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/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <asm/uaccess.h>
36 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
38 /* [Feb-1997 T. Schoebel-Theuer]
39 * Fundamental changes in the pathname lookup mechanisms (namei)
40 * were necessary because of omirr. The reason is that omirr needs
41 * to know the _real_ pathname, not the user-supplied one, in case
42 * of symlinks (and also when transname replacements occur).
44 * The new code replaces the old recursive symlink resolution with
45 * an iterative one (in case of non-nested symlink chains). It does
46 * this with calls to <fs>_follow_link().
47 * As a side effect, dir_namei(), _namei() and follow_link() are now
48 * replaced with a single function lookup_dentry() that can handle all
49 * the special cases of the former code.
51 * With the new dcache, the pathname is stored at each inode, at least as
52 * long as the refcount of the inode is positive. As a side effect, the
53 * size of the dcache depends on the inode cache and thus is dynamic.
55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
56 * resolution to correspond with current state of the code.
58 * Note that the symlink resolution is not *completely* iterative.
59 * There is still a significant amount of tail- and mid- recursion in
60 * the algorithm. Also, note that <fs>_readlink() is not used in
61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
62 * may return different results than <fs>_follow_link(). Many virtual
63 * filesystems (including /proc) exhibit this behavior.
66 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
68 * and the name already exists in form of a symlink, try to create the new
69 * name indicated by the symlink. The old code always complained that the
70 * name already exists, due to not following the symlink even if its target
71 * is nonexistent. The new semantics affects also mknod() and link() when
72 * the name is a symlink pointing to a non-existant name.
74 * I don't know which semantics is the right one, since I have no access
75 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
77 * "old" one. Personally, I think the new semantics is much more logical.
78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
79 * file does succeed in both HP-UX and SunOs, but not in Solaris
80 * and in the old Linux semantics.
83 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
84 * semantics. See the comments in "open_namei" and "do_link" below.
86 * [10-Sep-98 Alan Modra] Another symlink change.
89 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
90 * inside the path - always follow.
91 * in the last component in creation/removal/renaming - never follow.
92 * if LOOKUP_FOLLOW passed - follow.
93 * if the pathname has trailing slashes - follow.
94 * otherwise - don't follow.
95 * (applied in that order).
97 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
98 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
99 * During the 2.4 we need to fix the userland stuff depending on it -
100 * hopefully we will be able to get rid of that wart in 2.5. So far only
101 * XEmacs seems to be relying on it...
104 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
105 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
106 * any extra contention...
109 static int __link_path_walk(const char *name
, struct nameidata
*nd
);
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user
*filename
, char *page
)
121 unsigned long len
= PATH_MAX
;
123 if (!segment_eq(get_fs(), KERNEL_DS
)) {
124 if ((unsigned long) filename
>= TASK_SIZE
)
126 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
127 len
= TASK_SIZE
- (unsigned long) filename
;
130 retval
= strncpy_from_user(page
, filename
, len
);
134 return -ENAMETOOLONG
;
140 char * getname(const char __user
* filename
)
144 result
= ERR_PTR(-ENOMEM
);
147 int retval
= do_getname(filename
, tmp
);
152 result
= ERR_PTR(retval
);
155 audit_getname(result
);
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name
)
162 if (unlikely(!audit_dummy_context()))
167 EXPORT_SYMBOL(putname
);
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
182 int generic_permission(struct inode
*inode
, int mask
,
183 int (*check_acl
)(struct inode
*inode
, int mask
))
185 umode_t mode
= inode
->i_mode
;
187 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
189 if (current_fsuid() == inode
->i_uid
)
192 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
193 int error
= check_acl(inode
, mask
);
194 if (error
== -EACCES
)
195 goto check_capabilities
;
196 else if (error
!= -EAGAIN
)
200 if (in_group_p(inode
->i_gid
))
205 * If the DACs are ok we don't need any capability check.
207 if ((mask
& ~mode
) == 0)
212 * Read/write DACs are always overridable.
213 * Executable DACs are overridable if at least one exec bit is set.
215 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
216 if (capable(CAP_DAC_OVERRIDE
))
220 * Searching includes executable on directories, else just read.
222 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
223 if (capable(CAP_DAC_READ_SEARCH
))
230 * inode_permission - check for access rights to a given inode
231 * @inode: inode to check permission on
232 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
234 * Used to check for read/write/execute permissions on an inode.
235 * We use "fsuid" for this, letting us set arbitrary permissions
236 * for filesystem access without changing the "normal" uids which
237 * are used for other things.
239 int inode_permission(struct inode
*inode
, int mask
)
243 if (mask
& MAY_WRITE
) {
244 umode_t mode
= inode
->i_mode
;
247 * Nobody gets write access to a read-only fs.
249 if (IS_RDONLY(inode
) &&
250 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
254 * Nobody gets write access to an immutable file.
256 if (IS_IMMUTABLE(inode
))
260 if (inode
->i_op
&& inode
->i_op
->permission
)
261 retval
= inode
->i_op
->permission(inode
, mask
);
263 retval
= generic_permission(inode
, mask
, NULL
);
268 retval
= devcgroup_inode_permission(inode
, mask
);
272 return security_inode_permission(inode
,
273 mask
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
|MAY_APPEND
));
277 * file_permission - check for additional access rights to a given file
278 * @file: file to check access rights for
279 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
281 * Used to check for read/write/execute permissions on an already opened
285 * Do not use this function in new code. All access checks should
286 * be done using inode_permission().
288 int file_permission(struct file
*file
, int mask
)
290 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
294 * get_write_access() gets write permission for a file.
295 * put_write_access() releases this write permission.
296 * This is used for regular files.
297 * We cannot support write (and maybe mmap read-write shared) accesses and
298 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
299 * can have the following values:
300 * 0: no writers, no VM_DENYWRITE mappings
301 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
302 * > 0: (i_writecount) users are writing to the file.
304 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
305 * except for the cases where we don't hold i_writecount yet. Then we need to
306 * use {get,deny}_write_access() - these functions check the sign and refuse
307 * to do the change if sign is wrong. Exclusion between them is provided by
308 * the inode->i_lock spinlock.
311 int get_write_access(struct inode
* inode
)
313 spin_lock(&inode
->i_lock
);
314 if (atomic_read(&inode
->i_writecount
) < 0) {
315 spin_unlock(&inode
->i_lock
);
318 atomic_inc(&inode
->i_writecount
);
319 spin_unlock(&inode
->i_lock
);
324 int deny_write_access(struct file
* file
)
326 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
328 spin_lock(&inode
->i_lock
);
329 if (atomic_read(&inode
->i_writecount
) > 0) {
330 spin_unlock(&inode
->i_lock
);
333 atomic_dec(&inode
->i_writecount
);
334 spin_unlock(&inode
->i_lock
);
340 * path_get - get a reference to a path
341 * @path: path to get the reference to
343 * Given a path increment the reference count to the dentry and the vfsmount.
345 void path_get(struct path
*path
)
350 EXPORT_SYMBOL(path_get
);
353 * path_put - put a reference to a path
354 * @path: path to put the reference to
356 * Given a path decrement the reference count to the dentry and the vfsmount.
358 void path_put(struct path
*path
)
363 EXPORT_SYMBOL(path_put
);
366 * release_open_intent - free up open intent resources
367 * @nd: pointer to nameidata
369 void release_open_intent(struct nameidata
*nd
)
371 if (nd
->intent
.open
.file
->f_path
.dentry
== NULL
)
372 put_filp(nd
->intent
.open
.file
);
374 fput(nd
->intent
.open
.file
);
377 static inline struct dentry
*
378 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
380 int status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
381 if (unlikely(status
<= 0)) {
383 * The dentry failed validation.
