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 * get_write_access() gets write permission for a file.
276 * put_write_access() releases this write permission.
277 * This is used for regular files.
278 * We cannot support write (and maybe mmap read-write shared) accesses and
279 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
280 * can have the following values:
281 * 0: no writers, no VM_DENYWRITE mappings
282 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
283 * > 0: (i_writecount) users are writing to the file.
285 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
286 * except for the cases where we don't hold i_writecount yet. Then we need to
287 * use {get,deny}_write_access() - these functions check the sign and refuse
288 * to do the change if sign is wrong. Exclusion between them is provided by
289 * the inode->i_lock spinlock.
292 int get_write_access(struct inode
* inode
)
294 spin_lock(&inode
->i_lock
);
295 if (atomic_read(&inode
->i_writecount
) < 0) {
296 spin_unlock(&inode
->i_lock
);
299 atomic_inc(&inode
->i_writecount
);
300 spin_unlock(&inode
->i_lock
);
305 int deny_write_access(struct file
* file
)
307 struct inode
*inode
= file
->f_dentry
->d_inode
;
309 spin_lock(&inode
->i_lock
);
310 if (atomic_read(&inode
->i_writecount
) > 0) {
311 spin_unlock(&inode
->i_lock
);
314 atomic_dec(&inode
->i_writecount
);
315 spin_unlock(&inode
->i_lock
);
320 void path_release(struct nameidata
*nd
)
327 * umount() mustn't call path_release()/mntput() as that would clear
330 void path_release_on_umount(struct nameidata
*nd
)
333 mntput_no_expire(nd
->mnt
);
337 * release_open_intent - free up open intent resources
338 * @nd: pointer to nameidata
340 void release_open_intent(struct nameidata
*nd
)
342 if (nd
->intent
.open
.file
->f_dentry
== NULL
)
343 put_filp(nd
->intent
.open
.file
);
345 fput(nd
->intent
.open
.file
);
349 * Internal lookup() using the new generic dcache.
352 static struct dentry
* cached_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
354 struct dentry
* dentry
= __d_lookup(parent
, name
);
356 /* lockess __d_lookup may fail due to concurrent d_move()
357 * in some unrelated directory, so try with d_lookup
360 dentry
= d_lookup(parent
, name
);
362 if (dentry
&& dentry
->d_op
&& dentry
->d_op
->d_revalidate
) {
363 if (!dentry
->d_op
->d_revalidate(dentry
, nd
) && !d_invalidate(dentry
)) {
372 * Short-cut version of permission(), for calling by
373 * path_walk(), when dcache lock is held. Combines parts
374 * of permission() and generic_permission(), and tests ONLY for
375 * MAY_EXEC permission.
377 * If appropriate, check DAC only. If not appropriate, or
378 * short-cut DAC fails, then call permission() to do more
379 * complete permission check.
381 static inline int exec_permission_lite(struct inode
*inode
,
382 struct nameidata
*nd
)
384 umode_t mode
= inode
->i_mode
;
386 if (inode
->i_op
&& inode
->i_op
->permission
)
389 if (current
->fsuid
== inode
->i_uid
)
391 else if (in_group_p(inode
->i_gid
))
397 if ((inode
->i_mode
& S_IXUGO
) && capable(CAP_DAC_OVERRIDE
))
400 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_OVERRIDE
))
403 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_READ_SEARCH
))
408 return security_inode_permission(inode
, MAY_EXEC
, nd
);
412 * This is called when everything else fails, and we actually have
413 * to go to the low-level filesystem to find out what we should do..
415 * We get the directory semaphore, and after getting that we also
416 * make sure that nobody added the entry to the dcache in the meantime..
419 static struct dentry
* real_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
421 struct dentry
* result
;
422 struct inode
*dir
= parent
->d_inode
;
426 * First re-do the cached lookup just in case it was created
427 * while we waited for the directory semaphore..
429 * FIXME! This could use version numbering or similar to
430 * avoid unnecessary cache lookups.
432 * The "dcache_lock" is purely to protect the RCU list walker
433 * from concurrent renames at this point (we mustn't get false
434 * negatives from the RCU list walk here, unlike the optimistic
437 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
439 result
= d_lookup(parent
, name
);
441 struct dentry
* dentry
= d_alloc(parent
, name
);
442 result
= ERR_PTR(-ENOMEM
);
444 result
= dir
->i_op
->lookup(dir
, dentry
, nd
);
455 * Uhhuh! Nasty case: the cache was re-populated while
456 * we waited on the semaphore. Need to revalidate.
459 if (result
->d_op
&& result
->d_op
->d_revalidate
) {
460 if (!result
->d_op
->d_revalidate(result
, nd
) && !d_invalidate(result
)) {
462 result
= ERR_PTR(-ENOENT
);
468 static int __emul_lookup_dentry(const char *, struct nameidata
*);
472 walk_init_root(const char *name
, struct nameidata
*nd
)
474 read_lock(¤t
->fs
->lock
);
475 if (current
->fs
->altroot
&& !(nd
->flags
& LOOKUP_NOALT
)) {
476 nd
->mnt
= mntget(current
->fs
->altrootmnt
);
477 nd
->dentry
= dget(current
->fs
->altroot
);
478 read_unlock(¤t
->fs
->lock
);
479 if (__emul_lookup_dentry(name
,nd
))
481 read_lock(¤t
->fs
->lock
);
483 nd
->mnt
= mntget(current
->fs
->rootmnt
);
484 nd
->dentry
= dget(current
->fs
->root
);
485 read_unlock(¤t
->fs
->lock
);
489 static inline int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
498 if (!walk_init_root(link
, nd
))
499 /* weird __emul_prefix() stuff did it */
502 res
= link_path_walk(link
, nd
);
504 if (nd
->depth
|| res
|| nd
->last_type
!=LAST_NORM
)
507 * If it is an iterative symlinks resolution in open_namei() we
508 * have to copy the last component. And all that crap because of
509 * bloody create() on broken symlinks. Furrfu...
512 if (unlikely(!name
)) {
516 strcpy(name
, nd
->last
.name
);
517 nd
->last
.name
= name
;
521 return PTR_ERR(link
);
525 struct vfsmount
*mnt
;
526 struct dentry
*dentry
;
529 static inline int __do_follow_link(struct path
*path
, struct nameidata
*nd
)
533 struct dentry
*dentry
= path
->dentry
;
535 touch_atime(path
->mnt
, dentry
);
536 nd_set_link(nd
, NULL
);
538 if (path
->mnt
== nd
->mnt
)
540 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
541 error
= PTR_ERR(cookie
);
542 if (!IS_ERR(cookie
)) {
543 char *s
= nd_get_link(nd
);
546 error
= __vfs_follow_link(nd
, s
);
547 if (dentry
->d_inode
->i_op
->put_link
)
548 dentry
->d_inode
->i_op
->put_link(dentry
, nd
, cookie
);
556 static inline void dput_path(struct path
*path
, struct nameidata
*nd
)
559 if (path
->mnt
!= nd
->mnt
)
563 static inline void path_to_nameidata(struct path
*path
, struct nameidata
*nd
)
566 if (nd
->mnt
!= path
->mnt
)
569 nd
->dentry
= path
->dentry
;
573 * This limits recursive symlink follows to 8, while
574 * limiting consecutive symlinks to 40.
