4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <asm/uaccess.h>
37 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 static int __link_path_walk(const char *name
, struct nameidata
*nd
);
112 /* In order to reduce some races, while at the same time doing additional
113 * checking and hopefully speeding things up, we copy filenames to the
114 * kernel data space before using them..
116 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
117 * PATH_MAX includes the nul terminator --RR.
119 static int do_getname(const char __user
*filename
, char *page
)
122 unsigned long len
= PATH_MAX
;
124 if (!segment_eq(get_fs(), KERNEL_DS
)) {
125 if ((unsigned long) filename
>= TASK_SIZE
)
127 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
128 len
= TASK_SIZE
- (unsigned long) filename
;
131 retval
= strncpy_from_user(page
, filename
, len
);
135 return -ENAMETOOLONG
;
141 char * getname(const char __user
* filename
)
145 result
= ERR_PTR(-ENOMEM
);
148 int retval
= do_getname(filename
, tmp
);
153 result
= ERR_PTR(retval
);
156 audit_getname(result
);
160 #ifdef CONFIG_AUDITSYSCALL
161 void putname(const char *name
)
163 if (unlikely(!audit_dummy_context()))
168 EXPORT_SYMBOL(putname
);
173 * generic_permission - check for access rights on a Posix-like filesystem
174 * @inode: inode to check access rights for
175 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
176 * @check_acl: optional callback to check for Posix ACLs
178 * Used to check for read/write/execute permissions on a file.
179 * We use "fsuid" for this, letting us set arbitrary permissions
180 * for filesystem access without changing the "normal" uids which
181 * are used for other things..
183 int generic_permission(struct inode
*inode
, int mask
,
184 int (*check_acl
)(struct inode
*inode
, int mask
))
186 umode_t mode
= inode
->i_mode
;
188 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
190 if (current_fsuid() == inode
->i_uid
)
193 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
194 int error
= check_acl(inode
, mask
);
195 if (error
== -EACCES
)
196 goto check_capabilities
;
197 else if (error
!= -EAGAIN
)
201 if (in_group_p(inode
->i_gid
))
206 * If the DACs are ok we don't need any capability check.
208 if ((mask
& ~mode
) == 0)
213 * Read/write DACs are always overridable.
214 * Executable DACs are overridable if at least one exec bit is set.
216 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
217 if (capable(CAP_DAC_OVERRIDE
))
221 * Searching includes executable on directories, else just read.
223 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
224 if (capable(CAP_DAC_READ_SEARCH
))
231 * inode_permission - check for access rights to a given inode
232 * @inode: inode to check permission on
233 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
235 * Used to check for read/write/execute permissions on an inode.
236 * We use "fsuid" for this, letting us set arbitrary permissions
237 * for filesystem access without changing the "normal" uids which
238 * are used for other things.
240 int inode_permission(struct inode
*inode
, int mask
)
244 if (mask
& MAY_WRITE
) {
245 umode_t mode
= inode
->i_mode
;
248 * Nobody gets write access to a read-only fs.
250 if (IS_RDONLY(inode
) &&
251 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
255 * Nobody gets write access to an immutable file.
257 if (IS_IMMUTABLE(inode
))
261 if (inode
->i_op
->permission
)
262 retval
= inode
->i_op
->permission(inode
, mask
);
264 retval
= generic_permission(inode
, mask
, NULL
);
269 retval
= devcgroup_inode_permission(inode
, mask
);
273 return security_inode_permission(inode
,
274 mask
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
|MAY_APPEND
));
278 * file_permission - check for additional access rights to a given file
279 * @file: file to check access rights for
280 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
282 * Used to check for read/write/execute permissions on an already opened
286 * Do not use this function in new code. All access checks should
287 * be done using inode_permission().
289 int file_permission(struct file
*file
, int mask
)
291 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
295 * get_write_access() gets write permission for a file.
296 * put_write_access() releases this write permission.
297 * This is used for regular files.
298 * We cannot support write (and maybe mmap read-write shared) accesses and
299 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
300 * can have the following values:
301 * 0: no writers, no VM_DENYWRITE mappings
302 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
303 * > 0: (i_writecount) users are writing to the file.
305 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
306 * except for the cases where we don't hold i_writecount yet. Then we need to
307 * use {get,deny}_write_access() - these functions check the sign and refuse
308 * to do the change if sign is wrong. Exclusion between them is provided by
309 * the inode->i_lock spinlock.
312 int get_write_access(struct inode
* inode
)
314 spin_lock(&inode
->i_lock
);
315 if (atomic_read(&inode
->i_writecount
) < 0) {
316 spin_unlock(&inode
->i_lock
);
319 atomic_inc(&inode
->i_writecount
);
320 spin_unlock(&inode
->i_lock
);
325 int deny_write_access(struct file
* file
)
327 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
329 spin_lock(&inode
->i_lock
);
330 if (atomic_read(&inode
->i_writecount
) > 0) {
331 spin_unlock(&inode
->i_lock
);
334 atomic_dec(&inode
->i_writecount
);
335 spin_unlock(&inode
->i_lock
);
341 * path_get - get a reference to a path
342 * @path: path to get the reference to
344 * Given a path increment the reference count to the dentry and the vfsmount.
346 void path_get(struct path
*path
)
351 EXPORT_SYMBOL(path_get
);
354 * path_put - put a reference to a path
355 * @path: path to put the reference to
357 * Given a path decrement the reference count to the dentry and the vfsmount.
359 void path_put(struct path
*path
)
364 EXPORT_SYMBOL(path_put
);
367 * release_open_intent - free up open intent resources
368 * @nd: pointer to nameidata
370 void release_open_intent(struct nameidata
*nd
)
372 if (nd
->intent
.open
.file
->f_path
.dentry
== NULL
)
373 put_filp(nd
->intent
.open
.file
);
375 fput(nd
->intent
.open
.file
);
378 static inline struct dentry
*
379 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
381 int status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
382 if (unlikely(status
<= 0)) {
384 * The dentry failed validation.
385 * If d_revalidate returned 0 attempt to invalidate
386 * the dentry otherwise d_revalidate is asking us
387 * to return a fail status.
390 if (!d_invalidate(dentry
)) {
396 dentry
= ERR_PTR(status
);
403 * Internal lookup() using the new generic dcache.
406 static struct dentry
* cached_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
408 struct dentry
* dentry
= __d_lookup(parent
, name
);
410 /* lockess __d_lookup may fail due to concurrent d_move()
411 * in some unrelated directory, so try with d_lookup
414 dentry
= d_lookup(parent
, name
);
416 if (dentry
&& dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
417 dentry
= do_revalidate(dentry
, nd
);
423 * Short-cut version of permission(), for calling by
424 * path_walk(), when dcache lock is held. Combines parts
425 * of permission() and generic_permission(), and tests ONLY for
426 * MAY_EXEC permission.
428 * If appropriate, check DAC only. If not appropriate, or
429 * short-cut DAC fails, then call permission() to do more
430 * complete permission check.
432 static int exec_permission_lite(struct inode
*inode
)
434 umode_t mode
= inode
->i_mode
;
436 if (inode
->i_op
->permission
)
439 if (current_fsuid() == inode
->i_uid
)
441 else if (in_group_p(inode
->i_gid
))
447 if ((inode
->i_mode
& S_IXUGO
) && capable(CAP_DAC_OVERRIDE
))
450 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_OVERRIDE
))
453 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_READ_SEARCH
))
458 return security_inode_permission(inode
, MAY_EXEC
);
462 * This is called when everything else fails, and we actually have
463 * to go to the low-level filesystem to find out what we should do..
