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 path
*path
, struct nameidata
*nd
)
504 struct dentry
*dentry
= path
->dentry
;
506 touch_atime(nd
->mnt
, dentry
);
507 nd_set_link(nd
, NULL
);
509 if (path
->mnt
== nd
->mnt
)
511 error
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
513 char *s
= nd_get_link(nd
);
515 error
= __vfs_follow_link(nd
, s
);
516 if (dentry
->d_inode
->i_op
->put_link
)
517 dentry
->d_inode
->i_op
->put_link(dentry
, nd
);
526 * This limits recursive symlink follows to 8, while
527 * limiting consecutive symlinks to 40.
529 * Without that kind of total limit, nasty chains of consecutive
530 * symlinks can cause almost arbitrarily long lookups.
532 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
535 if (current
->link_count
>= MAX_NESTED_LINKS
)
537 if (current
->total_link_count
>= 40)
539 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
541 err
= security_inode_follow_link(path
->dentry
, nd
);
544 current
->link_count
++;
545 current
->total_link_count
++;
547 err
= __do_follow_link(path
, nd
);
548 current
->link_count
--;
553 if (path
->mnt
!= nd
->mnt
)
559 int follow_up(struct vfsmount
**mnt
, struct dentry
**dentry
)
561 struct vfsmount
*parent
;
562 struct dentry
*mountpoint
;
563 spin_lock(&vfsmount_lock
);
564 parent
=(*mnt
)->mnt_parent
;
565 if (parent
== *mnt
) {
566 spin_unlock(&vfsmount_lock
);
570 mountpoint
=dget((*mnt
)->mnt_mountpoint
);
571 spin_unlock(&vfsmount_lock
);
573 *dentry
= mountpoint
;
579 /* no need for dcache_lock, as serialization is taken care in
582 static int __follow_mount(struct path
*path
)
585 while (d_mountpoint(path
->dentry
)) {
586 struct vfsmount
*mounted
= lookup_mnt(path
->mnt
, path
->dentry
);
593 path
->dentry
= dget(mounted
->mnt_root
);
599 static int follow_mount(struct vfsmount
**mnt
, struct dentry
**dentry
)
602 while (d_mountpoint(*dentry
)) {
603 struct vfsmount
*mounted
= lookup_mnt(*mnt
, *dentry
);
609 *dentry
= dget(mounted
->mnt_root
);
615 /* no need for dcache_lock, as serialization is taken care in
618 int follow_down(struct vfsmount
**mnt
, struct dentry
**dentry
)
620 struct vfsmount
*mounted
;
622 mounted
= lookup_mnt(*mnt
, *dentry
);
627 *dentry
= dget(mounted
->mnt_root
);
633 static inline void follow_dotdot(struct vfsmount
**mnt
, struct dentry
**dentry
)
636 struct vfsmount
*parent
;
637 struct dentry
*old
= *dentry
;
639 read_lock(¤t
->fs
->lock
);
640 if (*dentry
== current
->fs
->root
&&
641 *mnt
== current
->fs
->rootmnt
) {
642 read_unlock(¤t
->fs
->lock
);
645 read_unlock(¤t
->fs
->lock
);
646 spin_lock(&dcache_lock
);
647 if (*dentry
!= (*mnt
)->mnt_root
) {
648 *dentry
= dget((*dentry
)->d_parent
);
649 spin_unlock(&dcache_lock
);
653 spin_unlock(&dcache_lock
);
654 spin_lock(&vfsmount_lock
);
655 parent
= (*mnt
)->mnt_parent
;
656 if (parent
== *mnt
) {
657 spin_unlock(&vfsmount_lock
);
661 *dentry
= dget((*mnt
)->mnt_mountpoint
);
662 spin_unlock(&vfsmount_lock
);
667 follow_mount(mnt
, dentry
);
671 * It's more convoluted than I'd like it to be, but... it's still fairly
672 * small and for now I'd prefer to have fast path as straight as possible.
673 * It _is_ time-critical.
675 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
678 struct vfsmount
*mnt
= nd
->mnt
;
679 struct dentry
*dentry
= __d_lookup(nd
->dentry
, name
);
683 if (dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
684 goto need_revalidate
;
687 path
->dentry
= dentry
;
691 dentry
= real_lookup(nd
->dentry
, name
, nd
);
697 if (dentry
->d_op
->d_revalidate(dentry
, nd
))
699 if (d_invalidate(dentry
))
705 return PTR_ERR(dentry
);
710 * This is the basic name resolution function, turning a pathname into
711 * the final dentry. We expect 'base' to be positive and a directory.
713 * Returns 0 and nd will have valid dentry and mnt on success.
714 * Returns error and drops reference to input namei data on failure.
716 static fastcall
int __link_path_walk(const char * name
, struct nameidata
*nd
)
721 unsigned int lookup_flags
= nd
->flags
;
728 inode
= nd
->dentry
->d_inode
;
730 lookup_flags
= LOOKUP_FOLLOW
;
732 /* At this point we know we have a real path component. */
738 err
= exec_permission_lite(inode
, nd
);
739 if (err
== -EAGAIN
) {
740 err
= permission(inode
, MAY_EXEC
, nd
);
746 c
= *(const unsigned char *)name
;
748 hash
= init_name_hash();
751 hash
= partial_name_hash(c
, hash
);
752 c
= *(const unsigned char *)name
;
753 } while (c
&& (c
!= '/'));
754 this.len
= name
- (const char *) this.name
;
755 this.hash
= end_name_hash(hash
);
757 /* remove trailing slashes? */
760 while (*++name
== '/');
762 goto last_with_slashes
;
765 * "." and ".." are special - ".." especially so because it has
766 * to be able to know about the current root directory and
767 * parent relationships.
769 if (this.name
[0] == '.') switch (this.len
) {
773 if (this.name
[1] != '.')
