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-existant 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 char * getname(const char __user
* filename
)
143 result
= ERR_PTR(-ENOMEM
);
146 int retval
= do_getname(filename
, tmp
);
151 result
= ERR_PTR(retval
);
154 audit_getname(result
);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name
)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname
);
170 * This does basic POSIX ACL permission checking
172 static int acl_permission_check(struct inode
*inode
, int mask
, unsigned int flags
,
173 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
175 umode_t mode
= inode
->i_mode
;
177 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
179 if (current_fsuid() == inode
->i_uid
)
182 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
183 int error
= check_acl(inode
, mask
, flags
);
184 if (error
!= -EAGAIN
)
188 if (in_group_p(inode
->i_gid
))
193 * If the DACs are ok we don't need any capability check.
195 if ((mask
& ~mode
) == 0)
201 * generic_permission - check for access rights on a Posix-like filesystem
202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
205 * @flags: IPERM_FLAG_ flags.
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
210 * are used for other things.
212 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
213 * request cannot be satisfied (eg. requires blocking or too much complexity).
214 * It would then be called again in ref-walk mode.
216 int generic_permission(struct inode
*inode
, int mask
, unsigned int flags
,
217 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
222 * Do the basic POSIX ACL permission checks.
224 ret
= acl_permission_check(inode
, mask
, flags
, check_acl
);
229 * Read/write DACs are always overridable.
230 * Executable DACs are overridable if at least one exec bit is set.
232 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
233 if (capable(CAP_DAC_OVERRIDE
))
237 * Searching includes executable on directories, else just read.
239 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
240 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
241 if (capable(CAP_DAC_READ_SEARCH
))
248 * inode_permission - check for access rights to a given inode
249 * @inode: inode to check permission on
250 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
252 * Used to check for read/write/execute permissions on an inode.
253 * We use "fsuid" for this, letting us set arbitrary permissions
254 * for filesystem access without changing the "normal" uids which
255 * are used for other things.
257 int inode_permission(struct inode
*inode
, int mask
)
261 if (mask
& MAY_WRITE
) {
262 umode_t mode
= inode
->i_mode
;
265 * Nobody gets write access to a read-only fs.
267 if (IS_RDONLY(inode
) &&
268 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
272 * Nobody gets write access to an immutable file.
274 if (IS_IMMUTABLE(inode
))
278 if (inode
->i_op
->permission
)
279 retval
= inode
->i_op
->permission(inode
, mask
, 0);
281 retval
= generic_permission(inode
, mask
, 0,
282 inode
->i_op
->check_acl
);
287 retval
= devcgroup_inode_permission(inode
, mask
);
291 return security_inode_permission(inode
, mask
);
295 * file_permission - check for additional access rights to a given file
296 * @file: file to check access rights for
297 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
299 * Used to check for read/write/execute permissions on an already opened
303 * Do not use this function in new code. All access checks should
304 * be done using inode_permission().
306 int file_permission(struct file
*file
, int mask
)
308 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
312 * get_write_access() gets write permission for a file.
313 * put_write_access() releases this write permission.
314 * This is used for regular files.
315 * We cannot support write (and maybe mmap read-write shared) accesses and
316 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
317 * can have the following values:
318 * 0: no writers, no VM_DENYWRITE mappings
319 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
320 * > 0: (i_writecount) users are writing to the file.
322 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
323 * except for the cases where we don't hold i_writecount yet. Then we need to
324 * use {get,deny}_write_access() - these functions check the sign and refuse
325 * to do the change if sign is wrong. Exclusion between them is provided by
326 * the inode->i_lock spinlock.
329 int get_write_access(struct inode
* inode
)
331 spin_lock(&inode
->i_lock
);
332 if (atomic_read(&inode
->i_writecount
) < 0) {
333 spin_unlock(&inode
->i_lock
);
336 atomic_inc(&inode
->i_writecount
);
337 spin_unlock(&inode
->i_lock
);
342 int deny_write_access(struct file
* file
)
344 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
346 spin_lock(&inode
->i_lock
);
347 if (atomic_read(&inode
->i_writecount
) > 0) {
348 spin_unlock(&inode
->i_lock
);
351 atomic_dec(&inode
->i_writecount
);
352 spin_unlock(&inode
->i_lock
);
358 * path_get - get a reference to a path
359 * @path: path to get the reference to
361 * Given a path increment the reference count to the dentry and the vfsmount.
363 void path_get(struct path
*path
)
368 EXPORT_SYMBOL(path_get
);
371 * path_put - put a reference to a path
372 * @path: path to put the reference to
374 * Given a path decrement the reference count to the dentry and the vfsmount.
376 void path_put(struct path
*path
)
381 EXPORT_SYMBOL(path_put
);
384 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
385 * @nd: nameidata pathwalk data to drop
386 * Returns: 0 on success, -ECHILD on failure
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
390 * to drop out of rcu-walk mode and take normal reference counts on dentries
391 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
392 * refcounts at the last known good point before rcu-walk got stuck, so
393 * ref-walk may continue from there. If this is not successful (eg. a seqcount
394 * has changed), then failure is returned and path walk restarts from the
395 * beginning in ref-walk mode.
397 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
398 * ref-walk. Must be called from rcu-walk context.
400 static int nameidata_drop_rcu(struct nameidata
*nd
)
402 struct fs_struct
*fs
= current
->fs
;
403 struct dentry
*dentry
= nd
->path
.dentry
;
405 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
407 spin_lock(&fs
->lock
);
408 if (nd
->root
.mnt
!= fs
->root
.mnt
||
409 nd
->root
.dentry
!= fs
->root
.dentry
)
412 spin_lock(&dentry
->d_lock
);
413 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
415 BUG_ON(nd
->inode
!= dentry
->d_inode
);
416 spin_unlock(&dentry
->d_lock
);
419 spin_unlock(&fs
->lock
);
421 mntget(nd
->path
.mnt
);
424 br_read_unlock(vfsmount_lock
);
425 nd
->flags
&= ~LOOKUP_RCU
;
428 spin_unlock(&dentry
->d_lock
);
431 spin_unlock(&fs
->lock
);
435 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
436 static inline int nameidata_drop_rcu_maybe(struct nameidata
*nd
)
438 if (nd
->flags
& LOOKUP_RCU
)
439 return nameidata_drop_rcu(nd
);
444 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
445 * @nd: nameidata pathwalk data to drop
446 * @dentry: dentry to drop
447 * Returns: 0 on success, -ECHILD on failure
449 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
450 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
451 * @nd. Must be called from rcu-walk context.
453 static int nameidata_dentry_drop_rcu(struct nameidata
*nd
, struct dentry
*dentry
)
455 struct fs_struct
*fs
= current
->fs
;
456 struct dentry
*parent
= nd
->path
.dentry
;
459 * It can be possible to revalidate the dentry that we started
460 * the path walk with. force_reval_path may also revalidate the
461 * dentry already committed to the nameidata.
463 if (unlikely(parent
== dentry
))
464 return nameidata_drop_rcu(nd
);
466 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
468 spin_lock(&fs
->lock
);
469 if (nd
->root
.mnt
!= fs
->root
.mnt
||
470 nd
->root
.dentry
!= fs
->root
.dentry
)
473 spin_lock(&parent
->d_lock
);
474 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
475 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
478 * If the sequence check on the child dentry passed, then the child has
479 * not been removed from its parent. This means the parent dentry must
480 * be valid and able to take a reference at this point.
482 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
483 BUG_ON(!parent
->d_count
);
485 spin_unlock(&dentry
->d_lock
);
486 spin_unlock(&parent
->d_lock
);
489 spin_unlock(&fs
->lock
);
491 mntget(nd
->path
.mnt
);
494 br_read_unlock(vfsmount_lock
);
495 nd
->flags
&= ~LOOKUP_RCU
;
498 spin_unlock(&dentry
->d_lock
);
499 spin_unlock(&parent
->d_lock
);
502 spin_unlock(&fs
->lock
);
506 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
507 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata
*nd
, struct dentry
*dentry
)
509 if (nd
->flags
& LOOKUP_RCU
)
510 return nameidata_dentry_drop_rcu(nd
, dentry
);
515 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
516 * @nd: nameidata pathwalk data to drop
517 * Returns: 0 on success, -ECHILD on failure
519 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
520 * nd->path should be the final element of the lookup, so nd->root is discarded.
521 * Must be called from rcu-walk context.
