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/dnotify.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 <asm/namei.h>
32 #include <asm/uaccess.h>
34 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
36 /* [Feb-1997 T. Schoebel-Theuer]
37 * Fundamental changes in the pathname lookup mechanisms (namei)
38 * were necessary because of omirr. The reason is that omirr needs
39 * to know the _real_ pathname, not the user-supplied one, in case
40 * of symlinks (and also when transname replacements occur).
42 * The new code replaces the old recursive symlink resolution with
43 * an iterative one (in case of non-nested symlink chains). It does
44 * this with calls to <fs>_follow_link().
45 * As a side effect, dir_namei(), _namei() and follow_link() are now
46 * replaced with a single function lookup_dentry() that can handle all
47 * the special cases of the former code.
49 * With the new dcache, the pathname is stored at each inode, at least as
50 * long as the refcount of the inode is positive. As a side effect, the
51 * size of the dcache depends on the inode cache and thus is dynamic.
53 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
54 * resolution to correspond with current state of the code.
56 * Note that the symlink resolution is not *completely* iterative.
57 * There is still a significant amount of tail- and mid- recursion in
58 * the algorithm. Also, note that <fs>_readlink() is not used in
59 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
60 * may return different results than <fs>_follow_link(). Many virtual
61 * filesystems (including /proc) exhibit this behavior.
64 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
65 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
66 * and the name already exists in form of a symlink, try to create the new
67 * name indicated by the symlink. The old code always complained that the
68 * name already exists, due to not following the symlink even if its target
69 * is nonexistent. The new semantics affects also mknod() and link() when
70 * the name is a symlink pointing to a non-existant name.
72 * I don't know which semantics is the right one, since I have no access
73 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
74 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
75 * "old" one. Personally, I think the new semantics is much more logical.
76 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
77 * file does succeed in both HP-UX and SunOs, but not in Solaris
78 * and in the old Linux semantics.
81 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
82 * semantics. See the comments in "open_namei" and "do_link" below.
84 * [10-Sep-98 Alan Modra] Another symlink change.
87 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
88 * inside the path - always follow.
89 * in the last component in creation/removal/renaming - never follow.
90 * if LOOKUP_FOLLOW passed - follow.
91 * if the pathname has trailing slashes - follow.
92 * otherwise - don't follow.
93 * (applied in that order).
95 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
96 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
97 * During the 2.4 we need to fix the userland stuff depending on it -
98 * hopefully we will be able to get rid of that wart in 2.5. So far only
99 * XEmacs seems to be relying on it...
102 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
103 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
104 * any extra contention...
107 /* In order to reduce some races, while at the same time doing additional
108 * checking and hopefully speeding things up, we copy filenames to the
109 * kernel data space before using them..
111 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
112 * PATH_MAX includes the nul terminator --RR.
114 static inline int do_getname(const char __user
*filename
, char *page
)
117 unsigned long len
= PATH_MAX
;
119 if (!segment_eq(get_fs(), KERNEL_DS
)) {
120 if ((unsigned long) filename
>= TASK_SIZE
)
122 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
123 len
= TASK_SIZE
- (unsigned long) filename
;
126 retval
= strncpy_from_user(page
, filename
, len
);
130 return -ENAMETOOLONG
;
136 char * getname(const char __user
* filename
)
140 result
= ERR_PTR(-ENOMEM
);
143 int retval
= do_getname(filename
, tmp
);
148 result
= ERR_PTR(retval
);
151 audit_getname(result
);
155 #ifdef CONFIG_AUDITSYSCALL
156 void putname(const char *name
)
158 if (unlikely(current
->audit_context
))
163 EXPORT_SYMBOL(putname
);
168 * generic_permission - check for access rights on a Posix-like filesystem
169 * @inode: inode to check access rights for
170 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
171 * @check_acl: optional callback to check for Posix ACLs
173 * Used to check for read/write/execute permissions on a file.
174 * We use "fsuid" for this, letting us set arbitrary permissions
175 * for filesystem access without changing the "normal" uids which
176 * are used for other things..
178 int generic_permission(struct inode
*inode
, int mask
,
179 int (*check_acl
)(struct inode
*inode
, int mask
))
181 umode_t mode
= inode
->i_mode
;
183 if (current
->fsuid
== inode
->i_uid
)
186 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
187 int error
= check_acl(inode
, mask
);
188 if (error
== -EACCES
)
189 goto check_capabilities
;
190 else if (error
!= -EAGAIN
)
194 if (in_group_p(inode
->i_gid
))
199 * If the DACs are ok we don't need any capability check.
201 if (((mode
& mask
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
)) == mask
))
206 * Read/write DACs are always overridable.
207 * Executable DACs are overridable if at least one exec bit is set.
209 if (!(mask
& MAY_EXEC
) ||
210 (inode
->i_mode
& S_IXUGO
) || S_ISDIR(inode
->i_mode
))
211 if (capable(CAP_DAC_OVERRIDE
))
215 * Searching includes executable on directories, else just read.
217 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
218 if (capable(CAP_DAC_READ_SEARCH
))
224 int permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
228 if (mask
& MAY_WRITE
) {
229 umode_t mode
= inode
->i_mode
;
232 * Nobody gets write access to a read-only fs.
234 if (IS_RDONLY(inode
) &&
235 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
239 * Nobody gets write access to an immutable file.
241 if (IS_IMMUTABLE(inode
))
246 /* Ordinary permission routines do not understand MAY_APPEND. */
247 submask
= mask
& ~MAY_APPEND
;
248 if (inode
->i_op
&& inode
->i_op
->permission
)
249 retval
= inode
->i_op
->permission(inode
, submask
, nd
);
251 retval
= generic_permission(inode
, submask
, NULL
);
255 return security_inode_permission(inode
, mask
, nd
);
259 * get_write_access() gets write permission for a file.
260 * put_write_access() releases this write permission.
261 * This is used for regular files.
262 * We cannot support write (and maybe mmap read-write shared) accesses and
263 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
264 * can have the following values:
265 * 0: no writers, no VM_DENYWRITE mappings
266 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
267 * > 0: (i_writecount) users are writing to the file.
269 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
270 * except for the cases where we don't hold i_writecount yet. Then we need to
271 * use {get,deny}_write_access() - these functions check the sign and refuse
272 * to do the change if sign is wrong. Exclusion between them is provided by
273 * the inode->i_lock spinlock.
276 int get_write_access(struct inode
* inode
)
278 spin_lock(&inode
->i_lock
);
279 if (atomic_read(&inode
->i_writecount
) < 0) {
280 spin_unlock(&inode
->i_lock
);
283 atomic_inc(&inode
->i_writecount
);
284 spin_unlock(&inode
->i_lock
);
289 int deny_write_access(struct file
* file
)
291 struct inode
*inode
= file
->f_dentry
->d_inode
;
293 spin_lock(&inode
->i_lock
);
294 if (atomic_read(&inode
->i_writecount
) > 0) {
295 spin_unlock(&inode
->i_lock
);
298 atomic_dec(&inode
->i_writecount
);
299 spin_unlock(&inode
->i_lock
);
304 void path_release(struct nameidata
*nd
)
311 * umount() mustn't call path_release()/mntput() as that would clear
314 void path_release_on_umount(struct nameidata
*nd
)
321 * Internal lookup() using the new generic dcache.
