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/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.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 <asm/uaccess.h>
36 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
38 /* [Feb-1997 T. Schoebel-Theuer]
39 * Fundamental changes in the pathname lookup mechanisms (namei)
40 * were necessary because of omirr. The reason is that omirr needs
41 * to know the _real_ pathname, not the user-supplied one, in case
42 * of symlinks (and also when transname replacements occur).
44 * The new code replaces the old recursive symlink resolution with
45 * an iterative one (in case of non-nested symlink chains). It does
46 * this with calls to <fs>_follow_link().
47 * As a side effect, dir_namei(), _namei() and follow_link() are now
48 * replaced with a single function lookup_dentry() that can handle all
49 * the special cases of the former code.
51 * With the new dcache, the pathname is stored at each inode, at least as
52 * long as the refcount of the inode is positive. As a side effect, the
53 * size of the dcache depends on the inode cache and thus is dynamic.
55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
56 * resolution to correspond with current state of the code.
58 * Note that the symlink resolution is not *completely* iterative.
59 * There is still a significant amount of tail- and mid- recursion in
60 * the algorithm. Also, note that <fs>_readlink() is not used in
61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
62 * may return different results than <fs>_follow_link(). Many virtual
63 * filesystems (including /proc) exhibit this behavior.
66 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
68 * and the name already exists in form of a symlink, try to create the new
69 * name indicated by the symlink. The old code always complained that the
70 * name already exists, due to not following the symlink even if its target
71 * is nonexistent. The new semantics affects also mknod() and link() when
72 * the name is a symlink pointing to a non-existant name.
74 * I don't know which semantics is the right one, since I have no access
75 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
77 * "old" one. Personally, I think the new semantics is much more logical.
78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
79 * file does succeed in both HP-UX and SunOs, but not in Solaris
80 * and in the old Linux semantics.
83 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
84 * semantics. See the comments in "open_namei" and "do_link" below.
86 * [10-Sep-98 Alan Modra] Another symlink change.
89 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
90 * inside the path - always follow.
91 * in the last component in creation/removal/renaming - never follow.
92 * if LOOKUP_FOLLOW passed - follow.
93 * if the pathname has trailing slashes - follow.
94 * otherwise - don't follow.
95 * (applied in that order).
97 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
98 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
99 * During the 2.4 we need to fix the userland stuff depending on it -
100 * hopefully we will be able to get rid of that wart in 2.5. So far only
101 * XEmacs seems to be relying on it...
104 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
105 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
106 * any extra contention...
109 static int __link_path_walk(const char *name
, struct nameidata
*nd
);
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user
*filename
, char *page
)
121 unsigned long len
= PATH_MAX
;
123 if (!segment_eq(get_fs(), KERNEL_DS
)) {
124 if ((unsigned long) filename
>= TASK_SIZE
)
126 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
127 len
= TASK_SIZE
- (unsigned long) filename
;
130 retval
= strncpy_from_user(page
, filename
, len
);
134 return -ENAMETOOLONG
;
140 char * getname(const char __user
* filename
)
144 result
= ERR_PTR(-ENOMEM
);
147 int retval
= do_getname(filename
, tmp
);
152 result
= ERR_PTR(retval
);
155 audit_getname(result
);
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name
)
162 if (unlikely(!audit_dummy_context()))
167 EXPORT_SYMBOL(putname
);
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
182 int generic_permission(struct inode
*inode
, int mask
,
183 int (*check_acl
)(struct inode
*inode
, int mask
))
185 umode_t mode
= inode
->i_mode
;
187 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
189 if (current
->fsuid
== inode
->i_uid
)
192 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
193 int error
= check_acl(inode
, mask
);
194 if (error
== -EACCES
)
195 goto check_capabilities
;
196 else if (error
!= -EAGAIN
)
200 if (in_group_p(inode
->i_gid
))
205 * If the DACs are ok we don't need any capability check.
207 if ((mask
& ~mode
) == 0)
212 * Read/write DACs are always overridable.
213 * Executable DACs are overridable if at least one exec bit is set.
215 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
216 if (capable(CAP_DAC_OVERRIDE
))
220 * Searching includes executable on directories, else just read.
222 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
223 if (capable(CAP_DAC_READ_SEARCH
))
229 int inode_permission(struct inode
*inode
, int mask
)
233 if (mask
& MAY_WRITE
) {
234 umode_t mode
= inode
->i_mode
;
237 * Nobody gets write access to a read-only fs.
239 if (IS_RDONLY(inode
) &&
240 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
244 * Nobody gets write access to an immutable file.
246 if (IS_IMMUTABLE(inode
))
250 /* Ordinary permission routines do not understand MAY_APPEND. */
251 if (inode
->i_op
&& inode
->i_op
->permission
)
252 retval
= inode
->i_op
->permission(inode
, mask
);
254 retval
= generic_permission(inode
, mask
, NULL
);
259 retval
= devcgroup_inode_permission(inode
, mask
);
263 return security_inode_permission(inode
,
264 mask
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
|MAY_APPEND
));
268 * vfs_permission - check for access rights to a given path
269 * @nd: lookup result that describes the path
270 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
272 * Used to check for read/write/execute permissions on a path.
273 * We use "fsuid" for this, letting us set arbitrary permissions
274 * for filesystem access without changing the "normal" uids which
275 * are used for other things.
277 int vfs_permission(struct nameidata
*nd
, int mask
)
279 return inode_permission(nd
->path
.dentry
->d_inode
, mask
);
283 * file_permission - check for additional access rights to a given file
284 * @file: file to check access rights for
285 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
287 * Used to check for read/write/execute permissions on an already opened
291 * Do not use this function in new code. All access checks should
292 * be done using vfs_permission().
294 int file_permission(struct file
*file
, int mask
)
296 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
300 * get_write_access() gets write permission for a file.
301 * put_write_access() releases this write permission.
302 * This is used for regular files.
303 * We cannot support write (and maybe mmap read-write shared) accesses and
304 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
305 * can have the following values:
306 * 0: no writers, no VM_DENYWRITE mappings
307 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
308 * > 0: (i_writecount) users are writing to the file.
310 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
311 * except for the cases where we don't hold i_writecount yet. Then we need to
312 * use {get,deny}_write_access() - these functions check the sign and refuse
313 * to do the change if sign is wrong. Exclusion between them is provided by
314 * the inode->i_lock spinlock.
317 int get_write_access(struct inode
* inode
)
319 spin_lock(&inode
->i_lock
);
320 if (atomic_read(&inode
->i_writecount
) < 0) {
321 spin_unlock(&inode
->i_lock
);
324 atomic_inc(&inode
->i_writecount
);
325 spin_unlock(&inode
->i_lock
);
330 int deny_write_access(struct file
* file
)
332 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
334 spin_lock(&inode
->i_lock
);
335 if (atomic_read(&inode
->i_writecount
) > 0) {
336 spin_unlock(&inode
->i_lock
);
339 atomic_dec(&inode
->i_writecount
);
340 spin_unlock(&inode
->i_lock
);
346 * path_get - get a reference to a path
347 * @path: path to get the reference to
349 * Given a path increment the reference count to the dentry and the vfsmount.
351 void path_get(struct path
*path
)
356 EXPORT_SYMBOL(path_get
);
359 * path_put - put a reference to a path
360 * @path: path to put the reference to
362 * Given a path decrement the reference count to the dentry and the vfsmount.
