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_get_long - get a long reference to a path
372 * @path: path to get the reference to
374 * Given a path increment the reference count to the dentry and the vfsmount.
376 void path_get_long(struct path
*path
)
378 mntget_long(path
->mnt
);
383 * path_put - put a reference to a path
384 * @path: path to put the reference to
386 * Given a path decrement the reference count to the dentry and the vfsmount.
388 void path_put(struct path
*path
)
393 EXPORT_SYMBOL(path_put
);
396 * path_put_long - put a long reference to a path
397 * @path: path to put the reference to
399 * Given a path decrement the reference count to the dentry and the vfsmount.
401 void path_put_long(struct path
*path
)
404 mntput_long(path
->mnt
);
408 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
409 * @nd: nameidata pathwalk data to drop
410 * Returns: 0 on success, -ECHILD on failure
412 * Path walking has 2 modes, rcu-walk and ref-walk (see
413 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
414 * to drop out of rcu-walk mode and take normal reference counts on dentries
415 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
416 * refcounts at the last known good point before rcu-walk got stuck, so
417 * ref-walk may continue from there. If this is not successful (eg. a seqcount
418 * has changed), then failure is returned and path walk restarts from the
419 * beginning in ref-walk mode.
421 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
422 * ref-walk. Must be called from rcu-walk context.
424 static int nameidata_drop_rcu(struct nameidata
*nd
)
426 struct fs_struct
*fs
= current
->fs
;
427 struct dentry
*dentry
= nd
->path
.dentry
;
429 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
431 spin_lock(&fs
->lock
);
432 if (nd
->root
.mnt
!= fs
->root
.mnt
||
433 nd
->root
.dentry
!= fs
->root
.dentry
)
436 spin_lock(&dentry
->d_lock
);
437 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
439 BUG_ON(nd
->inode
!= dentry
->d_inode
);
440 spin_unlock(&dentry
->d_lock
);
443 spin_unlock(&fs
->lock
);
445 mntget(nd
->path
.mnt
);
448 br_read_unlock(vfsmount_lock
);
449 nd
->flags
&= ~LOOKUP_RCU
;
452 spin_unlock(&dentry
->d_lock
);
455 spin_unlock(&fs
->lock
);
459 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
460 static inline int nameidata_drop_rcu_maybe(struct nameidata
*nd
)
462 if (nd
->flags
& LOOKUP_RCU
)
463 return nameidata_drop_rcu(nd
);
468 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
469 * @nd: nameidata pathwalk data to drop
470 * @dentry: dentry to drop
471 * Returns: 0 on success, -ECHILD on failure
473 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
474 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
475 * @nd. Must be called from rcu-walk context.
477 static int nameidata_dentry_drop_rcu(struct nameidata
*nd
, struct dentry
*dentry
)
479 struct fs_struct
*fs
= current
->fs
;
480 struct dentry
*parent
= nd
->path
.dentry
;
483 * It can be possible to revalidate the dentry that we started
484 * the path walk with. force_reval_path may also revalidate the
485 * dentry already committed to the nameidata.
487 if (unlikely(parent
== dentry
))
488 return nameidata_drop_rcu(nd
);
490 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
492 spin_lock(&fs
->lock
);
493 if (nd
->root
.mnt
!= fs
->root
.mnt
||
494 nd
->root
.dentry
!= fs
->root
.dentry
)
497 spin_lock(&parent
->d_lock
);
498 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
499 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
502 * If the sequence check on the child dentry passed, then the child has
503 * not been removed from its parent. This means the parent dentry must
504 * be valid and able to take a reference at this point.
506 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
507 BUG_ON(!parent
->d_count
);
509 spin_unlock(&dentry
->d_lock
);
510 spin_unlock(&parent
->d_lock
);
513 spin_unlock(&fs
->lock
);
515 mntget(nd
->path
.mnt
);
518 br_read_unlock(vfsmount_lock
);
519 nd
->flags
&= ~LOOKUP_RCU
;
522 spin_unlock(&dentry
->d_lock
);
523 spin_unlock(&parent
->d_lock
);
526 spin_unlock(&fs
->lock
);
530 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
531 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata
*nd
, struct dentry
*dentry
)
533 if (nd
->flags
& LOOKUP_RCU
)
534 return nameidata_dentry_drop_rcu(nd
, dentry
);
539 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
540 * @nd: nameidata pathwalk data to drop
541 * Returns: 0 on success, -ECHILD on failure
543 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
544 * nd->path should be the final element of the lookup, so nd->root is discarded.
545 * Must be called from rcu-walk context.
547 static int nameidata_drop_rcu_last(struct nameidata
*nd
)
549 struct dentry
*dentry
= nd
->path
.dentry
;
551 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
552 nd
->flags
&= ~LOOKUP_RCU
;
554 spin_lock(&dentry
->d_lock
);
555 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
557 BUG_ON(nd
->inode
!= dentry
->d_inode
);
558 spin_unlock(&dentry
->d_lock
);
560 mntget(nd
->path
.mnt
);
563 br_read_unlock(vfsmount_lock
);
568 spin_unlock(&dentry
->d_lock
);
570 br_read_unlock(vfsmount_lock
);
574 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
575 static inline int nameidata_drop_rcu_last_maybe(struct nameidata
*nd
)
577 if (likely(nd
->flags
& LOOKUP_RCU
))
578 return nameidata_drop_rcu_last(nd
);
583 * release_open_intent - free up open intent resources
584 * @nd: pointer to nameidata
586 void release_open_intent(struct nameidata
*nd
)
588 if (nd
->intent
.open
.file
->f_path
.dentry
== NULL
)
589 put_filp(nd
->intent
.open
.file
);
591 fput(nd
->intent
.open
.file
);
595 * Call d_revalidate and handle filesystems that request rcu-walk
596 * to be dropped. This may be called and return in rcu-walk mode,
597 * regardless of success or error. If -ECHILD is returned, the caller
598 * must return -ECHILD back up the path walk stack so path walk may
599 * be restarted in ref-walk mode.
601 static int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
605 status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
606 if (status
== -ECHILD
) {
607 if (nameidata_dentry_drop_rcu(nd
, dentry
))
609 status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
615 static inline struct dentry
*
616 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
620 status
= d_revalidate(dentry
, nd
);
621 if (unlikely(status
<= 0)) {
623 * The dentry failed validation.
624 * If d_revalidate returned 0 attempt to invalidate
625 * the dentry otherwise d_revalidate is asking us
626 * to return a fail status.
629 /* If we're in rcu-walk, we don't have a ref */
630 if (!(nd
->flags
& LOOKUP_RCU
))
632 dentry
= ERR_PTR(status
);
635 /* Don't d_invalidate in rcu-walk mode */
636 if (nameidata_dentry_drop_rcu_maybe(nd
, dentry
))
637 return ERR_PTR(-ECHILD
);
638 if (!d_invalidate(dentry
)) {
647 static inline int need_reval_dot(struct dentry
*dentry
)
649 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
652 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
659 * force_reval_path - force revalidation of a dentry
661 * In some situations the path walking code will trust dentries without
662 * revalidating them. This causes problems for filesystems that depend on
663 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
664 * (which indicates that it's possible for the dentry to go stale), force
665 * a d_revalidate call before proceeding.
667 * Returns 0 if the revalidation was successful. If the revalidation fails,
668 * either return the error returned by d_revalidate or -ESTALE if the
669 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
670 * invalidate the dentry. It's up to the caller to handle putting references
671 * to the path if necessary.
674 force_reval_path(struct path
*path
, struct nameidata
*nd
)
677 struct dentry
*dentry
= path
->dentry
;
680 * only check on filesystems where it's possible for the dentry to
683 if (!need_reval_dot(dentry
))
686 status
= d_revalidate(dentry
, nd
);
691 /* Don't d_invalidate in rcu-walk mode */
692 if (nameidata_drop_rcu(nd
))
694 d_invalidate(dentry
);
701 * Short-cut version of permission(), for calling on directories
702 * during pathname resolution. Combines parts of permission()
703 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
705 * If appropriate, check DAC only. If not appropriate, or
706 * short-cut DAC fails, then call ->permission() to do more
707 * complete permission check.
