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/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.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 <linux/fs_struct.h>
35 #include <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existent name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user
*filename
, char *page
)
120 unsigned long len
= PATH_MAX
;
122 if (!segment_eq(get_fs(), KERNEL_DS
)) {
123 if ((unsigned long) filename
>= TASK_SIZE
)
125 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
126 len
= TASK_SIZE
- (unsigned long) filename
;
129 retval
= strncpy_from_user(page
, filename
, len
);
133 return -ENAMETOOLONG
;
139 static char *getname_flags(const char __user
* filename
, int flags
)
143 result
= ERR_PTR(-ENOMEM
);
146 int retval
= do_getname(filename
, tmp
);
150 if (retval
!= -ENOENT
|| !(flags
& LOOKUP_EMPTY
)) {
152 result
= ERR_PTR(retval
);
156 audit_getname(result
);
160 char *getname(const char __user
* filename
)
162 return getname_flags(filename
, 0);
165 #ifdef CONFIG_AUDITSYSCALL
166 void putname(const char *name
)
168 if (unlikely(!audit_dummy_context()))
173 EXPORT_SYMBOL(putname
);
177 * This does basic POSIX ACL permission checking
179 static int acl_permission_check(struct inode
*inode
, int mask
, unsigned int flags
,
180 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
182 unsigned int mode
= inode
->i_mode
;
184 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
186 if (current_user_ns() != inode_userns(inode
))
189 if (current_fsuid() == inode
->i_uid
)
192 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
193 int error
= check_acl(inode
, mask
, flags
);
194 if (error
!= -EAGAIN
)
198 if (in_group_p(inode
->i_gid
))
204 * If the DACs are ok we don't need any capability check.
206 if ((mask
& ~mode
) == 0)
212 * generic_permission - check for access rights on a Posix-like filesystem
213 * @inode: inode to check access rights for
214 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
215 * @check_acl: optional callback to check for Posix ACLs
216 * @flags: IPERM_FLAG_ flags.
218 * Used to check for read/write/execute permissions on a file.
219 * We use "fsuid" for this, letting us set arbitrary permissions
220 * for filesystem access without changing the "normal" uids which
221 * are used for other things.
223 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
224 * request cannot be satisfied (eg. requires blocking or too much complexity).
225 * It would then be called again in ref-walk mode.
227 int generic_permission(struct inode
*inode
, int mask
, unsigned int flags
,
228 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
233 * Do the basic POSIX ACL permission checks.
235 ret
= acl_permission_check(inode
, mask
, flags
, check_acl
);
240 * Read/write DACs are always overridable.
241 * Executable DACs are overridable for all directories and
242 * for non-directories that have least one exec bit set.
244 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
245 if (ns_capable(inode_userns(inode
), CAP_DAC_OVERRIDE
))
249 * Searching includes executable on directories, else just read.
251 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
252 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
253 if (ns_capable(inode_userns(inode
), CAP_DAC_READ_SEARCH
))
260 * inode_permission - check for access rights to a given inode
261 * @inode: inode to check permission on
262 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
264 * Used to check for read/write/execute permissions on an inode.
265 * We use "fsuid" for this, letting us set arbitrary permissions
266 * for filesystem access without changing the "normal" uids which
267 * are used for other things.
269 int inode_permission(struct inode
*inode
, int mask
)
273 if (mask
& MAY_WRITE
) {
274 umode_t mode
= inode
->i_mode
;
277 * Nobody gets write access to a read-only fs.
279 if (IS_RDONLY(inode
) &&
280 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
284 * Nobody gets write access to an immutable file.
286 if (IS_IMMUTABLE(inode
))
290 if (inode
->i_op
->permission
)
291 retval
= inode
->i_op
->permission(inode
, mask
, 0);
293 retval
= generic_permission(inode
, mask
, 0,
294 inode
->i_op
->check_acl
);
299 retval
= devcgroup_inode_permission(inode
, mask
);
303 return security_inode_permission(inode
, mask
);
307 * file_permission - check for additional access rights to a given file
308 * @file: file to check access rights for
309 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
311 * Used to check for read/write/execute permissions on an already opened
315 * Do not use this function in new code. All access checks should
316 * be done using inode_permission().