465 * We get the directory semaphore, and after getting that we also
466 * make sure that nobody added the entry to the dcache in the meantime..
469 static struct dentry
* real_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
471 struct dentry
* result
;
472 struct inode
*dir
= parent
->d_inode
;
474 mutex_lock(&dir
->i_mutex
);
476 * First re-do the cached lookup just in case it was created
477 * while we waited for the directory semaphore..
479 * FIXME! This could use version numbering or similar to
480 * avoid unnecessary cache lookups.
482 * The "dcache_lock" is purely to protect the RCU list walker
483 * from concurrent renames at this point (we mustn't get false
484 * negatives from the RCU list walk here, unlike the optimistic
487 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
489 result
= d_lookup(parent
, name
);
491 struct dentry
*dentry
;
493 /* Don't create child dentry for a dead directory. */
494 result
= ERR_PTR(-ENOENT
);
498 dentry
= d_alloc(parent
, name
);
499 result
= ERR_PTR(-ENOMEM
);
501 result
= dir
->i_op
->lookup(dir
, dentry
, nd
);
508 mutex_unlock(&dir
->i_mutex
);
513 * Uhhuh! Nasty case: the cache was re-populated while
514 * we waited on the semaphore. Need to revalidate.
516 mutex_unlock(&dir
->i_mutex
);
517 if (result
->d_op
&& result
->d_op
->d_revalidate
) {
518 result
= do_revalidate(result
, nd
);
520 result
= ERR_PTR(-ENOENT
);
526 * Wrapper to retry pathname resolution whenever the underlying
527 * file system returns an ESTALE.
529 * Retry the whole path once, forcing real lookup requests
530 * instead of relying on the dcache.
532 static __always_inline
int link_path_walk(const char *name
, struct nameidata
*nd
)
534 struct path save
= nd
->path
;
537 /* make sure the stuff we saved doesn't go away */
540 result
= __link_path_walk(name
, nd
);
541 if (result
== -ESTALE
) {
542 /* nd->path had been dropped */
545 nd
->flags
|= LOOKUP_REVAL
;
546 result
= __link_path_walk(name
, nd
);
554 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
562 struct fs_struct
*fs
= current
->fs
;
566 read_lock(&fs
->lock
);
569 read_unlock(&fs
->lock
);
572 res
= link_path_walk(link
, nd
);
573 if (nd
->depth
|| res
|| nd
->last_type
!=LAST_NORM
)
576 * If it is an iterative symlinks resolution in open_namei() we
577 * have to copy the last component. And all that crap because of
578 * bloody create() on broken symlinks. Furrfu...
581 if (unlikely(!name
)) {
585 strcpy(name
, nd
->last
.name
);
586 nd
->last
.name
= name
;
590 return PTR_ERR(link
);
593 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
596 if (path
->mnt
!= nd
->path
.mnt
)
600 static inline void path_to_nameidata(struct path
*path
, struct nameidata
*nd
)
602 dput(nd
->path
.dentry
);
603 if (nd
->path
.mnt
!= path
->mnt
)
604 mntput(nd
->path
.mnt
);
605 nd
->path
.mnt
= path
->mnt
;
606 nd
->path
.dentry
= path
->dentry
;
609 static __always_inline
int __do_follow_link(struct path
*path
, struct nameidata
*nd
)
613 struct dentry
*dentry
= path
->dentry
;
615 touch_atime(path
->mnt
, dentry
);
616 nd_set_link(nd
, NULL
);
618 if (path
->mnt
!= nd
->path
.mnt
) {
619 path_to_nameidata(path
, nd
);
623 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
624 error
= PTR_ERR(cookie
);
625 if (!IS_ERR(cookie
)) {
626 char *s
= nd_get_link(nd
);
629 error
= __vfs_follow_link(nd
, s
);
630 if (dentry
->d_inode
->i_op
->put_link
)
631 dentry
->d_inode
->i_op
->put_link(dentry
, nd
, cookie
);
639 * This limits recursive symlink follows to 8, while
640 * limiting consecutive symlinks to 40.
642 * Without that kind of total limit, nasty chains of consecutive
643 * symlinks can cause almost arbitrarily long lookups.
645 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
648 if (current
->link_count
>= MAX_NESTED_LINKS
)
650 if (current
->total_link_count
>= 40)
652 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
654 err
= security_inode_follow_link(path
->dentry
, nd
);
657 current
->link_count
++;
658 current
->total_link_count
++;
660 err
= __do_follow_link(path
, nd
);
661 current
->link_count
--;
665 path_put_conditional(path
, nd
);
670 int follow_up(struct vfsmount
**mnt
, struct dentry
**dentry
)
672 struct vfsmount
*parent
;
673 struct dentry
*mountpoint
;
674 spin_lock(&vfsmount_lock
);
675 parent
=(*mnt
)->mnt_parent
;
676 if (parent
== *mnt
) {
677 spin_unlock(&vfsmount_lock
);
681 mountpoint
=dget((*mnt
)->mnt_mountpoint
);
682 spin_unlock(&vfsmount_lock
);
684 *dentry
= mountpoint
;
690 /* no need for dcache_lock, as serialization is taken care in
693 static int __follow_mount(struct path
*path
)
696 while (d_mountpoint(path
->dentry
)) {
697 struct vfsmount
*mounted
= lookup_mnt(path
->mnt
, path
->dentry
);
704 path
->dentry
= dget(mounted
->mnt_root
);
710 static void follow_mount(struct vfsmount
**mnt
, struct dentry
**dentry
)
712 while (d_mountpoint(*dentry
)) {
713 struct vfsmount
*mounted
= lookup_mnt(*mnt
, *dentry
);
719 *dentry
= dget(mounted
->mnt_root
);
723 /* no need for dcache_lock, as serialization is taken care in
726 int follow_down(struct vfsmount
**mnt
, struct dentry
**dentry
)
728 struct vfsmount
*mounted
;
730 mounted
= lookup_mnt(*mnt
, *dentry
);
735 *dentry
= dget(mounted
->mnt_root
);
741 static __always_inline
void follow_dotdot(struct nameidata
*nd
)
743 struct fs_struct
*fs
= current
->fs
;
746 struct vfsmount
*parent
;
747 struct dentry
*old
= nd
->path
.dentry
;
749 read_lock(&fs
->lock
);
750 if (nd
->path
.dentry
== fs
->root
.dentry
&&
751 nd
->path
.mnt
== fs
->root
.mnt
) {
752 read_unlock(&fs
->lock
);
755 read_unlock(&fs
->lock
);
756 spin_lock(&dcache_lock
);
757 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
758 nd
->path
.dentry
= dget(nd
->path
.dentry
->d_parent
);
759 spin_unlock(&dcache_lock
);
763 spin_unlock(&dcache_lock
);
764 spin_lock(&vfsmount_lock
);
765 parent
= nd
->path
.mnt
->mnt_parent
;
766 if (parent
== nd
->path
.mnt
) {
767 spin_unlock(&vfsmount_lock
);
771 nd
->path
.dentry
= dget(nd
->path
.mnt
->mnt_mountpoint
);
772 spin_unlock(&vfsmount_lock
);
774 mntput(nd
->path
.mnt
);
775 nd
->path
.mnt
= parent
;
777 follow_mount(&nd
->path
.mnt
, &nd
->path
.dentry
);
781 * It's more convoluted than I'd like it to be, but... it's still fairly
782 * small and for now I'd prefer to have fast path as straight as possible.
