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
) ||
216 (inode
->i_mode
& S_IXUGO
) || S_ISDIR(inode
->i_mode
))
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
))
230 int permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
233 struct vfsmount
*mnt
= NULL
;
238 if (mask
& MAY_WRITE
) {
239 umode_t mode
= inode
->i_mode
;
242 * Nobody gets write access to a read-only fs.
244 if (IS_RDONLY(inode
) &&
245 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
249 * Nobody gets write access to an immutable file.
251 if (IS_IMMUTABLE(inode
))
255 /* Ordinary permission routines do not understand MAY_APPEND. */
256 if (inode
->i_op
&& inode
->i_op
->permission
) {
257 retval
= inode
->i_op
->permission(inode
, mask
);
260 * Exec permission on a regular file is denied if none
261 * of the execute bits are set.
263 * This check should be done by the ->permission()
266 if ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
) &&
267 !(inode
->i_mode
& S_IXUGO
))
271 retval
= generic_permission(inode
, mask
, NULL
);
276 retval
= devcgroup_inode_permission(inode
, mask
);
280 return security_inode_permission(inode
,
281 mask
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
285 * vfs_permission - check for access rights to a given path
286 * @nd: lookup result that describes the path
287 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
289 * Used to check for read/write/execute permissions on a path.
290 * We use "fsuid" for this, letting us set arbitrary permissions
291 * for filesystem access without changing the "normal" uids which
292 * are used for other things.
294 int vfs_permission(struct nameidata
*nd
, int mask
)
296 return permission(nd
->path
.dentry
->d_inode
, mask
, nd
);
300 * file_permission - check for additional access rights to a given file
301 * @file: file to check access rights for
302 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
304 * Used to check for read/write/execute permissions on an already opened
308 * Do not use this function in new code. All access checks should
309 * be done using vfs_permission().
311 int file_permission(struct file
*file
, int mask
)
313 return permission(file
->f_path
.dentry
->d_inode
, mask
, NULL
);
317 * get_write_access() gets write permission for a file.
318 * put_write_access() releases this write permission.
319 * This is used for regular files.
320 * We cannot support write (and maybe mmap read-write shared) accesses and
321 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
322 * can have the following values:
323 * 0: no writers, no VM_DENYWRITE mappings
324 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
325 * > 0: (i_writecount) users are writing to the file.
327 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
328 * except for the cases where we don't hold i_writecount yet. Then we need to
329 * use {get,deny}_write_access() - these functions check the sign and refuse
330 * to do the change if sign is wrong. Exclusion between them is provided by
331 * the inode->i_lock spinlock.
334 int get_write_access(struct inode
* inode
)
336 spin_lock(&inode
->i_lock
);
337 if (atomic_read(&inode
->i_writecount
) < 0) {
338 spin_unlock(&inode
->i_lock
);
341 atomic_inc(&inode
->i_writecount
);
342 spin_unlock(&inode
->i_lock
);
347 int deny_write_access(struct file
* file
)
349 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
351 spin_lock(&inode
->i_lock
);
352 if (atomic_read(&inode
->i_writecount
) > 0) {
353 spin_unlock(&inode
->i_lock
);
356 atomic_dec(&inode
->i_writecount
);
357 spin_unlock(&inode
->i_lock
);
363 * path_get - get a reference to a path
364 * @path: path to get the reference to
366 * Given a path increment the reference count to the dentry and the vfsmount.
368 void path_get(struct path
*path
)
373 EXPORT_SYMBOL(path_get
);
376 * path_put - put a reference to a path
377 * @path: path to put the reference to
379 * Given a path decrement the reference count to the dentry and the vfsmount.
381 void path_put(struct path
*path
)
386 EXPORT_SYMBOL(path_put
);
389 * release_open_intent - free up open intent resources
390 * @nd: pointer to nameidata
392 void release_open_intent(struct nameidata
*nd
)
394 if (nd
->intent
.open
.file
->f_path
.dentry
== NULL
)
395 put_filp(nd
->intent
.open
.file
);
397 fput(nd
->intent
.open
.file
);
400 static inline struct dentry
*
401 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
403 int status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
404 if (unlikely(status
<= 0)) {
406 * The dentry failed validation.
407 * If d_revalidate returned 0 attempt to invalidate
408 * the dentry otherwise d_revalidate is asking us
409 * to return a fail status.
412 if (!d_invalidate(dentry
)) {
418 dentry
= ERR_PTR(status
);
425 * Internal lookup() using the new generic dcache.
428 static struct dentry
* cached_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
430 struct dentry
* dentry
= __d_lookup(parent
, name
);
432 /* lockess __d_lookup may fail due to concurrent d_move()
433 * in some unrelated directory, so try with d_lookup
436 dentry
= d_lookup(parent
, name
);
438 if (dentry
&& dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
439 dentry
= do_revalidate(dentry
, nd
);
445 * Short-cut version of permission(), for calling by
446 * path_walk(), when dcache lock is held. Combines parts
447 * of permission() and generic_permission(), and tests ONLY for
448 * MAY_EXEC permission.
450 * If appropriate, check DAC only. If not appropriate, or
451 * short-cut DAC fails, then call permission() to do more
452 * complete permission check.
454 static int exec_permission_lite(struct inode
*inode
)
456 umode_t mode
= inode
->i_mode
;
458 if (inode
->i_op
&& inode
->i_op
->permission
)
461 if (current
->fsuid
== inode
->i_uid
)
463 else if (in_group_p(inode
->i_gid
))
469 if ((inode
->i_mode
& S_IXUGO
) && capable(CAP_DAC_OVERRIDE
))
472 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_OVERRIDE
))
475 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_READ_SEARCH
))
480 return security_inode_permission(inode
, MAY_EXEC
);
484 * This is called when everything else fails, and we actually have
485 * to go to the low-level filesystem to find out what we should do..
487 * We get the directory semaphore, and after getting that we also
488 * make sure that nobody added the entry to the dcache in the meantime..
491 static struct dentry
* real_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
493 struct dentry
* result
;
494 struct inode
*dir
= parent
->d_inode
;
496 mutex_lock(&dir
->i_mutex
);
498 * First re-do the cached lookup just in case it was created
499 * while we waited for the directory semaphore..
501 * FIXME! This could use version numbering or similar to
502 * avoid unnecessary cache lookups.
504 * The "dcache_lock" is purely to protect the RCU list walker
505 * from concurrent renames at this point (we mustn't get false
506 * negatives from the RCU list walk here, unlike the optimistic
509 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
511 result
= d_lookup(parent
, name
);
513 struct dentry
*dentry
;
515 /* Don't create child dentry for a dead directory. */
516 result
= ERR_PTR(-ENOENT
);
520 dentry
= d_alloc(parent
, name
);
521 result
= ERR_PTR(-ENOMEM
);
523 result
= dir
->i_op
->lookup(dir
, dentry
, nd
);
530 mutex_unlock(&dir
->i_mutex
);
535 * Uhhuh! Nasty case: the cache was re-populated while
536 * we waited on the semaphore. Need to revalidate.
538 mutex_unlock(&dir
->i_mutex
);
539 if (result
->d_op
&& result
->d_op
->d_revalidate
) {
540 result
= do_revalidate(result
, nd
);
542 result
= ERR_PTR(-ENOENT
);
548 static __always_inline
void
549 walk_init_root(const char *name
, struct nameidata
*nd
)
551 struct fs_struct
*fs
= current
->fs
;
553 read_lock(&fs
->lock
);
556 read_unlock(&fs
->lock
);
560 * Wrapper to retry pathname resolution whenever the underlying
561 * file system returns an ESTALE.