384 * If d_revalidate returned 0 attempt to invalidate
385 * the dentry otherwise d_revalidate is asking us
386 * to return a fail status.
389 if (!d_invalidate(dentry
)) {
395 dentry
= ERR_PTR(status
);
402 * Internal lookup() using the new generic dcache.
405 static struct dentry
* cached_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
407 struct dentry
* dentry
= __d_lookup(parent
, name
);
409 /* lockess __d_lookup may fail due to concurrent d_move()
410 * in some unrelated directory, so try with d_lookup
413 dentry
= d_lookup(parent
, name
);
415 if (dentry
&& dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
416 dentry
= do_revalidate(dentry
, nd
);
422 * Short-cut version of permission(), for calling by
423 * path_walk(), when dcache lock is held. Combines parts
424 * of permission() and generic_permission(), and tests ONLY for
425 * MAY_EXEC permission.
427 * If appropriate, check DAC only. If not appropriate, or
428 * short-cut DAC fails, then call permission() to do more
429 * complete permission check.
431 static int exec_permission_lite(struct inode
*inode
)
433 umode_t mode
= inode
->i_mode
;
435 if (inode
->i_op
&& inode
->i_op
->permission
)
438 if (current_fsuid() == inode
->i_uid
)
440 else if (in_group_p(inode
->i_gid
))
446 if ((inode
->i_mode
& S_IXUGO
) && capable(CAP_DAC_OVERRIDE
))
449 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_OVERRIDE
))
452 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_READ_SEARCH
))
457 return security_inode_permission(inode
, MAY_EXEC
);
461 * This is called when everything else fails, and we actually have
462 * to go to the low-level filesystem to find out what we should do..
464 * We get the directory semaphore, and after getting that we also
465 * make sure that nobody added the entry to the dcache in the meantime..
468 static struct dentry
* real_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
470 struct dentry
* result
;
471 struct inode
*dir
= parent
->d_inode
;
473 mutex_lock(&dir
->i_mutex
);
475 * First re-do the cached lookup just in case it was created
476 * while we waited for the directory semaphore..
478 * FIXME! This could use version numbering or similar to
479 * avoid unnecessary cache lookups.
481 * The "dcache_lock" is purely to protect the RCU list walker
482 * from concurrent renames at this point (we mustn't get false
483 * negatives from the RCU list walk here, unlike the optimistic
486 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
488 result
= d_lookup(parent
, name
);
490 struct dentry
*dentry
;
492 /* Don't create child dentry for a dead directory. */
493 result
= ERR_PTR(-ENOENT
);
497 dentry
= d_alloc(parent
, name
);
498 result
= ERR_PTR(-ENOMEM
);
500 result
= dir
->i_op
->lookup(dir
, dentry
, nd
);
507 mutex_unlock(&dir
->i_mutex
);
512 * Uhhuh! Nasty case: the cache was re-populated while
513 * we waited on the semaphore. Need to revalidate.
515 mutex_unlock(&dir
->i_mutex
);
516 if (result
->d_op
&& result
->d_op
->d_revalidate
) {
517 result
= do_revalidate(result
, nd
);
519 result
= ERR_PTR(-ENOENT
);
525 * Wrapper to retry pathname resolution whenever the underlying
526 * file system returns an ESTALE.
528 * Retry the whole path once, forcing real lookup requests
529 * instead of relying on the dcache.
531 static __always_inline
int link_path_walk(const char *name
, struct nameidata
*nd
)
533 struct path save
= nd
->path
;
536 /* make sure the stuff we saved doesn't go away */
539 result
= __link_path_walk(name
, nd
);
540 if (result
== -ESTALE
) {
541 /* nd->path had been dropped */
544 nd
->flags
|= LOOKUP_REVAL
;
545 result
= __link_path_walk(name
, nd
);
553 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
561 struct fs_struct
*fs
= current
->fs
;
565 read_lock(&fs
->lock
);
568 read_unlock(&fs
->lock
);
571 res
= link_path_walk(link
, nd
);
572 if (nd
->depth
|| res
|| nd
->last_type
!=LAST_NORM
)
575 * If it is an iterative symlinks resolution in open_namei() we
576 * have to copy the last component. And all that crap because of
577 * bloody create() on broken symlinks. Furrfu...
580 if (unlikely(!name
)) {
584 strcpy(name
, nd
->last
.name
);
585 nd
->last
.name
= name
;
589 return PTR_ERR(link
);
592 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
595 if (path
->mnt
!= nd
->path
.mnt
)
599 static inline void path_to_nameidata(struct path
*path
, struct nameidata
*nd
)
601 dput(nd
->path
.dentry
);
602 if (nd
->path
.mnt
!= path
->mnt
)
603 mntput(nd
->path
.mnt
);
604 nd
->path
.mnt
= path
->mnt
;
605 nd
->path
.dentry
= path
->dentry
;
608 static __always_inline
int __do_follow_link(struct path
*path
, struct nameidata
*nd
)
612 struct dentry
*dentry
= path
->dentry
;
614 touch_atime(path
->mnt
, dentry
);
615 nd_set_link(nd
, NULL
);
617 if (path
->mnt
!= nd
->path
.mnt
) {
618 path_to_nameidata(path
, nd
);
622 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
623 error
= PTR_ERR(cookie
);
624 if (!IS_ERR(cookie
)) {
625 char *s
= nd_get_link(nd
);
628 error
= __vfs_follow_link(nd
, s
);
629 if (dentry
->d_inode
->i_op
->put_link
)
630 dentry
->d_inode
->i_op
->put_link(dentry
, nd
, cookie
);
638 * This limits recursive symlink follows to 8, while
639 * limiting consecutive symlinks to 40.
641 * Without that kind of total limit, nasty chains of consecutive
642 * symlinks can cause almost arbitrarily long lookups.
644 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
647 if (current
->link_count
>= MAX_NESTED_LINKS
)
649 if (current
->total_link_count
>= 40)
651 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
653 err
= security_inode_follow_link(path
->dentry
, nd
);
656 current
->link_count
++;
657 current
->total_link_count
++;
659 err
= __do_follow_link(path
, nd
);
660 current
->link_count
--;
664 path_put_conditional(path
, nd
);
669 int follow_up(struct vfsmount
**mnt
, struct dentry
**dentry
)
671 struct vfsmount
*parent
;
672 struct dentry
*mountpoint
;
673 spin_lock(&vfsmount_lock
);
674 parent
=(*mnt
)->mnt_parent
;
675 if (parent
== *mnt
) {
676 spin_unlock(&vfsmount_lock
);
680 mountpoint
=dget((*mnt
)->mnt_mountpoint
);
681 spin_unlock(&vfsmount_lock
);
683 *dentry
= mountpoint
;
689 /* no need for dcache_lock, as serialization is taken care in
692 static int __follow_mount(struct path
*path
)
695 while (d_mountpoint(path
->dentry
)) {
696 struct vfsmount
*mounted
= lookup_mnt(path
->mnt
, path
->dentry
);
703 path
->dentry
= dget(mounted
->mnt_root
);
709 static void follow_mount(struct vfsmount
**mnt
, struct dentry
**dentry
)
711 while (d_mountpoint(*dentry
)) {
712 struct vfsmount
*mounted
= lookup_mnt(*mnt
, *dentry
);
718 *dentry
= dget(mounted
->mnt_root
);
722 /* no need for dcache_lock, as serialization is taken care in
725 int follow_down(struct vfsmount
**mnt
, struct dentry
**dentry
)
727 struct vfsmount
*mounted
;
729 mounted
= lookup_mnt(*mnt
, *dentry
);
734 *dentry
= dget(mounted
->mnt_root
);
740 static __always_inline
void follow_dotdot(struct nameidata
*nd
)
742 struct fs_struct
*fs
= current
->fs
;
745 struct vfsmount
*parent
;
746 struct dentry
*old
= nd
->path
.dentry
;
748 read_lock(&fs
->lock
);
749 if (nd
->path
.dentry
== fs
->root
.dentry
&&
750 nd
->path
.mnt
== fs
->root
.mnt
) {
751 read_unlock(&fs
->lock
);
754 read_unlock(&fs
->lock
);
755 spin_lock(&dcache_lock
);
756 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
757 nd
->path
.dentry
= dget(nd
->path
.dentry
->d_parent
);
758 spin_unlock(&dcache_lock
);
762 spin_unlock(&dcache_lock
);
763 spin_lock(&vfsmount_lock
);
764 parent
= nd
->path
.mnt
->mnt_parent
;
765 if (parent
== nd
->path
.mnt
) {
766 spin_unlock(&vfsmount_lock
);
770 nd
->path
.dentry
= dget(nd
->path
.mnt
->mnt_mountpoint
);
771 spin_unlock(&vfsmount_lock
);
773 mntput(nd
->path
.mnt
);
774 nd
->path
.mnt
= parent
;
776 follow_mount(&nd
->path
.mnt
, &nd
->path
.dentry
);
780 * It's more convoluted than I'd like it to be, but... it's still fairly
781 * small and for now I'd prefer to have fast path as straight as possible.