709 static inline int exec_permission(struct inode
*inode
, unsigned int flags
)
713 if (inode
->i_op
->permission
) {
714 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
, flags
);
716 ret
= acl_permission_check(inode
, MAY_EXEC
, flags
,
717 inode
->i_op
->check_acl
);
724 if (capable(CAP_DAC_OVERRIDE
) || capable(CAP_DAC_READ_SEARCH
))
729 return security_inode_exec_permission(inode
, flags
);
732 static __always_inline
void set_root(struct nameidata
*nd
)
735 get_fs_root(current
->fs
, &nd
->root
);
738 static int link_path_walk(const char *, struct nameidata
*);
740 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
743 struct fs_struct
*fs
= current
->fs
;
747 seq
= read_seqcount_begin(&fs
->seq
);
749 } while (read_seqcount_retry(&fs
->seq
, seq
));
753 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
766 nd
->inode
= nd
->path
.dentry
->d_inode
;
768 ret
= link_path_walk(link
, nd
);
772 return PTR_ERR(link
);
775 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
778 if (path
->mnt
!= nd
->path
.mnt
)
782 static inline void path_to_nameidata(const struct path
*path
,
783 struct nameidata
*nd
)
785 if (!(nd
->flags
& LOOKUP_RCU
)) {
786 dput(nd
->path
.dentry
);
787 if (nd
->path
.mnt
!= path
->mnt
)
788 mntput(nd
->path
.mnt
);
790 nd
->path
.mnt
= path
->mnt
;
791 nd
->path
.dentry
= path
->dentry
;
794 static __always_inline
int
795 __do_follow_link(const struct path
*link
, struct nameidata
*nd
, void **p
)
798 struct dentry
*dentry
= link
->dentry
;
800 touch_atime(link
->mnt
, dentry
);
801 nd_set_link(nd
, NULL
);
803 if (link
->mnt
== nd
->path
.mnt
)
806 nd
->last_type
= LAST_BIND
;
807 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
810 char *s
= nd_get_link(nd
);
813 error
= __vfs_follow_link(nd
, s
);
814 else if (nd
->last_type
== LAST_BIND
) {
815 error
= force_reval_path(&nd
->path
, nd
);
824 * This limits recursive symlink follows to 8, while
825 * limiting consecutive symlinks to 40.
827 * Without that kind of total limit, nasty chains of consecutive
828 * symlinks can cause almost arbitrarily long lookups.
830 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
834 if (current
->link_count
>= MAX_NESTED_LINKS
)
836 if (current
->total_link_count
>= 40)
838 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
840 err
= security_inode_follow_link(path
->dentry
, nd
);
843 current
->link_count
++;
844 current
->total_link_count
++;
846 err
= __do_follow_link(path
, nd
, &cookie
);
847 if (!IS_ERR(cookie
) && path
->dentry
->d_inode
->i_op
->put_link
)
848 path
->dentry
->d_inode
->i_op
->put_link(path
->dentry
, nd
, cookie
);
850 current
->link_count
--;
854 path_put_conditional(path
, nd
);
859 static int follow_up_rcu(struct path
*path
)
861 struct vfsmount
*parent
;
862 struct dentry
*mountpoint
;
864 parent
= path
->mnt
->mnt_parent
;
865 if (parent
== path
->mnt
)
867 mountpoint
= path
->mnt
->mnt_mountpoint
;
868 path
->dentry
= mountpoint
;
873 int follow_up(struct path
*path
)
875 struct vfsmount
*parent
;
876 struct dentry
*mountpoint
;
878 br_read_lock(vfsmount_lock
);
879 parent
= path
->mnt
->mnt_parent
;
880 if (parent
== path
->mnt
) {
881 br_read_unlock(vfsmount_lock
);
885 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
886 br_read_unlock(vfsmount_lock
);
888 path
->dentry
= mountpoint
;
895 * Perform an automount
896 * - return -EISDIR to tell follow_managed() to stop and return the path we
899 static int follow_automount(struct path
*path
, unsigned flags
,
902 struct vfsmount
*mnt
;
905 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
908 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
909 * and this is the terminal part of the path.
911 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
912 return -EISDIR
; /* we actually want to stop here */
914 /* We want to mount if someone is trying to open/create a file of any
915 * type under the mountpoint, wants to traverse through the mountpoint
916 * or wants to open the mounted directory.
918 * We don't want to mount if someone's just doing a stat and they've
919 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
920 * appended a '/' to the name.
922 if (!(flags
& LOOKUP_FOLLOW
) &&
923 !(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
924 LOOKUP_OPEN
| LOOKUP_CREATE
)))
927 current
->total_link_count
++;
928 if (current
->total_link_count
>= 40)
931 mnt
= path
->dentry
->d_op
->d_automount(path
);
934 * The filesystem is allowed to return -EISDIR here to indicate
935 * it doesn't want to automount. For instance, autofs would do
936 * this so that its userspace daemon can mount on this dentry.
938 * However, we can only permit this if it's a terminal point in
939 * the path being looked up; if it wasn't then the remainder of
940 * the path is inaccessible and we should say so.
942 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_CONTINUE
))
947 if (!mnt
) /* mount collision */
950 /* The new mount record should have at least 2 refs to prevent it being
951 * expired before we get a chance to add it
953 BUG_ON(mnt_get_count(mnt
) < 2);
955 if (mnt
->mnt_sb
== path
->mnt
->mnt_sb
&&
956 mnt
->mnt_root
== path
->dentry
) {
957 mnt_clear_expiry(mnt
);
963 /* We need to add the mountpoint to the parent. The filesystem may
964 * have placed it on an expiry list, and so we need to make sure it
965 * won't be expired under us if do_add_mount() fails (do_add_mount()
966 * will eat a reference unconditionally).
969 err
= do_add_mount(mnt
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
972 /* Someone else made a mount here whilst we were busy */
975 mnt_clear_expiry(mnt
);
985 path
->dentry
= dget(mnt
->mnt_root
);
992 * Handle a dentry that is managed in some way.
993 * - Flagged for transit management (autofs)
994 * - Flagged as mountpoint
995 * - Flagged as automount point
997 * This may only be called in refwalk mode.
999 * Serialization is taken care of in namespace.c
1001 static int follow_managed(struct path
*path
, unsigned flags
)
1004 bool need_mntput
= false;
1007 /* Given that we're not holding a lock here, we retain the value in a
1008 * local variable for each dentry as we look at it so that we don't see
1009 * the components of that value change under us */
1010 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1011 managed
&= DCACHE_MANAGED_DENTRY
,
1012 unlikely(managed
!= 0)) {
1013 /* Allow the filesystem to manage the transit without i_mutex
1015 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1016 BUG_ON(!path
->dentry
->d_op
);
1017 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1018 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
,
1021 return ret
== -EISDIR
? 0 : ret
;
1024 /* Transit to a mounted filesystem. */
1025 if (managed
& DCACHE_MOUNTED
) {
1026 struct vfsmount
*mounted
= lookup_mnt(path
);
1031 path
->mnt
= mounted
;
1032 path
->dentry
= dget(mounted
->mnt_root
);
1037 /* Something is mounted on this dentry in another
1038 * namespace and/or whatever was mounted there in this
1039 * namespace got unmounted before we managed to get the
1043 /* Handle an automount point */
1044 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1045 ret
= follow_automount(path
, flags
, &need_mntput
);
1047 return ret
== -EISDIR
? 0 : ret
;
1051 /* We didn't change the current path point */
1057 int follow_down_one(struct path
*path
)
1059 struct vfsmount
*mounted
;
1061 mounted
= lookup_mnt(path
);
1065 path
->mnt
= mounted
;
1066 path
->dentry
= dget(mounted
->mnt_root
);
1073 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
1074 * meet a managed dentry and we're not walking to "..". True is returned to
1075 * continue, false to abort.
1077 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1078 struct inode
**inode
, bool reverse_transit
)
1080 while (d_mountpoint(path
->dentry
)) {
1081 struct vfsmount
*mounted
;
1082 if (unlikely(path
->dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) &&
1084 path
->dentry
->d_op
->d_manage(path
->dentry
, false, true) < 0)
1086 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1089 path
->mnt
= mounted
;
1090 path
->dentry
= mounted
->mnt_root
;
1091 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1092 *inode
= path
->dentry
->d_inode
;
1095 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1096 return reverse_transit
;
1100 static int follow_dotdot_rcu(struct nameidata
*nd
)
1102 struct inode
*inode
= nd
->inode
;
1107 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1108 nd
->path
.mnt
== nd
->root
.mnt
) {
1111 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1112 struct dentry
*old
= nd
->path
.dentry
;
1113 struct dentry
*parent
= old
->d_parent
;
1116 seq
= read_seqcount_begin(&parent
->d_seq
);
1117 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1119 inode
= parent
->d_inode
;
1120 nd
->path
.dentry
= parent
;
1124 if (!follow_up_rcu(&nd
->path
))
1126 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1127 inode
= nd
->path
.dentry
->d_inode
;
1129 __follow_mount_rcu(nd
, &nd
->path
, &inode
, true);
1136 * Follow down to the covering mount currently visible to userspace. At each
1137 * point, the filesystem owning that dentry may be queried as to whether the
1138 * caller is permitted to proceed or not.