523 static int nameidata_drop_rcu_last(struct nameidata
*nd
)
525 struct dentry
*dentry
= nd
->path
.dentry
;
527 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
528 nd
->flags
&= ~LOOKUP_RCU
;
530 spin_lock(&dentry
->d_lock
);
531 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
533 BUG_ON(nd
->inode
!= dentry
->d_inode
);
534 spin_unlock(&dentry
->d_lock
);
536 mntget(nd
->path
.mnt
);
539 br_read_unlock(vfsmount_lock
);
544 spin_unlock(&dentry
->d_lock
);
546 br_read_unlock(vfsmount_lock
);
550 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
551 static inline int nameidata_drop_rcu_last_maybe(struct nameidata
*nd
)
553 if (likely(nd
->flags
& LOOKUP_RCU
))
554 return nameidata_drop_rcu_last(nd
);
559 * release_open_intent - free up open intent resources
560 * @nd: pointer to nameidata
562 void release_open_intent(struct nameidata
*nd
)
564 struct file
*file
= nd
->intent
.open
.file
;
566 if (file
&& !IS_ERR(file
)) {
567 if (file
->f_path
.dentry
== NULL
)
575 * Call d_revalidate and handle filesystems that request rcu-walk
576 * to be dropped. This may be called and return in rcu-walk mode,
577 * regardless of success or error. If -ECHILD is returned, the caller
578 * must return -ECHILD back up the path walk stack so path walk may
579 * be restarted in ref-walk mode.
581 static int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
585 status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
586 if (status
== -ECHILD
) {
587 if (nameidata_dentry_drop_rcu(nd
, dentry
))
589 status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
595 static struct dentry
*
596 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
598 int status
= d_revalidate(dentry
, nd
);
599 if (unlikely(status
<= 0)) {
601 * The dentry failed validation.
602 * If d_revalidate returned 0 attempt to invalidate
603 * the dentry otherwise d_revalidate is asking us
604 * to return a fail status.
608 dentry
= ERR_PTR(status
);
609 } else if (!d_invalidate(dentry
)) {
617 static inline struct dentry
*
618 do_revalidate_rcu(struct dentry
*dentry
, struct nameidata
*nd
)
620 int status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
621 if (likely(status
> 0))
623 if (status
== -ECHILD
) {
624 if (nameidata_dentry_drop_rcu(nd
, dentry
))
625 return ERR_PTR(-ECHILD
);
626 return do_revalidate(dentry
, nd
);
629 return ERR_PTR(status
);
630 /* Don't d_invalidate in rcu-walk mode */
631 if (nameidata_dentry_drop_rcu(nd
, dentry
))
632 return ERR_PTR(-ECHILD
);
633 if (!d_invalidate(dentry
)) {
640 static inline int need_reval_dot(struct dentry
*dentry
)
642 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
645 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
652 * force_reval_path - force revalidation of a dentry
654 * In some situations the path walking code will trust dentries without
655 * revalidating them. This causes problems for filesystems that depend on
656 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
657 * (which indicates that it's possible for the dentry to go stale), force
658 * a d_revalidate call before proceeding.
660 * Returns 0 if the revalidation was successful. If the revalidation fails,
661 * either return the error returned by d_revalidate or -ESTALE if the
662 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
663 * invalidate the dentry. It's up to the caller to handle putting references
664 * to the path if necessary.
667 force_reval_path(struct path
*path
, struct nameidata
*nd
)
670 struct dentry
*dentry
= path
->dentry
;
673 * only check on filesystems where it's possible for the dentry to
676 if (!need_reval_dot(dentry
))
679 status
= d_revalidate(dentry
, nd
);
684 d_invalidate(dentry
);
691 * Short-cut version of permission(), for calling on directories
692 * during pathname resolution. Combines parts of permission()
693 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
695 * If appropriate, check DAC only. If not appropriate, or
696 * short-cut DAC fails, then call ->permission() to do more
697 * complete permission check.
699 static inline int exec_permission(struct inode
*inode
, unsigned int flags
)
703 if (inode
->i_op
->permission
) {
704 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
, flags
);
706 ret
= acl_permission_check(inode
, MAY_EXEC
, flags
,
707 inode
->i_op
->check_acl
);
714 if (capable(CAP_DAC_OVERRIDE
) || capable(CAP_DAC_READ_SEARCH
))
719 return security_inode_exec_permission(inode
, flags
);
722 static __always_inline
void set_root(struct nameidata
*nd
)
725 get_fs_root(current
->fs
, &nd
->root
);
728 static int link_path_walk(const char *, struct nameidata
*);
730 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
733 struct fs_struct
*fs
= current
->fs
;
737 seq
= read_seqcount_begin(&fs
->seq
);
739 } while (read_seqcount_retry(&fs
->seq
, seq
));
743 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
756 nd
->inode
= nd
->path
.dentry
->d_inode
;
758 ret
= link_path_walk(link
, nd
);
762 return PTR_ERR(link
);
765 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
768 if (path
->mnt
!= nd
->path
.mnt
)
772 static inline void path_to_nameidata(const struct path
*path
,
773 struct nameidata
*nd
)
775 if (!(nd
->flags
& LOOKUP_RCU
)) {
776 dput(nd
->path
.dentry
);
777 if (nd
->path
.mnt
!= path
->mnt
)
778 mntput(nd
->path
.mnt
);
780 nd
->path
.mnt
= path
->mnt
;
781 nd
->path
.dentry
= path
->dentry
;
784 static __always_inline
int
785 __do_follow_link(const struct path
*link
, struct nameidata
*nd
, void **p
)
788 struct dentry
*dentry
= link
->dentry
;
790 BUG_ON(nd
->flags
& LOOKUP_RCU
);
792 touch_atime(link
->mnt
, dentry
);
793 nd_set_link(nd
, NULL
);
795 if (link
->mnt
== nd
->path
.mnt
)
798 nd
->last_type
= LAST_BIND
;
799 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
802 char *s
= nd_get_link(nd
);
805 error
= __vfs_follow_link(nd
, s
);
806 else if (nd
->last_type
== LAST_BIND
) {
807 error
= force_reval_path(&nd
->path
, nd
);
816 * This limits recursive symlink follows to 8, while
817 * limiting consecutive symlinks to 40.
819 * Without that kind of total limit, nasty chains of consecutive
820 * symlinks can cause almost arbitrarily long lookups.
822 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
827 /* We drop rcu-walk here */
828 if (nameidata_dentry_drop_rcu_maybe(nd
, path
->dentry
))
831 if (current
->link_count
>= MAX_NESTED_LINKS
)
833 if (current
->total_link_count
>= 40)
835 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
837 err
= security_inode_follow_link(path
->dentry
, nd
);
840 current
->link_count
++;
841 current
->total_link_count
++;
843 err
= __do_follow_link(path
, nd
, &cookie
);
844 if (!IS_ERR(cookie
) && path
->dentry
->d_inode
->i_op
->put_link
)
845 path
->dentry
->d_inode
->i_op
->put_link(path
->dentry
, nd
, cookie
);
847 current
->link_count
--;
851 path_put_conditional(path
, nd
);
856 static int follow_up_rcu(struct path
*path
)
858 struct vfsmount
*parent
;
859 struct dentry
*mountpoint
;
861 parent
= path
->mnt
->mnt_parent
;
862 if (parent
== path
->mnt
)
864 mountpoint
= path
->mnt
->mnt_mountpoint
;
865 path
->dentry
= mountpoint
;
870 int follow_up(struct path
*path
)
872 struct vfsmount
*parent
;
873 struct dentry
*mountpoint
;
875 br_read_lock(vfsmount_lock
);
876 parent
= path
->mnt
->mnt_parent
;
877 if (parent
== path
->mnt
) {
878 br_read_unlock(vfsmount_lock
);
882 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
883 br_read_unlock(vfsmount_lock
);
885 path
->dentry
= mountpoint
;
892 * Perform an automount
893 * - return -EISDIR to tell follow_managed() to stop and return the path we
896 static int follow_automount(struct path
*path
, unsigned flags
,
899 struct vfsmount
*mnt
;
902 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
905 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
906 * and this is the terminal part of the path.
908 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
909 return -EISDIR
; /* we actually want to stop here */
911 /* We want to mount if someone is trying to open/create a file of any
912 * type under the mountpoint, wants to traverse through the mountpoint
913 * or wants to open the mounted directory.
915 * We don't want to mount if someone's just doing a stat and they've
916 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
917 * appended a '/' to the name.
919 if (!(flags
& LOOKUP_FOLLOW
) &&
920 !(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
921 LOOKUP_OPEN
| LOOKUP_CREATE
)))
924 current
->total_link_count
++;
925 if (current
->total_link_count
>= 40)
928 mnt
= path
->dentry
->d_op
->d_automount(path
);
931 * The filesystem is allowed to return -EISDIR here to indicate
932 * it doesn't want to automount. For instance, autofs would do
933 * this so that its userspace daemon can mount on this dentry.
935 * However, we can only permit this if it's a terminal point in
936 * the path being looked up; if it wasn't then the remainder of
937 * the path is inaccessible and we should say so.
939 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_CONTINUE
))
944 if (!mnt
) /* mount collision */
947 err
= finish_automount(mnt
, path
);
951 /* Someone else made a mount here whilst we were busy */
958 path
->dentry
= dget(mnt
->mnt_root
);
968 * Handle a dentry that is managed in some way.
969 * - Flagged for transit management (autofs)
970 * - Flagged as mountpoint
971 * - Flagged as automount point
973 * This may only be called in refwalk mode.