324 static struct dentry
* cached_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
326 struct dentry
* dentry
= __d_lookup(parent
, name
);
328 /* lockess __d_lookup may fail due to concurrent d_move()
329 * in some unrelated directory, so try with d_lookup
332 dentry
= d_lookup(parent
, name
);
334 if (dentry
&& dentry
->d_op
&& dentry
->d_op
->d_revalidate
) {
335 if (!dentry
->d_op
->d_revalidate(dentry
, nd
) && !d_invalidate(dentry
)) {
344 * Short-cut version of permission(), for calling by
345 * path_walk(), when dcache lock is held. Combines parts
346 * of permission() and generic_permission(), and tests ONLY for
347 * MAY_EXEC permission.
349 * If appropriate, check DAC only. If not appropriate, or
350 * short-cut DAC fails, then call permission() to do more
351 * complete permission check.
353 static inline int exec_permission_lite(struct inode
*inode
,
354 struct nameidata
*nd
)
356 umode_t mode
= inode
->i_mode
;
358 if (inode
->i_op
&& inode
->i_op
->permission
)
361 if (current
->fsuid
== inode
->i_uid
)
363 else if (in_group_p(inode
->i_gid
))
369 if ((inode
->i_mode
& S_IXUGO
) && capable(CAP_DAC_OVERRIDE
))
372 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_OVERRIDE
))
375 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_READ_SEARCH
))
380 return security_inode_permission(inode
, MAY_EXEC
, nd
);
384 * This is called when everything else fails, and we actually have
385 * to go to the low-level filesystem to find out what we should do..
387 * We get the directory semaphore, and after getting that we also
388 * make sure that nobody added the entry to the dcache in the meantime..
391 static struct dentry
* real_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
393 struct dentry
* result
;
394 struct inode
*dir
= parent
->d_inode
;
398 * First re-do the cached lookup just in case it was created
399 * while we waited for the directory semaphore..
401 * FIXME! This could use version numbering or similar to
402 * avoid unnecessary cache lookups.
404 * The "dcache_lock" is purely to protect the RCU list walker
405 * from concurrent renames at this point (we mustn't get false
406 * negatives from the RCU list walk here, unlike the optimistic
409 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
411 result
= d_lookup(parent
, name
);
413 struct dentry
* dentry
= d_alloc(parent
, name
);
414 result
= ERR_PTR(-ENOMEM
);
416 result
= dir
->i_op
->lookup(dir
, dentry
, nd
);
427 * Uhhuh! Nasty case: the cache was re-populated while
428 * we waited on the semaphore. Need to revalidate.
431 if (result
->d_op
&& result
->d_op
->d_revalidate
) {
432 if (!result
->d_op
->d_revalidate(result
, nd
) && !d_invalidate(result
)) {
434 result
= ERR_PTR(-ENOENT
);
440 static int __emul_lookup_dentry(const char *, struct nameidata
*);
444 walk_init_root(const char *name
, struct nameidata
*nd
)
446 read_lock(¤t
->fs
->lock
);
447 if (current
->fs
->altroot
&& !(nd
->flags
& LOOKUP_NOALT
)) {
448 nd
->mnt
= mntget(current
->fs
->altrootmnt
);
449 nd
->dentry
= dget(current
->fs
->altroot
);
450 read_unlock(¤t
->fs
->lock
);
451 if (__emul_lookup_dentry(name
,nd
))
453 read_lock(¤t
->fs
->lock
);
455 nd
->mnt
= mntget(current
->fs
->rootmnt
);
456 nd
->dentry
= dget(current
->fs
->root
);
457 read_unlock(¤t
->fs
->lock
);
461 static inline int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
470 if (!walk_init_root(link
, nd
))
471 /* weird __emul_prefix() stuff did it */
474 res
= link_path_walk(link
, nd
);
476 if (nd
->depth
|| res
|| nd
->last_type
!=LAST_NORM
)
479 * If it is an iterative symlinks resolution in open_namei() we
480 * have to copy the last component. And all that crap because of
481 * bloody create() on broken symlinks. Furrfu...
484 if (unlikely(!name
)) {
488 strcpy(name
, nd
->last
.name
);
489 nd
->last
.name
= name
;
493 return PTR_ERR(link
);
497 struct vfsmount
*mnt
;
498 struct dentry
*dentry
;
501 static inline int __do_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
505 touch_atime(nd
->mnt
, dentry
);
506 nd_set_link(nd
, NULL
);
507 error
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
509 char *s
= nd_get_link(nd
);
511 error
= __vfs_follow_link(nd
, s
);
512 if (dentry
->d_inode
->i_op
->put_link
)
513 dentry
->d_inode
->i_op
->put_link(dentry
, nd
);
520 * This limits recursive symlink follows to 8, while
521 * limiting consecutive symlinks to 40.
523 * Without that kind of total limit, nasty chains of consecutive
524 * symlinks can cause almost arbitrarily long lookups.
526 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
529 if (current
->link_count
>= MAX_NESTED_LINKS
)
531 if (current
->total_link_count
>= 40)
533 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
535 err
= security_inode_follow_link(path
->dentry
, nd
);
538 current
->link_count
++;
539 current
->total_link_count
++;
542 err
= __do_follow_link(path
->dentry
, nd
);
545 current
->link_count
--;
554 int follow_up(struct vfsmount
**mnt
, struct dentry
**dentry
)
556 struct vfsmount
*parent
;
557 struct dentry
*mountpoint
;
558 spin_lock(&vfsmount_lock
);
559 parent
=(*mnt
)->mnt_parent
;
560 if (parent
== *mnt
) {
561 spin_unlock(&vfsmount_lock
);
565 mountpoint
=dget((*mnt
)->mnt_mountpoint
);
566 spin_unlock(&vfsmount_lock
);
568 *dentry
= mountpoint
;
574 /* no need for dcache_lock, as serialization is taken care in
577 static int follow_mount(struct vfsmount
**mnt
, struct dentry
**dentry
)
580 while (d_mountpoint(*dentry
)) {
581 struct vfsmount
*mounted
= lookup_mnt(*mnt
, *dentry
);
587 *dentry
= dget(mounted
->mnt_root
);
593 /* no need for dcache_lock, as serialization is taken care in
596 static inline int __follow_down(struct vfsmount
**mnt
, struct dentry
**dentry
)
598 struct vfsmount
*mounted
;
600 mounted
= lookup_mnt(*mnt
, *dentry
);
605 *dentry
= dget(mounted
->mnt_root
);
611 int follow_down(struct vfsmount
**mnt
, struct dentry
**dentry
)
613 return __follow_down(mnt
,dentry
);
616 static inline void follow_dotdot(struct vfsmount
**mnt
, struct dentry
**dentry
)
619 struct vfsmount
*parent
;
620 struct dentry
*old
= *dentry
;
622 read_lock(¤t
->fs
->lock
);
623 if (*dentry
== current
->fs
->root
&&
624 *mnt
== current
->fs
->rootmnt
) {
625 read_unlock(¤t
->fs
->lock
);
628 read_unlock(¤t
->fs
->lock
);
629 spin_lock(&dcache_lock
);
630 if (*dentry
!= (*mnt
)->mnt_root
) {
631 *dentry
= dget((*dentry
)->d_parent
);
632 spin_unlock(&dcache_lock
);
636 spin_unlock(&dcache_lock
);
637 spin_lock(&vfsmount_lock
);
638 parent
= (*mnt
)->mnt_parent
;
639 if (parent
== *mnt
) {
640 spin_unlock(&vfsmount_lock
);
644 *dentry
= dget((*mnt
)->mnt_mountpoint
);
645 spin_unlock(&vfsmount_lock
);
650 follow_mount(mnt
, dentry
);
654 * It's more convoluted than I'd like it to be, but... it's still fairly
655 * small and for now I'd prefer to have fast path as straight as possible.
