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/smp_lock.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/file.h>
32 #include <asm/namei.h>
33 #include <asm/uaccess.h>
35 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
37 /* [Feb-1997 T. Schoebel-Theuer]
38 * Fundamental changes in the pathname lookup mechanisms (namei)
39 * were necessary because of omirr. The reason is that omirr needs
40 * to know the _real_ pathname, not the user-supplied one, in case
41 * of symlinks (and also when transname replacements occur).
43 * The new code replaces the old recursive symlink resolution with
44 * an iterative one (in case of non-nested symlink chains). It does
45 * this with calls to <fs>_follow_link().
46 * As a side effect, dir_namei(), _namei() and follow_link() are now
47 * replaced with a single function lookup_dentry() that can handle all
48 * the special cases of the former code.
50 * With the new dcache, the pathname is stored at each inode, at least as
51 * long as the refcount of the inode is positive. As a side effect, the
52 * size of the dcache depends on the inode cache and thus is dynamic.
54 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
55 * resolution to correspond with current state of the code.
57 * Note that the symlink resolution is not *completely* iterative.
58 * There is still a significant amount of tail- and mid- recursion in
59 * the algorithm. Also, note that <fs>_readlink() is not used in
60 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
61 * may return different results than <fs>_follow_link(). Many virtual
62 * filesystems (including /proc) exhibit this behavior.
65 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
66 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
67 * and the name already exists in form of a symlink, try to create the new
68 * name indicated by the symlink. The old code always complained that the
69 * name already exists, due to not following the symlink even if its target
70 * is nonexistent. The new semantics affects also mknod() and link() when
71 * the name is a symlink pointing to a non-existant name.
73 * I don't know which semantics is the right one, since I have no access
74 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
75 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
76 * "old" one. Personally, I think the new semantics is much more logical.
77 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
78 * file does succeed in both HP-UX and SunOs, but not in Solaris
79 * and in the old Linux semantics.
82 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
83 * semantics. See the comments in "open_namei" and "do_link" below.
85 * [10-Sep-98 Alan Modra] Another symlink change.
88 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
89 * inside the path - always follow.
90 * in the last component in creation/removal/renaming - never follow.
91 * if LOOKUP_FOLLOW passed - follow.
92 * if the pathname has trailing slashes - follow.
93 * otherwise - don't follow.
94 * (applied in that order).
96 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
97 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
98 * During the 2.4 we need to fix the userland stuff depending on it -
99 * hopefully we will be able to get rid of that wart in 2.5. So far only
100 * XEmacs seems to be relying on it...
103 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
104 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
105 * any extra contention...
108 /* In order to reduce some races, while at the same time doing additional
109 * checking and hopefully speeding things up, we copy filenames to the
110 * kernel data space before using them..
112 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
113 * PATH_MAX includes the nul terminator --RR.
115 static inline int do_getname(const char __user
*filename
, char *page
)
118 unsigned long len
= PATH_MAX
;
120 if (!segment_eq(get_fs(), KERNEL_DS
)) {
121 if ((unsigned long) filename
>= TASK_SIZE
)
123 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
124 len
= TASK_SIZE
- (unsigned long) filename
;
127 retval
= strncpy_from_user(page
, filename
, len
);
131 return -ENAMETOOLONG
;
137 char * getname(const char __user
* filename
)
141 result
= ERR_PTR(-ENOMEM
);
144 int retval
= do_getname(filename
, tmp
);
149 result
= ERR_PTR(retval
);
152 audit_getname(result
);
156 #ifdef CONFIG_AUDITSYSCALL
157 void putname(const char *name
)
159 if (unlikely(current
->audit_context
))
164 EXPORT_SYMBOL(putname
);
169 * generic_permission - check for access rights on a Posix-like filesystem
170 * @inode: inode to check access rights for
171 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
172 * @check_acl: optional callback to check for Posix ACLs
174 * Used to check for read/write/execute permissions on a file.
175 * We use "fsuid" for this, letting us set arbitrary permissions
176 * for filesystem access without changing the "normal" uids which
177 * are used for other things..
179 int generic_permission(struct inode
*inode
, int mask
,
180 int (*check_acl
)(struct inode
*inode
, int mask
))
182 umode_t mode
= inode
->i_mode
;
184 if (current
->fsuid
== inode
->i_uid
)
187 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
188 int error
= check_acl(inode
, mask
);
189 if (error
== -EACCES
)
190 goto check_capabilities
;
191 else if (error
!= -EAGAIN
)
195 if (in_group_p(inode
->i_gid
))
200 * If the DACs are ok we don't need any capability check.
202 if (((mode
& mask
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
)) == mask
))
207 * Read/write DACs are always overridable.
208 * Executable DACs are overridable if at least one exec bit is set.
210 if (!(mask
& MAY_EXEC
) ||
211 (inode
->i_mode
& S_IXUGO
) || S_ISDIR(inode
->i_mode
))
212 if (capable(CAP_DAC_OVERRIDE
))
216 * Searching includes executable on directories, else just read.
218 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
219 if (capable(CAP_DAC_READ_SEARCH
))
225 int permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
229 if (mask
& MAY_WRITE
) {
230 umode_t mode
= inode
->i_mode
;
233 * Nobody gets write access to a read-only fs.
235 if (IS_RDONLY(inode
) &&
236 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
240 * Nobody gets write access to an immutable file.
242 if (IS_IMMUTABLE(inode
))
247 /* Ordinary permission routines do not understand MAY_APPEND. */
248 submask
= mask
& ~MAY_APPEND
;
249 if (inode
->i_op
&& inode
->i_op
->permission
)
250 retval
= inode
->i_op
->permission(inode
, submask
, nd
);
252 retval
= generic_permission(inode
, submask
, NULL
);
256 return security_inode_permission(inode
, mask
, nd
);
260 * vfs_permission - check for access rights to a given path
261 * @nd: lookup result that describes the path
262 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
264 * Used to check for read/write/execute permissions on a path.
265 * We use "fsuid" for this, letting us set arbitrary permissions
266 * for filesystem access without changing the "normal" uids which
267 * are used for other things.
269 int vfs_permission(struct nameidata
*nd
, int mask
)
271 return permission(nd
->dentry
->d_inode
, mask
, nd
);
275 * file_permission - check for additional access rights to a given file
276 * @file: file to check access rights for
277 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
279 * Used to check for read/write/execute permissions on an already opened
283 * Do not use this function in new code. All access checks should
284 * be done using vfs_permission().
286 int file_permission(struct file
*file
, int mask
)
288 return permission(file
->f_dentry
->d_inode
, mask
, NULL
);
292 * get_write_access() gets write permission for a file.
293 * put_write_access() releases this write permission.
294 * This is used for regular files.
295 * We cannot support write (and maybe mmap read-write shared) accesses and
296 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
297 * can have the following values:
298 * 0: no writers, no VM_DENYWRITE mappings
299 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
300 * > 0: (i_writecount) users are writing to the file.
302 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
303 * except for the cases where we don't hold i_writecount yet. Then we need to
304 * use {get,deny}_write_access() - these functions check the sign and refuse
305 * to do the change if sign is wrong. Exclusion between them is provided by
306 * the inode->i_lock spinlock.
