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 <asm/namei.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 if (current
->fsuid
== inode
->i_uid
)
190 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
191 int error
= check_acl(inode
, mask
);
192 if (error
== -EACCES
)
193 goto check_capabilities
;
194 else if (error
!= -EAGAIN
)
198 if (in_group_p(inode
->i_gid
))
203 * If the DACs are ok we don't need any capability check.
205 if (((mode
& mask
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
)) == mask
))
210 * Read/write DACs are always overridable.
211 * Executable DACs are overridable if at least one exec bit is set.
213 if (!(mask
& MAY_EXEC
) ||
214 (inode
->i_mode
& S_IXUGO
) || S_ISDIR(inode
->i_mode
))
215 if (capable(CAP_DAC_OVERRIDE
))
219 * Searching includes executable on directories, else just read.
221 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
222 if (capable(CAP_DAC_READ_SEARCH
))
228 int permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
231 struct vfsmount
*mnt
= NULL
;
236 if (mask
& MAY_WRITE
) {
237 umode_t mode
= inode
->i_mode
;
240 * Nobody gets write access to a read-only fs.
242 if (IS_RDONLY(inode
) &&
243 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
247 * Nobody gets write access to an immutable file.
249 if (IS_IMMUTABLE(inode
))
253 if ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
)) {
255 * MAY_EXEC on regular files is denied if the fs is mounted
256 * with the "noexec" flag.
258 if (mnt
&& (mnt
->mnt_flags
& MNT_NOEXEC
))
262 /* Ordinary permission routines do not understand MAY_APPEND. */
263 submask
= mask
& ~MAY_APPEND
;
264 if (inode
->i_op
&& inode
->i_op
->permission
) {
265 retval
= inode
->i_op
->permission(inode
, submask
, nd
);
268 * Exec permission on a regular file is denied if none
269 * of the execute bits are set.
271 * This check should be done by the ->permission()
274 if ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
) &&
275 !(inode
->i_mode
& S_IXUGO
))
279 retval
= generic_permission(inode
, submask
, NULL
);
284 return security_inode_permission(inode
, mask
, nd
);
288 * vfs_permission - check for access rights to a given path
289 * @nd: lookup result that describes the path
290 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
292 * Used to check for read/write/execute permissions on a path.
293 * We use "fsuid" for this, letting us set arbitrary permissions
294 * for filesystem access without changing the "normal" uids which
295 * are used for other things.
297 int vfs_permission(struct nameidata
*nd
, int mask
)
299 return permission(nd
->path
.dentry
->d_inode
, mask
, nd
);
303 * file_permission - check for additional access rights to a given file
304 * @file: file to check access rights for
305 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
307 * Used to check for read/write/execute permissions on an already opened
311 * Do not use this function in new code. All access checks should
312 * be done using vfs_permission().
314 int file_permission(struct file
*file
, int mask
)
316 return permission(file
->f_path
.dentry
->d_inode
, mask
, NULL
);
320 * get_write_access() gets write permission for a file.
321 * put_write_access() releases this write permission.
322 * This is used for regular files.
323 * We cannot support write (and maybe mmap read-write shared) accesses and
324 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
325 * can have the following values:
326 * 0: no writers, no VM_DENYWRITE mappings
327 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
328 * > 0: (i_writecount) users are writing to the file.
330 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
331 * except for the cases where we don't hold i_writecount yet. Then we need to
332 * use {get,deny}_write_access() - these functions check the sign and refuse
333 * to do the change if sign is wrong. Exclusion between them is provided by
334 * the inode->i_lock spinlock.
337 int get_write_access(struct inode
* inode
)
339 spin_lock(&inode
->i_lock
);
340 if (atomic_read(&inode
->i_writecount
) < 0) {
341 spin_unlock(&inode
->i_lock
);
344 atomic_inc(&inode
->i_writecount
);
345 spin_unlock(&inode
->i_lock
);
350 int deny_write_access(struct file
* file
)
352 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
354 spin_lock(&inode
->i_lock
);
355 if (atomic_read(&inode
->i_writecount
) > 0) {
356 spin_unlock(&inode
->i_lock
);
359 atomic_dec(&inode
->i_writecount
);
360 spin_unlock(&inode
->i_lock
);
366 * path_get - get a reference to a path
367 * @path: path to get the reference to
369 * Given a path increment the reference count to the dentry and the vfsmount.
371 void path_get(struct path
*path
)
376 EXPORT_SYMBOL(path_get
);
379 * path_put - put a reference to a path
380 * @path: path to put the reference to
382 * Given a path decrement the reference count to the dentry and the vfsmount.
384 void path_put(struct path
*path
)
389 EXPORT_SYMBOL(path_put
);
392 * release_open_intent - free up open intent resources
393 * @nd: pointer to nameidata
395 void release_open_intent(struct nameidata
*nd
)
397 if (nd
->intent
.open
.file
->f_path
.dentry
== NULL
)
398 put_filp(nd
->intent
.open
.file
);
400 fput(nd
->intent
.open
.file
);
403 static inline struct dentry
*
404 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
406 int status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
407 if (unlikely(status
<= 0)) {
409 * The dentry failed validation.
410 * If d_revalidate returned 0 attempt to invalidate
411 * the dentry otherwise d_revalidate is asking us
412 * to return a fail status.
415 if (!d_invalidate(dentry
)) {
421 dentry
= ERR_PTR(status
);
428 * Internal lookup() using the new generic dcache.
431 static struct dentry
* cached_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
433 struct dentry
* dentry
= __d_lookup(parent
, name
);
435 /* lockess __d_lookup may fail due to concurrent d_move()
436 * in some unrelated directory, so try with d_lookup
439 dentry
= d_lookup(parent
, name
);
441 if (dentry
&& dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
442 dentry
= do_revalidate(dentry
, nd
);
448 * Short-cut version of permission(), for calling by
449 * path_walk(), when dcache lock is held. Combines parts
450 * of permission() and generic_permission(), and tests ONLY for
451 * MAY_EXEC permission.
453 * If appropriate, check DAC only. If not appropriate, or
454 * short-cut DAC fails, then call permission() to do more
455 * complete permission check.
457 static int exec_permission_lite(struct inode
*inode
,
458 struct nameidata
*nd
)
460 umode_t mode
= inode
->i_mode
;
462 if (inode
->i_op
&& inode
->i_op
->permission
)
465 if (current
->fsuid
== inode
->i_uid
)
467 else if (in_group_p(inode
->i_gid
))
473 if ((inode
->i_mode
& S_IXUGO
) && capable(CAP_DAC_OVERRIDE
))
476 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_OVERRIDE
))
479 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_READ_SEARCH
))
484 return security_inode_permission(inode
, MAY_EXEC
, nd
);
488 * This is called when everything else fails, and we actually have
489 * to go to the low-level filesystem to find out what we should do..
491 * We get the directory semaphore, and after getting that we also
492 * make sure that nobody added the entry to the dcache in the meantime..
495 static struct dentry
* real_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
497 struct dentry
* result
;
498 struct inode
*dir
= parent
->d_inode
;
500 mutex_lock(&dir
->i_mutex
);
502 * First re-do the cached lookup just in case it was created
503 * while we waited for the directory semaphore..
505 * FIXME! This could use version numbering or similar to
506 * avoid unnecessary cache lookups.
508 * The "dcache_lock" is purely to protect the RCU list walker
509 * from concurrent renames at this point (we mustn't get false
510 * negatives from the RCU list walk here, unlike the optimistic
513 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
515 result
= d_lookup(parent
, name
);
517 struct dentry
* dentry
= d_alloc(parent
, name
);
518 result
= ERR_PTR(-ENOMEM
);
520 result
= dir
->i_op
->lookup(dir
, dentry
, nd
);
526 mutex_unlock(&dir
->i_mutex
);
531 * Uhhuh! Nasty case: the cache was re-populated while
532 * we waited on the semaphore. Need to revalidate.