576 * Without that kind of total limit, nasty chains of consecutive
577 * symlinks can cause almost arbitrarily long lookups.
579 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
582 if (current
->link_count
>= MAX_NESTED_LINKS
)
584 if (current
->total_link_count
>= 40)
586 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
588 err
= security_inode_follow_link(path
->dentry
, nd
);
591 current
->link_count
++;
592 current
->total_link_count
++;
594 err
= __do_follow_link(path
, nd
);
595 current
->link_count
--;
604 int follow_up(struct vfsmount
**mnt
, struct dentry
**dentry
)
606 struct vfsmount
*parent
;
607 struct dentry
*mountpoint
;
608 spin_lock(&vfsmount_lock
);
609 parent
=(*mnt
)->mnt_parent
;
610 if (parent
== *mnt
) {
611 spin_unlock(&vfsmount_lock
);
615 mountpoint
=dget((*mnt
)->mnt_mountpoint
);
616 spin_unlock(&vfsmount_lock
);
618 *dentry
= mountpoint
;
624 /* no need for dcache_lock, as serialization is taken care in
627 static int __follow_mount(struct path
*path
)
630 while (d_mountpoint(path
->dentry
)) {
631 struct vfsmount
*mounted
= lookup_mnt(path
->mnt
, path
->dentry
);
638 path
->dentry
= dget(mounted
->mnt_root
);
644 static void follow_mount(struct vfsmount
**mnt
, struct dentry
**dentry
)
646 while (d_mountpoint(*dentry
)) {
647 struct vfsmount
*mounted
= lookup_mnt(*mnt
, *dentry
);
653 *dentry
= dget(mounted
->mnt_root
);
657 /* no need for dcache_lock, as serialization is taken care in
660 int follow_down(struct vfsmount
**mnt
, struct dentry
**dentry
)
662 struct vfsmount
*mounted
;
664 mounted
= lookup_mnt(*mnt
, *dentry
);
669 *dentry
= dget(mounted
->mnt_root
);
675 static inline void follow_dotdot(struct nameidata
*nd
)
678 struct vfsmount
*parent
;
679 struct dentry
*old
= nd
->dentry
;
681 read_lock(¤t
->fs
->lock
);
682 if (nd
->dentry
== current
->fs
->root
&&
683 nd
->mnt
== current
->fs
->rootmnt
) {
684 read_unlock(¤t
->fs
->lock
);
687 read_unlock(¤t
->fs
->lock
);
688 spin_lock(&dcache_lock
);
689 if (nd
->dentry
!= nd
->mnt
->mnt_root
) {
690 nd
->dentry
= dget(nd
->dentry
->d_parent
);
691 spin_unlock(&dcache_lock
);
695 spin_unlock(&dcache_lock
);
696 spin_lock(&vfsmount_lock
);
697 parent
= nd
->mnt
->mnt_parent
;
698 if (parent
== nd
->mnt
) {
699 spin_unlock(&vfsmount_lock
);
703 nd
->dentry
= dget(nd
->mnt
->mnt_mountpoint
);
704 spin_unlock(&vfsmount_lock
);
709 follow_mount(&nd
->mnt
, &nd
->dentry
);
713 * It's more convoluted than I'd like it to be, but... it's still fairly
714 * small and for now I'd prefer to have fast path as straight as possible.
715 * It _is_ time-critical.
717 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
720 struct vfsmount
*mnt
= nd
->mnt
;
721 struct dentry
*dentry
= __d_lookup(nd
->dentry
, name
);
725 if (dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
726 goto need_revalidate
;
729 path
->dentry
= dentry
;
730 __follow_mount(path
);
734 dentry
= real_lookup(nd
->dentry
, name
, nd
);
740 if (dentry
->d_op
->d_revalidate(dentry
, nd
))
742 if (d_invalidate(dentry
))
748 return PTR_ERR(dentry
);
753 * This is the basic name resolution function, turning a pathname into
754 * the final dentry. We expect 'base' to be positive and a directory.
756 * Returns 0 and nd will have valid dentry and mnt on success.
757 * Returns error and drops reference to input namei data on failure.
759 static fastcall
int __link_path_walk(const char * name
, struct nameidata
*nd
)
764 unsigned int lookup_flags
= nd
->flags
;
771 inode
= nd
->dentry
->d_inode
;
773 lookup_flags
= LOOKUP_FOLLOW
;
775 /* At this point we know we have a real path component. */
781 nd
->flags
|= LOOKUP_CONTINUE
;
782 err
= exec_permission_lite(inode
, nd
);
784 err
= vfs_permission(nd
, MAY_EXEC
);
789 c
= *(const unsigned char *)name
;
791 hash
= init_name_hash();
794 hash
= partial_name_hash(c
, hash
);
795 c
= *(const unsigned char *)name
;
796 } while (c
&& (c
!= '/'));
797 this.len
= name
- (const char *) this.name
;
798 this.hash
= end_name_hash(hash
);
800 /* remove trailing slashes? */
803 while (*++name
== '/');
805 goto last_with_slashes
;
808 * "." and ".." are special - ".." especially so because it has
809 * to be able to know about the current root directory and
810 * parent relationships.
812 if (this.name
[0] == '.') switch (this.len
) {
816 if (this.name
[1] != '.')
819 inode
= nd
->dentry
->d_inode
;
825 * See if the low-level filesystem might want
826 * to use its own hash..
828 if (nd
->dentry
->d_op
&& nd
->dentry
->d_op
->d_hash
) {
829 err
= nd
->dentry
->d_op
->d_hash(nd
->dentry
, &this);
833 /* This does the actual lookups.. */
834 err
= do_lookup(nd
, &this, &next
);
839 inode
= next
.dentry
->d_inode
;
846 if (inode
->i_op
->follow_link
) {
847 err
= do_follow_link(&next
, nd
);
851 inode
= nd
->dentry
->d_inode
;
858 path_to_nameidata(&next
, nd
);
860 if (!inode
->i_op
->lookup
)
863 /* here ends the main loop */
866 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
868 nd
->flags
&= ~LOOKUP_CONTINUE
;
869 if (lookup_flags
& LOOKUP_PARENT
)
871 if (this.name
[0] == '.') switch (this.len
) {
875 if (this.name
[1] != '.')