775 follow_dotdot(&nd
->mnt
, &nd
->dentry
);
776 inode
= nd
->dentry
->d_inode
;
782 * See if the low-level filesystem might want
783 * to use its own hash..
785 if (nd
->dentry
->d_op
&& nd
->dentry
->d_op
->d_hash
) {
786 err
= nd
->dentry
->d_op
->d_hash(nd
->dentry
, &this);
790 nd
->flags
|= LOOKUP_CONTINUE
;
791 /* This does the actual lookups.. */
792 err
= do_lookup(nd
, &this, &next
);
795 /* Check mountpoints.. */
796 __follow_mount(&next
);
799 inode
= next
.dentry
->d_inode
;
806 if (inode
->i_op
->follow_link
) {
807 err
= do_follow_link(&next
, nd
);
811 inode
= nd
->dentry
->d_inode
;
819 if (nd
->mnt
!= next
.mnt
)
822 nd
->dentry
= next
.dentry
;
825 if (!inode
->i_op
->lookup
)
828 /* here ends the main loop */
831 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
833 nd
->flags
&= ~LOOKUP_CONTINUE
;
834 if (lookup_flags
& LOOKUP_PARENT
)
836 if (this.name
[0] == '.') switch (this.len
) {
840 if (this.name
[1] != '.')
842 follow_dotdot(&nd
->mnt
, &nd
->dentry
);
843 inode
= nd
->dentry
->d_inode
;
848 if (nd
->dentry
->d_op
&& nd
->dentry
->d_op
->d_hash
) {
849 err
= nd
->dentry
->d_op
->d_hash(nd
->dentry
, &this);
853 err
= do_lookup(nd
, &this, &next
);
856 __follow_mount(&next
);
857 inode
= next
.dentry
->d_inode
;
858 if ((lookup_flags
& LOOKUP_FOLLOW
)
859 && inode
&& inode
->i_op
&& inode
->i_op
->follow_link
) {
860 err
= do_follow_link(&next
, nd
);
863 inode
= nd
->dentry
->d_inode
;
866 if (nd
->mnt
!= next
.mnt
)
869 nd
->dentry
= next
.dentry
;
874 if (lookup_flags
& LOOKUP_DIRECTORY
) {
876 if (!inode
->i_op
|| !inode
->i_op
->lookup
)
882 nd
->last_type
= LAST_NORM
;
883 if (this.name
[0] != '.')
886 nd
->last_type
= LAST_DOT
;
887 else if (this.len
== 2 && this.name
[1] == '.')
888 nd
->last_type
= LAST_DOTDOT
;
893 * We bypassed the ordinary revalidation routines.
894 * We may need to check the cached dentry for staleness.
896 if (nd
->dentry
&& nd
->dentry
->d_sb
&&
897 (nd
->dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)) {
899 /* Note: we do not d_invalidate() */
900 if (!nd
->dentry
->d_op
->d_revalidate(nd
->dentry
, nd
))
907 if (nd
->mnt
!= next
.mnt
)
917 * Wrapper to retry pathname resolution whenever the underlying
918 * file system returns an ESTALE.
920 * Retry the whole path once, forcing real lookup requests
921 * instead of relying on the dcache.
923 int fastcall
link_path_walk(const char *name
, struct nameidata
*nd
)
925 struct nameidata save
= *nd
;
928 /* make sure the stuff we saved doesn't go away */
932 result
= __link_path_walk(name
, nd
);
933 if (result
== -ESTALE
) {
937 nd
->flags
|= LOOKUP_REVAL
;
938 result
= __link_path_walk(name
, nd
);
947 int fastcall
path_walk(const char * name
, struct nameidata
*nd
)
949 current
->total_link_count
= 0;
950 return link_path_walk(name
, nd
);
954 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
955 * everything is done. Returns 0 and drops input nd, if lookup failed;
957 static int __emul_lookup_dentry(const char *name
, struct nameidata
*nd
)
959 if (path_walk(name
, nd
))
960 return 0; /* something went wrong... */
962 if (!nd
->dentry
->d_inode
|| S_ISDIR(nd
->dentry
->d_inode
->i_mode
)) {
963 struct dentry
*old_dentry
= nd
->dentry
;
964 struct vfsmount
*old_mnt
= nd
->mnt
;
965 struct qstr last
= nd
->last
;
966 int last_type
= nd
->last_type
;
968 * NAME was not found in alternate root or it's a directory. Try to find
969 * it in the normal root:
971 nd
->last_type
= LAST_ROOT
;
972 read_lock(¤t
->fs
->lock
);
973 nd
->mnt
= mntget(current
->fs
->rootmnt
);
974 nd
->dentry
= dget(current
->fs
->root
);
975 read_unlock(¤t
->fs
->lock
);
976 if (path_walk(name
, nd
) == 0) {
977 if (nd
->dentry
->d_inode
) {
984 nd
->dentry
= old_dentry
;
987 nd
->last_type
= last_type
;
992 void set_fs_altroot(void)
994 char *emul
= __emul_prefix();
996 struct vfsmount
*mnt
= NULL
, *oldmnt
;
997 struct dentry
*dentry
= NULL
, *olddentry
;
1002 err
= path_lookup(emul
, LOOKUP_FOLLOW
|LOOKUP_DIRECTORY
|LOOKUP_NOALT
, &nd
);
1008 write_lock(¤t
->fs
->lock
);
1009 oldmnt
= current
->fs
->altrootmnt
;
1010 olddentry
= current
->fs
->altroot
;
1011 current
->fs
->altrootmnt
= mnt
;
1012 current
->fs
->altroot
= dentry
;
1013 write_unlock(¤t
->fs
->lock
);
1020 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1021 int fastcall
path_lookup(const char *name
, unsigned int flags
, struct nameidata
*nd
)
1025 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1029 read_lock(¤t
->fs
->lock
);
1031 if (current
->fs
->altroot
&& !(nd
->flags
& LOOKUP_NOALT
)) {
1032 nd
->mnt
= mntget(current
->fs
->altrootmnt
);
1033 nd
->dentry
= dget(current
->fs
->altroot
);
1034 read_unlock(¤t
->fs
->lock
);
1035 if (__emul_lookup_dentry(name
,nd
))
1036 goto out
; /* found in altroot */
1037 read_lock(¤t
->fs
->lock
);
1039 nd
->mnt
= mntget(current
->fs
->rootmnt
);
1040 nd
->dentry
= dget(current
->fs
->root
);
1042 nd
->mnt
= mntget(current
->fs
->pwdmnt
);
1043 nd
->dentry
= dget(current
->fs
->pwd
);
1045 read_unlock(¤t
->fs
->lock
);
1046 current
->total_link_count
= 0;
1047 retval
= link_path_walk(name
, nd
);
1049 if (unlikely(current
->audit_context
1050 && nd
&& nd
->dentry
&& nd
->dentry
->d_inode
))
1051 audit_inode(name
, nd
->dentry
->d_inode
);
1056 * Restricted form of lookup. Doesn't follow links, single-component only,
1057 * needs parent already locked. Doesn't follow mounts.