309 int get_write_access(struct inode
* inode
)
311 spin_lock(&inode
->i_lock
);
312 if (atomic_read(&inode
->i_writecount
) < 0) {
313 spin_unlock(&inode
->i_lock
);
316 atomic_inc(&inode
->i_writecount
);
317 spin_unlock(&inode
->i_lock
);
322 int deny_write_access(struct file
* file
)
324 struct inode
*inode
= file
->f_dentry
->d_inode
;
326 spin_lock(&inode
->i_lock
);
327 if (atomic_read(&inode
->i_writecount
) > 0) {
328 spin_unlock(&inode
->i_lock
);
331 atomic_dec(&inode
->i_writecount
);
332 spin_unlock(&inode
->i_lock
);
337 void path_release(struct nameidata
*nd
)
344 * umount() mustn't call path_release()/mntput() as that would clear
347 void path_release_on_umount(struct nameidata
*nd
)
350 mntput_no_expire(nd
->mnt
);
354 * release_open_intent - free up open intent resources
355 * @nd: pointer to nameidata
357 void release_open_intent(struct nameidata
*nd
)
359 if (nd
->intent
.open
.file
->f_dentry
== NULL
)
360 put_filp(nd
->intent
.open
.file
);
362 fput(nd
->intent
.open
.file
);
366 * Internal lookup() using the new generic dcache.
369 static struct dentry
* cached_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
371 struct dentry
* dentry
= __d_lookup(parent
, name
);
373 /* lockess __d_lookup may fail due to concurrent d_move()
374 * in some unrelated directory, so try with d_lookup
377 dentry
= d_lookup(parent
, name
);
379 if (dentry
&& dentry
->d_op
&& dentry
->d_op
->d_revalidate
) {
380 if (!dentry
->d_op
->d_revalidate(dentry
, nd
) && !d_invalidate(dentry
)) {
389 * Short-cut version of permission(), for calling by
390 * path_walk(), when dcache lock is held. Combines parts
391 * of permission() and generic_permission(), and tests ONLY for
392 * MAY_EXEC permission.
394 * If appropriate, check DAC only. If not appropriate, or
395 * short-cut DAC fails, then call permission() to do more
396 * complete permission check.
398 static inline int exec_permission_lite(struct inode
*inode
,
399 struct nameidata
*nd
)
401 umode_t mode
= inode
->i_mode
;
403 if (inode
->i_op
&& inode
->i_op
->permission
)
406 if (current
->fsuid
== inode
->i_uid
)
408 else if (in_group_p(inode
->i_gid
))
414 if ((inode
->i_mode
& S_IXUGO
) && capable(CAP_DAC_OVERRIDE
))
417 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_OVERRIDE
))
420 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_READ_SEARCH
))
425 return security_inode_permission(inode
, MAY_EXEC
, nd
);
429 * This is called when everything else fails, and we actually have
430 * to go to the low-level filesystem to find out what we should do..
432 * We get the directory semaphore, and after getting that we also
433 * make sure that nobody added the entry to the dcache in the meantime..
436 static struct dentry
* real_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
438 struct dentry
* result
;
439 struct inode
*dir
= parent
->d_inode
;
443 * First re-do the cached lookup just in case it was created
444 * while we waited for the directory semaphore..
446 * FIXME! This could use version numbering or similar to
447 * avoid unnecessary cache lookups.
449 * The "dcache_lock" is purely to protect the RCU list walker
450 * from concurrent renames at this point (we mustn't get false
451 * negatives from the RCU list walk here, unlike the optimistic
454 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
456 result
= d_lookup(parent
, name
);
458 struct dentry
* dentry
= d_alloc(parent
, name
);
459 result
= ERR_PTR(-ENOMEM
);
461 result
= dir
->i_op
->lookup(dir
, dentry
, nd
);
472 * Uhhuh! Nasty case: the cache was re-populated while
473 * we waited on the semaphore. Need to revalidate.
476 if (result
->d_op
&& result
->d_op
->d_revalidate
) {
477 if (!result
->d_op
->d_revalidate(result
, nd
) && !d_invalidate(result
)) {
479 result
= ERR_PTR(-ENOENT
);
485 static int __emul_lookup_dentry(const char *, struct nameidata
*);
489 walk_init_root(const char *name
, struct nameidata
*nd
)
491 read_lock(¤t
->fs
->lock
);
492 if (current
->fs
->altroot
&& !(nd
->flags
& LOOKUP_NOALT
)) {
493 nd
->mnt
= mntget(current
->fs
->altrootmnt
);
494 nd
->dentry
= dget(current
->fs
->altroot
);
495 read_unlock(¤t
->fs
->lock
);
496 if (__emul_lookup_dentry(name
,nd
))
498 read_lock(¤t
->fs
->lock
);
500 nd
->mnt
= mntget(current
->fs
->rootmnt
);
501 nd
->dentry
= dget(current
->fs
->root
);
502 read_unlock(¤t
->fs
->lock
);
506 static inline int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
515 if (!walk_init_root(link
, nd
))
516 /* weird __emul_prefix() stuff did it */
519 res
= link_path_walk(link
, nd
);
521 if (nd
->depth
|| res
|| nd
->last_type
!=LAST_NORM
)
524 * If it is an iterative symlinks resolution in open_namei() we
525 * have to copy the last component. And all that crap because of
526 * bloody create() on broken symlinks. Furrfu...
529 if (unlikely(!name
)) {
533 strcpy(name
, nd
->last
.name
);
534 nd
->last
.name
= name
;
538 return PTR_ERR(link
);
542 struct vfsmount
*mnt
;
543 struct dentry
*dentry
;
546 static inline int __do_follow_link(struct path
*path
, struct nameidata
*nd
)
550 struct dentry
*dentry
= path
->dentry
;
552 touch_atime(path
->mnt
, dentry
);
553 nd_set_link(nd
, NULL
);
555 if (path
->mnt
== nd
->mnt
)
557 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
558 error
= PTR_ERR(cookie
);
559 if (!IS_ERR(cookie
)) {
560 char *s
= nd_get_link(nd
);
563 error
= __vfs_follow_link(nd
, s
);
564 if (dentry
->d_inode
->i_op
->put_link
)
565 dentry
->d_inode
->i_op
->put_link(dentry
, nd
, cookie
);
573 static inline void dput_path(struct path
*path
, struct nameidata
*nd
)
576 if (path
->mnt
!= nd
->mnt
)
580 static inline void path_to_nameidata(struct path
*path
, struct nameidata
*nd
)
583 if (nd
->mnt
!= path
->mnt
)
586 nd
->dentry
= path
->dentry
;
590 * This limits recursive symlink follows to 8, while
591 * limiting consecutive symlinks to 40.
593 * Without that kind of total limit, nasty chains of consecutive
594 * symlinks can cause almost arbitrarily long lookups.