783 * It _is_ time-critical.
785 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
788 struct vfsmount
*mnt
= nd
->path
.mnt
;
789 struct dentry
*dentry
= __d_lookup(nd
->path
.dentry
, name
);
793 if (dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
794 goto need_revalidate
;
797 path
->dentry
= dentry
;
798 __follow_mount(path
);
802 dentry
= real_lookup(nd
->path
.dentry
, name
, nd
);
808 dentry
= do_revalidate(dentry
, nd
);
816 return PTR_ERR(dentry
);
821 * This is the basic name resolution function, turning a pathname into
822 * the final dentry. We expect 'base' to be positive and a directory.
824 * Returns 0 and nd will have valid dentry and mnt on success.
825 * Returns error and drops reference to input namei data on failure.
827 static int __link_path_walk(const char *name
, struct nameidata
*nd
)
832 unsigned int lookup_flags
= nd
->flags
;
839 inode
= nd
->path
.dentry
->d_inode
;
841 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
843 /* At this point we know we have a real path component. */
849 nd
->flags
|= LOOKUP_CONTINUE
;
850 err
= exec_permission_lite(inode
);
852 err
= inode_permission(nd
->path
.dentry
->d_inode
,
855 err
= ima_path_check(&nd
->path
, MAY_EXEC
);
860 c
= *(const unsigned char *)name
;
862 hash
= init_name_hash();
865 hash
= partial_name_hash(c
, hash
);
866 c
= *(const unsigned char *)name
;
867 } while (c
&& (c
!= '/'));
868 this.len
= name
- (const char *) this.name
;
869 this.hash
= end_name_hash(hash
);
871 /* remove trailing slashes? */
874 while (*++name
== '/');
876 goto last_with_slashes
;
879 * "." and ".." are special - ".." especially so because it has
880 * to be able to know about the current root directory and
881 * parent relationships.
883 if (this.name
[0] == '.') switch (this.len
) {
887 if (this.name
[1] != '.')
890 inode
= nd
->path
.dentry
->d_inode
;
896 * See if the low-level filesystem might want
897 * to use its own hash..
899 if (nd
->path
.dentry
->d_op
&& nd
->path
.dentry
->d_op
->d_hash
) {
900 err
= nd
->path
.dentry
->d_op
->d_hash(nd
->path
.dentry
,
905 /* This does the actual lookups.. */
906 err
= do_lookup(nd
, &this, &next
);
911 inode
= next
.dentry
->d_inode
;
915 if (inode
->i_op
->follow_link
) {
916 err
= do_follow_link(&next
, nd
);
920 inode
= nd
->path
.dentry
->d_inode
;
924 path_to_nameidata(&next
, nd
);
926 if (!inode
->i_op
->lookup
)
929 /* here ends the main loop */
932 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
934 /* Clear LOOKUP_CONTINUE iff it was previously unset */
935 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
936 if (lookup_flags
& LOOKUP_PARENT
)
938 if (this.name
[0] == '.') switch (this.len
) {
942 if (this.name
[1] != '.')
945 inode
= nd
->path
.dentry
->d_inode
;
950 if (nd
->path
.dentry
->d_op
&& nd
->path
.dentry
->d_op
->d_hash
) {
951 err
= nd
->path
.dentry
->d_op
->d_hash(nd
->path
.dentry
,
956 err
= do_lookup(nd
, &this, &next
);
959 inode
= next
.dentry
->d_inode
;
960 if ((lookup_flags
& LOOKUP_FOLLOW
)
961 && inode
&& inode
->i_op
->follow_link
) {
962 err
= do_follow_link(&next
, nd
);
965 inode
= nd
->path
.dentry
->d_inode
;
967 path_to_nameidata(&next
, nd
);
971 if (lookup_flags
& LOOKUP_DIRECTORY
) {
973 if (!inode
->i_op
->lookup
)
979 nd
->last_type
= LAST_NORM
;
980 if (this.name
[0] != '.')
983 nd
->last_type
= LAST_DOT
;
984 else if (this.len
== 2 && this.name
[1] == '.')
985 nd
->last_type
= LAST_DOTDOT
;
990 * We bypassed the ordinary revalidation routines.
991 * We may need to check the cached dentry for staleness.
993 if (nd
->path
.dentry
&& nd
->path
.dentry
->d_sb
&&
994 (nd
->path
.dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)) {
996 /* Note: we do not d_invalidate() */
997 if (!nd
->path
.dentry
->d_op
->d_revalidate(
998 nd
->path
.dentry
, nd
))
1004 path_put_conditional(&next
, nd
);
1007 path_put(&nd
->path
);
1012 static int path_walk(const char *name
, struct nameidata
*nd
)
1014 current
->total_link_count
= 0;
1015 return link_path_walk(name
, nd
);
1018 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1019 static int do_path_lookup(int dfd
, const char *name
,
1020 unsigned int flags
, struct nameidata
*nd
)
1025 struct fs_struct
*fs
= current
->fs
;
1027 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1032 read_lock(&fs
->lock
);
1033 nd
->path
= fs
->root
;
1034 path_get(&fs
->root
);
1035 read_unlock(&fs
->lock
);
1036 } else if (dfd
== AT_FDCWD
) {
1037 read_lock(&fs
->lock
);
1040 read_unlock(&fs
->lock
);
1042 struct dentry
*dentry
;
1044 file
= fget_light(dfd
, &fput_needed
);
1049 dentry
= file
->f_path
.dentry
;
1052 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1055 retval
= file_permission(file
, MAY_EXEC
);
1059 nd
->path
= file
->f_path
;
1060 path_get(&file
->f_path
);
1062 fput_light(file
, fput_needed
);
1065 retval
= path_walk(name
, nd
);
1066 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1067 nd
->path
.dentry
->d_inode
))
1068 audit_inode(name
, nd
->path
.dentry
);
1073 fput_light(file
, fput_needed
);
1077 int path_lookup(const char *name
, unsigned int flags
,
1078 struct nameidata
*nd
)
1080 return do_path_lookup(AT_FDCWD
, name
, flags
, nd
);
1083 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1085 struct nameidata nd
;
1086 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1093 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1094 * @dentry: pointer to dentry of the base directory
1095 * @mnt: pointer to vfs mount of the base directory
1096 * @name: pointer to file name
1097 * @flags: lookup flags
1098 * @nd: pointer to nameidata
1100 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1101 const char *name
, unsigned int flags
,
1102 struct nameidata
*nd
)
1106 /* same as do_path_lookup */
1107 nd
->last_type
= LAST_ROOT
;
1111 nd
->path
.dentry
= dentry
;
1113 path_get(&nd
->path
);
1115 retval
= path_walk(name
, nd
);
1116 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1117 nd
->path
.dentry
->d_inode
))
1118 audit_inode(name
, nd
->path
.dentry
);
1125 * path_lookup_open - lookup a file path with open intent
1126 * @dfd: the directory to use as base, or AT_FDCWD
1127 * @name: pointer to file name
1128 * @lookup_flags: lookup intent flags
1129 * @nd: pointer to nameidata
1130 * @open_flags: open intent flags
1132 int path_lookup_open(int dfd
, const char *name
, unsigned int lookup_flags
,
1133 struct nameidata
*nd
, int open_flags
)
1135 struct file
*filp
= get_empty_filp();
1140 nd
->intent
.open
.file
= filp
;
1141 nd
->intent
.open
.flags
= open_flags
;
1142 nd
->intent
.open
.create_mode
= 0;
1143 err
= do_path_lookup(dfd
, name
, lookup_flags
|LOOKUP_OPEN
, nd
);
1144 if (IS_ERR(nd
->intent
.open
.file
)) {
1146 err
= PTR_ERR(nd
->intent
.open
.file
);
1147 path_put(&nd
->path
);
1149 } else if (err
!= 0)
1150 release_open_intent(nd
);
1154 static struct dentry
*__lookup_hash(struct qstr
*name
,
1155 struct dentry
*base
, struct nameidata
*nd
)
1157 struct dentry
*dentry
;
1158 struct inode
*inode
;
1161 inode
= base
->d_inode
;
1164 * See if the low-level filesystem might want
1165 * to use its own hash..