656 * It _is_ time-critical.
658 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
661 struct vfsmount
*mnt
= nd
->mnt
;
662 struct dentry
*dentry
= __d_lookup(nd
->dentry
, name
);
666 if (dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
667 goto need_revalidate
;
670 path
->dentry
= dentry
;
674 dentry
= real_lookup(nd
->dentry
, name
, nd
);
680 if (dentry
->d_op
->d_revalidate(dentry
, nd
))
682 if (d_invalidate(dentry
))
688 return PTR_ERR(dentry
);
693 * This is the basic name resolution function, turning a pathname into
694 * the final dentry. We expect 'base' to be positive and a directory.
696 * Returns 0 and nd will have valid dentry and mnt on success.
697 * Returns error and drops reference to input namei data on failure.
699 static fastcall
int __link_path_walk(const char * name
, struct nameidata
*nd
)
704 unsigned int lookup_flags
= nd
->flags
;
711 inode
= nd
->dentry
->d_inode
;
713 lookup_flags
= LOOKUP_FOLLOW
;
715 /* At this point we know we have a real path component. */
721 err
= exec_permission_lite(inode
, nd
);
722 if (err
== -EAGAIN
) {
723 err
= permission(inode
, MAY_EXEC
, nd
);
729 c
= *(const unsigned char *)name
;
731 hash
= init_name_hash();
734 hash
= partial_name_hash(c
, hash
);
735 c
= *(const unsigned char *)name
;
736 } while (c
&& (c
!= '/'));
737 this.len
= name
- (const char *) this.name
;
738 this.hash
= end_name_hash(hash
);
740 /* remove trailing slashes? */
743 while (*++name
== '/');
745 goto last_with_slashes
;
748 * "." and ".." are special - ".." especially so because it has
749 * to be able to know about the current root directory and
750 * parent relationships.
752 if (this.name
[0] == '.') switch (this.len
) {
756 if (this.name
[1] != '.')
758 follow_dotdot(&nd
->mnt
, &nd
->dentry
);
759 inode
= nd
->dentry
->d_inode
;
765 * See if the low-level filesystem might want
766 * to use its own hash..
768 if (nd
->dentry
->d_op
&& nd
->dentry
->d_op
->d_hash
) {
769 err
= nd
->dentry
->d_op
->d_hash(nd
->dentry
, &this);
773 nd
->flags
|= LOOKUP_CONTINUE
;
774 /* This does the actual lookups.. */
775 err
= do_lookup(nd
, &this, &next
);
778 /* Check mountpoints.. */
779 follow_mount(&next
.mnt
, &next
.dentry
);
782 inode
= next
.dentry
->d_inode
;
789 if (inode
->i_op
->follow_link
) {
790 err
= do_follow_link(&next
, nd
);
794 inode
= nd
->dentry
->d_inode
;
803 nd
->dentry
= next
.dentry
;
806 if (!inode
->i_op
->lookup
)
809 /* here ends the main loop */
812 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
814 nd
->flags
&= ~LOOKUP_CONTINUE
;
815 if (lookup_flags
& LOOKUP_PARENT
)
817 if (this.name
[0] == '.') switch (this.len
) {
821 if (this.name
[1] != '.')
823 follow_dotdot(&nd
->mnt
, &nd
->dentry
);
824 inode
= nd
->dentry
->d_inode
;
829 if (nd
->dentry
->d_op
&& nd
->dentry
->d_op
->d_hash
) {
830 err
= nd
->dentry
->d_op
->d_hash(nd
->dentry
, &this);
834 err
= do_lookup(nd
, &this, &next
);
837 follow_mount(&next
.mnt
, &next
.dentry
);
838 inode
= next
.dentry
->d_inode
;
839 if ((lookup_flags
& LOOKUP_FOLLOW
)
840 && inode
&& inode
->i_op
&& inode
->i_op
->follow_link
) {
841 err
= do_follow_link(&next
, nd
);
844 inode
= nd
->dentry
->d_inode
;
848 nd
->dentry
= next
.dentry
;
853 if (lookup_flags
& LOOKUP_DIRECTORY
) {
855 if (!inode
->i_op
|| !inode
->i_op
->lookup
)
861 nd
->last_type
= LAST_NORM
;
862 if (this.name
[0] != '.')
865 nd
->last_type
= LAST_DOT
;
866 else if (this.len
== 2 && this.name
[1] == '.')
867 nd
->last_type
= LAST_DOTDOT
;
872 * We bypassed the ordinary revalidation routines.
873 * We may need to check the cached dentry for staleness.
875 if (nd
->dentry
&& nd
->dentry
->d_sb
&&
876 (nd
->dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)) {
878 /* Note: we do not d_invalidate() */
879 if (!nd
->dentry
->d_op
->d_revalidate(nd
->dentry
, nd
))
894 * Wrapper to retry pathname resolution whenever the underlying
895 * file system returns an ESTALE.
897 * Retry the whole path once, forcing real lookup requests
898 * instead of relying on the dcache.
