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
, 0);
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 * @flags: IPERM_FLAG_ 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
, unsigned int flags
)
320 struct user_namespace
*ns
= inode_userns(inode
);
322 if (flags
& IPERM_FLAG_RCU
)
323 mask
|= MAY_NOT_BLOCK
;
325 if (inode
->i_op
->permission
) {
326 ret
= inode
->i_op
->permission(inode
, mask
, flags
);
330 ret
= acl_permission_check(inode
, mask
);
335 if (ns_capable(ns
, CAP_DAC_OVERRIDE
) ||
336 ns_capable(ns
, CAP_DAC_READ_SEARCH
))
341 return security_inode_exec_permission(inode
, flags
);
345 * path_get - get a reference to a path
346 * @path: path to get the reference to
348 * Given a path increment the reference count to the dentry and the vfsmount.
350 void path_get(struct path
*path
)
355 EXPORT_SYMBOL(path_get
);
358 * path_put - put a reference to a path
359 * @path: path to put the reference to
361 * Given a path decrement the reference count to the dentry and the vfsmount.
363 void path_put(struct path
*path
)
368 EXPORT_SYMBOL(path_put
);
371 * Path walking has 2 modes, rcu-walk and ref-walk (see
372 * Documentation/filesystems/path-lookup.txt). In situations when we can't
373 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
374 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
375 * mode. Refcounts are grabbed at the last known good point before rcu-walk
376 * got stuck, so ref-walk may continue from there. If this is not successful
377 * (eg. a seqcount has changed), then failure is returned and it's up to caller
378 * to restart the path walk from the beginning in ref-walk mode.
382 * unlazy_walk - try to switch to ref-walk mode.
383 * @nd: nameidata pathwalk data
384 * @dentry: child of nd->path.dentry or NULL
385 * Returns: 0 on success, -ECHILD on failure
387 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
388 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
389 * @nd or NULL. Must be called from rcu-walk context.
391 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
393 struct fs_struct
*fs
= current
->fs
;
394 struct dentry
*parent
= nd
->path
.dentry
;
397 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
398 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
400 spin_lock(&fs
->lock
);
401 if (nd
->root
.mnt
!= fs
->root
.mnt
||
402 nd
->root
.dentry
!= fs
->root
.dentry
)
405 spin_lock(&parent
->d_lock
);
407 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
409 BUG_ON(nd
->inode
!= parent
->d_inode
);
411 if (dentry
->d_parent
!= parent
)
413 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
414 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
417 * If the sequence check on the child dentry passed, then
418 * the child has not been removed from its parent. This
419 * means the parent dentry must be valid and able to take
420 * a reference at this point.
422 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
423 BUG_ON(!parent
->d_count
);
425 spin_unlock(&dentry
->d_lock
);
427 spin_unlock(&parent
->d_lock
);
430 spin_unlock(&fs
->lock
);
432 mntget(nd
->path
.mnt
);
435 br_read_unlock(vfsmount_lock
);
436 nd
->flags
&= ~LOOKUP_RCU
;
440 spin_unlock(&dentry
->d_lock
);
442 spin_unlock(&parent
->d_lock
);
445 spin_unlock(&fs
->lock
);
450 * release_open_intent - free up open intent resources
451 * @nd: pointer to nameidata
453 void release_open_intent(struct nameidata
*nd
)
455 struct file
*file
= nd
->intent
.open
.file
;
457 if (file
&& !IS_ERR(file
)) {
458 if (file
->f_path
.dentry
== NULL
)
465 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
467 return dentry
->d_op
->d_revalidate(dentry
, nd
);
470 static struct dentry
*
471 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
473 int status
= d_revalidate(dentry
, nd
);
474 if (unlikely(status
<= 0)) {
476 * The dentry failed validation.
477 * If d_revalidate returned 0 attempt to invalidate
478 * the dentry otherwise d_revalidate is asking us
479 * to return a fail status.
483 dentry
= ERR_PTR(status
);
484 } else if (!d_invalidate(dentry
)) {
493 * complete_walk - successful completion of path walk
494 * @nd: pointer nameidata
496 * If we had been in RCU mode, drop out of it and legitimize nd->path.
497 * Revalidate the final result, unless we'd already done that during
498 * the path walk or the filesystem doesn't ask for it. Return 0 on
499 * success, -error on failure. In case of failure caller does not
500 * need to drop nd->path.
502 static int complete_walk(struct nameidata
*nd
)
504 struct dentry
*dentry
= nd
->path
.dentry
;
507 if (nd
->flags
& LOOKUP_RCU
) {
508 nd
->flags
&= ~LOOKUP_RCU
;
509 if (!(nd
->flags
& LOOKUP_ROOT
))
511 spin_lock(&dentry
->d_lock
);
512 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
513 spin_unlock(&dentry
->d_lock
);
515 br_read_unlock(vfsmount_lock
);
518 BUG_ON(nd
->inode
!= dentry
->d_inode
);
519 spin_unlock(&dentry
->d_lock
);
520 mntget(nd
->path
.mnt
);
522 br_read_unlock(vfsmount_lock
);
525 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
528 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
531 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
534 /* Note: we do not d_invalidate() */
535 status
= d_revalidate(dentry
, nd
);
546 static __always_inline
void set_root(struct nameidata
*nd
)
549 get_fs_root(current
->fs
, &nd
->root
);
552 static int link_path_walk(const char *, struct nameidata
*);
554 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
557 struct fs_struct
*fs
= current
->fs
;
561 seq
= read_seqcount_begin(&fs
->seq
);
563 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
564 } while (read_seqcount_retry(&fs
->seq
, seq
));
568 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
580 nd
->flags
|= LOOKUP_JUMPED
;
582 nd
->inode
= nd
->path
.dentry
->d_inode
;
584 ret
= link_path_walk(link
, nd
);
588 return PTR_ERR(link
);
591 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
594 if (path
->mnt
!= nd
->path
.mnt
)
598 static inline void path_to_nameidata(const struct path
*path
,
599 struct nameidata
*nd
)
601 if (!(nd
->flags
& LOOKUP_RCU
)) {
602 dput(nd
->path
.dentry
);
603 if (nd
->path
.mnt
!= path
->mnt
)
604 mntput(nd
->path
.mnt
);
606 nd
->path
.mnt
= path
->mnt
;
607 nd
->path
.dentry
= path
->dentry
;
610 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
612 struct inode
*inode
= link
->dentry
->d_inode
;
613 if (!IS_ERR(cookie
) && inode
->i_op
->put_link
)
614 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
618 static __always_inline
int
619 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
622 struct dentry
*dentry
= link
->dentry
;
624 BUG_ON(nd
->flags
& LOOKUP_RCU
);
626 if (link
->mnt
== nd
->path
.mnt
)
629 if (unlikely(current
->total_link_count
>= 40)) {
630 *p
= ERR_PTR(-ELOOP
); /* no ->put_link(), please */
635 current
->total_link_count
++;
637 touch_atime(link
->mnt
, dentry
);
638 nd_set_link(nd
, NULL
);
640 error
= security_inode_follow_link(link
->dentry
, nd
);
642 *p
= ERR_PTR(error
); /* no ->put_link(), please */
647 nd
->last_type
= LAST_BIND
;
648 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
651 char *s
= nd_get_link(nd
);
654 error
= __vfs_follow_link(nd
, s
);
655 else if (nd
->last_type
== LAST_BIND
) {
656 nd
->flags
|= LOOKUP_JUMPED
;
657 nd
->inode
= nd
->path
.dentry
->d_inode
;
658 if (nd
->inode
->i_op
->follow_link
) {
659 /* stepped on a _really_ weird one */
668 static int follow_up_rcu(struct path
*path
)
670 struct vfsmount
*parent
;
671 struct dentry
*mountpoint
;
673 parent
= path
->mnt
->mnt_parent
;
674 if (parent
== path
->mnt
)
676 mountpoint
= path
->mnt
->mnt_mountpoint
;
677 path
->dentry
= mountpoint
;
682 int follow_up(struct path
*path
)
684 struct vfsmount
*parent
;
685 struct dentry
*mountpoint
;
687 br_read_lock(vfsmount_lock
);
688 parent
= path
->mnt
->mnt_parent
;
689 if (parent
== path
->mnt
) {
690 br_read_unlock(vfsmount_lock
);
694 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
695 br_read_unlock(vfsmount_lock
);
697 path
->dentry
= mountpoint
;
704 * Perform an automount
705 * - return -EISDIR to tell follow_managed() to stop and return the path we
708 static int follow_automount(struct path
*path
, unsigned flags
,
711 struct vfsmount
*mnt
;
714 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
717 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
718 * and this is the terminal part of the path.