1140 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1143 int follow_down(struct path
*path
, bool mounting_here
)
1148 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1149 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1150 /* Allow the filesystem to manage the transit without i_mutex
1153 * We indicate to the filesystem if someone is trying to mount
1154 * something here. This gives autofs the chance to deny anyone
1155 * other than its daemon the right to mount on its
1158 * The filesystem may sleep at this point.
1160 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1161 BUG_ON(!path
->dentry
->d_op
);
1162 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1163 ret
= path
->dentry
->d_op
->d_manage(
1164 path
->dentry
, mounting_here
, false);
1166 return ret
== -EISDIR
? 0 : ret
;
1169 /* Transit to a mounted filesystem. */
1170 if (managed
& DCACHE_MOUNTED
) {
1171 struct vfsmount
*mounted
= lookup_mnt(path
);
1176 path
->mnt
= mounted
;
1177 path
->dentry
= dget(mounted
->mnt_root
);
1181 /* Don't handle automount points here */
1188 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1190 static void follow_mount(struct path
*path
)
1192 while (d_mountpoint(path
->dentry
)) {
1193 struct vfsmount
*mounted
= lookup_mnt(path
);
1198 path
->mnt
= mounted
;
1199 path
->dentry
= dget(mounted
->mnt_root
);
1203 static void follow_dotdot(struct nameidata
*nd
)
1208 struct dentry
*old
= nd
->path
.dentry
;
1210 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1211 nd
->path
.mnt
== nd
->root
.mnt
) {
1214 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1215 /* rare case of legitimate dget_parent()... */
1216 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1220 if (!follow_up(&nd
->path
))
1223 follow_mount(&nd
->path
);
1224 nd
->inode
= nd
->path
.dentry
->d_inode
;
1228 * Allocate a dentry with name and parent, and perform a parent
1229 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1230 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1231 * have verified that no child exists while under i_mutex.
1233 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1234 struct qstr
*name
, struct nameidata
*nd
)
1236 struct inode
*inode
= parent
->d_inode
;
1237 struct dentry
*dentry
;
1240 /* Don't create child dentry for a dead directory. */
1241 if (unlikely(IS_DEADDIR(inode
)))
1242 return ERR_PTR(-ENOENT
);
1244 dentry
= d_alloc(parent
, name
);
1245 if (unlikely(!dentry
))
1246 return ERR_PTR(-ENOMEM
);
1248 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1249 if (unlikely(old
)) {
1257 * It's more convoluted than I'd like it to be, but... it's still fairly
1258 * small and for now I'd prefer to have fast path as straight as possible.
1259 * It _is_ time-critical.
1261 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1262 struct path
*path
, struct inode
**inode
)
1264 struct vfsmount
*mnt
= nd
->path
.mnt
;
1265 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1270 * See if the low-level filesystem might want
1271 * to use its own hash..
1273 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1274 err
= parent
->d_op
->d_hash(parent
, nd
->inode
, name
);
1280 * Rename seqlock is not required here because in the off chance
1281 * of a false negative due to a concurrent rename, we're going to
1282 * do the non-racy lookup, below.
1284 if (nd
->flags
& LOOKUP_RCU
) {
1288 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1290 if (nameidata_drop_rcu(nd
))
1294 /* Memory barrier in read_seqcount_begin of child is enough */
1295 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1299 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
)
1300 goto need_revalidate
;
1303 path
->dentry
= dentry
;
1304 if (likely(__follow_mount_rcu(nd
, path
, inode
, false)))
1306 if (nameidata_drop_rcu(nd
))
1310 dentry
= __d_lookup(parent
, name
);
1314 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
)
1315 goto need_revalidate
;
1318 path
->dentry
= dentry
;
1319 err
= follow_managed(path
, nd
->flags
);
1320 if (unlikely(err
< 0))
1322 *inode
= path
->dentry
->d_inode
;
1326 dir
= parent
->d_inode
;
1327 BUG_ON(nd
->inode
!= dir
);
1329 mutex_lock(&dir
->i_mutex
);
1331 * First re-do the cached lookup just in case it was created
1332 * while we waited for the directory semaphore, or the first
1333 * lookup failed due to an unrelated rename.
1335 * This could use version numbering or similar to avoid unnecessary
1336 * cache lookups, but then we'd have to do the first lookup in the
1337 * non-racy way. However in the common case here, everything should
1338 * be hot in cache, so would it be a big win?
1340 dentry
= d_lookup(parent
, name
);
1341 if (likely(!dentry
)) {
1342 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1343 mutex_unlock(&dir
->i_mutex
);
1349 * Uhhuh! Nasty case: the cache was re-populated while
1350 * we waited on the semaphore. Need to revalidate.
1352 mutex_unlock(&dir
->i_mutex
);
1356 dentry
= do_revalidate(dentry
, nd
);
1361 if (nd
->flags
& LOOKUP_RCU
)
1366 return PTR_ERR(dentry
);
1371 * This is the basic name resolution function, turning a pathname into
1372 * the final dentry. We expect 'base' to be positive and a directory.
1374 * Returns 0 and nd will have valid dentry and mnt on success.
1375 * Returns error and drops reference to input namei data on failure.
1377 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1381 unsigned int lookup_flags
= nd
->flags
;
1389 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
1391 /* At this point we know we have a real path component. */
1393 struct inode
*inode
;
1398 nd
->flags
|= LOOKUP_CONTINUE
;
1399 if (nd
->flags
& LOOKUP_RCU
) {
1400 err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1401 if (err
== -ECHILD
) {
1402 if (nameidata_drop_rcu(nd
))
1408 err
= exec_permission(nd
->inode
, 0);
1414 c
= *(const unsigned char *)name
;
1416 hash
= init_name_hash();
1419 hash
= partial_name_hash(c
, hash
);
1420 c
= *(const unsigned char *)name
;
1421 } while (c
&& (c
!= '/'));
1422 this.len
= name
- (const char *) this.name
;
1423 this.hash
= end_name_hash(hash
);
1425 /* remove trailing slashes? */
1427 goto last_component
;
1428 while (*++name
== '/');
1430 goto last_with_slashes
;
1433 * "." and ".." are special - ".." especially so because it has
1434 * to be able to know about the current root directory and
1435 * parent relationships.
1437 if (this.name
[0] == '.') switch (this.len
) {
1441 if (this.name
[1] != '.')
1443 if (nd
->flags
& LOOKUP_RCU
) {
1444 if (follow_dotdot_rcu(nd
))
1452 /* This does the actual lookups.. */
1453 err
= do_lookup(nd
, &this, &next
, &inode
);
1460 if (inode
->i_op
->follow_link
) {
1461 /* We commonly drop rcu-walk here */
1462 if (nameidata_dentry_drop_rcu_maybe(nd
, next
.dentry
))
1464 BUG_ON(inode
!= next
.dentry
->d_inode
);
1465 err
= do_follow_link(&next
, nd
);
1468 nd
->inode
= nd
->path
.dentry
->d_inode
;
1473 path_to_nameidata(&next
, nd
);
1477 if (!nd
->inode
->i_op
->lookup
)
1480 /* here ends the main loop */
1483 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1485 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1486 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1487 if (lookup_flags
& LOOKUP_PARENT
)
1489 if (this.name
[0] == '.') switch (this.len
) {
1493 if (this.name
[1] != '.')
1495 if (nd
->flags
& LOOKUP_RCU
) {
1496 if (follow_dotdot_rcu(nd
))
1504 err
= do_lookup(nd
, &this, &next
, &inode
);
1507 if (inode
&& unlikely(inode
->i_op
->follow_link
) &&
1508 (lookup_flags
& LOOKUP_FOLLOW
)) {
1509 if (nameidata_dentry_drop_rcu_maybe(nd
, next
.dentry
))
1511 BUG_ON(inode
!= next
.dentry
->d_inode
);
1512 err
= do_follow_link(&next
, nd
);
1515 nd
->inode
= nd
->path
.dentry
->d_inode
;
1517 path_to_nameidata(&next
, nd
);
1523 if (lookup_flags
& LOOKUP_DIRECTORY
) {
1525 if (!nd
->inode
->i_op
->lookup
)
1531 nd
->last_type
= LAST_NORM
;
1532 if (this.name
[0] != '.')
1535 nd
->last_type
= LAST_DOT
;
1536 else if (this.len
== 2 && this.name
[1] == '.')
1537 nd
->last_type
= LAST_DOTDOT
;
1542 * We bypassed the ordinary revalidation routines.
1543 * We may need to check the cached dentry for staleness.