318 int file_permission(struct file
*file
, int mask
)
320 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
324 * get_write_access() gets write permission for a file.
325 * put_write_access() releases this write permission.
326 * This is used for regular files.
327 * We cannot support write (and maybe mmap read-write shared) accesses and
328 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
329 * can have the following values:
330 * 0: no writers, no VM_DENYWRITE mappings
331 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
332 * > 0: (i_writecount) users are writing to the file.
334 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
335 * except for the cases where we don't hold i_writecount yet. Then we need to
336 * use {get,deny}_write_access() - these functions check the sign and refuse
337 * to do the change if sign is wrong. Exclusion between them is provided by
338 * the inode->i_lock spinlock.
341 int get_write_access(struct inode
* inode
)
343 spin_lock(&inode
->i_lock
);
344 if (atomic_read(&inode
->i_writecount
) < 0) {
345 spin_unlock(&inode
->i_lock
);
348 atomic_inc(&inode
->i_writecount
);
349 spin_unlock(&inode
->i_lock
);
354 int deny_write_access(struct file
* file
)
356 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
358 spin_lock(&inode
->i_lock
);
359 if (atomic_read(&inode
->i_writecount
) > 0) {
360 spin_unlock(&inode
->i_lock
);
363 atomic_dec(&inode
->i_writecount
);
364 spin_unlock(&inode
->i_lock
);
370 * path_get - get a reference to a path
371 * @path: path to get the reference to
373 * Given a path increment the reference count to the dentry and the vfsmount.
375 void path_get(struct path
*path
)
380 EXPORT_SYMBOL(path_get
);
383 * path_put - put a reference to a path
384 * @path: path to put the reference to
386 * Given a path decrement the reference count to the dentry and the vfsmount.
388 void path_put(struct path
*path
)
393 EXPORT_SYMBOL(path_put
);
396 * Path walking has 2 modes, rcu-walk and ref-walk (see
397 * Documentation/filesystems/path-lookup.txt). In situations when we can't
398 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
399 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
400 * mode. Refcounts are grabbed at the last known good point before rcu-walk
401 * got stuck, so ref-walk may continue from there. If this is not successful
402 * (eg. a seqcount has changed), then failure is returned and it's up to caller
403 * to restart the path walk from the beginning in ref-walk mode.
407 * unlazy_walk - try to switch to ref-walk mode.
408 * @nd: nameidata pathwalk data
409 * @dentry: child of nd->path.dentry or NULL
410 * Returns: 0 on success, -ECHILD on failure
412 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
413 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
414 * @nd or NULL. Must be called from rcu-walk context.
416 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
418 struct fs_struct
*fs
= current
->fs
;
419 struct dentry
*parent
= nd
->path
.dentry
;
422 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
423 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
425 spin_lock(&fs
->lock
);
426 if (nd
->root
.mnt
!= fs
->root
.mnt
||
427 nd
->root
.dentry
!= fs
->root
.dentry
)
430 spin_lock(&parent
->d_lock
);
432 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
434 BUG_ON(nd
->inode
!= parent
->d_inode
);
436 if (dentry
->d_parent
!= parent
)
438 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
439 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
442 * If the sequence check on the child dentry passed, then
443 * the child has not been removed from its parent. This
444 * means the parent dentry must be valid and able to take
445 * a reference at this point.
447 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
448 BUG_ON(!parent
->d_count
);
450 spin_unlock(&dentry
->d_lock
);
452 spin_unlock(&parent
->d_lock
);
455 spin_unlock(&fs
->lock
);
457 mntget(nd
->path
.mnt
);
460 br_read_unlock(vfsmount_lock
);
461 nd
->flags
&= ~LOOKUP_RCU
;
465 spin_unlock(&dentry
->d_lock
);
467 spin_unlock(&parent
->d_lock
);
470 spin_unlock(&fs
->lock
);
475 * release_open_intent - free up open intent resources
476 * @nd: pointer to nameidata
478 void release_open_intent(struct nameidata
*nd
)
480 struct file
*file
= nd
->intent
.open
.file
;
482 if (file
&& !IS_ERR(file
)) {
483 if (file
->f_path
.dentry
== NULL
)
490 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
492 return dentry
->d_op
->d_revalidate(dentry
, nd
);
495 static struct dentry
*
496 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
498 int status
= d_revalidate(dentry
, nd
);
499 if (unlikely(status
<= 0)) {
501 * The dentry failed validation.