364 void path_put(struct path
*path
)
369 EXPORT_SYMBOL(path_put
);
372 * release_open_intent - free up open intent resources
373 * @nd: pointer to nameidata
375 void release_open_intent(struct nameidata
*nd
)
377 if (nd
->intent
.open
.file
->f_path
.dentry
== NULL
)
378 put_filp(nd
->intent
.open
.file
);
380 fput(nd
->intent
.open
.file
);
382 EXPORT_SYMBOL_GPL(release_open_intent
);
384 static inline struct dentry
*
385 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
387 int status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
388 if (unlikely(status
<= 0)) {
390 * The dentry failed validation.
391 * If d_revalidate returned 0 attempt to invalidate
392 * the dentry otherwise d_revalidate is asking us
393 * to return a fail status.
396 if (!d_invalidate(dentry
)) {
402 dentry
= ERR_PTR(status
);
409 * Internal lookup() using the new generic dcache.
412 static struct dentry
* cached_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
414 struct dentry
* dentry
= __d_lookup(parent
, name
);
416 /* lockess __d_lookup may fail due to concurrent d_move()
417 * in some unrelated directory, so try with d_lookup
420 dentry
= d_lookup(parent
, name
);
422 if (dentry
&& dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
423 dentry
= do_revalidate(dentry
, nd
);
429 * Short-cut version of permission(), for calling by
430 * path_walk(), when dcache lock is held. Combines parts
431 * of permission() and generic_permission(), and tests ONLY for
432 * MAY_EXEC permission.
434 * If appropriate, check DAC only. If not appropriate, or
435 * short-cut DAC fails, then call permission() to do more
436 * complete permission check.
438 static int exec_permission_lite(struct inode
*inode
)
440 umode_t mode
= inode
->i_mode
;
442 if (inode
->i_op
&& inode
->i_op
->permission
)
445 if (current
->fsuid
== inode
->i_uid
)
447 else if (in_group_p(inode
->i_gid
))
453 if ((inode
->i_mode
& S_IXUGO
) && capable(CAP_DAC_OVERRIDE
))
456 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_OVERRIDE
))
459 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_READ_SEARCH
))
464 return security_inode_permission(inode
, MAY_EXEC
);
468 * This is called when everything else fails, and we actually have
469 * to go to the low-level filesystem to find out what we should do..
471 * We get the directory semaphore, and after getting that we also
472 * make sure that nobody added the entry to the dcache in the meantime..
475 static struct dentry
* real_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
477 struct dentry
* result
;
478 struct inode
*dir
= parent
->d_inode
;
480 mutex_lock(&dir
->i_mutex
);
482 * First re-do the cached lookup just in case it was created
483 * while we waited for the directory semaphore..
485 * FIXME! This could use version numbering or similar to
486 * avoid unnecessary cache lookups.
488 * The "dcache_lock" is purely to protect the RCU list walker
489 * from concurrent renames at this point (we mustn't get false
490 * negatives from the RCU list walk here, unlike the optimistic
493 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
495 result
= d_lookup(parent
, name
);
497 struct dentry
*dentry
;
499 /* Don't create child dentry for a dead directory. */
500 result
= ERR_PTR(-ENOENT
);
504 dentry
= d_alloc(parent
, name
);
505 result
= ERR_PTR(-ENOMEM
);
507 result
= dir
->i_op
->lookup(dir
, dentry
, nd
);
514 mutex_unlock(&dir
->i_mutex
);
519 * Uhhuh! Nasty case: the cache was re-populated while
520 * we waited on the semaphore. Need to revalidate.
522 mutex_unlock(&dir
->i_mutex
);
523 if (result
->d_op
&& result
->d_op
->d_revalidate
) {
524 result
= do_revalidate(result
, nd
);
526 result
= ERR_PTR(-ENOENT
);
532 static __always_inline
void
533 walk_init_root(const char *name
, struct nameidata
*nd
)
535 struct fs_struct
*fs
= current
->fs
;
537 read_lock(&fs
->lock
);
540 read_unlock(&fs
->lock
);
544 * Wrapper to retry pathname resolution whenever the underlying
545 * file system returns an ESTALE.
547 * Retry the whole path once, forcing real lookup requests
548 * instead of relying on the dcache.
550 static __always_inline
int link_path_walk(const char *name
, struct nameidata
*nd
)
552 struct path save
= nd
->path
;
555 /* make sure the stuff we saved doesn't go away */
558 result
= __link_path_walk(name
, nd
);
559 if (result
== -ESTALE
) {
560 /* nd->path had been dropped */
563 nd
->flags
|= LOOKUP_REVAL
;
564 result
= __link_path_walk(name
, nd
);
572 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
581 walk_init_root(link
, nd
);
583 res
= link_path_walk(link
, nd
);
584 if (nd
->depth
|| res
|| nd
->last_type
!=LAST_NORM
)
587 * If it is an iterative symlinks resolution in open_namei() we
588 * have to copy the last component. And all that crap because of
589 * bloody create() on broken symlinks. Furrfu...
592 if (unlikely(!name
)) {
596 strcpy(name
, nd
->last
.name
);
597 nd
->last
.name
= name
;
601 return PTR_ERR(link
);
604 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
607 if (path
->mnt
!= nd
->path
.mnt
)
611 static inline void path_to_nameidata(struct path
*path
, struct nameidata
*nd
)
613 dput(nd
->path
.dentry
);
614 if (nd
->path
.mnt
!= path
->mnt
)
615 mntput(nd
->path
.mnt
);
616 nd
->path
.mnt
= path
->mnt
;
617 nd
->path
.dentry
= path
->dentry
;
620 static __always_inline
int __do_follow_link(struct path
*path
, struct nameidata
*nd
)
624 struct dentry
*dentry
= path
->dentry
;
626 touch_atime(path
->mnt
, dentry
);
627 nd_set_link(nd
, NULL
);
629 if (path
->mnt
!= nd
->path
.mnt
) {
630 path_to_nameidata(path
, nd
);
634 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
635 error
= PTR_ERR(cookie
);
636 if (!IS_ERR(cookie
)) {
637 char *s
= nd_get_link(nd
);
640 error
= __vfs_follow_link(nd
, s
);
641 if (dentry
->d_inode
->i_op
->put_link
)
642 dentry
->d_inode
->i_op
->put_link(dentry
, nd
, cookie
);
650 * This limits recursive symlink follows to 8, while
651 * limiting consecutive symlinks to 40.
653 * Without that kind of total limit, nasty chains of consecutive
654 * symlinks can cause almost arbitrarily long lookups.