1545 if (need_reval_dot(nd
->path
.dentry
)) {
1546 /* Note: we do not d_invalidate() */
1547 err
= d_revalidate(nd
->path
.dentry
, nd
);
1554 if (nameidata_drop_rcu_last_maybe(nd
))
1558 if (!(nd
->flags
& LOOKUP_RCU
))
1559 path_put_conditional(&next
, nd
);
1562 if (!(nd
->flags
& LOOKUP_RCU
))
1563 path_put(&nd
->path
);
1568 static inline int path_walk_rcu(const char *name
, struct nameidata
*nd
)
1570 current
->total_link_count
= 0;
1572 return link_path_walk(name
, nd
);
1575 static inline int path_walk_simple(const char *name
, struct nameidata
*nd
)
1577 current
->total_link_count
= 0;
1579 return link_path_walk(name
, nd
);
1582 static int path_walk(const char *name
, struct nameidata
*nd
)
1584 struct path save
= nd
->path
;
1587 current
->total_link_count
= 0;
1589 /* make sure the stuff we saved doesn't go away */
1592 result
= link_path_walk(name
, nd
);
1593 if (result
== -ESTALE
) {
1594 /* nd->path had been dropped */
1595 current
->total_link_count
= 0;
1597 path_get(&nd
->path
);
1598 nd
->flags
|= LOOKUP_REVAL
;
1599 result
= link_path_walk(name
, nd
);
1607 static void path_finish_rcu(struct nameidata
*nd
)
1609 if (nd
->flags
& LOOKUP_RCU
) {
1610 /* RCU dangling. Cancel it. */
1611 nd
->flags
&= ~LOOKUP_RCU
;
1612 nd
->root
.mnt
= NULL
;
1614 br_read_unlock(vfsmount_lock
);
1620 static int path_init_rcu(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1626 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1627 nd
->flags
= flags
| LOOKUP_RCU
;
1629 nd
->root
.mnt
= NULL
;
1633 struct fs_struct
*fs
= current
->fs
;
1636 br_read_lock(vfsmount_lock
);
1640 seq
= read_seqcount_begin(&fs
->seq
);
1641 nd
->root
= fs
->root
;
1642 nd
->path
= nd
->root
;
1643 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1644 } while (read_seqcount_retry(&fs
->seq
, seq
));
1646 } else if (dfd
== AT_FDCWD
) {
1647 struct fs_struct
*fs
= current
->fs
;
1650 br_read_lock(vfsmount_lock
);
1654 seq
= read_seqcount_begin(&fs
->seq
);
1656 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1657 } while (read_seqcount_retry(&fs
->seq
, seq
));
1660 struct dentry
*dentry
;
1662 file
= fget_light(dfd
, &fput_needed
);
1667 dentry
= file
->f_path
.dentry
;
1670 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1673 retval
= file_permission(file
, MAY_EXEC
);
1677 nd
->path
= file
->f_path
;
1681 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1682 br_read_lock(vfsmount_lock
);
1685 nd
->inode
= nd
->path
.dentry
->d_inode
;
1689 fput_light(file
, fput_needed
);
1694 static int path_init(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1700 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1703 nd
->root
.mnt
= NULL
;
1707 nd
->path
= nd
->root
;
1708 path_get(&nd
->root
);
1709 } else if (dfd
== AT_FDCWD
) {
1710 get_fs_pwd(current
->fs
, &nd
->path
);
1712 struct dentry
*dentry
;
1714 file
= fget_light(dfd
, &fput_needed
);
1719 dentry
= file
->f_path
.dentry
;
1722 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1725 retval
= file_permission(file
, MAY_EXEC
);
1729 nd
->path
= file
->f_path
;
1730 path_get(&file
->f_path
);
1732 fput_light(file
, fput_needed
);
1734 nd
->inode
= nd
->path
.dentry
->d_inode
;
1738 fput_light(file
, fput_needed
);
1743 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1744 static int do_path_lookup(int dfd
, const char *name
,
1745 unsigned int flags
, struct nameidata
*nd
)
1750 * Path walking is largely split up into 2 different synchronisation
1751 * schemes, rcu-walk and ref-walk (explained in
1752 * Documentation/filesystems/path-lookup.txt). These share much of the
1753 * path walk code, but some things particularly setup, cleanup, and
1754 * following mounts are sufficiently divergent that functions are
1755 * duplicated. Typically there is a function foo(), and its RCU
1756 * analogue, foo_rcu().
1758 * -ECHILD is the error number of choice (just to avoid clashes) that
1759 * is returned if some aspect of an rcu-walk fails. Such an error must
1760 * be handled by restarting a traditional ref-walk (which will always
1761 * be able to complete).
1763 retval
= path_init_rcu(dfd
, name
, flags
, nd
);
1764 if (unlikely(retval
))
1766 retval
= path_walk_rcu(name
, nd
);
1767 path_finish_rcu(nd
);
1769 path_put(&nd
->root
);
1770 nd
->root
.mnt
= NULL
;
1773 if (unlikely(retval
== -ECHILD
|| retval
== -ESTALE
)) {
1774 /* slower, locked walk */
1775 if (retval
== -ESTALE
)
1776 flags
|= LOOKUP_REVAL
;
1777 retval
= path_init(dfd
, name
, flags
, nd
);
1778 if (unlikely(retval
))
1780 retval
= path_walk(name
, nd
);
1782 path_put(&nd
->root
);
1783 nd
->root
.mnt
= NULL
;
1787 if (likely(!retval
)) {
1788 if (unlikely(!audit_dummy_context())) {
1789 if (nd
->path
.dentry
&& nd
->inode
)
1790 audit_inode(name
, nd
->path
.dentry
);
1797 int path_lookup(const char *name
, unsigned int flags
,
1798 struct nameidata
*nd
)
1800 return do_path_lookup(AT_FDCWD
, name
, flags
, nd
);
1803 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1805 struct nameidata nd
;
1806 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1813 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1814 * @dentry: pointer to dentry of the base directory
1815 * @mnt: pointer to vfs mount of the base directory
1816 * @name: pointer to file name
1817 * @flags: lookup flags
1818 * @nd: pointer to nameidata
1820 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1821 const char *name
, unsigned int flags
,
1822 struct nameidata
*nd
)
1826 /* same as do_path_lookup */
1827 nd
->last_type
= LAST_ROOT
;
1831 nd
->path
.dentry
= dentry
;
1833 path_get(&nd
->path
);
1834 nd
->root
= nd
->path
;
1835 path_get(&nd
->root
);
1836 nd
->inode
= nd
->path
.dentry
->d_inode
;
1838 retval
= path_walk(name
, nd
);
1839 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1841 audit_inode(name
, nd
->path
.dentry
);
1843 path_put(&nd
->root
);
1844 nd
->root
.mnt
= NULL
;
1849 static struct dentry
*__lookup_hash(struct qstr
*name
,
1850 struct dentry
*base
, struct nameidata
*nd
)
1852 struct inode
*inode
= base
->d_inode
;
1853 struct dentry
*dentry
;
1856 err
= exec_permission(inode
, 0);
1858 return ERR_PTR(err
);
1861 * See if the low-level filesystem might want
1862 * to use its own hash..
1864 if (base
->d_flags
& DCACHE_OP_HASH
) {
1865 err
= base
->d_op
->d_hash(base
, inode
, name
);
1866 dentry
= ERR_PTR(err
);
1872 * Don't bother with __d_lookup: callers are for creat as
1873 * well as unlink, so a lot of the time it would cost
1876 dentry
= d_lookup(base
, name
);
1878 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1879 dentry
= do_revalidate(dentry
, nd
);
1882 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1888 * Restricted form of lookup. Doesn't follow links, single-component only,
1889 * needs parent already locked. Doesn't follow mounts.
1892 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1894 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1897 static int __lookup_one_len(const char *name
, struct qstr
*this,
1898 struct dentry
*base
, int len
)
1908 hash
= init_name_hash();
1910 c
= *(const unsigned char *)name
++;
1911 if (c
== '/' || c
== '\0')
1913 hash
= partial_name_hash(c
, hash
);
1915 this->hash
= end_name_hash(hash
);
1920 * lookup_one_len - filesystem helper to lookup single pathname component
1921 * @name: pathname component to lookup
1922 * @base: base directory to lookup from
1923 * @len: maximum length @len should be interpreted to
1925 * Note that this routine is purely a helper for filesystem usage and should
1926 * not be called by generic code. Also note that by using this function the
1927 * nameidata argument is passed to the filesystem methods and a filesystem
1928 * using this helper needs to be prepared for that.