596 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
599 if (current
->link_count
>= MAX_NESTED_LINKS
)
601 if (current
->total_link_count
>= 40)
603 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
605 err
= security_inode_follow_link(path
->dentry
, nd
);
608 current
->link_count
++;
609 current
->total_link_count
++;
611 err
= __do_follow_link(path
, nd
);
612 current
->link_count
--;
621 int follow_up(struct vfsmount
**mnt
, struct dentry
**dentry
)
623 struct vfsmount
*parent
;
624 struct dentry
*mountpoint
;
625 spin_lock(&vfsmount_lock
);
626 parent
=(*mnt
)->mnt_parent
;
627 if (parent
== *mnt
) {
628 spin_unlock(&vfsmount_lock
);
632 mountpoint
=dget((*mnt
)->mnt_mountpoint
);
633 spin_unlock(&vfsmount_lock
);
635 *dentry
= mountpoint
;
641 /* no need for dcache_lock, as serialization is taken care in
644 static int __follow_mount(struct path
*path
)
647 while (d_mountpoint(path
->dentry
)) {
648 struct vfsmount
*mounted
= lookup_mnt(path
->mnt
, path
->dentry
);
655 path
->dentry
= dget(mounted
->mnt_root
);
661 static void follow_mount(struct vfsmount
**mnt
, struct dentry
**dentry
)
663 while (d_mountpoint(*dentry
)) {
664 struct vfsmount
*mounted
= lookup_mnt(*mnt
, *dentry
);
670 *dentry
= dget(mounted
->mnt_root
);
674 /* no need for dcache_lock, as serialization is taken care in
677 int follow_down(struct vfsmount
**mnt
, struct dentry
**dentry
)
679 struct vfsmount
*mounted
;
681 mounted
= lookup_mnt(*mnt
, *dentry
);
686 *dentry
= dget(mounted
->mnt_root
);
692 static inline void follow_dotdot(struct nameidata
*nd
)
695 struct vfsmount
*parent
;
696 struct dentry
*old
= nd
->dentry
;
698 read_lock(¤t
->fs
->lock
);
699 if (nd
->dentry
== current
->fs
->root
&&
700 nd
->mnt
== current
->fs
->rootmnt
) {
701 read_unlock(¤t
->fs
->lock
);
704 read_unlock(¤t
->fs
->lock
);
705 spin_lock(&dcache_lock
);
706 if (nd
->dentry
!= nd
->mnt
->mnt_root
) {
707 nd
->dentry
= dget(nd
->dentry
->d_parent
);
708 spin_unlock(&dcache_lock
);
712 spin_unlock(&dcache_lock
);
713 spin_lock(&vfsmount_lock
);
714 parent
= nd
->mnt
->mnt_parent
;
715 if (parent
== nd
->mnt
) {
716 spin_unlock(&vfsmount_lock
);
720 nd
->dentry
= dget(nd
->mnt
->mnt_mountpoint
);
721 spin_unlock(&vfsmount_lock
);
726 follow_mount(&nd
->mnt
, &nd
->dentry
);
730 * It's more convoluted than I'd like it to be, but... it's still fairly
731 * small and for now I'd prefer to have fast path as straight as possible.
732 * It _is_ time-critical.
734 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
737 struct vfsmount
*mnt
= nd
->mnt
;
738 struct dentry
*dentry
= __d_lookup(nd
->dentry
, name
);
742 if (dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
743 goto need_revalidate
;
746 path
->dentry
= dentry
;
747 __follow_mount(path
);
751 dentry
= real_lookup(nd
->dentry
, name
, nd
);
757 if (dentry
->d_op
->d_revalidate(dentry
, nd
))
759 if (d_invalidate(dentry
))
765 return PTR_ERR(dentry
);
770 * This is the basic name resolution function, turning a pathname into
771 * the final dentry. We expect 'base' to be positive and a directory.
773 * Returns 0 and nd will have valid dentry and mnt on success.
774 * Returns error and drops reference to input namei data on failure.
776 static fastcall
int __link_path_walk(const char * name
, struct nameidata
*nd
)
781 unsigned int lookup_flags
= nd
->flags
;
788 inode
= nd
->dentry
->d_inode
;
790 lookup_flags
= LOOKUP_FOLLOW
;
792 /* At this point we know we have a real path component. */
798 nd
->flags
|= LOOKUP_CONTINUE
;
799 err
= exec_permission_lite(inode
, nd
);
801 err
= vfs_permission(nd
, MAY_EXEC
);
806 c
= *(const unsigned char *)name
;
808 hash
= init_name_hash();
811 hash
= partial_name_hash(c
, hash
);
812 c
= *(const unsigned char *)name
;
813 } while (c
&& (c
!= '/'));
814 this.len
= name
- (const char *) this.name
;
815 this.hash
= end_name_hash(hash
);
817 /* remove trailing slashes? */
820 while (*++name
== '/');
822 goto last_with_slashes
;
825 * "." and ".." are special - ".." especially so because it has
826 * to be able to know about the current root directory and
827 * parent relationships.
829 if (this.name
[0] == '.') switch (this.len
) {
833 if (this.name
[1] != '.')
836 inode
= nd
->dentry
->d_inode
;
842 * See if the low-level filesystem might want
843 * to use its own hash..
845 if (nd
->dentry
->d_op
&& nd
->dentry
->d_op
->d_hash
) {
846 err
= nd
->dentry
->d_op
->d_hash(nd
->dentry
, &this);
850 /* This does the actual lookups.. */
851 err
= do_lookup(nd
, &this, &next
);
856 inode
= next
.dentry
->d_inode
;
863 if (inode
->i_op
->follow_link
) {
864 err
= do_follow_link(&next
, nd
);
868 inode
= nd
->dentry
->d_inode
;
875 path_to_nameidata(&next
, nd
);
877 if (!inode
->i_op
->lookup
)
880 /* here ends the main loop */
883 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
885 nd
->flags
&= ~LOOKUP_CONTINUE
;
886 if (lookup_flags
& LOOKUP_PARENT
)
888 if (this.name
[0] == '.') switch (this.len
) {
892 if (this.name
[1] != '.')
895 inode
= nd
->dentry
->d_inode
;
900 if (nd
->dentry
->d_op
&& nd
->dentry
->d_op
->d_hash
) {
901 err
= nd
->dentry
->d_op
->d_hash(nd
->dentry
, &this);
905 err
= do_lookup(nd
, &this, &next
);
908 inode
= next
.dentry
->d_inode
;
909 if ((lookup_flags
& LOOKUP_FOLLOW
)
910 && inode
&& inode
->i_op
&& inode
->i_op
->follow_link
) {
911 err
= do_follow_link(&next
, nd
);
914 inode
= nd
->dentry
->d_inode
;
916 path_to_nameidata(&next
, nd
);
920 if (lookup_flags
& LOOKUP_DIRECTORY
) {
922 if (!inode
->i_op
|| !inode
->i_op
->lookup
)
928 nd
->last_type
= LAST_NORM
;
929 if (this.name
[0] != '.')
932 nd
->last_type
= LAST_DOT
;
933 else if (this.len
== 2 && this.name
[1] == '.')
934 nd
->last_type
= LAST_DOTDOT
;
939 * We bypassed the ordinary revalidation routines.
940 * We may need to check the cached dentry for staleness.
942 if (nd
->dentry
&& nd
->dentry
->d_sb
&&
943 (nd
->dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)) {
945 /* Note: we do not d_invalidate() */
946 if (!nd
->dentry
->d_op
->d_revalidate(nd
->dentry
, nd
))
952 dput_path(&next
, nd
);
961 * Wrapper to retry pathname resolution whenever the underlying
962 * file system returns an ESTALE.
964 * Retry the whole path once, forcing real lookup requests
965 * instead of relying on the dcache.