563 * Retry the whole path once, forcing real lookup requests
564 * instead of relying on the dcache.
566 static __always_inline
int link_path_walk(const char *name
, struct nameidata
*nd
)
568 struct path save
= nd
->path
;
571 /* make sure the stuff we saved doesn't go away */
574 result
= __link_path_walk(name
, nd
);
575 if (result
== -ESTALE
) {
576 /* nd->path had been dropped */
579 nd
->flags
|= LOOKUP_REVAL
;
580 result
= __link_path_walk(name
, nd
);
588 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
597 walk_init_root(link
, nd
);
599 res
= link_path_walk(link
, nd
);
600 if (nd
->depth
|| res
|| nd
->last_type
!=LAST_NORM
)
603 * If it is an iterative symlinks resolution in open_namei() we
604 * have to copy the last component. And all that crap because of
605 * bloody create() on broken symlinks. Furrfu...
608 if (unlikely(!name
)) {
612 strcpy(name
, nd
->last
.name
);
613 nd
->last
.name
= name
;
617 return PTR_ERR(link
);
620 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
623 if (path
->mnt
!= nd
->path
.mnt
)
627 static inline void path_to_nameidata(struct path
*path
, struct nameidata
*nd
)
629 dput(nd
->path
.dentry
);
630 if (nd
->path
.mnt
!= path
->mnt
)
631 mntput(nd
->path
.mnt
);
632 nd
->path
.mnt
= path
->mnt
;
633 nd
->path
.dentry
= path
->dentry
;
636 static __always_inline
int __do_follow_link(struct path
*path
, struct nameidata
*nd
)
640 struct dentry
*dentry
= path
->dentry
;
642 touch_atime(path
->mnt
, dentry
);
643 nd_set_link(nd
, NULL
);
645 if (path
->mnt
!= nd
->path
.mnt
) {
646 path_to_nameidata(path
, nd
);
650 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
651 error
= PTR_ERR(cookie
);
652 if (!IS_ERR(cookie
)) {
653 char *s
= nd_get_link(nd
);
656 error
= __vfs_follow_link(nd
, s
);
657 if (dentry
->d_inode
->i_op
->put_link
)
658 dentry
->d_inode
->i_op
->put_link(dentry
, nd
, cookie
);
666 * This limits recursive symlink follows to 8, while
667 * limiting consecutive symlinks to 40.
669 * Without that kind of total limit, nasty chains of consecutive
670 * symlinks can cause almost arbitrarily long lookups.
672 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
675 if (current
->link_count
>= MAX_NESTED_LINKS
)
677 if (current
->total_link_count
>= 40)
679 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
681 err
= security_inode_follow_link(path
->dentry
, nd
);
684 current
->link_count
++;
685 current
->total_link_count
++;
687 err
= __do_follow_link(path
, nd
);
688 current
->link_count
--;
692 path_put_conditional(path
, nd
);
697 int follow_up(struct vfsmount
**mnt
, struct dentry
**dentry
)
699 struct vfsmount
*parent
;
700 struct dentry
*mountpoint
;
701 spin_lock(&vfsmount_lock
);
702 parent
=(*mnt
)->mnt_parent
;
703 if (parent
== *mnt
) {
704 spin_unlock(&vfsmount_lock
);
708 mountpoint
=dget((*mnt
)->mnt_mountpoint
);
709 spin_unlock(&vfsmount_lock
);
711 *dentry
= mountpoint
;
717 /* no need for dcache_lock, as serialization is taken care in
720 static int __follow_mount(struct path
*path
)
723 while (d_mountpoint(path
->dentry
)) {
724 struct vfsmount
*mounted
= lookup_mnt(path
->mnt
, path
->dentry
);
731 path
->dentry
= dget(mounted
->mnt_root
);
737 static void follow_mount(struct vfsmount
**mnt
, struct dentry
**dentry
)
739 while (d_mountpoint(*dentry
)) {
740 struct vfsmount
*mounted
= lookup_mnt(*mnt
, *dentry
);
746 *dentry
= dget(mounted
->mnt_root
);
750 /* no need for dcache_lock, as serialization is taken care in
753 int follow_down(struct vfsmount
**mnt
, struct dentry
**dentry
)
755 struct vfsmount
*mounted
;
757 mounted
= lookup_mnt(*mnt
, *dentry
);
762 *dentry
= dget(mounted
->mnt_root
);
768 static __always_inline
void follow_dotdot(struct nameidata
*nd
)
770 struct fs_struct
*fs
= current
->fs
;
773 struct vfsmount
*parent
;
774 struct dentry
*old
= nd
->path
.dentry
;
776 read_lock(&fs
->lock
);
777 if (nd
->path
.dentry
== fs
->root
.dentry
&&
778 nd
->path
.mnt
== fs
->root
.mnt
) {
779 read_unlock(&fs
->lock
);
782 read_unlock(&fs
->lock
);
783 spin_lock(&dcache_lock
);
784 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
785 nd
->path
.dentry
= dget(nd
->path
.dentry
->d_parent
);
786 spin_unlock(&dcache_lock
);
790 spin_unlock(&dcache_lock
);
791 spin_lock(&vfsmount_lock
);
792 parent
= nd
->path
.mnt
->mnt_parent
;
793 if (parent
== nd
->path
.mnt
) {
794 spin_unlock(&vfsmount_lock
);
798 nd
->path
.dentry
= dget(nd
->path
.mnt
->mnt_mountpoint
);
799 spin_unlock(&vfsmount_lock
);
801 mntput(nd
->path
.mnt
);
802 nd
->path
.mnt
= parent
;
804 follow_mount(&nd
->path
.mnt
, &nd
->path
.dentry
);
808 * It's more convoluted than I'd like it to be, but... it's still fairly
809 * small and for now I'd prefer to have fast path as straight as possible.
810 * It _is_ time-critical.
812 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
815 struct vfsmount
*mnt
= nd
->path
.mnt
;
816 struct dentry
*dentry
= __d_lookup(nd
->path
.dentry
, name
);
820 if (dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
821 goto need_revalidate
;
824 path
->dentry
= dentry
;
825 __follow_mount(path
);
829 dentry
= real_lookup(nd
->path
.dentry
, name
, nd
);
835 dentry
= do_revalidate(dentry
, nd
);
843 return PTR_ERR(dentry
);
848 * This is the basic name resolution function, turning a pathname into
849 * the final dentry. We expect 'base' to be positive and a directory.
851 * Returns 0 and nd will have valid dentry and mnt on success.
852 * Returns error and drops reference to input namei data on failure.
854 static int __link_path_walk(const char *name
, struct nameidata
*nd
)
859 unsigned int lookup_flags
= nd
->flags
;
866 inode
= nd
->path
.dentry
->d_inode
;
868 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
870 /* At this point we know we have a real path component. */
876 nd
->flags
|= LOOKUP_CONTINUE
;
877 err
= exec_permission_lite(inode
);
879 err
= vfs_permission(nd
, MAY_EXEC
);
884 c
= *(const unsigned char *)name
;
886 hash
= init_name_hash();
889 hash
= partial_name_hash(c
, hash
);
890 c
= *(const unsigned char *)name
;
891 } while (c
&& (c
!= '/'));
892 this.len
= name
- (const char *) this.name
;
893 this.hash
= end_name_hash(hash
);
895 /* remove trailing slashes? */
898 while (*++name
== '/');
900 goto last_with_slashes
;
903 * "." and ".." are special - ".." especially so because it has
904 * to be able to know about the current root directory and
905 * parent relationships.