656 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
659 if (current
->link_count
>= MAX_NESTED_LINKS
)
661 if (current
->total_link_count
>= 40)
663 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
665 err
= security_inode_follow_link(path
->dentry
, nd
);
668 current
->link_count
++;
669 current
->total_link_count
++;
671 err
= __do_follow_link(path
, nd
);
672 current
->link_count
--;
676 path_put_conditional(path
, nd
);
681 int follow_up(struct vfsmount
**mnt
, struct dentry
**dentry
)
683 struct vfsmount
*parent
;
684 struct dentry
*mountpoint
;
685 spin_lock(&vfsmount_lock
);
686 parent
=(*mnt
)->mnt_parent
;
687 if (parent
== *mnt
) {
688 spin_unlock(&vfsmount_lock
);
692 mountpoint
=dget((*mnt
)->mnt_mountpoint
);
693 spin_unlock(&vfsmount_lock
);
695 *dentry
= mountpoint
;
701 /* no need for dcache_lock, as serialization is taken care in
704 static int __follow_mount(struct path
*path
)
707 while (d_mountpoint(path
->dentry
)) {
708 struct vfsmount
*mounted
= lookup_mnt(path
->mnt
, path
->dentry
);
715 path
->dentry
= dget(mounted
->mnt_root
);
721 static void follow_mount(struct vfsmount
**mnt
, struct dentry
**dentry
)
723 while (d_mountpoint(*dentry
)) {
724 struct vfsmount
*mounted
= lookup_mnt(*mnt
, *dentry
);
730 *dentry
= dget(mounted
->mnt_root
);
734 /* no need for dcache_lock, as serialization is taken care in
737 int follow_down(struct vfsmount
**mnt
, struct dentry
**dentry
)
739 struct vfsmount
*mounted
;
741 mounted
= lookup_mnt(*mnt
, *dentry
);
746 *dentry
= dget(mounted
->mnt_root
);
752 static __always_inline
void follow_dotdot(struct nameidata
*nd
)
754 struct fs_struct
*fs
= current
->fs
;
757 struct vfsmount
*parent
;
758 struct dentry
*old
= nd
->path
.dentry
;
760 read_lock(&fs
->lock
);
761 if (nd
->path
.dentry
== fs
->root
.dentry
&&
762 nd
->path
.mnt
== fs
->root
.mnt
) {
763 read_unlock(&fs
->lock
);
766 read_unlock(&fs
->lock
);
767 spin_lock(&dcache_lock
);
768 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
769 nd
->path
.dentry
= dget(nd
->path
.dentry
->d_parent
);
770 spin_unlock(&dcache_lock
);
774 spin_unlock(&dcache_lock
);
775 spin_lock(&vfsmount_lock
);
776 parent
= nd
->path
.mnt
->mnt_parent
;
777 if (parent
== nd
->path
.mnt
) {
778 spin_unlock(&vfsmount_lock
);
782 nd
->path
.dentry
= dget(nd
->path
.mnt
->mnt_mountpoint
);
783 spin_unlock(&vfsmount_lock
);
785 mntput(nd
->path
.mnt
);
786 nd
->path
.mnt
= parent
;
788 follow_mount(&nd
->path
.mnt
, &nd
->path
.dentry
);
792 * It's more convoluted than I'd like it to be, but... it's still fairly
793 * small and for now I'd prefer to have fast path as straight as possible.
794 * It _is_ time-critical.
796 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
799 struct vfsmount
*mnt
= nd
->path
.mnt
;
800 struct dentry
*dentry
= __d_lookup(nd
->path
.dentry
, name
);
804 if (dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
805 goto need_revalidate
;
808 path
->dentry
= dentry
;
809 __follow_mount(path
);
813 dentry
= real_lookup(nd
->path
.dentry
, name
, nd
);
819 dentry
= do_revalidate(dentry
, nd
);
827 return PTR_ERR(dentry
);
832 * This is the basic name resolution function, turning a pathname into
833 * the final dentry. We expect 'base' to be positive and a directory.
835 * Returns 0 and nd will have valid dentry and mnt on success.
836 * Returns error and drops reference to input namei data on failure.
838 static int __link_path_walk(const char *name
, struct nameidata
*nd
)
843 unsigned int lookup_flags
= nd
->flags
;
850 inode
= nd
->path
.dentry
->d_inode
;
852 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
854 /* At this point we know we have a real path component. */
860 nd
->flags
|= LOOKUP_CONTINUE
;
861 err
= exec_permission_lite(inode
);
863 err
= vfs_permission(nd
, MAY_EXEC
);
868 c
= *(const unsigned char *)name
;
870 hash
= init_name_hash();
873 hash
= partial_name_hash(c
, hash
);
874 c
= *(const unsigned char *)name
;
875 } while (c
&& (c
!= '/'));
876 this.len
= name
- (const char *) this.name
;
877 this.hash
= end_name_hash(hash
);
879 /* remove trailing slashes? */
882 while (*++name
== '/');
884 goto last_with_slashes
;
887 * "." and ".." are special - ".." especially so because it has
888 * to be able to know about the current root directory and
889 * parent relationships.
891 if (this.name
[0] == '.') switch (this.len
) {
895 if (this.name
[1] != '.')
898 inode
= nd
->path
.dentry
->d_inode
;
904 * See if the low-level filesystem might want
905 * to use its own hash..
907 if (nd
->path
.dentry
->d_op
&& nd
->path
.dentry
->d_op
->d_hash
) {
908 err
= nd
->path
.dentry
->d_op
->d_hash(nd
->path
.dentry
,
913 /* This does the actual lookups.. */
914 err
= do_lookup(nd
, &this, &next
);
919 inode
= next
.dentry
->d_inode
;
926 if (inode
->i_op
->follow_link
) {
927 err
= do_follow_link(&next
, nd
);
931 inode
= nd
->path
.dentry
->d_inode
;
938 path_to_nameidata(&next
, nd
);
940 if (!inode
->i_op
->lookup
)
943 /* here ends the main loop */
946 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
948 /* Clear LOOKUP_CONTINUE iff it was previously unset */
949 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
950 if (lookup_flags
& LOOKUP_PARENT
)
952 if (this.name
[0] == '.') switch (this.len
) {
956 if (this.name
[1] != '.')
959 inode
= nd
->path
.dentry
->d_inode
;
964 if (nd
->path
.dentry
->d_op
&& nd
->path
.dentry
->d_op
->d_hash
) {
965 err
= nd
->path
.dentry
->d_op
->d_hash(nd
->path
.dentry
,
970 err
= do_lookup(nd
, &this, &next
);
973 inode
= next
.dentry
->d_inode
;
974 if ((lookup_flags
& LOOKUP_FOLLOW
)
975 && inode
&& inode
->i_op
&& inode
->i_op
->follow_link
) {
976 err
= do_follow_link(&next
, nd
);
979 inode
= nd
->path
.dentry
->d_inode
;
981 path_to_nameidata(&next
, nd
);
985 if (lookup_flags
& LOOKUP_DIRECTORY
) {
987 if (!inode
->i_op
|| !inode
->i_op
->lookup
)
993 nd
->last_type
= LAST_NORM
;
994 if (this.name
[0] != '.')
997 nd
->last_type
= LAST_DOT
;
998 else if (this.len
== 2 && this.name
[1] == '.')
999 nd
->last_type
= LAST_DOTDOT
;
1004 * We bypassed the ordinary revalidation routines.
1005 * We may need to check the cached dentry for staleness.