534 mutex_unlock(&dir
->i_mutex
);
535 if (result
->d_op
&& result
->d_op
->d_revalidate
) {
536 result
= do_revalidate(result
, nd
);
538 result
= ERR_PTR(-ENOENT
);
543 static int __emul_lookup_dentry(const char *, struct nameidata
*);
546 static __always_inline
int
547 walk_init_root(const char *name
, struct nameidata
*nd
)
549 struct fs_struct
*fs
= current
->fs
;
551 read_lock(&fs
->lock
);
552 if (fs
->altroot
.dentry
&& !(nd
->flags
& LOOKUP_NOALT
)) {
553 nd
->path
= fs
->altroot
;
554 path_get(&fs
->altroot
);
555 read_unlock(&fs
->lock
);
556 if (__emul_lookup_dentry(name
,nd
))
558 read_lock(&fs
->lock
);
562 read_unlock(&fs
->lock
);
567 * Wrapper to retry pathname resolution whenever the underlying
568 * file system returns an ESTALE.
570 * Retry the whole path once, forcing real lookup requests
571 * instead of relying on the dcache.
573 static __always_inline
int link_path_walk(const char *name
, struct nameidata
*nd
)
575 struct path save
= nd
->path
;
578 /* make sure the stuff we saved doesn't go away */
582 result
= __link_path_walk(name
, nd
);
583 if (result
== -ESTALE
) {
584 /* nd->path had been dropped */
586 dget(nd
->path
.dentry
);
587 mntget(nd
->path
.mnt
);
588 nd
->flags
|= LOOKUP_REVAL
;
589 result
= __link_path_walk(name
, nd
);
597 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
606 if (!walk_init_root(link
, nd
))
607 /* weird __emul_prefix() stuff did it */
610 res
= link_path_walk(link
, nd
);
612 if (nd
->depth
|| res
|| nd
->last_type
!=LAST_NORM
)
615 * If it is an iterative symlinks resolution in open_namei() we
616 * have to copy the last component. And all that crap because of
617 * bloody create() on broken symlinks. Furrfu...
620 if (unlikely(!name
)) {
624 strcpy(name
, nd
->last
.name
);
625 nd
->last
.name
= name
;
629 return PTR_ERR(link
);
632 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
635 if (path
->mnt
!= nd
->path
.mnt
)
639 static inline void path_to_nameidata(struct path
*path
, struct nameidata
*nd
)
641 dput(nd
->path
.dentry
);
642 if (nd
->path
.mnt
!= path
->mnt
)
643 mntput(nd
->path
.mnt
);
644 nd
->path
.mnt
= path
->mnt
;
645 nd
->path
.dentry
= path
->dentry
;
648 static __always_inline
int __do_follow_link(struct path
*path
, struct nameidata
*nd
)
652 struct dentry
*dentry
= path
->dentry
;
654 touch_atime(path
->mnt
, dentry
);
655 nd_set_link(nd
, NULL
);
657 if (path
->mnt
!= nd
->path
.mnt
) {
658 path_to_nameidata(path
, nd
);
662 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
663 error
= PTR_ERR(cookie
);
664 if (!IS_ERR(cookie
)) {
665 char *s
= nd_get_link(nd
);
668 error
= __vfs_follow_link(nd
, s
);
669 if (dentry
->d_inode
->i_op
->put_link
)
670 dentry
->d_inode
->i_op
->put_link(dentry
, nd
, cookie
);
678 * This limits recursive symlink follows to 8, while
679 * limiting consecutive symlinks to 40.
681 * Without that kind of total limit, nasty chains of consecutive
682 * symlinks can cause almost arbitrarily long lookups.
684 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
687 if (current
->link_count
>= MAX_NESTED_LINKS
)
689 if (current
->total_link_count
>= 40)
691 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
693 err
= security_inode_follow_link(path
->dentry
, nd
);
696 current
->link_count
++;
697 current
->total_link_count
++;
699 err
= __do_follow_link(path
, nd
);
700 current
->link_count
--;
704 path_put_conditional(path
, nd
);
709 int follow_up(struct vfsmount
**mnt
, struct dentry
**dentry
)
711 struct vfsmount
*parent
;
712 struct dentry
*mountpoint
;
713 spin_lock(&vfsmount_lock
);
714 parent
=(*mnt
)->mnt_parent
;
715 if (parent
== *mnt
) {
716 spin_unlock(&vfsmount_lock
);
720 mountpoint
=dget((*mnt
)->mnt_mountpoint
);
721 spin_unlock(&vfsmount_lock
);
723 *dentry
= mountpoint
;
729 /* no need for dcache_lock, as serialization is taken care in
732 static int __follow_mount(struct path
*path
)
735 while (d_mountpoint(path
->dentry
)) {
736 struct vfsmount
*mounted
= lookup_mnt(path
->mnt
, path
->dentry
);
743 path
->dentry
= dget(mounted
->mnt_root
);
749 static void follow_mount(struct vfsmount
**mnt
, struct dentry
**dentry
)
751 while (d_mountpoint(*dentry
)) {
752 struct vfsmount
*mounted
= lookup_mnt(*mnt
, *dentry
);
758 *dentry
= dget(mounted
->mnt_root
);
762 /* no need for dcache_lock, as serialization is taken care in
765 int follow_down(struct vfsmount
**mnt
, struct dentry
**dentry
)
767 struct vfsmount
*mounted
;
769 mounted
= lookup_mnt(*mnt
, *dentry
);
774 *dentry
= dget(mounted
->mnt_root
);
780 static __always_inline
void follow_dotdot(struct nameidata
*nd
)
782 struct fs_struct
*fs
= current
->fs
;
785 struct vfsmount
*parent
;
786 struct dentry
*old
= nd
->path
.dentry
;
788 read_lock(&fs
->lock
);
789 if (nd
->path
.dentry
== fs
->root
.dentry
&&
790 nd
->path
.mnt
== fs
->root
.mnt
) {
791 read_unlock(&fs
->lock
);
794 read_unlock(&fs
->lock
);
795 spin_lock(&dcache_lock
);
796 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
797 nd
->path
.dentry
= dget(nd
->path
.dentry
->d_parent
);
798 spin_unlock(&dcache_lock
);
802 spin_unlock(&dcache_lock
);
803 spin_lock(&vfsmount_lock
);
804 parent
= nd
->path
.mnt
->mnt_parent
;
805 if (parent
== nd
->path
.mnt
) {
806 spin_unlock(&vfsmount_lock
);
810 nd
->path
.dentry
= dget(nd
->path
.mnt
->mnt_mountpoint
);
811 spin_unlock(&vfsmount_lock
);
813 mntput(nd
->path
.mnt
);
814 nd
->path
.mnt
= parent
;
816 follow_mount(&nd
->path
.mnt
, &nd
->path
.dentry
);
820 * It's more convoluted than I'd like it to be, but... it's still fairly
821 * small and for now I'd prefer to have fast path as straight as possible.
822 * It _is_ time-critical.
824 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
827 struct vfsmount
*mnt
= nd
->path
.mnt
;
828 struct dentry
*dentry
= __d_lookup(nd
->path
.dentry
, name
);
832 if (dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
833 goto need_revalidate
;
836 path
->dentry
= dentry
;
837 __follow_mount(path
);
841 dentry
= real_lookup(nd
->path
.dentry
, name
, nd
);
847 dentry
= do_revalidate(dentry
, nd
);
855 return PTR_ERR(dentry
);
860 * This is the basic name resolution function, turning a pathname into
861 * the final dentry. We expect 'base' to be positive and a directory.
863 * Returns 0 and nd will have valid dentry and mnt on success.
864 * Returns error and drops reference to input namei data on failure.