1167 if (base
->d_op
&& base
->d_op
->d_hash
) {
1168 err
= base
->d_op
->d_hash(base
, name
);
1169 dentry
= ERR_PTR(err
);
1174 dentry
= cached_lookup(base
, name
, nd
);
1178 /* Don't create child dentry for a dead directory. */
1179 dentry
= ERR_PTR(-ENOENT
);
1180 if (IS_DEADDIR(inode
))
1183 new = d_alloc(base
, name
);
1184 dentry
= ERR_PTR(-ENOMEM
);
1187 dentry
= inode
->i_op
->lookup(inode
, new, nd
);
1198 * Restricted form of lookup. Doesn't follow links, single-component only,
1199 * needs parent already locked. Doesn't follow mounts.
1202 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1206 err
= inode_permission(nd
->path
.dentry
->d_inode
, MAY_EXEC
);
1208 return ERR_PTR(err
);
1209 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1212 static int __lookup_one_len(const char *name
, struct qstr
*this,
1213 struct dentry
*base
, int len
)
1223 hash
= init_name_hash();
1225 c
= *(const unsigned char *)name
++;
1226 if (c
== '/' || c
== '\0')
1228 hash
= partial_name_hash(c
, hash
);
1230 this->hash
= end_name_hash(hash
);
1235 * lookup_one_len - filesystem helper to lookup single pathname component
1236 * @name: pathname component to lookup
1237 * @base: base directory to lookup from
1238 * @len: maximum length @len should be interpreted to
1240 * Note that this routine is purely a helper for filesystem usage and should
1241 * not be called by generic code. Also note that by using this function the
1242 * nameidata argument is passed to the filesystem methods and a filesystem
1243 * using this helper needs to be prepared for that.
1245 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1250 err
= __lookup_one_len(name
, &this, base
, len
);
1252 return ERR_PTR(err
);
1254 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
1256 return ERR_PTR(err
);
1257 return __lookup_hash(&this, base
, NULL
);
1261 * lookup_one_noperm - bad hack for sysfs
1262 * @name: pathname component to lookup
1263 * @base: base directory to lookup from
1265 * This is a variant of lookup_one_len that doesn't perform any permission
1266 * checks. It's a horrible hack to work around the braindead sysfs
1267 * architecture and should not be used anywhere else.
1269 * DON'T USE THIS FUNCTION EVER, thanks.
1271 struct dentry
*lookup_one_noperm(const char *name
, struct dentry
*base
)
1276 err
= __lookup_one_len(name
, &this, base
, strlen(name
));
1278 return ERR_PTR(err
);
1279 return __lookup_hash(&this, base
, NULL
);
1282 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1285 struct nameidata nd
;
1286 char *tmp
= getname(name
);
1287 int err
= PTR_ERR(tmp
);
1290 BUG_ON(flags
& LOOKUP_PARENT
);
1292 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1300 static int user_path_parent(int dfd
, const char __user
*path
,
1301 struct nameidata
*nd
, char **name
)
1303 char *s
= getname(path
);
1309 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1319 * It's inline, so penalty for filesystems that don't use sticky bit is
1322 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1324 uid_t fsuid
= current_fsuid();
1326 if (!(dir
->i_mode
& S_ISVTX
))
1328 if (inode
->i_uid
== fsuid
)
1330 if (dir
->i_uid
== fsuid
)
1332 return !capable(CAP_FOWNER
);
1336 * Check whether we can remove a link victim from directory dir, check
1337 * whether the type of victim is right.
1338 * 1. We can't do it if dir is read-only (done in permission())
1339 * 2. We should have write and exec permissions on dir
1340 * 3. We can't remove anything from append-only dir
1341 * 4. We can't do anything with immutable dir (done in permission())
1342 * 5. If the sticky bit on dir is set we should either
1343 * a. be owner of dir, or
1344 * b. be owner of victim, or
1345 * c. have CAP_FOWNER capability
1346 * 6. If the victim is append-only or immutable we can't do antyhing with
1347 * links pointing to it.
1348 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1349 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1350 * 9. We can't remove a root or mountpoint.
1351 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1352 * nfs_async_unlink().
1354 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1358 if (!victim
->d_inode
)
1361 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1362 audit_inode_child(victim
->d_name
.name
, victim
, dir
);
1364 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1369 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1370 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1373 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1375 if (IS_ROOT(victim
))
1377 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1379 if (IS_DEADDIR(dir
))
1381 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1386 /* Check whether we can create an object with dentry child in directory
1388 * 1. We can't do it if child already exists (open has special treatment for
1389 * this case, but since we are inlined it's OK)
1390 * 2. We can't do it if dir is read-only (done in permission())
1391 * 3. We should have write and exec permissions on dir
1392 * 4. We can't do it if dir is immutable (done in permission())
1394 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1398 if (IS_DEADDIR(dir
))
1400 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1404 * O_DIRECTORY translates into forcing a directory lookup.
1406 static inline int lookup_flags(unsigned int f
)
1408 unsigned long retval
= LOOKUP_FOLLOW
;
1411 retval
&= ~LOOKUP_FOLLOW
;
1413 if (f
& O_DIRECTORY
)
1414 retval
|= LOOKUP_DIRECTORY
;
1420 * p1 and p2 should be directories on the same fs.
1422 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1427 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1431 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1433 p
= d_ancestor(p2
, p1
);
1435 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1436 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1440 p
= d_ancestor(p1
, p2
);
1442 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1443 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1447 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1448 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1452 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1454 mutex_unlock(&p1
->d_inode
->i_mutex
);
1456 mutex_unlock(&p2
->d_inode
->i_mutex
);
1457 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1461 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1462 struct nameidata
*nd
)
1464 int error
= may_create(dir
, dentry
);
1469 if (!dir
->i_op
->create
)
1470 return -EACCES
; /* shouldn't it be ENOSYS? */
1473 error
= security_inode_create(dir
, dentry
, mode
);
1477 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1479 fsnotify_create(dir
, dentry
);
1483 int may_open(struct path
*path
, int acc_mode
, int flag
)
1485 struct dentry
*dentry
= path
->dentry
;
1486 struct inode
*inode
= dentry
->d_inode
;
1492 if (S_ISLNK(inode
->i_mode
))
1495 if (S_ISDIR(inode
->i_mode
) && (acc_mode
& MAY_WRITE
))
1499 * FIFO's, sockets and device files are special: they don't
1500 * actually live on the filesystem itself, and as such you
1501 * can write to them even if the filesystem is read-only.