1060 static struct dentry
* __lookup_hash(struct qstr
*name
, struct dentry
* base
, struct nameidata
*nd
)
1062 struct dentry
* dentry
;
1063 struct inode
*inode
;
1066 inode
= base
->d_inode
;
1067 err
= permission(inode
, MAY_EXEC
, nd
);
1068 dentry
= ERR_PTR(err
);
1073 * See if the low-level filesystem might want
1074 * to use its own hash..
1076 if (base
->d_op
&& base
->d_op
->d_hash
) {
1077 err
= base
->d_op
->d_hash(base
, name
);
1078 dentry
= ERR_PTR(err
);
1083 dentry
= cached_lookup(base
, name
, nd
);
1085 struct dentry
*new = d_alloc(base
, name
);
1086 dentry
= ERR_PTR(-ENOMEM
);
1089 dentry
= inode
->i_op
->lookup(inode
, new, nd
);
1099 struct dentry
* lookup_hash(struct qstr
*name
, struct dentry
* base
)
1101 return __lookup_hash(name
, base
, NULL
);
1105 struct dentry
* lookup_one_len(const char * name
, struct dentry
* base
, int len
)
1116 hash
= init_name_hash();
1118 c
= *(const unsigned char *)name
++;
1119 if (c
== '/' || c
== '\0')
1121 hash
= partial_name_hash(c
, hash
);
1123 this.hash
= end_name_hash(hash
);
1125 return lookup_hash(&this, base
);
1127 return ERR_PTR(-EACCES
);
1133 * is used by most simple commands to get the inode of a specified name.
1134 * Open, link etc use their own routines, but this is enough for things
1137 * namei exists in two versions: namei/lnamei. The only difference is
1138 * that namei follows links, while lnamei does not.
1141 int fastcall
__user_walk(const char __user
*name
, unsigned flags
, struct nameidata
*nd
)
1143 char *tmp
= getname(name
);
1144 int err
= PTR_ERR(tmp
);
1147 err
= path_lookup(tmp
, flags
, nd
);
1154 * It's inline, so penalty for filesystems that don't use sticky bit is
1157 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1159 if (!(dir
->i_mode
& S_ISVTX
))
1161 if (inode
->i_uid
== current
->fsuid
)
1163 if (dir
->i_uid
== current
->fsuid
)
1165 return !capable(CAP_FOWNER
);
1169 * Check whether we can remove a link victim from directory dir, check
1170 * whether the type of victim is right.
1171 * 1. We can't do it if dir is read-only (done in permission())
1172 * 2. We should have write and exec permissions on dir
1173 * 3. We can't remove anything from append-only dir
1174 * 4. We can't do anything with immutable dir (done in permission())
1175 * 5. If the sticky bit on dir is set we should either
1176 * a. be owner of dir, or
1177 * b. be owner of victim, or
1178 * c. have CAP_FOWNER capability
1179 * 6. If the victim is append-only or immutable we can't do antyhing with
1180 * links pointing to it.
1181 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1182 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1183 * 9. We can't remove a root or mountpoint.
1184 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1185 * nfs_async_unlink().
1187 static inline int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1191 if (!victim
->d_inode
)
1194 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1196 error
= permission(dir
,MAY_WRITE
| MAY_EXEC
, NULL
);
1201 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1202 IS_IMMUTABLE(victim
->d_inode
))
1205 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1207 if (IS_ROOT(victim
))
1209 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1211 if (IS_DEADDIR(dir
))
1213 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1218 /* Check whether we can create an object with dentry child in directory
1220 * 1. We can't do it if child already exists (open has special treatment for
1221 * this case, but since we are inlined it's OK)
1222 * 2. We can't do it if dir is read-only (done in permission())
1223 * 3. We should have write and exec permissions on dir
1224 * 4. We can't do it if dir is immutable (done in permission())
1226 static inline int may_create(struct inode
*dir
, struct dentry
*child
,
1227 struct nameidata
*nd
)
1231 if (IS_DEADDIR(dir
))
1233 return permission(dir
,MAY_WRITE
| MAY_EXEC
, nd
);
1237 * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1240 * O_DIRECTORY translates into forcing a directory lookup.
1242 static inline int lookup_flags(unsigned int f
)
1244 unsigned long retval
= LOOKUP_FOLLOW
;
1247 retval
&= ~LOOKUP_FOLLOW
;
1249 if ((f
& (O_CREAT
|O_EXCL
)) == (O_CREAT
|O_EXCL
))
1250 retval
&= ~LOOKUP_FOLLOW
;
1252 if (f
& O_DIRECTORY
)
1253 retval
|= LOOKUP_DIRECTORY
;
1259 * p1 and p2 should be directories on the same fs.