866 static int __link_path_walk(const char *name
, struct nameidata
*nd
)
871 unsigned int lookup_flags
= nd
->flags
;
878 inode
= nd
->path
.dentry
->d_inode
;
880 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
882 /* At this point we know we have a real path component. */
888 nd
->flags
|= LOOKUP_CONTINUE
;
889 err
= exec_permission_lite(inode
, nd
);
891 err
= vfs_permission(nd
, MAY_EXEC
);
896 c
= *(const unsigned char *)name
;
898 hash
= init_name_hash();
901 hash
= partial_name_hash(c
, hash
);
902 c
= *(const unsigned char *)name
;
903 } while (c
&& (c
!= '/'));
904 this.len
= name
- (const char *) this.name
;
905 this.hash
= end_name_hash(hash
);
907 /* remove trailing slashes? */
910 while (*++name
== '/');
912 goto last_with_slashes
;
915 * "." and ".." are special - ".." especially so because it has
916 * to be able to know about the current root directory and
917 * parent relationships.
919 if (this.name
[0] == '.') switch (this.len
) {
923 if (this.name
[1] != '.')
926 inode
= nd
->path
.dentry
->d_inode
;
932 * See if the low-level filesystem might want
933 * to use its own hash..
935 if (nd
->path
.dentry
->d_op
&& nd
->path
.dentry
->d_op
->d_hash
) {
936 err
= nd
->path
.dentry
->d_op
->d_hash(nd
->path
.dentry
,
941 /* This does the actual lookups.. */
942 err
= do_lookup(nd
, &this, &next
);
947 inode
= next
.dentry
->d_inode
;
954 if (inode
->i_op
->follow_link
) {
955 err
= do_follow_link(&next
, nd
);
959 inode
= nd
->path
.dentry
->d_inode
;
966 path_to_nameidata(&next
, nd
);
968 if (!inode
->i_op
->lookup
)
971 /* here ends the main loop */
974 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
976 /* Clear LOOKUP_CONTINUE iff it was previously unset */
977 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
978 if (lookup_flags
& LOOKUP_PARENT
)
980 if (this.name
[0] == '.') switch (this.len
) {
984 if (this.name
[1] != '.')
987 inode
= nd
->path
.dentry
->d_inode
;
992 if (nd
->path
.dentry
->d_op
&& nd
->path
.dentry
->d_op
->d_hash
) {
993 err
= nd
->path
.dentry
->d_op
->d_hash(nd
->path
.dentry
,
998 err
= do_lookup(nd
, &this, &next
);
1001 inode
= next
.dentry
->d_inode
;
1002 if ((lookup_flags
& LOOKUP_FOLLOW
)
1003 && inode
&& inode
->i_op
&& inode
->i_op
->follow_link
) {
1004 err
= do_follow_link(&next
, nd
);
1007 inode
= nd
->path
.dentry
->d_inode
;
1009 path_to_nameidata(&next
, nd
);
1013 if (lookup_flags
& LOOKUP_DIRECTORY
) {
1015 if (!inode
->i_op
|| !inode
->i_op
->lookup
)
1021 nd
->last_type
= LAST_NORM
;
1022 if (this.name
[0] != '.')
1025 nd
->last_type
= LAST_DOT
;
1026 else if (this.len
== 2 && this.name
[1] == '.')
1027 nd
->last_type
= LAST_DOTDOT
;
1032 * We bypassed the ordinary revalidation routines.
1033 * We may need to check the cached dentry for staleness.
1035 if (nd
->path
.dentry
&& nd
->path
.dentry
->d_sb
&&
1036 (nd
->path
.dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)) {
1038 /* Note: we do not d_invalidate() */
1039 if (!nd
->path
.dentry
->d_op
->d_revalidate(
1040 nd
->path
.dentry
, nd
))
1046 path_put_conditional(&next
, nd
);
1049 path_put(&nd
->path
);
1054 static int path_walk(const char *name
, struct nameidata
*nd
)
1056 current
->total_link_count
= 0;
1057 return link_path_walk(name
, nd
);
1061 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1062 * everything is done. Returns 0 and drops input nd, if lookup failed;
1064 static int __emul_lookup_dentry(const char *name
, struct nameidata
*nd
)
1066 if (path_walk(name
, nd
))
1067 return 0; /* something went wrong... */
1069 if (!nd
->path
.dentry
->d_inode
||
1070 S_ISDIR(nd
->path
.dentry
->d_inode
->i_mode
)) {
1071 struct path old_path
= nd
->path
;
1072 struct qstr last
= nd
->last
;
1073 int last_type
= nd
->last_type
;
1074 struct fs_struct
*fs
= current
->fs
;
1077 * NAME was not found in alternate root or it's a directory.
1078 * Try to find it in the normal root:
1080 nd
->last_type
= LAST_ROOT
;
1081 read_lock(&fs
->lock
);
1082 nd
->path
= fs
->root
;
1083 path_get(&fs
->root
);
1084 read_unlock(&fs
->lock
);
1085 if (path_walk(name
, nd
) == 0) {
1086 if (nd
->path
.dentry
->d_inode
) {
1087 path_put(&old_path
);
1090 path_put(&nd
->path
);
1092 nd
->path
= old_path
;
1094 nd
->last_type
= last_type
;
1099 void set_fs_altroot(void)
1101 char *emul
= __emul_prefix();
1102 struct nameidata nd
;
1103 struct path path
= {}, old_path
;
1105 struct fs_struct
*fs
= current
->fs
;
1109 err
= path_lookup(emul
, LOOKUP_FOLLOW
|LOOKUP_DIRECTORY
|LOOKUP_NOALT
, &nd
);
1113 write_lock(&fs
->lock
);
1114 old_path
= fs
->altroot
;
1116 write_unlock(&fs
->lock
);
1117 if (old_path
.dentry
)
1118 path_put(&old_path
);
1121 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1122 static int do_path_lookup(int dfd
, const char *name
,
1123 unsigned int flags
, struct nameidata
*nd
)
1128 struct fs_struct
*fs
= current
->fs
;
1130 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1135 read_lock(&fs
->lock
);
1136 if (fs
->altroot
.dentry
&& !(nd
->flags
& LOOKUP_NOALT
)) {
1137 nd
->path
= fs
->altroot
;
1138 path_get(&fs
->altroot
);
1139 read_unlock(&fs
->lock
);
1140 if (__emul_lookup_dentry(name
,nd
))
1141 goto out
; /* found in altroot */
1142 read_lock(&fs
->lock
);
1144 nd
->path
= fs
->root
;
1145 path_get(&fs
->root
);
1146 read_unlock(&fs
->lock
);
1147 } else if (dfd
== AT_FDCWD
) {
1148 read_lock(&fs
->lock
);
1151 read_unlock(&fs
->lock
);
1153 struct dentry
*dentry
;
1155 file
= fget_light(dfd
, &fput_needed
);
1160 dentry
= file
->f_path
.dentry
;
1163 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1166 retval
= file_permission(file
, MAY_EXEC
);
1170 nd
->path
= file
->f_path
;
1171 path_get(&file
->f_path
);
1173 fput_light(file
, fput_needed
);
1176 retval
= path_walk(name
, nd
);
1178 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1179 nd
->path
.dentry
->d_inode
))
1180 audit_inode(name
, nd
->path
.dentry
);
1185 fput_light(file
, fput_needed
);
1189 int path_lookup(const char *name
, unsigned int flags
,
1190 struct nameidata
*nd
)
1192 return do_path_lookup(AT_FDCWD
, name
, flags
, nd
);
1196 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1197 * @dentry: pointer to dentry of the base directory
1198 * @mnt: pointer to vfs mount of the base directory
1199 * @name: pointer to file name
1200 * @flags: lookup flags
1201 * @nd: pointer to nameidata
1203 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1204 const char *name
, unsigned int flags
,