975 * Serialization is taken care of in namespace.c
977 static int follow_managed(struct path
*path
, unsigned flags
)
980 bool need_mntput
= false;
983 /* Given that we're not holding a lock here, we retain the value in a
984 * local variable for each dentry as we look at it so that we don't see
985 * the components of that value change under us */
986 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
987 managed
&= DCACHE_MANAGED_DENTRY
,
988 unlikely(managed
!= 0)) {
989 /* Allow the filesystem to manage the transit without i_mutex
991 if (managed
& DCACHE_MANAGE_TRANSIT
) {
992 BUG_ON(!path
->dentry
->d_op
);
993 BUG_ON(!path
->dentry
->d_op
->d_manage
);
994 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
,
997 return ret
== -EISDIR
? 0 : ret
;
1000 /* Transit to a mounted filesystem. */
1001 if (managed
& DCACHE_MOUNTED
) {
1002 struct vfsmount
*mounted
= lookup_mnt(path
);
1007 path
->mnt
= mounted
;
1008 path
->dentry
= dget(mounted
->mnt_root
);
1013 /* Something is mounted on this dentry in another
1014 * namespace and/or whatever was mounted there in this
1015 * namespace got unmounted before we managed to get the
1019 /* Handle an automount point */
1020 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1021 ret
= follow_automount(path
, flags
, &need_mntput
);
1023 return ret
== -EISDIR
? 0 : ret
;
1027 /* We didn't change the current path point */
1033 int follow_down_one(struct path
*path
)
1035 struct vfsmount
*mounted
;
1037 mounted
= lookup_mnt(path
);
1041 path
->mnt
= mounted
;
1042 path
->dentry
= dget(mounted
->mnt_root
);
1049 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
1050 * meet a managed dentry and we're not walking to "..". True is returned to
1051 * continue, false to abort.
1053 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1054 struct inode
**inode
, bool reverse_transit
)
1056 while (d_mountpoint(path
->dentry
)) {
1057 struct vfsmount
*mounted
;
1058 if (unlikely(path
->dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) &&
1060 path
->dentry
->d_op
->d_manage(path
->dentry
, false, true) < 0)
1062 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1065 path
->mnt
= mounted
;
1066 path
->dentry
= mounted
->mnt_root
;
1067 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1068 *inode
= path
->dentry
->d_inode
;
1071 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1072 return reverse_transit
;
1076 static int follow_dotdot_rcu(struct nameidata
*nd
)
1078 struct inode
*inode
= nd
->inode
;
1083 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1084 nd
->path
.mnt
== nd
->root
.mnt
) {
1087 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1088 struct dentry
*old
= nd
->path
.dentry
;
1089 struct dentry
*parent
= old
->d_parent
;
1092 seq
= read_seqcount_begin(&parent
->d_seq
);
1093 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1095 inode
= parent
->d_inode
;
1096 nd
->path
.dentry
= parent
;
1100 if (!follow_up_rcu(&nd
->path
))
1102 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1103 inode
= nd
->path
.dentry
->d_inode
;
1105 __follow_mount_rcu(nd
, &nd
->path
, &inode
, true);
1112 * Follow down to the covering mount currently visible to userspace. At each
1113 * point, the filesystem owning that dentry may be queried as to whether the
1114 * caller is permitted to proceed or not.
1116 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1119 int follow_down(struct path
*path
, bool mounting_here
)
1124 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1125 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1126 /* Allow the filesystem to manage the transit without i_mutex
1129 * We indicate to the filesystem if someone is trying to mount
1130 * something here. This gives autofs the chance to deny anyone
1131 * other than its daemon the right to mount on its
1134 * The filesystem may sleep at this point.
1136 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1137 BUG_ON(!path
->dentry
->d_op
);
1138 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1139 ret
= path
->dentry
->d_op
->d_manage(
1140 path
->dentry
, mounting_here
, false);
1142 return ret
== -EISDIR
? 0 : ret
;
1145 /* Transit to a mounted filesystem. */
1146 if (managed
& DCACHE_MOUNTED
) {
1147 struct vfsmount
*mounted
= lookup_mnt(path
);
1152 path
->mnt
= mounted
;
1153 path
->dentry
= dget(mounted
->mnt_root
);
1157 /* Don't handle automount points here */
1164 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1166 static void follow_mount(struct path
*path
)
1168 while (d_mountpoint(path
->dentry
)) {
1169 struct vfsmount
*mounted
= lookup_mnt(path
);
1174 path
->mnt
= mounted
;
1175 path
->dentry
= dget(mounted
->mnt_root
);
1179 static void follow_dotdot(struct nameidata
*nd
)
1184 struct dentry
*old
= nd
->path
.dentry
;
1186 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1187 nd
->path
.mnt
== nd
->root
.mnt
) {
1190 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1191 /* rare case of legitimate dget_parent()... */
1192 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1196 if (!follow_up(&nd
->path
))
1199 follow_mount(&nd
->path
);
1200 nd
->inode
= nd
->path
.dentry
->d_inode
;
1204 * Allocate a dentry with name and parent, and perform a parent
1205 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1206 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1207 * have verified that no child exists while under i_mutex.
1209 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1210 struct qstr
*name
, struct nameidata
*nd
)
1212 struct inode
*inode
= parent
->d_inode
;
1213 struct dentry
*dentry
;
1216 /* Don't create child dentry for a dead directory. */
1217 if (unlikely(IS_DEADDIR(inode
)))
1218 return ERR_PTR(-ENOENT
);
1220 dentry
= d_alloc(parent
, name
);
1221 if (unlikely(!dentry
))
1222 return ERR_PTR(-ENOMEM
);
1224 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1225 if (unlikely(old
)) {
1233 * It's more convoluted than I'd like it to be, but... it's still fairly
1234 * small and for now I'd prefer to have fast path as straight as possible.
1235 * It _is_ time-critical.
1237 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1238 struct path
*path
, struct inode
**inode
)
1240 struct vfsmount
*mnt
= nd
->path
.mnt
;
1241 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1246 * See if the low-level filesystem might want
1247 * to use its own hash..
1249 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1250 err
= parent
->d_op
->d_hash(parent
, nd
->inode
, name
);
1256 * Rename seqlock is not required here because in the off chance
1257 * of a false negative due to a concurrent rename, we're going to
1258 * do the non-racy lookup, below.
1260 if (nd
->flags
& LOOKUP_RCU
) {
1264 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1266 if (nameidata_drop_rcu(nd
))
1270 /* Memory barrier in read_seqcount_begin of child is enough */
1271 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1275 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1276 dentry
= do_revalidate_rcu(dentry
, nd
);
1281 if (!(nd
->flags
& LOOKUP_RCU
))
1285 path
->dentry
= dentry
;
1286 if (likely(__follow_mount_rcu(nd
, path
, inode
, false)))
1288 if (nameidata_drop_rcu(nd
))
1292 dentry
= __d_lookup(parent
, name
);
1296 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1297 dentry
= do_revalidate(dentry
, nd
);
1305 path
->dentry
= dentry
;
1306 err
= follow_managed(path
, nd
->flags
);
1307 if (unlikely(err
< 0)) {
1308 path_put_conditional(path
, nd
);
1311 *inode
= path
->dentry
->d_inode
;
1315 dir
= parent
->d_inode
;
1316 BUG_ON(nd
->inode
!= dir
);
1318 mutex_lock(&dir
->i_mutex
);
1320 * First re-do the cached lookup just in case it was created
1321 * while we waited for the directory semaphore, or the first
1322 * lookup failed due to an unrelated rename.
1324 * This could use version numbering or similar to avoid unnecessary
1325 * cache lookups, but then we'd have to do the first lookup in the
1326 * non-racy way. However in the common case here, everything should
1327 * be hot in cache, so would it be a big win?
1329 dentry
= d_lookup(parent
, name
);
1330 if (likely(!dentry
)) {
1331 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1332 mutex_unlock(&dir
->i_mutex
);
1338 * Uhhuh! Nasty case: the cache was re-populated while
1339 * we waited on the semaphore. Need to revalidate.
1341 mutex_unlock(&dir
->i_mutex
);
1345 return PTR_ERR(dentry
);
1350 * This is the basic name resolution function, turning a pathname into
1351 * the final dentry. We expect 'base' to be positive and a directory.
1353 * Returns 0 and nd will have valid dentry and mnt on success.
1354 * Returns error and drops reference to input namei data on failure.
1356 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1360 unsigned int lookup_flags
= nd
->flags
;
1368 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
1370 /* At this point we know we have a real path component. */
1372 struct inode
*inode
;
1377 nd
->flags
|= LOOKUP_CONTINUE
;
1378 if (nd
->flags
& LOOKUP_RCU
) {
1379 err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1380 if (err
== -ECHILD
) {
1381 if (nameidata_drop_rcu(nd
))
1387 err
= exec_permission(nd
->inode
, 0);
1393 c
= *(const unsigned char *)name
;
1395 hash
= init_name_hash();
1398 hash
= partial_name_hash(c
, hash
);
1399 c
= *(const unsigned char *)name
;
1400 } while (c
&& (c
!= '/'));
1401 this.len
= name
- (const char *) this.name
;
1402 this.hash
= end_name_hash(hash
);
1404 /* remove trailing slashes? */
1406 goto last_component
;
1407 while (*++name
== '/');
1409 goto last_with_slashes
;
1412 * "." and ".." are special - ".." especially so because it has
1413 * to be able to know about the current root directory and
1414 * parent relationships.