878 inode
= nd
->dentry
->d_inode
;
883 if (nd
->dentry
->d_op
&& nd
->dentry
->d_op
->d_hash
) {
884 err
= nd
->dentry
->d_op
->d_hash(nd
->dentry
, &this);
888 err
= do_lookup(nd
, &this, &next
);
891 inode
= next
.dentry
->d_inode
;
892 if ((lookup_flags
& LOOKUP_FOLLOW
)
893 && inode
&& inode
->i_op
&& inode
->i_op
->follow_link
) {
894 err
= do_follow_link(&next
, nd
);
897 inode
= nd
->dentry
->d_inode
;
899 path_to_nameidata(&next
, nd
);
903 if (lookup_flags
& LOOKUP_DIRECTORY
) {
905 if (!inode
->i_op
|| !inode
->i_op
->lookup
)
911 nd
->last_type
= LAST_NORM
;
912 if (this.name
[0] != '.')
915 nd
->last_type
= LAST_DOT
;
916 else if (this.len
== 2 && this.name
[1] == '.')
917 nd
->last_type
= LAST_DOTDOT
;
922 * We bypassed the ordinary revalidation routines.
923 * We may need to check the cached dentry for staleness.
925 if (nd
->dentry
&& nd
->dentry
->d_sb
&&
926 (nd
->dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)) {
928 /* Note: we do not d_invalidate() */
929 if (!nd
->dentry
->d_op
->d_revalidate(nd
->dentry
, nd
))
935 dput_path(&next
, nd
);
944 * Wrapper to retry pathname resolution whenever the underlying
945 * file system returns an ESTALE.
947 * Retry the whole path once, forcing real lookup requests
948 * instead of relying on the dcache.
950 int fastcall
link_path_walk(const char *name
, struct nameidata
*nd
)
952 struct nameidata save
= *nd
;
955 /* make sure the stuff we saved doesn't go away */
959 result
= __link_path_walk(name
, nd
);
960 if (result
== -ESTALE
) {
964 nd
->flags
|= LOOKUP_REVAL
;
965 result
= __link_path_walk(name
, nd
);
974 int fastcall
path_walk(const char * name
, struct nameidata
*nd
)
976 current
->total_link_count
= 0;
977 return link_path_walk(name
, nd
);
981 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
982 * everything is done. Returns 0 and drops input nd, if lookup failed;
984 static int __emul_lookup_dentry(const char *name
, struct nameidata
*nd
)
986 if (path_walk(name
, nd
))
987 return 0; /* something went wrong... */
989 if (!nd
->dentry
->d_inode
|| S_ISDIR(nd
->dentry
->d_inode
->i_mode
)) {
990 struct dentry
*old_dentry
= nd
->dentry
;
991 struct vfsmount
*old_mnt
= nd
->mnt
;
992 struct qstr last
= nd
->last
;
993 int last_type
= nd
->last_type
;
995 * NAME was not found in alternate root or it's a directory. Try to find
996 * it in the normal root:
998 nd
->last_type
= LAST_ROOT
;
999 read_lock(¤t
->fs
->lock
);
1000 nd
->mnt
= mntget(current
->fs
->rootmnt
);
1001 nd
->dentry
= dget(current
->fs
->root
);
1002 read_unlock(¤t
->fs
->lock
);
1003 if (path_walk(name
, nd
) == 0) {
1004 if (nd
->dentry
->d_inode
) {
1011 nd
->dentry
= old_dentry
;
1014 nd
->last_type
= last_type
;
1019 void set_fs_altroot(void)
1021 char *emul
= __emul_prefix();
1022 struct nameidata nd
;
1023 struct vfsmount
*mnt
= NULL
, *oldmnt
;
1024 struct dentry
*dentry
= NULL
, *olddentry
;
1029 err
= path_lookup(emul
, LOOKUP_FOLLOW
|LOOKUP_DIRECTORY
|LOOKUP_NOALT
, &nd
);
1035 write_lock(¤t
->fs
->lock
);
1036 oldmnt
= current
->fs
->altrootmnt
;
1037 olddentry
= current
->fs
->altroot
;
1038 current
->fs
->altrootmnt
= mnt
;
1039 current
->fs
->altroot
= dentry
;
1040 write_unlock(¤t
->fs
->lock
);
1047 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1048 int fastcall
path_lookup(const char *name
, unsigned int flags
, struct nameidata
*nd
)
1052 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1056 read_lock(¤t
->fs
->lock
);
1058 if (current
->fs
->altroot
&& !(nd
->flags
& LOOKUP_NOALT
)) {
1059 nd
->mnt
= mntget(current
->fs
->altrootmnt
);
1060 nd
->dentry
= dget(current
->fs
->altroot
);
1061 read_unlock(¤t
->fs
->lock
);
1062 if (__emul_lookup_dentry(name
,nd
))
1063 goto out
; /* found in altroot */
1064 read_lock(¤t
->fs
->lock
);
1066 nd
->mnt
= mntget(current
->fs
->rootmnt
);
1067 nd
->dentry
= dget(current
->fs
->root
);
1069 nd
->mnt
= mntget(current
->fs
->pwdmnt
);
1070 nd
->dentry
= dget(current
->fs
->pwd
);
1072 read_unlock(¤t
->fs
->lock
);
1073 current
->total_link_count
= 0;
1074 retval
= link_path_walk(name
, nd
);
1076 if (unlikely(current
->audit_context
1077 && nd
&& nd
->dentry
&& nd
->dentry
->d_inode
))
1078 audit_inode(name
, nd
->dentry
->d_inode
, flags
);
1082 static int __path_lookup_intent_open(const char *name
, unsigned int lookup_flags
,
1083 struct nameidata
*nd
, int open_flags
, int create_mode
)
1085 struct file
*filp
= get_empty_filp();
1090 nd
->intent
.open
.file
= filp
;
1091 nd
->intent
.open
.flags
= open_flags
;
1092 nd
->intent
.open
.create_mode
= create_mode
;
1093 err
= path_lookup(name
, lookup_flags
|LOOKUP_OPEN
, nd
);
1094 if (IS_ERR(nd
->intent
.open
.file
)) {
1096 err
= PTR_ERR(nd
->intent
.open
.file
);
1099 } else if (err
!= 0)
1100 release_open_intent(nd
);
1105 * path_lookup_open - lookup a file path with open intent
1106 * @name: pointer to file name
1107 * @lookup_flags: lookup intent flags
1108 * @nd: pointer to nameidata
1109 * @open_flags: open intent flags
1111 int path_lookup_open(const char *name
, unsigned int lookup_flags
,
1112 struct nameidata
*nd
, int open_flags
)
1114 return __path_lookup_intent_open(name
, lookup_flags
, nd
,
1119 * path_lookup_create - lookup a file path with open + create intent
1120 * @name: pointer to file name
1121 * @lookup_flags: lookup intent flags
1122 * @nd: pointer to nameidata
1123 * @open_flags: open intent flags
1124 * @create_mode: create intent flags
1126 int path_lookup_create(const char *name
, unsigned int lookup_flags
,
1127 struct nameidata
*nd
, int open_flags
, int create_mode
)
1129 return __path_lookup_intent_open(name
, lookup_flags
|LOOKUP_CREATE
, nd
,
1130 open_flags
, create_mode
);
1133 int __user_path_lookup_open(const char __user
*name
, unsigned int lookup_flags
,
1134 struct nameidata
*nd
, int open_flags
)
1136 char *tmp
= getname(name
);
1137 int err
= PTR_ERR(tmp
);
1140 err
= __path_lookup_intent_open(tmp
, lookup_flags
, nd
, open_flags
, 0);
1147 * Restricted form of lookup. Doesn't follow links, single-component only,
1148 * needs parent already locked. Doesn't follow mounts.