1205 struct nameidata
*nd
)
1209 /* same as do_path_lookup */
1210 nd
->last_type
= LAST_ROOT
;
1214 nd
->path
.mnt
= mntget(mnt
);
1215 nd
->path
.dentry
= dget(dentry
);
1217 retval
= path_walk(name
, nd
);
1218 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1219 nd
->path
.dentry
->d_inode
))
1220 audit_inode(name
, nd
->path
.dentry
);
1226 static int __path_lookup_intent_open(int dfd
, const char *name
,
1227 unsigned int lookup_flags
, struct nameidata
*nd
,
1228 int open_flags
, int create_mode
)
1230 struct file
*filp
= get_empty_filp();
1235 nd
->intent
.open
.file
= filp
;
1236 nd
->intent
.open
.flags
= open_flags
;
1237 nd
->intent
.open
.create_mode
= create_mode
;
1238 err
= do_path_lookup(dfd
, name
, lookup_flags
|LOOKUP_OPEN
, nd
);
1239 if (IS_ERR(nd
->intent
.open
.file
)) {
1241 err
= PTR_ERR(nd
->intent
.open
.file
);
1242 path_put(&nd
->path
);
1244 } else if (err
!= 0)
1245 release_open_intent(nd
);
1250 * path_lookup_open - lookup a file path with open intent
1251 * @dfd: the directory to use as base, or AT_FDCWD
1252 * @name: pointer to file name
1253 * @lookup_flags: lookup intent flags
1254 * @nd: pointer to nameidata
1255 * @open_flags: open intent flags
1257 int path_lookup_open(int dfd
, const char *name
, unsigned int lookup_flags
,
1258 struct nameidata
*nd
, int open_flags
)
1260 return __path_lookup_intent_open(dfd
, name
, lookup_flags
, nd
,
1265 * path_lookup_create - lookup a file path with open + create intent
1266 * @dfd: the directory to use as base, or AT_FDCWD
1267 * @name: pointer to file name
1268 * @lookup_flags: lookup intent flags
1269 * @nd: pointer to nameidata
1270 * @open_flags: open intent flags
1271 * @create_mode: create intent flags
1273 static int path_lookup_create(int dfd
, const char *name
,
1274 unsigned int lookup_flags
, struct nameidata
*nd
,
1275 int open_flags
, int create_mode
)
1277 return __path_lookup_intent_open(dfd
, name
, lookup_flags
|LOOKUP_CREATE
,
1278 nd
, open_flags
, create_mode
);
1281 int __user_path_lookup_open(const char __user
*name
, unsigned int lookup_flags
,
1282 struct nameidata
*nd
, int open_flags
)
1284 char *tmp
= getname(name
);
1285 int err
= PTR_ERR(tmp
);
1288 err
= __path_lookup_intent_open(AT_FDCWD
, tmp
, lookup_flags
, nd
, open_flags
, 0);
1294 static struct dentry
*__lookup_hash(struct qstr
*name
,
1295 struct dentry
*base
, struct nameidata
*nd
)
1297 struct dentry
*dentry
;
1298 struct inode
*inode
;
1301 inode
= base
->d_inode
;
1304 * See if the low-level filesystem might want
1305 * to use its own hash..
1307 if (base
->d_op
&& base
->d_op
->d_hash
) {
1308 err
= base
->d_op
->d_hash(base
, name
);
1309 dentry
= ERR_PTR(err
);
1314 dentry
= cached_lookup(base
, name
, nd
);
1316 struct dentry
*new = d_alloc(base
, name
);
1317 dentry
= ERR_PTR(-ENOMEM
);
1320 dentry
= inode
->i_op
->lookup(inode
, new, nd
);
1331 * Restricted form of lookup. Doesn't follow links, single-component only,
1332 * needs parent already locked. Doesn't follow mounts.
1335 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1339 err
= permission(nd
->path
.dentry
->d_inode
, MAY_EXEC
, nd
);
1341 return ERR_PTR(err
);
1342 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1345 static int __lookup_one_len(const char *name
, struct qstr
*this,
1346 struct dentry
*base
, int len
)
1356 hash
= init_name_hash();
1358 c
= *(const unsigned char *)name
++;
1359 if (c
== '/' || c
== '\0')
1361 hash
= partial_name_hash(c
, hash
);
1363 this->hash
= end_name_hash(hash
);
1368 * lookup_one_len - filesystem helper to lookup single pathname component
1369 * @name: pathname component to lookup
1370 * @base: base directory to lookup from
1371 * @len: maximum length @len should be interpreted to
1373 * Note that this routine is purely a helper for filesystem usage and should
1374 * not be called by generic code. Also note that by using this function the
1375 * nameidata argument is passed to the filesystem methods and a filesystem
1376 * using this helper needs to be prepared for that.
1378 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1383 err
= __lookup_one_len(name
, &this, base
, len
);
1385 return ERR_PTR(err
);
1387 err
= permission(base
->d_inode
, MAY_EXEC
, NULL
);
1389 return ERR_PTR(err
);
1390 return __lookup_hash(&this, base
, NULL
);
1394 * lookup_one_noperm - bad hack for sysfs
1395 * @name: pathname component to lookup
1396 * @base: base directory to lookup from
1398 * This is a variant of lookup_one_len that doesn't perform any permission
1399 * checks. It's a horrible hack to work around the braindead sysfs
1400 * architecture and should not be used anywhere else.
1402 * DON'T USE THIS FUNCTION EVER, thanks.
1404 struct dentry
*lookup_one_noperm(const char *name
, struct dentry
*base
)
1409 err
= __lookup_one_len(name
, &this, base
, strlen(name
));
1411 return ERR_PTR(err
);
1412 return __lookup_hash(&this, base
, NULL
);
1415 int __user_walk_fd(int dfd
, const char __user
*name
, unsigned flags
,
1416 struct nameidata
*nd
)
1418 char *tmp
= getname(name
);
1419 int err
= PTR_ERR(tmp
);
1422 err
= do_path_lookup(dfd
, tmp
, flags
, nd
);
1428 int __user_walk(const char __user
*name
, unsigned flags
, struct nameidata
*nd
)
1430 return __user_walk_fd(AT_FDCWD
, name
, flags
, nd
);
1434 * It's inline, so penalty for filesystems that don't use sticky bit is
1437 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1439 if (!(dir
->i_mode
& S_ISVTX
))
1441 if (inode
->i_uid
== current
->fsuid
)
1443 if (dir
->i_uid
== current
->fsuid
)
1445 return !capable(CAP_FOWNER
);
1449 * Check whether we can remove a link victim from directory dir, check
1450 * whether the type of victim is right.
1451 * 1. We can't do it if dir is read-only (done in permission())
1452 * 2. We should have write and exec permissions on dir
1453 * 3. We can't remove anything from append-only dir
1454 * 4. We can't do anything with immutable dir (done in permission())
1455 * 5. If the sticky bit on dir is set we should either
1456 * a. be owner of dir, or
1457 * b. be owner of victim, or
1458 * c. have CAP_FOWNER capability
1459 * 6. If the victim is append-only or immutable we can't do antyhing with
1460 * links pointing to it.
1461 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1462 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1463 * 9. We can't remove a root or mountpoint.
1464 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1465 * nfs_async_unlink().