900 int fastcall
link_path_walk(const char *name
, struct nameidata
*nd
)
902 struct nameidata save
= *nd
;
905 /* make sure the stuff we saved doesn't go away */
909 result
= __link_path_walk(name
, nd
);
910 if (result
== -ESTALE
) {
914 nd
->flags
|= LOOKUP_REVAL
;
915 result
= __link_path_walk(name
, nd
);
924 int fastcall
path_walk(const char * name
, struct nameidata
*nd
)
926 current
->total_link_count
= 0;
927 return link_path_walk(name
, nd
);
931 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
932 * everything is done. Returns 0 and drops input nd, if lookup failed;
934 static int __emul_lookup_dentry(const char *name
, struct nameidata
*nd
)
936 if (path_walk(name
, nd
))
937 return 0; /* something went wrong... */
939 if (!nd
->dentry
->d_inode
|| S_ISDIR(nd
->dentry
->d_inode
->i_mode
)) {
940 struct dentry
*old_dentry
= nd
->dentry
;
941 struct vfsmount
*old_mnt
= nd
->mnt
;
942 struct qstr last
= nd
->last
;
943 int last_type
= nd
->last_type
;
945 * NAME was not found in alternate root or it's a directory. Try to find
946 * it in the normal root:
948 nd
->last_type
= LAST_ROOT
;
949 read_lock(¤t
->fs
->lock
);
950 nd
->mnt
= mntget(current
->fs
->rootmnt
);
951 nd
->dentry
= dget(current
->fs
->root
);
952 read_unlock(¤t
->fs
->lock
);
953 if (path_walk(name
, nd
) == 0) {
954 if (nd
->dentry
->d_inode
) {
961 nd
->dentry
= old_dentry
;
964 nd
->last_type
= last_type
;
969 void set_fs_altroot(void)
971 char *emul
= __emul_prefix();
973 struct vfsmount
*mnt
= NULL
, *oldmnt
;
974 struct dentry
*dentry
= NULL
, *olddentry
;
979 err
= path_lookup(emul
, LOOKUP_FOLLOW
|LOOKUP_DIRECTORY
|LOOKUP_NOALT
, &nd
);
985 write_lock(¤t
->fs
->lock
);
986 oldmnt
= current
->fs
->altrootmnt
;
987 olddentry
= current
->fs
->altroot
;
988 current
->fs
->altrootmnt
= mnt
;
989 current
->fs
->altroot
= dentry
;
990 write_unlock(¤t
->fs
->lock
);
997 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
998 int fastcall
path_lookup(const char *name
, unsigned int flags
, struct nameidata
*nd
)
1002 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1006 read_lock(¤t
->fs
->lock
);
1008 if (current
->fs
->altroot
&& !(nd
->flags
& LOOKUP_NOALT
)) {
1009 nd
->mnt
= mntget(current
->fs
->altrootmnt
);
1010 nd
->dentry
= dget(current
->fs
->altroot
);
1011 read_unlock(¤t
->fs
->lock
);
1012 if (__emul_lookup_dentry(name
,nd
))
1013 goto out
; /* found in altroot */
1014 read_lock(¤t
->fs
->lock
);
1016 nd
->mnt
= mntget(current
->fs
->rootmnt
);
1017 nd
->dentry
= dget(current
->fs
->root
);
1019 nd
->mnt
= mntget(current
->fs
->pwdmnt
);
1020 nd
->dentry
= dget(current
->fs
->pwd
);
1022 read_unlock(¤t
->fs
->lock
);
1023 current
->total_link_count
= 0;
1024 retval
= link_path_walk(name
, nd
);
1026 if (unlikely(current
->audit_context
1027 && nd
&& nd
->dentry
&& nd
->dentry
->d_inode
))
1028 audit_inode(name
, nd
->dentry
->d_inode
);
1033 * Restricted form of lookup. Doesn't follow links, single-component only,
1034 * needs parent already locked. Doesn't follow mounts.
1037 static struct dentry
* __lookup_hash(struct qstr
*name
, struct dentry
* base
, struct nameidata
*nd
)
1039 struct dentry
* dentry
;
1040 struct inode
*inode
;
1043 inode
= base
->d_inode
;
1044 err
= permission(inode
, MAY_EXEC
, nd
);
1045 dentry
= ERR_PTR(err
);
1050 * See if the low-level filesystem might want
1051 * to use its own hash..
1053 if (base
->d_op
&& base
->d_op
->d_hash
) {
1054 err
= base
->d_op
->d_hash(base
, name
);
1055 dentry
= ERR_PTR(err
);
1060 dentry
= cached_lookup(base
, name
, nd
);
1062 struct dentry
*new = d_alloc(base
, name
);
1063 dentry
= ERR_PTR(-ENOMEM
);
1066 dentry
= inode
->i_op
->lookup(inode
, new, nd
);
1076 struct dentry
* lookup_hash(struct qstr
*name
, struct dentry
* base
)
1078 return __lookup_hash(name
, base
, NULL
);
1082 struct dentry
* lookup_one_len(const char * name
, struct dentry
* base
, int len
)
1093 hash
= init_name_hash();
1095 c
= *(const unsigned char *)name
++;
1096 if (c
== '/' || c
== '\0')
1098 hash
= partial_name_hash(c
, hash
);
1100 this.hash
= end_name_hash(hash
);
1102 return lookup_hash(&this, base
);
1104 return ERR_PTR(-EACCES
);
1110 * is used by most simple commands to get the inode of a specified name.
1111 * Open, link etc use their own routines, but this is enough for things
1114 * namei exists in two versions: namei/lnamei. The only difference is
1115 * that namei follows links, while lnamei does not.
1118 int fastcall
__user_walk(const char __user
*name
, unsigned flags
, struct nameidata
*nd
)
1120 char *tmp
= getname(name
);
1121 int err
= PTR_ERR(tmp
);
1124 err
= path_lookup(tmp
, flags
, nd
);
1131 * It's inline, so penalty for filesystems that don't use sticky bit is
1134 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1136 if (!(dir
->i_mode
& S_ISVTX
))
1138 if (inode
->i_uid
== current
->fsuid
)
1140 if (dir
->i_uid
== current
->fsuid
)
1142 return !capable(CAP_FOWNER
);
1146 * Check whether we can remove a link victim from directory dir, check
1147 * whether the type of victim is right.
1148 * 1. We can't do it if dir is read-only (done in permission())
1149 * 2. We should have write and exec permissions on dir
1150 * 3. We can't remove anything from append-only dir
1151 * 4. We can't do anything with immutable dir (done in permission())
1152 * 5. If the sticky bit on dir is set we should either
1153 * a. be owner of dir, or
1154 * b. be owner of victim, or
1155 * c. have CAP_FOWNER capability
1156 * 6. If the victim is append-only or immutable we can't do antyhing with
1157 * links pointing to it.
1158 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1159 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1160 * 9. We can't remove a root or mountpoint.
1161 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1162 * nfs_async_unlink().
1164 static inline int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1168 if (!victim
->d_inode
)
1171 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1173 error
= permission(dir
,MAY_WRITE
| MAY_EXEC
, NULL
);
1178 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1179 IS_IMMUTABLE(victim
->d_inode
))
1182 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1184 if (IS_ROOT(victim
))
1186 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1188 if (IS_DEADDIR(dir
))
1190 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1195 /* Check whether we can create an object with dentry child in directory
1197 * 1. We can't do it if child already exists (open has special treatment for
1198 * this case, but since we are inlined it's OK)
1199 * 2. We can't do it if dir is read-only (done in permission())
1200 * 3. We should have write and exec permissions on dir
1201 * 4. We can't do it if dir is immutable (done in permission())
1203 static inline int may_create(struct inode
*dir
, struct dentry
*child
,
1204 struct nameidata
*nd
)
1208 if (IS_DEADDIR(dir
))
1210 return permission(dir
,MAY_WRITE
| MAY_EXEC
, nd
);
1214 * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1217 * O_DIRECTORY translates into forcing a directory lookup.
1219 static inline int lookup_flags(unsigned int f
)
1221 unsigned long retval
= LOOKUP_FOLLOW
;
1224 retval
&= ~LOOKUP_FOLLOW
;
1226 if ((f
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1227 retval
&= ~LOOKUP_FOLLOW
;
1229 if (f
& O_DIRECTORY
)
1230 retval
|= LOOKUP_DIRECTORY
;
1236 * p1 and p2 should be directories on the same fs.