720 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
721 return -EISDIR
; /* we actually want to stop here */
723 /* We want to mount if someone is trying to open/create a file of any
724 * type under the mountpoint, wants to traverse through the mountpoint
725 * or wants to open the mounted directory.
727 * We don't want to mount if someone's just doing a stat and they've
728 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
729 * appended a '/' to the name.
731 if (!(flags
& LOOKUP_FOLLOW
) &&
732 !(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
733 LOOKUP_OPEN
| LOOKUP_CREATE
)))
736 current
->total_link_count
++;
737 if (current
->total_link_count
>= 40)
740 mnt
= path
->dentry
->d_op
->d_automount(path
);
743 * The filesystem is allowed to return -EISDIR here to indicate
744 * it doesn't want to automount. For instance, autofs would do
745 * this so that its userspace daemon can mount on this dentry.
747 * However, we can only permit this if it's a terminal point in
748 * the path being looked up; if it wasn't then the remainder of
749 * the path is inaccessible and we should say so.
751 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_CONTINUE
))
756 if (!mnt
) /* mount collision */
760 /* lock_mount() may release path->mnt on error */
764 err
= finish_automount(mnt
, path
);
768 /* Someone else made a mount here whilst we were busy */
773 path
->dentry
= dget(mnt
->mnt_root
);
782 * Handle a dentry that is managed in some way.
783 * - Flagged for transit management (autofs)
784 * - Flagged as mountpoint
785 * - Flagged as automount point
787 * This may only be called in refwalk mode.
789 * Serialization is taken care of in namespace.c
791 static int follow_managed(struct path
*path
, unsigned flags
)
793 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
795 bool need_mntput
= false;
798 /* Given that we're not holding a lock here, we retain the value in a
799 * local variable for each dentry as we look at it so that we don't see
800 * the components of that value change under us */
801 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
802 managed
&= DCACHE_MANAGED_DENTRY
,
803 unlikely(managed
!= 0)) {
804 /* Allow the filesystem to manage the transit without i_mutex
806 if (managed
& DCACHE_MANAGE_TRANSIT
) {
807 BUG_ON(!path
->dentry
->d_op
);
808 BUG_ON(!path
->dentry
->d_op
->d_manage
);
809 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
814 /* Transit to a mounted filesystem. */
815 if (managed
& DCACHE_MOUNTED
) {
816 struct vfsmount
*mounted
= lookup_mnt(path
);
822 path
->dentry
= dget(mounted
->mnt_root
);
827 /* Something is mounted on this dentry in another
828 * namespace and/or whatever was mounted there in this
829 * namespace got unmounted before we managed to get the
833 /* Handle an automount point */
834 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
835 ret
= follow_automount(path
, flags
, &need_mntput
);
841 /* We didn't change the current path point */
845 if (need_mntput
&& path
->mnt
== mnt
)
852 int follow_down_one(struct path
*path
)
854 struct vfsmount
*mounted
;
856 mounted
= lookup_mnt(path
);
861 path
->dentry
= dget(mounted
->mnt_root
);
867 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
869 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
870 dentry
->d_op
->d_manage(dentry
, true) < 0);
874 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
875 * we meet a managed dentry that would need blocking.
877 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
878 struct inode
**inode
)
881 struct vfsmount
*mounted
;
883 * Don't forget we might have a non-mountpoint managed dentry
884 * that wants to block transit.
886 if (unlikely(managed_dentry_might_block(path
->dentry
)))
889 if (!d_mountpoint(path
->dentry
))
892 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
896 path
->dentry
= mounted
->mnt_root
;
897 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
899 * Update the inode too. We don't need to re-check the
900 * dentry sequence number here after this d_inode read,
901 * because a mount-point is always pinned.
903 *inode
= path
->dentry
->d_inode
;
908 static void follow_mount_rcu(struct nameidata
*nd
)
910 while (d_mountpoint(nd
->path
.dentry
)) {
911 struct vfsmount
*mounted
;
912 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
915 nd
->path
.mnt
= mounted
;
916 nd
->path
.dentry
= mounted
->mnt_root
;
917 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
921 static int follow_dotdot_rcu(struct nameidata
*nd
)
926 if (nd
->path
.dentry
== nd
->root
.dentry
&&
927 nd
->path
.mnt
== nd
->root
.mnt
) {
930 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
931 struct dentry
*old
= nd
->path
.dentry
;
932 struct dentry
*parent
= old
->d_parent
;
935 seq
= read_seqcount_begin(&parent
->d_seq
);
936 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
938 nd
->path
.dentry
= parent
;
942 if (!follow_up_rcu(&nd
->path
))
944 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
946 follow_mount_rcu(nd
);
947 nd
->inode
= nd
->path
.dentry
->d_inode
;
951 nd
->flags
&= ~LOOKUP_RCU
;
952 if (!(nd
->flags
& LOOKUP_ROOT
))
955 br_read_unlock(vfsmount_lock
);
960 * Follow down to the covering mount currently visible to userspace. At each
961 * point, the filesystem owning that dentry may be queried as to whether the
962 * caller is permitted to proceed or not.
964 int follow_down(struct path
*path
)
969 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
970 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
971 /* Allow the filesystem to manage the transit without i_mutex
974 * We indicate to the filesystem if someone is trying to mount
975 * something here. This gives autofs the chance to deny anyone
976 * other than its daemon the right to mount on its
979 * The filesystem may sleep at this point.
981 if (managed
& DCACHE_MANAGE_TRANSIT
) {
982 BUG_ON(!path
->dentry
->d_op
);
983 BUG_ON(!path
->dentry
->d_op
->d_manage
);
984 ret
= path
->dentry
->d_op
->d_manage(
985 path
->dentry
, false);
987 return ret
== -EISDIR
? 0 : ret
;
990 /* Transit to a mounted filesystem. */
991 if (managed
& DCACHE_MOUNTED
) {
992 struct vfsmount
*mounted
= lookup_mnt(path
);
998 path
->dentry
= dget(mounted
->mnt_root
);
1002 /* Don't handle automount points here */
1009 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1011 static void follow_mount(struct path
*path
)
1013 while (d_mountpoint(path
->dentry
)) {
1014 struct vfsmount
*mounted
= lookup_mnt(path
);
1019 path
->mnt
= mounted
;
1020 path
->dentry
= dget(mounted
->mnt_root
);
1024 static void follow_dotdot(struct nameidata
*nd
)
1029 struct dentry
*old
= nd
->path
.dentry
;
1031 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1032 nd
->path
.mnt
== nd
->root
.mnt
) {
1035 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1036 /* rare case of legitimate dget_parent()... */
1037 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1041 if (!follow_up(&nd
->path
))
1044 follow_mount(&nd
->path
);
1045 nd
->inode
= nd
->path
.dentry
->d_inode
;
1049 * Allocate a dentry with name and parent, and perform a parent
1050 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1051 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1052 * have verified that no child exists while under i_mutex.