1467 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1471 if (!victim
->d_inode
)
1474 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1475 audit_inode_child(victim
->d_name
.name
, victim
, dir
);
1477 error
= permission(dir
,MAY_WRITE
| MAY_EXEC
, NULL
);
1482 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1483 IS_IMMUTABLE(victim
->d_inode
))
1486 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1488 if (IS_ROOT(victim
))
1490 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1492 if (IS_DEADDIR(dir
))
1494 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1499 /* Check whether we can create an object with dentry child in directory
1501 * 1. We can't do it if child already exists (open has special treatment for
1502 * this case, but since we are inlined it's OK)
1503 * 2. We can't do it if dir is read-only (done in permission())
1504 * 3. We should have write and exec permissions on dir
1505 * 4. We can't do it if dir is immutable (done in permission())
1507 static inline int may_create(struct inode
*dir
, struct dentry
*child
,
1508 struct nameidata
*nd
)
1512 if (IS_DEADDIR(dir
))
1514 return permission(dir
,MAY_WRITE
| MAY_EXEC
, nd
);
1518 * O_DIRECTORY translates into forcing a directory lookup.
1520 static inline int lookup_flags(unsigned int f
)
1522 unsigned long retval
= LOOKUP_FOLLOW
;
1525 retval
&= ~LOOKUP_FOLLOW
;
1527 if (f
& O_DIRECTORY
)
1528 retval
|= LOOKUP_DIRECTORY
;
1534 * p1 and p2 should be directories on the same fs.
1536 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1541 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1545 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1547 for (p
= p1
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1548 if (p
->d_parent
== p2
) {
1549 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1550 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1555 for (p
= p2
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1556 if (p
->d_parent
== p1
) {
1557 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1558 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1563 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1564 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1568 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1570 mutex_unlock(&p1
->d_inode
->i_mutex
);
1572 mutex_unlock(&p2
->d_inode
->i_mutex
);
1573 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1577 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1578 struct nameidata
*nd
)
1580 int error
= may_create(dir
, dentry
, nd
);
1585 if (!dir
->i_op
|| !dir
->i_op
->create
)
1586 return -EACCES
; /* shouldn't it be ENOSYS? */
1589 error
= security_inode_create(dir
, dentry
, mode
);
1593 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1595 fsnotify_create(dir
, dentry
);
1599 int may_open(struct nameidata
*nd
, int acc_mode
, int flag
)
1601 struct dentry
*dentry
= nd
->path
.dentry
;
1602 struct inode
*inode
= dentry
->d_inode
;
1608 if (S_ISLNK(inode
->i_mode
))
1611 if (S_ISDIR(inode
->i_mode
) && (acc_mode
& MAY_WRITE
))
1615 * FIFO's, sockets and device files are special: they don't
1616 * actually live on the filesystem itself, and as such you
1617 * can write to them even if the filesystem is read-only.
1619 if (S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
1621 } else if (S_ISBLK(inode
->i_mode
) || S_ISCHR(inode
->i_mode
)) {
1622 if (nd
->path
.mnt
->mnt_flags
& MNT_NODEV
)
1626 } else if (IS_RDONLY(inode
) && (acc_mode
& MAY_WRITE
))
1629 error
= vfs_permission(nd
, acc_mode
);
1633 * An append-only file must be opened in append mode for writing.
1635 if (IS_APPEND(inode
)) {
1636 if ((flag
& FMODE_WRITE
) && !(flag
& O_APPEND
))
1642 /* O_NOATIME can only be set by the owner or superuser */
1643 if (flag
& O_NOATIME
)
1644 if (!is_owner_or_cap(inode
))
1648 * Ensure there are no outstanding leases on the file.
1650 error
= break_lease(inode
, flag
);
1654 if (flag
& O_TRUNC
) {
1655 error
= get_write_access(inode
);
1660 * Refuse to truncate files with mandatory locks held on them.
1662 error
= locks_verify_locked(inode
);
1666 error
= do_truncate(dentry
, 0,
1667 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
1670 put_write_access(inode
);
1674 if (flag
& FMODE_WRITE
)
1681 * Be careful about ever adding any more callers of this
1682 * function. Its flags must be in the namei format, not
1683 * what get passed to sys_open().
1685 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
1689 struct dentry
*dir
= nd
->path
.dentry
;
1691 if (!IS_POSIXACL(dir
->d_inode
))
1692 mode
&= ~current
->fs
->umask
;
1693 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
1694 mutex_unlock(&dir
->d_inode
->i_mutex
);
1695 dput(nd
->path
.dentry
);
1696 nd
->path
.dentry
= path
->dentry
;
1699 /* Don't check for write permission, don't truncate */
1700 return may_open(nd
, 0, flag
& ~O_TRUNC
);
1704 * Note that while the flag value (low two bits) for sys_open means:
1709 * it is changed into
1710 * 00 - no permissions needed
1711 * 01 - read-permission
1712 * 10 - write-permission
1714 * for the internal routines (ie open_namei()/follow_link() etc)
1715 * This is more logical, and also allows the 00 "no perm needed"
1716 * to be used for symlinks (where the permissions are checked
1720 static inline int open_to_namei_flags(int flag
)
1722 if ((flag
+1) & O_ACCMODE
)
1728 * Note that the low bits of "flag" aren't the same as in the open
1729 * system call. See open_to_namei_flags().
1731 struct file
*do_filp_open(int dfd
, const char *pathname
,
1732 int open_flag
, int mode
)
1734 struct nameidata nd
;
1735 int acc_mode
, error
;
1739 int flag
= open_to_namei_flags(open_flag
);
1741 acc_mode
= ACC_MODE(flag
);
1743 /* O_TRUNC implies we need access checks for write permissions */
1745 acc_mode
|= MAY_WRITE
;
1747 /* Allow the LSM permission hook to distinguish append
1748 access from general write access. */
1749 if (flag
& O_APPEND
)
1750 acc_mode
|= MAY_APPEND
;
1753 * The simplest case - just a plain lookup.
1755 if (!(flag
& O_CREAT
)) {
1756 error
= path_lookup_open(dfd
, pathname
, lookup_flags(flag
),
1759 return ERR_PTR(error
);
1764 * Create - we need to know the parent.
1766 error
= path_lookup_create(dfd
, pathname
, LOOKUP_PARENT
,
1769 return ERR_PTR(error
);
1772 * We have the parent and last component. First of all, check
1773 * that we are not asked to creat(2) an obvious directory - that
1777 if (nd
.last_type
!= LAST_NORM
|| nd
.last
.name
[nd
.last
.len
])
1780 dir
= nd
.path
.dentry
;
1781 nd
.flags
&= ~LOOKUP_PARENT
;
1782 mutex_lock(&dir
->d_inode
->i_mutex
);
1783 path
.dentry
= lookup_hash(&nd
);
1784 path
.mnt
= nd
.path
.mnt
;
1787 error
= PTR_ERR(path
.dentry
);
1788 if (IS_ERR(path
.dentry
)) {
1789 mutex_unlock(&dir
->d_inode
->i_mutex
);
1793 if (IS_ERR(nd
.intent
.open
.file
)) {
1794 mutex_unlock(&dir
->d_inode
->i_mutex
);
1795 error
= PTR_ERR(nd
.intent
.open
.file
);
1799 /* Negative dentry, just create the file */
1800 if (!path
.dentry
->d_inode
) {
1801 error
= __open_namei_create(&nd
, &path
, flag
, mode
);
1804 return nameidata_to_filp(&nd
, open_flag
);
1808 * It already exists.
1810 mutex_unlock(&dir
->d_inode
->i_mutex
);
1811 audit_inode(pathname
, path
.dentry
);
1817 if (__follow_mount(&path
)) {
1819 if (flag
& O_NOFOLLOW
)
1824 if (!path
.dentry
->d_inode
)
1826 if (path
.dentry
->d_inode
->i_op
&& path
.dentry
->d_inode
->i_op
->follow_link
)
1829 path_to_nameidata(&path
, &nd
);
1831 if (path
.dentry
->d_inode
&& S_ISDIR(path
.dentry
->d_inode
->i_mode
))
1834 error
= may_open(&nd
, acc_mode
, flag
);
1837 return nameidata_to_filp(&nd
, open_flag
);
1840 path_put_conditional(&path
, &nd
);
1842 if (!IS_ERR(nd
.intent
.open
.file
))
1843 release_open_intent(&nd
);
1845 return ERR_PTR(error
);
1849 if (flag
& O_NOFOLLOW
)
1852 * This is subtle. Instead of calling do_follow_link() we do the
1853 * thing by hands. The reason is that this way we have zero link_count
1854 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1855 * After that we have the parent and last component, i.e.