502 * If d_revalidate returned 0 attempt to invalidate
503 * the dentry otherwise d_revalidate is asking us
504 * to return a fail status.
508 dentry
= ERR_PTR(status
);
509 } else if (!d_invalidate(dentry
)) {
518 * complete_walk - successful completion of path walk
519 * @nd: pointer nameidata
521 * If we had been in RCU mode, drop out of it and legitimize nd->path.
522 * Revalidate the final result, unless we'd already done that during
523 * the path walk or the filesystem doesn't ask for it. Return 0 on
524 * success, -error on failure. In case of failure caller does not
525 * need to drop nd->path.
527 static int complete_walk(struct nameidata
*nd
)
529 struct dentry
*dentry
= nd
->path
.dentry
;
532 if (nd
->flags
& LOOKUP_RCU
) {
533 nd
->flags
&= ~LOOKUP_RCU
;
534 if (!(nd
->flags
& LOOKUP_ROOT
))
536 spin_lock(&dentry
->d_lock
);
537 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
538 spin_unlock(&dentry
->d_lock
);
540 br_read_unlock(vfsmount_lock
);
543 BUG_ON(nd
->inode
!= dentry
->d_inode
);
544 spin_unlock(&dentry
->d_lock
);
545 mntget(nd
->path
.mnt
);
547 br_read_unlock(vfsmount_lock
);
550 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
553 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
556 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
559 /* Note: we do not d_invalidate() */
560 status
= d_revalidate(dentry
, nd
);
572 * Short-cut version of permission(), for calling on directories
573 * during pathname resolution. Combines parts of permission()
574 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
576 * If appropriate, check DAC only. If not appropriate, or
577 * short-cut DAC fails, then call ->permission() to do more
578 * complete permission check.
580 static inline int exec_permission(struct inode
*inode
, unsigned int flags
)
583 struct user_namespace
*ns
= inode_userns(inode
);
585 if (inode
->i_op
->permission
) {
586 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
, flags
);
590 ret
= acl_permission_check(inode
, MAY_EXEC
, flags
,
591 inode
->i_op
->check_acl
);
596 if (ns_capable(ns
, CAP_DAC_OVERRIDE
) ||
597 ns_capable(ns
, CAP_DAC_READ_SEARCH
))
602 return security_inode_exec_permission(inode
, flags
);
605 static __always_inline
void set_root(struct nameidata
*nd
)
608 get_fs_root(current
->fs
, &nd
->root
);
611 static int link_path_walk(const char *, struct nameidata
*);
613 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
616 struct fs_struct
*fs
= current
->fs
;
620 seq
= read_seqcount_begin(&fs
->seq
);
622 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
623 } while (read_seqcount_retry(&fs
->seq
, seq
));
627 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
639 nd
->flags
|= LOOKUP_JUMPED
;
641 nd
->inode
= nd
->path
.dentry
->d_inode
;
643 ret
= link_path_walk(link
, nd
);
647 return PTR_ERR(link
);
650 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
653 if (path
->mnt
!= nd
->path
.mnt
)
657 static inline void path_to_nameidata(const struct path
*path
,
658 struct nameidata
*nd
)
660 if (!(nd
->flags
& LOOKUP_RCU
)) {
661 dput(nd
->path
.dentry
);
662 if (nd
->path
.mnt
!= path
->mnt
)
663 mntput(nd
->path
.mnt
);
665 nd
->path
.mnt
= path
->mnt
;
666 nd
->path
.dentry
= path
->dentry
;
669 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
671 struct inode
*inode
= link
->dentry
->d_inode
;
672 if (!IS_ERR(cookie
) && inode
->i_op
->put_link
)
673 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
677 static __always_inline
int