907 if (this.name
[0] == '.') switch (this.len
) {
911 if (this.name
[1] != '.')
914 inode
= nd
->path
.dentry
->d_inode
;
920 * See if the low-level filesystem might want
921 * to use its own hash..
923 if (nd
->path
.dentry
->d_op
&& nd
->path
.dentry
->d_op
->d_hash
) {
924 err
= nd
->path
.dentry
->d_op
->d_hash(nd
->path
.dentry
,
929 /* This does the actual lookups.. */
930 err
= do_lookup(nd
, &this, &next
);
935 inode
= next
.dentry
->d_inode
;
942 if (inode
->i_op
->follow_link
) {
943 err
= do_follow_link(&next
, nd
);
947 inode
= nd
->path
.dentry
->d_inode
;
954 path_to_nameidata(&next
, nd
);
956 if (!inode
->i_op
->lookup
)
959 /* here ends the main loop */
962 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
964 /* Clear LOOKUP_CONTINUE iff it was previously unset */
965 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
966 if (lookup_flags
& LOOKUP_PARENT
)
968 if (this.name
[0] == '.') switch (this.len
) {
972 if (this.name
[1] != '.')
975 inode
= nd
->path
.dentry
->d_inode
;
980 if (nd
->path
.dentry
->d_op
&& nd
->path
.dentry
->d_op
->d_hash
) {
981 err
= nd
->path
.dentry
->d_op
->d_hash(nd
->path
.dentry
,
986 err
= do_lookup(nd
, &this, &next
);
989 inode
= next
.dentry
->d_inode
;
990 if ((lookup_flags
& LOOKUP_FOLLOW
)
991 && inode
&& inode
->i_op
&& inode
->i_op
->follow_link
) {
992 err
= do_follow_link(&next
, nd
);
995 inode
= nd
->path
.dentry
->d_inode
;
997 path_to_nameidata(&next
, nd
);
1001 if (lookup_flags
& LOOKUP_DIRECTORY
) {
1003 if (!inode
->i_op
|| !inode
->i_op
->lookup
)
1009 nd
->last_type
= LAST_NORM
;
1010 if (this.name
[0] != '.')
1013 nd
->last_type
= LAST_DOT
;
1014 else if (this.len
== 2 && this.name
[1] == '.')
1015 nd
->last_type
= LAST_DOTDOT
;
1020 * We bypassed the ordinary revalidation routines.
1021 * We may need to check the cached dentry for staleness.
1023 if (nd
->path
.dentry
&& nd
->path
.dentry
->d_sb
&&
1024 (nd
->path
.dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)) {
1026 /* Note: we do not d_invalidate() */
1027 if (!nd
->path
.dentry
->d_op
->d_revalidate(
1028 nd
->path
.dentry
, nd
))
1034 path_put_conditional(&next
, nd
);
1037 path_put(&nd
->path
);
1042 static int path_walk(const char *name
, struct nameidata
*nd
)
1044 current
->total_link_count
= 0;
1045 return link_path_walk(name
, nd
);
1048 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1049 static int do_path_lookup(int dfd
, const char *name
,
1050 unsigned int flags
, struct nameidata
*nd
)
1055 struct fs_struct
*fs
= current
->fs
;
1057 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1062 read_lock(&fs
->lock
);
1063 nd
->path
= fs
->root
;
1064 path_get(&fs
->root
);
1065 read_unlock(&fs
->lock
);
1066 } else if (dfd
== AT_FDCWD
) {
1067 read_lock(&fs
->lock
);
1070 read_unlock(&fs
->lock
);
1072 struct dentry
*dentry
;
1074 file
= fget_light(dfd
, &fput_needed
);
1079 dentry
= file
->f_path
.dentry
;
1082 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1085 retval
= file_permission(file
, MAY_EXEC
);
1089 nd
->path
= file
->f_path
;
1090 path_get(&file
->f_path
);
1092 fput_light(file
, fput_needed
);
1095 retval
= path_walk(name
, nd
);
1096 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1097 nd
->path
.dentry
->d_inode
))
1098 audit_inode(name
, nd
->path
.dentry
);
1103 fput_light(file
, fput_needed
);
1107 int path_lookup(const char *name
, unsigned int flags
,
1108 struct nameidata
*nd
)
1110 return do_path_lookup(AT_FDCWD
, name
, flags
, nd
);
1114 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1115 * @dentry: pointer to dentry of the base directory
1116 * @mnt: pointer to vfs mount of the base directory
1117 * @name: pointer to file name
1118 * @flags: lookup flags
1119 * @nd: pointer to nameidata
1121 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1122 const char *name
, unsigned int flags
,
1123 struct nameidata
*nd
)
1127 /* same as do_path_lookup */
1128 nd
->last_type
= LAST_ROOT
;
1132 nd
->path
.dentry
= dentry
;
1134 path_get(&nd
->path
);
1136 retval
= path_walk(name
, nd
);
1137 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1138 nd
->path
.dentry
->d_inode
))
1139 audit_inode(name
, nd
->path
.dentry
);
1145 static int __path_lookup_intent_open(int dfd
, const char *name
,
1146 unsigned int lookup_flags
, struct nameidata
*nd
,
1147 int open_flags
, int create_mode
)
1149 struct file
*filp
= get_empty_filp();
1154 nd
->intent
.open
.file
= filp
;
1155 nd
->intent
.open
.flags
= open_flags
;
1156 nd
->intent
.open
.create_mode
= create_mode
;
1157 err
= do_path_lookup(dfd
, name
, lookup_flags
|LOOKUP_OPEN
, nd
);
1158 if (IS_ERR(nd
->intent
.open
.file
)) {
1160 err
= PTR_ERR(nd
->intent
.open
.file
);
1161 path_put(&nd
->path
);
1163 } else if (err
!= 0)
1164 release_open_intent(nd
);
1169 * path_lookup_open - lookup a file path with open intent
1170 * @dfd: the directory to use as base, or AT_FDCWD
1171 * @name: pointer to file name
1172 * @lookup_flags: lookup intent flags
1173 * @nd: pointer to nameidata
1174 * @open_flags: open intent flags
1176 int path_lookup_open(int dfd
, const char *name
, unsigned int lookup_flags
,
1177 struct nameidata
*nd
, int open_flags
)
1179 return __path_lookup_intent_open(dfd
, name
, lookup_flags
, nd
,
1184 * path_lookup_create - lookup a file path with open + create intent
1185 * @dfd: the directory to use as base, or AT_FDCWD
1186 * @name: pointer to file name
1187 * @lookup_flags: lookup intent flags
1188 * @nd: pointer to nameidata
1189 * @open_flags: open intent flags
1190 * @create_mode: create intent flags
1192 static int path_lookup_create(int dfd
, const char *name
,
1193 unsigned int lookup_flags
, struct nameidata
*nd
,
1194 int open_flags
, int create_mode
)
1196 return __path_lookup_intent_open(dfd
, name
, lookup_flags
|LOOKUP_CREATE
,
1197 nd
, open_flags
, create_mode
);
1200 int __user_path_lookup_open(const char __user
*name
, unsigned int lookup_flags
,
1201 struct nameidata
*nd
, int open_flags
)
1203 char *tmp
= getname(name
);
1204 int err
= PTR_ERR(tmp
);
1207 err
= __path_lookup_intent_open(AT_FDCWD
, tmp
, lookup_flags
, nd
, open_flags
, 0);
1213 static struct dentry
*__lookup_hash(struct qstr
*name
,
1214 struct dentry
*base
, struct nameidata
*nd
)
1216 struct dentry
*dentry
;
1217 struct inode
*inode
;
1220 inode
= base
->d_inode
;
1223 * See if the low-level filesystem might want
1224 * to use its own hash..