1856 * we are in the same situation as after the first path_walk().
1857 * Well, almost - if the last component is normal we get its copy
1858 * stored in nd->last.name and we will have to putname() it when we
1859 * are done. Procfs-like symlinks just set LAST_BIND.
1861 nd
.flags
|= LOOKUP_PARENT
;
1862 error
= security_inode_follow_link(path
.dentry
, &nd
);
1865 error
= __do_follow_link(&path
, &nd
);
1867 /* Does someone understand code flow here? Or it is only
1868 * me so stupid? Anathema to whoever designed this non-sense
1869 * with "intent.open".
1871 release_open_intent(&nd
);
1872 return ERR_PTR(error
);
1874 nd
.flags
&= ~LOOKUP_PARENT
;
1875 if (nd
.last_type
== LAST_BIND
)
1878 if (nd
.last_type
!= LAST_NORM
)
1880 if (nd
.last
.name
[nd
.last
.len
]) {
1881 __putname(nd
.last
.name
);
1886 __putname(nd
.last
.name
);
1889 dir
= nd
.path
.dentry
;
1890 mutex_lock(&dir
->d_inode
->i_mutex
);
1891 path
.dentry
= lookup_hash(&nd
);
1892 path
.mnt
= nd
.path
.mnt
;
1893 __putname(nd
.last
.name
);
1898 * filp_open - open file and return file pointer
1900 * @filename: path to open
1901 * @flags: open flags as per the open(2) second argument
1902 * @mode: mode for the new file if O_CREAT is set, else ignored
1904 * This is the helper to open a file from kernelspace if you really
1905 * have to. But in generally you should not do this, so please move
1906 * along, nothing to see here..
1908 struct file
*filp_open(const char *filename
, int flags
, int mode
)
1910 return do_filp_open(AT_FDCWD
, filename
, flags
, mode
);
1912 EXPORT_SYMBOL(filp_open
);
1915 * lookup_create - lookup a dentry, creating it if it doesn't exist
1916 * @nd: nameidata info
1917 * @is_dir: directory flag
1919 * Simple function to lookup and return a dentry and create it
1920 * if it doesn't exist. Is SMP-safe.
1922 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1924 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
1926 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
1928 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1930 * Yucky last component or no last component at all?
1931 * (foo/., foo/.., /////)
1933 if (nd
->last_type
!= LAST_NORM
)
1935 nd
->flags
&= ~LOOKUP_PARENT
;
1936 nd
->flags
|= LOOKUP_CREATE
;
1937 nd
->intent
.open
.flags
= O_EXCL
;
1940 * Do the final lookup.
1942 dentry
= lookup_hash(nd
);
1947 * Special case - lookup gave negative, but... we had foo/bar/
1948 * From the vfs_mknod() POV we just have a negative dentry -
1949 * all is fine. Let's be bastards - you had / on the end, you've
1950 * been asking for (non-existent) directory. -ENOENT for you.
1952 if (!is_dir
&& nd
->last
.name
[nd
->last
.len
] && !dentry
->d_inode
)
1957 dentry
= ERR_PTR(-ENOENT
);
1961 EXPORT_SYMBOL_GPL(lookup_create
);
1963 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1965 int error
= may_create(dir
, dentry
, NULL
);
1970 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
1973 if (!dir
->i_op
|| !dir
->i_op
->mknod
)
1976 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
1981 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
1983 fsnotify_create(dir
, dentry
);
1987 static int may_mknod(mode_t mode
)
1989 switch (mode
& S_IFMT
) {
1995 case 0: /* zero mode translates to S_IFREG */
2004 asmlinkage
long sys_mknodat(int dfd
, const char __user
*filename
, int mode
,
2009 struct dentry
* dentry
;
2010 struct nameidata nd
;
2014 tmp
= getname(filename
);
2016 return PTR_ERR(tmp
);
2018 error
= do_path_lookup(dfd
, tmp
, LOOKUP_PARENT
, &nd
);
2021 dentry
= lookup_create(&nd
, 0);
2022 if (IS_ERR(dentry
)) {
2023 error
= PTR_ERR(dentry
);
2026 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2027 mode
&= ~current
->fs
->umask
;
2028 error
= may_mknod(mode
);
2031 error
= mnt_want_write(nd
.path
.mnt
);
2034 switch (mode
& S_IFMT
) {
2035 case 0: case S_IFREG
:
2036 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2038 case S_IFCHR
: case S_IFBLK
:
2039 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2040 new_decode_dev(dev
));
2042 case S_IFIFO
: case S_IFSOCK
:
2043 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2046 mnt_drop_write(nd
.path
.mnt
);
2050 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2058 asmlinkage
long sys_mknod(const char __user
*filename
, int mode
, unsigned dev
)
2060 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2063 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2065 int error
= may_create(dir
, dentry
, NULL
);
2070 if (!dir
->i_op
|| !dir
->i_op
->mkdir
)
2073 mode
&= (S_IRWXUGO
|S_ISVTX
);
2074 error
= security_inode_mkdir(dir
, dentry
, mode
);
2079 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2081 fsnotify_mkdir(dir
, dentry
);
2085 asmlinkage
long sys_mkdirat(int dfd
, const char __user
*pathname
, int mode
)
2089 struct dentry
*dentry
;
2090 struct nameidata nd
;
2092 tmp
= getname(pathname
);
2093 error
= PTR_ERR(tmp
);
2097 error
= do_path_lookup(dfd
, tmp
, LOOKUP_PARENT
, &nd
);
2100 dentry
= lookup_create(&nd
, 1);
2101 error
= PTR_ERR(dentry
);
2105 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2106 mode
&= ~current
->fs
->umask
;
2107 error
= mnt_want_write(nd
.path
.mnt
);
2110 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2111 mnt_drop_write(nd
.path
.mnt
);
2115 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2123 asmlinkage
long sys_mkdir(const char __user
*pathname
, int mode
)
2125 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2129 * We try to drop the dentry early: we should have
2130 * a usage count of 2 if we're the only user of this
2131 * dentry, and if that is true (possibly after pruning
2132 * the dcache), then we drop the dentry now.
2134 * A low-level filesystem can, if it choses, legally
2137 * if (!d_unhashed(dentry))
2140 * if it cannot handle the case of removing a directory
2141 * that is still in use by something else..