1416 if (this.name
[0] == '.') switch (this.len
) {
1420 if (this.name
[1] != '.')
1422 if (nd
->flags
& LOOKUP_RCU
) {
1423 if (follow_dotdot_rcu(nd
))
1431 /* This does the actual lookups.. */
1432 err
= do_lookup(nd
, &this, &next
, &inode
);
1439 if (inode
->i_op
->follow_link
) {
1440 BUG_ON(inode
!= next
.dentry
->d_inode
);
1441 err
= do_follow_link(&next
, nd
);
1444 nd
->inode
= nd
->path
.dentry
->d_inode
;
1449 path_to_nameidata(&next
, nd
);
1453 if (!nd
->inode
->i_op
->lookup
)
1456 /* here ends the main loop */
1459 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1461 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1462 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1463 if (lookup_flags
& LOOKUP_PARENT
)
1465 if (this.name
[0] == '.') switch (this.len
) {
1469 if (this.name
[1] != '.')
1471 if (nd
->flags
& LOOKUP_RCU
) {
1472 if (follow_dotdot_rcu(nd
))
1480 err
= do_lookup(nd
, &this, &next
, &inode
);
1483 if (inode
&& unlikely(inode
->i_op
->follow_link
) &&
1484 (lookup_flags
& LOOKUP_FOLLOW
)) {
1485 BUG_ON(inode
!= next
.dentry
->d_inode
);
1486 err
= do_follow_link(&next
, nd
);
1489 nd
->inode
= nd
->path
.dentry
->d_inode
;
1491 path_to_nameidata(&next
, nd
);
1497 if (lookup_flags
& LOOKUP_DIRECTORY
) {
1499 if (!nd
->inode
->i_op
->lookup
)
1505 nd
->last_type
= LAST_NORM
;
1506 if (this.name
[0] != '.')
1509 nd
->last_type
= LAST_DOT
;
1510 else if (this.len
== 2 && this.name
[1] == '.')
1511 nd
->last_type
= LAST_DOTDOT
;
1516 * We bypassed the ordinary revalidation routines.
1517 * We may need to check the cached dentry for staleness.
1519 if (need_reval_dot(nd
->path
.dentry
)) {
1520 /* Note: we do not d_invalidate() */
1521 err
= d_revalidate(nd
->path
.dentry
, nd
);
1528 if (nameidata_drop_rcu_last_maybe(nd
))
1532 if (!(nd
->flags
& LOOKUP_RCU
))
1533 path_put_conditional(&next
, nd
);
1536 if (!(nd
->flags
& LOOKUP_RCU
))
1537 path_put(&nd
->path
);
1542 static inline int path_walk_rcu(const char *name
, struct nameidata
*nd
)
1544 current
->total_link_count
= 0;
1546 return link_path_walk(name
, nd
);
1549 static inline int path_walk_simple(const char *name
, struct nameidata
*nd
)
1551 current
->total_link_count
= 0;
1553 return link_path_walk(name
, nd
);
1556 static int path_walk(const char *name
, struct nameidata
*nd
)
1558 struct path save
= nd
->path
;
1561 current
->total_link_count
= 0;
1563 /* make sure the stuff we saved doesn't go away */
1566 result
= link_path_walk(name
, nd
);
1567 if (result
== -ESTALE
) {
1568 /* nd->path had been dropped */
1569 current
->total_link_count
= 0;
1571 path_get(&nd
->path
);
1572 nd
->flags
|= LOOKUP_REVAL
;
1573 result
= link_path_walk(name
, nd
);
1581 static void path_finish_rcu(struct nameidata
*nd
)
1583 if (nd
->flags
& LOOKUP_RCU
) {
1584 /* RCU dangling. Cancel it. */
1585 nd
->flags
&= ~LOOKUP_RCU
;
1586 nd
->root
.mnt
= NULL
;
1588 br_read_unlock(vfsmount_lock
);
1594 static int path_init_rcu(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1600 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1601 nd
->flags
= flags
| LOOKUP_RCU
;
1603 nd
->root
.mnt
= NULL
;
1607 struct fs_struct
*fs
= current
->fs
;
1610 br_read_lock(vfsmount_lock
);
1614 seq
= read_seqcount_begin(&fs
->seq
);
1615 nd
->root
= fs
->root
;
1616 nd
->path
= nd
->root
;
1617 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1618 } while (read_seqcount_retry(&fs
->seq
, seq
));
1620 } else if (dfd
== AT_FDCWD
) {
1621 struct fs_struct
*fs
= current
->fs
;
1624 br_read_lock(vfsmount_lock
);
1628 seq
= read_seqcount_begin(&fs
->seq
);
1630 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1631 } while (read_seqcount_retry(&fs
->seq
, seq
));
1634 struct dentry
*dentry
;
1636 file
= fget_light(dfd
, &fput_needed
);
1641 dentry
= file
->f_path
.dentry
;
1644 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1647 retval
= file_permission(file
, MAY_EXEC
);
1651 nd
->path
= file
->f_path
;
1655 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1656 br_read_lock(vfsmount_lock
);
1659 nd
->inode
= nd
->path
.dentry
->d_inode
;
1663 fput_light(file
, fput_needed
);
1668 static int path_init(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1674 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1677 nd
->root
.mnt
= NULL
;
1681 nd
->path
= nd
->root
;
1682 path_get(&nd
->root
);
1683 } else if (dfd
== AT_FDCWD
) {
1684 get_fs_pwd(current
->fs
, &nd
->path
);
1686 struct dentry
*dentry
;
1688 file
= fget_light(dfd
, &fput_needed
);
1693 dentry
= file
->f_path
.dentry
;
1696 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1699 retval
= file_permission(file
, MAY_EXEC
);
1703 nd
->path
= file
->f_path
;
1704 path_get(&file
->f_path
);
1706 fput_light(file
, fput_needed
);
1708 nd
->inode
= nd
->path
.dentry
->d_inode
;
1712 fput_light(file
, fput_needed
);
1717 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1718 static int do_path_lookup(int dfd
, const char *name
,
1719 unsigned int flags
, struct nameidata
*nd
)
1724 * Path walking is largely split up into 2 different synchronisation
1725 * schemes, rcu-walk and ref-walk (explained in
1726 * Documentation/filesystems/path-lookup.txt). These share much of the
1727 * path walk code, but some things particularly setup, cleanup, and
1728 * following mounts are sufficiently divergent that functions are
1729 * duplicated. Typically there is a function foo(), and its RCU
1730 * analogue, foo_rcu().
1732 * -ECHILD is the error number of choice (just to avoid clashes) that
1733 * is returned if some aspect of an rcu-walk fails. Such an error must
1734 * be handled by restarting a traditional ref-walk (which will always
1735 * be able to complete).
1737 retval
= path_init_rcu(dfd
, name
, flags
, nd
);
1738 if (unlikely(retval
))
1740 retval
= path_walk_rcu(name
, nd
);
1741 path_finish_rcu(nd
);
1743 path_put(&nd
->root
);
1744 nd
->root
.mnt
= NULL
;
1747 if (unlikely(retval
== -ECHILD
|| retval
== -ESTALE
)) {
1748 /* slower, locked walk */
1749 if (retval
== -ESTALE
)
1750 flags
|= LOOKUP_REVAL
;
1751 retval
= path_init(dfd
, name
, flags
, nd
);
1752 if (unlikely(retval
))
1754 retval
= path_walk(name
, nd
);
1756 path_put(&nd
->root
);
1757 nd
->root
.mnt
= NULL
;
1761 if (likely(!retval
)) {
1762 if (unlikely(!audit_dummy_context())) {
1763 if (nd
->path
.dentry
&& nd
->inode
)
1764 audit_inode(name
, nd
->path
.dentry
);
1771 int path_lookup(const char *name
, unsigned int flags
,
1772 struct nameidata
*nd
)
1774 return do_path_lookup(AT_FDCWD
, name
, flags
, nd
);
1777 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1779 struct nameidata nd
;
1780 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1787 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1788 * @dentry: pointer to dentry of the base directory
1789 * @mnt: pointer to vfs mount of the base directory
1790 * @name: pointer to file name
1791 * @flags: lookup flags
1792 * @nd: pointer to nameidata
1794 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1795 const char *name
, unsigned int flags
,
1796 struct nameidata
*nd
)
1800 /* same as do_path_lookup */
1801 nd
->last_type
= LAST_ROOT
;
1805 nd
->path
.dentry
= dentry
;
1807 path_get(&nd
->path
);
1808 nd
->root
= nd
->path
;
1809 path_get(&nd
->root
);
1810 nd
->inode
= nd
->path
.dentry
->d_inode
;
1812 retval
= path_walk(name
, nd
);
1813 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1815 audit_inode(name
, nd
->path
.dentry
);
1817 path_put(&nd
->root
);
1818 nd
->root
.mnt
= NULL
;
1823 static struct dentry
*__lookup_hash(struct qstr
*name
,
1824 struct dentry
*base
, struct nameidata
*nd
)
1826 struct inode
*inode
= base
->d_inode
;
1827 struct dentry
*dentry
;
1830 err
= exec_permission(inode
, 0);
1832 return ERR_PTR(err
);
1835 * See if the low-level filesystem might want
1836 * to use its own hash..