1226 if (base
->d_op
&& base
->d_op
->d_hash
) {
1227 err
= base
->d_op
->d_hash(base
, name
);
1228 dentry
= ERR_PTR(err
);
1233 dentry
= cached_lookup(base
, name
, nd
);
1237 /* Don't create child dentry for a dead directory. */
1238 dentry
= ERR_PTR(-ENOENT
);
1239 if (IS_DEADDIR(inode
))
1242 new = d_alloc(base
, name
);
1243 dentry
= ERR_PTR(-ENOMEM
);
1246 dentry
= inode
->i_op
->lookup(inode
, new, nd
);
1257 * Restricted form of lookup. Doesn't follow links, single-component only,
1258 * needs parent already locked. Doesn't follow mounts.
1261 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1265 err
= permission(nd
->path
.dentry
->d_inode
, MAY_EXEC
, nd
);
1267 return ERR_PTR(err
);
1268 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1271 static int __lookup_one_len(const char *name
, struct qstr
*this,
1272 struct dentry
*base
, int len
)
1282 hash
= init_name_hash();
1284 c
= *(const unsigned char *)name
++;
1285 if (c
== '/' || c
== '\0')
1287 hash
= partial_name_hash(c
, hash
);
1289 this->hash
= end_name_hash(hash
);
1294 * lookup_one_len - filesystem helper to lookup single pathname component
1295 * @name: pathname component to lookup
1296 * @base: base directory to lookup from
1297 * @len: maximum length @len should be interpreted to
1299 * Note that this routine is purely a helper for filesystem usage and should
1300 * not be called by generic code. Also note that by using this function the
1301 * nameidata argument is passed to the filesystem methods and a filesystem
1302 * using this helper needs to be prepared for that.
1304 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1309 err
= __lookup_one_len(name
, &this, base
, len
);
1311 return ERR_PTR(err
);
1313 err
= permission(base
->d_inode
, MAY_EXEC
, NULL
);
1315 return ERR_PTR(err
);
1316 return __lookup_hash(&this, base
, NULL
);
1320 * lookup_one_noperm - bad hack for sysfs
1321 * @name: pathname component to lookup
1322 * @base: base directory to lookup from
1324 * This is a variant of lookup_one_len that doesn't perform any permission
1325 * checks. It's a horrible hack to work around the braindead sysfs
1326 * architecture and should not be used anywhere else.
1328 * DON'T USE THIS FUNCTION EVER, thanks.
1330 struct dentry
*lookup_one_noperm(const char *name
, struct dentry
*base
)
1335 err
= __lookup_one_len(name
, &this, base
, strlen(name
));
1337 return ERR_PTR(err
);
1338 return __lookup_hash(&this, base
, NULL
);
1341 int __user_walk_fd(int dfd
, const char __user
*name
, unsigned flags
,
1342 struct nameidata
*nd
)
1344 char *tmp
= getname(name
);
1345 int err
= PTR_ERR(tmp
);
1348 err
= do_path_lookup(dfd
, tmp
, flags
, nd
);
1354 int __user_walk(const char __user
*name
, unsigned flags
, struct nameidata
*nd
)
1356 return __user_walk_fd(AT_FDCWD
, name
, flags
, nd
);
1360 * It's inline, so penalty for filesystems that don't use sticky bit is
1363 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1365 if (!(dir
->i_mode
& S_ISVTX
))
1367 if (inode
->i_uid
== current
->fsuid
)
1369 if (dir
->i_uid
== current
->fsuid
)
1371 return !capable(CAP_FOWNER
);
1375 * Check whether we can remove a link victim from directory dir, check
1376 * whether the type of victim is right.
1377 * 1. We can't do it if dir is read-only (done in permission())
1378 * 2. We should have write and exec permissions on dir
1379 * 3. We can't remove anything from append-only dir
1380 * 4. We can't do anything with immutable dir (done in permission())
1381 * 5. If the sticky bit on dir is set we should either
1382 * a. be owner of dir, or
1383 * b. be owner of victim, or
1384 * c. have CAP_FOWNER capability
1385 * 6. If the victim is append-only or immutable we can't do antyhing with
1386 * links pointing to it.
1387 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1388 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1389 * 9. We can't remove a root or mountpoint.
1390 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1391 * nfs_async_unlink().
1393 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1397 if (!victim
->d_inode
)
1400 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1401 audit_inode_child(victim
->d_name
.name
, victim
, dir
);
1403 error
= permission(dir
,MAY_WRITE
| MAY_EXEC
, NULL
);
1408 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1409 IS_IMMUTABLE(victim
->d_inode
))
1412 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1414 if (IS_ROOT(victim
))
1416 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1418 if (IS_DEADDIR(dir
))
1420 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1425 /* Check whether we can create an object with dentry child in directory
1427 * 1. We can't do it if child already exists (open has special treatment for
1428 * this case, but since we are inlined it's OK)
1429 * 2. We can't do it if dir is read-only (done in permission())
1430 * 3. We should have write and exec permissions on dir
1431 * 4. We can't do it if dir is immutable (done in permission())
1433 static inline int may_create(struct inode
*dir
, struct dentry
*child
,
1434 struct nameidata
*nd
)
1438 if (IS_DEADDIR(dir
))
1440 return permission(dir
,MAY_WRITE
| MAY_EXEC
, nd
);
1444 * O_DIRECTORY translates into forcing a directory lookup.
1446 static inline int lookup_flags(unsigned int f
)
1448 unsigned long retval
= LOOKUP_FOLLOW
;
1451 retval
&= ~LOOKUP_FOLLOW
;
1453 if (f
& O_DIRECTORY
)
1454 retval
|= LOOKUP_DIRECTORY
;
1460 * p1 and p2 should be directories on the same fs.
1462 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1467 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1471 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1473 for (p
= p1
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1474 if (p
->d_parent
== p2
) {
1475 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1476 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1481 for (p
= p2
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1482 if (p
->d_parent
== p1
) {
1483 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1484 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1489 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1490 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1494 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1496 mutex_unlock(&p1
->d_inode
->i_mutex
);
1498 mutex_unlock(&p2
->d_inode
->i_mutex
);
1499 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1503 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1504 struct nameidata
*nd
)
1506 int error
= may_create(dir
, dentry
, nd
);
1511 if (!dir
->i_op
|| !dir
->i_op
->create
)
1512 return -EACCES
; /* shouldn't it be ENOSYS? */
1515 error
= security_inode_create(dir
, dentry
, mode
);
1519 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1521 fsnotify_create(dir
, dentry
);
1525 int may_open(struct nameidata
*nd
, int acc_mode
, int flag
)
1527 struct dentry
*dentry
= nd
->path
.dentry
;
1528 struct inode
*inode
= dentry
->d_inode
;
1534 if (S_ISLNK(inode
->i_mode
))
1537 if (S_ISDIR(inode
->i_mode
) && (acc_mode
& MAY_WRITE
))
1541 * FIFO's, sockets and device files are special: they don't
1542 * actually live on the filesystem itself, and as such you
1543 * can write to them even if the filesystem is read-only.
1545 if (S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
1547 } else if (S_ISBLK(inode
->i_mode
) || S_ISCHR(inode
->i_mode
)) {
1548 if (nd
->path
.mnt
->mnt_flags
& MNT_NODEV
)
1554 error
= vfs_permission(nd
, acc_mode
);
1558 * An append-only file must be opened in append mode for writing.