1238 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1243 down(&p1
->d_inode
->i_sem
);
1247 down(&p1
->d_inode
->i_sb
->s_vfs_rename_sem
);
1249 for (p
= p1
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1250 if (p
->d_parent
== p2
) {
1251 down(&p2
->d_inode
->i_sem
);
1252 down(&p1
->d_inode
->i_sem
);
1257 for (p
= p2
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1258 if (p
->d_parent
== p1
) {
1259 down(&p1
->d_inode
->i_sem
);
1260 down(&p2
->d_inode
->i_sem
);
1265 down(&p1
->d_inode
->i_sem
);
1266 down(&p2
->d_inode
->i_sem
);
1270 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1272 up(&p1
->d_inode
->i_sem
);
1274 up(&p2
->d_inode
->i_sem
);
1275 up(&p1
->d_inode
->i_sb
->s_vfs_rename_sem
);
1279 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1280 struct nameidata
*nd
)
1282 int error
= may_create(dir
, dentry
, nd
);
1287 if (!dir
->i_op
|| !dir
->i_op
->create
)
1288 return -EACCES
; /* shouldn't it be ENOSYS? */
1291 error
= security_inode_create(dir
, dentry
, mode
);
1295 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1297 inode_dir_notify(dir
, DN_CREATE
);
1298 security_inode_post_create(dir
, dentry
, mode
);
1303 int may_open(struct nameidata
*nd
, int acc_mode
, int flag
)
1305 struct dentry
*dentry
= nd
->dentry
;
1306 struct inode
*inode
= dentry
->d_inode
;
1312 if (S_ISLNK(inode
->i_mode
))
1315 if (S_ISDIR(inode
->i_mode
) && (flag
& FMODE_WRITE
))
1318 error
= permission(inode
, acc_mode
, nd
);
1323 * FIFO's, sockets and device files are special: they don't
1324 * actually live on the filesystem itself, and as such you
1325 * can write to them even if the filesystem is read-only.
1327 if (S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
1329 } else if (S_ISBLK(inode
->i_mode
) || S_ISCHR(inode
->i_mode
)) {
1330 if (nd
->mnt
->mnt_flags
& MNT_NODEV
)
1334 } else if (IS_RDONLY(inode
) && (flag
& FMODE_WRITE
))
1337 * An append-only file must be opened in append mode for writing.
1339 if (IS_APPEND(inode
)) {
1340 if ((flag
& FMODE_WRITE
) && !(flag
& O_APPEND
))
1346 /* O_NOATIME can only be set by the owner or superuser */
1347 if (flag
& O_NOATIME
)
1348 if (current
->fsuid
!= inode
->i_uid
&& !capable(CAP_FOWNER
))
1352 * Ensure there are no outstanding leases on the file.
1354 error
= break_lease(inode
, flag
);
1358 if (flag
& O_TRUNC
) {
1359 error
= get_write_access(inode
);
1364 * Refuse to truncate files with mandatory locks held on them.
1366 error
= locks_verify_locked(inode
);
1370 error
= do_truncate(dentry
, 0);
1372 put_write_access(inode
);
1376 if (flag
& FMODE_WRITE
)
1385 * namei for open - this is in fact almost the whole open-routine.
1387 * Note that the low bits of "flag" aren't the same as in the open
1388 * system call - they are 00 - no permissions needed
1389 * 01 - read permission needed
1390 * 10 - write permission needed
1391 * 11 - read/write permissions needed
1392 * which is a lot more logical, and also allows the "no perm" needed
1393 * for symlinks (where the permissions are checked later).
1396 int open_namei(const char * pathname
, int flag
, int mode
, struct nameidata
*nd
)
1398 int acc_mode
, error
= 0;
1403 acc_mode
= ACC_MODE(flag
);
1405 /* Allow the LSM permission hook to distinguish append
1406 access from general write access. */
1407 if (flag
& O_APPEND
)
1408 acc_mode
|= MAY_APPEND
;
1410 /* Fill in the open() intent data */
1411 nd
->intent
.open
.flags
= flag
;
1412 nd
->intent
.open
.create_mode
= mode
;
1415 * The simplest case - just a plain lookup.
1417 if (!(flag
& O_CREAT
)) {
1418 error
= path_lookup(pathname
, lookup_flags(flag
)|LOOKUP_OPEN
, nd
);
1425 * Create - we need to know the parent.
1427 error
= path_lookup(pathname
, LOOKUP_PARENT
|LOOKUP_OPEN
|LOOKUP_CREATE
, nd
);
1432 * We have the parent and last component. First of all, check
1433 * that we are not asked to creat(2) an obvious directory - that
1437 if (nd
->last_type
!= LAST_NORM
|| nd
->last
.name
[nd
->last
.len
])
1441 nd
->flags
&= ~LOOKUP_PARENT
;
1442 down(&dir
->d_inode
->i_sem
);
1443 path
.dentry
= __lookup_hash(&nd
->last
, nd
->dentry
, nd
);
1447 error
= PTR_ERR(path
.dentry
);
1448 if (IS_ERR(path
.dentry
)) {
1449 up(&dir
->d_inode
->i_sem
);
1453 /* Negative dentry, just create the file */
1454 if (!path
.dentry
->d_inode
) {
1455 if (!IS_POSIXACL(dir
->d_inode
))
1456 mode
&= ~current
->fs
->umask
;
1457 error
= vfs_create(dir
->d_inode
, path
.dentry
, mode
, nd
);
1458 up(&dir
->d_inode
->i_sem
);
1460 nd
->dentry
= path
.dentry
;
1463 /* Don't check for write permission, don't truncate */
1470 * It already exists.
1472 up(&dir
->d_inode
->i_sem
);
1478 if (d_mountpoint(path
.dentry
)) {
1480 if (flag
& O_NOFOLLOW
)
1482 while (__follow_down(&path
.mnt
,&path
.dentry
) && d_mountpoint(path
.dentry
));
1486 if (!path
.dentry
->d_inode
)
1488 if (path
.dentry
->d_inode
->i_op
&& path
.dentry
->d_inode
->i_op
->follow_link
)
1492 nd
->dentry
= path
.dentry
;
1494 if (path
.dentry
->d_inode
&& S_ISDIR(path
.dentry
->d_inode
->i_mode
))
1497 error
= may_open(nd
, acc_mode
, flag
);
1510 if (flag
& O_NOFOLLOW
)
1513 * This is subtle. Instead of calling do_follow_link() we do the
1514 * thing by hands. The reason is that this way we have zero link_count
1515 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1516 * After that we have the parent and last component, i.e.
1517 * we are in the same situation as after the first path_walk().
1518 * Well, almost - if the last component is normal we get its copy
1519 * stored in nd->last.name and we will have to putname() it when we
1520 * are done. Procfs-like symlinks just set LAST_BIND.