1930 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1935 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1937 err
= __lookup_one_len(name
, &this, base
, len
);
1939 return ERR_PTR(err
);
1941 return __lookup_hash(&this, base
, NULL
);
1944 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1947 struct nameidata nd
;
1948 char *tmp
= getname(name
);
1949 int err
= PTR_ERR(tmp
);
1952 BUG_ON(flags
& LOOKUP_PARENT
);
1954 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1962 static int user_path_parent(int dfd
, const char __user
*path
,
1963 struct nameidata
*nd
, char **name
)
1965 char *s
= getname(path
);
1971 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1981 * It's inline, so penalty for filesystems that don't use sticky bit is
1984 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1986 uid_t fsuid
= current_fsuid();
1988 if (!(dir
->i_mode
& S_ISVTX
))
1990 if (inode
->i_uid
== fsuid
)
1992 if (dir
->i_uid
== fsuid
)
1994 return !capable(CAP_FOWNER
);
1998 * Check whether we can remove a link victim from directory dir, check
1999 * whether the type of victim is right.
2000 * 1. We can't do it if dir is read-only (done in permission())
2001 * 2. We should have write and exec permissions on dir
2002 * 3. We can't remove anything from append-only dir
2003 * 4. We can't do anything with immutable dir (done in permission())
2004 * 5. If the sticky bit on dir is set we should either
2005 * a. be owner of dir, or
2006 * b. be owner of victim, or
2007 * c. have CAP_FOWNER capability
2008 * 6. If the victim is append-only or immutable we can't do antyhing with
2009 * links pointing to it.
2010 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2011 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2012 * 9. We can't remove a root or mountpoint.
2013 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2014 * nfs_async_unlink().
2016 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
2020 if (!victim
->d_inode
)
2023 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2024 audit_inode_child(victim
, dir
);
2026 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2031 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
2032 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
2035 if (!S_ISDIR(victim
->d_inode
->i_mode
))
2037 if (IS_ROOT(victim
))
2039 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
2041 if (IS_DEADDIR(dir
))
2043 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2048 /* Check whether we can create an object with dentry child in directory
2050 * 1. We can't do it if child already exists (open has special treatment for
2051 * this case, but since we are inlined it's OK)
2052 * 2. We can't do it if dir is read-only (done in permission())
2053 * 3. We should have write and exec permissions on dir
2054 * 4. We can't do it if dir is immutable (done in permission())
2056 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2060 if (IS_DEADDIR(dir
))
2062 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2066 * p1 and p2 should be directories on the same fs.
2068 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2073 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2077 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2079 p
= d_ancestor(p2
, p1
);
2081 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2082 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2086 p
= d_ancestor(p1
, p2
);
2088 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2089 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2093 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2094 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2098 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2100 mutex_unlock(&p1
->d_inode
->i_mutex
);
2102 mutex_unlock(&p2
->d_inode
->i_mutex
);
2103 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2107 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
2108 struct nameidata
*nd
)
2110 int error
= may_create(dir
, dentry
);
2115 if (!dir
->i_op
->create
)
2116 return -EACCES
; /* shouldn't it be ENOSYS? */
2119 error
= security_inode_create(dir
, dentry
, mode
);
2122 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
2124 fsnotify_create(dir
, dentry
);
2128 int may_open(struct path
*path
, int acc_mode
, int flag
)
2130 struct dentry
*dentry
= path
->dentry
;
2131 struct inode
*inode
= dentry
->d_inode
;
2137 switch (inode
->i_mode
& S_IFMT
) {
2141 if (acc_mode
& MAY_WRITE
)
2146 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2155 error
= inode_permission(inode
, acc_mode
);
2160 * An append-only file must be opened in append mode for writing.
2162 if (IS_APPEND(inode
)) {
2163 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2169 /* O_NOATIME can only be set by the owner or superuser */
2170 if (flag
& O_NOATIME
&& !is_owner_or_cap(inode
))
2174 * Ensure there are no outstanding leases on the file.
2176 return break_lease(inode
, flag
);
2179 static int handle_truncate(struct file
*filp
)
2181 struct path
*path
= &filp
->f_path
;
2182 struct inode
*inode
= path
->dentry
->d_inode
;
2183 int error
= get_write_access(inode
);
2187 * Refuse to truncate files with mandatory locks held on them.
2189 error
= locks_verify_locked(inode
);
2191 error
= security_path_truncate(path
);
2193 error
= do_truncate(path
->dentry
, 0,
2194 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2197 put_write_access(inode
);
2202 * Be careful about ever adding any more callers of this
2203 * function. Its flags must be in the namei format, not
2204 * what get passed to sys_open().
2206 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
2207 int open_flag
, int mode
)
2210 struct dentry
*dir
= nd
->path
.dentry
;
2212 if (!IS_POSIXACL(dir
->d_inode
))
2213 mode
&= ~current_umask();
2214 error
= security_path_mknod(&nd
->path
, path
->dentry
, mode
, 0);
2217 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
2219 mutex_unlock(&dir
->d_inode
->i_mutex
);
2220 dput(nd
->path
.dentry
);
2221 nd
->path
.dentry
= path
->dentry
;
2225 /* Don't check for write permission, don't truncate */
2226 return may_open(&nd
->path
, 0, open_flag
& ~O_TRUNC
);
2230 * Note that while the flag value (low two bits) for sys_open means:
2235 * it is changed into
2236 * 00 - no permissions needed
2237 * 01 - read-permission
2238 * 10 - write-permission
2240 * for the internal routines (ie open_namei()/follow_link() etc)
2241 * This is more logical, and also allows the 00 "no perm needed"
2242 * to be used for symlinks (where the permissions are checked
2246 static inline int open_to_namei_flags(int flag
)
2248 if ((flag
+1) & O_ACCMODE
)
2253 static int open_will_truncate(int flag
, struct inode
*inode
)
2256 * We'll never write to the fs underlying
2259 if (special_file(inode
->i_mode
))
2261 return (flag
& O_TRUNC
);
2264 static struct file
*finish_open(struct nameidata
*nd
,
2265 int open_flag
, int acc_mode
)
2271 will_truncate
= open_will_truncate(open_flag
, nd
->path
.dentry
->d_inode
);
2272 if (will_truncate
) {
2273 error
= mnt_want_write(nd
->path
.mnt
);
2277 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2280 mnt_drop_write(nd
->path
.mnt
);
2283 filp
= nameidata_to_filp(nd
);
2284 if (!IS_ERR(filp
)) {
2285 error
= ima_file_check(filp
, acc_mode
);
2288 filp
= ERR_PTR(error
);
2291 if (!IS_ERR(filp
)) {
2292 if (will_truncate
) {
2293 error
= handle_truncate(filp
);
2296 filp
= ERR_PTR(error
);
2301 * It is now safe to drop the mnt write
2302 * because the filp has had a write taken
2306 mnt_drop_write(nd
->path
.mnt
);
2307 path_put(&nd
->path
);
2311 if (!IS_ERR(nd
->intent
.open
.file
))
2312 release_open_intent(nd
);
2313 path_put(&nd
->path
);
2314 return ERR_PTR(error
);
2318 * Handle O_CREAT case for do_filp_open
2320 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2321 int open_flag
, int acc_mode
,
2322 int mode
, const char *pathname
)
2324 struct dentry
*dir
= nd
->path
.dentry
;
2326 int error
= -EISDIR
;
2328 switch (nd
->last_type
) {
2331 dir
= nd
->path
.dentry
;
2333 if (need_reval_dot(dir
)) {
2334 int status
= d_revalidate(nd
->path
.dentry
, nd
);
2346 audit_inode(pathname
, dir
);
2350 /* trailing slashes? */
2351 if (nd
->last
.name
[nd
->last
.len
])
2354 mutex_lock(&dir
->d_inode
->i_mutex
);
2356 path
->dentry
= lookup_hash(nd
);
2357 path
->mnt
= nd
->path
.mnt
;
2359 error
= PTR_ERR(path
->dentry
);
2360 if (IS_ERR(path
->dentry
)) {
2361 mutex_unlock(&dir
->d_inode
->i_mutex
);
2365 if (IS_ERR(nd
->intent
.open
.file
)) {
2366 error
= PTR_ERR(nd
->intent
.open
.file
);
2367 goto exit_mutex_unlock
;
2370 /* Negative dentry, just create the file */
2371 if (!path
->dentry
->d_inode
) {
2373 * This write is needed to ensure that a
2374 * ro->rw transition does not occur between
2375 * the time when the file is created and when
2376 * a permanent write count is taken through
2377 * the 'struct file' in nameidata_to_filp().