1261 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1266 down(&p1
->d_inode
->i_sem
);
1270 down(&p1
->d_inode
->i_sb
->s_vfs_rename_sem
);
1272 for (p
= p1
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1273 if (p
->d_parent
== p2
) {
1274 down(&p2
->d_inode
->i_sem
);
1275 down(&p1
->d_inode
->i_sem
);
1280 for (p
= p2
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1281 if (p
->d_parent
== p1
) {
1282 down(&p1
->d_inode
->i_sem
);
1283 down(&p2
->d_inode
->i_sem
);
1288 down(&p1
->d_inode
->i_sem
);
1289 down(&p2
->d_inode
->i_sem
);
1293 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1295 up(&p1
->d_inode
->i_sem
);
1297 up(&p2
->d_inode
->i_sem
);
1298 up(&p1
->d_inode
->i_sb
->s_vfs_rename_sem
);
1302 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1303 struct nameidata
*nd
)
1305 int error
= may_create(dir
, dentry
, nd
);
1310 if (!dir
->i_op
|| !dir
->i_op
->create
)
1311 return -EACCES
; /* shouldn't it be ENOSYS? */
1314 error
= security_inode_create(dir
, dentry
, mode
);
1318 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1320 inode_dir_notify(dir
, DN_CREATE
);
1321 security_inode_post_create(dir
, dentry
, mode
);
1326 int may_open(struct nameidata
*nd
, int acc_mode
, int flag
)
1328 struct dentry
*dentry
= nd
->dentry
;
1329 struct inode
*inode
= dentry
->d_inode
;
1335 if (S_ISLNK(inode
->i_mode
))
1338 if (S_ISDIR(inode
->i_mode
) && (flag
& FMODE_WRITE
))
1341 error
= permission(inode
, acc_mode
, nd
);
1346 * FIFO's, sockets and device files are special: they don't
1347 * actually live on the filesystem itself, and as such you
1348 * can write to them even if the filesystem is read-only.
1350 if (S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
1352 } else if (S_ISBLK(inode
->i_mode
) || S_ISCHR(inode
->i_mode
)) {
1353 if (nd
->mnt
->mnt_flags
& MNT_NODEV
)
1357 } else if (IS_RDONLY(inode
) && (flag
& FMODE_WRITE
))
1360 * An append-only file must be opened in append mode for writing.
1362 if (IS_APPEND(inode
)) {
1363 if ((flag
& FMODE_WRITE
) && !(flag
& O_APPEND
))
1369 /* O_NOATIME can only be set by the owner or superuser */
1370 if (flag
& O_NOATIME
)
1371 if (current
->fsuid
!= inode
->i_uid
&& !capable(CAP_FOWNER
))
1375 * Ensure there are no outstanding leases on the file.
1377 error
= break_lease(inode
, flag
);
1381 if (flag
& O_TRUNC
) {
1382 error
= get_write_access(inode
);
1387 * Refuse to truncate files with mandatory locks held on them.
1389 error
= locks_verify_locked(inode
);
1393 error
= do_truncate(dentry
, 0);
1395 put_write_access(inode
);
1399 if (flag
& FMODE_WRITE
)
1408 * namei for open - this is in fact almost the whole open-routine.
1410 * Note that the low bits of "flag" aren't the same as in the open
1411 * system call - they are 00 - no permissions needed
1412 * 01 - read permission needed
1413 * 10 - write permission needed
1414 * 11 - read/write permissions needed
1415 * which is a lot more logical, and also allows the "no perm" needed
1416 * for symlinks (where the permissions are checked later).
1419 int open_namei(const char * pathname
, int flag
, int mode
, struct nameidata
*nd
)
1421 int acc_mode
, error
= 0;
1426 acc_mode
= ACC_MODE(flag
);
1428 /* Allow the LSM permission hook to distinguish append
1429 access from general write access. */
1430 if (flag
& O_APPEND
)
1431 acc_mode
|= MAY_APPEND
;
1433 /* Fill in the open() intent data */
1434 nd
->intent
.open
.flags
= flag
;
1435 nd
->intent
.open
.create_mode
= mode
;
1438 * The simplest case - just a plain lookup.
1440 if (!(flag
& O_CREAT
)) {
1441 error
= path_lookup(pathname
, lookup_flags(flag
)|LOOKUP_OPEN
, nd
);
1448 * Create - we need to know the parent.
1450 error
= path_lookup(pathname
, LOOKUP_PARENT
|LOOKUP_OPEN
|LOOKUP_CREATE
, nd
);
1455 * We have the parent and last component. First of all, check
1456 * that we are not asked to creat(2) an obvious directory - that
1460 if (nd
->last_type
!= LAST_NORM
|| nd
->last
.name
[nd
->last
.len
])
1464 nd
->flags
&= ~LOOKUP_PARENT
;
1465 down(&dir
->d_inode
->i_sem
);
1466 path
.dentry
= __lookup_hash(&nd
->last
, nd
->dentry
, nd
);
1470 error
= PTR_ERR(path
.dentry
);
1471 if (IS_ERR(path
.dentry
)) {
1472 up(&dir
->d_inode
->i_sem
);
1476 /* Negative dentry, just create the file */
1477 if (!path
.dentry
->d_inode
) {
1478 if (!IS_POSIXACL(dir
->d_inode
))
1479 mode
&= ~current
->fs
->umask
;
1480 error
= vfs_create(dir
->d_inode
, path
.dentry
, mode
, nd
);
1481 up(&dir
->d_inode
->i_sem
);
1483 nd
->dentry
= path
.dentry
;
1486 /* Don't check for write permission, don't truncate */
1493 * It already exists.