1151 static struct dentry
* __lookup_hash(struct qstr
*name
, struct dentry
* base
, struct nameidata
*nd
)
1153 struct dentry
* dentry
;
1154 struct inode
*inode
;
1157 inode
= base
->d_inode
;
1158 err
= permission(inode
, MAY_EXEC
, nd
);
1159 dentry
= ERR_PTR(err
);
1164 * See if the low-level filesystem might want
1165 * to use its own hash..
1167 if (base
->d_op
&& base
->d_op
->d_hash
) {
1168 err
= base
->d_op
->d_hash(base
, name
);
1169 dentry
= ERR_PTR(err
);
1174 dentry
= cached_lookup(base
, name
, nd
);
1176 struct dentry
*new = d_alloc(base
, name
);
1177 dentry
= ERR_PTR(-ENOMEM
);
1180 dentry
= inode
->i_op
->lookup(inode
, new, nd
);
1190 struct dentry
* lookup_hash(struct qstr
*name
, struct dentry
* base
)
1192 return __lookup_hash(name
, base
, NULL
);
1196 struct dentry
* lookup_one_len(const char * name
, struct dentry
* base
, int len
)
1207 hash
= init_name_hash();
1209 c
= *(const unsigned char *)name
++;
1210 if (c
== '/' || c
== '\0')
1212 hash
= partial_name_hash(c
, hash
);
1214 this.hash
= end_name_hash(hash
);
1216 return lookup_hash(&this, base
);
1218 return ERR_PTR(-EACCES
);
1224 * is used by most simple commands to get the inode of a specified name.
1225 * Open, link etc use their own routines, but this is enough for things
1228 * namei exists in two versions: namei/lnamei. The only difference is
1229 * that namei follows links, while lnamei does not.
1232 int fastcall
__user_walk(const char __user
*name
, unsigned flags
, struct nameidata
*nd
)
1234 char *tmp
= getname(name
);
1235 int err
= PTR_ERR(tmp
);
1238 err
= path_lookup(tmp
, flags
, nd
);
1245 * It's inline, so penalty for filesystems that don't use sticky bit is
1248 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1250 if (!(dir
->i_mode
& S_ISVTX
))
1252 if (inode
->i_uid
== current
->fsuid
)
1254 if (dir
->i_uid
== current
->fsuid
)
1256 return !capable(CAP_FOWNER
);
1260 * Check whether we can remove a link victim from directory dir, check
1261 * whether the type of victim is right.
1262 * 1. We can't do it if dir is read-only (done in permission())
1263 * 2. We should have write and exec permissions on dir
1264 * 3. We can't remove anything from append-only dir
1265 * 4. We can't do anything with immutable dir (done in permission())
1266 * 5. If the sticky bit on dir is set we should either
1267 * a. be owner of dir, or
1268 * b. be owner of victim, or
1269 * c. have CAP_FOWNER capability
1270 * 6. If the victim is append-only or immutable we can't do antyhing with
1271 * links pointing to it.
1272 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1273 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1274 * 9. We can't remove a root or mountpoint.
1275 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1276 * nfs_async_unlink().
1278 static inline int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1282 if (!victim
->d_inode
)
1285 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1287 error
= permission(dir
,MAY_WRITE
| MAY_EXEC
, NULL
);
1292 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1293 IS_IMMUTABLE(victim
->d_inode
))
1296 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1298 if (IS_ROOT(victim
))
1300 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1302 if (IS_DEADDIR(dir
))
1304 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1309 /* Check whether we can create an object with dentry child in directory
1311 * 1. We can't do it if child already exists (open has special treatment for
1312 * this case, but since we are inlined it's OK)
1313 * 2. We can't do it if dir is read-only (done in permission())
1314 * 3. We should have write and exec permissions on dir
1315 * 4. We can't do it if dir is immutable (done in permission())
1317 static inline int may_create(struct inode
*dir
, struct dentry
*child
,
1318 struct nameidata
*nd
)
1322 if (IS_DEADDIR(dir
))
1324 return permission(dir
,MAY_WRITE
| MAY_EXEC
, nd
);
1328 * O_DIRECTORY translates into forcing a directory lookup.
1330 static inline int lookup_flags(unsigned int f
)
1332 unsigned long retval
= LOOKUP_FOLLOW
;
1335 retval
&= ~LOOKUP_FOLLOW
;
1337 if (f
& O_DIRECTORY
)
1338 retval
|= LOOKUP_DIRECTORY
;
1344 * p1 and p2 should be directories on the same fs.
1346 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1351 down(&p1
->d_inode
->i_sem
);
1355 down(&p1
->d_inode
->i_sb
->s_vfs_rename_sem
);
1357 for (p
= p1
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1358 if (p
->d_parent
== p2
) {
1359 down(&p2
->d_inode
->i_sem
);
1360 down(&p1
->d_inode
->i_sem
);
1365 for (p
= p2
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1366 if (p
->d_parent
== p1
) {
1367 down(&p1
->d_inode
->i_sem
);
1368 down(&p2
->d_inode
->i_sem
);
1373 down(&p1
->d_inode
->i_sem
);
1374 down(&p2
->d_inode
->i_sem
);
1378 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1380 up(&p1
->d_inode
->i_sem
);
1382 up(&p2
->d_inode
->i_sem
);
1383 up(&p1
->d_inode
->i_sb
->s_vfs_rename_sem
);
1387 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1388 struct nameidata
*nd
)
1390 int error
= may_create(dir
, dentry
, nd
);
1395 if (!dir
->i_op
|| !dir
->i_op
->create
)
1396 return -EACCES
; /* shouldn't it be ENOSYS? */
1399 error
= security_inode_create(dir
, dentry
, mode
);
1403 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1405 fsnotify_create(dir
, dentry
->d_name
.name
);
1409 int may_open(struct nameidata
*nd
, int acc_mode
, int flag
)
1411 struct dentry
*dentry
= nd
->dentry
;
1412 struct inode
*inode
= dentry
->d_inode
;
1418 if (S_ISLNK(inode
->i_mode
))
1421 if (S_ISDIR(inode
->i_mode
) && (flag
& FMODE_WRITE
))
1424 error
= vfs_permission(nd
, acc_mode
);
1429 * FIFO's, sockets and device files are special: they don't
1430 * actually live on the filesystem itself, and as such you
1431 * can write to them even if the filesystem is read-only.