678 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
681 struct dentry
*dentry
= link
->dentry
;
683 BUG_ON(nd
->flags
& LOOKUP_RCU
);
685 if (link
->mnt
== nd
->path
.mnt
)
688 if (unlikely(current
->total_link_count
>= 40)) {
689 *p
= ERR_PTR(-ELOOP
); /* no ->put_link(), please */
694 current
->total_link_count
++;
696 touch_atime(link
->mnt
, dentry
);
697 nd_set_link(nd
, NULL
);
699 error
= security_inode_follow_link(link
->dentry
, nd
);
701 *p
= ERR_PTR(error
); /* no ->put_link(), please */
706 nd
->last_type
= LAST_BIND
;
707 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
710 char *s
= nd_get_link(nd
);
713 error
= __vfs_follow_link(nd
, s
);
714 else if (nd
->last_type
== LAST_BIND
) {
715 nd
->flags
|= LOOKUP_JUMPED
;
716 nd
->inode
= nd
->path
.dentry
->d_inode
;
717 if (nd
->inode
->i_op
->follow_link
) {
718 /* stepped on a _really_ weird one */
727 static int follow_up_rcu(struct path
*path
)
729 struct vfsmount
*parent
;
730 struct dentry
*mountpoint
;
732 parent
= path
->mnt
->mnt_parent
;
733 if (parent
== path
->mnt
)
735 mountpoint
= path
->mnt
->mnt_mountpoint
;
736 path
->dentry
= mountpoint
;
741 int follow_up(struct path
*path
)
743 struct vfsmount
*parent
;
744 struct dentry
*mountpoint
;
746 br_read_lock(vfsmount_lock
);
747 parent
= path
->mnt
->mnt_parent
;
748 if (parent
== path
->mnt
) {
749 br_read_unlock(vfsmount_lock
);
753 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
754 br_read_unlock(vfsmount_lock
);
756 path
->dentry
= mountpoint
;
763 * Perform an automount
764 * - return -EISDIR to tell follow_managed() to stop and return the path we
767 static int follow_automount(struct path
*path
, unsigned flags
,
770 struct vfsmount
*mnt
;
773 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
776 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
777 * and this is the terminal part of the path.
779 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
780 return -EISDIR
; /* we actually want to stop here */
782 /* We want to mount if someone is trying to open/create a file of any
783 * type under the mountpoint, wants to traverse through the mountpoint
784 * or wants to open the mounted directory.
786 * We don't want to mount if someone's just doing a stat and they've
787 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
788 * appended a '/' to the name.
790 if (!(flags
& LOOKUP_FOLLOW
) &&
791 !(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
792 LOOKUP_OPEN
| LOOKUP_CREATE
)))
795 current
->total_link_count
++;
796 if (current
->total_link_count
>= 40)
799 mnt
= path
->dentry
->d_op
->d_automount(path
);
802 * The filesystem is allowed to return -EISDIR here to indicate
803 * it doesn't want to automount. For instance, autofs would do
804 * this so that its userspace daemon can mount on this dentry.
806 * However, we can only permit this if it's a terminal point in
807 * the path being looked up; if it wasn't then the remainder of
808 * the path is inaccessible and we should say so.
810 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_CONTINUE
))
815 if (!mnt
) /* mount collision */
819 /* lock_mount() may release path->mnt on error */
823 err
= finish_automount(mnt
, path
);
827 /* Someone else made a mount here whilst we were busy */
832 path
->dentry
= dget(mnt
->mnt_root
);
841 * Handle a dentry that is managed in some way.
842 * - Flagged for transit management (autofs)
843 * - Flagged as mountpoint
844 * - Flagged as automount point
846 * This may only be called in refwalk mode.