1495 up(&dir
->d_inode
->i_sem
);
1501 if (__follow_mount(&path
)) {
1503 if (flag
& O_NOFOLLOW
)
1507 if (!path
.dentry
->d_inode
)
1509 if (path
.dentry
->d_inode
->i_op
&& path
.dentry
->d_inode
->i_op
->follow_link
)
1513 nd
->dentry
= path
.dentry
;
1514 if (nd
->mnt
!= path
.mnt
)
1518 if (path
.dentry
->d_inode
&& S_ISDIR(path
.dentry
->d_inode
->i_mode
))
1521 error
= may_open(nd
, acc_mode
, flag
);
1528 if (nd
->mnt
!= path
.mnt
)
1536 if (flag
& O_NOFOLLOW
)
1539 * This is subtle. Instead of calling do_follow_link() we do the
1540 * thing by hands. The reason is that this way we have zero link_count
1541 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1542 * After that we have the parent and last component, i.e.
1543 * we are in the same situation as after the first path_walk().
1544 * Well, almost - if the last component is normal we get its copy
1545 * stored in nd->last.name and we will have to putname() it when we
1546 * are done. Procfs-like symlinks just set LAST_BIND.
1548 nd
->flags
|= LOOKUP_PARENT
;
1549 error
= security_inode_follow_link(path
.dentry
, nd
);
1552 error
= __do_follow_link(&path
, nd
);
1555 nd
->flags
&= ~LOOKUP_PARENT
;
1556 if (nd
->last_type
== LAST_BIND
)
1559 if (nd
->last_type
!= LAST_NORM
)
1561 if (nd
->last
.name
[nd
->last
.len
]) {
1562 putname(nd
->last
.name
);
1567 putname(nd
->last
.name
);
1571 down(&dir
->d_inode
->i_sem
);
1572 path
.dentry
= __lookup_hash(&nd
->last
, nd
->dentry
, nd
);
1574 putname(nd
->last
.name
);
1579 * lookup_create - lookup a dentry, creating it if it doesn't exist
1580 * @nd: nameidata info
1581 * @is_dir: directory flag
1583 * Simple function to lookup and return a dentry and create it
1584 * if it doesn't exist. Is SMP-safe.
1586 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
1588 struct dentry
*dentry
;
1590 down(&nd
->dentry
->d_inode
->i_sem
);
1591 dentry
= ERR_PTR(-EEXIST
);
1592 if (nd
->last_type
!= LAST_NORM
)
1594 nd
->flags
&= ~LOOKUP_PARENT
;
1595 dentry
= lookup_hash(&nd
->last
, nd
->dentry
);
1598 if (!is_dir
&& nd
->last
.name
[nd
->last
.len
] && !dentry
->d_inode
)
1603 dentry
= ERR_PTR(-ENOENT
);
1607 EXPORT_SYMBOL_GPL(lookup_create
);
1609 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1611 int error
= may_create(dir
, dentry
, NULL
);
1616 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
1619 if (!dir
->i_op
|| !dir
->i_op
->mknod
)
1622 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
1627 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
1629 inode_dir_notify(dir
, DN_CREATE
);
1630 security_inode_post_mknod(dir
, dentry
, mode
, dev
);
1635 asmlinkage
long sys_mknod(const char __user
* filename
, int mode
, unsigned dev
)
1639 struct dentry
* dentry
;
1640 struct nameidata nd
;
1644 tmp
= getname(filename
);
1646 return PTR_ERR(tmp
);
1648 error
= path_lookup(tmp
, LOOKUP_PARENT
, &nd
);
1651 dentry
= lookup_create(&nd
, 0);
1652 error
= PTR_ERR(dentry
);
1654 if (!IS_POSIXACL(nd
.dentry
->d_inode
))
1655 mode
&= ~current
->fs
->umask
;
1656 if (!IS_ERR(dentry
)) {
1657 switch (mode
& S_IFMT
) {
1658 case 0: case S_IFREG
:
1659 error
= vfs_create(nd
.dentry
->d_inode
,dentry
,mode
,&nd
);
1661 case S_IFCHR
: case S_IFBLK
:
1662 error
= vfs_mknod(nd
.dentry
->d_inode
,dentry
,mode
,
1663 new_decode_dev(dev
));
1665 case S_IFIFO
: case S_IFSOCK
:
1666 error
= vfs_mknod(nd
.dentry
->d_inode
,dentry
,mode
,0);
1676 up(&nd
.dentry
->d_inode
->i_sem
);
1684 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1686 int error
= may_create(dir
, dentry
, NULL
);
1691 if (!dir
->i_op
|| !dir
->i_op
->mkdir
)
1694 mode
&= (S_IRWXUGO
|S_ISVTX
);
1695 error
= security_inode_mkdir(dir
, dentry
, mode
);
1700 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
1702 inode_dir_notify(dir
, DN_CREATE
);
1703 security_inode_post_mkdir(dir
,dentry
, mode
);
1708 asmlinkage
long sys_mkdir(const char __user
* pathname
, int mode
)
1713 tmp
= getname(pathname
);
1714 error
= PTR_ERR(tmp
);
1716 struct dentry
*dentry
;
1717 struct nameidata nd
;
1719 error
= path_lookup(tmp
, LOOKUP_PARENT
, &nd
);
1722 dentry
= lookup_create(&nd
, 1);
1723 error
= PTR_ERR(dentry
);
1724 if (!IS_ERR(dentry
)) {
1725 if (!IS_POSIXACL(nd
.dentry
->d_inode
))
1726 mode
&= ~current
->fs
->umask
;
1727 error
= vfs_mkdir(nd
.dentry
->d_inode
, dentry
, mode
);
1730 up(&nd
.dentry
->d_inode
->i_sem
);
1740 * We try to drop the dentry early: we should have
1741 * a usage count of 2 if we're the only user of this
1742 * dentry, and if that is true (possibly after pruning
1743 * the dcache), then we drop the dentry now.