1007 if (nd
->path
.dentry
&& nd
->path
.dentry
->d_sb
&&
1008 (nd
->path
.dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)) {
1010 /* Note: we do not d_invalidate() */
1011 if (!nd
->path
.dentry
->d_op
->d_revalidate(
1012 nd
->path
.dentry
, nd
))
1018 path_put_conditional(&next
, nd
);
1021 path_put(&nd
->path
);
1026 static int path_walk(const char *name
, struct nameidata
*nd
)
1028 current
->total_link_count
= 0;
1029 return link_path_walk(name
, nd
);
1032 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1033 static int do_path_lookup(int dfd
, const char *name
,
1034 unsigned int flags
, struct nameidata
*nd
)
1039 struct fs_struct
*fs
= current
->fs
;
1041 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1046 read_lock(&fs
->lock
);
1047 nd
->path
= fs
->root
;
1048 path_get(&fs
->root
);
1049 read_unlock(&fs
->lock
);
1050 } else if (dfd
== AT_FDCWD
) {
1051 read_lock(&fs
->lock
);
1054 read_unlock(&fs
->lock
);
1056 struct dentry
*dentry
;
1058 file
= fget_light(dfd
, &fput_needed
);
1063 dentry
= file
->f_path
.dentry
;
1066 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1069 retval
= file_permission(file
, MAY_EXEC
);
1073 nd
->path
= file
->f_path
;
1074 path_get(&file
->f_path
);
1076 fput_light(file
, fput_needed
);
1079 retval
= path_walk(name
, nd
);
1080 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1081 nd
->path
.dentry
->d_inode
))
1082 audit_inode(name
, nd
->path
.dentry
);
1087 fput_light(file
, fput_needed
);
1091 int path_lookup(const char *name
, unsigned int flags
,
1092 struct nameidata
*nd
)
1094 return do_path_lookup(AT_FDCWD
, name
, flags
, nd
);
1097 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1099 struct nameidata nd
;
1100 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1107 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1108 * @dentry: pointer to dentry of the base directory
1109 * @mnt: pointer to vfs mount of the base directory
1110 * @name: pointer to file name
1111 * @flags: lookup flags
1112 * @nd: pointer to nameidata
1114 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1115 const char *name
, unsigned int flags
,
1116 struct nameidata
*nd
)
1120 /* same as do_path_lookup */
1121 nd
->last_type
= LAST_ROOT
;
1125 nd
->path
.dentry
= dentry
;
1127 path_get(&nd
->path
);
1129 retval
= path_walk(name
, nd
);
1130 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1131 nd
->path
.dentry
->d_inode
))
1132 audit_inode(name
, nd
->path
.dentry
);
1139 * path_lookup_open - lookup a file path with open intent
1140 * @dfd: the directory to use as base, or AT_FDCWD
1141 * @name: pointer to file name
1142 * @lookup_flags: lookup intent flags
1143 * @nd: pointer to nameidata
1144 * @open_flags: open intent flags
1146 int path_lookup_open(int dfd
, const char *name
, unsigned int lookup_flags
,
1147 struct nameidata
*nd
, int open_flags
)
1149 struct file
*filp
= get_empty_filp();
1154 nd
->intent
.open
.file
= filp
;
1155 nd
->intent
.open
.flags
= open_flags
;
1156 nd
->intent
.open
.create_mode
= 0;
1157 err
= do_path_lookup(dfd
, name
, lookup_flags
|LOOKUP_OPEN
, nd
);
1158 if (IS_ERR(nd
->intent
.open
.file
)) {
1160 err
= PTR_ERR(nd
->intent
.open
.file
);
1161 path_put(&nd
->path
);
1163 } else if (err
!= 0)
1164 release_open_intent(nd
);
1168 struct dentry
*__lookup_hash(struct qstr
*name
, struct dentry
*base
, struct nameidata
*nd
)
1170 struct dentry
*dentry
;
1171 struct inode
*inode
;
1174 inode
= base
->d_inode
;
1177 * See if the low-level filesystem might want
1178 * to use its own hash..
1180 if (base
->d_op
&& base
->d_op
->d_hash
) {
1181 err
= base
->d_op
->d_hash(base
, name
);
1182 dentry
= ERR_PTR(err
);
1187 dentry
= cached_lookup(base
, name
, nd
);
1191 /* Don't create child dentry for a dead directory. */
1192 dentry
= ERR_PTR(-ENOENT
);
1193 if (IS_DEADDIR(inode
))
1196 new = d_alloc(base
, name
);
1197 dentry
= ERR_PTR(-ENOMEM
);
1200 dentry
= inode
->i_op
->lookup(inode
, new, nd
);
1211 * Restricted form of lookup. Doesn't follow links, single-component only,
1212 * needs parent already locked. Doesn't follow mounts.
1215 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1219 err
= inode_permission(nd
->path
.dentry
->d_inode
, MAY_EXEC
);
1221 return ERR_PTR(err
);
1222 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1225 static int __lookup_one_len(const char *name
, struct qstr
*this,
1226 struct dentry
*base
, int len
)
1236 hash
= init_name_hash();
1238 c
= *(const unsigned char *)name
++;
1239 if (c
== '/' || c
== '\0')
1241 hash
= partial_name_hash(c
, hash
);
1243 this->hash
= end_name_hash(hash
);
1248 * lookup_one_len - filesystem helper to lookup single pathname component
1249 * @name: pathname component to lookup
1250 * @base: base directory to lookup from
1251 * @len: maximum length @len should be interpreted to
1253 * Note that this routine is purely a helper for filesystem usage and should
1254 * not be called by generic code. Also note that by using this function the
1255 * nameidata argument is passed to the filesystem methods and a filesystem
1256 * using this helper needs to be prepared for that.
1258 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1263 err
= __lookup_one_len(name
, &this, base
, len
);
1265 return ERR_PTR(err
);
1267 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
1269 return ERR_PTR(err
);
1270 return __lookup_hash(&this, base
, NULL
);
1274 * lookup_one_noperm - bad hack for sysfs
1275 * @name: pathname component to lookup
1276 * @base: base directory to lookup from
1278 * This is a variant of lookup_one_len that doesn't perform any permission
1279 * checks. It's a horrible hack to work around the braindead sysfs
1280 * architecture and should not be used anywhere else.
1282 * DON'T USE THIS FUNCTION EVER, thanks.
1284 struct dentry
*lookup_one_noperm(const char *name
, struct dentry
*base
)
1289 err
= __lookup_one_len(name
, &this, base
, strlen(name
));
1291 return ERR_PTR(err
);
1292 return __lookup_hash(&this, base
, NULL
);
1295 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1298 struct nameidata nd
;
1299 char *tmp
= getname(name
);
1300 int err
= PTR_ERR(tmp
);
1303 BUG_ON(flags
& LOOKUP_PARENT
);
1305 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1313 static int user_path_parent(int dfd
, const char __user
*path
,
1314 struct nameidata
*nd
, char **name
)
1316 char *s
= getname(path
);
1322 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1332 * It's inline, so penalty for filesystems that don't use sticky bit is
1335 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1337 if (!(dir
->i_mode
& S_ISVTX
))
1339 if (inode
->i_uid
== current
->fsuid
)
1341 if (dir
->i_uid
== current
->fsuid
)
1343 return !capable(CAP_FOWNER
);
1347 * Check whether we can remove a link victim from directory dir, check
1348 * whether the type of victim is right.
1349 * 1. We can't do it if dir is read-only (done in permission())
1350 * 2. We should have write and exec permissions on dir
1351 * 3. We can't remove anything from append-only dir
1352 * 4. We can't do anything with immutable dir (done in permission())
1353 * 5. If the sticky bit on dir is set we should either
1354 * a. be owner of dir, or
1355 * b. be owner of victim, or
1356 * c. have CAP_FOWNER capability
1357 * 6. If the victim is append-only or immutable we can't do antyhing with
1358 * links pointing to it.
1359 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1360 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1361 * 9. We can't remove a root or mountpoint.
1362 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1363 * nfs_async_unlink().
1365 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1369 if (!victim
->d_inode
)
1372 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1373 audit_inode_child(victim
->d_name
.name
, victim
, dir
);
1375 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1380 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1381 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1384 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1386 if (IS_ROOT(victim
))
1388 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1390 if (IS_DEADDIR(dir
))
1392 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1397 /* Check whether we can create an object with dentry child in directory
1399 * 1. We can't do it if child already exists (open has special treatment for
1400 * this case, but since we are inlined it's OK)
1401 * 2. We can't do it if dir is read-only (done in permission())
1402 * 3. We should have write and exec permissions on dir
1403 * 4. We can't do it if dir is immutable (done in permission())
1405 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1409 if (IS_DEADDIR(dir
))
1411 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1415 * O_DIRECTORY translates into forcing a directory lookup.
1417 static inline int lookup_flags(unsigned int f
)
1419 unsigned long retval
= LOOKUP_FOLLOW
;
1422 retval
&= ~LOOKUP_FOLLOW
;
1424 if (f
& O_DIRECTORY
)
1425 retval
|= LOOKUP_DIRECTORY
;
1431 * p1 and p2 should be directories on the same fs.