782 * It _is_ time-critical.
784 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
787 struct vfsmount
*mnt
= nd
->path
.mnt
;
788 struct dentry
*dentry
= __d_lookup(nd
->path
.dentry
, name
);
792 if (dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
793 goto need_revalidate
;
796 path
->dentry
= dentry
;
797 __follow_mount(path
);
801 dentry
= real_lookup(nd
->path
.dentry
, name
, nd
);
807 dentry
= do_revalidate(dentry
, nd
);
815 return PTR_ERR(dentry
);
820 * This is the basic name resolution function, turning a pathname into
821 * the final dentry. We expect 'base' to be positive and a directory.
823 * Returns 0 and nd will have valid dentry and mnt on success.
824 * Returns error and drops reference to input namei data on failure.
826 static int __link_path_walk(const char *name
, struct nameidata
*nd
)
831 unsigned int lookup_flags
= nd
->flags
;
838 inode
= nd
->path
.dentry
->d_inode
;
840 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
842 /* At this point we know we have a real path component. */
848 nd
->flags
|= LOOKUP_CONTINUE
;
849 err
= exec_permission_lite(inode
);
851 err
= inode_permission(nd
->path
.dentry
->d_inode
,
857 c
= *(const unsigned char *)name
;
859 hash
= init_name_hash();
862 hash
= partial_name_hash(c
, hash
);
863 c
= *(const unsigned char *)name
;
864 } while (c
&& (c
!= '/'));
865 this.len
= name
- (const char *) this.name
;
866 this.hash
= end_name_hash(hash
);
868 /* remove trailing slashes? */
871 while (*++name
== '/');
873 goto last_with_slashes
;
876 * "." and ".." are special - ".." especially so because it has
877 * to be able to know about the current root directory and
878 * parent relationships.
880 if (this.name
[0] == '.') switch (this.len
) {
884 if (this.name
[1] != '.')
887 inode
= nd
->path
.dentry
->d_inode
;
893 * See if the low-level filesystem might want
894 * to use its own hash..
896 if (nd
->path
.dentry
->d_op
&& nd
->path
.dentry
->d_op
->d_hash
) {
897 err
= nd
->path
.dentry
->d_op
->d_hash(nd
->path
.dentry
,
902 /* This does the actual lookups.. */
903 err
= do_lookup(nd
, &this, &next
);
908 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
;
927 path_to_nameidata(&next
, nd
);
929 if (!inode
->i_op
->lookup
)
932 /* here ends the main loop */
935 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
937 /* Clear LOOKUP_CONTINUE iff it was previously unset */
938 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
939 if (lookup_flags
& LOOKUP_PARENT
)
941 if (this.name
[0] == '.') switch (this.len
) {
945 if (this.name
[1] != '.')
948 inode
= nd
->path
.dentry
->d_inode
;
953 if (nd
->path
.dentry
->d_op
&& nd
->path
.dentry
->d_op
->d_hash
) {
954 err
= nd
->path
.dentry
->d_op
->d_hash(nd
->path
.dentry
,
959 err
= do_lookup(nd
, &this, &next
);
962 inode
= next
.dentry
->d_inode
;
963 if ((lookup_flags
& LOOKUP_FOLLOW
)
964 && inode
&& inode
->i_op
&& inode
->i_op
->follow_link
) {
965 err
= do_follow_link(&next
, nd
);
968 inode
= nd
->path
.dentry
->d_inode
;
970 path_to_nameidata(&next
, nd
);
974 if (lookup_flags
& LOOKUP_DIRECTORY
) {
976 if (!inode
->i_op
|| !inode
->i_op
->lookup
)
982 nd
->last_type
= LAST_NORM
;
983 if (this.name
[0] != '.')
986 nd
->last_type
= LAST_DOT
;
987 else if (this.len
== 2 && this.name
[1] == '.')
988 nd
->last_type
= LAST_DOTDOT
;
993 * We bypassed the ordinary revalidation routines.
994 * We may need to check the cached dentry for staleness.
996 if (nd
->path
.dentry
&& nd
->path
.dentry
->d_sb
&&
997 (nd
->path
.dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)) {
999 /* Note: we do not d_invalidate() */
1000 if (!nd
->path
.dentry
->d_op
->d_revalidate(
1001 nd
->path
.dentry
, nd
))
1007 path_put_conditional(&next
, nd
);
1010 path_put(&nd
->path
);
1015 static int path_walk(const char *name
, struct nameidata
*nd
)
1017 current
->total_link_count
= 0;
1018 return link_path_walk(name
, nd
);
1021 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1022 static int do_path_lookup(int dfd
, const char *name
,
1023 unsigned int flags
, struct nameidata
*nd
)
1028 struct fs_struct
*fs
= current
->fs
;
1030 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1035 read_lock(&fs
->lock
);
1036 nd
->path
= fs
->root
;
1037 path_get(&fs
->root
);
1038 read_unlock(&fs
->lock
);
1039 } else if (dfd
== AT_FDCWD
) {
1040 read_lock(&fs
->lock
);
1043 read_unlock(&fs
->lock
);
1045 struct dentry
*dentry
;
1047 file
= fget_light(dfd
, &fput_needed
);
1052 dentry
= file
->f_path
.dentry
;
1055 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1058 retval
= file_permission(file
, MAY_EXEC
);
1062 nd
->path
= file
->f_path
;
1063 path_get(&file
->f_path
);
1065 fput_light(file
, fput_needed
);
1068 retval
= path_walk(name
, nd
);
1069 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1070 nd
->path
.dentry
->d_inode
))
1071 audit_inode(name
, nd
->path
.dentry
);
1076 fput_light(file
, fput_needed
);
1080 int path_lookup(const char *name
, unsigned int flags
,
1081 struct nameidata
*nd
)
1083 return do_path_lookup(AT_FDCWD
, name
, flags
, nd
);
1086 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1088 struct nameidata nd
;
1089 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1096 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1097 * @dentry: pointer to dentry of the base directory
1098 * @mnt: pointer to vfs mount of the base directory
1099 * @name: pointer to file name
1100 * @flags: lookup flags
1101 * @nd: pointer to nameidata
1103 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1104 const char *name
, unsigned int flags
,
1105 struct nameidata
*nd
)
1109 /* same as do_path_lookup */
1110 nd
->last_type
= LAST_ROOT
;
1114 nd
->path
.dentry
= dentry
;
1116 path_get(&nd
->path
);
1118 retval
= path_walk(name
, nd
);
1119 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1120 nd
->path
.dentry
->d_inode
))
1121 audit_inode(name
, nd
->path
.dentry
);
1128 * path_lookup_open - lookup a file path with open intent
1129 * @dfd: the directory to use as base, or AT_FDCWD
1130 * @name: pointer to file name
1131 * @lookup_flags: lookup intent flags
1132 * @nd: pointer to nameidata
1133 * @open_flags: open intent flags
1135 int path_lookup_open(int dfd
, const char *name
, unsigned int lookup_flags
,
1136 struct nameidata
*nd
, int open_flags
)
1138 struct file
*filp
= get_empty_filp();
1143 nd
->intent
.open
.file
= filp
;
1144 nd
->intent
.open
.flags
= open_flags
;
1145 nd
->intent
.open
.create_mode
= 0;
1146 err
= do_path_lookup(dfd
, name
, lookup_flags
|LOOKUP_OPEN
, nd
);
1147 if (IS_ERR(nd
->intent
.open
.file
)) {
1149 err
= PTR_ERR(nd
->intent
.open
.file
);
1150 path_put(&nd
->path
);
1152 } else if (err
!= 0)
1153 release_open_intent(nd
);
1157 static struct dentry
*__lookup_hash(struct qstr
*name
,
1158 struct dentry
*base
, struct nameidata
*nd
)
1160 struct dentry
*dentry
;
1161 struct inode
*inode
;
1164 inode
= base
->d_inode
;
1167 * See if the low-level filesystem might want
1168 * to use its own hash..