1745 * A low-level filesystem can, if it choses, legally
1748 * if (!d_unhashed(dentry))
1751 * if it cannot handle the case of removing a directory
1752 * that is still in use by something else..
1754 void dentry_unhash(struct dentry
*dentry
)
1757 if (atomic_read(&dentry
->d_count
))
1758 shrink_dcache_parent(dentry
);
1759 spin_lock(&dcache_lock
);
1760 spin_lock(&dentry
->d_lock
);
1761 if (atomic_read(&dentry
->d_count
) == 2)
1763 spin_unlock(&dentry
->d_lock
);
1764 spin_unlock(&dcache_lock
);
1767 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1769 int error
= may_delete(dir
, dentry
, 1);
1774 if (!dir
->i_op
|| !dir
->i_op
->rmdir
)
1779 down(&dentry
->d_inode
->i_sem
);
1780 dentry_unhash(dentry
);
1781 if (d_mountpoint(dentry
))
1784 error
= security_inode_rmdir(dir
, dentry
);
1786 error
= dir
->i_op
->rmdir(dir
, dentry
);
1788 dentry
->d_inode
->i_flags
|= S_DEAD
;
1791 up(&dentry
->d_inode
->i_sem
);
1793 inode_dir_notify(dir
, DN_DELETE
);
1801 asmlinkage
long sys_rmdir(const char __user
* pathname
)
1805 struct dentry
*dentry
;
1806 struct nameidata nd
;
1808 name
= getname(pathname
);
1810 return PTR_ERR(name
);
1812 error
= path_lookup(name
, LOOKUP_PARENT
, &nd
);
1816 switch(nd
.last_type
) {
1827 down(&nd
.dentry
->d_inode
->i_sem
);
1828 dentry
= lookup_hash(&nd
.last
, nd
.dentry
);
1829 error
= PTR_ERR(dentry
);
1830 if (!IS_ERR(dentry
)) {
1831 error
= vfs_rmdir(nd
.dentry
->d_inode
, dentry
);
1834 up(&nd
.dentry
->d_inode
->i_sem
);
1842 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1844 int error
= may_delete(dir
, dentry
, 0);
1849 if (!dir
->i_op
|| !dir
->i_op
->unlink
)
1854 down(&dentry
->d_inode
->i_sem
);
1855 if (d_mountpoint(dentry
))
1858 error
= security_inode_unlink(dir
, dentry
);
1860 error
= dir
->i_op
->unlink(dir
, dentry
);
1862 up(&dentry
->d_inode
->i_sem
);
1864 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1865 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
1867 inode_dir_notify(dir
, DN_DELETE
);
1873 * Make sure that the actual truncation of the file will occur outside its
1874 * directory's i_sem. Truncate can take a long time if there is a lot of
1875 * writeout happening, and we don't want to prevent access to the directory
1876 * while waiting on the I/O.
1878 asmlinkage
long sys_unlink(const char __user
* pathname
)
1882 struct dentry
*dentry
;
1883 struct nameidata nd
;
1884 struct inode
*inode
= NULL
;
1886 name
= getname(pathname
);
1888 return PTR_ERR(name
);
1890 error
= path_lookup(name
, LOOKUP_PARENT
, &nd
);
1894 if (nd
.last_type
!= LAST_NORM
)
1896 down(&nd
.dentry
->d_inode
->i_sem
);
1897 dentry
= lookup_hash(&nd
.last
, nd
.dentry
);
1898 error
= PTR_ERR(dentry
);
1899 if (!IS_ERR(dentry
)) {
1900 /* Why not before? Because we want correct error value */
1901 if (nd
.last
.name
[nd
.last
.len
])
1903 inode
= dentry
->d_inode
;
1905 atomic_inc(&inode
->i_count
);
1906 error
= vfs_unlink(nd
.dentry
->d_inode
, dentry
);
1910 up(&nd
.dentry
->d_inode
->i_sem
);
1912 iput(inode
); /* truncate the inode here */
1920 error
= !dentry
->d_inode
? -ENOENT
:
1921 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
1925 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
, int mode
)
1927 int error
= may_create(dir
, dentry
, NULL
);
1932 if (!dir
->i_op
|| !dir
->i_op
->symlink
)
1935 error
= security_inode_symlink(dir
, dentry
, oldname
);
1940 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
1942 inode_dir_notify(dir
, DN_CREATE
);
1943 security_inode_post_symlink(dir
, dentry
, oldname
);
1948 asmlinkage
long sys_symlink(const char __user
* oldname
, const char __user
* newname
)
1954 from
= getname(oldname
);
1956 return PTR_ERR(from
);
1957 to
= getname(newname
);
1958 error
= PTR_ERR(to
);
1960 struct dentry
*dentry
;
1961 struct nameidata nd
;
1963 error
= path_lookup(to
, LOOKUP_PARENT
, &nd
);
1966 dentry
= lookup_create(&nd
, 0);
1967 error
= PTR_ERR(dentry
);
1968 if (!IS_ERR(dentry
)) {
1969 error
= vfs_symlink(nd
.dentry
->d_inode
, dentry
, from
, S_IALLUGO
);
1972 up(&nd
.dentry
->d_inode
->i_sem
);
1981 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
1983 struct inode
*inode
= old_dentry
->d_inode
;
1989 error
= may_create(dir
, new_dentry
, NULL
);
1993 if (dir
->i_sb
!= inode
->i_sb
)
1997 * A link to an append-only or immutable file cannot be created.