1560 if (IS_APPEND(inode
)) {
1561 if ((flag
& FMODE_WRITE
) && !(flag
& O_APPEND
))
1567 /* O_NOATIME can only be set by the owner or superuser */
1568 if (flag
& O_NOATIME
)
1569 if (!is_owner_or_cap(inode
))
1573 * Ensure there are no outstanding leases on the file.
1575 error
= break_lease(inode
, flag
);
1579 if (flag
& O_TRUNC
) {
1580 error
= get_write_access(inode
);
1585 * Refuse to truncate files with mandatory locks held on them.
1587 error
= locks_verify_locked(inode
);
1591 error
= do_truncate(dentry
, 0,
1592 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
1595 put_write_access(inode
);
1599 if (flag
& FMODE_WRITE
)
1606 * Be careful about ever adding any more callers of this
1607 * function. Its flags must be in the namei format, not
1608 * what get passed to sys_open().
1610 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
1614 struct dentry
*dir
= nd
->path
.dentry
;
1616 if (!IS_POSIXACL(dir
->d_inode
))
1617 mode
&= ~current
->fs
->umask
;
1618 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
1619 mutex_unlock(&dir
->d_inode
->i_mutex
);
1620 dput(nd
->path
.dentry
);
1621 nd
->path
.dentry
= path
->dentry
;
1624 /* Don't check for write permission, don't truncate */
1625 return may_open(nd
, 0, flag
& ~O_TRUNC
);
1629 * Note that while the flag value (low two bits) for sys_open means:
1634 * it is changed into
1635 * 00 - no permissions needed
1636 * 01 - read-permission
1637 * 10 - write-permission
1639 * for the internal routines (ie open_namei()/follow_link() etc)
1640 * This is more logical, and also allows the 00 "no perm needed"
1641 * to be used for symlinks (where the permissions are checked
1645 static inline int open_to_namei_flags(int flag
)
1647 if ((flag
+1) & O_ACCMODE
)
1652 static int open_will_write_to_fs(int flag
, struct inode
*inode
)
1655 * We'll never write to the fs underlying
1658 if (special_file(inode
->i_mode
))
1660 return (flag
& O_TRUNC
);
1664 * Note that the low bits of the passed in "open_flag"
1665 * are not the same as in the local variable "flag". See
1666 * open_to_namei_flags() for more details.
1668 struct file
*do_filp_open(int dfd
, const char *pathname
,
1669 int open_flag
, int mode
)
1672 struct nameidata nd
;
1673 int acc_mode
, error
;
1678 int flag
= open_to_namei_flags(open_flag
);
1680 acc_mode
= MAY_OPEN
| ACC_MODE(flag
);
1682 /* O_TRUNC implies we need access checks for write permissions */
1684 acc_mode
|= MAY_WRITE
;
1686 /* Allow the LSM permission hook to distinguish append
1687 access from general write access. */
1688 if (flag
& O_APPEND
)
1689 acc_mode
|= MAY_APPEND
;
1692 * The simplest case - just a plain lookup.
1694 if (!(flag
& O_CREAT
)) {
1695 error
= path_lookup_open(dfd
, pathname
, lookup_flags(flag
),
1698 return ERR_PTR(error
);
1703 * Create - we need to know the parent.
1705 error
= path_lookup_create(dfd
, pathname
, LOOKUP_PARENT
,
1708 return ERR_PTR(error
);
1711 * We have the parent and last component. First of all, check
1712 * that we are not asked to creat(2) an obvious directory - that
1716 if (nd
.last_type
!= LAST_NORM
|| nd
.last
.name
[nd
.last
.len
])
1719 dir
= nd
.path
.dentry
;
1720 nd
.flags
&= ~LOOKUP_PARENT
;
1721 mutex_lock(&dir
->d_inode
->i_mutex
);
1722 path
.dentry
= lookup_hash(&nd
);
1723 path
.mnt
= nd
.path
.mnt
;
1726 error
= PTR_ERR(path
.dentry
);
1727 if (IS_ERR(path
.dentry
)) {
1728 mutex_unlock(&dir
->d_inode
->i_mutex
);
1732 if (IS_ERR(nd
.intent
.open
.file
)) {
1733 error
= PTR_ERR(nd
.intent
.open
.file
);
1734 goto exit_mutex_unlock
;
1737 /* Negative dentry, just create the file */
1738 if (!path
.dentry
->d_inode
) {
1740 * This write is needed to ensure that a
1741 * ro->rw transition does not occur between
1742 * the time when the file is created and when
1743 * a permanent write count is taken through
1744 * the 'struct file' in nameidata_to_filp().
1746 error
= mnt_want_write(nd
.path
.mnt
);
1748 goto exit_mutex_unlock
;
1749 error
= __open_namei_create(&nd
, &path
, flag
, mode
);
1751 mnt_drop_write(nd
.path
.mnt
);
1754 filp
= nameidata_to_filp(&nd
, open_flag
);
1755 mnt_drop_write(nd
.path
.mnt
);
1760 * It already exists.
1762 mutex_unlock(&dir
->d_inode
->i_mutex
);
1763 audit_inode(pathname
, path
.dentry
);
1769 if (__follow_mount(&path
)) {
1771 if (flag
& O_NOFOLLOW
)
1776 if (!path
.dentry
->d_inode
)
1778 if (path
.dentry
->d_inode
->i_op
&& path
.dentry
->d_inode
->i_op
->follow_link
)
1781 path_to_nameidata(&path
, &nd
);
1783 if (path
.dentry
->d_inode
&& S_ISDIR(path
.dentry
->d_inode
->i_mode
))
1788 * 1. may_open() truncates a file
1789 * 2. a rw->ro mount transition occurs
1790 * 3. nameidata_to_filp() fails due to
1792 * That would be inconsistent, and should
1793 * be avoided. Taking this mnt write here
1794 * ensures that (2) can not occur.
1796 will_write
= open_will_write_to_fs(flag
, nd
.path
.dentry
->d_inode
);
1798 error
= mnt_want_write(nd
.path
.mnt
);
1802 error
= may_open(&nd
, acc_mode
, flag
);
1805 mnt_drop_write(nd
.path
.mnt
);
1808 filp
= nameidata_to_filp(&nd
, open_flag
);
1810 * It is now safe to drop the mnt write
1811 * because the filp has had a write taken
1815 mnt_drop_write(nd
.path
.mnt
);
1819 mutex_unlock(&dir
->d_inode
->i_mutex
);
1821 path_put_conditional(&path
, &nd
);
1823 if (!IS_ERR(nd
.intent
.open
.file
))
1824 release_open_intent(&nd
);
1826 return ERR_PTR(error
);
1830 if (flag
& O_NOFOLLOW
)
1833 * This is subtle. Instead of calling do_follow_link() we do the
1834 * thing by hands. The reason is that this way we have zero link_count
1835 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1836 * After that we have the parent and last component, i.e.
1837 * we are in the same situation as after the first path_walk().
1838 * Well, almost - if the last component is normal we get its copy
1839 * stored in nd->last.name and we will have to putname() it when we
1840 * are done. Procfs-like symlinks just set LAST_BIND.
1842 nd
.flags
|= LOOKUP_PARENT
;
1843 error
= security_inode_follow_link(path
.dentry
, &nd
);
1846 error
= __do_follow_link(&path
, &nd
);
1848 /* Does someone understand code flow here? Or it is only
1849 * me so stupid? Anathema to whoever designed this non-sense
1850 * with "intent.open".