1433 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1438 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1442 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1444 p
= d_ancestor(p2
, p1
);
1446 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1447 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1451 p
= d_ancestor(p1
, p2
);
1453 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1454 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1458 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1459 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1463 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1465 mutex_unlock(&p1
->d_inode
->i_mutex
);
1467 mutex_unlock(&p2
->d_inode
->i_mutex
);
1468 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1472 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1473 struct nameidata
*nd
)
1475 int error
= may_create(dir
, dentry
);
1480 if (!dir
->i_op
|| !dir
->i_op
->create
)
1481 return -EACCES
; /* shouldn't it be ENOSYS? */
1484 error
= security_inode_create(dir
, dentry
, mode
);
1488 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1490 fsnotify_create(dir
, dentry
);
1494 int may_open(struct nameidata
*nd
, int acc_mode
, int flag
)
1496 struct dentry
*dentry
= nd
->path
.dentry
;
1497 struct inode
*inode
= dentry
->d_inode
;
1503 if (S_ISLNK(inode
->i_mode
))
1506 if (S_ISDIR(inode
->i_mode
) && (acc_mode
& MAY_WRITE
))
1510 * FIFO's, sockets and device files are special: they don't
1511 * actually live on the filesystem itself, and as such you
1512 * can write to them even if the filesystem is read-only.
1514 if (S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
1516 } else if (S_ISBLK(inode
->i_mode
) || S_ISCHR(inode
->i_mode
)) {
1517 if (nd
->path
.mnt
->mnt_flags
& MNT_NODEV
)
1523 error
= vfs_permission(nd
, acc_mode
);
1527 * An append-only file must be opened in append mode for writing.
1529 if (IS_APPEND(inode
)) {
1530 if ((flag
& FMODE_WRITE
) && !(flag
& O_APPEND
))
1536 /* O_NOATIME can only be set by the owner or superuser */
1537 if (flag
& O_NOATIME
)
1538 if (!is_owner_or_cap(inode
))
1542 * Ensure there are no outstanding leases on the file.
1544 error
= break_lease(inode
, flag
);
1548 if (flag
& O_TRUNC
) {
1549 error
= get_write_access(inode
);
1554 * Refuse to truncate files with mandatory locks held on them.
1556 error
= locks_verify_locked(inode
);
1560 error
= do_truncate(dentry
, 0,
1561 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
1564 put_write_access(inode
);
1568 if (flag
& FMODE_WRITE
)
1575 * Be careful about ever adding any more callers of this
1576 * function. Its flags must be in the namei format, not
1577 * what get passed to sys_open().
1579 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
1583 struct dentry
*dir
= nd
->path
.dentry
;
1585 if (!IS_POSIXACL(dir
->d_inode
))
1586 mode
&= ~current
->fs
->umask
;
1587 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
1588 mutex_unlock(&dir
->d_inode
->i_mutex
);
1589 dput(nd
->path
.dentry
);
1590 nd
->path
.dentry
= path
->dentry
;
1593 /* Don't check for write permission, don't truncate */
1594 return may_open(nd
, 0, flag
& ~O_TRUNC
);
1598 * Note that while the flag value (low two bits) for sys_open means:
1603 * it is changed into
1604 * 00 - no permissions needed
1605 * 01 - read-permission
1606 * 10 - write-permission
1608 * for the internal routines (ie open_namei()/follow_link() etc)
1609 * This is more logical, and also allows the 00 "no perm needed"
1610 * to be used for symlinks (where the permissions are checked
1614 static inline int open_to_namei_flags(int flag
)
1616 if ((flag
+1) & O_ACCMODE
)
1621 static int open_will_write_to_fs(int flag
, struct inode
*inode
)
1624 * We'll never write to the fs underlying
1627 if (special_file(inode
->i_mode
))
1629 return (flag
& O_TRUNC
);
1633 * Note that the low bits of the passed in "open_flag"
1634 * are not the same as in the local variable "flag". See
1635 * open_to_namei_flags() for more details.
1637 struct file
*do_filp_open(int dfd
, const char *pathname
,
1638 int open_flag
, int mode
)
1641 struct nameidata nd
;
1642 int acc_mode
, error
;
1647 int flag
= open_to_namei_flags(open_flag
);
1649 acc_mode
= MAY_OPEN
| ACC_MODE(flag
);
1651 /* O_TRUNC implies we need access checks for write permissions */
1653 acc_mode
|= MAY_WRITE
;
1655 /* Allow the LSM permission hook to distinguish append
1656 access from general write access. */
1657 if (flag
& O_APPEND
)
1658 acc_mode
|= MAY_APPEND
;
1661 * The simplest case - just a plain lookup.
1663 if (!(flag
& O_CREAT
)) {
1664 error
= path_lookup_open(dfd
, pathname
, lookup_flags(flag
),
1667 return ERR_PTR(error
);
1672 * Create - we need to know the parent.
1674 error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
, &nd
);
1676 return ERR_PTR(error
);
1679 * We have the parent and last component. First of all, check
1680 * that we are not asked to creat(2) an obvious directory - that
1684 if (nd
.last_type
!= LAST_NORM
|| nd
.last
.name
[nd
.last
.len
])
1688 filp
= get_empty_filp();
1691 nd
.intent
.open
.file
= filp
;
1692 nd
.intent
.open
.flags
= flag
;
1693 nd
.intent
.open
.create_mode
= mode
;
1694 dir
= nd
.path
.dentry
;
1695 nd
.flags
&= ~LOOKUP_PARENT
;
1696 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_OPEN
;
1698 nd
.flags
|= LOOKUP_EXCL
;
1699 mutex_lock(&dir
->d_inode
->i_mutex
);
1700 path
.dentry
= lookup_hash(&nd
);
1701 path
.mnt
= nd
.path
.mnt
;
1704 error
= PTR_ERR(path
.dentry
);
1705 if (IS_ERR(path
.dentry
)) {
1706 mutex_unlock(&dir
->d_inode
->i_mutex
);
1710 if (IS_ERR(nd
.intent
.open
.file
)) {
1711 error
= PTR_ERR(nd
.intent
.open
.file
);
1712 goto exit_mutex_unlock
;
1715 /* Negative dentry, just create the file */
1716 if (!path
.dentry
->d_inode
) {
1718 * This write is needed to ensure that a
1719 * ro->rw transition does not occur between
1720 * the time when the file is created and when
1721 * a permanent write count is taken through
1722 * the 'struct file' in nameidata_to_filp().
1724 error
= mnt_want_write(nd
.path
.mnt
);
1726 goto exit_mutex_unlock
;
1727 error
= __open_namei_create(&nd
, &path
, flag
, mode
);
1729 mnt_drop_write(nd
.path
.mnt
);
1732 filp
= nameidata_to_filp(&nd
, open_flag
);
1733 mnt_drop_write(nd
.path
.mnt
);
1738 * It already exists.
1740 mutex_unlock(&dir
->d_inode
->i_mutex
);
1741 audit_inode(pathname
, path
.dentry
);
1747 if (__follow_mount(&path
)) {
1749 if (flag
& O_NOFOLLOW
)
1754 if (!path
.dentry
->d_inode
)
1756 if (path
.dentry
->d_inode
->i_op
&& path
.dentry
->d_inode
->i_op
->follow_link
)
1759 path_to_nameidata(&path
, &nd
);
1761 if (path
.dentry
->d_inode
&& S_ISDIR(path
.dentry
->d_inode
->i_mode
))
1766 * 1. may_open() truncates a file
1767 * 2. a rw->ro mount transition occurs
1768 * 3. nameidata_to_filp() fails due to
1770 * That would be inconsistent, and should
1771 * be avoided. Taking this mnt write here
1772 * ensures that (2) can not occur.