1503 if (S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
1505 } else if (S_ISBLK(inode
->i_mode
) || S_ISCHR(inode
->i_mode
)) {
1506 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
1512 error
= inode_permission(inode
, acc_mode
);
1516 error
= ima_path_check(path
,
1517 acc_mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
));
1521 * An append-only file must be opened in append mode for writing.
1523 if (IS_APPEND(inode
)) {
1524 if ((flag
& FMODE_WRITE
) && !(flag
& O_APPEND
))
1530 /* O_NOATIME can only be set by the owner or superuser */
1531 if (flag
& O_NOATIME
)
1532 if (!is_owner_or_cap(inode
))
1536 * Ensure there are no outstanding leases on the file.
1538 error
= break_lease(inode
, flag
);
1542 if (flag
& O_TRUNC
) {
1543 error
= get_write_access(inode
);
1548 * Refuse to truncate files with mandatory locks held on them.
1550 error
= locks_verify_locked(inode
);
1552 error
= security_path_truncate(path
, 0,
1553 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
);
1557 error
= do_truncate(dentry
, 0,
1558 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
1561 put_write_access(inode
);
1565 if (flag
& FMODE_WRITE
)
1572 * Be careful about ever adding any more callers of this
1573 * function. Its flags must be in the namei format, not
1574 * what get passed to sys_open().
1576 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
1580 struct dentry
*dir
= nd
->path
.dentry
;
1582 if (!IS_POSIXACL(dir
->d_inode
))
1583 mode
&= ~current
->fs
->umask
;
1584 error
= security_path_mknod(&nd
->path
, path
->dentry
, mode
, 0);
1587 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
1589 mutex_unlock(&dir
->d_inode
->i_mutex
);
1590 dput(nd
->path
.dentry
);
1591 nd
->path
.dentry
= path
->dentry
;
1594 /* Don't check for write permission, don't truncate */
1595 return may_open(&nd
->path
, 0, flag
& ~O_TRUNC
);
1599 * Note that while the flag value (low two bits) for sys_open means:
1604 * it is changed into
1605 * 00 - no permissions needed
1606 * 01 - read-permission
1607 * 10 - write-permission
1609 * for the internal routines (ie open_namei()/follow_link() etc)
1610 * This is more logical, and also allows the 00 "no perm needed"
1611 * to be used for symlinks (where the permissions are checked
1615 static inline int open_to_namei_flags(int flag
)
1617 if ((flag
+1) & O_ACCMODE
)
1622 static int open_will_write_to_fs(int flag
, struct inode
*inode
)
1625 * We'll never write to the fs underlying
1628 if (special_file(inode
->i_mode
))
1630 return (flag
& O_TRUNC
);
1634 * Note that the low bits of the passed in "open_flag"
1635 * are not the same as in the local variable "flag". See
1636 * open_to_namei_flags() for more details.
1638 struct file
*do_filp_open(int dfd
, const char *pathname
,
1639 int open_flag
, int mode
)
1642 struct nameidata nd
;
1643 int acc_mode
, error
;
1648 int flag
= open_to_namei_flags(open_flag
);
1650 acc_mode
= MAY_OPEN
| ACC_MODE(flag
);
1652 /* O_TRUNC implies we need access checks for write permissions */
1654 acc_mode
|= MAY_WRITE
;
1656 /* Allow the LSM permission hook to distinguish append
1657 access from general write access. */
1658 if (flag
& O_APPEND
)
1659 acc_mode
|= MAY_APPEND
;
1662 * The simplest case - just a plain lookup.
1664 if (!(flag
& O_CREAT
)) {
1665 error
= path_lookup_open(dfd
, pathname
, lookup_flags(flag
),
1668 return ERR_PTR(error
);
1673 * Create - we need to know the parent.
1675 error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
, &nd
);
1677 return ERR_PTR(error
);
1680 * We have the parent and last component. First of all, check
1681 * that we are not asked to creat(2) an obvious directory - that
1685 if (nd
.last_type
!= LAST_NORM
|| nd
.last
.name
[nd
.last
.len
])
1689 filp
= get_empty_filp();
1692 nd
.intent
.open
.file
= filp
;
1693 nd
.intent
.open
.flags
= flag
;
1694 nd
.intent
.open
.create_mode
= mode
;
1695 dir
= nd
.path
.dentry
;
1696 nd
.flags
&= ~LOOKUP_PARENT
;
1697 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_OPEN
;
1699 nd
.flags
|= LOOKUP_EXCL
;
1700 mutex_lock(&dir
->d_inode
->i_mutex
);
1701 path
.dentry
= lookup_hash(&nd
);
1702 path
.mnt
= nd
.path
.mnt
;
1705 error
= PTR_ERR(path
.dentry
);
1706 if (IS_ERR(path
.dentry
)) {
1707 mutex_unlock(&dir
->d_inode
->i_mutex
);
1711 if (IS_ERR(nd
.intent
.open
.file
)) {
1712 error
= PTR_ERR(nd
.intent
.open
.file
);
1713 goto exit_mutex_unlock
;
1716 /* Negative dentry, just create the file */
1717 if (!path
.dentry
->d_inode
) {
1719 * This write is needed to ensure that a
1720 * ro->rw transition does not occur between
1721 * the time when the file is created and when
1722 * a permanent write count is taken through
1723 * the 'struct file' in nameidata_to_filp().
1725 error
= mnt_want_write(nd
.path
.mnt
);
1727 goto exit_mutex_unlock
;
1728 error
= __open_namei_create(&nd
, &path
, flag
, mode
);
1730 mnt_drop_write(nd
.path
.mnt
);
1733 filp
= nameidata_to_filp(&nd
, open_flag
);
1734 mnt_drop_write(nd
.path
.mnt
);
1739 * It already exists.
1741 mutex_unlock(&dir
->d_inode
->i_mutex
);
1742 audit_inode(pathname
, path
.dentry
);
1748 if (__follow_mount(&path
)) {
1750 if (flag
& O_NOFOLLOW
)
1755 if (!path
.dentry
->d_inode
)
1757 if (path
.dentry
->d_inode
->i_op
->follow_link
)
1760 path_to_nameidata(&path
, &nd
);
1762 if (path
.dentry
->d_inode
&& S_ISDIR(path
.dentry
->d_inode
->i_mode
))
1767 * 1. may_open() truncates a file
1768 * 2. a rw->ro mount transition occurs
1769 * 3. nameidata_to_filp() fails due to
1771 * That would be inconsistent, and should
1772 * be avoided. Taking this mnt write here
1773 * ensures that (2) can not occur.