1170 if (base
->d_op
&& base
->d_op
->d_hash
) {
1171 err
= base
->d_op
->d_hash(base
, name
);
1172 dentry
= ERR_PTR(err
);
1177 dentry
= cached_lookup(base
, name
, nd
);
1181 /* Don't create child dentry for a dead directory. */
1182 dentry
= ERR_PTR(-ENOENT
);
1183 if (IS_DEADDIR(inode
))
1186 new = d_alloc(base
, name
);
1187 dentry
= ERR_PTR(-ENOMEM
);
1190 dentry
= inode
->i_op
->lookup(inode
, new, nd
);
1201 * Restricted form of lookup. Doesn't follow links, single-component only,
1202 * needs parent already locked. Doesn't follow mounts.
1205 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1209 err
= inode_permission(nd
->path
.dentry
->d_inode
, MAY_EXEC
);
1211 return ERR_PTR(err
);
1212 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1215 static int __lookup_one_len(const char *name
, struct qstr
*this,
1216 struct dentry
*base
, int len
)
1226 hash
= init_name_hash();
1228 c
= *(const unsigned char *)name
++;
1229 if (c
== '/' || c
== '\0')
1231 hash
= partial_name_hash(c
, hash
);
1233 this->hash
= end_name_hash(hash
);
1238 * lookup_one_len - filesystem helper to lookup single pathname component
1239 * @name: pathname component to lookup
1240 * @base: base directory to lookup from
1241 * @len: maximum length @len should be interpreted to
1243 * Note that this routine is purely a helper for filesystem usage and should
1244 * not be called by generic code. Also note that by using this function the
1245 * nameidata argument is passed to the filesystem methods and a filesystem
1246 * using this helper needs to be prepared for that.
1248 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1253 err
= __lookup_one_len(name
, &this, base
, len
);
1255 return ERR_PTR(err
);
1257 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
1259 return ERR_PTR(err
);
1260 return __lookup_hash(&this, base
, NULL
);
1264 * lookup_one_noperm - bad hack for sysfs
1265 * @name: pathname component to lookup
1266 * @base: base directory to lookup from
1268 * This is a variant of lookup_one_len that doesn't perform any permission
1269 * checks. It's a horrible hack to work around the braindead sysfs
1270 * architecture and should not be used anywhere else.
1272 * DON'T USE THIS FUNCTION EVER, thanks.
1274 struct dentry
*lookup_one_noperm(const char *name
, struct dentry
*base
)
1279 err
= __lookup_one_len(name
, &this, base
, strlen(name
));
1281 return ERR_PTR(err
);
1282 return __lookup_hash(&this, base
, NULL
);
1285 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1288 struct nameidata nd
;
1289 char *tmp
= getname(name
);
1290 int err
= PTR_ERR(tmp
);
1293 BUG_ON(flags
& LOOKUP_PARENT
);
1295 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1303 static int user_path_parent(int dfd
, const char __user
*path
,
1304 struct nameidata
*nd
, char **name
)
1306 char *s
= getname(path
);
1312 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1322 * It's inline, so penalty for filesystems that don't use sticky bit is
1325 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1327 uid_t fsuid
= current_fsuid();
1329 if (!(dir
->i_mode
& S_ISVTX
))
1331 if (inode
->i_uid
== fsuid
)
1333 if (dir
->i_uid
== fsuid
)
1335 return !capable(CAP_FOWNER
);
1339 * Check whether we can remove a link victim from directory dir, check
1340 * whether the type of victim is right.
1341 * 1. We can't do it if dir is read-only (done in permission())
1342 * 2. We should have write and exec permissions on dir
1343 * 3. We can't remove anything from append-only dir
1344 * 4. We can't do anything with immutable dir (done in permission())
1345 * 5. If the sticky bit on dir is set we should either
1346 * a. be owner of dir, or
1347 * b. be owner of victim, or
1348 * c. have CAP_FOWNER capability
1349 * 6. If the victim is append-only or immutable we can't do antyhing with
1350 * links pointing to it.
1351 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1352 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1353 * 9. We can't remove a root or mountpoint.
1354 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1355 * nfs_async_unlink().
1357 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1361 if (!victim
->d_inode
)
1364 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1365 audit_inode_child(victim
->d_name
.name
, victim
, dir
);
1367 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1372 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1373 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1376 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1378 if (IS_ROOT(victim
))
1380 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1382 if (IS_DEADDIR(dir
))
1384 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1389 /* Check whether we can create an object with dentry child in directory
1391 * 1. We can't do it if child already exists (open has special treatment for
1392 * this case, but since we are inlined it's OK)
1393 * 2. We can't do it if dir is read-only (done in permission())
1394 * 3. We should have write and exec permissions on dir
1395 * 4. We can't do it if dir is immutable (done in permission())
1397 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1401 if (IS_DEADDIR(dir
))
1403 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1407 * O_DIRECTORY translates into forcing a directory lookup.
1409 static inline int lookup_flags(unsigned int f
)
1411 unsigned long retval
= LOOKUP_FOLLOW
;
1414 retval
&= ~LOOKUP_FOLLOW
;
1416 if (f
& O_DIRECTORY
)
1417 retval
|= LOOKUP_DIRECTORY
;
1423 * p1 and p2 should be directories on the same fs.