1433 if (S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
1435 } else if (S_ISBLK(inode
->i_mode
) || S_ISCHR(inode
->i_mode
)) {
1436 if (nd
->mnt
->mnt_flags
& MNT_NODEV
)
1440 } else if (IS_RDONLY(inode
) && (flag
& FMODE_WRITE
))
1443 * An append-only file must be opened in append mode for writing.
1445 if (IS_APPEND(inode
)) {
1446 if ((flag
& FMODE_WRITE
) && !(flag
& O_APPEND
))
1452 /* O_NOATIME can only be set by the owner or superuser */
1453 if (flag
& O_NOATIME
)
1454 if (current
->fsuid
!= inode
->i_uid
&& !capable(CAP_FOWNER
))
1458 * Ensure there are no outstanding leases on the file.
1460 error
= break_lease(inode
, flag
);
1464 if (flag
& O_TRUNC
) {
1465 error
= get_write_access(inode
);
1470 * Refuse to truncate files with mandatory locks held on them.
1472 error
= locks_verify_locked(inode
);
1476 error
= do_truncate(dentry
, 0, NULL
);
1478 put_write_access(inode
);
1482 if (flag
& FMODE_WRITE
)
1491 * namei for open - this is in fact almost the whole open-routine.
1493 * Note that the low bits of "flag" aren't the same as in the open
1494 * system call - they are 00 - no permissions needed
1495 * 01 - read permission needed
1496 * 10 - write permission needed
1497 * 11 - read/write permissions needed
1498 * which is a lot more logical, and also allows the "no perm" needed
1499 * for symlinks (where the permissions are checked later).
1502 int open_namei(const char * pathname
, int flag
, int mode
, struct nameidata
*nd
)
1504 int acc_mode
, error
;
1509 acc_mode
= ACC_MODE(flag
);
1511 /* O_TRUNC implies we need access checks for write permissions */
1513 acc_mode
|= MAY_WRITE
;
1515 /* Allow the LSM permission hook to distinguish append
1516 access from general write access. */
1517 if (flag
& O_APPEND
)
1518 acc_mode
|= MAY_APPEND
;
1521 * The simplest case - just a plain lookup.
1523 if (!(flag
& O_CREAT
)) {
1524 error
= path_lookup_open(pathname
, lookup_flags(flag
), nd
, flag
);
1531 * Create - we need to know the parent.
1533 error
= path_lookup_create(pathname
, LOOKUP_PARENT
, nd
, flag
, mode
);
1538 * We have the parent and last component. First of all, check
1539 * that we are not asked to creat(2) an obvious directory - that
1543 if (nd
->last_type
!= LAST_NORM
|| nd
->last
.name
[nd
->last
.len
])
1547 nd
->flags
&= ~LOOKUP_PARENT
;
1548 down(&dir
->d_inode
->i_sem
);
1549 path
.dentry
= __lookup_hash(&nd
->last
, nd
->dentry
, nd
);
1553 error
= PTR_ERR(path
.dentry
);
1554 if (IS_ERR(path
.dentry
)) {
1555 up(&dir
->d_inode
->i_sem
);
1559 /* Negative dentry, just create the file */
1560 if (!path
.dentry
->d_inode
) {
1561 if (!IS_POSIXACL(dir
->d_inode
))
1562 mode
&= ~current
->fs
->umask
;
1563 error
= vfs_create(dir
->d_inode
, path
.dentry
, mode
, nd
);
1564 up(&dir
->d_inode
->i_sem
);
1566 nd
->dentry
= path
.dentry
;
1569 /* Don't check for write permission, don't truncate */
1576 * It already exists.
1578 up(&dir
->d_inode
->i_sem
);
1584 if (__follow_mount(&path
)) {
1586 if (flag
& O_NOFOLLOW
)
1590 if (!path
.dentry
->d_inode
)
1592 if (path
.dentry
->d_inode
->i_op
&& path
.dentry
->d_inode
->i_op
->follow_link
)
1595 path_to_nameidata(&path
, nd
);
1597 if (path
.dentry
->d_inode
&& S_ISDIR(path
.dentry
->d_inode
->i_mode
))
1600 error
= may_open(nd
, acc_mode
, flag
);
1606 dput_path(&path
, nd
);
1608 if (!IS_ERR(nd
->intent
.open
.file
))
1609 release_open_intent(nd
);
1615 if (flag
& O_NOFOLLOW
)
1618 * This is subtle. Instead of calling do_follow_link() we do the
1619 * thing by hands. The reason is that this way we have zero link_count
1620 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1621 * After that we have the parent and last component, i.e.
1622 * we are in the same situation as after the first path_walk().
1623 * Well, almost - if the last component is normal we get its copy
1624 * stored in nd->last.name and we will have to putname() it when we
1625 * are done. Procfs-like symlinks just set LAST_BIND.
1627 nd
->flags
|= LOOKUP_PARENT
;
1628 error
= security_inode_follow_link(path
.dentry
, nd
);
1631 error
= __do_follow_link(&path
, nd
);
1634 nd
->flags
&= ~LOOKUP_PARENT
;
1635 if (nd
->last_type
== LAST_BIND
)
1638 if (nd
->last_type
!= LAST_NORM
)
1640 if (nd
->last
.name
[nd
->last
.len
]) {
1641 __putname(nd
->last
.name
);
1646 __putname(nd
->last
.name
);
1650 down(&dir
->d_inode
->i_sem
);
1651 path
.dentry
= __lookup_hash(&nd
->last
, nd
->dentry
, nd
);
1653 __putname(nd
->last
.name
);
1658 * lookup_create - lookup a dentry, creating it if it doesn't exist
1659 * @nd: nameidata info
1660 * @is_dir: directory flag
1662 * Simple function to lookup and return a dentry and create it
1663 * if it doesn't exist. Is SMP-safe.
1665 * Returns with nd->dentry->d_inode->i_sem locked.
1667 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
1669 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
1671 down(&nd
->dentry
->d_inode
->i_sem
);
1673 * Yucky last component or no last component at all?
1674 * (foo/., foo/.., /////)
1676 if (nd
->last_type
!= LAST_NORM
)
1678 nd
->flags
&= ~LOOKUP_PARENT
;
1681 * Do the final lookup.