1775 will_write
= open_will_write_to_fs(flag
, nd
.path
.dentry
->d_inode
);
1777 error
= mnt_want_write(nd
.path
.mnt
);
1781 error
= may_open(&nd
.path
, acc_mode
, flag
);
1784 mnt_drop_write(nd
.path
.mnt
);
1787 filp
= nameidata_to_filp(&nd
, open_flag
);
1789 * It is now safe to drop the mnt write
1790 * because the filp has had a write taken
1794 mnt_drop_write(nd
.path
.mnt
);
1798 mutex_unlock(&dir
->d_inode
->i_mutex
);
1800 path_put_conditional(&path
, &nd
);
1802 if (!IS_ERR(nd
.intent
.open
.file
))
1803 release_open_intent(&nd
);
1806 return ERR_PTR(error
);
1810 if (flag
& O_NOFOLLOW
)
1813 * This is subtle. Instead of calling do_follow_link() we do the
1814 * thing by hands. The reason is that this way we have zero link_count
1815 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1816 * After that we have the parent and last component, i.e.
1817 * we are in the same situation as after the first path_walk().
1818 * Well, almost - if the last component is normal we get its copy
1819 * stored in nd->last.name and we will have to putname() it when we
1820 * are done. Procfs-like symlinks just set LAST_BIND.
1822 nd
.flags
|= LOOKUP_PARENT
;
1823 error
= security_inode_follow_link(path
.dentry
, &nd
);
1826 error
= __do_follow_link(&path
, &nd
);
1828 /* Does someone understand code flow here? Or it is only
1829 * me so stupid? Anathema to whoever designed this non-sense
1830 * with "intent.open".
1832 release_open_intent(&nd
);
1833 return ERR_PTR(error
);
1835 nd
.flags
&= ~LOOKUP_PARENT
;
1836 if (nd
.last_type
== LAST_BIND
)
1839 if (nd
.last_type
!= LAST_NORM
)
1841 if (nd
.last
.name
[nd
.last
.len
]) {
1842 __putname(nd
.last
.name
);
1847 __putname(nd
.last
.name
);
1850 dir
= nd
.path
.dentry
;
1851 mutex_lock(&dir
->d_inode
->i_mutex
);
1852 path
.dentry
= lookup_hash(&nd
);
1853 path
.mnt
= nd
.path
.mnt
;
1854 __putname(nd
.last
.name
);
1859 * filp_open - open file and return file pointer
1861 * @filename: path to open
1862 * @flags: open flags as per the open(2) second argument
1863 * @mode: mode for the new file if O_CREAT is set, else ignored
1865 * This is the helper to open a file from kernelspace if you really
1866 * have to. But in generally you should not do this, so please move
1867 * along, nothing to see here..
1869 struct file
*filp_open(const char *filename
, int flags
, int mode
)
1871 return do_filp_open(AT_FDCWD
, filename
, flags
, mode
);
1873 EXPORT_SYMBOL(filp_open
);
1876 * lookup_create - lookup a dentry, creating it if it doesn't exist
1877 * @nd: nameidata info
1878 * @is_dir: directory flag
1880 * Simple function to lookup and return a dentry and create it
1881 * if it doesn't exist. Is SMP-safe.
1883 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1885 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
1887 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
1889 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1891 * Yucky last component or no last component at all?
1892 * (foo/., foo/.., /////)
1894 if (nd
->last_type
!= LAST_NORM
)
1896 nd
->flags
&= ~LOOKUP_PARENT
;
1897 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
1898 nd
->intent
.open
.flags
= O_EXCL
;
1901 * Do the final lookup.
1903 dentry
= lookup_hash(nd
);
1907 if (dentry
->d_inode
)
1910 * Special case - lookup gave negative, but... we had foo/bar/
1911 * From the vfs_mknod() POV we just have a negative dentry -
1912 * all is fine. Let's be bastards - you had / on the end, you've
1913 * been asking for (non-existent) directory. -ENOENT for you.
1915 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
1917 dentry
= ERR_PTR(-ENOENT
);
1922 dentry
= ERR_PTR(-EEXIST
);
1926 EXPORT_SYMBOL_GPL(lookup_create
);
1928 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1930 int error
= may_create(dir
, dentry
);
1935 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
1938 if (!dir
->i_op
->mknod
)
1941 error
= devcgroup_inode_mknod(mode
, dev
);
1945 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
1950 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
1952 fsnotify_create(dir
, dentry
);
1956 static int may_mknod(mode_t mode
)
1958 switch (mode
& S_IFMT
) {
1964 case 0: /* zero mode translates to S_IFREG */
1973 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
1978 struct dentry
*dentry
;
1979 struct nameidata nd
;
1984 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
1988 dentry
= lookup_create(&nd
, 0);
1989 if (IS_ERR(dentry
)) {
1990 error
= PTR_ERR(dentry
);
1993 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
1994 mode
&= ~current
->fs
->umask
;
1995 error
= may_mknod(mode
);
1998 error
= mnt_want_write(nd
.path
.mnt
);
2001 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2003 goto out_drop_write
;
2004 switch (mode
& S_IFMT
) {
2005 case 0: case S_IFREG
:
2006 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2008 case S_IFCHR
: case S_IFBLK
:
2009 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2010 new_decode_dev(dev
));
2012 case S_IFIFO
: case S_IFSOCK
:
2013 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2017 mnt_drop_write(nd
.path
.mnt
);
2021 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2028 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2030 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2033 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2035 int error
= may_create(dir
, dentry
);
2040 if (!dir
->i_op
->mkdir
)
2043 mode
&= (S_IRWXUGO
|S_ISVTX
);
2044 error
= security_inode_mkdir(dir
, dentry
, mode
);
2049 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2051 fsnotify_mkdir(dir
, dentry
);
2055 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2059 struct dentry
*dentry
;
2060 struct nameidata nd
;
2062 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2066 dentry
= lookup_create(&nd
, 1);
2067 error
= PTR_ERR(dentry
);
2071 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2072 mode
&= ~current
->fs
->umask
;
2073 error
= mnt_want_write(nd
.path
.mnt
);
2076 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2078 goto out_drop_write
;
2079 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2081 mnt_drop_write(nd
.path
.mnt
);
2085 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2092 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2094 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2098 * We try to drop the dentry early: we should have
2099 * a usage count of 2 if we're the only user of this
2100 * dentry, and if that is true (possibly after pruning
2101 * the dcache), then we drop the dentry now.
2103 * A low-level filesystem can, if it choses, legally
2106 * if (!d_unhashed(dentry))
2109 * if it cannot handle the case of removing a directory
2110 * that is still in use by something else..