967 int fastcall
link_path_walk(const char *name
, struct nameidata
*nd
)
969 struct nameidata save
= *nd
;
972 /* make sure the stuff we saved doesn't go away */
976 result
= __link_path_walk(name
, nd
);
977 if (result
== -ESTALE
) {
981 nd
->flags
|= LOOKUP_REVAL
;
982 result
= __link_path_walk(name
, nd
);
991 int fastcall
path_walk(const char * name
, struct nameidata
*nd
)
993 current
->total_link_count
= 0;
994 return link_path_walk(name
, nd
);
998 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
999 * everything is done. Returns 0 and drops input nd, if lookup failed;
1001 static int __emul_lookup_dentry(const char *name
, struct nameidata
*nd
)
1003 if (path_walk(name
, nd
))
1004 return 0; /* something went wrong... */
1006 if (!nd
->dentry
->d_inode
|| S_ISDIR(nd
->dentry
->d_inode
->i_mode
)) {
1007 struct dentry
*old_dentry
= nd
->dentry
;
1008 struct vfsmount
*old_mnt
= nd
->mnt
;
1009 struct qstr last
= nd
->last
;
1010 int last_type
= nd
->last_type
;
1012 * NAME was not found in alternate root or it's a directory. Try to find
1013 * it in the normal root:
1015 nd
->last_type
= LAST_ROOT
;
1016 read_lock(¤t
->fs
->lock
);
1017 nd
->mnt
= mntget(current
->fs
->rootmnt
);
1018 nd
->dentry
= dget(current
->fs
->root
);
1019 read_unlock(¤t
->fs
->lock
);
1020 if (path_walk(name
, nd
) == 0) {
1021 if (nd
->dentry
->d_inode
) {
1028 nd
->dentry
= old_dentry
;
1031 nd
->last_type
= last_type
;
1036 void set_fs_altroot(void)
1038 char *emul
= __emul_prefix();
1039 struct nameidata nd
;
1040 struct vfsmount
*mnt
= NULL
, *oldmnt
;
1041 struct dentry
*dentry
= NULL
, *olddentry
;
1046 err
= path_lookup(emul
, LOOKUP_FOLLOW
|LOOKUP_DIRECTORY
|LOOKUP_NOALT
, &nd
);
1052 write_lock(¤t
->fs
->lock
);
1053 oldmnt
= current
->fs
->altrootmnt
;
1054 olddentry
= current
->fs
->altroot
;
1055 current
->fs
->altrootmnt
= mnt
;
1056 current
->fs
->altroot
= dentry
;
1057 write_unlock(¤t
->fs
->lock
);
1064 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1065 int fastcall
path_lookup(const char *name
, unsigned int flags
, struct nameidata
*nd
)
1069 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1073 read_lock(¤t
->fs
->lock
);
1075 if (current
->fs
->altroot
&& !(nd
->flags
& LOOKUP_NOALT
)) {
1076 nd
->mnt
= mntget(current
->fs
->altrootmnt
);
1077 nd
->dentry
= dget(current
->fs
->altroot
);
1078 read_unlock(¤t
->fs
->lock
);
1079 if (__emul_lookup_dentry(name
,nd
))
1080 goto out
; /* found in altroot */
1081 read_lock(¤t
->fs
->lock
);
1083 nd
->mnt
= mntget(current
->fs
->rootmnt
);
1084 nd
->dentry
= dget(current
->fs
->root
);
1086 nd
->mnt
= mntget(current
->fs
->pwdmnt
);
1087 nd
->dentry
= dget(current
->fs
->pwd
);
1089 read_unlock(¤t
->fs
->lock
);
1090 current
->total_link_count
= 0;
1091 retval
= link_path_walk(name
, nd
);
1093 if (unlikely(current
->audit_context
1094 && nd
&& nd
->dentry
&& nd
->dentry
->d_inode
))
1095 audit_inode(name
, nd
->dentry
->d_inode
, flags
);
1099 static int __path_lookup_intent_open(const char *name
, unsigned int lookup_flags
,
1100 struct nameidata
*nd
, int open_flags
, int create_mode
)
1102 struct file
*filp
= get_empty_filp();
1107 nd
->intent
.open
.file
= filp
;
1108 nd
->intent
.open
.flags
= open_flags
;
1109 nd
->intent
.open
.create_mode
= create_mode
;
1110 err
= path_lookup(name
, lookup_flags
|LOOKUP_OPEN
, nd
);
1111 if (IS_ERR(nd
->intent
.open
.file
)) {
1113 err
= PTR_ERR(nd
->intent
.open
.file
);
1116 } else if (err
!= 0)
1117 release_open_intent(nd
);
1122 * path_lookup_open - lookup a file path with open intent
1123 * @name: pointer to file name
1124 * @lookup_flags: lookup intent flags
1125 * @nd: pointer to nameidata
1126 * @open_flags: open intent flags
1128 int path_lookup_open(const char *name
, unsigned int lookup_flags
,
1129 struct nameidata
*nd
, int open_flags
)
1131 return __path_lookup_intent_open(name
, lookup_flags
, nd
,
1136 * path_lookup_create - lookup a file path with open + create intent
1137 * @name: pointer to file name
1138 * @lookup_flags: lookup intent flags
1139 * @nd: pointer to nameidata
1140 * @open_flags: open intent flags
1141 * @create_mode: create intent flags
1143 static int path_lookup_create(const char *name
, unsigned int lookup_flags
,
1144 struct nameidata
*nd
, int open_flags
,
1147 return __path_lookup_intent_open(name
, lookup_flags
|LOOKUP_CREATE
, nd
,
1148 open_flags
, create_mode
);
1151 int __user_path_lookup_open(const char __user
*name
, unsigned int lookup_flags
,
1152 struct nameidata
*nd
, int open_flags
)
1154 char *tmp
= getname(name
);
1155 int err
= PTR_ERR(tmp
);
1158 err
= __path_lookup_intent_open(tmp
, lookup_flags
, nd
, open_flags
, 0);
1165 * Restricted form of lookup. Doesn't follow links, single-component only,
1166 * needs parent already locked. Doesn't follow mounts.
1169 static struct dentry
* __lookup_hash(struct qstr
*name
, struct dentry
* base
, struct nameidata
*nd
)
1171 struct dentry
* dentry
;
1172 struct inode
*inode
;
1175 inode
= base
->d_inode
;
1176 err
= permission(inode
, MAY_EXEC
, nd
);
1177 dentry
= ERR_PTR(err
);
1182 * See if the low-level filesystem might want
1183 * to use its own hash..
1185 if (base
->d_op
&& base
->d_op
->d_hash
) {
1186 err
= base
->d_op
->d_hash(base
, name
);
1187 dentry
= ERR_PTR(err
);
1192 dentry
= cached_lookup(base
, name
, nd
);
1194 struct dentry
*new = d_alloc(base
, name
);
1195 dentry
= ERR_PTR(-ENOMEM
);
1198 dentry
= inode
->i_op
->lookup(inode
, new, nd
);
1208 struct dentry
* lookup_hash(struct nameidata
*nd
)
1210 return __lookup_hash(&nd
->last
, nd
->dentry
, nd
);
1214 struct dentry
* lookup_one_len(const char * name
, struct dentry
* base
, int len
)
1225 hash
= init_name_hash();
1227 c
= *(const unsigned char *)name
++;
1228 if (c
== '/' || c
== '\0')
1230 hash
= partial_name_hash(c
, hash
);
1232 this.hash
= end_name_hash(hash
);
1234 return __lookup_hash(&this, base
, NULL
);
1236 return ERR_PTR(-EACCES
);
1242 * is used by most simple commands to get the inode of a specified name.
1243 * Open, link etc use their own routines, but this is enough for things
1246 * namei exists in two versions: namei/lnamei. The only difference is
1247 * that namei follows links, while lnamei does not.
1250 int fastcall
__user_walk(const char __user
*name
, unsigned flags
, struct nameidata
*nd
)
1252 char *tmp
= getname(name
);
1253 int err
= PTR_ERR(tmp
);
1256 err
= path_lookup(tmp
, flags
, nd
);
1263 * It's inline, so penalty for filesystems that don't use sticky bit is
1266 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1268 if (!(dir
->i_mode
& S_ISVTX
))
1270 if (inode
->i_uid
== current
->fsuid
)
1272 if (dir
->i_uid
== current
->fsuid
)
1274 return !capable(CAP_FOWNER
);
1278 * Check whether we can remove a link victim from directory dir, check
1279 * whether the type of victim is right.