1425 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1430 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1434 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1436 p
= d_ancestor(p2
, p1
);
1438 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1439 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1443 p
= d_ancestor(p1
, p2
);
1445 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1446 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1450 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1451 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1455 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1457 mutex_unlock(&p1
->d_inode
->i_mutex
);
1459 mutex_unlock(&p2
->d_inode
->i_mutex
);
1460 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1464 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1465 struct nameidata
*nd
)
1467 int error
= may_create(dir
, dentry
);
1472 if (!dir
->i_op
|| !dir
->i_op
->create
)
1473 return -EACCES
; /* shouldn't it be ENOSYS? */
1476 error
= security_inode_create(dir
, dentry
, mode
);
1480 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1482 fsnotify_create(dir
, dentry
);
1486 int may_open(struct path
*path
, int acc_mode
, int flag
)
1488 struct dentry
*dentry
= path
->dentry
;
1489 struct inode
*inode
= dentry
->d_inode
;
1495 if (S_ISLNK(inode
->i_mode
))
1498 if (S_ISDIR(inode
->i_mode
) && (acc_mode
& MAY_WRITE
))
1502 * FIFO's, sockets and device files are special: they don't
1503 * actually live on the filesystem itself, and as such you
1504 * can write to them even if the filesystem is read-only.
1506 if (S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
1508 } else if (S_ISBLK(inode
->i_mode
) || S_ISCHR(inode
->i_mode
)) {
1509 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
1515 error
= inode_permission(inode
, acc_mode
);
1519 * An append-only file must be opened in append mode for writing.
1521 if (IS_APPEND(inode
)) {
1522 if ((flag
& FMODE_WRITE
) && !(flag
& O_APPEND
))
1528 /* O_NOATIME can only be set by the owner or superuser */
1529 if (flag
& O_NOATIME
)
1530 if (!is_owner_or_cap(inode
))
1534 * Ensure there are no outstanding leases on the file.
1536 error
= break_lease(inode
, flag
);
1540 if (flag
& O_TRUNC
) {
1541 error
= get_write_access(inode
);
1546 * Refuse to truncate files with mandatory locks held on them.
1548 error
= locks_verify_locked(inode
);
1550 error
= security_path_truncate(path
, 0,
1551 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
);
1555 error
= do_truncate(dentry
, 0,
1556 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
1559 put_write_access(inode
);
1563 if (flag
& FMODE_WRITE
)
1570 * Be careful about ever adding any more callers of this
1571 * function. Its flags must be in the namei format, not
1572 * what get passed to sys_open().
1574 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
1578 struct dentry
*dir
= nd
->path
.dentry
;
1580 if (!IS_POSIXACL(dir
->d_inode
))
1581 mode
&= ~current
->fs
->umask
;
1582 error
= security_path_mknod(&nd
->path
, path
->dentry
, mode
, 0);
1585 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
1587 mutex_unlock(&dir
->d_inode
->i_mutex
);
1588 dput(nd
->path
.dentry
);
1589 nd
->path
.dentry
= path
->dentry
;
1592 /* Don't check for write permission, don't truncate */
1593 return may_open(&nd
->path
, 0, flag
& ~O_TRUNC
);
1597 * Note that while the flag value (low two bits) for sys_open means:
1602 * it is changed into
1603 * 00 - no permissions needed
1604 * 01 - read-permission
1605 * 10 - write-permission
1607 * for the internal routines (ie open_namei()/follow_link() etc)
1608 * This is more logical, and also allows the 00 "no perm needed"
1609 * to be used for symlinks (where the permissions are checked
1613 static inline int open_to_namei_flags(int flag
)
1615 if ((flag
+1) & O_ACCMODE
)
1620 static int open_will_write_to_fs(int flag
, struct inode
*inode
)
1623 * We'll never write to the fs underlying
1626 if (special_file(inode
->i_mode
))
1628 return (flag
& O_TRUNC
);
1632 * Note that the low bits of the passed in "open_flag"
1633 * are not the same as in the local variable "flag". See
1634 * open_to_namei_flags() for more details.
1636 struct file
*do_filp_open(int dfd
, const char *pathname
,
1637 int open_flag
, int mode
)
1640 struct nameidata nd
;
1641 int acc_mode
, error
;
1646 int flag
= open_to_namei_flags(open_flag
);
1648 acc_mode
= MAY_OPEN
| ACC_MODE(flag
);
1650 /* O_TRUNC implies we need access checks for write permissions */
1652 acc_mode
|= MAY_WRITE
;
1654 /* Allow the LSM permission hook to distinguish append
1655 access from general write access. */
1656 if (flag
& O_APPEND
)
1657 acc_mode
|= MAY_APPEND
;
1660 * The simplest case - just a plain lookup.
1662 if (!(flag
& O_CREAT
)) {
1663 error
= path_lookup_open(dfd
, pathname
, lookup_flags(flag
),
1666 return ERR_PTR(error
);
1671 * Create - we need to know the parent.
1673 error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
, &nd
);
1675 return ERR_PTR(error
);
1678 * We have the parent and last component. First of all, check
1679 * that we are not asked to creat(2) an obvious directory - that
1683 if (nd
.last_type
!= LAST_NORM
|| nd
.last
.name
[nd
.last
.len
])
1687 filp
= get_empty_filp();
1690 nd
.intent
.open
.file
= filp
;
1691 nd
.intent
.open
.flags
= flag
;
1692 nd
.intent
.open
.create_mode
= mode
;
1693 dir
= nd
.path
.dentry
;
1694 nd
.flags
&= ~LOOKUP_PARENT
;
1695 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_OPEN
;
1697 nd
.flags
|= LOOKUP_EXCL
;
1698 mutex_lock(&dir
->d_inode
->i_mutex
);
1699 path
.dentry
= lookup_hash(&nd
);
1700 path
.mnt
= nd
.path
.mnt
;
1703 error
= PTR_ERR(path
.dentry
);
1704 if (IS_ERR(path
.dentry
)) {
1705 mutex_unlock(&dir
->d_inode
->i_mutex
);
1709 if (IS_ERR(nd
.intent
.open
.file
)) {
1710 error
= PTR_ERR(nd
.intent
.open
.file
);
1711 goto exit_mutex_unlock
;
1714 /* Negative dentry, just create the file */
1715 if (!path
.dentry
->d_inode
) {
1717 * This write is needed to ensure that a
1718 * ro->rw transition does not occur between
1719 * the time when the file is created and when
1720 * a permanent write count is taken through
1721 * the 'struct file' in nameidata_to_filp().
1723 error
= mnt_want_write(nd
.path
.mnt
);
1725 goto exit_mutex_unlock
;
1726 error
= __open_namei_create(&nd
, &path
, flag
, mode
);
1728 mnt_drop_write(nd
.path
.mnt
);
1731 filp
= nameidata_to_filp(&nd
, open_flag
);
1732 mnt_drop_write(nd
.path
.mnt
);
1737 * It already exists.
1739 mutex_unlock(&dir
->d_inode
->i_mutex
);
1740 audit_inode(pathname
, path
.dentry
);
1746 if (__follow_mount(&path
)) {
1748 if (flag
& O_NOFOLLOW
)
1753 if (!path
.dentry
->d_inode
)
1755 if (path
.dentry
->d_inode
->i_op
&& path
.dentry
->d_inode
->i_op
->follow_link
)
1758 path_to_nameidata(&path
, &nd
);
1760 if (path
.dentry
->d_inode
&& S_ISDIR(path
.dentry
->d_inode
->i_mode
))
1765 * 1. may_open() truncates a file
1766 * 2. a rw->ro mount transition occurs
1767 * 3. nameidata_to_filp() fails due to
1769 * That would be inconsistent, and should
1770 * be avoided. Taking this mnt write here
1771 * ensures that (2) can not occur.