1774 will_write
= open_will_write_to_fs(flag
, nd
.path
.dentry
->d_inode
);
1776 error
= mnt_want_write(nd
.path
.mnt
);
1780 error
= may_open(&nd
, acc_mode
, flag
);
1783 mnt_drop_write(nd
.path
.mnt
);
1786 filp
= nameidata_to_filp(&nd
, open_flag
);
1788 * It is now safe to drop the mnt write
1789 * because the filp has had a write taken
1793 mnt_drop_write(nd
.path
.mnt
);
1797 mutex_unlock(&dir
->d_inode
->i_mutex
);
1799 path_put_conditional(&path
, &nd
);
1801 if (!IS_ERR(nd
.intent
.open
.file
))
1802 release_open_intent(&nd
);
1805 return ERR_PTR(error
);
1809 if (flag
& O_NOFOLLOW
)
1812 * This is subtle. Instead of calling do_follow_link() we do the
1813 * thing by hands. The reason is that this way we have zero link_count
1814 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1815 * After that we have the parent and last component, i.e.
1816 * we are in the same situation as after the first path_walk().
1817 * Well, almost - if the last component is normal we get its copy
1818 * stored in nd->last.name and we will have to putname() it when we
1819 * are done. Procfs-like symlinks just set LAST_BIND.
1821 nd
.flags
|= LOOKUP_PARENT
;
1822 error
= security_inode_follow_link(path
.dentry
, &nd
);
1825 error
= __do_follow_link(&path
, &nd
);
1827 /* Does someone understand code flow here? Or it is only
1828 * me so stupid? Anathema to whoever designed this non-sense
1829 * with "intent.open".
1831 release_open_intent(&nd
);
1832 return ERR_PTR(error
);
1834 nd
.flags
&= ~LOOKUP_PARENT
;
1835 if (nd
.last_type
== LAST_BIND
)
1838 if (nd
.last_type
!= LAST_NORM
)
1840 if (nd
.last
.name
[nd
.last
.len
]) {
1841 __putname(nd
.last
.name
);
1846 __putname(nd
.last
.name
);
1849 dir
= nd
.path
.dentry
;
1850 mutex_lock(&dir
->d_inode
->i_mutex
);
1851 path
.dentry
= lookup_hash(&nd
);
1852 path
.mnt
= nd
.path
.mnt
;
1853 __putname(nd
.last
.name
);
1858 * filp_open - open file and return file pointer
1860 * @filename: path to open
1861 * @flags: open flags as per the open(2) second argument
1862 * @mode: mode for the new file if O_CREAT is set, else ignored
1864 * This is the helper to open a file from kernelspace if you really
1865 * have to. But in generally you should not do this, so please move
1866 * along, nothing to see here..
1868 struct file
*filp_open(const char *filename
, int flags
, int mode
)
1870 return do_filp_open(AT_FDCWD
, filename
, flags
, mode
);
1872 EXPORT_SYMBOL(filp_open
);
1875 * lookup_create - lookup a dentry, creating it if it doesn't exist
1876 * @nd: nameidata info
1877 * @is_dir: directory flag
1879 * Simple function to lookup and return a dentry and create it
1880 * if it doesn't exist. Is SMP-safe.
1882 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1884 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
1886 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
1888 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1890 * Yucky last component or no last component at all?
1891 * (foo/., foo/.., /////)
1893 if (nd
->last_type
!= LAST_NORM
)
1895 nd
->flags
&= ~LOOKUP_PARENT
;
1896 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
1897 nd
->intent
.open
.flags
= O_EXCL
;
1900 * Do the final lookup.
1902 dentry
= lookup_hash(nd
);
1906 if (dentry
->d_inode
)
1909 * Special case - lookup gave negative, but... we had foo/bar/
1910 * From the vfs_mknod() POV we just have a negative dentry -
1911 * all is fine. Let's be bastards - you had / on the end, you've
1912 * been asking for (non-existent) directory. -ENOENT for you.
1914 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
1916 dentry
= ERR_PTR(-ENOENT
);
1921 dentry
= ERR_PTR(-EEXIST
);
1925 EXPORT_SYMBOL_GPL(lookup_create
);
1927 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1929 int error
= may_create(dir
, dentry
);
1934 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
1937 if (!dir
->i_op
|| !dir
->i_op
->mknod
)
1940 error
= devcgroup_inode_mknod(mode
, dev
);
1944 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
1949 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
1951 fsnotify_create(dir
, dentry
);
1955 static int may_mknod(mode_t mode
)
1957 switch (mode
& S_IFMT
) {
1963 case 0: /* zero mode translates to S_IFREG */
1972 asmlinkage
long sys_mknodat(int dfd
, const char __user
*filename
, int mode
,
1977 struct dentry
*dentry
;
1978 struct nameidata nd
;
1983 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
1987 dentry
= lookup_create(&nd
, 0);
1988 if (IS_ERR(dentry
)) {
1989 error
= PTR_ERR(dentry
);
1992 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
1993 mode
&= ~current
->fs
->umask
;
1994 error
= may_mknod(mode
);
1997 error
= mnt_want_write(nd
.path
.mnt
);
2000 switch (mode
& S_IFMT
) {
2001 case 0: case S_IFREG
:
2002 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2004 case S_IFCHR
: case S_IFBLK
:
2005 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2006 new_decode_dev(dev
));
2008 case S_IFIFO
: case S_IFSOCK
:
2009 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2012 mnt_drop_write(nd
.path
.mnt
);
2016 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2023 asmlinkage
long sys_mknod(const char __user
*filename
, int mode
, unsigned dev
)
2025 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2028 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2030 int error
= may_create(dir
, dentry
);
2035 if (!dir
->i_op
|| !dir
->i_op
->mkdir
)
2038 mode
&= (S_IRWXUGO
|S_ISVTX
);
2039 error
= security_inode_mkdir(dir
, dentry
, mode
);
2044 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2046 fsnotify_mkdir(dir
, dentry
);
2050 asmlinkage
long sys_mkdirat(int dfd
, const char __user
*pathname
, int mode
)
2054 struct dentry
*dentry
;
2055 struct nameidata nd
;
2057 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2061 dentry
= lookup_create(&nd
, 1);
2062 error
= PTR_ERR(dentry
);
2066 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2067 mode
&= ~current
->fs
->umask
;
2068 error
= mnt_want_write(nd
.path
.mnt
);
2071 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2072 mnt_drop_write(nd
.path
.mnt
);
2076 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2083 asmlinkage
long sys_mkdir(const char __user
*pathname
, int mode
)
2085 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2089 * We try to drop the dentry early: we should have
2090 * a usage count of 2 if we're the only user of this
2091 * dentry, and if that is true (possibly after pruning
2092 * the dcache), then we drop the dentry now.
2094 * A low-level filesystem can, if it choses, legally
2097 * if (!d_unhashed(dentry))
2100 * if it cannot handle the case of removing a directory
2101 * that is still in use by something else..