2379 error
= mnt_want_write(nd
->path
.mnt
);
2381 goto exit_mutex_unlock
;
2382 error
= __open_namei_create(nd
, path
, open_flag
, mode
);
2384 mnt_drop_write(nd
->path
.mnt
);
2387 filp
= nameidata_to_filp(nd
);
2388 mnt_drop_write(nd
->path
.mnt
);
2389 path_put(&nd
->path
);
2390 if (!IS_ERR(filp
)) {
2391 error
= ima_file_check(filp
, acc_mode
);
2394 filp
= ERR_PTR(error
);
2401 * It already exists.
2403 mutex_unlock(&dir
->d_inode
->i_mutex
);
2404 audit_inode(pathname
, path
->dentry
);
2407 if (open_flag
& O_EXCL
)
2410 error
= follow_managed(path
, nd
->flags
);
2415 if (!path
->dentry
->d_inode
)
2418 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2421 path_to_nameidata(path
, nd
);
2422 nd
->inode
= path
->dentry
->d_inode
;
2424 if (S_ISDIR(nd
->inode
->i_mode
))
2427 filp
= finish_open(nd
, open_flag
, acc_mode
);
2431 mutex_unlock(&dir
->d_inode
->i_mutex
);
2433 path_put_conditional(path
, nd
);
2435 if (!IS_ERR(nd
->intent
.open
.file
))
2436 release_open_intent(nd
);
2437 path_put(&nd
->path
);
2438 return ERR_PTR(error
);
2442 * Note that the low bits of the passed in "open_flag"
2443 * are not the same as in the local variable "flag". See
2444 * open_to_namei_flags() for more details.
2446 struct file
*do_filp_open(int dfd
, const char *pathname
,
2447 int open_flag
, int mode
, int acc_mode
)
2450 struct nameidata nd
;
2454 int flag
= open_to_namei_flags(open_flag
);
2457 if (!(open_flag
& O_CREAT
))
2460 /* Must never be set by userspace */
2461 open_flag
&= ~FMODE_NONOTIFY
;
2464 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2465 * check for O_DSYNC if the need any syncing at all we enforce it's
2466 * always set instead of having to deal with possibly weird behaviour
2467 * for malicious applications setting only __O_SYNC.
2469 if (open_flag
& __O_SYNC
)
2470 open_flag
|= O_DSYNC
;
2473 acc_mode
= MAY_OPEN
| ACC_MODE(open_flag
);
2475 /* O_TRUNC implies we need access checks for write permissions */
2476 if (open_flag
& O_TRUNC
)
2477 acc_mode
|= MAY_WRITE
;
2479 /* Allow the LSM permission hook to distinguish append
2480 access from general write access. */
2481 if (open_flag
& O_APPEND
)
2482 acc_mode
|= MAY_APPEND
;
2484 flags
= LOOKUP_OPEN
;
2485 if (open_flag
& O_CREAT
) {
2486 flags
|= LOOKUP_CREATE
;
2487 if (open_flag
& O_EXCL
)
2488 flags
|= LOOKUP_EXCL
;
2490 if (open_flag
& O_DIRECTORY
)
2491 flags
|= LOOKUP_DIRECTORY
;
2492 if (!(open_flag
& O_NOFOLLOW
))
2493 flags
|= LOOKUP_FOLLOW
;
2495 filp
= get_empty_filp();
2497 return ERR_PTR(-ENFILE
);
2499 filp
->f_flags
= open_flag
;
2500 nd
.intent
.open
.file
= filp
;
2501 nd
.intent
.open
.flags
= flag
;
2502 nd
.intent
.open
.create_mode
= mode
;
2504 if (open_flag
& O_CREAT
)
2507 /* !O_CREAT, simple open */
2508 error
= do_path_lookup(dfd
, pathname
, flags
, &nd
);
2509 if (unlikely(error
))
2512 if (!(nd
.flags
& LOOKUP_FOLLOW
)) {
2513 if (nd
.inode
->i_op
->follow_link
)
2517 if (nd
.flags
& LOOKUP_DIRECTORY
) {
2518 if (!nd
.inode
->i_op
->lookup
)
2521 audit_inode(pathname
, nd
.path
.dentry
);
2522 filp
= finish_open(&nd
, open_flag
, acc_mode
);
2526 /* OK, have to create the file. Find the parent. */
2527 error
= path_init_rcu(dfd
, pathname
,
2528 LOOKUP_PARENT
| (flags
& LOOKUP_REVAL
), &nd
);
2531 error
= path_walk_rcu(pathname
, &nd
);
2532 path_finish_rcu(&nd
);
2533 if (unlikely(error
== -ECHILD
|| error
== -ESTALE
)) {
2534 /* slower, locked walk */
2535 if (error
== -ESTALE
) {
2537 flags
|= LOOKUP_REVAL
;
2539 error
= path_init(dfd
, pathname
,
2540 LOOKUP_PARENT
| (flags
& LOOKUP_REVAL
), &nd
);
2544 error
= path_walk_simple(pathname
, &nd
);
2546 if (unlikely(error
))
2548 if (unlikely(!audit_dummy_context()))
2549 audit_inode(pathname
, nd
.path
.dentry
);
2552 * We have the parent and last component.
2555 filp
= do_last(&nd
, &path
, open_flag
, acc_mode
, mode
, pathname
);
2556 while (unlikely(!filp
)) { /* trailing symlink */
2557 struct path link
= path
;
2558 struct inode
*linki
= link
.dentry
->d_inode
;
2561 if (!(nd
.flags
& LOOKUP_FOLLOW
))
2566 * This is subtle. Instead of calling do_follow_link() we do
2567 * the thing by hands. The reason is that this way we have zero
2568 * link_count and path_walk() (called from ->follow_link)
2569 * honoring LOOKUP_PARENT. After that we have the parent and
2570 * last component, i.e. we are in the same situation as after
2571 * the first path_walk(). Well, almost - if the last component
2572 * is normal we get its copy stored in nd->last.name and we will
2573 * have to putname() it when we are done. Procfs-like symlinks
2574 * just set LAST_BIND.
2576 nd
.flags
|= LOOKUP_PARENT
;
2577 error
= security_inode_follow_link(link
.dentry
, &nd
);
2580 error
= __do_follow_link(&link
, &nd
, &cookie
);
2581 if (unlikely(error
)) {
2582 if (!IS_ERR(cookie
) && linki
->i_op
->put_link
)
2583 linki
->i_op
->put_link(link
.dentry
, &nd
, cookie
);
2584 /* nd.path had been dropped */
2588 nd
.flags
&= ~LOOKUP_PARENT
;
2589 filp
= do_last(&nd
, &path
, open_flag
, acc_mode
, mode
, pathname
);
2590 if (linki
->i_op
->put_link
)
2591 linki
->i_op
->put_link(link
.dentry
, &nd
, cookie
);
2597 if (filp
== ERR_PTR(-ESTALE
) && !(flags
& LOOKUP_REVAL
))
2602 path_put_conditional(&path
, &nd
);
2606 if (!IS_ERR(nd
.intent
.open
.file
))
2607 release_open_intent(&nd
);
2608 filp
= ERR_PTR(error
);
2613 * filp_open - open file and return file pointer
2615 * @filename: path to open
2616 * @flags: open flags as per the open(2) second argument
2617 * @mode: mode for the new file if O_CREAT is set, else ignored
2619 * This is the helper to open a file from kernelspace if you really
2620 * have to. But in generally you should not do this, so please move
2621 * along, nothing to see here..
2623 struct file
*filp_open(const char *filename
, int flags
, int mode
)
2625 return do_filp_open(AT_FDCWD
, filename
, flags
, mode
, 0);
2627 EXPORT_SYMBOL(filp_open
);
2630 * lookup_create - lookup a dentry, creating it if it doesn't exist
2631 * @nd: nameidata info
2632 * @is_dir: directory flag
2634 * Simple function to lookup and return a dentry and create it
2635 * if it doesn't exist. Is SMP-safe.
2637 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2639 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2641 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2643 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2645 * Yucky last component or no last component at all?
2646 * (foo/., foo/.., /////)
2648 if (nd
->last_type
!= LAST_NORM
)
2650 nd
->flags
&= ~LOOKUP_PARENT
;
2651 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2652 nd
->intent
.open
.flags
= O_EXCL
;
2655 * Do the final lookup.
2657 dentry
= lookup_hash(nd
);
2661 if (dentry
->d_inode
)
2664 * Special case - lookup gave negative, but... we had foo/bar/
2665 * From the vfs_mknod() POV we just have a negative dentry -
2666 * all is fine. Let's be bastards - you had / on the end, you've
2667 * been asking for (non-existent) directory. -ENOENT for you.