1838 if (base
->d_flags
& DCACHE_OP_HASH
) {
1839 err
= base
->d_op
->d_hash(base
, inode
, name
);
1840 dentry
= ERR_PTR(err
);
1846 * Don't bother with __d_lookup: callers are for creat as
1847 * well as unlink, so a lot of the time it would cost
1850 dentry
= d_lookup(base
, name
);
1852 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1853 dentry
= do_revalidate(dentry
, nd
);
1856 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1862 * Restricted form of lookup. Doesn't follow links, single-component only,
1863 * needs parent already locked. Doesn't follow mounts.
1866 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1868 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1871 static int __lookup_one_len(const char *name
, struct qstr
*this,
1872 struct dentry
*base
, int len
)
1882 hash
= init_name_hash();
1884 c
= *(const unsigned char *)name
++;
1885 if (c
== '/' || c
== '\0')
1887 hash
= partial_name_hash(c
, hash
);
1889 this->hash
= end_name_hash(hash
);
1894 * lookup_one_len - filesystem helper to lookup single pathname component
1895 * @name: pathname component to lookup
1896 * @base: base directory to lookup from
1897 * @len: maximum length @len should be interpreted to
1899 * Note that this routine is purely a helper for filesystem usage and should
1900 * not be called by generic code. Also note that by using this function the
1901 * nameidata argument is passed to the filesystem methods and a filesystem
1902 * using this helper needs to be prepared for that.
1904 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1909 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1911 err
= __lookup_one_len(name
, &this, base
, len
);
1913 return ERR_PTR(err
);
1915 return __lookup_hash(&this, base
, NULL
);
1918 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1921 struct nameidata nd
;
1922 char *tmp
= getname(name
);
1923 int err
= PTR_ERR(tmp
);
1926 BUG_ON(flags
& LOOKUP_PARENT
);
1928 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1936 static int user_path_parent(int dfd
, const char __user
*path
,
1937 struct nameidata
*nd
, char **name
)
1939 char *s
= getname(path
);
1945 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1955 * It's inline, so penalty for filesystems that don't use sticky bit is
1958 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1960 uid_t fsuid
= current_fsuid();
1962 if (!(dir
->i_mode
& S_ISVTX
))
1964 if (inode
->i_uid
== fsuid
)
1966 if (dir
->i_uid
== fsuid
)
1968 return !capable(CAP_FOWNER
);
1972 * Check whether we can remove a link victim from directory dir, check
1973 * whether the type of victim is right.
1974 * 1. We can't do it if dir is read-only (done in permission())
1975 * 2. We should have write and exec permissions on dir
1976 * 3. We can't remove anything from append-only dir
1977 * 4. We can't do anything with immutable dir (done in permission())
1978 * 5. If the sticky bit on dir is set we should either
1979 * a. be owner of dir, or
1980 * b. be owner of victim, or
1981 * c. have CAP_FOWNER capability
1982 * 6. If the victim is append-only or immutable we can't do antyhing with
1983 * links pointing to it.
1984 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1985 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1986 * 9. We can't remove a root or mountpoint.
1987 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1988 * nfs_async_unlink().
1990 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1994 if (!victim
->d_inode
)
1997 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1998 audit_inode_child(victim
, dir
);
2000 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2005 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
2006 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
2009 if (!S_ISDIR(victim
->d_inode
->i_mode
))
2011 if (IS_ROOT(victim
))
2013 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
2015 if (IS_DEADDIR(dir
))
2017 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2022 /* Check whether we can create an object with dentry child in directory
2024 * 1. We can't do it if child already exists (open has special treatment for
2025 * this case, but since we are inlined it's OK)
2026 * 2. We can't do it if dir is read-only (done in permission())
2027 * 3. We should have write and exec permissions on dir
2028 * 4. We can't do it if dir is immutable (done in permission())
2030 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2034 if (IS_DEADDIR(dir
))
2036 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2040 * p1 and p2 should be directories on the same fs.
2042 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2047 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2051 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2053 p
= d_ancestor(p2
, p1
);
2055 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2056 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2060 p
= d_ancestor(p1
, p2
);
2062 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2063 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2067 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2068 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2072 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2074 mutex_unlock(&p1
->d_inode
->i_mutex
);
2076 mutex_unlock(&p2
->d_inode
->i_mutex
);
2077 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2081 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
2082 struct nameidata
*nd
)
2084 int error
= may_create(dir
, dentry
);
2089 if (!dir
->i_op
->create
)
2090 return -EACCES
; /* shouldn't it be ENOSYS? */
2093 error
= security_inode_create(dir
, dentry
, mode
);
2096 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
2098 fsnotify_create(dir
, dentry
);
2102 int may_open(struct path
*path
, int acc_mode
, int flag
)
2104 struct dentry
*dentry
= path
->dentry
;
2105 struct inode
*inode
= dentry
->d_inode
;
2111 switch (inode
->i_mode
& S_IFMT
) {
2115 if (acc_mode
& MAY_WRITE
)
2120 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2129 error
= inode_permission(inode
, acc_mode
);
2134 * An append-only file must be opened in append mode for writing.
2136 if (IS_APPEND(inode
)) {
2137 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2143 /* O_NOATIME can only be set by the owner or superuser */
2144 if (flag
& O_NOATIME
&& !is_owner_or_cap(inode
))
2148 * Ensure there are no outstanding leases on the file.
2150 return break_lease(inode
, flag
);
2153 static int handle_truncate(struct file
*filp
)
2155 struct path
*path
= &filp
->f_path
;
2156 struct inode
*inode
= path
->dentry
->d_inode
;
2157 int error
= get_write_access(inode
);
2161 * Refuse to truncate files with mandatory locks held on them.
2163 error
= locks_verify_locked(inode
);
2165 error
= security_path_truncate(path
);
2167 error
= do_truncate(path
->dentry
, 0,
2168 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2171 put_write_access(inode
);
2176 * Be careful about ever adding any more callers of this
2177 * function. Its flags must be in the namei format, not
2178 * what get passed to sys_open().
2180 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
2181 int open_flag
, int mode
)
2184 struct dentry
*dir
= nd
->path
.dentry
;
2186 if (!IS_POSIXACL(dir
->d_inode
))
2187 mode
&= ~current_umask();
2188 error
= security_path_mknod(&nd
->path
, path
->dentry
, mode
, 0);
2191 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
2193 mutex_unlock(&dir
->d_inode
->i_mutex
);
2194 dput(nd
->path
.dentry
);
2195 nd
->path
.dentry
= path
->dentry
;
2199 /* Don't check for write permission, don't truncate */
2200 return may_open(&nd
->path
, 0, open_flag
& ~O_TRUNC
);
2204 * Note that while the flag value (low two bits) for sys_open means:
2209 * it is changed into
2210 * 00 - no permissions needed
2211 * 01 - read-permission
2212 * 10 - write-permission
2214 * for the internal routines (ie open_namei()/follow_link() etc)
2215 * This is more logical, and also allows the 00 "no perm needed"
2216 * to be used for symlinks (where the permissions are checked
2220 static inline int open_to_namei_flags(int flag
)
2222 if ((flag
+1) & O_ACCMODE
)
2227 static int open_will_truncate(int flag
, struct inode
*inode
)
2230 * We'll never write to the fs underlying
2233 if (special_file(inode
->i_mode
))
2235 return (flag
& O_TRUNC
);
2238 static struct file
*finish_open(struct nameidata
*nd
,
2239 int open_flag
, int acc_mode
)
2245 will_truncate
= open_will_truncate(open_flag
, nd
->path
.dentry
->d_inode
);
2246 if (will_truncate
) {
2247 error
= mnt_want_write(nd
->path
.mnt
);
2251 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2254 mnt_drop_write(nd
->path
.mnt
);
2257 filp
= nameidata_to_filp(nd
);
2258 if (!IS_ERR(filp
)) {
2259 error
= ima_file_check(filp
, acc_mode
);
2262 filp
= ERR_PTR(error
);
2265 if (!IS_ERR(filp
)) {
2266 if (will_truncate
) {
2267 error
= handle_truncate(filp
);
2270 filp
= ERR_PTR(error
);
2275 * It is now safe to drop the mnt write
2276 * because the filp has had a write taken
2280 mnt_drop_write(nd
->path
.mnt
);
2281 path_put(&nd
->path
);
2285 path_put(&nd
->path
);
2286 return ERR_PTR(error
);
2290 * Handle O_CREAT case for do_filp_open
2292 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2293 int open_flag
, int acc_mode
,
2294 int mode
, const char *pathname
)
2296 struct dentry
*dir
= nd
->path
.dentry
;
2298 int error
= -EISDIR
;
2300 switch (nd
->last_type
) {
2303 dir
= nd
->path
.dentry
;
2305 if (need_reval_dot(dir
)) {
2306 int status
= d_revalidate(nd
->path
.dentry
, nd
);
2318 audit_inode(pathname
, dir
);
2322 /* trailing slashes? */
2323 if (nd
->last
.name
[nd
->last
.len
])
2326 mutex_lock(&dir
->d_inode
->i_mutex
);
2328 path
->dentry
= lookup_hash(nd
);
2329 path
->mnt
= nd
->path
.mnt
;
2331 error
= PTR_ERR(path
->dentry
);
2332 if (IS_ERR(path
->dentry
)) {
2333 mutex_unlock(&dir
->d_inode
->i_mutex
);
2337 if (IS_ERR(nd
->intent
.open
.file
)) {
2338 error
= PTR_ERR(nd
->intent
.open
.file
);
2339 goto exit_mutex_unlock
;
2342 /* Negative dentry, just create the file */
2343 if (!path
->dentry
->d_inode
) {
2345 * This write is needed to ensure that a
2346 * ro->rw transition does not occur between
2347 * the time when the file is created and when
2348 * a permanent write count is taken through
2349 * the 'struct file' in nameidata_to_filp().