2143 void dentry_unhash(struct dentry
*dentry
)
2146 shrink_dcache_parent(dentry
);
2147 spin_lock(&dcache_lock
);
2148 spin_lock(&dentry
->d_lock
);
2149 if (atomic_read(&dentry
->d_count
) == 2)
2151 spin_unlock(&dentry
->d_lock
);
2152 spin_unlock(&dcache_lock
);
2155 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2157 int error
= may_delete(dir
, dentry
, 1);
2162 if (!dir
->i_op
|| !dir
->i_op
->rmdir
)
2167 mutex_lock(&dentry
->d_inode
->i_mutex
);
2168 dentry_unhash(dentry
);
2169 if (d_mountpoint(dentry
))
2172 error
= security_inode_rmdir(dir
, dentry
);
2174 error
= dir
->i_op
->rmdir(dir
, dentry
);
2176 dentry
->d_inode
->i_flags
|= S_DEAD
;
2179 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2188 static long do_rmdir(int dfd
, const char __user
*pathname
)
2192 struct dentry
*dentry
;
2193 struct nameidata nd
;
2195 name
= getname(pathname
);
2197 return PTR_ERR(name
);
2199 error
= do_path_lookup(dfd
, name
, LOOKUP_PARENT
, &nd
);
2203 switch(nd
.last_type
) {
2214 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2215 dentry
= lookup_hash(&nd
);
2216 error
= PTR_ERR(dentry
);
2219 error
= mnt_want_write(nd
.path
.mnt
);
2222 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2223 mnt_drop_write(nd
.path
.mnt
);
2227 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2235 asmlinkage
long sys_rmdir(const char __user
*pathname
)
2237 return do_rmdir(AT_FDCWD
, pathname
);
2240 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2242 int error
= may_delete(dir
, dentry
, 0);
2247 if (!dir
->i_op
|| !dir
->i_op
->unlink
)
2252 mutex_lock(&dentry
->d_inode
->i_mutex
);
2253 if (d_mountpoint(dentry
))
2256 error
= security_inode_unlink(dir
, dentry
);
2258 error
= dir
->i_op
->unlink(dir
, dentry
);
2260 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2262 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2263 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2264 fsnotify_link_count(dentry
->d_inode
);
2272 * Make sure that the actual truncation of the file will occur outside its
2273 * directory's i_mutex. Truncate can take a long time if there is a lot of
2274 * writeout happening, and we don't want to prevent access to the directory
2275 * while waiting on the I/O.
2277 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2281 struct dentry
*dentry
;
2282 struct nameidata nd
;
2283 struct inode
*inode
= NULL
;
2285 name
= getname(pathname
);
2287 return PTR_ERR(name
);
2289 error
= do_path_lookup(dfd
, name
, LOOKUP_PARENT
, &nd
);
2293 if (nd
.last_type
!= LAST_NORM
)
2295 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2296 dentry
= lookup_hash(&nd
);
2297 error
= PTR_ERR(dentry
);
2298 if (!IS_ERR(dentry
)) {
2299 /* Why not before? Because we want correct error value */
2300 if (nd
.last
.name
[nd
.last
.len
])
2302 inode
= dentry
->d_inode
;
2304 atomic_inc(&inode
->i_count
);
2305 error
= mnt_want_write(nd
.path
.mnt
);
2308 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2309 mnt_drop_write(nd
.path
.mnt
);
2313 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2315 iput(inode
); /* truncate the inode here */
2323 error
= !dentry
->d_inode
? -ENOENT
:
2324 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2328 asmlinkage
long sys_unlinkat(int dfd
, const char __user
*pathname
, int flag
)
2330 if ((flag
& ~AT_REMOVEDIR
) != 0)
2333 if (flag
& AT_REMOVEDIR
)
2334 return do_rmdir(dfd
, pathname
);
2336 return do_unlinkat(dfd
, pathname
);
2339 asmlinkage
long sys_unlink(const char __user
*pathname
)
2341 return do_unlinkat(AT_FDCWD
, pathname
);
2344 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
, int mode
)
2346 int error
= may_create(dir
, dentry
, NULL
);
2351 if (!dir
->i_op
|| !dir
->i_op
->symlink
)
2354 error
= security_inode_symlink(dir
, dentry
, oldname
);
2359 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2361 fsnotify_create(dir
, dentry
);
2365 asmlinkage
long sys_symlinkat(const char __user
*oldname
,
2366 int newdfd
, const char __user
*newname
)
2371 struct dentry
*dentry
;
2372 struct nameidata nd
;
2374 from
= getname(oldname
);
2376 return PTR_ERR(from
);
2377 to
= getname(newname
);
2378 error
= PTR_ERR(to
);
2382 error
= do_path_lookup(newdfd
, to
, LOOKUP_PARENT
, &nd
);
2385 dentry
= lookup_create(&nd
, 0);
2386 error
= PTR_ERR(dentry
);
2390 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
, S_IALLUGO
);
2393 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2402 asmlinkage
long sys_symlink(const char __user
*oldname
, const char __user
*newname
)
2404 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2407 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2409 struct inode
*inode
= old_dentry
->d_inode
;
2415 error
= may_create(dir
, new_dentry
, NULL
);
2419 if (dir
->i_sb
!= inode
->i_sb
)
2423 * A link to an append-only or immutable file cannot be created.
2425 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2427 if (!dir
->i_op
|| !dir
->i_op
->link
)
2429 if (S_ISDIR(old_dentry
->d_inode
->i_mode
))
2432 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2436 mutex_lock(&old_dentry
->d_inode
->i_mutex
);
2438 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2439 mutex_unlock(&old_dentry
->d_inode
->i_mutex
);
2441 fsnotify_link(dir
, old_dentry
->d_inode
, new_dentry
);
2446 * Hardlinks are often used in delicate situations. We avoid
2447 * security-related surprises by not following symlinks on the
2450 * We don't follow them on the oldname either to be compatible
2451 * with linux 2.0, and to avoid hard-linking to directories
2452 * and other special files. --ADM
2454 asmlinkage
long sys_linkat(int olddfd
, const char __user
*oldname
,
2455 int newdfd
, const char __user
*newname
,
2458 struct dentry
*new_dentry
;
2459 struct nameidata nd
, old_nd
;
2463 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
2466 to
= getname(newname
);
2470 error
= __user_walk_fd(olddfd
, oldname
,
2471 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
2475 error
= do_path_lookup(newdfd
, to
, LOOKUP_PARENT
, &nd
);
2479 if (old_nd
.path
.mnt
!= nd
.path
.mnt
)
2481 new_dentry
= lookup_create(&nd
, 0);
2482 error
= PTR_ERR(new_dentry
);
2483 if (IS_ERR(new_dentry
))
2485 error
= vfs_link(old_nd
.path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2488 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2492 path_put(&old_nd
.path
);
2499 asmlinkage
long sys_link(const char __user
*oldname
, const char __user
*newname
)
2501 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2505 * The worst of all namespace operations - renaming directory. "Perverted"
2506 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2508 * a) we can get into loop creation. Check is done in is_subdir().
2509 * b) race potential - two innocent renames can create a loop together.
2510 * That's where 4.4 screws up. Current fix: serialization on
2511 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2513 * c) we have to lock _three_ objects - parents and victim (if it exists).
2514 * And that - after we got ->i_mutex on parents (until then we don't know
2515 * whether the target exists). Solution: try to be smart with locking
2516 * order for inodes. We rely on the fact that tree topology may change
2517 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2518 * move will be locked. Thus we can rank directories by the tree
2519 * (ancestors first) and rank all non-directories after them.
2520 * That works since everybody except rename does "lock parent, lookup,
2521 * lock child" and rename is under ->s_vfs_rename_mutex.
2522 * HOWEVER, it relies on the assumption that any object with ->lookup()
2523 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2524 * we'd better make sure that there's no link(2) for them.
2525 * d) some filesystems don't support opened-but-unlinked directories,
2526 * either because of layout or because they are not ready to deal with
2527 * all cases correctly. The latter will be fixed (taking this sort of
2528 * stuff into VFS), but the former is not going away. Solution: the same
2529 * trick as in rmdir().
2530 * e) conversion from fhandle to dentry may come in the wrong moment - when
2531 * we are removing the target. Solution: we will have to grab ->i_mutex
2532 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2533 * ->i_mutex on parents, which works but leads to some truely excessive
2536 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2537 struct inode
*new_dir
, struct dentry
*new_dentry
)
2540 struct inode
*target
;
2543 * If we are going to change the parent - check write permissions,
2544 * we'll need to flip '..'.