848 * Serialization is taken care of in namespace.c
850 static int follow_managed(struct path
*path
, unsigned flags
)
852 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
854 bool need_mntput
= false;
857 /* Given that we're not holding a lock here, we retain the value in a
858 * local variable for each dentry as we look at it so that we don't see
859 * the components of that value change under us */
860 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
861 managed
&= DCACHE_MANAGED_DENTRY
,
862 unlikely(managed
!= 0)) {
863 /* Allow the filesystem to manage the transit without i_mutex
865 if (managed
& DCACHE_MANAGE_TRANSIT
) {
866 BUG_ON(!path
->dentry
->d_op
);
867 BUG_ON(!path
->dentry
->d_op
->d_manage
);
868 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
873 /* Transit to a mounted filesystem. */
874 if (managed
& DCACHE_MOUNTED
) {
875 struct vfsmount
*mounted
= lookup_mnt(path
);
881 path
->dentry
= dget(mounted
->mnt_root
);
886 /* Something is mounted on this dentry in another
887 * namespace and/or whatever was mounted there in this
888 * namespace got unmounted before we managed to get the
892 /* Handle an automount point */
893 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
894 ret
= follow_automount(path
, flags
, &need_mntput
);
900 /* We didn't change the current path point */
904 if (need_mntput
&& path
->mnt
== mnt
)
911 int follow_down_one(struct path
*path
)
913 struct vfsmount
*mounted
;
915 mounted
= lookup_mnt(path
);
920 path
->dentry
= dget(mounted
->mnt_root
);
926 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
928 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
929 dentry
->d_op
->d_manage(dentry
, true) < 0);
933 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
934 * we meet a managed dentry that would need blocking.
936 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
937 struct inode
**inode
)
940 struct vfsmount
*mounted
;
942 * Don't forget we might have a non-mountpoint managed dentry
943 * that wants to block transit.
945 if (unlikely(managed_dentry_might_block(path
->dentry
)))
948 if (!d_mountpoint(path
->dentry
))
951 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
955 path
->dentry
= mounted
->mnt_root
;
956 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
958 * Update the inode too. We don't need to re-check the
959 * dentry sequence number here after this d_inode read,
960 * because a mount-point is always pinned.
962 *inode
= path
->dentry
->d_inode
;
967 static void follow_mount_rcu(struct nameidata
*nd
)
969 while (d_mountpoint(nd
->path
.dentry
)) {
970 struct vfsmount
*mounted
;
971 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
974 nd
->path
.mnt
= mounted
;
975 nd
->path
.dentry
= mounted
->mnt_root
;
976 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
980 static int follow_dotdot_rcu(struct nameidata
*nd
)
985 if (nd
->path
.dentry
== nd
->root
.dentry
&&
986 nd
->path
.mnt
== nd
->root
.mnt
) {
989 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
990 struct dentry
*old
= nd
->path
.dentry
;
991 struct dentry
*parent
= old
->d_parent
;
994 seq
= read_seqcount_begin(&parent
->d_seq
);
995 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
997 nd
->path
.dentry
= parent
;
1001 if (!follow_up_rcu(&nd
->path
))
1003 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1005 follow_mount_rcu(nd
);
1006 nd
->inode
= nd
->path
.dentry
->d_inode
;
1010 nd
->flags
&= ~LOOKUP_RCU
;
1011 if (!(nd
->flags
& LOOKUP_ROOT
))
1012 nd
->root
.mnt
= NULL
;
1014 br_read_unlock(vfsmount_lock
);
1019 * Follow down to the covering mount currently visible to userspace. At each
1020 * point, the filesystem owning that dentry may be queried as to whether the
1021 * caller is permitted to proceed or not.
1023 int follow_down(struct path
*path
)
1028 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1029 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1030 /* Allow the filesystem to manage the transit without i_mutex
1033 * We indicate to the filesystem if someone is trying to mount
1034 * something here. This gives autofs the chance to deny anyone
1035 * other than its daemon the right to mount on its
1038 * The filesystem may sleep at this point.
1040 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1041 BUG_ON(!path
->dentry
->d_op
);
1042 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1043 ret
= path
->dentry
->d_op
->d_manage(
1044 path
->dentry
, false);
1046 return ret
== -EISDIR
? 0 : ret
;
1049 /* Transit to a mounted filesystem. */
1050 if (managed
& DCACHE_MOUNTED
) {
1051 struct vfsmount
*mounted
= lookup_mnt(path
);
1056 path
->mnt
= mounted
;
1057 path
->dentry
= dget(mounted
->mnt_root
);
1061 /* Don't handle automount points here */
1068 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1070 static void follow_mount(struct path
*path
)
1072 while (d_mountpoint(path
->dentry
)) {
1073 struct vfsmount
*mounted
= lookup_mnt(path
);
1078 path
->mnt
= mounted
;
1079 path
->dentry
= dget(mounted
->mnt_root
);
1083 static void follow_dotdot(struct nameidata
*nd
)
1088 struct dentry
*old
= nd
->path
.dentry
;
1090 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1091 nd
->path
.mnt
== nd
->root
.mnt
) {
1094 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1095 /* rare case of legitimate dget_parent()... */
1096 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1100 if (!follow_up(&nd
->path
))
1103 follow_mount(&nd
->path
);
1104 nd
->inode
= nd
->path
.dentry
->d_inode
;
1108 * Allocate a dentry with name and parent, and perform a parent
1109 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1110 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1111 * have verified that no child exists while under i_mutex.