1054 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1055 struct qstr
*name
, struct nameidata
*nd
)
1057 struct inode
*inode
= parent
->d_inode
;
1058 struct dentry
*dentry
;
1061 /* Don't create child dentry for a dead directory. */
1062 if (unlikely(IS_DEADDIR(inode
)))
1063 return ERR_PTR(-ENOENT
);
1065 dentry
= d_alloc(parent
, name
);
1066 if (unlikely(!dentry
))
1067 return ERR_PTR(-ENOMEM
);
1069 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1070 if (unlikely(old
)) {
1078 * We already have a dentry, but require a lookup to be performed on the parent
1079 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1080 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1081 * child exists while under i_mutex.
1083 static struct dentry
*d_inode_lookup(struct dentry
*parent
, struct dentry
*dentry
,
1084 struct nameidata
*nd
)
1086 struct inode
*inode
= parent
->d_inode
;
1089 /* Don't create child dentry for a dead directory. */
1090 if (unlikely(IS_DEADDIR(inode
)))
1091 return ERR_PTR(-ENOENT
);
1093 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1094 if (unlikely(old
)) {
1102 * It's more convoluted than I'd like it to be, but... it's still fairly
1103 * small and for now I'd prefer to have fast path as straight as possible.
1104 * It _is_ time-critical.
1106 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1107 struct path
*path
, struct inode
**inode
)
1109 struct vfsmount
*mnt
= nd
->path
.mnt
;
1110 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1116 * Rename seqlock is not required here because in the off chance
1117 * of a false negative due to a concurrent rename, we're going to
1118 * do the non-racy lookup, below.
1120 if (nd
->flags
& LOOKUP_RCU
) {
1123 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1127 /* Memory barrier in read_seqcount_begin of child is enough */
1128 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1132 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1133 status
= d_revalidate(dentry
, nd
);
1134 if (unlikely(status
<= 0)) {
1135 if (status
!= -ECHILD
)
1140 if (unlikely(d_need_lookup(dentry
)))
1143 path
->dentry
= dentry
;
1144 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1146 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1150 if (unlazy_walk(nd
, dentry
))
1153 dentry
= __d_lookup(parent
, name
);
1156 if (dentry
&& unlikely(d_need_lookup(dentry
))) {
1161 if (unlikely(!dentry
)) {
1162 struct inode
*dir
= parent
->d_inode
;
1163 BUG_ON(nd
->inode
!= dir
);
1165 mutex_lock(&dir
->i_mutex
);
1166 dentry
= d_lookup(parent
, name
);
1167 if (likely(!dentry
)) {
1168 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1169 if (IS_ERR(dentry
)) {
1170 mutex_unlock(&dir
->i_mutex
);
1171 return PTR_ERR(dentry
);
1176 } else if (unlikely(d_need_lookup(dentry
))) {
1177 dentry
= d_inode_lookup(parent
, dentry
, nd
);
1178 if (IS_ERR(dentry
)) {
1179 mutex_unlock(&dir
->i_mutex
);
1180 return PTR_ERR(dentry
);
1186 mutex_unlock(&dir
->i_mutex
);
1188 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1189 status
= d_revalidate(dentry
, nd
);
1190 if (unlikely(status
<= 0)) {
1195 if (!d_invalidate(dentry
)) {
1204 path
->dentry
= dentry
;
1205 err
= follow_managed(path
, nd
->flags
);
1206 if (unlikely(err
< 0)) {
1207 path_put_conditional(path
, nd
);
1210 *inode
= path
->dentry
->d_inode
;
1214 static inline int may_lookup(struct nameidata
*nd
)
1216 if (nd
->flags
& LOOKUP_RCU
) {
1217 int err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1220 if (unlazy_walk(nd
, NULL
))
1223 return exec_permission(nd
->inode
, 0);
1226 static inline int handle_dots(struct nameidata
*nd
, int type
)
1228 if (type
== LAST_DOTDOT
) {
1229 if (nd
->flags
& LOOKUP_RCU
) {
1230 if (follow_dotdot_rcu(nd
))
1238 static void terminate_walk(struct nameidata
*nd
)
1240 if (!(nd
->flags
& LOOKUP_RCU
)) {
1241 path_put(&nd
->path
);
1243 nd
->flags
&= ~LOOKUP_RCU
;
1244 if (!(nd
->flags
& LOOKUP_ROOT
))
1245 nd
->root
.mnt
= NULL
;
1247 br_read_unlock(vfsmount_lock
);
1251 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1252 struct qstr
*name
, int type
, int follow
)
1254 struct inode
*inode
;
1257 * "." and ".." are special - ".." especially so because it has
1258 * to be able to know about the current root directory and
1259 * parent relationships.
1261 if (unlikely(type
!= LAST_NORM
))
1262 return handle_dots(nd
, type
);
1263 err
= do_lookup(nd
, name
, path
, &inode
);
1264 if (unlikely(err
)) {
1269 path_to_nameidata(path
, nd
);
1273 if (unlikely(inode
->i_op
->follow_link
) && follow
) {
1274 if (nd
->flags
& LOOKUP_RCU
) {
1275 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1280 BUG_ON(inode
!= path
->dentry
->d_inode
);
1283 path_to_nameidata(path
, nd
);
1289 * This limits recursive symlink follows to 8, while
1290 * limiting consecutive symlinks to 40.
1292 * Without that kind of total limit, nasty chains of consecutive
1293 * symlinks can cause almost arbitrarily long lookups.
1295 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1299 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1300 path_put_conditional(path
, nd
);
1301 path_put(&nd
->path
);
1304 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1307 current
->link_count
++;
1310 struct path link
= *path
;
1313 res
= follow_link(&link
, nd
, &cookie
);
1315 res
= walk_component(nd
, path
, &nd
->last
,
1316 nd
->last_type
, LOOKUP_FOLLOW
);
1317 put_link(nd
, &link
, cookie
);
1320 current
->link_count
--;
1327 * This is the basic name resolution function, turning a pathname into
1328 * the final dentry. We expect 'base' to be positive and a directory.
1330 * Returns 0 and nd will have valid dentry and mnt on success.
1331 * Returns error and drops reference to input namei data on failure.