1522 nd
->flags
|= LOOKUP_PARENT
;
1523 error
= security_inode_follow_link(path
.dentry
, nd
);
1527 error
= __do_follow_link(path
.dentry
, nd
);
1533 nd
->flags
&= ~LOOKUP_PARENT
;
1534 if (nd
->last_type
== LAST_BIND
) {
1535 path
.dentry
= nd
->dentry
;
1539 if (nd
->last_type
!= LAST_NORM
)
1541 if (nd
->last
.name
[nd
->last
.len
]) {
1542 putname(nd
->last
.name
);
1547 putname(nd
->last
.name
);
1551 down(&dir
->d_inode
->i_sem
);
1552 path
.dentry
= __lookup_hash(&nd
->last
, nd
->dentry
, nd
);
1553 putname(nd
->last
.name
);
1558 * lookup_create - lookup a dentry, creating it if it doesn't exist
1559 * @nd: nameidata info
1560 * @is_dir: directory flag
1562 * Simple function to lookup and return a dentry and create it
1563 * if it doesn't exist. Is SMP-safe.
1565 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
1567 struct dentry
*dentry
;
1569 down(&nd
->dentry
->d_inode
->i_sem
);
1570 dentry
= ERR_PTR(-EEXIST
);
1571 if (nd
->last_type
!= LAST_NORM
)
1573 nd
->flags
&= ~LOOKUP_PARENT
;
1574 dentry
= lookup_hash(&nd
->last
, nd
->dentry
);
1577 if (!is_dir
&& nd
->last
.name
[nd
->last
.len
] && !dentry
->d_inode
)
1582 dentry
= ERR_PTR(-ENOENT
);
1586 EXPORT_SYMBOL_GPL(lookup_create
);
1588 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1590 int error
= may_create(dir
, dentry
, NULL
);
1595 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
1598 if (!dir
->i_op
|| !dir
->i_op
->mknod
)
1601 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
1606 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
1608 inode_dir_notify(dir
, DN_CREATE
);
1609 security_inode_post_mknod(dir
, dentry
, mode
, dev
);
1614 asmlinkage
long sys_mknod(const char __user
* filename
, int mode
, unsigned dev
)
1618 struct dentry
* dentry
;
1619 struct nameidata nd
;
1623 tmp
= getname(filename
);
1625 return PTR_ERR(tmp
);
1627 error
= path_lookup(tmp
, LOOKUP_PARENT
, &nd
);
1630 dentry
= lookup_create(&nd
, 0);
1631 error
= PTR_ERR(dentry
);
1633 if (!IS_POSIXACL(nd
.dentry
->d_inode
))
1634 mode
&= ~current
->fs
->umask
;
1635 if (!IS_ERR(dentry
)) {
1636 switch (mode
& S_IFMT
) {
1637 case 0: case S_IFREG
:
1638 error
= vfs_create(nd
.dentry
->d_inode
,dentry
,mode
,&nd
);
1640 case S_IFCHR
: case S_IFBLK
:
1641 error
= vfs_mknod(nd
.dentry
->d_inode
,dentry
,mode
,
1642 new_decode_dev(dev
));
1644 case S_IFIFO
: case S_IFSOCK
:
1645 error
= vfs_mknod(nd
.dentry
->d_inode
,dentry
,mode
,0);
1655 up(&nd
.dentry
->d_inode
->i_sem
);
1663 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1665 int error
= may_create(dir
, dentry
, NULL
);
1670 if (!dir
->i_op
|| !dir
->i_op
->mkdir
)
1673 mode
&= (S_IRWXUGO
|S_ISVTX
);
1674 error
= security_inode_mkdir(dir
, dentry
, mode
);
1679 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
1681 inode_dir_notify(dir
, DN_CREATE
);
1682 security_inode_post_mkdir(dir
,dentry
, mode
);
1687 asmlinkage
long sys_mkdir(const char __user
* pathname
, int mode
)
1692 tmp
= getname(pathname
);
1693 error
= PTR_ERR(tmp
);
1695 struct dentry
*dentry
;
1696 struct nameidata nd
;
1698 error
= path_lookup(tmp
, LOOKUP_PARENT
, &nd
);
1701 dentry
= lookup_create(&nd
, 1);
1702 error
= PTR_ERR(dentry
);
1703 if (!IS_ERR(dentry
)) {
1704 if (!IS_POSIXACL(nd
.dentry
->d_inode
))
1705 mode
&= ~current
->fs
->umask
;
1706 error
= vfs_mkdir(nd
.dentry
->d_inode
, dentry
, mode
);
1709 up(&nd
.dentry
->d_inode
->i_sem
);
1719 * We try to drop the dentry early: we should have
1720 * a usage count of 2 if we're the only user of this
1721 * dentry, and if that is true (possibly after pruning
1722 * the dcache), then we drop the dentry now.
1724 * A low-level filesystem can, if it choses, legally
1727 * if (!d_unhashed(dentry))
1730 * if it cannot handle the case of removing a directory
1731 * that is still in use by something else..
1733 void dentry_unhash(struct dentry
*dentry
)
1736 if (atomic_read(&dentry
->d_count
))
1737 shrink_dcache_parent(dentry
);
1738 spin_lock(&dcache_lock
);
1739 spin_lock(&dentry
->d_lock
);
1740 if (atomic_read(&dentry
->d_count
) == 2)
1742 spin_unlock(&dentry
->d_lock
);
1743 spin_unlock(&dcache_lock
);
1746 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1748 int error
= may_delete(dir
, dentry
, 1);
1753 if (!dir
->i_op
|| !dir
->i_op
->rmdir
)
1758 down(&dentry
->d_inode
->i_sem
);
1759 dentry_unhash(dentry
);
1760 if (d_mountpoint(dentry
))
1763 error
= security_inode_rmdir(dir
, dentry
);
1765 error
= dir
->i_op
->rmdir(dir
, dentry
);
1767 dentry
->d_inode
->i_flags
|= S_DEAD
;
1770 up(&dentry
->d_inode
->i_sem
);
1772 inode_dir_notify(dir
, DN_DELETE
);
1780 asmlinkage
long sys_rmdir(const char __user
* pathname
)
1784 struct dentry
*dentry
;
1785 struct nameidata nd
;
1787 name
= getname(pathname
);
1789 return PTR_ERR(name
);
1791 error
= path_lookup(name
, LOOKUP_PARENT
, &nd
);
1795 switch(nd
.last_type
) {
1806 down(&nd
.dentry
->d_inode
->i_sem
);
1807 dentry
= lookup_hash(&nd
.last
, nd
.dentry
);
1808 error
= PTR_ERR(dentry
);
1809 if (!IS_ERR(dentry
)) {
1810 error
= vfs_rmdir(nd
.dentry
->d_inode
, dentry
);
1813 up(&nd
.dentry
->d_inode
->i_sem
);
1821 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1823 int error
= may_delete(dir
, dentry
, 0);
1828 if (!dir
->i_op
|| !dir
->i_op
->unlink
)
1833 down(&dentry
->d_inode
->i_sem
);
1834 if (d_mountpoint(dentry
))
1837 error
= security_inode_unlink(dir
, dentry
);
1839 error
= dir
->i_op
->unlink(dir
, dentry
);
1841 up(&dentry
->d_inode
->i_sem
);
1843 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1844 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
1846 inode_dir_notify(dir
, DN_DELETE
);
1852 * Make sure that the actual truncation of the file will occur outside its
1853 * directory's i_sem. Truncate can take a long time if there is a lot of
1854 * writeout happening, and we don't want to prevent access to the directory
1855 * while waiting on the I/O.