1999 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2001 if (!dir
->i_op
|| !dir
->i_op
->link
)
2003 if (S_ISDIR(old_dentry
->d_inode
->i_mode
))
2006 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2010 down(&old_dentry
->d_inode
->i_sem
);
2012 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2013 up(&old_dentry
->d_inode
->i_sem
);
2015 inode_dir_notify(dir
, DN_CREATE
);
2016 security_inode_post_link(old_dentry
, dir
, new_dentry
);
2022 * Hardlinks are often used in delicate situations. We avoid
2023 * security-related surprises by not following symlinks on the
2026 * We don't follow them on the oldname either to be compatible
2027 * with linux 2.0, and to avoid hard-linking to directories
2028 * and other special files. --ADM
2030 asmlinkage
long sys_link(const char __user
* oldname
, const char __user
* newname
)
2032 struct dentry
*new_dentry
;
2033 struct nameidata nd
, old_nd
;
2037 to
= getname(newname
);
2041 error
= __user_walk(oldname
, 0, &old_nd
);
2044 error
= path_lookup(to
, LOOKUP_PARENT
, &nd
);
2048 if (old_nd
.mnt
!= nd
.mnt
)
2050 new_dentry
= lookup_create(&nd
, 0);
2051 error
= PTR_ERR(new_dentry
);
2052 if (!IS_ERR(new_dentry
)) {
2053 error
= vfs_link(old_nd
.dentry
, nd
.dentry
->d_inode
, new_dentry
);
2056 up(&nd
.dentry
->d_inode
->i_sem
);
2060 path_release(&old_nd
);
2068 * The worst of all namespace operations - renaming directory. "Perverted"
2069 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2071 * a) we can get into loop creation. Check is done in is_subdir().
2072 * b) race potential - two innocent renames can create a loop together.
2073 * That's where 4.4 screws up. Current fix: serialization on
2074 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2076 * c) we have to lock _three_ objects - parents and victim (if it exists).
2077 * And that - after we got ->i_sem on parents (until then we don't know
2078 * whether the target exists). Solution: try to be smart with locking
2079 * order for inodes. We rely on the fact that tree topology may change
2080 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2081 * move will be locked. Thus we can rank directories by the tree
2082 * (ancestors first) and rank all non-directories after them.
2083 * That works since everybody except rename does "lock parent, lookup,
2084 * lock child" and rename is under ->s_vfs_rename_sem.
2085 * HOWEVER, it relies on the assumption that any object with ->lookup()
2086 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2087 * we'd better make sure that there's no link(2) for them.
2088 * d) some filesystems don't support opened-but-unlinked directories,
2089 * either because of layout or because they are not ready to deal with
2090 * all cases correctly. The latter will be fixed (taking this sort of
2091 * stuff into VFS), but the former is not going away. Solution: the same
2092 * trick as in rmdir().
2093 * e) conversion from fhandle to dentry may come in the wrong moment - when
2094 * we are removing the target. Solution: we will have to grab ->i_sem
2095 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2096 * ->i_sem on parents, which works but leads to some truely excessive
2099 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2100 struct inode
*new_dir
, struct dentry
*new_dentry
)
2103 struct inode
*target
;
2106 * If we are going to change the parent - check write permissions,
2107 * we'll need to flip '..'.
2109 if (new_dir
!= old_dir
) {
2110 error
= permission(old_dentry
->d_inode
, MAY_WRITE
, NULL
);
2115 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2119 target
= new_dentry
->d_inode
;
2121 down(&target
->i_sem
);
2122 dentry_unhash(new_dentry
);
2124 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2127 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2130 target
->i_flags
|= S_DEAD
;
2132 if (d_unhashed(new_dentry
))
2133 d_rehash(new_dentry
);
2137 d_move(old_dentry
,new_dentry
);
2138 security_inode_post_rename(old_dir
, old_dentry
,
2139 new_dir
, new_dentry
);
2144 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
2145 struct inode
*new_dir
, struct dentry
*new_dentry
)
2147 struct inode
*target
;
2150 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2155 target
= new_dentry
->d_inode
;
2157 down(&target
->i_sem
);
2158 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2161 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2163 /* The following d_move() should become unconditional */
2164 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_ODD_RENAME
))
2165 d_move(old_dentry
, new_dentry
);
2166 security_inode_post_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2174 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2175 struct inode
*new_dir
, struct dentry
*new_dentry
)
2178 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2180 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
2183 error
= may_delete(old_dir
, old_dentry
, is_dir
);
2187 if (!new_dentry
->d_inode
)
2188 error
= may_create(new_dir
, new_dentry
, NULL
);
2190 error
= may_delete(new_dir
, new_dentry
, is_dir
);
2194 if (!old_dir
->i_op
|| !old_dir
->i_op
->rename
)
2197 DQUOT_INIT(old_dir
);
2198 DQUOT_INIT(new_dir
);
2201 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
2203 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
2205 if (old_dir
== new_dir
)
2206 inode_dir_notify(old_dir
, DN_RENAME
);
2208 inode_dir_notify(old_dir
, DN_DELETE
);
2209 inode_dir_notify(new_dir
, DN_CREATE
);
2215 static inline int do_rename(const char * oldname
, const char * newname
)
2218 struct dentry
* old_dir
, * new_dir
;
2219 struct dentry
* old_dentry
, *new_dentry
;
2220 struct dentry
* trap
;
2221 struct nameidata oldnd
, newnd
;
2223 error
= path_lookup(oldname
, LOOKUP_PARENT
, &oldnd
);
2227 error
= path_lookup(newname
, LOOKUP_PARENT
, &newnd
);
2232 if (oldnd
.mnt
!= newnd
.mnt
)
2235 old_dir
= oldnd
.dentry
;
2237 if (oldnd
.last_type
!= LAST_NORM
)
2240 new_dir
= newnd
.dentry
;
2241 if (newnd
.last_type
!= LAST_NORM
)
2244 trap
= lock_rename(new_dir
, old_dir
);
2246 old_dentry
= lookup_hash(&oldnd
.last
, old_dir
);
2247 error
= PTR_ERR(old_dentry
);
2248 if (IS_ERR(old_dentry
))
2250 /* source must exist */
2252 if (!old_dentry
->d_inode
)
2254 /* unless the source is a directory trailing slashes give -ENOTDIR */
2255 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
2257 if (oldnd
.last
.name
[oldnd
.last
.len
])
2259 if (newnd
.last
.name
[newnd
.last
.len
])
2262 /* source should not be ancestor of target */
2264 if (old_dentry
== trap
)
2266 new_dentry
= lookup_hash(&newnd
.last
, new_dir
);
2267 error
= PTR_ERR(new_dentry
);
2268 if (IS_ERR(new_dentry
))
2270 /* target should not be an ancestor of source */
2272 if (new_dentry
== trap
)
2275 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
2276 new_dir
->d_inode
, new_dentry
);
2282 unlock_rename(new_dir
, old_dir
);
2284 path_release(&newnd
);
2286 path_release(&oldnd
);
2291 asmlinkage
long sys_rename(const char __user
* oldname
, const char __user
* newname
)
2297 from
= getname(oldname
);
2299 return PTR_ERR(from
);
2300 to
= getname(newname
);
2301 error
= PTR_ERR(to
);
2303 error
= do_rename(from
,to
);
2310 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
2314 len
= PTR_ERR(link
);
2319 if (len
> (unsigned) buflen
)
2321 if (copy_to_user(buffer
, link
, len
))
2328 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2329 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2330 * using) it for any given inode is up to filesystem.