1333 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1337 unsigned int lookup_flags
= nd
->flags
;
1344 /* At this point we know we have a real path component. */
1351 nd
->flags
|= LOOKUP_CONTINUE
;
1353 err
= may_lookup(nd
);
1358 c
= *(const unsigned char *)name
;
1360 hash
= init_name_hash();
1363 hash
= partial_name_hash(c
, hash
);
1364 c
= *(const unsigned char *)name
;
1365 } while (c
&& (c
!= '/'));
1366 this.len
= name
- (const char *) this.name
;
1367 this.hash
= end_name_hash(hash
);
1370 if (this.name
[0] == '.') switch (this.len
) {
1372 if (this.name
[1] == '.') {
1374 nd
->flags
|= LOOKUP_JUMPED
;
1380 if (likely(type
== LAST_NORM
)) {
1381 struct dentry
*parent
= nd
->path
.dentry
;
1382 nd
->flags
&= ~LOOKUP_JUMPED
;
1383 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1384 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1391 /* remove trailing slashes? */
1393 goto last_component
;
1394 while (*++name
== '/');
1396 goto last_component
;
1398 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1403 err
= nested_symlink(&next
, nd
);
1408 if (!nd
->inode
->i_op
->lookup
)
1411 /* here ends the main loop */
1414 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1415 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1417 nd
->last_type
= type
;
1424 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1425 struct nameidata
*nd
, struct file
**fp
)
1431 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1432 nd
->flags
= flags
| LOOKUP_JUMPED
;
1434 if (flags
& LOOKUP_ROOT
) {
1435 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1437 if (!inode
->i_op
->lookup
)
1439 retval
= inode_permission(inode
, MAY_EXEC
);
1443 nd
->path
= nd
->root
;
1445 if (flags
& LOOKUP_RCU
) {
1446 br_read_lock(vfsmount_lock
);
1448 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1450 path_get(&nd
->path
);
1455 nd
->root
.mnt
= NULL
;
1458 if (flags
& LOOKUP_RCU
) {
1459 br_read_lock(vfsmount_lock
);
1464 path_get(&nd
->root
);
1466 nd
->path
= nd
->root
;
1467 } else if (dfd
== AT_FDCWD
) {
1468 if (flags
& LOOKUP_RCU
) {
1469 struct fs_struct
*fs
= current
->fs
;
1472 br_read_lock(vfsmount_lock
);
1476 seq
= read_seqcount_begin(&fs
->seq
);
1478 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1479 } while (read_seqcount_retry(&fs
->seq
, seq
));
1481 get_fs_pwd(current
->fs
, &nd
->path
);
1484 struct dentry
*dentry
;
1486 file
= fget_raw_light(dfd
, &fput_needed
);
1491 dentry
= file
->f_path
.dentry
;
1495 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1498 retval
= exec_permission(dentry
->d_inode
, 0);
1503 nd
->path
= file
->f_path
;
1504 if (flags
& LOOKUP_RCU
) {
1507 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1508 br_read_lock(vfsmount_lock
);
1511 path_get(&file
->f_path
);
1512 fput_light(file
, fput_needed
);
1516 nd
->inode
= nd
->path
.dentry
->d_inode
;
1520 fput_light(file
, fput_needed
);
1525 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1527 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1528 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1530 nd
->flags
&= ~LOOKUP_PARENT
;
1531 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1532 nd
->flags
& LOOKUP_FOLLOW
);
1535 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1536 static int path_lookupat(int dfd
, const char *name
,
1537 unsigned int flags
, struct nameidata
*nd
)
1539 struct file
*base
= NULL
;
1544 * Path walking is largely split up into 2 different synchronisation
1545 * schemes, rcu-walk and ref-walk (explained in
1546 * Documentation/filesystems/path-lookup.txt). These share much of the
1547 * path walk code, but some things particularly setup, cleanup, and
1548 * following mounts are sufficiently divergent that functions are
1549 * duplicated. Typically there is a function foo(), and its RCU
1550 * analogue, foo_rcu().
1552 * -ECHILD is the error number of choice (just to avoid clashes) that
1553 * is returned if some aspect of an rcu-walk fails. Such an error must
1554 * be handled by restarting a traditional ref-walk (which will always
1555 * be able to complete).
1557 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1562 current
->total_link_count
= 0;
1563 err
= link_path_walk(name
, nd
);
1565 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1566 err
= lookup_last(nd
, &path
);
1569 struct path link
= path
;
1570 nd
->flags
|= LOOKUP_PARENT
;
1571 err
= follow_link(&link
, nd
, &cookie
);
1573 err
= lookup_last(nd
, &path
);
1574 put_link(nd
, &link
, cookie
);
1579 err
= complete_walk(nd
);
1581 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1582 if (!nd
->inode
->i_op
->lookup
) {
1583 path_put(&nd
->path
);
1591 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1592 path_put(&nd
->root
);
1593 nd
->root
.mnt
= NULL
;
1598 static int do_path_lookup(int dfd
, const char *name
,
1599 unsigned int flags
, struct nameidata
*nd
)
1601 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1602 if (unlikely(retval
== -ECHILD
))
1603 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1604 if (unlikely(retval
== -ESTALE
))
1605 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1607 if (likely(!retval
)) {
1608 if (unlikely(!audit_dummy_context())) {
1609 if (nd
->path
.dentry
&& nd
->inode
)
1610 audit_inode(name
, nd
->path
.dentry
);
1616 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1618 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1621 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1623 struct nameidata nd
;
1624 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1631 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1632 * @dentry: pointer to dentry of the base directory
1633 * @mnt: pointer to vfs mount of the base directory
1634 * @name: pointer to file name
1635 * @flags: lookup flags
1636 * @nd: pointer to nameidata
1638 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1639 const char *name
, unsigned int flags
,
1640 struct nameidata
*nd
)
1642 nd
->root
.dentry
= dentry
;
1644 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1645 return do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, nd
);
1648 static struct dentry
*__lookup_hash(struct qstr
*name
,
1649 struct dentry
*base
, struct nameidata
*nd
)
1651 struct inode
*inode
= base
->d_inode
;
1652 struct dentry
*dentry
;
1655 err
= exec_permission(inode
, 0);
1657 return ERR_PTR(err
);
1660 * Don't bother with __d_lookup: callers are for creat as
1661 * well as unlink, so a lot of the time it would cost
1664 dentry
= d_lookup(base
, name
);
1666 if (dentry
&& d_need_lookup(dentry
)) {
1668 * __lookup_hash is called with the parent dir's i_mutex already
1669 * held, so we are good to go here.
1671 dentry
= d_inode_lookup(base
, dentry
, nd
);
1676 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1677 dentry
= do_revalidate(dentry
, nd
);
1680 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1686 * Restricted form of lookup. Doesn't follow links, single-component only,
1687 * needs parent already locked. Doesn't follow mounts.
1690 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1692 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1696 * lookup_one_len - filesystem helper to lookup single pathname component
1697 * @name: pathname component to lookup
1698 * @base: base directory to lookup from
1699 * @len: maximum length @len should be interpreted to
1701 * Note that this routine is purely a helper for filesystem usage and should
1702 * not be called by generic code. Also note that by using this function the
1703 * nameidata argument is passed to the filesystem methods and a filesystem
1704 * using this helper needs to be prepared for that.
1706 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1712 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1717 return ERR_PTR(-EACCES
);
1719 hash
= init_name_hash();
1721 c
= *(const unsigned char *)name
++;
1722 if (c
== '/' || c
== '\0')
1723 return ERR_PTR(-EACCES
);
1724 hash
= partial_name_hash(c
, hash
);
1726 this.hash
= end_name_hash(hash
);
1728 * See if the low-level filesystem might want
1729 * to use its own hash..