1857 asmlinkage
long sys_unlink(const char __user
* pathname
)
1861 struct dentry
*dentry
;
1862 struct nameidata nd
;
1863 struct inode
*inode
= NULL
;
1865 name
= getname(pathname
);
1867 return PTR_ERR(name
);
1869 error
= path_lookup(name
, LOOKUP_PARENT
, &nd
);
1873 if (nd
.last_type
!= LAST_NORM
)
1875 down(&nd
.dentry
->d_inode
->i_sem
);
1876 dentry
= lookup_hash(&nd
.last
, nd
.dentry
);
1877 error
= PTR_ERR(dentry
);
1878 if (!IS_ERR(dentry
)) {
1879 /* Why not before? Because we want correct error value */
1880 if (nd
.last
.name
[nd
.last
.len
])
1882 inode
= dentry
->d_inode
;
1884 atomic_inc(&inode
->i_count
);
1885 error
= vfs_unlink(nd
.dentry
->d_inode
, dentry
);
1889 up(&nd
.dentry
->d_inode
->i_sem
);
1891 iput(inode
); /* truncate the inode here */
1899 error
= !dentry
->d_inode
? -ENOENT
:
1900 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
1904 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
, int mode
)
1906 int error
= may_create(dir
, dentry
, NULL
);
1911 if (!dir
->i_op
|| !dir
->i_op
->symlink
)
1914 error
= security_inode_symlink(dir
, dentry
, oldname
);
1919 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
1921 inode_dir_notify(dir
, DN_CREATE
);
1922 security_inode_post_symlink(dir
, dentry
, oldname
);
1927 asmlinkage
long sys_symlink(const char __user
* oldname
, const char __user
* newname
)
1933 from
= getname(oldname
);
1935 return PTR_ERR(from
);
1936 to
= getname(newname
);
1937 error
= PTR_ERR(to
);
1939 struct dentry
*dentry
;
1940 struct nameidata nd
;
1942 error
= path_lookup(to
, LOOKUP_PARENT
, &nd
);
1945 dentry
= lookup_create(&nd
, 0);
1946 error
= PTR_ERR(dentry
);
1947 if (!IS_ERR(dentry
)) {
1948 error
= vfs_symlink(nd
.dentry
->d_inode
, dentry
, from
, S_IALLUGO
);
1951 up(&nd
.dentry
->d_inode
->i_sem
);
1960 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
1962 struct inode
*inode
= old_dentry
->d_inode
;
1968 error
= may_create(dir
, new_dentry
, NULL
);
1972 if (dir
->i_sb
!= inode
->i_sb
)
1976 * A link to an append-only or immutable file cannot be created.
1978 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
1980 if (!dir
->i_op
|| !dir
->i_op
->link
)
1982 if (S_ISDIR(old_dentry
->d_inode
->i_mode
))
1985 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
1989 down(&old_dentry
->d_inode
->i_sem
);
1991 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
1992 up(&old_dentry
->d_inode
->i_sem
);
1994 inode_dir_notify(dir
, DN_CREATE
);
1995 security_inode_post_link(old_dentry
, dir
, new_dentry
);
2001 * Hardlinks are often used in delicate situations. We avoid
2002 * security-related surprises by not following symlinks on the
2005 * We don't follow them on the oldname either to be compatible
2006 * with linux 2.0, and to avoid hard-linking to directories
2007 * and other special files. --ADM
2009 asmlinkage
long sys_link(const char __user
* oldname
, const char __user
* newname
)
2011 struct dentry
*new_dentry
;
2012 struct nameidata nd
, old_nd
;
2016 to
= getname(newname
);
2020 error
= __user_walk(oldname
, 0, &old_nd
);
2023 error
= path_lookup(to
, LOOKUP_PARENT
, &nd
);
2027 if (old_nd
.mnt
!= nd
.mnt
)
2029 new_dentry
= lookup_create(&nd
, 0);
2030 error
= PTR_ERR(new_dentry
);
2031 if (!IS_ERR(new_dentry
)) {
2032 error
= vfs_link(old_nd
.dentry
, nd
.dentry
->d_inode
, new_dentry
);
2035 up(&nd
.dentry
->d_inode
->i_sem
);
2039 path_release(&old_nd
);
2047 * The worst of all namespace operations - renaming directory. "Perverted"
2048 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2050 * a) we can get into loop creation. Check is done in is_subdir().
2051 * b) race potential - two innocent renames can create a loop together.
2052 * That's where 4.4 screws up. Current fix: serialization on
2053 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2055 * c) we have to lock _three_ objects - parents and victim (if it exists).
2056 * And that - after we got ->i_sem on parents (until then we don't know
2057 * whether the target exists). Solution: try to be smart with locking
2058 * order for inodes. We rely on the fact that tree topology may change
2059 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2060 * move will be locked. Thus we can rank directories by the tree
2061 * (ancestors first) and rank all non-directories after them.
2062 * That works since everybody except rename does "lock parent, lookup,
2063 * lock child" and rename is under ->s_vfs_rename_sem.
2064 * HOWEVER, it relies on the assumption that any object with ->lookup()
2065 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2066 * we'd better make sure that there's no link(2) for them.
2067 * d) some filesystems don't support opened-but-unlinked directories,
2068 * either because of layout or because they are not ready to deal with
2069 * all cases correctly. The latter will be fixed (taking this sort of
2070 * stuff into VFS), but the former is not going away. Solution: the same
2071 * trick as in rmdir().
2072 * e) conversion from fhandle to dentry may come in the wrong moment - when
2073 * we are removing the target. Solution: we will have to grab ->i_sem
2074 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2075 * ->i_sem on parents, which works but leads to some truely excessive
2078 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2079 struct inode
*new_dir
, struct dentry
*new_dentry
)
2082 struct inode
*target
;
2085 * If we are going to change the parent - check write permissions,
2086 * we'll need to flip '..'.