2351 error
= mnt_want_write(nd
->path
.mnt
);
2353 goto exit_mutex_unlock
;
2354 error
= __open_namei_create(nd
, path
, open_flag
, mode
);
2356 mnt_drop_write(nd
->path
.mnt
);
2359 filp
= nameidata_to_filp(nd
);
2360 mnt_drop_write(nd
->path
.mnt
);
2361 path_put(&nd
->path
);
2362 if (!IS_ERR(filp
)) {
2363 error
= ima_file_check(filp
, acc_mode
);
2366 filp
= ERR_PTR(error
);
2373 * It already exists.
2375 mutex_unlock(&dir
->d_inode
->i_mutex
);
2376 audit_inode(pathname
, path
->dentry
);
2379 if (open_flag
& O_EXCL
)
2382 error
= follow_managed(path
, nd
->flags
);
2387 if (!path
->dentry
->d_inode
)
2390 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2393 path_to_nameidata(path
, nd
);
2394 nd
->inode
= path
->dentry
->d_inode
;
2396 if (S_ISDIR(nd
->inode
->i_mode
))
2399 filp
= finish_open(nd
, open_flag
, acc_mode
);
2403 mutex_unlock(&dir
->d_inode
->i_mutex
);
2405 path_put_conditional(path
, nd
);
2407 path_put(&nd
->path
);
2408 return ERR_PTR(error
);
2412 * Note that the low bits of the passed in "open_flag"
2413 * are not the same as in the local variable "flag". See
2414 * open_to_namei_flags() for more details.
2416 struct file
*do_filp_open(int dfd
, const char *pathname
,
2417 int open_flag
, int mode
, int acc_mode
)
2420 struct nameidata nd
;
2424 int flag
= open_to_namei_flags(open_flag
);
2427 if (!(open_flag
& O_CREAT
))
2430 /* Must never be set by userspace */
2431 open_flag
&= ~FMODE_NONOTIFY
;
2434 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2435 * check for O_DSYNC if the need any syncing at all we enforce it's
2436 * always set instead of having to deal with possibly weird behaviour
2437 * for malicious applications setting only __O_SYNC.
2439 if (open_flag
& __O_SYNC
)
2440 open_flag
|= O_DSYNC
;
2443 acc_mode
= MAY_OPEN
| ACC_MODE(open_flag
);
2445 /* O_TRUNC implies we need access checks for write permissions */
2446 if (open_flag
& O_TRUNC
)
2447 acc_mode
|= MAY_WRITE
;
2449 /* Allow the LSM permission hook to distinguish append
2450 access from general write access. */
2451 if (open_flag
& O_APPEND
)
2452 acc_mode
|= MAY_APPEND
;
2454 flags
= LOOKUP_OPEN
;
2455 if (open_flag
& O_CREAT
) {
2456 flags
|= LOOKUP_CREATE
;
2457 if (open_flag
& O_EXCL
)
2458 flags
|= LOOKUP_EXCL
;
2460 if (open_flag
& O_DIRECTORY
)
2461 flags
|= LOOKUP_DIRECTORY
;
2462 if (!(open_flag
& O_NOFOLLOW
))
2463 flags
|= LOOKUP_FOLLOW
;
2465 filp
= get_empty_filp();
2467 return ERR_PTR(-ENFILE
);
2469 filp
->f_flags
= open_flag
;
2470 nd
.intent
.open
.file
= filp
;
2471 nd
.intent
.open
.flags
= flag
;
2472 nd
.intent
.open
.create_mode
= mode
;
2474 if (open_flag
& O_CREAT
)
2477 /* !O_CREAT, simple open */
2478 error
= do_path_lookup(dfd
, pathname
, flags
, &nd
);
2479 if (unlikely(error
))
2482 if (!(nd
.flags
& LOOKUP_FOLLOW
)) {
2483 if (nd
.inode
->i_op
->follow_link
)
2487 if (nd
.flags
& LOOKUP_DIRECTORY
) {
2488 if (!nd
.inode
->i_op
->lookup
)
2491 audit_inode(pathname
, nd
.path
.dentry
);
2492 filp
= finish_open(&nd
, open_flag
, acc_mode
);
2493 release_open_intent(&nd
);
2497 /* OK, have to create the file. Find the parent. */
2498 error
= path_init_rcu(dfd
, pathname
,
2499 LOOKUP_PARENT
| (flags
& LOOKUP_REVAL
), &nd
);
2502 error
= path_walk_rcu(pathname
, &nd
);
2503 path_finish_rcu(&nd
);
2504 if (unlikely(error
== -ECHILD
|| error
== -ESTALE
)) {
2505 /* slower, locked walk */
2506 if (error
== -ESTALE
) {
2508 flags
|= LOOKUP_REVAL
;
2510 error
= path_init(dfd
, pathname
,
2511 LOOKUP_PARENT
| (flags
& LOOKUP_REVAL
), &nd
);
2515 error
= path_walk_simple(pathname
, &nd
);
2517 if (unlikely(error
))
2519 if (unlikely(!audit_dummy_context()))
2520 audit_inode(pathname
, nd
.path
.dentry
);
2523 * We have the parent and last component.
2526 filp
= do_last(&nd
, &path
, open_flag
, acc_mode
, mode
, pathname
);
2527 while (unlikely(!filp
)) { /* trailing symlink */
2528 struct path link
= path
;
2529 struct inode
*linki
= link
.dentry
->d_inode
;
2532 if (!(nd
.flags
& LOOKUP_FOLLOW
))
2537 * This is subtle. Instead of calling do_follow_link() we do
2538 * the thing by hands. The reason is that this way we have zero
2539 * link_count and path_walk() (called from ->follow_link)
2540 * honoring LOOKUP_PARENT. After that we have the parent and
2541 * last component, i.e. we are in the same situation as after
2542 * the first path_walk(). Well, almost - if the last component
2543 * is normal we get its copy stored in nd->last.name and we will
2544 * have to putname() it when we are done. Procfs-like symlinks
2545 * just set LAST_BIND.
2547 nd
.flags
|= LOOKUP_PARENT
;
2548 error
= security_inode_follow_link(link
.dentry
, &nd
);
2551 error
= __do_follow_link(&link
, &nd
, &cookie
);
2552 if (unlikely(error
)) {
2553 if (!IS_ERR(cookie
) && linki
->i_op
->put_link
)
2554 linki
->i_op
->put_link(link
.dentry
, &nd
, cookie
);
2555 /* nd.path had been dropped */
2559 nd
.flags
&= ~LOOKUP_PARENT
;
2560 filp
= do_last(&nd
, &path
, open_flag
, acc_mode
, mode
, pathname
);
2561 if (linki
->i_op
->put_link
)
2562 linki
->i_op
->put_link(link
.dentry
, &nd
, cookie
);
2568 if (filp
== ERR_PTR(-ESTALE
) && !(flags
& LOOKUP_REVAL
))
2570 release_open_intent(&nd
);
2574 path_put_conditional(&path
, &nd
);
2578 filp
= ERR_PTR(error
);
2583 * filp_open - open file and return file pointer
2585 * @filename: path to open
2586 * @flags: open flags as per the open(2) second argument
2587 * @mode: mode for the new file if O_CREAT is set, else ignored
2589 * This is the helper to open a file from kernelspace if you really
2590 * have to. But in generally you should not do this, so please move
2591 * along, nothing to see here..
2593 struct file
*filp_open(const char *filename
, int flags
, int mode
)
2595 return do_filp_open(AT_FDCWD
, filename
, flags
, mode
, 0);
2597 EXPORT_SYMBOL(filp_open
);
2600 * lookup_create - lookup a dentry, creating it if it doesn't exist
2601 * @nd: nameidata info
2602 * @is_dir: directory flag
2604 * Simple function to lookup and return a dentry and create it
2605 * if it doesn't exist. Is SMP-safe.