1731 if (base
->d_flags
& DCACHE_OP_HASH
) {
1732 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1734 return ERR_PTR(err
);
1737 return __lookup_hash(&this, base
, NULL
);
1740 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1743 struct nameidata nd
;
1744 char *tmp
= getname_flags(name
, flags
);
1745 int err
= PTR_ERR(tmp
);
1748 BUG_ON(flags
& LOOKUP_PARENT
);
1750 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1758 static int user_path_parent(int dfd
, const char __user
*path
,
1759 struct nameidata
*nd
, char **name
)
1761 char *s
= getname(path
);
1767 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1777 * It's inline, so penalty for filesystems that don't use sticky bit is
1780 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1782 uid_t fsuid
= current_fsuid();
1784 if (!(dir
->i_mode
& S_ISVTX
))
1786 if (current_user_ns() != inode_userns(inode
))
1788 if (inode
->i_uid
== fsuid
)
1790 if (dir
->i_uid
== fsuid
)
1794 return !ns_capable(inode_userns(inode
), CAP_FOWNER
);
1798 * Check whether we can remove a link victim from directory dir, check
1799 * whether the type of victim is right.
1800 * 1. We can't do it if dir is read-only (done in permission())
1801 * 2. We should have write and exec permissions on dir
1802 * 3. We can't remove anything from append-only dir
1803 * 4. We can't do anything with immutable dir (done in permission())
1804 * 5. If the sticky bit on dir is set we should either
1805 * a. be owner of dir, or
1806 * b. be owner of victim, or
1807 * c. have CAP_FOWNER capability
1808 * 6. If the victim is append-only or immutable we can't do antyhing with
1809 * links pointing to it.
1810 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1811 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1812 * 9. We can't remove a root or mountpoint.
1813 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1814 * nfs_async_unlink().
1816 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1820 if (!victim
->d_inode
)
1823 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1824 audit_inode_child(victim
, dir
);
1826 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1831 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1832 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1835 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1837 if (IS_ROOT(victim
))
1839 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1841 if (IS_DEADDIR(dir
))
1843 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1848 /* Check whether we can create an object with dentry child in directory
1850 * 1. We can't do it if child already exists (open has special treatment for
1851 * this case, but since we are inlined it's OK)
1852 * 2. We can't do it if dir is read-only (done in permission())
1853 * 3. We should have write and exec permissions on dir
1854 * 4. We can't do it if dir is immutable (done in permission())
1856 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1860 if (IS_DEADDIR(dir
))
1862 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1866 * p1 and p2 should be directories on the same fs.
1868 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1873 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1877 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1879 p
= d_ancestor(p2
, p1
);
1881 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1882 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1886 p
= d_ancestor(p1
, p2
);
1888 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1889 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1893 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1894 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1898 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1900 mutex_unlock(&p1
->d_inode
->i_mutex
);
1902 mutex_unlock(&p2
->d_inode
->i_mutex
);
1903 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1907 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1908 struct nameidata
*nd
)
1910 int error
= may_create(dir
, dentry
);
1915 if (!dir
->i_op
->create
)
1916 return -EACCES
; /* shouldn't it be ENOSYS? */
1919 error
= security_inode_create(dir
, dentry
, mode
);
1922 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1924 fsnotify_create(dir
, dentry
);
1928 static int may_open(struct path
*path
, int acc_mode
, int flag
)
1930 struct dentry
*dentry
= path
->dentry
;
1931 struct inode
*inode
= dentry
->d_inode
;
1941 switch (inode
->i_mode
& S_IFMT
) {
1945 if (acc_mode
& MAY_WRITE
)
1950 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
1959 error
= inode_permission(inode
, acc_mode
);
1964 * An append-only file must be opened in append mode for writing.
1966 if (IS_APPEND(inode
)) {
1967 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
1973 /* O_NOATIME can only be set by the owner or superuser */
1974 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
1978 * Ensure there are no outstanding leases on the file.
1980 return break_lease(inode
, flag
);
1983 static int handle_truncate(struct file
*filp
)
1985 struct path
*path
= &filp
->f_path
;
1986 struct inode
*inode
= path
->dentry
->d_inode
;
1987 int error
= get_write_access(inode
);
1991 * Refuse to truncate files with mandatory locks held on them.
1993 error
= locks_verify_locked(inode
);
1995 error
= security_path_truncate(path
);
1997 error
= do_truncate(path
->dentry
, 0,
1998 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2001 put_write_access(inode
);
2006 * Note that while the flag value (low two bits) for sys_open means:
2011 * it is changed into
2012 * 00 - no permissions needed
2013 * 01 - read-permission
2014 * 10 - write-permission
2016 * for the internal routines (ie open_namei()/follow_link() etc)
2017 * This is more logical, and also allows the 00 "no perm needed"
2018 * to be used for symlinks (where the permissions are checked
2022 static inline int open_to_namei_flags(int flag
)
2024 if ((flag
+1) & O_ACCMODE
)
2030 * Handle the last step of open()
2032 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2033 const struct open_flags
*op
, const char *pathname
)
2035 struct dentry
*dir
= nd
->path
.dentry
;
2036 struct dentry
*dentry
;
2037 int open_flag
= op
->open_flag
;
2038 int will_truncate
= open_flag
& O_TRUNC
;
2040 int acc_mode
= op
->acc_mode
;
2044 nd
->flags
&= ~LOOKUP_PARENT
;
2045 nd
->flags
|= op
->intent
;
2047 switch (nd
->last_type
) {
2050 error
= handle_dots(nd
, nd
->last_type
);
2052 return ERR_PTR(error
);
2055 error
= complete_walk(nd
);
2057 return ERR_PTR(error
);
2058 audit_inode(pathname
, nd
->path
.dentry
);
2059 if (open_flag
& O_CREAT
) {
2065 error
= complete_walk(nd
);
2067 return ERR_PTR(error
);
2068 audit_inode(pathname
, dir
);
2072 if (!(open_flag
& O_CREAT
)) {
2074 if (nd
->last
.name
[nd
->last
.len
])
2075 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2076 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2078 /* we _can_ be in RCU mode here */
2079 error
= walk_component(nd
, path
, &nd
->last
, LAST_NORM
,
2082 return ERR_PTR(error
);
2083 if (error
) /* symlink */
2086 error
= complete_walk(nd
);
2088 return ERR_PTR(-ECHILD
);
2091 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2092 if (!nd
->inode
->i_op
->lookup
)
2095 audit_inode(pathname
, nd
->path
.dentry
);
2099 /* create side of things */
2100 error
= complete_walk(nd
);
2102 return ERR_PTR(error
);
2104 audit_inode(pathname
, dir
);
2106 /* trailing slashes? */
2107 if (nd
->last
.name
[nd
->last
.len
])
2110 mutex_lock(&dir
->d_inode
->i_mutex
);
2112 dentry
= lookup_hash(nd
);
2113 error
= PTR_ERR(dentry
);
2114 if (IS_ERR(dentry
)) {
2115 mutex_unlock(&dir
->d_inode
->i_mutex
);
2119 path
->dentry
= dentry
;
2120 path
->mnt
= nd
->path
.mnt
;
2122 /* Negative dentry, just create the file */
2123 if (!dentry
->d_inode
) {
2124 int mode
= op
->mode
;
2125 if (!IS_POSIXACL(dir
->d_inode
))
2126 mode
&= ~current_umask();
2128 * This write is needed to ensure that a
2129 * rw->ro transition does not occur between
2130 * the time when the file is created and when
2131 * a permanent write count is taken through
2132 * the 'struct file' in nameidata_to_filp().