2088 if (new_dir
!= old_dir
) {
2089 error
= permission(old_dentry
->d_inode
, MAY_WRITE
, NULL
);
2094 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2098 target
= new_dentry
->d_inode
;
2100 down(&target
->i_sem
);
2101 dentry_unhash(new_dentry
);
2103 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2106 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2109 target
->i_flags
|= S_DEAD
;
2111 if (d_unhashed(new_dentry
))
2112 d_rehash(new_dentry
);
2116 d_move(old_dentry
,new_dentry
);
2117 security_inode_post_rename(old_dir
, old_dentry
,
2118 new_dir
, new_dentry
);
2123 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
2124 struct inode
*new_dir
, struct dentry
*new_dentry
)
2126 struct inode
*target
;
2129 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2134 target
= new_dentry
->d_inode
;
2136 down(&target
->i_sem
);
2137 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2140 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2142 /* The following d_move() should become unconditional */
2143 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_ODD_RENAME
))
2144 d_move(old_dentry
, new_dentry
);
2145 security_inode_post_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2153 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2154 struct inode
*new_dir
, struct dentry
*new_dentry
)
2157 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2159 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
2162 error
= may_delete(old_dir
, old_dentry
, is_dir
);
2166 if (!new_dentry
->d_inode
)
2167 error
= may_create(new_dir
, new_dentry
, NULL
);
2169 error
= may_delete(new_dir
, new_dentry
, is_dir
);
2173 if (!old_dir
->i_op
|| !old_dir
->i_op
->rename
)
2176 DQUOT_INIT(old_dir
);
2177 DQUOT_INIT(new_dir
);
2180 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
2182 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
2184 if (old_dir
== new_dir
)
2185 inode_dir_notify(old_dir
, DN_RENAME
);
2187 inode_dir_notify(old_dir
, DN_DELETE
);
2188 inode_dir_notify(new_dir
, DN_CREATE
);
2194 static inline int do_rename(const char * oldname
, const char * newname
)
2197 struct dentry
* old_dir
, * new_dir
;
2198 struct dentry
* old_dentry
, *new_dentry
;
2199 struct dentry
* trap
;
2200 struct nameidata oldnd
, newnd
;
2202 error
= path_lookup(oldname
, LOOKUP_PARENT
, &oldnd
);
2206 error
= path_lookup(newname
, LOOKUP_PARENT
, &newnd
);
2211 if (oldnd
.mnt
!= newnd
.mnt
)
2214 old_dir
= oldnd
.dentry
;
2216 if (oldnd
.last_type
!= LAST_NORM
)
2219 new_dir
= newnd
.dentry
;
2220 if (newnd
.last_type
!= LAST_NORM
)
2223 trap
= lock_rename(new_dir
, old_dir
);
2225 old_dentry
= lookup_hash(&oldnd
.last
, old_dir
);
2226 error
= PTR_ERR(old_dentry
);
2227 if (IS_ERR(old_dentry
))
2229 /* source must exist */
2231 if (!old_dentry
->d_inode
)
2233 /* unless the source is a directory trailing slashes give -ENOTDIR */
2234 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
2236 if (oldnd
.last
.name
[oldnd
.last
.len
])
2238 if (newnd
.last
.name
[newnd
.last
.len
])
2241 /* source should not be ancestor of target */
2243 if (old_dentry
== trap
)
2245 new_dentry
= lookup_hash(&newnd
.last
, new_dir
);
2246 error
= PTR_ERR(new_dentry
);
2247 if (IS_ERR(new_dentry
))
2249 /* target should not be an ancestor of source */
2251 if (new_dentry
== trap
)
2254 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
2255 new_dir
->d_inode
, new_dentry
);
2261 unlock_rename(new_dir
, old_dir
);
2263 path_release(&newnd
);
2265 path_release(&oldnd
);
2270 asmlinkage
long sys_rename(const char __user
* oldname
, const char __user
* newname
)
2276 from
= getname(oldname
);
2278 return PTR_ERR(from
);
2279 to
= getname(newname
);
2280 error
= PTR_ERR(to
);
2282 error
= do_rename(from
,to
);
2289 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
2293 len
= PTR_ERR(link
);
2298 if (len
> (unsigned) buflen
)
2300 if (copy_to_user(buffer
, link
, len
))
2307 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2308 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2309 * using) it for any given inode is up to filesystem.
2311 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2313 struct nameidata nd
;
2316 res
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
2318 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
2319 if (dentry
->d_inode
->i_op
->put_link
)
2320 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
);
2325 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
2327 return __vfs_follow_link(nd
, link
);
2330 /* get the link contents into pagecache */
2331 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
2334 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
2335 page
= read_cache_page(mapping
, 0, (filler_t
*)mapping
->a_ops
->readpage
,
2339 wait_on_page_locked(page
);
2340 if (!PageUptodate(page
))
2346 page_cache_release(page
);
2347 return ERR_PTR(-EIO
);
2353 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2355 struct page
*page
= NULL
;
2356 char *s
= page_getlink(dentry
, &page
);
2357 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
2360 page_cache_release(page
);
2365 int page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
2368 nd_set_link(nd
, page_getlink(dentry
, &page
));
2372 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
)
2374 if (!IS_ERR(nd_get_link(nd
))) {
2376 page
= find_get_page(dentry
->d_inode
->i_mapping
, 0);
2380 page_cache_release(page
);
2381 page_cache_release(page
);
2385 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
2387 struct address_space
*mapping
= inode
->i_mapping
;
2388 struct page
*page
= grab_cache_page(mapping
, 0);
2394 err
= mapping
->a_ops
->prepare_write(NULL
, page
, 0, len
-1);
2397 kaddr
= kmap_atomic(page
, KM_USER0
);
2398 memcpy(kaddr
, symname
, len
-1);
2399 kunmap_atomic(kaddr
, KM_USER0
);
2400 mapping
->a_ops
->commit_write(NULL
, page
, 0, len
-1);
2402 * Notice that we are _not_ going to block here - end of page is
2403 * unmapped, so this will only try to map the rest of page, see
2404 * that it is unmapped (typically even will not look into inode -
2405 * ->i_size will be enough for everything) and zero it out.
2406 * OTOH it's obviously correct and should make the page up-to-date.
2408 if (!PageUptodate(page
)) {
2409 err
= mapping
->a_ops
->readpage(NULL
, page
);
2410 wait_on_page_locked(page
);
2414 page_cache_release(page
);
2417 mark_inode_dirty(inode
);
2421 page_cache_release(page
);
2426 struct inode_operations page_symlink_inode_operations
= {
2427 .readlink
= generic_readlink
,
2428 .follow_link
= page_follow_link_light
,
2429 .put_link
= page_put_link
,
2432 EXPORT_SYMBOL(__user_walk
);
2433 EXPORT_SYMBOL(follow_down
);
2434 EXPORT_SYMBOL(follow_up
);
2435 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
2436 EXPORT_SYMBOL(getname
);
2437 EXPORT_SYMBOL(lock_rename
);
2438 EXPORT_SYMBOL(lookup_hash
);
2439 EXPORT_SYMBOL(lookup_one_len
);
2440 EXPORT_SYMBOL(page_follow_link_light
);
2441 EXPORT_SYMBOL(page_put_link
);
2442 EXPORT_SYMBOL(page_readlink
);
2443 EXPORT_SYMBOL(page_symlink
);
2444 EXPORT_SYMBOL(page_symlink_inode_operations
);
2445 EXPORT_SYMBOL(path_lookup
);
2446 EXPORT_SYMBOL(path_release
);
2447 EXPORT_SYMBOL(path_walk
);
2448 EXPORT_SYMBOL(permission
);
2449 EXPORT_SYMBOL(unlock_rename
);
2450 EXPORT_SYMBOL(vfs_create
);
2451 EXPORT_SYMBOL(vfs_follow_link
);
2452 EXPORT_SYMBOL(vfs_link
);
2453 EXPORT_SYMBOL(vfs_mkdir
);
2454 EXPORT_SYMBOL(vfs_mknod
);
2455 EXPORT_SYMBOL(generic_permission
);
2456 EXPORT_SYMBOL(vfs_readlink
);
2457 EXPORT_SYMBOL(vfs_rename
);
2458 EXPORT_SYMBOL(vfs_rmdir
);
2459 EXPORT_SYMBOL(vfs_symlink
);
2460 EXPORT_SYMBOL(vfs_unlink
);
2461 EXPORT_SYMBOL(dentry_unhash
);
2462 EXPORT_SYMBOL(generic_readlink
);