1773 will_write
= open_will_write_to_fs(flag
, nd
.path
.dentry
->d_inode
);
1775 error
= mnt_want_write(nd
.path
.mnt
);
1779 error
= may_open(&nd
.path
, acc_mode
, flag
);
1782 mnt_drop_write(nd
.path
.mnt
);
1785 filp
= nameidata_to_filp(&nd
, open_flag
);
1787 * It is now safe to drop the mnt write
1788 * because the filp has had a write taken
1792 mnt_drop_write(nd
.path
.mnt
);
1796 mutex_unlock(&dir
->d_inode
->i_mutex
);
1798 path_put_conditional(&path
, &nd
);
1800 if (!IS_ERR(nd
.intent
.open
.file
))
1801 release_open_intent(&nd
);
1804 return ERR_PTR(error
);
1808 if (flag
& O_NOFOLLOW
)
1811 * This is subtle. Instead of calling do_follow_link() we do the
1812 * thing by hands. The reason is that this way we have zero link_count
1813 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1814 * After that we have the parent and last component, i.e.
1815 * we are in the same situation as after the first path_walk().
1816 * Well, almost - if the last component is normal we get its copy
1817 * stored in nd->last.name and we will have to putname() it when we
1818 * are done. Procfs-like symlinks just set LAST_BIND.
1820 nd
.flags
|= LOOKUP_PARENT
;
1821 error
= security_inode_follow_link(path
.dentry
, &nd
);
1824 error
= __do_follow_link(&path
, &nd
);
1826 /* Does someone understand code flow here? Or it is only
1827 * me so stupid? Anathema to whoever designed this non-sense
1828 * with "intent.open".
1830 release_open_intent(&nd
);
1831 return ERR_PTR(error
);
1833 nd
.flags
&= ~LOOKUP_PARENT
;
1834 if (nd
.last_type
== LAST_BIND
)
1837 if (nd
.last_type
!= LAST_NORM
)
1839 if (nd
.last
.name
[nd
.last
.len
]) {
1840 __putname(nd
.last
.name
);
1845 __putname(nd
.last
.name
);
1848 dir
= nd
.path
.dentry
;
1849 mutex_lock(&dir
->d_inode
->i_mutex
);
1850 path
.dentry
= lookup_hash(&nd
);
1851 path
.mnt
= nd
.path
.mnt
;
1852 __putname(nd
.last
.name
);
1857 * filp_open - open file and return file pointer
1859 * @filename: path to open
1860 * @flags: open flags as per the open(2) second argument
1861 * @mode: mode for the new file if O_CREAT is set, else ignored
1863 * This is the helper to open a file from kernelspace if you really
1864 * have to. But in generally you should not do this, so please move
1865 * along, nothing to see here..
1867 struct file
*filp_open(const char *filename
, int flags
, int mode
)
1869 return do_filp_open(AT_FDCWD
, filename
, flags
, mode
);
1871 EXPORT_SYMBOL(filp_open
);
1874 * lookup_create - lookup a dentry, creating it if it doesn't exist
1875 * @nd: nameidata info
1876 * @is_dir: directory flag
1878 * Simple function to lookup and return a dentry and create it
1879 * if it doesn't exist. Is SMP-safe.
1881 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1883 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
1885 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
1887 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1889 * Yucky last component or no last component at all?
1890 * (foo/., foo/.., /////)
1892 if (nd
->last_type
!= LAST_NORM
)
1894 nd
->flags
&= ~LOOKUP_PARENT
;
1895 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
1896 nd
->intent
.open
.flags
= O_EXCL
;
1899 * Do the final lookup.
1901 dentry
= lookup_hash(nd
);
1905 if (dentry
->d_inode
)
1908 * Special case - lookup gave negative, but... we had foo/bar/
1909 * From the vfs_mknod() POV we just have a negative dentry -
1910 * all is fine. Let's be bastards - you had / on the end, you've
1911 * been asking for (non-existent) directory. -ENOENT for you.
1913 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
1915 dentry
= ERR_PTR(-ENOENT
);
1920 dentry
= ERR_PTR(-EEXIST
);
1924 EXPORT_SYMBOL_GPL(lookup_create
);
1926 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1928 int error
= may_create(dir
, dentry
);
1933 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
1936 if (!dir
->i_op
|| !dir
->i_op
->mknod
)
1939 error
= devcgroup_inode_mknod(mode
, dev
);
1943 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
1948 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
1950 fsnotify_create(dir
, dentry
);
1954 static int may_mknod(mode_t mode
)
1956 switch (mode
& S_IFMT
) {
1962 case 0: /* zero mode translates to S_IFREG */
1971 asmlinkage
long sys_mknodat(int dfd
, const char __user
*filename
, int mode
,
1976 struct dentry
*dentry
;
1977 struct nameidata nd
;
1982 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
1986 dentry
= lookup_create(&nd
, 0);
1987 if (IS_ERR(dentry
)) {
1988 error
= PTR_ERR(dentry
);
1991 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
1992 mode
&= ~current
->fs
->umask
;
1993 error
= may_mknod(mode
);
1996 error
= mnt_want_write(nd
.path
.mnt
);
1999 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2001 goto out_drop_write
;
2002 switch (mode
& S_IFMT
) {
2003 case 0: case S_IFREG
:
2004 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2006 case S_IFCHR
: case S_IFBLK
:
2007 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2008 new_decode_dev(dev
));
2010 case S_IFIFO
: case S_IFSOCK
:
2011 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2015 mnt_drop_write(nd
.path
.mnt
);
2019 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2026 asmlinkage
long sys_mknod(const char __user
*filename
, int mode
, unsigned dev
)
2028 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2031 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2033 int error
= may_create(dir
, dentry
);
2038 if (!dir
->i_op
|| !dir
->i_op
->mkdir
)
2041 mode
&= (S_IRWXUGO
|S_ISVTX
);
2042 error
= security_inode_mkdir(dir
, dentry
, mode
);
2047 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2049 fsnotify_mkdir(dir
, dentry
);
2053 asmlinkage
long sys_mkdirat(int dfd
, const char __user
*pathname
, int mode
)
2057 struct dentry
*dentry
;
2058 struct nameidata nd
;
2060 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2064 dentry
= lookup_create(&nd
, 1);
2065 error
= PTR_ERR(dentry
);
2069 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2070 mode
&= ~current
->fs
->umask
;
2071 error
= mnt_want_write(nd
.path
.mnt
);
2074 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2076 goto out_drop_write
;
2077 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2079 mnt_drop_write(nd
.path
.mnt
);
2083 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2090 asmlinkage
long sys_mkdir(const char __user
*pathname
, int mode
)
2092 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2096 * We try to drop the dentry early: we should have
2097 * a usage count of 2 if we're the only user of this
2098 * dentry, and if that is true (possibly after pruning
2099 * the dcache), then we drop the dentry now.
2101 * A low-level filesystem can, if it choses, legally
2104 * if (!d_unhashed(dentry))
2107 * if it cannot handle the case of removing a directory
2108 * that is still in use by something else..