2134 error
= mnt_want_write(nd
->path
.mnt
);
2136 goto exit_mutex_unlock
;
2138 /* Don't check for write permission, don't truncate */
2139 open_flag
&= ~O_TRUNC
;
2141 acc_mode
= MAY_OPEN
;
2142 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2144 goto exit_mutex_unlock
;
2145 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2147 goto exit_mutex_unlock
;
2148 mutex_unlock(&dir
->d_inode
->i_mutex
);
2149 dput(nd
->path
.dentry
);
2150 nd
->path
.dentry
= dentry
;
2155 * It already exists.
2157 mutex_unlock(&dir
->d_inode
->i_mutex
);
2158 audit_inode(pathname
, path
->dentry
);
2161 if (open_flag
& O_EXCL
)
2164 error
= follow_managed(path
, nd
->flags
);
2169 if (!path
->dentry
->d_inode
)
2172 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2175 path_to_nameidata(path
, nd
);
2176 nd
->inode
= path
->dentry
->d_inode
;
2178 if (S_ISDIR(nd
->inode
->i_mode
))
2181 if (!S_ISREG(nd
->inode
->i_mode
))
2184 if (will_truncate
) {
2185 error
= mnt_want_write(nd
->path
.mnt
);
2191 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2194 filp
= nameidata_to_filp(nd
);
2195 if (!IS_ERR(filp
)) {
2196 error
= ima_file_check(filp
, op
->acc_mode
);
2199 filp
= ERR_PTR(error
);
2202 if (!IS_ERR(filp
)) {
2203 if (will_truncate
) {
2204 error
= handle_truncate(filp
);
2207 filp
= ERR_PTR(error
);
2213 mnt_drop_write(nd
->path
.mnt
);
2214 path_put(&nd
->path
);
2218 mutex_unlock(&dir
->d_inode
->i_mutex
);
2220 path_put_conditional(path
, nd
);
2222 filp
= ERR_PTR(error
);
2226 static struct file
*path_openat(int dfd
, const char *pathname
,
2227 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2229 struct file
*base
= NULL
;
2234 filp
= get_empty_filp();
2236 return ERR_PTR(-ENFILE
);
2238 filp
->f_flags
= op
->open_flag
;
2239 nd
->intent
.open
.file
= filp
;
2240 nd
->intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2241 nd
->intent
.open
.create_mode
= op
->mode
;
2243 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2244 if (unlikely(error
))
2247 current
->total_link_count
= 0;
2248 error
= link_path_walk(pathname
, nd
);
2249 if (unlikely(error
))
2252 filp
= do_last(nd
, &path
, op
, pathname
);
2253 while (unlikely(!filp
)) { /* trailing symlink */
2254 struct path link
= path
;
2256 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2257 path_put_conditional(&path
, nd
);
2258 path_put(&nd
->path
);
2259 filp
= ERR_PTR(-ELOOP
);
2262 nd
->flags
|= LOOKUP_PARENT
;
2263 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2264 error
= follow_link(&link
, nd
, &cookie
);
2265 if (unlikely(error
))
2266 filp
= ERR_PTR(error
);
2268 filp
= do_last(nd
, &path
, op
, pathname
);
2269 put_link(nd
, &link
, cookie
);
2272 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2273 path_put(&nd
->root
);
2276 release_open_intent(nd
);
2280 filp
= ERR_PTR(error
);
2284 struct file
*do_filp_open(int dfd
, const char *pathname
,
2285 const struct open_flags
*op
, int flags
)
2287 struct nameidata nd
;
2290 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2291 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2292 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2293 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2294 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2298 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2299 const char *name
, const struct open_flags
*op
, int flags
)
2301 struct nameidata nd
;
2305 nd
.root
.dentry
= dentry
;
2307 flags
|= LOOKUP_ROOT
;
2309 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2310 return ERR_PTR(-ELOOP
);
2312 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2313 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2314 file
= path_openat(-1, name
, &nd
, op
, flags
);
2315 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2316 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2321 * lookup_create - lookup a dentry, creating it if it doesn't exist
2322 * @nd: nameidata info
2323 * @is_dir: directory flag
2325 * Simple function to lookup and return a dentry and create it
2326 * if it doesn't exist. Is SMP-safe.
2328 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2330 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2332 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2334 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2336 * Yucky last component or no last component at all?
2337 * (foo/., foo/.., /////)
2339 if (nd
->last_type
!= LAST_NORM
)
2341 nd
->flags
&= ~LOOKUP_PARENT
;
2342 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2343 nd
->intent
.open
.flags
= O_EXCL
;
2346 * Do the final lookup.
2348 dentry
= lookup_hash(nd
);
2352 if (dentry
->d_inode
)
2355 * Special case - lookup gave negative, but... we had foo/bar/
2356 * From the vfs_mknod() POV we just have a negative dentry -
2357 * all is fine. Let's be bastards - you had / on the end, you've
2358 * been asking for (non-existent) directory. -ENOENT for you.
2360 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2362 dentry
= ERR_PTR(-ENOENT
);
2367 dentry
= ERR_PTR(-EEXIST
);
2371 EXPORT_SYMBOL_GPL(lookup_create
);
2373 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2375 int error
= may_create(dir
, dentry
);
2380 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) &&
2381 !ns_capable(inode_userns(dir
), CAP_MKNOD
))
2384 if (!dir
->i_op
->mknod
)
2387 error
= devcgroup_inode_mknod(mode
, dev
);
2391 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2395 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2397 fsnotify_create(dir
, dentry
);
2401 static int may_mknod(mode_t mode
)
2403 switch (mode
& S_IFMT
) {
2409 case 0: /* zero mode translates to S_IFREG */
2418 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2423 struct dentry
*dentry
;
2424 struct nameidata nd
;
2429 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2433 dentry
= lookup_create(&nd
, 0);
2434 if (IS_ERR(dentry
)) {
2435 error
= PTR_ERR(dentry
);
2438 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2439 mode
&= ~current_umask();
2440 error
= may_mknod(mode
);
2443 error
= mnt_want_write(nd
.path
.mnt
);
2446 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2448 goto out_drop_write
;
2449 switch (mode
& S_IFMT
) {
2450 case 0: case S_IFREG
:
2451 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2453 case S_IFCHR
: case S_IFBLK
:
2454 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2455 new_decode_dev(dev
));
2457 case S_IFIFO
: case S_IFSOCK
:
2458 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2462 mnt_drop_write(nd
.path
.mnt
);
2466 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2473 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2475 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2478 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2480 int error
= may_create(dir
, dentry
);
2485 if (!dir
->i_op
->mkdir
)
2488 mode
&= (S_IRWXUGO
|S_ISVTX
);
2489 error
= security_inode_mkdir(dir
, dentry
, mode
);
2493 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2495 fsnotify_mkdir(dir
, dentry
);
2499 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2503 struct dentry
*dentry
;
2504 struct nameidata nd
;
2506 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2510 dentry
= lookup_create(&nd
, 1);
2511 error
= PTR_ERR(dentry
);
2515 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2516 mode
&= ~current_umask();
2517 error
= mnt_want_write(nd
.path
.mnt
);
2520 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2522 goto out_drop_write
;
2523 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2525 mnt_drop_write(nd
.path
.mnt
);
2529 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2536 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2538 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2542 * The dentry_unhash() helper will try to drop the dentry early: we
2543 * should have a usage count of 2 if we're the only user of this
2544 * dentry, and if that is true (possibly after pruning the dcache),
2545 * then we drop the dentry now.
2547 * A low-level filesystem can, if it choses, legally
2550 * if (!d_unhashed(dentry))
2553 * if it cannot handle the case of removing a directory
2554 * that is still in use by something else..