1852 release_open_intent(&nd
);
1853 return ERR_PTR(error
);
1855 nd
.flags
&= ~LOOKUP_PARENT
;
1856 if (nd
.last_type
== LAST_BIND
)
1859 if (nd
.last_type
!= LAST_NORM
)
1861 if (nd
.last
.name
[nd
.last
.len
]) {
1862 __putname(nd
.last
.name
);
1867 __putname(nd
.last
.name
);
1870 dir
= nd
.path
.dentry
;
1871 mutex_lock(&dir
->d_inode
->i_mutex
);
1872 path
.dentry
= lookup_hash(&nd
);
1873 path
.mnt
= nd
.path
.mnt
;
1874 __putname(nd
.last
.name
);
1879 * filp_open - open file and return file pointer
1881 * @filename: path to open
1882 * @flags: open flags as per the open(2) second argument
1883 * @mode: mode for the new file if O_CREAT is set, else ignored
1885 * This is the helper to open a file from kernelspace if you really
1886 * have to. But in generally you should not do this, so please move
1887 * along, nothing to see here..
1889 struct file
*filp_open(const char *filename
, int flags
, int mode
)
1891 return do_filp_open(AT_FDCWD
, filename
, flags
, mode
);
1893 EXPORT_SYMBOL(filp_open
);
1896 * lookup_create - lookup a dentry, creating it if it doesn't exist
1897 * @nd: nameidata info
1898 * @is_dir: directory flag
1900 * Simple function to lookup and return a dentry and create it
1901 * if it doesn't exist. Is SMP-safe.
1903 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1905 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
1907 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
1909 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1911 * Yucky last component or no last component at all?
1912 * (foo/., foo/.., /////)
1914 if (nd
->last_type
!= LAST_NORM
)
1916 nd
->flags
&= ~LOOKUP_PARENT
;
1917 nd
->flags
|= LOOKUP_CREATE
;
1918 nd
->intent
.open
.flags
= O_EXCL
;
1921 * Do the final lookup.
1923 dentry
= lookup_hash(nd
);
1927 if (dentry
->d_inode
)
1930 * Special case - lookup gave negative, but... we had foo/bar/
1931 * From the vfs_mknod() POV we just have a negative dentry -
1932 * all is fine. Let's be bastards - you had / on the end, you've
1933 * been asking for (non-existent) directory. -ENOENT for you.
1935 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
1937 dentry
= ERR_PTR(-ENOENT
);
1942 dentry
= ERR_PTR(-EEXIST
);
1946 EXPORT_SYMBOL_GPL(lookup_create
);
1948 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1950 int error
= may_create(dir
, dentry
, NULL
);
1955 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
1958 if (!dir
->i_op
|| !dir
->i_op
->mknod
)
1961 error
= devcgroup_inode_mknod(mode
, dev
);
1965 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
1970 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
1972 fsnotify_create(dir
, dentry
);
1976 static int may_mknod(mode_t mode
)
1978 switch (mode
& S_IFMT
) {
1984 case 0: /* zero mode translates to S_IFREG */
1993 asmlinkage
long sys_mknodat(int dfd
, const char __user
*filename
, int mode
,
1998 struct dentry
* dentry
;
1999 struct nameidata nd
;
2003 tmp
= getname(filename
);
2005 return PTR_ERR(tmp
);
2007 error
= do_path_lookup(dfd
, tmp
, LOOKUP_PARENT
, &nd
);
2010 dentry
= lookup_create(&nd
, 0);
2011 if (IS_ERR(dentry
)) {
2012 error
= PTR_ERR(dentry
);
2015 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2016 mode
&= ~current
->fs
->umask
;
2017 error
= may_mknod(mode
);
2020 error
= mnt_want_write(nd
.path
.mnt
);
2023 switch (mode
& S_IFMT
) {
2024 case 0: case S_IFREG
:
2025 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2027 case S_IFCHR
: case S_IFBLK
:
2028 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2029 new_decode_dev(dev
));
2031 case S_IFIFO
: case S_IFSOCK
:
2032 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2035 mnt_drop_write(nd
.path
.mnt
);
2039 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2047 asmlinkage
long sys_mknod(const char __user
*filename
, int mode
, unsigned dev
)
2049 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2052 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2054 int error
= may_create(dir
, dentry
, NULL
);
2059 if (!dir
->i_op
|| !dir
->i_op
->mkdir
)
2062 mode
&= (S_IRWXUGO
|S_ISVTX
);
2063 error
= security_inode_mkdir(dir
, dentry
, mode
);
2068 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2070 fsnotify_mkdir(dir
, dentry
);
2074 asmlinkage
long sys_mkdirat(int dfd
, const char __user
*pathname
, int mode
)
2078 struct dentry
*dentry
;
2079 struct nameidata nd
;
2081 tmp
= getname(pathname
);
2082 error
= PTR_ERR(tmp
);
2086 error
= do_path_lookup(dfd
, tmp
, LOOKUP_PARENT
, &nd
);
2089 dentry
= lookup_create(&nd
, 1);
2090 error
= PTR_ERR(dentry
);
2094 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2095 mode
&= ~current
->fs
->umask
;
2096 error
= mnt_want_write(nd
.path
.mnt
);
2099 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2100 mnt_drop_write(nd
.path
.mnt
);
2104 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2112 asmlinkage
long sys_mkdir(const char __user
*pathname
, int mode
)
2114 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2118 * We try to drop the dentry early: we should have
2119 * a usage count of 2 if we're the only user of this
2120 * dentry, and if that is true (possibly after pruning
2121 * the dcache), then we drop the dentry now.
2123 * A low-level filesystem can, if it choses, legally
2126 * if (!d_unhashed(dentry))
2129 * if it cannot handle the case of removing a directory
2130 * that is still in use by something else..
2132 void dentry_unhash(struct dentry
*dentry
)
2135 shrink_dcache_parent(dentry
);
2136 spin_lock(&dcache_lock
);
2137 spin_lock(&dentry
->d_lock
);
2138 if (atomic_read(&dentry
->d_count
) == 2)
2140 spin_unlock(&dentry
->d_lock
);
2141 spin_unlock(&dcache_lock
);
2144 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2146 int error
= may_delete(dir
, dentry
, 1);
2151 if (!dir
->i_op
|| !dir
->i_op
->rmdir
)
2156 mutex_lock(&dentry
->d_inode
->i_mutex
);
2157 dentry_unhash(dentry
);
2158 if (d_mountpoint(dentry
))
2161 error
= security_inode_rmdir(dir
, dentry
);
2163 error
= dir
->i_op
->rmdir(dir
, dentry
);
2165 dentry
->d_inode
->i_flags
|= S_DEAD
;
2168 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2177 static long do_rmdir(int dfd
, const char __user
*pathname
)
2181 struct dentry
*dentry
;
2182 struct nameidata nd
;
2184 name
= getname(pathname
);
2186 return PTR_ERR(name
);
2188 error
= do_path_lookup(dfd
, name
, LOOKUP_PARENT
, &nd
);
2192 switch(nd
.last_type
) {
2203 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2204 dentry
= lookup_hash(&nd
);
2205 error
= PTR_ERR(dentry
);
2208 error
= mnt_want_write(nd
.path
.mnt
);
2211 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2212 mnt_drop_write(nd
.path
.mnt
);
2216 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2224 asmlinkage
long sys_rmdir(const char __user
*pathname
)
2226 return do_rmdir(AT_FDCWD
, pathname
);
2229 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2231 int error
= may_delete(dir
, dentry
, 0);
2236 if (!dir
->i_op
|| !dir
->i_op
->unlink
)
2241 mutex_lock(&dentry
->d_inode
->i_mutex
);
2242 if (d_mountpoint(dentry
))
2245 error
= security_inode_unlink(dir
, dentry
);
2247 error
= dir
->i_op
->unlink(dir
, dentry
);
2249 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2251 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2252 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2253 fsnotify_link_count(dentry
->d_inode
);
2261 * Make sure that the actual truncation of the file will occur outside its
2262 * directory's i_mutex. Truncate can take a long time if there is a lot of
2263 * writeout happening, and we don't want to prevent access to the directory
2264 * while waiting on the I/O.
