046 if (oldnd
.last_type
!= LAST_NORM
)
3049 new_dir
= newnd
.path
.dentry
;
3050 if (newnd
.last_type
!= LAST_NORM
)
3053 oldnd
.flags
&= ~LOOKUP_PARENT
;
3054 newnd
.flags
&= ~LOOKUP_PARENT
;
3055 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3057 trap
= lock_rename(new_dir
, old_dir
);
3059 old_dentry
= lookup_hash(&oldnd
);
3060 error
= PTR_ERR(old_dentry
);
3061 if (IS_ERR(old_dentry
))
3063 /* source must exist */
3065 if (!old_dentry
->d_inode
)
3067 /* unless the source is a directory trailing slashes give -ENOTDIR */
3068 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3070 if (oldnd
.last
.name
[oldnd
.last
.len
])
3072 if (newnd
.last
.name
[newnd
.last
.len
])
3075 /* source should not be ancestor of target */
3077 if (old_dentry
== trap
)
3079 new_dentry
= lookup_hash(&newnd
);
3080 error
= PTR_ERR(new_dentry
);
3081 if (IS_ERR(new_dentry
))
3083 /* target should not be an ancestor of source */
3085 if (new_dentry
== trap
)
3088 error
= mnt_want_write(oldnd
.path
.mnt
);
3091 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3092 &newnd
.path
, new_dentry
);
3095 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3096 new_dir
->d_inode
, new_dentry
);
3098 mnt_drop_write(oldnd
.path
.mnt
);
3104 unlock_rename(new_dir
, old_dir
);
3106 path_put(&newnd
.path
);
3109 path_put(&oldnd
.path
);
3115 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3117 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3120 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3124 len
= PTR_ERR(link
);
3129 if (len
> (unsigned) buflen
)
3131 if (copy_to_user(buffer
, link
, len
))
3138 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3139 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3140 * using) it for any given inode is up to filesystem.
3142 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3144 struct nameidata nd
;
3149 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3151 return PTR_ERR(cookie
);
3153 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3154 if (dentry
->d_inode
->i_op
->put_link
)
3155 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3159 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3161 return __vfs_follow_link(nd
, link
);
3164 /* get the link contents into pagecache */
3165 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3169 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3170 page
= read_mapping_page(mapping
, 0, NULL
);
3175 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3179 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3181 struct page
*page
= NULL
;
3182 char *s
= page_getlink(dentry
, &page
);
3183 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3186 page_cache_release(page
);
3191 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3193 struct page
*page
= NULL
;
3194 nd_set_link(nd
, page_getlink(dentry
, &page
));
3198 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3200 struct page
*page
= cookie
;
3204 page_cache_release(page
);
3209 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3211 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3213 struct address_space
*mapping
= inode
->i_mapping
;
3218 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3220 flags
|= AOP_FLAG_NOFS
;
3223 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3224 flags
, &page
, &fsdata
);
3228 kaddr
= kmap_atomic(page
, KM_USER0
);
3229 memcpy(kaddr
, symname
, len
-1);
3230 kunmap_atomic(kaddr
, KM_USER0
);
3232 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3239 mark_inode_dirty(inode
);
3245 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3247 return __page_symlink(inode
, symname
, len
,
3248 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3251 const struct inode_operations page_symlink_inode_operations
= {
3252 .readlink
= generic_readlink
,
3253 .follow_link
= page_follow_link_light
,
3254 .put_link
= page_put_link
,
3257 EXPORT_SYMBOL(user_path_at
);
3258 EXPORT_SYMBOL(follow_down_one
);
3259 EXPORT_SYMBOL(follow_down
);
3260 EXPORT_SYMBOL(follow_up
);
3261 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3262 EXPORT_SYMBOL(getname
);
3263 EXPORT_SYMBOL(lock_rename
);
3264 EXPORT_SYMBOL(lookup_one_len
);
3265 EXPORT_SYMBOL(page_follow_link_light
);
3266 EXPORT_SYMBOL(page_put_link
);
3267 EXPORT_SYMBOL(page_readlink
);
3268 EXPORT_SYMBOL(__page_symlink
);
3269 EXPORT_SYMBOL(page_symlink
);
3270 EXPORT_SYMBOL(page_symlink_inode_operations
);
3271 EXPORT_SYMBOL(kern_path
);
3272 EXPORT_SYMBOL(vfs_path_lookup
);
3273 EXPORT_SYMBOL(inode_permission
);
3274 EXPORT_SYMBOL(unlock_rename
);
3275 EXPORT_SYMBOL(vfs_create
);
3276 EXPORT_SYMBOL(vfs_follow_link
);
3277 EXPORT_SYMBOL(vfs_link
);
3278 EXPORT_SYMBOL(vfs_mkdir
);
3279 EXPORT_SYMBOL(vfs_mknod
);
3280 EXPORT_SYMBOL(generic_permission
);
3281 EXPORT_SYMBOL(vfs_readlink
);
3282 EXPORT_SYMBOL(vfs_rename
);
3283 EXPORT_SYMBOL(vfs_rmdir
);
3284 EXPORT_SYMBOL(vfs_symlink
);
3285 EXPORT_SYMBOL(vfs_unlink
);
3286 EXPORT_SYMBOL(dentry_unhash
);
3287 EXPORT_SYMBOL(generic_readlink
);