1280 * 1. We can't do it if dir is read-only (done in permission())
1281 * 2. We should have write and exec permissions on dir
1282 * 3. We can't remove anything from append-only dir
1283 * 4. We can't do anything with immutable dir (done in permission())
1284 * 5. If the sticky bit on dir is set we should either
1285 * a. be owner of dir, or
1286 * b. be owner of victim, or
1287 * c. have CAP_FOWNER capability
1288 * 6. If the victim is append-only or immutable we can't do antyhing with
1289 * links pointing to it.
1290 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1291 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1292 * 9. We can't remove a root or mountpoint.
1293 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1294 * nfs_async_unlink().
1296 static inline int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1300 if (!victim
->d_inode
)
1303 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1305 error
= permission(dir
,MAY_WRITE
| MAY_EXEC
, NULL
);
1310 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1311 IS_IMMUTABLE(victim
->d_inode
))
1314 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1316 if (IS_ROOT(victim
))
1318 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1320 if (IS_DEADDIR(dir
))
1322 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1327 /* Check whether we can create an object with dentry child in directory
1329 * 1. We can't do it if child already exists (open has special treatment for
1330 * this case, but since we are inlined it's OK)
1331 * 2. We can't do it if dir is read-only (done in permission())
1332 * 3. We should have write and exec permissions on dir
1333 * 4. We can't do it if dir is immutable (done in permission())
1335 static inline int may_create(struct inode
*dir
, struct dentry
*child
,
1336 struct nameidata
*nd
)
1340 if (IS_DEADDIR(dir
))
1342 return permission(dir
,MAY_WRITE
| MAY_EXEC
, nd
);
1346 * O_DIRECTORY translates into forcing a directory lookup.
1348 static inline int lookup_flags(unsigned int f
)
1350 unsigned long retval
= LOOKUP_FOLLOW
;
1353 retval
&= ~LOOKUP_FOLLOW
;
1355 if (f
& O_DIRECTORY
)
1356 retval
|= LOOKUP_DIRECTORY
;
1362 * p1 and p2 should be directories on the same fs.
1364 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1369 down(&p1
->d_inode
->i_sem
);
1373 down(&p1
->d_inode
->i_sb
->s_vfs_rename_sem
);
1375 for (p
= p1
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1376 if (p
->d_parent
== p2
) {
1377 down(&p2
->d_inode
->i_sem
);
1378 down(&p1
->d_inode
->i_sem
);
1383 for (p
= p2
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1384 if (p
->d_parent
== p1
) {
1385 down(&p1
->d_inode
->i_sem
);
1386 down(&p2
->d_inode
->i_sem
);
1391 down(&p1
->d_inode
->i_sem
);
1392 down(&p2
->d_inode
->i_sem
);
1396 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1398 up(&p1
->d_inode
->i_sem
);
1400 up(&p2
->d_inode
->i_sem
);
1401 up(&p1
->d_inode
->i_sb
->s_vfs_rename_sem
);
1405 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1406 struct nameidata
*nd
)
1408 int error
= may_create(dir
, dentry
, nd
);
1413 if (!dir
->i_op
|| !dir
->i_op
->create
)
1414 return -EACCES
; /* shouldn't it be ENOSYS? */
1417 error
= security_inode_create(dir
, dentry
, mode
);
1421 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1423 fsnotify_create(dir
, dentry
->d_name
.name
);
1427 int may_open(struct nameidata
*nd
, int acc_mode
, int flag
)
1429 struct dentry
*dentry
= nd
->dentry
;
1430 struct inode
*inode
= dentry
->d_inode
;
1436 if (S_ISLNK(inode
->i_mode
))
1439 if (S_ISDIR(inode
->i_mode
) && (flag
& FMODE_WRITE
))
1442 error
= vfs_permission(nd
, acc_mode
);
1447 * FIFO's, sockets and device files are special: they don't
1448 * actually live on the filesystem itself, and as such you
1449 * can write to them even if the filesystem is read-only.
1451 if (S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
1453 } else if (S_ISBLK(inode
->i_mode
) || S_ISCHR(inode
->i_mode
)) {
1454 if (nd
->mnt
->mnt_flags
& MNT_NODEV
)
1458 } else if (IS_RDONLY(inode
) && (flag
& FMODE_WRITE
))
1461 * An append-only file must be opened in append mode for writing.
1463 if (IS_APPEND(inode
)) {
1464 if ((flag
& FMODE_WRITE
) && !(flag
& O_APPEND
))
1470 /* O_NOATIME can only be set by the owner or superuser */
1471 if (flag
& O_NOATIME
)
1472 if (current
->fsuid
!= inode
->i_uid
&& !capable(CAP_FOWNER
))
1476 * Ensure there are no outstanding leases on the file.
1478 error
= break_lease(inode
, flag
);
1482 if (flag
& O_TRUNC
) {
1483 error
= get_write_access(inode
);
1488 * Refuse to truncate files with mandatory locks held on them.
1490 error
= locks_verify_locked(inode
);
1494 error
= do_truncate(dentry
, 0, ATTR_MTIME
|ATTR_CTIME
, NULL
);
1496 put_write_access(inode
);
1500 if (flag
& FMODE_WRITE
)
1509 * namei for open - this is in fact almost the whole open-routine.
1511 * Note that the low bits of "flag" aren't the same as in the open
1512 * system call - they are 00 - no permissions needed
1513 * 01 - read permission needed
1514 * 10 - write permission needed
1515 * 11 - read/write permissions needed
1516 * which is a lot more logical, and also allows the "no perm" needed
1517 * for symlinks (where the permissions are checked later).
1520 int open_namei(const char * pathname
, int flag
, int mode
, struct nameidata
*nd
)
1522 int acc_mode
, error
;
1527 acc_mode
= ACC_MODE(flag
);
1529 /* O_TRUNC implies we need access checks for write permissions */
1531 acc_mode
|= MAY_WRITE
;
1533 /* Allow the LSM permission hook to distinguish append
1534 access from general write access. */
1535 if (flag
& O_APPEND
)
1536 acc_mode
|= MAY_APPEND
;
1539 * The simplest case - just a plain lookup.
1541 if (!(flag
& O_CREAT
)) {
1542 error
= path_lookup_open(pathname
, lookup_flags(flag
), nd
, flag
);
1549 * Create - we need to know the parent.
1551 error
= path_lookup_create(pathname
, LOOKUP_PARENT
, nd
, flag
, mode
);
1556 * We have the parent and last component. First of all, check
1557 * that we are not asked to creat(2) an obvious directory - that
1561 if (nd
->last_type
!= LAST_NORM
|| nd
->last
.name
[nd
->last
.len
])
1565 nd
->flags
&= ~LOOKUP_PARENT
;
1566 down(&dir
->d_inode
->i_sem
);
1567 path
.dentry
= lookup_hash(nd
);
1571 error
= PTR_ERR(path
.dentry
);
1572 if (IS_ERR(path
.dentry
)) {
1573 up(&dir
->d_inode
->i_sem
);
1577 /* Negative dentry, just create the file */
1578 if (!path
.dentry
->d_inode
) {
1579 if (!IS_POSIXACL(dir
->d_inode
))
1580 mode
&= ~current
->fs
->umask
;
1581 error
= vfs_create(dir
->d_inode
, path
.dentry
, mode
, nd
);
1582 up(&dir
->d_inode
->i_sem
);
1584 nd
->dentry
= path
.dentry
;
1587 /* Don't check for write permission, don't truncate */
1594 * It already exists.