2607 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2609 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2611 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2613 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2615 * Yucky last component or no last component at all?
2616 * (foo/., foo/.., /////)
2618 if (nd
->last_type
!= LAST_NORM
)
2620 nd
->flags
&= ~LOOKUP_PARENT
;
2621 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2622 nd
->intent
.open
.flags
= O_EXCL
;
2625 * Do the final lookup.
2627 dentry
= lookup_hash(nd
);
2631 if (dentry
->d_inode
)
2634 * Special case - lookup gave negative, but... we had foo/bar/
2635 * From the vfs_mknod() POV we just have a negative dentry -
2636 * all is fine. Let's be bastards - you had / on the end, you've
2637 * been asking for (non-existent) directory. -ENOENT for you.
2639 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2641 dentry
= ERR_PTR(-ENOENT
);
2646 dentry
= ERR_PTR(-EEXIST
);
2650 EXPORT_SYMBOL_GPL(lookup_create
);
2652 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2654 int error
= may_create(dir
, dentry
);
2659 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
2662 if (!dir
->i_op
->mknod
)
2665 error
= devcgroup_inode_mknod(mode
, dev
);
2669 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2673 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2675 fsnotify_create(dir
, dentry
);
2679 static int may_mknod(mode_t mode
)
2681 switch (mode
& S_IFMT
) {
2687 case 0: /* zero mode translates to S_IFREG */
2696 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2701 struct dentry
*dentry
;
2702 struct nameidata nd
;
2707 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2711 dentry
= lookup_create(&nd
, 0);
2712 if (IS_ERR(dentry
)) {
2713 error
= PTR_ERR(dentry
);
2716 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2717 mode
&= ~current_umask();
2718 error
= may_mknod(mode
);
2721 error
= mnt_want_write(nd
.path
.mnt
);
2724 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2726 goto out_drop_write
;
2727 switch (mode
& S_IFMT
) {
2728 case 0: case S_IFREG
:
2729 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2731 case S_IFCHR
: case S_IFBLK
:
2732 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2733 new_decode_dev(dev
));
2735 case S_IFIFO
: case S_IFSOCK
:
2736 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2740 mnt_drop_write(nd
.path
.mnt
);
2744 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2751 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2753 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2756 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2758 int error
= may_create(dir
, dentry
);
2763 if (!dir
->i_op
->mkdir
)
2766 mode
&= (S_IRWXUGO
|S_ISVTX
);
2767 error
= security_inode_mkdir(dir
, dentry
, mode
);
2771 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2773 fsnotify_mkdir(dir
, dentry
);
2777 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2781 struct dentry
*dentry
;
2782 struct nameidata nd
;
2784 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2788 dentry
= lookup_create(&nd
, 1);
2789 error
= PTR_ERR(dentry
);
2793 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2794 mode
&= ~current_umask();
2795 error
= mnt_want_write(nd
.path
.mnt
);
2798 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2800 goto out_drop_write
;
2801 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2803 mnt_drop_write(nd
.path
.mnt
);
2807 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2814 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2816 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2820 * We try to drop the dentry early: we should have
2821 * a usage count of 2 if we're the only user of this
2822 * dentry, and if that is true (possibly after pruning
2823 * the dcache), then we drop the dentry now.
2825 * A low-level filesystem can, if it choses, legally
2828 * if (!d_unhashed(dentry))
2831 * if it cannot handle the case of removing a directory
2832 * that is still in use by something else..
2834 void dentry_unhash(struct dentry
*dentry
)
2837 shrink_dcache_parent(dentry
);
2838 spin_lock(&dentry
->d_lock
);
2839 if (dentry
->d_count
== 2)
2841 spin_unlock(&dentry
->d_lock
);
2844 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2846 int error
= may_delete(dir
, dentry
, 1);
2851 if (!dir
->i_op
->rmdir
)
2854 mutex_lock(&dentry
->d_inode
->i_mutex
);
2855 dentry_unhash(dentry
);
2856 if (d_mountpoint(dentry
))
2859 error
= security_inode_rmdir(dir
, dentry
);
2861 error
= dir
->i_op
->rmdir(dir
, dentry
);
2863 dentry
->d_inode
->i_flags
|= S_DEAD
;
2868 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2877 static long do_rmdir(int dfd
, const char __user
*pathname
)
2881 struct dentry
*dentry
;
2882 struct nameidata nd
;
2884 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2888 switch(nd
.last_type
) {
2900 nd
.flags
&= ~LOOKUP_PARENT
;
2902 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2903 dentry
= lookup_hash(&nd
);
2904 error
= PTR_ERR(dentry
);
2907 error
= mnt_want_write(nd
.path
.mnt
);
2910 error
= security_path_rmdir(&nd
.path
, dentry
);
2913 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2915 mnt_drop_write(nd
.path
.mnt
);
2919 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2926 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2928 return do_rmdir(AT_FDCWD
, pathname
);
2931 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2933 int error
= may_delete(dir
, dentry
, 0);
2938 if (!dir
->i_op
->unlink
)
2941 mutex_lock(&dentry
->d_inode
->i_mutex
);
2942 if (d_mountpoint(dentry
))
2945 error
= security_inode_unlink(dir
, dentry
);
2947 error
= dir
->i_op
->unlink(dir
, dentry
);
2952 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2954 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2955 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2956 fsnotify_link_count(dentry
->d_inode
);
2964 * Make sure that the actual truncation of the file will occur outside its
2965 * directory's i_mutex. Truncate can take a long time if there is a lot of
2966 * writeout happening, and we don't want to prevent access to the directory
2967 * while waiting on the I/O.
2969 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2973 struct dentry
*dentry
;
2974 struct nameidata nd
;
2975 struct inode
*inode
= NULL
;
2977 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2982 if (nd
.last_type
!= LAST_NORM
)
2985 nd
.flags
&= ~LOOKUP_PARENT
;
2987 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2988 dentry
= lookup_hash(&nd
);
2989 error
= PTR_ERR(dentry
);
2990 if (!IS_ERR(dentry
)) {
2991 /* Why not before? Because we want correct error value */
2992 if (nd
.last
.name
[nd
.last
.len
])
2994 inode
= dentry
->d_inode
;
2997 error
= mnt_want_write(nd
.path
.mnt
);
3000 error
= security_path_unlink(&nd
.path
, dentry
);
3003 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
3005 mnt_drop_write(nd
.path
.mnt
);
3009 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3011 iput(inode
); /* truncate the inode here */
3018 error
= !dentry
->d_inode
? -ENOENT
:
3019 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
3023 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3025 if ((flag
& ~AT_REMOVEDIR
) != 0)
3028 if (flag
& AT_REMOVEDIR
)
3029 return do_rmdir(dfd
, pathname
);
3031 return do_unlinkat(dfd
, pathname
);
3034 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3036 return do_unlinkat(AT_FDCWD
, pathname
);
3039 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3041 int error
= may_create(dir
, dentry
);
3046 if (!dir
->i_op
->symlink
)
3049 error
= security_inode_symlink(dir
, dentry
, oldname
);
3053 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3055 fsnotify_create(dir
, dentry
);
3059 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3060 int, newdfd
, const char __user
*, newname
)
3065 struct dentry
*dentry
;
3066 struct nameidata nd
;
3068 from
= getname(oldname
);
3070 return PTR_ERR(from
);
3072 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
3076 dentry
= lookup_create(&nd
, 0);
3077 error
= PTR_ERR(dentry
);
3081 error
= mnt_want_write(nd
.path
.mnt
);
3084 error
= security_path_symlink(&nd
.path
, dentry
, from
);
3086 goto out_drop_write
;
3087 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
3089 mnt_drop_write(nd
.path
.mnt
);
3093 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3101 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3103 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3106 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3108 struct inode
*inode
= old_dentry
->d_inode
;
3114 error
= may_create(dir
, new_dentry
);
3118 if (dir
->i_sb
!= inode
->i_sb
)
3122 * A link to an append-only or immutable file cannot be created.
3124 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3126 if (!dir
->i_op
->link
)
3128 if (S_ISDIR(inode
->i_mode
))
3131 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3135 mutex_lock(&inode
->i_mutex
);
3136 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3137 mutex_unlock(&inode
->i_mutex
);
3139 fsnotify_link(dir
, inode
, new_dentry
);
3144 * Hardlinks are often used in delicate situations. We avoid
3145 * security-related surprises by not following symlinks on the
3148 * We don't follow them on the oldname either to be compatible
3149 * with linux 2.0, and to avoid hard-linking to directories
3150 * and other special files. --ADM
3152 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3153 int, newdfd
, const char __user
*, newname
, int, flags
)
3155 struct dentry
*new_dentry
;
3156 struct nameidata nd
;
3157 struct path old_path
;
3161 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
3164 error
= user_path_at(olddfd
, oldname
,
3165 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
3170 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
3174 if (old_path
.mnt
!= nd
.path
.mnt
)
3176 new_dentry
= lookup_create(&nd
, 0);
3177 error
= PTR_ERR(new_dentry
);
3178 if (IS_ERR(new_dentry
))
3180 error
= mnt_want_write(nd
.path
.mnt
);
3183 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
3185 goto out_drop_write
;
3186 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
3188 mnt_drop_write(nd
.path
.mnt
);
3192 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3197 path_put(&old_path
);
3202 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3204 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3208 * The worst of all namespace operations - renaming directory. "Perverted"
3209 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3211 * a) we can get into loop creation. Check is done in is_subdir().