2110 void dentry_unhash(struct dentry
*dentry
)
2113 shrink_dcache_parent(dentry
);
2114 spin_lock(&dcache_lock
);
2115 spin_lock(&dentry
->d_lock
);
2116 if (atomic_read(&dentry
->d_count
) == 2)
2118 spin_unlock(&dentry
->d_lock
);
2119 spin_unlock(&dcache_lock
);
2122 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2124 int error
= may_delete(dir
, dentry
, 1);
2129 if (!dir
->i_op
|| !dir
->i_op
->rmdir
)
2134 mutex_lock(&dentry
->d_inode
->i_mutex
);
2135 dentry_unhash(dentry
);
2136 if (d_mountpoint(dentry
))
2139 error
= security_inode_rmdir(dir
, dentry
);
2141 error
= dir
->i_op
->rmdir(dir
, dentry
);
2143 dentry
->d_inode
->i_flags
|= S_DEAD
;
2146 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2155 static long do_rmdir(int dfd
, const char __user
*pathname
)
2159 struct dentry
*dentry
;
2160 struct nameidata nd
;
2162 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2166 switch(nd
.last_type
) {
2178 nd
.flags
&= ~LOOKUP_PARENT
;
2180 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2181 dentry
= lookup_hash(&nd
);
2182 error
= PTR_ERR(dentry
);
2185 error
= mnt_want_write(nd
.path
.mnt
);
2188 error
= security_path_rmdir(&nd
.path
, dentry
);
2191 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2193 mnt_drop_write(nd
.path
.mnt
);
2197 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2204 asmlinkage
long sys_rmdir(const char __user
*pathname
)
2206 return do_rmdir(AT_FDCWD
, pathname
);
2209 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2211 int error
= may_delete(dir
, dentry
, 0);
2216 if (!dir
->i_op
|| !dir
->i_op
->unlink
)
2221 mutex_lock(&dentry
->d_inode
->i_mutex
);
2222 if (d_mountpoint(dentry
))
2225 error
= security_inode_unlink(dir
, dentry
);
2227 error
= dir
->i_op
->unlink(dir
, dentry
);
2229 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2231 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2232 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2233 fsnotify_link_count(dentry
->d_inode
);
2241 * Make sure that the actual truncation of the file will occur outside its
2242 * directory's i_mutex. Truncate can take a long time if there is a lot of
2243 * writeout happening, and we don't want to prevent access to the directory
2244 * while waiting on the I/O.
2246 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2250 struct dentry
*dentry
;
2251 struct nameidata nd
;
2252 struct inode
*inode
= NULL
;
2254 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2259 if (nd
.last_type
!= LAST_NORM
)
2262 nd
.flags
&= ~LOOKUP_PARENT
;
2264 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2265 dentry
= lookup_hash(&nd
);
2266 error
= PTR_ERR(dentry
);
2267 if (!IS_ERR(dentry
)) {
2268 /* Why not before? Because we want correct error value */
2269 if (nd
.last
.name
[nd
.last
.len
])
2271 inode
= dentry
->d_inode
;
2273 atomic_inc(&inode
->i_count
);
2274 error
= mnt_want_write(nd
.path
.mnt
);
2277 error
= security_path_unlink(&nd
.path
, dentry
);
2280 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2282 mnt_drop_write(nd
.path
.mnt
);
2286 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2288 iput(inode
); /* truncate the inode here */
2295 error
= !dentry
->d_inode
? -ENOENT
:
2296 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2300 asmlinkage
long sys_unlinkat(int dfd
, const char __user
*pathname
, int flag
)
2302 if ((flag
& ~AT_REMOVEDIR
) != 0)
2305 if (flag
& AT_REMOVEDIR
)
2306 return do_rmdir(dfd
, pathname
);
2308 return do_unlinkat(dfd
, pathname
);
2311 asmlinkage
long sys_unlink(const char __user
*pathname
)
2313 return do_unlinkat(AT_FDCWD
, pathname
);
2316 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2318 int error
= may_create(dir
, dentry
);
2323 if (!dir
->i_op
|| !dir
->i_op
->symlink
)
2326 error
= security_inode_symlink(dir
, dentry
, oldname
);
2331 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2333 fsnotify_create(dir
, dentry
);
2337 asmlinkage
long sys_symlinkat(const char __user
*oldname
,
2338 int newdfd
, const char __user
*newname
)
2343 struct dentry
*dentry
;
2344 struct nameidata nd
;
2346 from
= getname(oldname
);
2348 return PTR_ERR(from
);
2350 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2354 dentry
= lookup_create(&nd
, 0);
2355 error
= PTR_ERR(dentry
);
2359 error
= mnt_want_write(nd
.path
.mnt
);
2362 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2364 goto out_drop_write
;
2365 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2367 mnt_drop_write(nd
.path
.mnt
);
2371 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2379 asmlinkage
long sys_symlink(const char __user
*oldname
, const char __user
*newname
)
2381 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2384 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2386 struct inode
*inode
= old_dentry
->d_inode
;
2392 error
= may_create(dir
, new_dentry
);
2396 if (dir
->i_sb
!= inode
->i_sb
)
2400 * A link to an append-only or immutable file cannot be created.
2402 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2404 if (!dir
->i_op
|| !dir
->i_op
->link
)
2406 if (S_ISDIR(inode
->i_mode
))
2409 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2413 mutex_lock(&inode
->i_mutex
);
2415 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2416 mutex_unlock(&inode
->i_mutex
);
2418 fsnotify_link(dir
, inode
, new_dentry
);
2423 * Hardlinks are often used in delicate situations. We avoid
2424 * security-related surprises by not following symlinks on the
2427 * We don't follow them on the oldname either to be compatible
2428 * with linux 2.0, and to avoid hard-linking to directories
2429 * and other special files. --ADM
2431 asmlinkage
long sys_linkat(int olddfd
, const char __user
*oldname
,
2432 int newdfd
, const char __user
*newname
,
2435 struct dentry
*new_dentry
;
2436 struct nameidata nd
;
2437 struct path old_path
;
2441 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
2444 error
= user_path_at(olddfd
, oldname
,
2445 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
2450 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2454 if (old_path
.mnt
!= nd
.path
.mnt
)
2456 new_dentry
= lookup_create(&nd
, 0);
2457 error
= PTR_ERR(new_dentry
);
2458 if (IS_ERR(new_dentry
))
2460 error
= mnt_want_write(nd
.path
.mnt
);
2463 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2465 goto out_drop_write
;
2466 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2468 mnt_drop_write(nd
.path
.mnt
);
2472 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2477 path_put(&old_path
);
2482 asmlinkage
long sys_link(const char __user
*oldname
, const char __user
*newname
)
2484 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2488 * The worst of all namespace operations - renaming directory. "Perverted"
2489 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2491 * a) we can get into loop creation. Check is done in is_subdir().
2492 * b) race potential - two innocent renames can create a loop together.
2493 * That's where 4.4 screws up. Current fix: serialization on
2494 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2496 * c) we have to lock _three_ objects - parents and victim (if it exists).
2497 * And that - after we got ->i_mutex on parents (until then we don't know
2498 * whether the target exists). Solution: try to be smart with locking
2499 * order for inodes. We rely on the fact that tree topology may change
2500 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2501 * move will be locked. Thus we can rank directories by the tree
2502 * (ancestors first) and rank all non-directories after them.
2503 * That works since everybody except rename does "lock parent, lookup,
2504 * lock child" and rename is under ->s_vfs_rename_mutex.
2505 * HOWEVER, it relies on the assumption that any object with ->lookup()
2506 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2507 * we'd better make sure that there's no link(2) for them.
2508 * d) some filesystems don't support opened-but-unlinked directories,
2509 * either because of layout or because they are not ready to deal with
2510 * all cases correctly. The latter will be fixed (taking this sort of
2511 * stuff into VFS), but the former is not going away. Solution: the same
2512 * trick as in rmdir().
2513 * e) conversion from fhandle to dentry may come in the wrong moment - when
2514 * we are removing the target. Solution: we will have to grab ->i_mutex
2515 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2516 * ->i_mutex on parents, which works but leads to some truely excessive
2519 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2520 struct inode
*new_dir
, struct dentry
*new_dentry
)
2523 struct inode
*target
;
2526 * If we are going to change the parent - check write permissions,
2527 * we'll need to flip '..'.