2669 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2671 dentry
= ERR_PTR(-ENOENT
);
2676 dentry
= ERR_PTR(-EEXIST
);
2680 EXPORT_SYMBOL_GPL(lookup_create
);
2682 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2684 int error
= may_create(dir
, dentry
);
2689 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
2692 if (!dir
->i_op
->mknod
)
2695 error
= devcgroup_inode_mknod(mode
, dev
);
2699 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2703 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2705 fsnotify_create(dir
, dentry
);
2709 static int may_mknod(mode_t mode
)
2711 switch (mode
& S_IFMT
) {
2717 case 0: /* zero mode translates to S_IFREG */
2726 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2731 struct dentry
*dentry
;
2732 struct nameidata nd
;
2737 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2741 dentry
= lookup_create(&nd
, 0);
2742 if (IS_ERR(dentry
)) {
2743 error
= PTR_ERR(dentry
);
2746 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2747 mode
&= ~current_umask();
2748 error
= may_mknod(mode
);
2751 error
= mnt_want_write(nd
.path
.mnt
);
2754 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2756 goto out_drop_write
;
2757 switch (mode
& S_IFMT
) {
2758 case 0: case S_IFREG
:
2759 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2761 case S_IFCHR
: case S_IFBLK
:
2762 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2763 new_decode_dev(dev
));
2765 case S_IFIFO
: case S_IFSOCK
:
2766 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2770 mnt_drop_write(nd
.path
.mnt
);
2774 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2781 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2783 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2786 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2788 int error
= may_create(dir
, dentry
);
2793 if (!dir
->i_op
->mkdir
)
2796 mode
&= (S_IRWXUGO
|S_ISVTX
);
2797 error
= security_inode_mkdir(dir
, dentry
, mode
);
2801 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2803 fsnotify_mkdir(dir
, dentry
);
2807 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2811 struct dentry
*dentry
;
2812 struct nameidata nd
;
2814 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2818 dentry
= lookup_create(&nd
, 1);
2819 error
= PTR_ERR(dentry
);
2823 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2824 mode
&= ~current_umask();
2825 error
= mnt_want_write(nd
.path
.mnt
);
2828 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2830 goto out_drop_write
;
2831 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2833 mnt_drop_write(nd
.path
.mnt
);
2837 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2844 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2846 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2850 * We try to drop the dentry early: we should have
2851 * a usage count of 2 if we're the only user of this
2852 * dentry, and if that is true (possibly after pruning
2853 * the dcache), then we drop the dentry now.
2855 * A low-level filesystem can, if it choses, legally
2858 * if (!d_unhashed(dentry))
2861 * if it cannot handle the case of removing a directory
2862 * that is still in use by something else..
2864 void dentry_unhash(struct dentry
*dentry
)
2867 shrink_dcache_parent(dentry
);
2868 spin_lock(&dentry
->d_lock
);
2869 if (dentry
->d_count
== 2)
2871 spin_unlock(&dentry
->d_lock
);
2874 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2876 int error
= may_delete(dir
, dentry
, 1);
2881 if (!dir
->i_op
->rmdir
)
2884 mutex_lock(&dentry
->d_inode
->i_mutex
);
2885 dentry_unhash(dentry
);
2886 if (d_mountpoint(dentry
))
2889 error
= security_inode_rmdir(dir
, dentry
);
2891 error
= dir
->i_op
->rmdir(dir
, dentry
);
2893 dentry
->d_inode
->i_flags
|= S_DEAD
;
2898 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2907 static long do_rmdir(int dfd
, const char __user
*pathname
)
2911 struct dentry
*dentry
;
2912 struct nameidata nd
;
2914 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2918 switch(nd
.last_type
) {
2930 nd
.flags
&= ~LOOKUP_PARENT
;
2932 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2933 dentry
= lookup_hash(&nd
);
2934 error
= PTR_ERR(dentry
);
2937 error
= mnt_want_write(nd
.path
.mnt
);
2940 error
= security_path_rmdir(&nd
.path
, dentry
);
2943 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2945 mnt_drop_write(nd
.path
.mnt
);
2949 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2956 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2958 return do_rmdir(AT_FDCWD
, pathname
);
2961 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2963 int error
= may_delete(dir
, dentry
, 0);
2968 if (!dir
->i_op
->unlink
)
2971 mutex_lock(&dentry
->d_inode
->i_mutex
);
2972 if (d_mountpoint(dentry
))
2975 error
= security_inode_unlink(dir
, dentry
);
2977 error
= dir
->i_op
->unlink(dir
, dentry
);
2982 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2984 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2985 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2986 fsnotify_link_count(dentry
->d_inode
);
2994 * Make sure that the actual truncation of the file will occur outside its
2995 * directory's i_mutex. Truncate can take a long time if there is a lot of
2996 * writeout happening, and we don't want to prevent access to the directory
2997 * while waiting on the I/O.
2999 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3003 struct dentry
*dentry
;
3004 struct nameidata nd
;
3005 struct inode
*inode
= NULL
;
3007 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
3012 if (nd
.last_type
!= LAST_NORM
)
3015 nd
.flags
&= ~LOOKUP_PARENT
;
3017 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3018 dentry
= lookup_hash(&nd
);
3019 error
= PTR_ERR(dentry
);
3020 if (!IS_ERR(dentry
)) {
3021 /* Why not before? Because we want correct error value */
3022 if (nd
.last
.name
[nd
.last
.len
])
3024 inode
= dentry
->d_inode
;
3027 error
= mnt_want_write(nd
.path
.mnt
);
3030 error
= security_path_unlink(&nd
.path
, dentry
);
3033 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
3035 mnt_drop_write(nd
.path
.mnt
);
3039 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3041 iput(inode
); /* truncate the inode here */
3048 error
= !dentry
->d_inode
? -ENOENT
:
3049 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
3053 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3055 if ((flag
& ~AT_REMOVEDIR
) != 0)
3058 if (flag
& AT_REMOVEDIR
)
3059 return do_rmdir(dfd
, pathname
);
3061 return do_unlinkat(dfd
, pathname
);
3064 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3066 return do_unlinkat(AT_FDCWD
, pathname
);
3069 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3071 int error
= may_create(dir
, dentry
);
3076 if (!dir
->i_op
->symlink
)
3079 error
= security_inode_symlink(dir
, dentry
, oldname
);
3083 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3085 fsnotify_create(dir
, dentry
);
3089 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3090 int, newdfd
, const char __user
*, newname
)
3095 struct dentry
*dentry
;
3096 struct nameidata nd
;
3098 from
= getname(oldname
);
3100 return PTR_ERR(from
);
3102 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
3106 dentry
= lookup_create(&nd
, 0);
3107 error
= PTR_ERR(dentry
);
3111 error
= mnt_want_write(nd
.path
.mnt
);
3114 error
= security_path_symlink(&nd
.path
, dentry
, from
);
3116 goto out_drop_write
;
3117 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
3119 mnt_drop_write(nd
.path
.mnt
);
3123 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3131 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3133 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3136 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3138 struct inode
*inode
= old_dentry
->d_inode
;
3144 error
= may_create(dir
, new_dentry
);
3148 if (dir
->i_sb
!= inode
->i_sb
)
3152 * A link to an append-only or immutable file cannot be created.
3154 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3156 if (!dir
->i_op
->link
)
3158 if (S_ISDIR(inode
->i_mode
))
3161 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3165 mutex_lock(&inode
->i_mutex
);
3166 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3167 mutex_unlock(&inode
->i_mutex
);
3169 fsnotify_link(dir
, inode
, new_dentry
);
3174 * Hardlinks are often used in delicate situations. We avoid
3175 * security-related surprises by not following symlinks on the
3178 * We don't follow them on the oldname either to be compatible
3179 * with linux 2.0, and to avoid hard-linking to directories
3180 * and other special files. --ADM
3182 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3183 int, newdfd
, const char __user
*, newname
, int, flags
)
3185 struct dentry
*new_dentry
;
3186 struct nameidata nd
;
3187 struct path old_path
;
3191 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
3194 error
= user_path_at(olddfd
, oldname
,
3195 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
3200 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
3204 if (old_path
.mnt
!= nd
.path
.mnt
)
3206 new_dentry
= lookup_create(&nd
, 0);
3207 error
= PTR_ERR(new_dentry
);
3208 if (IS_ERR(new_dentry
))
3210 error
= mnt_want_write(nd
.path
.mnt
);
3213 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
3215 goto out_drop_write
;
3216 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
3218 mnt_drop_write(nd
.path
.mnt
);
3222 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3227 path_put(&old_path
);
3232 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3234 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3238 * The worst of all namespace operations - renaming directory. "Perverted"
3239 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3241 * a) we can get into loop creation. Check is done in is_subdir().
3242 * b) race potential - two innocent renames can create a loop together.