2556 void dentry_unhash(struct dentry
*dentry
)
2558 shrink_dcache_parent(dentry
);
2559 spin_lock(&dentry
->d_lock
);
2560 if (dentry
->d_count
== 1)
2562 spin_unlock(&dentry
->d_lock
);
2565 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2567 int error
= may_delete(dir
, dentry
, 1);
2572 if (!dir
->i_op
->rmdir
)
2575 mutex_lock(&dentry
->d_inode
->i_mutex
);
2578 if (d_mountpoint(dentry
))
2581 error
= security_inode_rmdir(dir
, dentry
);
2585 shrink_dcache_parent(dentry
);
2586 error
= dir
->i_op
->rmdir(dir
, dentry
);
2590 dentry
->d_inode
->i_flags
|= S_DEAD
;
2594 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2600 static long do_rmdir(int dfd
, const char __user
*pathname
)
2604 struct dentry
*dentry
;
2605 struct nameidata nd
;
2607 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2611 switch(nd
.last_type
) {
2623 nd
.flags
&= ~LOOKUP_PARENT
;
2625 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2626 dentry
= lookup_hash(&nd
);
2627 error
= PTR_ERR(dentry
);
2630 if (!dentry
->d_inode
) {
2634 error
= mnt_want_write(nd
.path
.mnt
);
2637 error
= security_path_rmdir(&nd
.path
, dentry
);
2640 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2642 mnt_drop_write(nd
.path
.mnt
);
2646 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2653 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2655 return do_rmdir(AT_FDCWD
, pathname
);
2658 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2660 int error
= may_delete(dir
, dentry
, 0);
2665 if (!dir
->i_op
->unlink
)
2668 mutex_lock(&dentry
->d_inode
->i_mutex
);
2669 if (d_mountpoint(dentry
))
2672 error
= security_inode_unlink(dir
, dentry
);
2674 error
= dir
->i_op
->unlink(dir
, dentry
);
2679 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2681 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2682 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2683 fsnotify_link_count(dentry
->d_inode
);
2691 * Make sure that the actual truncation of the file will occur outside its
2692 * directory's i_mutex. Truncate can take a long time if there is a lot of
2693 * writeout happening, and we don't want to prevent access to the directory
2694 * while waiting on the I/O.
2696 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2700 struct dentry
*dentry
;
2701 struct nameidata nd
;
2702 struct inode
*inode
= NULL
;
2704 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2709 if (nd
.last_type
!= LAST_NORM
)
2712 nd
.flags
&= ~LOOKUP_PARENT
;
2714 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2715 dentry
= lookup_hash(&nd
);
2716 error
= PTR_ERR(dentry
);
2717 if (!IS_ERR(dentry
)) {
2718 /* Why not before? Because we want correct error value */
2719 if (nd
.last
.name
[nd
.last
.len
])
2721 inode
= dentry
->d_inode
;
2725 error
= mnt_want_write(nd
.path
.mnt
);
2728 error
= security_path_unlink(&nd
.path
, dentry
);
2731 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2733 mnt_drop_write(nd
.path
.mnt
);
2737 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2739 iput(inode
); /* truncate the inode here */
2746 error
= !dentry
->d_inode
? -ENOENT
:
2747 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2751 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2753 if ((flag
& ~AT_REMOVEDIR
) != 0)
2756 if (flag
& AT_REMOVEDIR
)
2757 return do_rmdir(dfd
, pathname
);
2759 return do_unlinkat(dfd
, pathname
);
2762 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2764 return do_unlinkat(AT_FDCWD
, pathname
);
2767 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2769 int error
= may_create(dir
, dentry
);
2774 if (!dir
->i_op
->symlink
)
2777 error
= security_inode_symlink(dir
, dentry
, oldname
);
2781 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2783 fsnotify_create(dir
, dentry
);
2787 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2788 int, newdfd
, const char __user
*, newname
)
2793 struct dentry
*dentry
;
2794 struct nameidata nd
;
2796 from
= getname(oldname
);
2798 return PTR_ERR(from
);
2800 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2804 dentry
= lookup_create(&nd
, 0);
2805 error
= PTR_ERR(dentry
);
2809 error
= mnt_want_write(nd
.path
.mnt
);
2812 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2814 goto out_drop_write
;
2815 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2817 mnt_drop_write(nd
.path
.mnt
);
2821 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2829 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2831 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2834 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2836 struct inode
*inode
= old_dentry
->d_inode
;
2842 error
= may_create(dir
, new_dentry
);
2846 if (dir
->i_sb
!= inode
->i_sb
)
2850 * A link to an append-only or immutable file cannot be created.
2852 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2854 if (!dir
->i_op
->link
)
2856 if (S_ISDIR(inode
->i_mode
))
2859 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2863 mutex_lock(&inode
->i_mutex
);
2864 /* Make sure we don't allow creating hardlink to an unlinked file */
2865 if (inode
->i_nlink
== 0)
2868 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2869 mutex_unlock(&inode
->i_mutex
);
2871 fsnotify_link(dir
, inode
, new_dentry
);
2876 * Hardlinks are often used in delicate situations. We avoid
2877 * security-related surprises by not following symlinks on the
2880 * We don't follow them on the oldname either to be compatible
2881 * with linux 2.0, and to avoid hard-linking to directories
2882 * and other special files. --ADM
2884 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2885 int, newdfd
, const char __user
*, newname
, int, flags
)
2887 struct dentry
*new_dentry
;
2888 struct nameidata nd
;
2889 struct path old_path
;
2894 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
2897 * To use null names we require CAP_DAC_READ_SEARCH
2898 * This ensures that not everyone will be able to create
2899 * handlink using the passed filedescriptor.
2901 if (flags
& AT_EMPTY_PATH
) {
2902 if (!capable(CAP_DAC_READ_SEARCH
))
2907 if (flags
& AT_SYMLINK_FOLLOW
)
2908 how
|= LOOKUP_FOLLOW
;
2910 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
2914 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2918 if (old_path
.mnt
!= nd
.path
.mnt
)
2920 new_dentry
= lookup_create(&nd
, 0);
2921 error
= PTR_ERR(new_dentry
);
2922 if (IS_ERR(new_dentry
))
2924 error
= mnt_want_write(nd
.path
.mnt
);
2927 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2929 goto out_drop_write
;
2930 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2932 mnt_drop_write(nd
.path
.mnt
);
2936 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2941 path_put(&old_path
);
2946 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
2948 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2952 * The worst of all namespace operations - renaming directory. "Perverted"
2953 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2955 * a) we can get into loop creation. Check is done in is_subdir().
2956 * b) race potential - two innocent renames can create a loop together.
2957 * That's where 4.4 screws up. Current fix: serialization on
2958 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2960 * c) we have to lock _three_ objects - parents and victim (if it exists).
2961 * And that - after we got ->i_mutex on parents (until then we don't know
2962 * whether the target exists). Solution: try to be smart with locking
2963 * order for inodes. We rely on the fact that tree topology may change
2964 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2965 * move will be locked. Thus we can rank directories by the tree
2966 * (ancestors first) and rank all non-directories after them.
2967 * That works since everybody except rename does "lock parent, lookup,
2968 * lock child" and rename is under ->s_vfs_rename_mutex.
2969 * HOWEVER, it relies on the assumption that any object with ->lookup()
2970 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2971 * we'd better make sure that there's no link(2) for them.