2546 if (new_dir
!= old_dir
) {
2547 error
= permission(old_dentry
->d_inode
, MAY_WRITE
, NULL
);
2552 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2556 target
= new_dentry
->d_inode
;
2558 mutex_lock(&target
->i_mutex
);
2559 dentry_unhash(new_dentry
);
2561 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2564 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2567 target
->i_flags
|= S_DEAD
;
2568 mutex_unlock(&target
->i_mutex
);
2569 if (d_unhashed(new_dentry
))
2570 d_rehash(new_dentry
);
2574 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2575 d_move(old_dentry
,new_dentry
);
2579 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
2580 struct inode
*new_dir
, struct dentry
*new_dentry
)
2582 struct inode
*target
;
2585 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2590 target
= new_dentry
->d_inode
;
2592 mutex_lock(&target
->i_mutex
);
2593 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2596 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2598 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2599 d_move(old_dentry
, new_dentry
);
2602 mutex_unlock(&target
->i_mutex
);
2607 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2608 struct inode
*new_dir
, struct dentry
*new_dentry
)
2611 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2612 const char *old_name
;
2614 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
2617 error
= may_delete(old_dir
, old_dentry
, is_dir
);
2621 if (!new_dentry
->d_inode
)
2622 error
= may_create(new_dir
, new_dentry
, NULL
);
2624 error
= may_delete(new_dir
, new_dentry
, is_dir
);
2628 if (!old_dir
->i_op
|| !old_dir
->i_op
->rename
)
2631 DQUOT_INIT(old_dir
);
2632 DQUOT_INIT(new_dir
);
2634 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
2637 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
2639 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
2641 const char *new_name
= old_dentry
->d_name
.name
;
2642 fsnotify_move(old_dir
, new_dir
, old_name
, new_name
, is_dir
,
2643 new_dentry
->d_inode
, old_dentry
);
2645 fsnotify_oldname_free(old_name
);
2650 static int do_rename(int olddfd
, const char *oldname
,
2651 int newdfd
, const char *newname
)
2654 struct dentry
* old_dir
, * new_dir
;
2655 struct dentry
* old_dentry
, *new_dentry
;
2656 struct dentry
* trap
;
2657 struct nameidata oldnd
, newnd
;
2659 error
= do_path_lookup(olddfd
, oldname
, LOOKUP_PARENT
, &oldnd
);
2663 error
= do_path_lookup(newdfd
, newname
, LOOKUP_PARENT
, &newnd
);
2668 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
2671 old_dir
= oldnd
.path
.dentry
;
2673 if (oldnd
.last_type
!= LAST_NORM
)
2676 new_dir
= newnd
.path
.dentry
;
2677 if (newnd
.last_type
!= LAST_NORM
)
2680 trap
= lock_rename(new_dir
, old_dir
);
2682 old_dentry
= lookup_hash(&oldnd
);
2683 error
= PTR_ERR(old_dentry
);
2684 if (IS_ERR(old_dentry
))
2686 /* source must exist */
2688 if (!old_dentry
->d_inode
)
2690 /* unless the source is a directory trailing slashes give -ENOTDIR */
2691 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
2693 if (oldnd
.last
.name
[oldnd
.last
.len
])
2695 if (newnd
.last
.name
[newnd
.last
.len
])
2698 /* source should not be ancestor of target */
2700 if (old_dentry
== trap
)
2702 new_dentry
= lookup_hash(&newnd
);
2703 error
= PTR_ERR(new_dentry
);
2704 if (IS_ERR(new_dentry
))
2706 /* target should not be an ancestor of source */
2708 if (new_dentry
== trap
)
2711 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
2712 new_dir
->d_inode
, new_dentry
);
2718 unlock_rename(new_dir
, old_dir
);
2720 path_put(&newnd
.path
);
2722 path_put(&oldnd
.path
);
2727 asmlinkage
long sys_renameat(int olddfd
, const char __user
*oldname
,
2728 int newdfd
, const char __user
*newname
)
2734 from
= getname(oldname
);
2736 return PTR_ERR(from
);
2737 to
= getname(newname
);
2738 error
= PTR_ERR(to
);
2740 error
= do_rename(olddfd
, from
, newdfd
, to
);
2747 asmlinkage
long sys_rename(const char __user
*oldname
, const char __user
*newname
)
2749 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
2752 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
2756 len
= PTR_ERR(link
);
2761 if (len
> (unsigned) buflen
)
2763 if (copy_to_user(buffer
, link
, len
))
2770 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2771 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2772 * using) it for any given inode is up to filesystem.
2774 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2776 struct nameidata nd
;
2780 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
2781 if (!IS_ERR(cookie
)) {
2782 int res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
2783 if (dentry
->d_inode
->i_op
->put_link
)
2784 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
2785 cookie
= ERR_PTR(res
);
2787 return PTR_ERR(cookie
);
2790 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
2792 return __vfs_follow_link(nd
, link
);
2795 /* get the link contents into pagecache */
2796 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
2799 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
2800 page
= read_mapping_page(mapping
, 0, NULL
);
2807 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2809 struct page
*page
= NULL
;
2810 char *s
= page_getlink(dentry
, &page
);
2811 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
2814 page_cache_release(page
);
2819 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
2821 struct page
*page
= NULL
;
2822 nd_set_link(nd
, page_getlink(dentry
, &page
));
2826 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2828 struct page
*page
= cookie
;
2832 page_cache_release(page
);
2836 int __page_symlink(struct inode
*inode
, const char *symname
, int len
,
2839 struct address_space
*mapping
= inode
->i_mapping
;
2846 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
2847 AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
2851 kaddr
= kmap_atomic(page
, KM_USER0
);
2852 memcpy(kaddr
, symname
, len
-1);
2853 kunmap_atomic(kaddr
, KM_USER0
);
2855 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
2862 mark_inode_dirty(inode
);
2868 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
2870 return __page_symlink(inode
, symname
, len
,
2871 mapping_gfp_mask(inode
->i_mapping
));
2874 const struct inode_operations page_symlink_inode_operations
= {
2875 .readlink
= generic_readlink
,
2876 .follow_link
= page_follow_link_light
,
2877 .put_link
= page_put_link
,
2880 EXPORT_SYMBOL(__user_walk
);
2881 EXPORT_SYMBOL(__user_walk_fd
);
2882 EXPORT_SYMBOL(follow_down
);
2883 EXPORT_SYMBOL(follow_up
);
2884 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
2885 EXPORT_SYMBOL(getname
);
2886 EXPORT_SYMBOL(lock_rename
);
2887 EXPORT_SYMBOL(lookup_one_len
);
2888 EXPORT_SYMBOL(page_follow_link_light
);
2889 EXPORT_SYMBOL(page_put_link
);
2890 EXPORT_SYMBOL(page_readlink
);
2891 EXPORT_SYMBOL(__page_symlink
);
2892 EXPORT_SYMBOL(page_symlink
);
2893 EXPORT_SYMBOL(page_symlink_inode_operations
);
2894 EXPORT_SYMBOL(path_lookup
);
2895 EXPORT_SYMBOL(vfs_path_lookup
);
2896 EXPORT_SYMBOL(permission
);
2897 EXPORT_SYMBOL(vfs_permission
);
2898 EXPORT_SYMBOL(file_permission
);
2899 EXPORT_SYMBOL(unlock_rename
);
2900 EXPORT_SYMBOL(vfs_create
);
2901 EXPORT_SYMBOL(vfs_follow_link
);
2902 EXPORT_SYMBOL(vfs_link
);
2903 EXPORT_SYMBOL(vfs_mkdir
);
2904 EXPORT_SYMBOL(vfs_mknod
);
2905 EXPORT_SYMBOL(generic_permission
);
2906 EXPORT_SYMBOL(vfs_readlink
);
2907 EXPORT_SYMBOL(vfs_rename
);
2908 EXPORT_SYMBOL(vfs_rmdir
);
2909 EXPORT_SYMBOL(vfs_symlink
);
2910 EXPORT_SYMBOL(vfs_unlink
);
2911 EXPORT_SYMBOL(dentry_unhash
);
2912 EXPORT_SYMBOL(generic_readlink
);