1596 up(&dir
->d_inode
->i_sem
);
1602 if (__follow_mount(&path
)) {
1604 if (flag
& O_NOFOLLOW
)
1608 if (!path
.dentry
->d_inode
)
1610 if (path
.dentry
->d_inode
->i_op
&& path
.dentry
->d_inode
->i_op
->follow_link
)
1613 path_to_nameidata(&path
, nd
);
1615 if (path
.dentry
->d_inode
&& S_ISDIR(path
.dentry
->d_inode
->i_mode
))
1618 error
= may_open(nd
, acc_mode
, flag
);
1624 dput_path(&path
, nd
);
1626 if (!IS_ERR(nd
->intent
.open
.file
))
1627 release_open_intent(nd
);
1633 if (flag
& O_NOFOLLOW
)
1636 * This is subtle. Instead of calling do_follow_link() we do the
1637 * thing by hands. The reason is that this way we have zero link_count
1638 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1639 * After that we have the parent and last component, i.e.
1640 * we are in the same situation as after the first path_walk().
1641 * Well, almost - if the last component is normal we get its copy
1642 * stored in nd->last.name and we will have to putname() it when we
1643 * are done. Procfs-like symlinks just set LAST_BIND.
1645 nd
->flags
|= LOOKUP_PARENT
;
1646 error
= security_inode_follow_link(path
.dentry
, nd
);
1649 error
= __do_follow_link(&path
, nd
);
1652 nd
->flags
&= ~LOOKUP_PARENT
;
1653 if (nd
->last_type
== LAST_BIND
)
1656 if (nd
->last_type
!= LAST_NORM
)
1658 if (nd
->last
.name
[nd
->last
.len
]) {
1659 __putname(nd
->last
.name
);
1664 __putname(nd
->last
.name
);
1668 down(&dir
->d_inode
->i_sem
);
1669 path
.dentry
= lookup_hash(nd
);
1671 __putname(nd
->last
.name
);
1676 * lookup_create - lookup a dentry, creating it if it doesn't exist
1677 * @nd: nameidata info
1678 * @is_dir: directory flag
1680 * Simple function to lookup and return a dentry and create it
1681 * if it doesn't exist. Is SMP-safe.
1683 * Returns with nd->dentry->d_inode->i_sem locked.
1685 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
1687 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
1689 down(&nd
->dentry
->d_inode
->i_sem
);
1691 * Yucky last component or no last component at all?
1692 * (foo/., foo/.., /////)
1694 if (nd
->last_type
!= LAST_NORM
)
1696 nd
->flags
&= ~LOOKUP_PARENT
;
1699 * Do the final lookup.
1701 dentry
= lookup_hash(nd
);
1706 * Special case - lookup gave negative, but... we had foo/bar/
1707 * From the vfs_mknod() POV we just have a negative dentry -
1708 * all is fine. Let's be bastards - you had / on the end, you've
1709 * been asking for (non-existent) directory. -ENOENT for you.
1711 if (!is_dir
&& nd
->last
.name
[nd
->last
.len
] && !dentry
->d_inode
)
1716 dentry
= ERR_PTR(-ENOENT
);
1720 EXPORT_SYMBOL_GPL(lookup_create
);
1722 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1724 int error
= may_create(dir
, dentry
, NULL
);
1729 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
1732 if (!dir
->i_op
|| !dir
->i_op
->mknod
)
1735 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
1740 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
1742 fsnotify_create(dir
, dentry
->d_name
.name
);
1746 asmlinkage
long sys_mknod(const char __user
* filename
, int mode
, unsigned dev
)
1750 struct dentry
* dentry
;
1751 struct nameidata nd
;
1755 tmp
= getname(filename
);
1757 return PTR_ERR(tmp
);
1759 error
= path_lookup(tmp
, LOOKUP_PARENT
, &nd
);
1762 dentry
= lookup_create(&nd
, 0);
1763 error
= PTR_ERR(dentry
);
1765 if (!IS_POSIXACL(nd
.dentry
->d_inode
))
1766 mode
&= ~current
->fs
->umask
;
1767 if (!IS_ERR(dentry
)) {
1768 switch (mode
& S_IFMT
) {
1769 case 0: case S_IFREG
:
1770 error
= vfs_create(nd
.dentry
->d_inode
,dentry
,mode
,&nd
);
1772 case S_IFCHR
: case S_IFBLK
:
1773 error
= vfs_mknod(nd
.dentry
->d_inode
,dentry
,mode
,
1774 new_decode_dev(dev
));
1776 case S_IFIFO
: case S_IFSOCK
:
1777 error
= vfs_mknod(nd
.dentry
->d_inode
,dentry
,mode
,0);
1787 up(&nd
.dentry
->d_inode
->i_sem
);
1795 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1797 int error
= may_create(dir
, dentry
, NULL
);
1802 if (!dir
->i_op
|| !dir
->i_op
->mkdir
)
1805 mode
&= (S_IRWXUGO
|S_ISVTX
);
1806 error
= security_inode_mkdir(dir
, dentry
, mode
);
1811 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
1813 fsnotify_mkdir(dir
, dentry
->d_name
.name
);
1817 asmlinkage
long sys_mkdir(const char __user
* pathname
, int mode
)
1822 tmp
= getname(pathname
);
1823 error
= PTR_ERR(tmp
);
1825 struct dentry
*dentry
;
1826 struct nameidata nd
;
1828 error
= path_lookup(tmp
, LOOKUP_PARENT
, &nd
);
1831 dentry
= lookup_create(&nd
, 1);
1832 error
= PTR_ERR(dentry
);
1833 if (!IS_ERR(dentry
)) {
1834 if (!IS_POSIXACL(nd
.dentry
->d_inode
))
1835 mode
&= ~current
->fs
->umask
;
1836 error
= vfs_mkdir(nd
.dentry
->d_inode
, dentry
, mode
);
1839 up(&nd
.dentry
->d_inode
->i_sem
);
1849 * We try to drop the dentry early: we should have
1850 * a usage count of 2 if we're the only user of this
1851 * dentry, and if that is true (possibly after pruning
1852 * the dcache), then we drop the dentry now.
1854 * A low-level filesystem can, if it choses, legally
1857 * if (!d_unhashed(dentry))
1860 * if it cannot handle the case of removing a directory
1861 * that is still in use by something else..
1863 void dentry_unhash(struct dentry
*dentry
)
1866 if (atomic_read(&dentry
->d_count
))
1867 shrink_dcache_parent(dentry
);
1868 spin_lock(&dcache_lock
);
1869 spin_lock(&dentry
->d_lock
);
1870 if (atomic_read(&dentry
->d_count
) == 2)
1872 spin_unlock(&dentry
->d_lock
);
1873 spin_unlock(&dcache_lock
);
1876 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1878 int error
= may_delete(dir
, dentry
, 1);
1883 if (!dir
->i_op
|| !dir
->i_op
->rmdir
)
1888 down(&dentry
->d_inode
->i_sem
);
1889 dentry_unhash(dentry
);
1890 if (d_mountpoint(dentry
))
1893 error
= security_inode_rmdir(dir
, dentry
);
1895 error
= dir
->i_op
->rmdir(dir
, dentry
);
1897 dentry
->d_inode
->i_flags
|= S_DEAD
;
1900 up(&dentry
->d_inode
->i_sem
);
1909 asmlinkage
long sys_rmdir(const char __user
* pathname
)
1913 struct dentry
*dentry
;
1914 struct nameidata nd
;
1916 name
= getname(pathname
);
1918 return PTR_ERR(name
);
1920 error
= path_lookup(name
, LOOKUP_PARENT
, &nd
);
1924 switch(nd
.last_type
) {
1935 down(&nd
.dentry
->d_inode
->i_sem
);
1936 dentry
= lookup_hash(&nd
);
1937 error
= PTR_ERR(dentry
);
1938 if (!IS_ERR(dentry
)) {
1939 error
= vfs_rmdir(nd
.dentry
->d_inode
, dentry
);
1942 up(&nd
.dentry
->d_inode
->i_sem
);
1950 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1952 int error
= may_delete(dir
, dentry
, 0);
1957 if (!dir
->i_op
|| !dir
->i_op
->unlink
)
1962 down(&dentry
->d_inode
->i_sem
);
1963 if (d_mountpoint(dentry
))
1966 error
= security_inode_unlink(dir
, dentry
);
1968 error
= dir
->i_op
->unlink(dir
, dentry
);
1970 up(&dentry
->d_inode
->i_sem
);
1972 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1973 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
1981 * Make sure that the actual truncation of the file will occur outside its
1982 * directory's i_sem. Truncate can take a long time if there is a lot of
1983 * writeout happening, and we don't want to prevent access to the directory
1984 * while waiting on the I/O.