2972 * d) conversion from fhandle to dentry may come in the wrong moment - when
2973 * we are removing the target. Solution: we will have to grab ->i_mutex
2974 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2975 * ->i_mutex on parents, which works but leads to some truly excessive
2978 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2979 struct inode
*new_dir
, struct dentry
*new_dentry
)
2982 struct inode
*target
= new_dentry
->d_inode
;
2985 * If we are going to change the parent - check write permissions,
2986 * we'll need to flip '..'.
2988 if (new_dir
!= old_dir
) {
2989 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
2994 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2999 mutex_lock(&target
->i_mutex
);
3002 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3006 shrink_dcache_parent(new_dentry
);
3007 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3012 target
->i_flags
|= S_DEAD
;
3013 dont_mount(new_dentry
);
3017 mutex_unlock(&target
->i_mutex
);
3019 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3020 d_move(old_dentry
,new_dentry
);
3024 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3025 struct inode
*new_dir
, struct dentry
*new_dentry
)
3027 struct inode
*target
= new_dentry
->d_inode
;
3030 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3036 mutex_lock(&target
->i_mutex
);
3039 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3042 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3047 dont_mount(new_dentry
);
3048 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3049 d_move(old_dentry
, new_dentry
);
3052 mutex_unlock(&target
->i_mutex
);
3057 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3058 struct inode
*new_dir
, struct dentry
*new_dentry
)
3061 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3062 const unsigned char *old_name
;
3064 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3067 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3071 if (!new_dentry
->d_inode
)
3072 error
= may_create(new_dir
, new_dentry
);
3074 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3078 if (!old_dir
->i_op
->rename
)
3081 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3084 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3086 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3088 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3089 new_dentry
->d_inode
, old_dentry
);
3090 fsnotify_oldname_free(old_name
);
3095 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3096 int, newdfd
, const char __user
*, newname
)
3098 struct dentry
*old_dir
, *new_dir
;
3099 struct dentry
*old_dentry
, *new_dentry
;
3100 struct dentry
*trap
;
3101 struct nameidata oldnd
, newnd
;
3106 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3110 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3115 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3118 old_dir
= oldnd
.path
.dentry
;
3120 if (oldnd
.last_type
!= LAST_NORM
)
3123 new_dir
= newnd
.path
.dentry
;
3124 if (newnd
.last_type
!= LAST_NORM
)
3127 oldnd
.flags
&= ~LOOKUP_PARENT
;
3128 newnd
.flags
&= ~LOOKUP_PARENT
;
3129 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3131 trap
= lock_rename(new_dir
, old_dir
);
3133 old_dentry
= lookup_hash(&oldnd
);
3134 error
= PTR_ERR(old_dentry
);
3135 if (IS_ERR(old_dentry
))
3137 /* source must exist */
3139 if (!old_dentry
->d_inode
)
3141 /* unless the source is a directory trailing slashes give -ENOTDIR */
3142 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3144 if (oldnd
.last
.name
[oldnd
.last
.len
])
3146 if (newnd
.last
.name
[newnd
.last
.len
])
3149 /* source should not be ancestor of target */
3151 if (old_dentry
== trap
)
3153 new_dentry
= lookup_hash(&newnd
);
3154 error
= PTR_ERR(new_dentry
);
3155 if (IS_ERR(new_dentry
))
3157 /* target should not be an ancestor of source */
3159 if (new_dentry
== trap
)
3162 error
= mnt_want_write(oldnd
.path
.mnt
);
3165 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3166 &newnd
.path
, new_dentry
);
3169 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3170 new_dir
->d_inode
, new_dentry
);
3172 mnt_drop_write(oldnd
.path
.mnt
);
3178 unlock_rename(new_dir
, old_dir
);
3180 path_put(&newnd
.path
);
3183 path_put(&oldnd
.path
);
3189 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3191 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3194 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3198 len
= PTR_ERR(link
);
3203 if (len
> (unsigned) buflen
)
3205 if (copy_to_user(buffer
, link
, len
))
3212 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3213 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3214 * using) it for any given inode is up to filesystem.
3216 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3218 struct nameidata nd
;
3223 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3225 return PTR_ERR(cookie
);
3227 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3228 if (dentry
->d_inode
->i_op
->put_link
)
3229 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3233 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3235 return __vfs_follow_link(nd
, link
);
3238 /* get the link contents into pagecache */
3239 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3243 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3244 page
= read_mapping_page(mapping
, 0, NULL
);
3249 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3253 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3255 struct page
*page
= NULL
;
3256 char *s
= page_getlink(dentry
, &page
);
3257 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3260 page_cache_release(page
);
3265 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3267 struct page
*page
= NULL
;
3268 nd_set_link(nd
, page_getlink(dentry
, &page
));
3272 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3274 struct page
*page
= cookie
;
3278 page_cache_release(page
);
3283 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3285 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3287 struct address_space
*mapping
= inode
->i_mapping
;
3292 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3294 flags
|= AOP_FLAG_NOFS
;
3297 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3298 flags
, &page
, &fsdata
);
3302 kaddr
= kmap_atomic(page
, KM_USER0
);
3303 memcpy(kaddr
, symname
, len
-1);
3304 kunmap_atomic(kaddr
, KM_USER0
);
3306 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3313 mark_inode_dirty(inode
);
3319 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3321 return __page_symlink(inode
, symname
, len
,
3322 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3325 const struct inode_operations page_symlink_inode_operations
= {
3326 .readlink
= generic_readlink
,
3327 .follow_link
= page_follow_link_light
,
3328 .put_link
= page_put_link
,
3331 EXPORT_SYMBOL(user_path_at
);
3332 EXPORT_SYMBOL(follow_down_one
);
3333 EXPORT_SYMBOL(follow_down
);
3334 EXPORT_SYMBOL(follow_up
);
3335 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3336 EXPORT_SYMBOL(getname
);
3337 EXPORT_SYMBOL(lock_rename
);
3338 EXPORT_SYMBOL(lookup_one_len
);
3339 EXPORT_SYMBOL(page_follow_link_light
);
3340 EXPORT_SYMBOL(page_put_link
);
3341 EXPORT_SYMBOL(page_readlink
);
3342 EXPORT_SYMBOL(__page_symlink
);
3343 EXPORT_SYMBOL(page_symlink
);
3344 EXPORT_SYMBOL(page_symlink_inode_operations
);
3345 EXPORT_SYMBOL(kern_path_parent
);
3346 EXPORT_SYMBOL(kern_path
);
3347 EXPORT_SYMBOL(vfs_path_lookup
);
3348 EXPORT_SYMBOL(inode_permission
);
3349 EXPORT_SYMBOL(unlock_rename
);
3350 EXPORT_SYMBOL(vfs_create
);
3351 EXPORT_SYMBOL(vfs_follow_link
);
3352 EXPORT_SYMBOL(vfs_link
);
3353 EXPORT_SYMBOL(vfs_mkdir
);
3354 EXPORT_SYMBOL(vfs_mknod
);
3355 EXPORT_SYMBOL(generic_permission
);
3356 EXPORT_SYMBOL(vfs_readlink
);
3357 EXPORT_SYMBOL(vfs_rename
);
3358 EXPORT_SYMBOL(vfs_rmdir
);
3359 EXPORT_SYMBOL(vfs_symlink
);
3360 EXPORT_SYMBOL(vfs_unlink
);
3361 EXPORT_SYMBOL(dentry_unhash
);
3362 EXPORT_SYMBOL(generic_readlink
);