2332 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2334 struct nameidata nd
;
2337 res
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
2339 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
2340 if (dentry
->d_inode
->i_op
->put_link
)
2341 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
);
2346 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
2348 return __vfs_follow_link(nd
, link
);
2351 /* get the link contents into pagecache */
2352 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
2355 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
2356 page
= read_cache_page(mapping
, 0, (filler_t
*)mapping
->a_ops
->readpage
,
2360 wait_on_page_locked(page
);
2361 if (!PageUptodate(page
))
2367 page_cache_release(page
);
2368 return ERR_PTR(-EIO
);
2374 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2376 struct page
*page
= NULL
;
2377 char *s
= page_getlink(dentry
, &page
);
2378 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
2381 page_cache_release(page
);
2386 int page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
2389 nd_set_link(nd
, page_getlink(dentry
, &page
));
2393 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
)
2395 if (!IS_ERR(nd_get_link(nd
))) {
2397 page
= find_get_page(dentry
->d_inode
->i_mapping
, 0);
2401 page_cache_release(page
);
2402 page_cache_release(page
);
2406 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
2408 struct address_space
*mapping
= inode
->i_mapping
;
2409 struct page
*page
= grab_cache_page(mapping
, 0);
2415 err
= mapping
->a_ops
->prepare_write(NULL
, page
, 0, len
-1);
2418 kaddr
= kmap_atomic(page
, KM_USER0
);
2419 memcpy(kaddr
, symname
, len
-1);
2420 kunmap_atomic(kaddr
, KM_USER0
);
2421 mapping
->a_ops
->commit_write(NULL
, page
, 0, len
-1);
2423 * Notice that we are _not_ going to block here - end of page is
2424 * unmapped, so this will only try to map the rest of page, see
2425 * that it is unmapped (typically even will not look into inode -
2426 * ->i_size will be enough for everything) and zero it out.
2427 * OTOH it's obviously correct and should make the page up-to-date.
2429 if (!PageUptodate(page
)) {
2430 err
= mapping
->a_ops
->readpage(NULL
, page
);
2431 wait_on_page_locked(page
);
2435 page_cache_release(page
);
2438 mark_inode_dirty(inode
);
2442 page_cache_release(page
);
2447 struct inode_operations page_symlink_inode_operations
= {
2448 .readlink
= generic_readlink
,
2449 .follow_link
= page_follow_link_light
,
2450 .put_link
= page_put_link
,
2453 EXPORT_SYMBOL(__user_walk
);
2454 EXPORT_SYMBOL(follow_down
);
2455 EXPORT_SYMBOL(follow_up
);
2456 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
2457 EXPORT_SYMBOL(getname
);
2458 EXPORT_SYMBOL(lock_rename
);
2459 EXPORT_SYMBOL(lookup_hash
);
2460 EXPORT_SYMBOL(lookup_one_len
);
2461 EXPORT_SYMBOL(page_follow_link_light
);
2462 EXPORT_SYMBOL(page_put_link
);
2463 EXPORT_SYMBOL(page_readlink
);
2464 EXPORT_SYMBOL(page_symlink
);
2465 EXPORT_SYMBOL(page_symlink_inode_operations
);
2466 EXPORT_SYMBOL(path_lookup
);
2467 EXPORT_SYMBOL(path_release
);
2468 EXPORT_SYMBOL(path_walk
);
2469 EXPORT_SYMBOL(permission
);
2470 EXPORT_SYMBOL(unlock_rename
);
2471 EXPORT_SYMBOL(vfs_create
);
2472 EXPORT_SYMBOL(vfs_follow_link
);
2473 EXPORT_SYMBOL(vfs_link
);
2474 EXPORT_SYMBOL(vfs_mkdir
);
2475 EXPORT_SYMBOL(vfs_mknod
);
2476 EXPORT_SYMBOL(generic_permission
);
2477 EXPORT_SYMBOL(vfs_readlink
);
2478 EXPORT_SYMBOL(vfs_rename
);
2479 EXPORT_SYMBOL(vfs_rmdir
);
2480 EXPORT_SYMBOL(vfs_symlink
);
2481 EXPORT_SYMBOL(vfs_unlink
);
2482 EXPORT_SYMBOL(dentry_unhash
);
2483 EXPORT_SYMBOL(generic_readlink
);