3212 * b) race potential - two innocent renames can create a loop together.
3213 * That's where 4.4 screws up. Current fix: serialization on
3214 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3216 * c) we have to lock _three_ objects - parents and victim (if it exists).
3217 * And that - after we got ->i_mutex on parents (until then we don't know
3218 * whether the target exists). Solution: try to be smart with locking
3219 * order for inodes. We rely on the fact that tree topology may change
3220 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3221 * move will be locked. Thus we can rank directories by the tree
3222 * (ancestors first) and rank all non-directories after them.
3223 * That works since everybody except rename does "lock parent, lookup,
3224 * lock child" and rename is under ->s_vfs_rename_mutex.
3225 * HOWEVER, it relies on the assumption that any object with ->lookup()
3226 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3227 * we'd better make sure that there's no link(2) for them.
3228 * d) some filesystems don't support opened-but-unlinked directories,
3229 * either because of layout or because they are not ready to deal with
3230 * all cases correctly. The latter will be fixed (taking this sort of
3231 * stuff into VFS), but the former is not going away. Solution: the same
3232 * trick as in rmdir().
3233 * e) conversion from fhandle to dentry may come in the wrong moment - when
3234 * we are removing the target. Solution: we will have to grab ->i_mutex
3235 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3236 * ->i_mutex on parents, which works but leads to some truly excessive
3239 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3240 struct inode
*new_dir
, struct dentry
*new_dentry
)
3243 struct inode
*target
;
3246 * If we are going to change the parent - check write permissions,
3247 * we'll need to flip '..'.
3249 if (new_dir
!= old_dir
) {
3250 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3255 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3259 target
= new_dentry
->d_inode
;
3261 mutex_lock(&target
->i_mutex
);
3262 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3266 dentry_unhash(new_dentry
);
3267 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3271 target
->i_flags
|= S_DEAD
;
3272 dont_mount(new_dentry
);
3274 mutex_unlock(&target
->i_mutex
);
3275 if (d_unhashed(new_dentry
))
3276 d_rehash(new_dentry
);
3280 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3281 d_move(old_dentry
,new_dentry
);
3285 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3286 struct inode
*new_dir
, struct dentry
*new_dentry
)
3288 struct inode
*target
;
3291 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3296 target
= new_dentry
->d_inode
;
3298 mutex_lock(&target
->i_mutex
);
3299 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3302 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3305 dont_mount(new_dentry
);
3306 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3307 d_move(old_dentry
, new_dentry
);
3310 mutex_unlock(&target
->i_mutex
);
3315 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3316 struct inode
*new_dir
, struct dentry
*new_dentry
)
3319 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3320 const unsigned char *old_name
;
3322 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3325 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3329 if (!new_dentry
->d_inode
)
3330 error
= may_create(new_dir
, new_dentry
);
3332 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3336 if (!old_dir
->i_op
->rename
)
3339 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3342 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3344 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3346 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3347 new_dentry
->d_inode
, old_dentry
);
3348 fsnotify_oldname_free(old_name
);
3353 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3354 int, newdfd
, const char __user
*, newname
)
3356 struct dentry
*old_dir
, *new_dir
;
3357 struct dentry
*old_dentry
, *new_dentry
;
3358 struct dentry
*trap
;
3359 struct nameidata oldnd
, newnd
;
3364 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3368 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3373 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3376 old_dir
= oldnd
.path
.dentry
;
3378 if (oldnd
.last_type
!= LAST_NORM
)
3381 new_dir
= newnd
.path
.dentry
;
3382 if (newnd
.last_type
!= LAST_NORM
)
3385 oldnd
.flags
&= ~LOOKUP_PARENT
;
3386 newnd
.flags
&= ~LOOKUP_PARENT
;
3387 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3389 trap
= lock_rename(new_dir
, old_dir
);
3391 old_dentry
= lookup_hash(&oldnd
);
3392 error
= PTR_ERR(old_dentry
);
3393 if (IS_ERR(old_dentry
))
3395 /* source must exist */
3397 if (!old_dentry
->d_inode
)
3399 /* unless the source is a directory trailing slashes give -ENOTDIR */
3400 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3402 if (oldnd
.last
.name
[oldnd
.last
.len
])
3404 if (newnd
.last
.name
[newnd
.last
.len
])
3407 /* source should not be ancestor of target */
3409 if (old_dentry
== trap
)
3411 new_dentry
= lookup_hash(&newnd
);
3412 error
= PTR_ERR(new_dentry
);
3413 if (IS_ERR(new_dentry
))
3415 /* target should not be an ancestor of source */
3417 if (new_dentry
== trap
)
3420 error
= mnt_want_write(oldnd
.path
.mnt
);
3423 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3424 &newnd
.path
, new_dentry
);
3427 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3428 new_dir
->d_inode
, new_dentry
);
3430 mnt_drop_write(oldnd
.path
.mnt
);
3436 unlock_rename(new_dir
, old_dir
);
3438 path_put(&newnd
.path
);
3441 path_put(&oldnd
.path
);
3447 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3449 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3452 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3456 len
= PTR_ERR(link
);
3461 if (len
> (unsigned) buflen
)
3463 if (copy_to_user(buffer
, link
, len
))
3470 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3471 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3472 * using) it for any given inode is up to filesystem.
3474 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3476 struct nameidata nd
;
3481 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3483 return PTR_ERR(cookie
);
3485 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3486 if (dentry
->d_inode
->i_op
->put_link
)
3487 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3491 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3493 return __vfs_follow_link(nd
, link
);
3496 /* get the link contents into pagecache */
3497 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3501 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3502 page
= read_mapping_page(mapping
, 0, NULL
);
3507 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3511 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3513 struct page
*page
= NULL
;
3514 char *s
= page_getlink(dentry
, &page
);
3515 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3518 page_cache_release(page
);
3523 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3525 struct page
*page
= NULL
;
3526 nd_set_link(nd
, page_getlink(dentry
, &page
));
3530 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3532 struct page
*page
= cookie
;
3536 page_cache_release(page
);
3541 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3543 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3545 struct address_space
*mapping
= inode
->i_mapping
;
3550 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3552 flags
|= AOP_FLAG_NOFS
;
3555 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3556 flags
, &page
, &fsdata
);
3560 kaddr
= kmap_atomic(page
, KM_USER0
);
3561 memcpy(kaddr
, symname
, len
-1);
3562 kunmap_atomic(kaddr
, KM_USER0
);
3564 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3571 mark_inode_dirty(inode
);
3577 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3579 return __page_symlink(inode
, symname
, len
,
3580 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3583 const struct inode_operations page_symlink_inode_operations
= {
3584 .readlink
= generic_readlink
,
3585 .follow_link
= page_follow_link_light
,
3586 .put_link
= page_put_link
,
3589 EXPORT_SYMBOL(user_path_at
);
3590 EXPORT_SYMBOL(follow_down_one
);
3591 EXPORT_SYMBOL(follow_down
);
3592 EXPORT_SYMBOL(follow_up
);
3593 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3594 EXPORT_SYMBOL(getname
);
3595 EXPORT_SYMBOL(lock_rename
);
3596 EXPORT_SYMBOL(lookup_one_len
);
3597 EXPORT_SYMBOL(page_follow_link_light
);
3598 EXPORT_SYMBOL(page_put_link
);
3599 EXPORT_SYMBOL(page_readlink
);
3600 EXPORT_SYMBOL(__page_symlink
);
3601 EXPORT_SYMBOL(page_symlink
);
3602 EXPORT_SYMBOL(page_symlink_inode_operations
);
3603 EXPORT_SYMBOL(path_lookup
);
3604 EXPORT_SYMBOL(kern_path
);
3605 EXPORT_SYMBOL(vfs_path_lookup
);
3606 EXPORT_SYMBOL(inode_permission
);
3607 EXPORT_SYMBOL(file_permission
);
3608 EXPORT_SYMBOL(unlock_rename
);
3609 EXPORT_SYMBOL(vfs_create
);
3610 EXPORT_SYMBOL(vfs_follow_link
);
3611 EXPORT_SYMBOL(vfs_link
);
3612 EXPORT_SYMBOL(vfs_mkdir
);
3613 EXPORT_SYMBOL(vfs_mknod
);
3614 EXPORT_SYMBOL(generic_permission
);
3615 EXPORT_SYMBOL(vfs_readlink
);
3616 EXPORT_SYMBOL(vfs_rename
);
3617 EXPORT_SYMBOL(vfs_rmdir
);
3618 EXPORT_SYMBOL(vfs_symlink
);
3619 EXPORT_SYMBOL(vfs_unlink
);
3620 EXPORT_SYMBOL(dentry_unhash
);
3621 EXPORT_SYMBOL(generic_readlink
);