2529 if (new_dir
!= old_dir
) {
2530 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
2535 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2539 target
= new_dentry
->d_inode
;
2541 mutex_lock(&target
->i_mutex
);
2542 dentry_unhash(new_dentry
);
2544 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2547 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2550 target
->i_flags
|= S_DEAD
;
2551 mutex_unlock(&target
->i_mutex
);
2552 if (d_unhashed(new_dentry
))
2553 d_rehash(new_dentry
);
2557 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2558 d_move(old_dentry
,new_dentry
);
2562 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
2563 struct inode
*new_dir
, struct dentry
*new_dentry
)
2565 struct inode
*target
;
2568 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2573 target
= new_dentry
->d_inode
;
2575 mutex_lock(&target
->i_mutex
);
2576 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2579 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2581 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2582 d_move(old_dentry
, new_dentry
);
2585 mutex_unlock(&target
->i_mutex
);
2590 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2591 struct inode
*new_dir
, struct dentry
*new_dentry
)
2594 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2595 const char *old_name
;
2597 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
2600 error
= may_delete(old_dir
, old_dentry
, is_dir
);
2604 if (!new_dentry
->d_inode
)
2605 error
= may_create(new_dir
, new_dentry
);
2607 error
= may_delete(new_dir
, new_dentry
, is_dir
);
2611 if (!old_dir
->i_op
|| !old_dir
->i_op
->rename
)
2614 DQUOT_INIT(old_dir
);
2615 DQUOT_INIT(new_dir
);
2617 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
2620 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
2622 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
2624 const char *new_name
= old_dentry
->d_name
.name
;
2625 fsnotify_move(old_dir
, new_dir
, old_name
, new_name
, is_dir
,
2626 new_dentry
->d_inode
, old_dentry
);
2628 fsnotify_oldname_free(old_name
);
2633 asmlinkage
long sys_renameat(int olddfd
, const char __user
*oldname
,
2634 int newdfd
, const char __user
*newname
)
2636 struct dentry
*old_dir
, *new_dir
;
2637 struct dentry
*old_dentry
, *new_dentry
;
2638 struct dentry
*trap
;
2639 struct nameidata oldnd
, newnd
;
2644 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
2648 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
2653 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
2656 old_dir
= oldnd
.path
.dentry
;
2658 if (oldnd
.last_type
!= LAST_NORM
)
2661 new_dir
= newnd
.path
.dentry
;
2662 if (newnd
.last_type
!= LAST_NORM
)
2665 oldnd
.flags
&= ~LOOKUP_PARENT
;
2666 newnd
.flags
&= ~LOOKUP_PARENT
;
2667 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
2669 trap
= lock_rename(new_dir
, old_dir
);
2671 old_dentry
= lookup_hash(&oldnd
);
2672 error
= PTR_ERR(old_dentry
);
2673 if (IS_ERR(old_dentry
))
2675 /* source must exist */
2677 if (!old_dentry
->d_inode
)
2679 /* unless the source is a directory trailing slashes give -ENOTDIR */
2680 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
2682 if (oldnd
.last
.name
[oldnd
.last
.len
])
2684 if (newnd
.last
.name
[newnd
.last
.len
])
2687 /* source should not be ancestor of target */
2689 if (old_dentry
== trap
)
2691 new_dentry
= lookup_hash(&newnd
);
2692 error
= PTR_ERR(new_dentry
);
2693 if (IS_ERR(new_dentry
))
2695 /* target should not be an ancestor of source */
2697 if (new_dentry
== trap
)
2700 error
= mnt_want_write(oldnd
.path
.mnt
);
2703 error
= security_path_rename(&oldnd
.path
, old_dentry
,
2704 &newnd
.path
, new_dentry
);
2707 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
2708 new_dir
->d_inode
, new_dentry
);
2710 mnt_drop_write(oldnd
.path
.mnt
);
2716 unlock_rename(new_dir
, old_dir
);
2718 path_put(&newnd
.path
);
2721 path_put(&oldnd
.path
);
2727 asmlinkage
long sys_rename(const char __user
*oldname
, const char __user
*newname
)
2729 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
2732 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
2736 len
= PTR_ERR(link
);
2741 if (len
> (unsigned) buflen
)
2743 if (copy_to_user(buffer
, link
, len
))
2750 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2751 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2752 * using) it for any given inode is up to filesystem.
2754 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2756 struct nameidata nd
;
2761 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
2763 return PTR_ERR(cookie
);
2765 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
2766 if (dentry
->d_inode
->i_op
->put_link
)
2767 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
2771 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
2773 return __vfs_follow_link(nd
, link
);
2776 /* get the link contents into pagecache */
2777 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
2781 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
2782 page
= read_mapping_page(mapping
, 0, NULL
);
2787 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
2791 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2793 struct page
*page
= NULL
;
2794 char *s
= page_getlink(dentry
, &page
);
2795 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
2798 page_cache_release(page
);
2803 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
2805 struct page
*page
= NULL
;
2806 nd_set_link(nd
, page_getlink(dentry
, &page
));
2810 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2812 struct page
*page
= cookie
;
2816 page_cache_release(page
);
2820 int __page_symlink(struct inode
*inode
, const char *symname
, int len
,
2823 struct address_space
*mapping
= inode
->i_mapping
;
2830 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
2831 AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
2835 kaddr
= kmap_atomic(page
, KM_USER0
);
2836 memcpy(kaddr
, symname
, len
-1);
2837 kunmap_atomic(kaddr
, KM_USER0
);
2839 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
2846 mark_inode_dirty(inode
);
2852 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
2854 return __page_symlink(inode
, symname
, len
,
2855 mapping_gfp_mask(inode
->i_mapping
));
2858 const struct inode_operations page_symlink_inode_operations
= {
2859 .readlink
= generic_readlink
,
2860 .follow_link
= page_follow_link_light
,
2861 .put_link
= page_put_link
,
2864 EXPORT_SYMBOL(user_path_at
);
2865 EXPORT_SYMBOL(follow_down
);
2866 EXPORT_SYMBOL(follow_up
);
2867 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
2868 EXPORT_SYMBOL(getname
);
2869 EXPORT_SYMBOL(lock_rename
);
2870 EXPORT_SYMBOL(lookup_one_len
);
2871 EXPORT_SYMBOL(page_follow_link_light
);
2872 EXPORT_SYMBOL(page_put_link
);
2873 EXPORT_SYMBOL(page_readlink
);
2874 EXPORT_SYMBOL(__page_symlink
);
2875 EXPORT_SYMBOL(page_symlink
);
2876 EXPORT_SYMBOL(page_symlink_inode_operations
);
2877 EXPORT_SYMBOL(path_lookup
);
2878 EXPORT_SYMBOL(kern_path
);
2879 EXPORT_SYMBOL(vfs_path_lookup
);
2880 EXPORT_SYMBOL(inode_permission
);
2881 EXPORT_SYMBOL(file_permission
);
2882 EXPORT_SYMBOL(unlock_rename
);
2883 EXPORT_SYMBOL(vfs_create
);
2884 EXPORT_SYMBOL(vfs_follow_link
);
2885 EXPORT_SYMBOL(vfs_link
);
2886 EXPORT_SYMBOL(vfs_mkdir
);
2887 EXPORT_SYMBOL(vfs_mknod
);
2888 EXPORT_SYMBOL(generic_permission
);
2889 EXPORT_SYMBOL(vfs_readlink
);
2890 EXPORT_SYMBOL(vfs_rename
);
2891 EXPORT_SYMBOL(vfs_rmdir
);
2892 EXPORT_SYMBOL(vfs_symlink
);
2893 EXPORT_SYMBOL(vfs_unlink
);
2894 EXPORT_SYMBOL(dentry_unhash
);
2895 EXPORT_SYMBOL(generic_readlink
);
2897 /* to be mentioned only in INIT_TASK */
2898 struct fs_struct init_fs
= {
2899 .count
= ATOMIC_INIT(1),
2900 .lock
= __RW_LOCK_UNLOCKED(init_fs
.lock
),