1683 dentry
= lookup_hash(&nd
->last
, nd
->dentry
);
1688 * Special case - lookup gave negative, but... we had foo/bar/
1689 * From the vfs_mknod() POV we just have a negative dentry -
1690 * all is fine. Let's be bastards - you had / on the end, you've
1691 * been asking for (non-existent) directory. -ENOENT for you.
1693 if (!is_dir
&& nd
->last
.name
[nd
->last
.len
] && !dentry
->d_inode
)
1698 dentry
= ERR_PTR(-ENOENT
);
1702 EXPORT_SYMBOL_GPL(lookup_create
);
1704 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1706 int error
= may_create(dir
, dentry
, NULL
);
1711 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
1714 if (!dir
->i_op
|| !dir
->i_op
->mknod
)
1717 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
1722 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
1724 fsnotify_create(dir
, dentry
->d_name
.name
);
1728 asmlinkage
long sys_mknod(const char __user
* filename
, int mode
, unsigned dev
)
1732 struct dentry
* dentry
;
1733 struct nameidata nd
;
1737 tmp
= getname(filename
);
1739 return PTR_ERR(tmp
);
1741 error
= path_lookup(tmp
, LOOKUP_PARENT
, &nd
);
1744 dentry
= lookup_create(&nd
, 0);
1745 error
= PTR_ERR(dentry
);
1747 if (!IS_POSIXACL(nd
.dentry
->d_inode
))
1748 mode
&= ~current
->fs
->umask
;
1749 if (!IS_ERR(dentry
)) {
1750 switch (mode
& S_IFMT
) {
1751 case 0: case S_IFREG
:
1752 error
= vfs_create(nd
.dentry
->d_inode
,dentry
,mode
,&nd
);
1754 case S_IFCHR
: case S_IFBLK
:
1755 error
= vfs_mknod(nd
.dentry
->d_inode
,dentry
,mode
,
1756 new_decode_dev(dev
));
1758 case S_IFIFO
: case S_IFSOCK
:
1759 error
= vfs_mknod(nd
.dentry
->d_inode
,dentry
,mode
,0);
1769 up(&nd
.dentry
->d_inode
->i_sem
);
1777 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1779 int error
= may_create(dir
, dentry
, NULL
);
1784 if (!dir
->i_op
|| !dir
->i_op
->mkdir
)
1787 mode
&= (S_IRWXUGO
|S_ISVTX
);
1788 error
= security_inode_mkdir(dir
, dentry
, mode
);
1793 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
1795 fsnotify_mkdir(dir
, dentry
->d_name
.name
);
1799 asmlinkage
long sys_mkdir(const char __user
* pathname
, int mode
)
1804 tmp
= getname(pathname
);
1805 error
= PTR_ERR(tmp
);
1807 struct dentry
*dentry
;
1808 struct nameidata nd
;
1810 error
= path_lookup(tmp
, LOOKUP_PARENT
, &nd
);
1813 dentry
= lookup_create(&nd
, 1);
1814 error
= PTR_ERR(dentry
);
1815 if (!IS_ERR(dentry
)) {
1816 if (!IS_POSIXACL(nd
.dentry
->d_inode
))
1817 mode
&= ~current
->fs
->umask
;
1818 error
= vfs_mkdir(nd
.dentry
->d_inode
, dentry
, mode
);
1821 up(&nd
.dentry
->d_inode
->i_sem
);
1831 * We try to drop the dentry early: we should have
1832 * a usage count of 2 if we're the only user of this
1833 * dentry, and if that is true (possibly after pruning
1834 * the dcache), then we drop the dentry now.
1836 * A low-level filesystem can, if it choses, legally
1839 * if (!d_unhashed(dentry))
1842 * if it cannot handle the case of removing a directory
1843 * that is still in use by something else..
1845 void dentry_unhash(struct dentry
*dentry
)
1848 if (atomic_read(&dentry
->d_count
))
1849 shrink_dcache_parent(dentry
);
1850 spin_lock(&dcache_lock
);
1851 spin_lock(&dentry
->d_lock
);
1852 if (atomic_read(&dentry
->d_count
) == 2)
1854 spin_unlock(&dentry
->d_lock
);
1855 spin_unlock(&dcache_lock
);
1858 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1860 int error
= may_delete(dir
, dentry
, 1);
1865 if (!dir
->i_op
|| !dir
->i_op
->rmdir
)
1870 down(&dentry
->d_inode
->i_sem
);
1871 dentry_unhash(dentry
);
1872 if (d_mountpoint(dentry
))
1875 error
= security_inode_rmdir(dir
, dentry
);
1877 error
= dir
->i_op
->rmdir(dir
, dentry
);
1879 dentry
->d_inode
->i_flags
|= S_DEAD
;
1882 up(&dentry
->d_inode
->i_sem
);
1891 asmlinkage
long sys_rmdir(const char __user
* pathname
)
1895 struct dentry
*dentry
;
1896 struct nameidata nd
;
1898 name
= getname(pathname
);
1900 return PTR_ERR(name
);
1902 error
= path_lookup(name
, LOOKUP_PARENT
, &nd
);
1906 switch(nd
.last_type
) {
1917 down(&nd
.dentry
->d_inode
->i_sem
);
1918 dentry
= lookup_hash(&nd
.last
, nd
.dentry
);
1919 error
= PTR_ERR(dentry
);
1920 if (!IS_ERR(dentry
)) {
1921 error
= vfs_rmdir(nd
.dentry
->d_inode
, dentry
);
1924 up(&nd
.dentry
->d_inode
->i_sem
);
1932 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1934 int error
= may_delete(dir
, dentry
, 0);
1939 if (!dir
->i_op
|| !dir
->i_op
->unlink
)
1944 down(&dentry
->d_inode
->i_sem
);
1945 if (d_mountpoint(dentry
))
1948 error
= security_inode_unlink(dir
, dentry
);
1950 error
= dir
->i_op
->unlink(dir
, dentry
);
1952 up(&dentry
->d_inode
->i_sem
);
1954 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1955 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
1963 * Make sure that the actual truncation of the file will occur outside its
1964 * directory's i_sem. Truncate can take a long time if there is a lot of
1965 * writeout happening, and we don't want to prevent access to the directory
1966 * while waiting on the I/O.