2103 void dentry_unhash(struct dentry
*dentry
)
2106 shrink_dcache_parent(dentry
);
2107 spin_lock(&dcache_lock
);
2108 spin_lock(&dentry
->d_lock
);
2109 if (atomic_read(&dentry
->d_count
) == 2)
2111 spin_unlock(&dentry
->d_lock
);
2112 spin_unlock(&dcache_lock
);
2115 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2117 int error
= may_delete(dir
, dentry
, 1);
2122 if (!dir
->i_op
|| !dir
->i_op
->rmdir
)
2127 mutex_lock(&dentry
->d_inode
->i_mutex
);
2128 dentry_unhash(dentry
);
2129 if (d_mountpoint(dentry
))
2132 error
= security_inode_rmdir(dir
, dentry
);
2134 error
= dir
->i_op
->rmdir(dir
, dentry
);
2136 dentry
->d_inode
->i_flags
|= S_DEAD
;
2139 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2148 static long do_rmdir(int dfd
, const char __user
*pathname
)
2152 struct dentry
*dentry
;
2153 struct nameidata nd
;
2155 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2159 switch(nd
.last_type
) {
2171 nd
.flags
&= ~LOOKUP_PARENT
;
2173 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2174 dentry
= lookup_hash(&nd
);
2175 error
= PTR_ERR(dentry
);
2178 error
= mnt_want_write(nd
.path
.mnt
);
2181 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2182 mnt_drop_write(nd
.path
.mnt
);
2186 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2193 asmlinkage
long sys_rmdir(const char __user
*pathname
)
2195 return do_rmdir(AT_FDCWD
, pathname
);
2198 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2200 int error
= may_delete(dir
, dentry
, 0);
2205 if (!dir
->i_op
|| !dir
->i_op
->unlink
)
2208 vfs_check_frozen(dir
->i_sb
, SB_FREEZE_WRITE
);
2212 mutex_lock(&dentry
->d_inode
->i_mutex
);
2213 if (d_mountpoint(dentry
))
2216 error
= security_inode_unlink(dir
, dentry
);
2218 error
= dir
->i_op
->unlink(dir
, dentry
);
2220 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2222 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2223 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2224 fsnotify_link_count(dentry
->d_inode
);
2232 * Make sure that the actual truncation of the file will occur outside its
2233 * directory's i_mutex. Truncate can take a long time if there is a lot of
2234 * writeout happening, and we don't want to prevent access to the directory
2235 * while waiting on the I/O.
2237 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2241 struct dentry
*dentry
;
2242 struct nameidata nd
;
2243 struct inode
*inode
= NULL
;
2245 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2250 if (nd
.last_type
!= LAST_NORM
)
2253 nd
.flags
&= ~LOOKUP_PARENT
;
2255 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2256 dentry
= lookup_hash(&nd
);
2257 error
= PTR_ERR(dentry
);
2258 if (!IS_ERR(dentry
)) {
2259 /* Why not before? Because we want correct error value */
2260 if (nd
.last
.name
[nd
.last
.len
])
2262 inode
= dentry
->d_inode
;
2264 atomic_inc(&inode
->i_count
);
2265 error
= mnt_want_write(nd
.path
.mnt
);
2268 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2269 mnt_drop_write(nd
.path
.mnt
);
2273 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2275 iput(inode
); /* truncate the inode here */
2282 error
= !dentry
->d_inode
? -ENOENT
:
2283 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2287 asmlinkage
long sys_unlinkat(int dfd
, const char __user
*pathname
, int flag
)
2289 if ((flag
& ~AT_REMOVEDIR
) != 0)
2292 if (flag
& AT_REMOVEDIR
)
2293 return do_rmdir(dfd
, pathname
);
2295 return do_unlinkat(dfd
, pathname
);
2298 asmlinkage
long sys_unlink(const char __user
*pathname
)
2300 return do_unlinkat(AT_FDCWD
, pathname
);
2303 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2305 int error
= may_create(dir
, dentry
);
2310 if (!dir
->i_op
|| !dir
->i_op
->symlink
)
2313 error
= security_inode_symlink(dir
, dentry
, oldname
);
2318 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2320 fsnotify_create(dir
, dentry
);
2324 asmlinkage
long sys_symlinkat(const char __user
*oldname
,
2325 int newdfd
, const char __user
*newname
)
2330 struct dentry
*dentry
;
2331 struct nameidata nd
;
2333 from
= getname(oldname
);
2335 return PTR_ERR(from
);
2337 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2341 dentry
= lookup_create(&nd
, 0);
2342 error
= PTR_ERR(dentry
);
2346 error
= mnt_want_write(nd
.path
.mnt
);
2349 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2350 mnt_drop_write(nd
.path
.mnt
);
2354 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2362 asmlinkage
long sys_symlink(const char __user
*oldname
, const char __user
*newname
)
2364 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2367 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2369 struct inode
*inode
= old_dentry
->d_inode
;
2375 error
= may_create(dir
, new_dentry
);
2379 if (dir
->i_sb
!= inode
->i_sb
)
2383 * A link to an append-only or immutable file cannot be created.
2385 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2387 if (!dir
->i_op
|| !dir
->i_op
->link
)
2389 if (S_ISDIR(inode
->i_mode
))
2392 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2396 mutex_lock(&inode
->i_mutex
);
2398 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2399 mutex_unlock(&inode
->i_mutex
);
2401 fsnotify_link(dir
, inode
, new_dentry
);
2406 * Hardlinks are often used in delicate situations. We avoid
2407 * security-related surprises by not following symlinks on the
2410 * We don't follow them on the oldname either to be compatible
2411 * with linux 2.0, and to avoid hard-linking to directories
2412 * and other special files. --ADM
2414 asmlinkage
long sys_linkat(int olddfd
, const char __user
*oldname
,
2415 int newdfd
, const char __user
*newname
,
2418 struct dentry
*new_dentry
;
2419 struct nameidata nd
;
2420 struct path old_path
;
2424 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
2427 error
= user_path_at(olddfd
, oldname
,
2428 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
2433 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2437 if (old_path
.mnt
!= nd
.path
.mnt
)
2439 new_dentry
= lookup_create(&nd
, 0);
2440 error
= PTR_ERR(new_dentry
);
2441 if (IS_ERR(new_dentry
))
2443 error
= mnt_want_write(nd
.path
.mnt
);
2446 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2447 mnt_drop_write(nd
.path
.mnt
);
2451 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2456 path_put(&old_path
);
2461 asmlinkage
long sys_link(const char __user
*oldname
, const char __user
*newname
)
2463 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2467 * The worst of all namespace operations - renaming directory. "Perverted"
2468 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2470 * a) we can get into loop creation. Check is done in is_subdir().
2471 * b) race potential - two innocent renames can create a loop together.
2472 * That's where 4.4 screws up. Current fix: serialization on
2473 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2475 * c) we have to lock _three_ objects - parents and victim (if it exists).
2476 * And that - after we got ->i_mutex on parents (until then we don't know
2477 * whether the target exists). Solution: try to be smart with locking
2478 * order for inodes. We rely on the fact that tree topology may change
2479 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2480 * move will be locked. Thus we can rank directories by the tree
2481 * (ancestors first) and rank all non-directories after them.
2482 * That works since everybody except rename does "lock parent, lookup,
2483 * lock child" and rename is under ->s_vfs_rename_mutex.
2484 * HOWEVER, it relies on the assumption that any object with ->lookup()
2485 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2486 * we'd better make sure that there's no link(2) for them.
2487 * d) some filesystems don't support opened-but-unlinked directories,
2488 * either because of layout or because they are not ready to deal with
2489 * all cases correctly. The latter will be fixed (taking this sort of
2490 * stuff into VFS), but the former is not going away. Solution: the same
2491 * trick as in rmdir().
2492 * e) conversion from fhandle to dentry may come in the wrong moment - when
2493 * we are removing the target. Solution: we will have to grab ->i_mutex
2494 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2495 * ->i_mutex on parents, which works but leads to some truely excessive
2498 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2499 struct inode
*new_dir
, struct dentry
*new_dentry
)
2502 struct inode
*target
;
2505 * If we are going to change the parent - check write permissions,
2506 * we'll need to flip '..'.