2112 void dentry_unhash(struct dentry
*dentry
)
2115 shrink_dcache_parent(dentry
);
2116 spin_lock(&dcache_lock
);
2117 spin_lock(&dentry
->d_lock
);
2118 if (atomic_read(&dentry
->d_count
) == 2)
2120 spin_unlock(&dentry
->d_lock
);
2121 spin_unlock(&dcache_lock
);
2124 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2126 int error
= may_delete(dir
, dentry
, 1);
2131 if (!dir
->i_op
->rmdir
)
2136 mutex_lock(&dentry
->d_inode
->i_mutex
);
2137 dentry_unhash(dentry
);
2138 if (d_mountpoint(dentry
))
2141 error
= security_inode_rmdir(dir
, dentry
);
2143 error
= dir
->i_op
->rmdir(dir
, dentry
);
2145 dentry
->d_inode
->i_flags
|= S_DEAD
;
2148 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2157 static long do_rmdir(int dfd
, const char __user
*pathname
)
2161 struct dentry
*dentry
;
2162 struct nameidata nd
;
2164 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2168 switch(nd
.last_type
) {
2180 nd
.flags
&= ~LOOKUP_PARENT
;
2182 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2183 dentry
= lookup_hash(&nd
);
2184 error
= PTR_ERR(dentry
);
2187 error
= mnt_want_write(nd
.path
.mnt
);
2190 error
= security_path_rmdir(&nd
.path
, dentry
);
2193 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2195 mnt_drop_write(nd
.path
.mnt
);
2199 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2206 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2208 return do_rmdir(AT_FDCWD
, pathname
);
2211 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2213 int error
= may_delete(dir
, dentry
, 0);
2218 if (!dir
->i_op
->unlink
)
2223 mutex_lock(&dentry
->d_inode
->i_mutex
);
2224 if (d_mountpoint(dentry
))
2227 error
= security_inode_unlink(dir
, dentry
);
2229 error
= dir
->i_op
->unlink(dir
, dentry
);
2231 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2233 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2234 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2235 fsnotify_link_count(dentry
->d_inode
);
2243 * Make sure that the actual truncation of the file will occur outside its
2244 * directory's i_mutex. Truncate can take a long time if there is a lot of
2245 * writeout happening, and we don't want to prevent access to the directory
2246 * while waiting on the I/O.
2248 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2252 struct dentry
*dentry
;
2253 struct nameidata nd
;
2254 struct inode
*inode
= NULL
;
2256 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2261 if (nd
.last_type
!= LAST_NORM
)
2264 nd
.flags
&= ~LOOKUP_PARENT
;
2266 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2267 dentry
= lookup_hash(&nd
);
2268 error
= PTR_ERR(dentry
);
2269 if (!IS_ERR(dentry
)) {
2270 /* Why not before? Because we want correct error value */
2271 if (nd
.last
.name
[nd
.last
.len
])
2273 inode
= dentry
->d_inode
;
2275 atomic_inc(&inode
->i_count
);
2276 error
= mnt_want_write(nd
.path
.mnt
);
2279 error
= security_path_unlink(&nd
.path
, dentry
);
2282 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2284 mnt_drop_write(nd
.path
.mnt
);
2288 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2290 iput(inode
); /* truncate the inode here */
2297 error
= !dentry
->d_inode
? -ENOENT
:
2298 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2302 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2304 if ((flag
& ~AT_REMOVEDIR
) != 0)
2307 if (flag
& AT_REMOVEDIR
)
2308 return do_rmdir(dfd
, pathname
);
2310 return do_unlinkat(dfd
, pathname
);
2313 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2315 return do_unlinkat(AT_FDCWD
, pathname
);
2318 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2320 int error
= may_create(dir
, dentry
);
2325 if (!dir
->i_op
->symlink
)
2328 error
= security_inode_symlink(dir
, dentry
, oldname
);
2333 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2335 fsnotify_create(dir
, dentry
);
2339 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2340 int, newdfd
, const char __user
*, newname
)
2345 struct dentry
*dentry
;
2346 struct nameidata nd
;
2348 from
= getname(oldname
);
2350 return PTR_ERR(from
);
2352 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2356 dentry
= lookup_create(&nd
, 0);
2357 error
= PTR_ERR(dentry
);
2361 error
= mnt_want_write(nd
.path
.mnt
);
2364 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2366 goto out_drop_write
;
2367 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2369 mnt_drop_write(nd
.path
.mnt
);
2373 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2381 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2383 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2386 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2388 struct inode
*inode
= old_dentry
->d_inode
;
2394 error
= may_create(dir
, new_dentry
);
2398 if (dir
->i_sb
!= inode
->i_sb
)
2402 * A link to an append-only or immutable file cannot be created.
2404 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2406 if (!dir
->i_op
->link
)
2408 if (S_ISDIR(inode
->i_mode
))
2411 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2415 mutex_lock(&inode
->i_mutex
);
2417 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2418 mutex_unlock(&inode
->i_mutex
);
2420 fsnotify_link(dir
, inode
, new_dentry
);
2425 * Hardlinks are often used in delicate situations. We avoid
2426 * security-related surprises by not following symlinks on the
2429 * We don't follow them on the oldname either to be compatible
2430 * with linux 2.0, and to avoid hard-linking to directories
2431 * and other special files. --ADM
2433 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2434 int, newdfd
, const char __user
*, newname
, int, flags
)
2436 struct dentry
*new_dentry
;
2437 struct nameidata nd
;
2438 struct path old_path
;
2442 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
2445 error
= user_path_at(olddfd
, oldname
,
2446 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
2451 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2455 if (old_path
.mnt
!= nd
.path
.mnt
)
2457 new_dentry
= lookup_create(&nd
, 0);
2458 error
= PTR_ERR(new_dentry
);
2459 if (IS_ERR(new_dentry
))
2461 error
= mnt_want_write(nd
.path
.mnt
);
2464 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2466 goto out_drop_write
;
2467 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2469 mnt_drop_write(nd
.path
.mnt
);
2473 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2478 path_put(&old_path
);
2483 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
2485 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2489 * The worst of all namespace operations - renaming directory. "Perverted"
2490 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2492 * a) we can get into loop creation. Check is done in is_subdir().
2493 * b) race potential - two innocent renames can create a loop together.
2494 * That's where 4.4 screws up. Current fix: serialization on
2495 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2497 * c) we have to lock _three_ objects - parents and victim (if it exists).
2498 * And that - after we got ->i_mutex on parents (until then we don't know
2499 * whether the target exists). Solution: try to be smart with locking
2500 * order for inodes. We rely on the fact that tree topology may change
2501 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2502 * move will be locked. Thus we can rank directories by the tree
2503 * (ancestors first) and rank all non-directories after them.
2504 * That works since everybody except rename does "lock parent, lookup,
2505 * lock child" and rename is under ->s_vfs_rename_mutex.
2506 * HOWEVER, it relies on the assumption that any object with ->lookup()
2507 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2508 * we'd better make sure that there's no link(2) for them.
2509 * d) some filesystems don't support opened-but-unlinked directories,
2510 * either because of layout or because they are not ready to deal with
2511 * all cases correctly. The latter will be fixed (taking this sort of
2512 * stuff into VFS), but the former is not going away. Solution: the same
2513 * trick as in rmdir().
2514 * e) conversion from fhandle to dentry may come in the wrong moment - when
2515 * we are removing the target. Solution: we will have to grab ->i_mutex
2516 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2517 * ->i_mutex on parents, which works but leads to some truely excessive
2520 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2521 struct inode
*new_dir
, struct dentry
*new_dentry
)
2524 struct inode
*target
;
2527 * If we are going to change the parent - check write permissions,
2528 * we'll need to flip '..'.