1968 asmlinkage
long sys_unlink(const char __user
* pathname
)
1972 struct dentry
*dentry
;
1973 struct nameidata nd
;
1974 struct inode
*inode
= NULL
;
1976 name
= getname(pathname
);
1978 return PTR_ERR(name
);
1980 error
= path_lookup(name
, LOOKUP_PARENT
, &nd
);
1984 if (nd
.last_type
!= LAST_NORM
)
1986 down(&nd
.dentry
->d_inode
->i_sem
);
1987 dentry
= lookup_hash(&nd
.last
, nd
.dentry
);
1988 error
= PTR_ERR(dentry
);
1989 if (!IS_ERR(dentry
)) {
1990 /* Why not before? Because we want correct error value */
1991 if (nd
.last
.name
[nd
.last
.len
])
1993 inode
= dentry
->d_inode
;
1995 atomic_inc(&inode
->i_count
);
1996 error
= vfs_unlink(nd
.dentry
->d_inode
, dentry
);
2000 up(&nd
.dentry
->d_inode
->i_sem
);
2002 iput(inode
); /* truncate the inode here */
2010 error
= !dentry
->d_inode
? -ENOENT
:
2011 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2015 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
, int mode
)
2017 int error
= may_create(dir
, dentry
, NULL
);
2022 if (!dir
->i_op
|| !dir
->i_op
->symlink
)
2025 error
= security_inode_symlink(dir
, dentry
, oldname
);
2030 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2032 fsnotify_create(dir
, dentry
->d_name
.name
);
2036 asmlinkage
long sys_symlink(const char __user
* oldname
, const char __user
* newname
)
2042 from
= getname(oldname
);
2044 return PTR_ERR(from
);
2045 to
= getname(newname
);
2046 error
= PTR_ERR(to
);
2048 struct dentry
*dentry
;
2049 struct nameidata nd
;
2051 error
= path_lookup(to
, LOOKUP_PARENT
, &nd
);
2054 dentry
= lookup_create(&nd
, 0);
2055 error
= PTR_ERR(dentry
);
2056 if (!IS_ERR(dentry
)) {
2057 error
= vfs_symlink(nd
.dentry
->d_inode
, dentry
, from
, S_IALLUGO
);
2060 up(&nd
.dentry
->d_inode
->i_sem
);
2069 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2071 struct inode
*inode
= old_dentry
->d_inode
;
2077 error
= may_create(dir
, new_dentry
, NULL
);
2081 if (dir
->i_sb
!= inode
->i_sb
)
2085 * A link to an append-only or immutable file cannot be created.
2087 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2089 if (!dir
->i_op
|| !dir
->i_op
->link
)
2091 if (S_ISDIR(old_dentry
->d_inode
->i_mode
))
2094 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2098 down(&old_dentry
->d_inode
->i_sem
);
2100 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2101 up(&old_dentry
->d_inode
->i_sem
);
2103 fsnotify_create(dir
, new_dentry
->d_name
.name
);
2108 * Hardlinks are often used in delicate situations. We avoid
2109 * security-related surprises by not following symlinks on the
2112 * We don't follow them on the oldname either to be compatible
2113 * with linux 2.0, and to avoid hard-linking to directories
2114 * and other special files. --ADM
2116 asmlinkage
long sys_link(const char __user
* oldname
, const char __user
* newname
)
2118 struct dentry
*new_dentry
;
2119 struct nameidata nd
, old_nd
;
2123 to
= getname(newname
);
2127 error
= __user_walk(oldname
, 0, &old_nd
);
2130 error
= path_lookup(to
, LOOKUP_PARENT
, &nd
);
2134 if (old_nd
.mnt
!= nd
.mnt
)
2136 new_dentry
= lookup_create(&nd
, 0);
2137 error
= PTR_ERR(new_dentry
);
2138 if (!IS_ERR(new_dentry
)) {
2139 error
= vfs_link(old_nd
.dentry
, nd
.dentry
->d_inode
, new_dentry
);
2142 up(&nd
.dentry
->d_inode
->i_sem
);
2146 path_release(&old_nd
);
2154 * The worst of all namespace operations - renaming directory. "Perverted"
2155 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2157 * a) we can get into loop creation. Check is done in is_subdir().
2158 * b) race potential - two innocent renames can create a loop together.
2159 * That's where 4.4 screws up. Current fix: serialization on
2160 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2162 * c) we have to lock _three_ objects - parents and victim (if it exists).
2163 * And that - after we got ->i_sem on parents (until then we don't know
2164 * whether the target exists). Solution: try to be smart with locking
2165 * order for inodes. We rely on the fact that tree topology may change
2166 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2167 * move will be locked. Thus we can rank directories by the tree
2168 * (ancestors first) and rank all non-directories after them.
2169 * That works since everybody except rename does "lock parent, lookup,
2170 * lock child" and rename is under ->s_vfs_rename_sem.
2171 * HOWEVER, it relies on the assumption that any object with ->lookup()
2172 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2173 * we'd better make sure that there's no link(2) for them.
2174 * d) some filesystems don't support opened-but-unlinked directories,
2175 * either because of layout or because they are not ready to deal with
2176 * all cases correctly. The latter will be fixed (taking this sort of
2177 * stuff into VFS), but the former is not going away. Solution: the same
2178 * trick as in rmdir().
2179 * e) conversion from fhandle to dentry may come in the wrong moment - when
2180 * we are removing the target. Solution: we will have to grab ->i_sem
2181 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2182 * ->i_sem on parents, which works but leads to some truely excessive
2185 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2186 struct inode
*new_dir
, struct dentry
*new_dentry
)
2189 struct inode
*target
;
2192 * If we are going to change the parent - check write permissions,
2193 * we'll need to flip '..'.