1986 asmlinkage
long sys_unlink(const char __user
* pathname
)
1990 struct dentry
*dentry
;
1991 struct nameidata nd
;
1992 struct inode
*inode
= NULL
;
1994 name
= getname(pathname
);
1996 return PTR_ERR(name
);
1998 error
= path_lookup(name
, LOOKUP_PARENT
, &nd
);
2002 if (nd
.last_type
!= LAST_NORM
)
2004 down(&nd
.dentry
->d_inode
->i_sem
);
2005 dentry
= lookup_hash(&nd
);
2006 error
= PTR_ERR(dentry
);
2007 if (!IS_ERR(dentry
)) {
2008 /* Why not before? Because we want correct error value */
2009 if (nd
.last
.name
[nd
.last
.len
])
2011 inode
= dentry
->d_inode
;
2013 atomic_inc(&inode
->i_count
);
2014 error
= vfs_unlink(nd
.dentry
->d_inode
, dentry
);
2018 up(&nd
.dentry
->d_inode
->i_sem
);
2020 iput(inode
); /* truncate the inode here */
2028 error
= !dentry
->d_inode
? -ENOENT
:
2029 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2033 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
, int mode
)
2035 int error
= may_create(dir
, dentry
, NULL
);
2040 if (!dir
->i_op
|| !dir
->i_op
->symlink
)
2043 error
= security_inode_symlink(dir
, dentry
, oldname
);
2048 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2050 fsnotify_create(dir
, dentry
->d_name
.name
);
2054 asmlinkage
long sys_symlink(const char __user
* oldname
, const char __user
* newname
)
2060 from
= getname(oldname
);
2062 return PTR_ERR(from
);
2063 to
= getname(newname
);
2064 error
= PTR_ERR(to
);
2066 struct dentry
*dentry
;
2067 struct nameidata nd
;
2069 error
= path_lookup(to
, LOOKUP_PARENT
, &nd
);
2072 dentry
= lookup_create(&nd
, 0);
2073 error
= PTR_ERR(dentry
);
2074 if (!IS_ERR(dentry
)) {
2075 error
= vfs_symlink(nd
.dentry
->d_inode
, dentry
, from
, S_IALLUGO
);
2078 up(&nd
.dentry
->d_inode
->i_sem
);
2087 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2089 struct inode
*inode
= old_dentry
->d_inode
;
2095 error
= may_create(dir
, new_dentry
, NULL
);
2099 if (dir
->i_sb
!= inode
->i_sb
)
2103 * A link to an append-only or immutable file cannot be created.
2105 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2107 if (!dir
->i_op
|| !dir
->i_op
->link
)
2109 if (S_ISDIR(old_dentry
->d_inode
->i_mode
))
2112 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2116 down(&old_dentry
->d_inode
->i_sem
);
2118 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2119 up(&old_dentry
->d_inode
->i_sem
);
2121 fsnotify_create(dir
, new_dentry
->d_name
.name
);
2126 * Hardlinks are often used in delicate situations. We avoid
2127 * security-related surprises by not following symlinks on the
2130 * We don't follow them on the oldname either to be compatible
2131 * with linux 2.0, and to avoid hard-linking to directories
2132 * and other special files. --ADM
2134 asmlinkage
long sys_link(const char __user
* oldname
, const char __user
* newname
)
2136 struct dentry
*new_dentry
;
2137 struct nameidata nd
, old_nd
;
2141 to
= getname(newname
);
2145 error
= __user_walk(oldname
, 0, &old_nd
);
2148 error
= path_lookup(to
, LOOKUP_PARENT
, &nd
);
2152 if (old_nd
.mnt
!= nd
.mnt
)
2154 new_dentry
= lookup_create(&nd
, 0);
2155 error
= PTR_ERR(new_dentry
);
2156 if (!IS_ERR(new_dentry
)) {
2157 error
= vfs_link(old_nd
.dentry
, nd
.dentry
->d_inode
, new_dentry
);
2160 up(&nd
.dentry
->d_inode
->i_sem
);
2164 path_release(&old_nd
);
2172 * The worst of all namespace operations - renaming directory. "Perverted"
2173 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2175 * a) we can get into loop creation. Check is done in is_subdir().
2176 * b) race potential - two innocent renames can create a loop together.
2177 * That's where 4.4 screws up. Current fix: serialization on
2178 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2180 * c) we have to lock _three_ objects - parents and victim (if it exists).
2181 * And that - after we got ->i_sem on parents (until then we don't know
2182 * whether the target exists). Solution: try to be smart with locking
2183 * order for inodes. We rely on the fact that tree topology may change
2184 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2185 * move will be locked. Thus we can rank directories by the tree
2186 * (ancestors first) and rank all non-directories after them.
2187 * That works since everybody except rename does "lock parent, lookup,
2188 * lock child" and rename is under ->s_vfs_rename_sem.
2189 * HOWEVER, it relies on the assumption that any object with ->lookup()
2190 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2191 * we'd better make sure that there's no link(2) for them.
2192 * d) some filesystems don't support opened-but-unlinked directories,
2193 * either because of layout or because they are not ready to deal with
2194 * all cases correctly. The latter will be fixed (taking this sort of
2195 * stuff into VFS), but the former is not going away. Solution: the same
2196 * trick as in rmdir().
2197 * e) conversion from fhandle to dentry may come in the wrong moment - when
2198 * we are removing the target. Solution: we will have to grab ->i_sem
2199 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2200 * ->i_sem on parents, which works but leads to some truely excessive
2203 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2204 struct inode
*new_dir
, struct dentry
*new_dentry
)
2207 struct inode
*target
;
2210 * If we are going to change the parent - check write permissions,
2211 * we'll need to flip '..'.