2530 if (new_dir
!= old_dir
) {
2531 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
2536 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2540 target
= new_dentry
->d_inode
;
2542 mutex_lock(&target
->i_mutex
);
2543 dentry_unhash(new_dentry
);
2545 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2548 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2551 target
->i_flags
|= S_DEAD
;
2552 mutex_unlock(&target
->i_mutex
);
2553 if (d_unhashed(new_dentry
))
2554 d_rehash(new_dentry
);
2558 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2559 d_move(old_dentry
,new_dentry
);
2563 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
2564 struct inode
*new_dir
, struct dentry
*new_dentry
)
2566 struct inode
*target
;
2569 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2574 target
= new_dentry
->d_inode
;
2576 mutex_lock(&target
->i_mutex
);
2577 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2580 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2582 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2583 d_move(old_dentry
, new_dentry
);
2586 mutex_unlock(&target
->i_mutex
);
2591 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2592 struct inode
*new_dir
, struct dentry
*new_dentry
)
2595 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2596 const char *old_name
;
2598 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
2601 error
= may_delete(old_dir
, old_dentry
, is_dir
);
2605 if (!new_dentry
->d_inode
)
2606 error
= may_create(new_dir
, new_dentry
);
2608 error
= may_delete(new_dir
, new_dentry
, is_dir
);
2612 if (!old_dir
->i_op
->rename
)
2615 DQUOT_INIT(old_dir
);
2616 DQUOT_INIT(new_dir
);
2618 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
2621 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
2623 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
2625 const char *new_name
= old_dentry
->d_name
.name
;
2626 fsnotify_move(old_dir
, new_dir
, old_name
, new_name
, is_dir
,
2627 new_dentry
->d_inode
, old_dentry
);
2629 fsnotify_oldname_free(old_name
);
2634 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
2635 int, newdfd
, const char __user
*, newname
)
2637 struct dentry
*old_dir
, *new_dir
;
2638 struct dentry
*old_dentry
, *new_dentry
;
2639 struct dentry
*trap
;
2640 struct nameidata oldnd
, newnd
;
2645 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
2649 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
2654 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
2657 old_dir
= oldnd
.path
.dentry
;
2659 if (oldnd
.last_type
!= LAST_NORM
)
2662 new_dir
= newnd
.path
.dentry
;
2663 if (newnd
.last_type
!= LAST_NORM
)
2666 oldnd
.flags
&= ~LOOKUP_PARENT
;
2667 newnd
.flags
&= ~LOOKUP_PARENT
;
2668 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
2670 trap
= lock_rename(new_dir
, old_dir
);
2672 old_dentry
= lookup_hash(&oldnd
);
2673 error
= PTR_ERR(old_dentry
);
2674 if (IS_ERR(old_dentry
))
2676 /* source must exist */
2678 if (!old_dentry
->d_inode
)
2680 /* unless the source is a directory trailing slashes give -ENOTDIR */
2681 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
2683 if (oldnd
.last
.name
[oldnd
.last
.len
])
2685 if (newnd
.last
.name
[newnd
.last
.len
])
2688 /* source should not be ancestor of target */
2690 if (old_dentry
== trap
)
2692 new_dentry
= lookup_hash(&newnd
);
2693 error
= PTR_ERR(new_dentry
);
2694 if (IS_ERR(new_dentry
))
2696 /* target should not be an ancestor of source */
2698 if (new_dentry
== trap
)
2701 error
= mnt_want_write(oldnd
.path
.mnt
);
2704 error
= security_path_rename(&oldnd
.path
, old_dentry
,
2705 &newnd
.path
, new_dentry
);
2708 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
2709 new_dir
->d_inode
, new_dentry
);
2711 mnt_drop_write(oldnd
.path
.mnt
);
2717 unlock_rename(new_dir
, old_dir
);
2719 path_put(&newnd
.path
);
2722 path_put(&oldnd
.path
);
2728 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
2730 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
2733 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
2737 len
= PTR_ERR(link
);
2742 if (len
> (unsigned) buflen
)
2744 if (copy_to_user(buffer
, link
, len
))
2751 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2752 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2753 * using) it for any given inode is up to filesystem.
2755 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2757 struct nameidata nd
;
2762 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
2764 return PTR_ERR(cookie
);
2766 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
2767 if (dentry
->d_inode
->i_op
->put_link
)
2768 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
2772 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
2774 return __vfs_follow_link(nd
, link
);
2777 /* get the link contents into pagecache */
2778 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
2782 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
2783 page
= read_mapping_page(mapping
, 0, NULL
);
2788 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
2792 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2794 struct page
*page
= NULL
;
2795 char *s
= page_getlink(dentry
, &page
);
2796 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
2799 page_cache_release(page
);
2804 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
2806 struct page
*page
= NULL
;
2807 nd_set_link(nd
, page_getlink(dentry
, &page
));
2811 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2813 struct page
*page
= cookie
;
2817 page_cache_release(page
);
2822 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2824 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
2826 struct address_space
*mapping
= inode
->i_mapping
;
2831 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
2833 flags
|= AOP_FLAG_NOFS
;
2836 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
2837 flags
, &page
, &fsdata
);
2841 kaddr
= kmap_atomic(page
, KM_USER0
);
2842 memcpy(kaddr
, symname
, len
-1);
2843 kunmap_atomic(kaddr
, KM_USER0
);
2845 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
2852 mark_inode_dirty(inode
);
2858 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
2860 return __page_symlink(inode
, symname
, len
,
2861 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
2864 const struct inode_operations page_symlink_inode_operations
= {
2865 .readlink
= generic_readlink
,
2866 .follow_link
= page_follow_link_light
,
2867 .put_link
= page_put_link
,
2870 EXPORT_SYMBOL(user_path_at
);
2871 EXPORT_SYMBOL(follow_down
);
2872 EXPORT_SYMBOL(follow_up
);
2873 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
2874 EXPORT_SYMBOL(getname
);
2875 EXPORT_SYMBOL(lock_rename
);
2876 EXPORT_SYMBOL(lookup_one_len
);
2877 EXPORT_SYMBOL(page_follow_link_light
);
2878 EXPORT_SYMBOL(page_put_link
);
2879 EXPORT_SYMBOL(page_readlink
);
2880 EXPORT_SYMBOL(__page_symlink
);
2881 EXPORT_SYMBOL(page_symlink
);
2882 EXPORT_SYMBOL(page_symlink_inode_operations
);
2883 EXPORT_SYMBOL(path_lookup
);
2884 EXPORT_SYMBOL(kern_path
);
2885 EXPORT_SYMBOL(vfs_path_lookup
);
2886 EXPORT_SYMBOL(inode_permission
);
2887 EXPORT_SYMBOL(file_permission
);
2888 EXPORT_SYMBOL(unlock_rename
);
2889 EXPORT_SYMBOL(vfs_create
);
2890 EXPORT_SYMBOL(vfs_follow_link
);
2891 EXPORT_SYMBOL(vfs_link
);
2892 EXPORT_SYMBOL(vfs_mkdir
);
2893 EXPORT_SYMBOL(vfs_mknod
);
2894 EXPORT_SYMBOL(generic_permission
);
2895 EXPORT_SYMBOL(vfs_readlink
);
2896 EXPORT_SYMBOL(vfs_rename
);
2897 EXPORT_SYMBOL(vfs_rmdir
);
2898 EXPORT_SYMBOL(vfs_symlink
);
2899 EXPORT_SYMBOL(vfs_unlink
);
2900 EXPORT_SYMBOL(dentry_unhash
);
2901 EXPORT_SYMBOL(generic_readlink
);
2903 /* to be mentioned only in INIT_TASK */
2904 struct fs_struct init_fs
= {
2905 .count
= ATOMIC_INIT(1),
2906 .lock
= __RW_LOCK_UNLOCKED(init_fs
.lock
),