2195 if (new_dir
!= old_dir
) {
2196 error
= permission(old_dentry
->d_inode
, MAY_WRITE
, NULL
);
2201 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2205 target
= new_dentry
->d_inode
;
2207 down(&target
->i_sem
);
2208 dentry_unhash(new_dentry
);
2210 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2213 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2216 target
->i_flags
|= S_DEAD
;
2218 if (d_unhashed(new_dentry
))
2219 d_rehash(new_dentry
);
2223 d_move(old_dentry
,new_dentry
);
2227 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
2228 struct inode
*new_dir
, struct dentry
*new_dentry
)
2230 struct inode
*target
;
2233 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2238 target
= new_dentry
->d_inode
;
2240 down(&target
->i_sem
);
2241 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2244 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2246 /* The following d_move() should become unconditional */
2247 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_ODD_RENAME
))
2248 d_move(old_dentry
, new_dentry
);
2256 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2257 struct inode
*new_dir
, struct dentry
*new_dentry
)
2260 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2261 const char *old_name
;
2263 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
2266 error
= may_delete(old_dir
, old_dentry
, is_dir
);
2270 if (!new_dentry
->d_inode
)
2271 error
= may_create(new_dir
, new_dentry
, NULL
);
2273 error
= may_delete(new_dir
, new_dentry
, is_dir
);
2277 if (!old_dir
->i_op
|| !old_dir
->i_op
->rename
)
2280 DQUOT_INIT(old_dir
);
2281 DQUOT_INIT(new_dir
);
2283 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
2286 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
2288 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
2290 const char *new_name
= old_dentry
->d_name
.name
;
2291 fsnotify_move(old_dir
, new_dir
, old_name
, new_name
, is_dir
,
2292 new_dentry
->d_inode
, old_dentry
->d_inode
);
2294 fsnotify_oldname_free(old_name
);
2299 static inline int do_rename(const char * oldname
, const char * newname
)
2302 struct dentry
* old_dir
, * new_dir
;
2303 struct dentry
* old_dentry
, *new_dentry
;
2304 struct dentry
* trap
;
2305 struct nameidata oldnd
, newnd
;
2307 error
= path_lookup(oldname
, LOOKUP_PARENT
, &oldnd
);
2311 error
= path_lookup(newname
, LOOKUP_PARENT
, &newnd
);
2316 if (oldnd
.mnt
!= newnd
.mnt
)
2319 old_dir
= oldnd
.dentry
;
2321 if (oldnd
.last_type
!= LAST_NORM
)
2324 new_dir
= newnd
.dentry
;
2325 if (newnd
.last_type
!= LAST_NORM
)
2328 trap
= lock_rename(new_dir
, old_dir
);
2330 old_dentry
= lookup_hash(&oldnd
.last
, old_dir
);
2331 error
= PTR_ERR(old_dentry
);
2332 if (IS_ERR(old_dentry
))
2334 /* source must exist */
2336 if (!old_dentry
->d_inode
)
2338 /* unless the source is a directory trailing slashes give -ENOTDIR */
2339 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
2341 if (oldnd
.last
.name
[oldnd
.last
.len
])
2343 if (newnd
.last
.name
[newnd
.last
.len
])
2346 /* source should not be ancestor of target */
2348 if (old_dentry
== trap
)
2350 new_dentry
= lookup_hash(&newnd
.last
, new_dir
);
2351 error
= PTR_ERR(new_dentry
);
2352 if (IS_ERR(new_dentry
))
2354 /* target should not be an ancestor of source */
2356 if (new_dentry
== trap
)
2359 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
2360 new_dir
->d_inode
, new_dentry
);
2366 unlock_rename(new_dir
, old_dir
);
2368 path_release(&newnd
);
2370 path_release(&oldnd
);
2375 asmlinkage
long sys_rename(const char __user
* oldname
, const char __user
* newname
)
2381 from
= getname(oldname
);
2383 return PTR_ERR(from
);
2384 to
= getname(newname
);
2385 error
= PTR_ERR(to
);
2387 error
= do_rename(from
,to
);
2394 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
2398 len
= PTR_ERR(link
);
2403 if (len
> (unsigned) buflen
)
2405 if (copy_to_user(buffer
, link
, len
))
2412 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2413 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2414 * using) it for any given inode is up to filesystem.
2416 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2418 struct nameidata nd
;
2422 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
2423 if (!IS_ERR(cookie
)) {
2424 int res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
2425 if (dentry
->d_inode
->i_op
->put_link
)
2426 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
2427 cookie
= ERR_PTR(res
);
2429 return PTR_ERR(cookie
);
2432 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
2434 return __vfs_follow_link(nd
, link
);
2437 /* get the link contents into pagecache */
2438 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
2441 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
2442 page
= read_cache_page(mapping
, 0, (filler_t
*)mapping
->a_ops
->readpage
,
2446 wait_on_page_locked(page
);
2447 if (!PageUptodate(page
))
2453 page_cache_release(page
);
2454 return ERR_PTR(-EIO
);
2460 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2462 struct page
*page
= NULL
;
2463 char *s
= page_getlink(dentry
, &page
);
2464 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
2467 page_cache_release(page
);
2472 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
2474 struct page
*page
= NULL
;
2475 nd_set_link(nd
, page_getlink(dentry
, &page
));
2479 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2481 struct page
*page
= cookie
;
2485 page_cache_release(page
);
2489 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
2491 struct address_space
*mapping
= inode
->i_mapping
;
2492 struct page
*page
= grab_cache_page(mapping
, 0);
2498 err
= mapping
->a_ops
->prepare_write(NULL
, page
, 0, len
-1);
2501 kaddr
= kmap_atomic(page
, KM_USER0
);
2502 memcpy(kaddr
, symname
, len
-1);
2503 kunmap_atomic(kaddr
, KM_USER0
);
2504 mapping
->a_ops
->commit_write(NULL
, page
, 0, len
-1);
2506 * Notice that we are _not_ going to block here - end of page is
2507 * unmapped, so this will only try to map the rest of page, see
2508 * that it is unmapped (typically even will not look into inode -
2509 * ->i_size will be enough for everything) and zero it out.
2510 * OTOH it's obviously correct and should make the page up-to-date.
2512 if (!PageUptodate(page
)) {
2513 err
= mapping
->a_ops
->readpage(NULL
, page
);
2514 wait_on_page_locked(page
);
2518 page_cache_release(page
);
2521 mark_inode_dirty(inode
);
2525 page_cache_release(page
);
2530 struct inode_operations page_symlink_inode_operations
= {
2531 .readlink
= generic_readlink
,
2532 .follow_link
= page_follow_link_light
,
2533 .put_link
= page_put_link
,
2536 EXPORT_SYMBOL(__user_walk
);
2537 EXPORT_SYMBOL(follow_down
);
2538 EXPORT_SYMBOL(follow_up
);
2539 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
2540 EXPORT_SYMBOL(getname
);
2541 EXPORT_SYMBOL(lock_rename
);
2542 EXPORT_SYMBOL(lookup_hash
);
2543 EXPORT_SYMBOL(lookup_one_len
);
2544 EXPORT_SYMBOL(page_follow_link_light
);
2545 EXPORT_SYMBOL(page_put_link
);
2546 EXPORT_SYMBOL(page_readlink
);
2547 EXPORT_SYMBOL(page_symlink
);
2548 EXPORT_SYMBOL(page_symlink_inode_operations
);
2549 EXPORT_SYMBOL(path_lookup
);
2550 EXPORT_SYMBOL(path_release
);
2551 EXPORT_SYMBOL(path_walk
);
2552 EXPORT_SYMBOL(permission
);
2553 EXPORT_SYMBOL(vfs_permission
);
2554 EXPORT_SYMBOL(unlock_rename
);
2555 EXPORT_SYMBOL(vfs_create
);
2556 EXPORT_SYMBOL(vfs_follow_link
);
2557 EXPORT_SYMBOL(vfs_link
);
2558 EXPORT_SYMBOL(vfs_mkdir
);
2559 EXPORT_SYMBOL(vfs_mknod
);
2560 EXPORT_SYMBOL(generic_permission
);
2561 EXPORT_SYMBOL(vfs_readlink
);
2562 EXPORT_SYMBOL(vfs_rename
);
2563 EXPORT_SYMBOL(vfs_rmdir
);
2564 EXPORT_SYMBOL(vfs_symlink
);
2565 EXPORT_SYMBOL(vfs_unlink
);
2566 EXPORT_SYMBOL(dentry_unhash
);
2567 EXPORT_SYMBOL(generic_readlink
);