2508 if (new_dir
!= old_dir
) {
2509 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
2514 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2518 target
= new_dentry
->d_inode
;
2520 mutex_lock(&target
->i_mutex
);
2521 dentry_unhash(new_dentry
);
2523 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2526 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2529 target
->i_flags
|= S_DEAD
;
2530 mutex_unlock(&target
->i_mutex
);
2531 if (d_unhashed(new_dentry
))
2532 d_rehash(new_dentry
);
2536 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2537 d_move(old_dentry
,new_dentry
);
2541 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
2542 struct inode
*new_dir
, struct dentry
*new_dentry
)
2544 struct inode
*target
;
2547 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2552 target
= new_dentry
->d_inode
;
2554 mutex_lock(&target
->i_mutex
);
2555 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2558 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2560 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2561 d_move(old_dentry
, new_dentry
);
2564 mutex_unlock(&target
->i_mutex
);
2569 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2570 struct inode
*new_dir
, struct dentry
*new_dentry
)
2573 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2574 const char *old_name
;
2576 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
2579 error
= may_delete(old_dir
, old_dentry
, is_dir
);
2583 if (!new_dentry
->d_inode
)
2584 error
= may_create(new_dir
, new_dentry
);
2586 error
= may_delete(new_dir
, new_dentry
, is_dir
);
2590 if (!old_dir
->i_op
|| !old_dir
->i_op
->rename
)
2593 DQUOT_INIT(old_dir
);
2594 DQUOT_INIT(new_dir
);
2596 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
2599 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
2601 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
2603 const char *new_name
= old_dentry
->d_name
.name
;
2604 fsnotify_move(old_dir
, new_dir
, old_name
, new_name
, is_dir
,
2605 new_dentry
->d_inode
, old_dentry
);
2607 fsnotify_oldname_free(old_name
);
2612 asmlinkage
long sys_renameat(int olddfd
, const char __user
*oldname
,
2613 int newdfd
, const char __user
*newname
)
2615 struct dentry
*old_dir
, *new_dir
;
2616 struct dentry
*old_dentry
, *new_dentry
;
2617 struct dentry
*trap
;
2618 struct nameidata oldnd
, newnd
;
2623 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
2627 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
2632 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
2635 old_dir
= oldnd
.path
.dentry
;
2637 if (oldnd
.last_type
!= LAST_NORM
)
2640 new_dir
= newnd
.path
.dentry
;
2641 if (newnd
.last_type
!= LAST_NORM
)
2644 oldnd
.flags
&= ~LOOKUP_PARENT
;
2645 newnd
.flags
&= ~LOOKUP_PARENT
;
2646 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
2648 trap
= lock_rename(new_dir
, old_dir
);
2650 old_dentry
= lookup_hash(&oldnd
);
2651 error
= PTR_ERR(old_dentry
);
2652 if (IS_ERR(old_dentry
))
2654 /* source must exist */
2656 if (!old_dentry
->d_inode
)
2658 /* unless the source is a directory trailing slashes give -ENOTDIR */
2659 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
2661 if (oldnd
.last
.name
[oldnd
.last
.len
])
2663 if (newnd
.last
.name
[newnd
.last
.len
])
2666 /* source should not be ancestor of target */
2668 if (old_dentry
== trap
)
2670 new_dentry
= lookup_hash(&newnd
);
2671 error
= PTR_ERR(new_dentry
);
2672 if (IS_ERR(new_dentry
))
2674 /* target should not be an ancestor of source */
2676 if (new_dentry
== trap
)
2679 error
= mnt_want_write(oldnd
.path
.mnt
);
2682 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
2683 new_dir
->d_inode
, new_dentry
);
2684 mnt_drop_write(oldnd
.path
.mnt
);
2690 unlock_rename(new_dir
, old_dir
);
2692 path_put(&newnd
.path
);
2695 path_put(&oldnd
.path
);
2701 asmlinkage
long sys_rename(const char __user
*oldname
, const char __user
*newname
)
2703 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
2706 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
2710 len
= PTR_ERR(link
);
2715 if (len
> (unsigned) buflen
)
2717 if (copy_to_user(buffer
, link
, len
))
2724 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2725 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2726 * using) it for any given inode is up to filesystem.
2728 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2730 struct nameidata nd
;
2735 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
2737 return PTR_ERR(cookie
);
2739 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
2740 if (dentry
->d_inode
->i_op
->put_link
)
2741 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
2745 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
2747 return __vfs_follow_link(nd
, link
);
2750 /* get the link contents into pagecache */
2751 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
2754 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
2755 page
= read_mapping_page(mapping
, 0, NULL
);
2762 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2764 struct page
*page
= NULL
;
2765 char *s
= page_getlink(dentry
, &page
);
2766 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
2769 page_cache_release(page
);
2774 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
2776 struct page
*page
= NULL
;
2777 nd_set_link(nd
, page_getlink(dentry
, &page
));
2781 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2783 struct page
*page
= cookie
;
2787 page_cache_release(page
);
2791 int __page_symlink(struct inode
*inode
, const char *symname
, int len
,
2794 struct address_space
*mapping
= inode
->i_mapping
;
2801 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
2802 AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
2806 kaddr
= kmap_atomic(page
, KM_USER0
);
2807 memcpy(kaddr
, symname
, len
-1);
2808 kunmap_atomic(kaddr
, KM_USER0
);
2810 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
2817 mark_inode_dirty(inode
);
2823 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
2825 return __page_symlink(inode
, symname
, len
,
2826 mapping_gfp_mask(inode
->i_mapping
));
2829 const struct inode_operations page_symlink_inode_operations
= {
2830 .readlink
= generic_readlink
,
2831 .follow_link
= page_follow_link_light
,
2832 .put_link
= page_put_link
,
2835 EXPORT_SYMBOL(user_path_at
);
2836 EXPORT_SYMBOL(follow_down
);
2837 EXPORT_SYMBOL(follow_up
);
2838 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
2839 EXPORT_SYMBOL(getname
);
2840 EXPORT_SYMBOL(lock_rename
);
2841 EXPORT_SYMBOL(__lookup_hash
);
2842 EXPORT_SYMBOL(lookup_one_len
);
2843 EXPORT_SYMBOL(page_follow_link_light
);
2844 EXPORT_SYMBOL(page_put_link
);
2845 EXPORT_SYMBOL(page_readlink
);
2846 EXPORT_SYMBOL(__page_symlink
);
2847 EXPORT_SYMBOL(page_symlink
);
2848 EXPORT_SYMBOL(page_symlink_inode_operations
);
2849 EXPORT_SYMBOL(path_lookup
);
2850 EXPORT_SYMBOL(kern_path
);
2851 EXPORT_SYMBOL(vfs_path_lookup
);
2852 EXPORT_SYMBOL(inode_permission
);
2853 EXPORT_SYMBOL(vfs_permission
);
2854 EXPORT_SYMBOL(file_permission
);
2855 EXPORT_SYMBOL(unlock_rename
);
2856 EXPORT_SYMBOL(vfs_create
);
2857 EXPORT_SYMBOL(vfs_follow_link
);
2858 EXPORT_SYMBOL(vfs_link
);
2859 EXPORT_SYMBOL(vfs_mkdir
);
2860 EXPORT_SYMBOL(vfs_mknod
);
2861 EXPORT_SYMBOL(generic_permission
);
2862 EXPORT_SYMBOL(vfs_readlink
);
2863 EXPORT_SYMBOL(vfs_rename
);
2864 EXPORT_SYMBOL(vfs_rmdir
);
2865 EXPORT_SYMBOL(vfs_symlink
);
2866 EXPORT_SYMBOL(vfs_unlink
);
2867 EXPORT_SYMBOL(dentry_unhash
);
2868 EXPORT_SYMBOL(generic_readlink
);
2869 EXPORT_SYMBOL(deny_write_access
);