2266 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2270 struct dentry
*dentry
;
2271 struct nameidata nd
;
2272 struct inode
*inode
= NULL
;
2274 name
= getname(pathname
);
2276 return PTR_ERR(name
);
2278 error
= do_path_lookup(dfd
, name
, LOOKUP_PARENT
, &nd
);
2282 if (nd
.last_type
!= LAST_NORM
)
2284 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2285 dentry
= lookup_hash(&nd
);
2286 error
= PTR_ERR(dentry
);
2287 if (!IS_ERR(dentry
)) {
2288 /* Why not before? Because we want correct error value */
2289 if (nd
.last
.name
[nd
.last
.len
])
2291 inode
= dentry
->d_inode
;
2293 atomic_inc(&inode
->i_count
);
2294 error
= mnt_want_write(nd
.path
.mnt
);
2297 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2298 mnt_drop_write(nd
.path
.mnt
);
2302 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2304 iput(inode
); /* truncate the inode here */
2312 error
= !dentry
->d_inode
? -ENOENT
:
2313 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2317 asmlinkage
long sys_unlinkat(int dfd
, const char __user
*pathname
, int flag
)
2319 if ((flag
& ~AT_REMOVEDIR
) != 0)
2322 if (flag
& AT_REMOVEDIR
)
2323 return do_rmdir(dfd
, pathname
);
2325 return do_unlinkat(dfd
, pathname
);
2328 asmlinkage
long sys_unlink(const char __user
*pathname
)
2330 return do_unlinkat(AT_FDCWD
, pathname
);
2333 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2335 int error
= may_create(dir
, dentry
, NULL
);
2340 if (!dir
->i_op
|| !dir
->i_op
->symlink
)
2343 error
= security_inode_symlink(dir
, dentry
, oldname
);
2348 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2350 fsnotify_create(dir
, dentry
);
2354 asmlinkage
long sys_symlinkat(const char __user
*oldname
,
2355 int newdfd
, const char __user
*newname
)
2360 struct dentry
*dentry
;
2361 struct nameidata nd
;
2363 from
= getname(oldname
);
2365 return PTR_ERR(from
);
2366 to
= getname(newname
);
2367 error
= PTR_ERR(to
);
2371 error
= do_path_lookup(newdfd
, to
, LOOKUP_PARENT
, &nd
);
2374 dentry
= lookup_create(&nd
, 0);
2375 error
= PTR_ERR(dentry
);
2379 error
= mnt_want_write(nd
.path
.mnt
);
2382 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2383 mnt_drop_write(nd
.path
.mnt
);
2387 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2396 asmlinkage
long sys_symlink(const char __user
*oldname
, const char __user
*newname
)
2398 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2401 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2403 struct inode
*inode
= old_dentry
->d_inode
;
2409 error
= may_create(dir
, new_dentry
, NULL
);
2413 if (dir
->i_sb
!= inode
->i_sb
)
2417 * A link to an append-only or immutable file cannot be created.
2419 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2421 if (!dir
->i_op
|| !dir
->i_op
->link
)
2423 if (S_ISDIR(inode
->i_mode
))
2426 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2430 mutex_lock(&inode
->i_mutex
);
2432 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2433 mutex_unlock(&inode
->i_mutex
);
2435 fsnotify_link(dir
, inode
, new_dentry
);
2440 * Hardlinks are often used in delicate situations. We avoid
2441 * security-related surprises by not following symlinks on the
2444 * We don't follow them on the oldname either to be compatible
2445 * with linux 2.0, and to avoid hard-linking to directories
2446 * and other special files. --ADM
2448 asmlinkage
long sys_linkat(int olddfd
, const char __user
*oldname
,
2449 int newdfd
, const char __user
*newname
,
2452 struct dentry
*new_dentry
;
2453 struct nameidata nd
, old_nd
;
2457 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
2460 to
= getname(newname
);
2464 error
= __user_walk_fd(olddfd
, oldname
,
2465 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
2469 error
= do_path_lookup(newdfd
, to
, LOOKUP_PARENT
, &nd
);
2473 if (old_nd
.path
.mnt
!= nd
.path
.mnt
)
2475 new_dentry
= lookup_create(&nd
, 0);
2476 error
= PTR_ERR(new_dentry
);
2477 if (IS_ERR(new_dentry
))
2479 error
= mnt_want_write(nd
.path
.mnt
);
2482 error
= vfs_link(old_nd
.path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2483 mnt_drop_write(nd
.path
.mnt
);
2487 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2491 path_put(&old_nd
.path
);
2498 asmlinkage
long sys_link(const char __user
*oldname
, const char __user
*newname
)
2500 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2504 * The worst of all namespace operations - renaming directory. "Perverted"
2505 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2507 * a) we can get into loop creation. Check is done in is_subdir().
2508 * b) race potential - two innocent renames can create a loop together.
2509 * That's where 4.4 screws up. Current fix: serialization on
2510 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2512 * c) we have to lock _three_ objects - parents and victim (if it exists).
2513 * And that - after we got ->i_mutex on parents (until then we don't know
2514 * whether the target exists). Solution: try to be smart with locking
2515 * order for inodes. We rely on the fact that tree topology may change
2516 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2517 * move will be locked. Thus we can rank directories by the tree
2518 * (ancestors first) and rank all non-directories after them.
2519 * That works since everybody except rename does "lock parent, lookup,
2520 * lock child" and rename is under ->s_vfs_rename_mutex.
2521 * HOWEVER, it relies on the assumption that any object with ->lookup()
2522 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2523 * we'd better make sure that there's no link(2) for them.
2524 * d) some filesystems don't support opened-but-unlinked directories,
2525 * either because of layout or because they are not ready to deal with
2526 * all cases correctly. The latter will be fixed (taking this sort of
2527 * stuff into VFS), but the former is not going away. Solution: the same
2528 * trick as in rmdir().
2529 * e) conversion from fhandle to dentry may come in the wrong moment - when
2530 * we are removing the target. Solution: we will have to grab ->i_mutex
2531 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2532 * ->i_mutex on parents, which works but leads to some truely excessive
2535 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2536 struct inode
*new_dir
, struct dentry
*new_dentry
)
2539 struct inode
*target
;
2542 * If we are going to change the parent - check write permissions,
2543 * we'll need to flip '..'.