3243 * That's where 4.4 screws up. Current fix: serialization on
3244 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3246 * c) we have to lock _three_ objects - parents and victim (if it exists).
3247 * And that - after we got ->i_mutex on parents (until then we don't know
3248 * whether the target exists). Solution: try to be smart with locking
3249 * order for inodes. We rely on the fact that tree topology may change
3250 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3251 * move will be locked. Thus we can rank directories by the tree
3252 * (ancestors first) and rank all non-directories after them.
3253 * That works since everybody except rename does "lock parent, lookup,
3254 * lock child" and rename is under ->s_vfs_rename_mutex.
3255 * HOWEVER, it relies on the assumption that any object with ->lookup()
3256 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3257 * we'd better make sure that there's no link(2) for them.
3258 * d) some filesystems don't support opened-but-unlinked directories,
3259 * either because of layout or because they are not ready to deal with
3260 * all cases correctly. The latter will be fixed (taking this sort of
3261 * stuff into VFS), but the former is not going away. Solution: the same
3262 * trick as in rmdir().
3263 * e) conversion from fhandle to dentry may come in the wrong moment - when
3264 * we are removing the target. Solution: we will have to grab ->i_mutex
3265 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3266 * ->i_mutex on parents, which works but leads to some truly excessive
3269 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3270 struct inode
*new_dir
, struct dentry
*new_dentry
)
3273 struct inode
*target
;
3276 * If we are going to change the parent - check write permissions,
3277 * we'll need to flip '..'.
3279 if (new_dir
!= old_dir
) {
3280 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3285 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3289 target
= new_dentry
->d_inode
;
3291 mutex_lock(&target
->i_mutex
);
3292 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3296 dentry_unhash(new_dentry
);
3297 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3301 target
->i_flags
|= S_DEAD
;
3302 dont_mount(new_dentry
);
3304 mutex_unlock(&target
->i_mutex
);
3305 if (d_unhashed(new_dentry
))
3306 d_rehash(new_dentry
);
3310 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3311 d_move(old_dentry
,new_dentry
);
3315 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3316 struct inode
*new_dir
, struct dentry
*new_dentry
)
3318 struct inode
*target
;
3321 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3326 target
= new_dentry
->d_inode
;
3328 mutex_lock(&target
->i_mutex
);
3329 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3332 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3335 dont_mount(new_dentry
);
3336 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3337 d_move(old_dentry
, new_dentry
);
3340 mutex_unlock(&target
->i_mutex
);
3345 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3346 struct inode
*new_dir
, struct dentry
*new_dentry
)
3349 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3350 const unsigned char *old_name
;
3352 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3355 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3359 if (!new_dentry
->d_inode
)
3360 error
= may_create(new_dir
, new_dentry
);
3362 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3366 if (!old_dir
->i_op
->rename
)
3369 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3372 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3374 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3376 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3377 new_dentry
->d_inode
, old_dentry
);
3378 fsnotify_oldname_free(old_name
);
3383 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3384 int, newdfd
, const char __user
*, newname
)
3386 struct dentry
*old_dir
, *new_dir
;
3387 struct dentry
*old_dentry
, *new_dentry
;
3388 struct dentry
*trap
;
3389 struct nameidata oldnd
, newnd
;
3394 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3398 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3403 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3406 old_dir
= oldnd
.path
.dentry
;
3408 if (oldnd
.last_type
!= LAST_NORM
)
3411 new_dir
= newnd
.path
.dentry
;
3412 if (newnd
.last_type
!= LAST_NORM
)
3415 oldnd
.flags
&= ~LOOKUP_PARENT
;
3416 newnd
.flags
&= ~LOOKUP_PARENT
;
3417 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3419 trap
= lock_rename(new_dir
, old_dir
);
3421 old_dentry
= lookup_hash(&oldnd
);
3422 error
= PTR_ERR(old_dentry
);
3423 if (IS_ERR(old_dentry
))
3425 /* source must exist */
3427 if (!old_dentry
->d_inode
)
3429 /* unless the source is a directory trailing slashes give -ENOTDIR */
3430 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3432 if (oldnd
.last
.name
[oldnd
.last
.len
])
3434 if (newnd
.last
.name
[newnd
.last
.len
])
3437 /* source should not be ancestor of target */
3439 if (old_dentry
== trap
)
3441 new_dentry
= lookup_hash(&newnd
);
3442 error
= PTR_ERR(new_dentry
);
3443 if (IS_ERR(new_dentry
))
3445 /* target should not be an ancestor of source */
3447 if (new_dentry
== trap
)
3450 error
= mnt_want_write(oldnd
.path
.mnt
);
3453 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3454 &newnd
.path
, new_dentry
);
3457 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3458 new_dir
->d_inode
, new_dentry
);
3460 mnt_drop_write(oldnd
.path
.mnt
);
3466 unlock_rename(new_dir
, old_dir
);
3468 path_put(&newnd
.path
);
3471 path_put(&oldnd
.path
);
3477 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3479 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3482 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3486 len
= PTR_ERR(link
);
3491 if (len
> (unsigned) buflen
)
3493 if (copy_to_user(buffer
, link
, len
))
3500 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3501 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3502 * using) it for any given inode is up to filesystem.
3504 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3506 struct nameidata nd
;
3511 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3513 return PTR_ERR(cookie
);
3515 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3516 if (dentry
->d_inode
->i_op
->put_link
)
3517 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3521 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3523 return __vfs_follow_link(nd
, link
);
3526 /* get the link contents into pagecache */
3527 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3531 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3532 page
= read_mapping_page(mapping
, 0, NULL
);
3537 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3541 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3543 struct page
*page
= NULL
;
3544 char *s
= page_getlink(dentry
, &page
);
3545 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3548 page_cache_release(page
);
3553 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3555 struct page
*page
= NULL
;
3556 nd_set_link(nd
, page_getlink(dentry
, &page
));
3560 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3562 struct page
*page
= cookie
;
3566 page_cache_release(page
);
3571 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3573 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3575 struct address_space
*mapping
= inode
->i_mapping
;
3580 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3582 flags
|= AOP_FLAG_NOFS
;
3585 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3586 flags
, &page
, &fsdata
);
3590 kaddr
= kmap_atomic(page
, KM_USER0
);
3591 memcpy(kaddr
, symname
, len
-1);
3592 kunmap_atomic(kaddr
, KM_USER0
);
3594 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3601 mark_inode_dirty(inode
);
3607 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3609 return __page_symlink(inode
, symname
, len
,
3610 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3613 const struct inode_operations page_symlink_inode_operations
= {
3614 .readlink
= generic_readlink
,
3615 .follow_link
= page_follow_link_light
,
3616 .put_link
= page_put_link
,
3619 EXPORT_SYMBOL(user_path_at
);
3620 EXPORT_SYMBOL(follow_down_one
);
3621 EXPORT_SYMBOL(follow_down
);
3622 EXPORT_SYMBOL(follow_up
);
3623 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3624 EXPORT_SYMBOL(getname
);
3625 EXPORT_SYMBOL(lock_rename
);
3626 EXPORT_SYMBOL(lookup_one_len
);
3627 EXPORT_SYMBOL(page_follow_link_light
);
3628 EXPORT_SYMBOL(page_put_link
);
3629 EXPORT_SYMBOL(page_readlink
);
3630 EXPORT_SYMBOL(__page_symlink
);
3631 EXPORT_SYMBOL(page_symlink
);
3632 EXPORT_SYMBOL(page_symlink_inode_operations
);
3633 EXPORT_SYMBOL(path_lookup
);
3634 EXPORT_SYMBOL(kern_path
);
3635 EXPORT_SYMBOL(vfs_path_lookup
);
3636 EXPORT_SYMBOL(inode_permission
);
3637 EXPORT_SYMBOL(file_permission
);
3638 EXPORT_SYMBOL(unlock_rename
);
3639 EXPORT_SYMBOL(vfs_create
);
3640 EXPORT_SYMBOL(vfs_follow_link
);
3641 EXPORT_SYMBOL(vfs_link
);
3642 EXPORT_SYMBOL(vfs_mkdir
);
3643 EXPORT_SYMBOL(vfs_mknod
);
3644 EXPORT_SYMBOL(generic_permission
);
3645 EXPORT_SYMBOL(vfs_readlink
);
3646 EXPORT_SYMBOL(vfs_rename
);
3647 EXPORT_SYMBOL(vfs_rmdir
);
3648 EXPORT_SYMBOL(vfs_symlink
);
3649 EXPORT_SYMBOL(vfs_unlink
);
3650 EXPORT_SYMBOL(dentry_unhash
);
3651 EXPORT_SYMBOL(generic_readlink
);