2213 if (new_dir
!= old_dir
) {
2214 error
= permission(old_dentry
->d_inode
, MAY_WRITE
, NULL
);
2219 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2223 target
= new_dentry
->d_inode
;
2225 down(&target
->i_sem
);
2226 dentry_unhash(new_dentry
);
2228 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2231 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2234 target
->i_flags
|= S_DEAD
;
2236 if (d_unhashed(new_dentry
))
2237 d_rehash(new_dentry
);
2241 d_move(old_dentry
,new_dentry
);
2245 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
2246 struct inode
*new_dir
, struct dentry
*new_dentry
)
2248 struct inode
*target
;
2251 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2256 target
= new_dentry
->d_inode
;
2258 down(&target
->i_sem
);
2259 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2262 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2264 /* The following d_move() should become unconditional */
2265 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_ODD_RENAME
))
2266 d_move(old_dentry
, new_dentry
);
2274 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2275 struct inode
*new_dir
, struct dentry
*new_dentry
)
2278 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2279 const char *old_name
;
2281 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
2284 error
= may_delete(old_dir
, old_dentry
, is_dir
);
2288 if (!new_dentry
->d_inode
)
2289 error
= may_create(new_dir
, new_dentry
, NULL
);
2291 error
= may_delete(new_dir
, new_dentry
, is_dir
);
2295 if (!old_dir
->i_op
|| !old_dir
->i_op
->rename
)
2298 DQUOT_INIT(old_dir
);
2299 DQUOT_INIT(new_dir
);
2301 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
2304 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
2306 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
2308 const char *new_name
= old_dentry
->d_name
.name
;
2309 fsnotify_move(old_dir
, new_dir
, old_name
, new_name
, is_dir
,
2310 new_dentry
->d_inode
, old_dentry
->d_inode
);
2312 fsnotify_oldname_free(old_name
);
2317 static inline int do_rename(const char * oldname
, const char * newname
)
2320 struct dentry
* old_dir
, * new_dir
;
2321 struct dentry
* old_dentry
, *new_dentry
;
2322 struct dentry
* trap
;
2323 struct nameidata oldnd
, newnd
;
2325 error
= path_lookup(oldname
, LOOKUP_PARENT
, &oldnd
);
2329 error
= path_lookup(newname
, LOOKUP_PARENT
, &newnd
);
2334 if (oldnd
.mnt
!= newnd
.mnt
)
2337 old_dir
= oldnd
.dentry
;
2339 if (oldnd
.last_type
!= LAST_NORM
)
2342 new_dir
= newnd
.dentry
;
2343 if (newnd
.last_type
!= LAST_NORM
)
2346 trap
= lock_rename(new_dir
, old_dir
);
2348 old_dentry
= lookup_hash(&oldnd
);
2349 error
= PTR_ERR(old_dentry
);
2350 if (IS_ERR(old_dentry
))
2352 /* source must exist */
2354 if (!old_dentry
->d_inode
)
2356 /* unless the source is a directory trailing slashes give -ENOTDIR */
2357 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
2359 if (oldnd
.last
.name
[oldnd
.last
.len
])
2361 if (newnd
.last
.name
[newnd
.last
.len
])
2364 /* source should not be ancestor of target */
2366 if (old_dentry
== trap
)
2368 new_dentry
= lookup_hash(&newnd
);
2369 error
= PTR_ERR(new_dentry
);
2370 if (IS_ERR(new_dentry
))
2372 /* target should not be an ancestor of source */
2374 if (new_dentry
== trap
)
2377 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
2378 new_dir
->d_inode
, new_dentry
);
2384 unlock_rename(new_dir
, old_dir
);
2386 path_release(&newnd
);
2388 path_release(&oldnd
);
2393 asmlinkage
long sys_rename(const char __user
* oldname
, const char __user
* newname
)
2399 from
= getname(oldname
);
2401 return PTR_ERR(from
);
2402 to
= getname(newname
);
2403 error
= PTR_ERR(to
);
2405 error
= do_rename(from
,to
);
2412 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
2416 len
= PTR_ERR(link
);
2421 if (len
> (unsigned) buflen
)
2423 if (copy_to_user(buffer
, link
, len
))
2430 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2431 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2432 * using) it for any given inode is up to filesystem.
2434 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2436 struct nameidata nd
;
2440 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
2441 if (!IS_ERR(cookie
)) {
2442 int res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
2443 if (dentry
->d_inode
->i_op
->put_link
)
2444 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
2445 cookie
= ERR_PTR(res
);
2447 return PTR_ERR(cookie
);
2450 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
2452 return __vfs_follow_link(nd
, link
);
2455 /* get the link contents into pagecache */
2456 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
2459 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
2460 page
= read_cache_page(mapping
, 0, (filler_t
*)mapping
->a_ops
->readpage
,
2464 wait_on_page_locked(page
);
2465 if (!PageUptodate(page
))
2471 page_cache_release(page
);
2472 return ERR_PTR(-EIO
);
2478 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2480 struct page
*page
= NULL
;
2481 char *s
= page_getlink(dentry
, &page
);
2482 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
2485 page_cache_release(page
);
2490 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
2492 struct page
*page
= NULL
;
2493 nd_set_link(nd
, page_getlink(dentry
, &page
));
2497 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2499 struct page
*page
= cookie
;
2503 page_cache_release(page
);
2507 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
2509 struct address_space
*mapping
= inode
->i_mapping
;
2510 struct page
*page
= grab_cache_page(mapping
, 0);
2516 err
= mapping
->a_ops
->prepare_write(NULL
, page
, 0, len
-1);
2519 kaddr
= kmap_atomic(page
, KM_USER0
);
2520 memcpy(kaddr
, symname
, len
-1);
2521 kunmap_atomic(kaddr
, KM_USER0
);
2522 mapping
->a_ops
->commit_write(NULL
, page
, 0, len
-1);
2524 * Notice that we are _not_ going to block here - end of page is
2525 * unmapped, so this will only try to map the rest of page, see
2526 * that it is unmapped (typically even will not look into inode -
2527 * ->i_size will be enough for everything) and zero it out.
2528 * OTOH it's obviously correct and should make the page up-to-date.
2530 if (!PageUptodate(page
)) {
2531 err
= mapping
->a_ops
->readpage(NULL
, page
);
2532 wait_on_page_locked(page
);
2536 page_cache_release(page
);
2539 mark_inode_dirty(inode
);
2543 page_cache_release(page
);
2548 struct inode_operations page_symlink_inode_operations
= {
2549 .readlink
= generic_readlink
,
2550 .follow_link
= page_follow_link_light
,
2551 .put_link
= page_put_link
,
2554 EXPORT_SYMBOL(__user_walk
);
2555 EXPORT_SYMBOL(follow_down
);
2556 EXPORT_SYMBOL(follow_up
);
2557 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
2558 EXPORT_SYMBOL(getname
);
2559 EXPORT_SYMBOL(lock_rename
);
2560 EXPORT_SYMBOL(lookup_hash
);
2561 EXPORT_SYMBOL(lookup_one_len
);
2562 EXPORT_SYMBOL(page_follow_link_light
);
2563 EXPORT_SYMBOL(page_put_link
);
2564 EXPORT_SYMBOL(page_readlink
);
2565 EXPORT_SYMBOL(page_symlink
);
2566 EXPORT_SYMBOL(page_symlink_inode_operations
);
2567 EXPORT_SYMBOL(path_lookup
);
2568 EXPORT_SYMBOL(path_release
);
2569 EXPORT_SYMBOL(path_walk
);
2570 EXPORT_SYMBOL(permission
);
2571 EXPORT_SYMBOL(vfs_permission
);
2572 EXPORT_SYMBOL(file_permission
);
2573 EXPORT_SYMBOL(unlock_rename
);
2574 EXPORT_SYMBOL(vfs_create
);
2575 EXPORT_SYMBOL(vfs_follow_link
);
2576 EXPORT_SYMBOL(vfs_link
);
2577 EXPORT_SYMBOL(vfs_mkdir
);
2578 EXPORT_SYMBOL(vfs_mknod
);
2579 EXPORT_SYMBOL(generic_permission
);
2580 EXPORT_SYMBOL(vfs_readlink
);
2581 EXPORT_SYMBOL(vfs_rename
);
2582 EXPORT_SYMBOL(vfs_rmdir
);
2583 EXPORT_SYMBOL(vfs_symlink
);
2584 EXPORT_SYMBOL(vfs_unlink
);
2585 EXPORT_SYMBOL(dentry_unhash
);
2586 EXPORT_SYMBOL(generic_readlink
);