2545 if (new_dir
!= old_dir
) {
2546 error
= permission(old_dentry
->d_inode
, MAY_WRITE
, NULL
);
2551 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2555 target
= new_dentry
->d_inode
;
2557 mutex_lock(&target
->i_mutex
);
2558 dentry_unhash(new_dentry
);
2560 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2563 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2566 target
->i_flags
|= S_DEAD
;
2567 mutex_unlock(&target
->i_mutex
);
2568 if (d_unhashed(new_dentry
))
2569 d_rehash(new_dentry
);
2573 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2574 d_move(old_dentry
,new_dentry
);
2578 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
2579 struct inode
*new_dir
, struct dentry
*new_dentry
)
2581 struct inode
*target
;
2584 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2589 target
= new_dentry
->d_inode
;
2591 mutex_lock(&target
->i_mutex
);
2592 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2595 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2597 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2598 d_move(old_dentry
, new_dentry
);
2601 mutex_unlock(&target
->i_mutex
);
2606 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2607 struct inode
*new_dir
, struct dentry
*new_dentry
)
2610 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2611 const char *old_name
;
2613 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
2616 error
= may_delete(old_dir
, old_dentry
, is_dir
);
2620 if (!new_dentry
->d_inode
)
2621 error
= may_create(new_dir
, new_dentry
, NULL
);
2623 error
= may_delete(new_dir
, new_dentry
, is_dir
);
2627 if (!old_dir
->i_op
|| !old_dir
->i_op
->rename
)
2630 DQUOT_INIT(old_dir
);
2631 DQUOT_INIT(new_dir
);
2633 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
2636 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
2638 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
2640 const char *new_name
= old_dentry
->d_name
.name
;
2641 fsnotify_move(old_dir
, new_dir
, old_name
, new_name
, is_dir
,
2642 new_dentry
->d_inode
, old_dentry
);
2644 fsnotify_oldname_free(old_name
);
2649 static int do_rename(int olddfd
, const char *oldname
,
2650 int newdfd
, const char *newname
)
2653 struct dentry
* old_dir
, * new_dir
;
2654 struct dentry
* old_dentry
, *new_dentry
;
2655 struct dentry
* trap
;
2656 struct nameidata oldnd
, newnd
;
2658 error
= do_path_lookup(olddfd
, oldname
, LOOKUP_PARENT
, &oldnd
);
2662 error
= do_path_lookup(newdfd
, newname
, LOOKUP_PARENT
, &newnd
);
2667 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
2670 old_dir
= oldnd
.path
.dentry
;
2672 if (oldnd
.last_type
!= LAST_NORM
)
2675 new_dir
= newnd
.path
.dentry
;
2676 if (newnd
.last_type
!= LAST_NORM
)
2679 trap
= lock_rename(new_dir
, old_dir
);
2681 old_dentry
= lookup_hash(&oldnd
);
2682 error
= PTR_ERR(old_dentry
);
2683 if (IS_ERR(old_dentry
))
2685 /* source must exist */
2687 if (!old_dentry
->d_inode
)
2689 /* unless the source is a directory trailing slashes give -ENOTDIR */
2690 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
2692 if (oldnd
.last
.name
[oldnd
.last
.len
])
2694 if (newnd
.last
.name
[newnd
.last
.len
])
2697 /* source should not be ancestor of target */
2699 if (old_dentry
== trap
)
2701 new_dentry
= lookup_hash(&newnd
);
2702 error
= PTR_ERR(new_dentry
);
2703 if (IS_ERR(new_dentry
))
2705 /* target should not be an ancestor of source */
2707 if (new_dentry
== trap
)
2710 error
= mnt_want_write(oldnd
.path
.mnt
);
2713 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
2714 new_dir
->d_inode
, new_dentry
);
2715 mnt_drop_write(oldnd
.path
.mnt
);
2721 unlock_rename(new_dir
, old_dir
);
2723 path_put(&newnd
.path
);
2725 path_put(&oldnd
.path
);
2730 asmlinkage
long sys_renameat(int olddfd
, const char __user
*oldname
,
2731 int newdfd
, const char __user
*newname
)
2737 from
= getname(oldname
);
2739 return PTR_ERR(from
);
2740 to
= getname(newname
);
2741 error
= PTR_ERR(to
);
2743 error
= do_rename(olddfd
, from
, newdfd
, to
);
2750 asmlinkage
long sys_rename(const char __user
*oldname
, const char __user
*newname
)
2752 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
2755 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
2759 len
= PTR_ERR(link
);
2764 if (len
> (unsigned) buflen
)
2766 if (copy_to_user(buffer
, link
, len
))
2773 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2774 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2775 * using) it for any given inode is up to filesystem.
2777 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2779 struct nameidata nd
;
2784 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
2786 return PTR_ERR(cookie
);
2788 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
2789 if (dentry
->d_inode
->i_op
->put_link
)
2790 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
2794 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
2796 return __vfs_follow_link(nd
, link
);
2799 /* get the link contents into pagecache */
2800 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
2803 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
2804 page
= read_mapping_page(mapping
, 0, NULL
);
2811 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2813 struct page
*page
= NULL
;
2814 char *s
= page_getlink(dentry
, &page
);
2815 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
2818 page_cache_release(page
);
2823 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
2825 struct page
*page
= NULL
;
2826 nd_set_link(nd
, page_getlink(dentry
, &page
));
2830 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2832 struct page
*page
= cookie
;
2836 page_cache_release(page
);
2840 int __page_symlink(struct inode
*inode
, const char *symname
, int len
,
2843 struct address_space
*mapping
= inode
->i_mapping
;
2850 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
2851 AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
2855 kaddr
= kmap_atomic(page
, KM_USER0
);
2856 memcpy(kaddr
, symname
, len
-1);
2857 kunmap_atomic(kaddr
, KM_USER0
);
2859 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
2866 mark_inode_dirty(inode
);
2872 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
2874 return __page_symlink(inode
, symname
, len
,
2875 mapping_gfp_mask(inode
->i_mapping
));
2878 const struct inode_operations page_symlink_inode_operations
= {
2879 .readlink
= generic_readlink
,
2880 .follow_link
= page_follow_link_light
,
2881 .put_link
= page_put_link
,
2884 EXPORT_SYMBOL(__user_walk
);
2885 EXPORT_SYMBOL(__user_walk_fd
);
2886 EXPORT_SYMBOL(follow_down
);
2887 EXPORT_SYMBOL(follow_up
);
2888 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
2889 EXPORT_SYMBOL(getname
);
2890 EXPORT_SYMBOL(lock_rename
);
2891 EXPORT_SYMBOL(lookup_one_len
);
2892 EXPORT_SYMBOL(page_follow_link_light
);
2893 EXPORT_SYMBOL(page_put_link
);
2894 EXPORT_SYMBOL(page_readlink
);
2895 EXPORT_SYMBOL(__page_symlink
);
2896 EXPORT_SYMBOL(page_symlink
);
2897 EXPORT_SYMBOL(page_symlink_inode_operations
);
2898 EXPORT_SYMBOL(path_lookup
);
2899 EXPORT_SYMBOL(vfs_path_lookup
);
2900 EXPORT_SYMBOL(permission
);
2901 EXPORT_SYMBOL(vfs_permission
);
2902 EXPORT_SYMBOL(file_permission
);
2903 EXPORT_SYMBOL(unlock_rename
);
2904 EXPORT_SYMBOL(vfs_create
);
2905 EXPORT_SYMBOL(vfs_follow_link
);
2906 EXPORT_SYMBOL(vfs_link
);
2907 EXPORT_SYMBOL(vfs_mkdir
);
2908 EXPORT_SYMBOL(vfs_mknod
);
2909 EXPORT_SYMBOL(generic_permission
);
2910 EXPORT_SYMBOL(vfs_readlink
);
2911 EXPORT_SYMBOL(vfs_rename
);
2912 EXPORT_SYMBOL(vfs_rmdir
);
2913 EXPORT_SYMBOL(vfs_symlink
);
2914 EXPORT_SYMBOL(vfs_unlink
);
2915 EXPORT_SYMBOL(dentry_unhash
);
2916 EXPORT_SYMBOL(generic_readlink
);