1113 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1114 struct qstr
*name
, struct nameidata
*nd
)
1116 struct inode
*inode
= parent
->d_inode
;
1117 struct dentry
*dentry
;
1120 /* Don't create child dentry for a dead directory. */
1121 if (unlikely(IS_DEADDIR(inode
)))
1122 return ERR_PTR(-ENOENT
);
1124 dentry
= d_alloc(parent
, name
);
1125 if (unlikely(!dentry
))
1126 return ERR_PTR(-ENOMEM
);
1128 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1129 if (unlikely(old
)) {
1137 * We already have a dentry, but require a lookup to be performed on the parent
1138 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1139 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1140 * child exists while under i_mutex.
1142 static struct dentry
*d_inode_lookup(struct dentry
*parent
, struct dentry
*dentry
,
1143 struct nameidata
*nd
)
1145 struct inode
*inode
= parent
->d_inode
;
1148 /* Don't create child dentry for a dead directory. */
1149 if (unlikely(IS_DEADDIR(inode
)))
1150 return ERR_PTR(-ENOENT
);
1152 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1153 if (unlikely(old
)) {
1161 * It's more convoluted than I'd like it to be, but... it's still fairly
1162 * small and for now I'd prefer to have fast path as straight as possible.
1163 * It _is_ time-critical.
1165 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1166 struct path
*path
, struct inode
**inode
)
1168 struct vfsmount
*mnt
= nd
->path
.mnt
;
1169 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1175 * Rename seqlock is not required here because in the off chance
1176 * of a false negative due to a concurrent rename, we're going to
1177 * do the non-racy lookup, below.
1179 if (nd
->flags
& LOOKUP_RCU
) {
1182 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1186 /* Memory barrier in read_seqcount_begin of child is enough */
1187 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1191 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1192 status
= d_revalidate(dentry
, nd
);
1193 if (unlikely(status
<= 0)) {
1194 if (status
!= -ECHILD
)
1199 if (unlikely(d_need_lookup(dentry
)))
1202 path
->dentry
= dentry
;
1203 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1205 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1209 if (unlazy_walk(nd
, dentry
))
1212 dentry
= __d_lookup(parent
, name
);
1215 if (dentry
&& unlikely(d_need_lookup(dentry
))) {
1220 if (unlikely(!dentry
)) {
1221 struct inode
*dir
= parent
->d_inode
;
1222 BUG_ON(nd
->inode
!= dir
);
1224 mutex_lock(&dir
->i_mutex
);
1225 dentry
= d_lookup(parent
, name
);
1226 if (likely(!dentry
)) {
1227 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1228 if (IS_ERR(dentry
)) {
1229 mutex_unlock(&dir
->i_mutex
);
1230 return PTR_ERR(dentry
);
1235 } else if (unlikely(d_need_lookup(dentry
))) {
1236 dentry
= d_inode_lookup(parent
, dentry
, nd
);
1237 if (IS_ERR(dentry
)) {
1238 mutex_unlock(&dir
->i_mutex
);
1239 return PTR_ERR(dentry
);
1245 mutex_unlock(&dir
->i_mutex
);
1247 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1248 status
= d_revalidate(dentry
, nd
);
1249 if (unlikely(status
<= 0)) {
1254 if (!d_invalidate(dentry
)) {
1263 path
->dentry
= dentry
;
1264 err
= follow_managed(path
, nd
->flags
);
1265 if (unlikely(err
< 0)) {
1266 path_put_conditional(path
, nd
);
1269 *inode
= path
->dentry
->d_inode
;
1273 static inline int may_lookup(struct nameidata
*nd
)
1275 if (nd
->flags
& LOOKUP_RCU
) {
1276 int err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1279 if (unlazy_walk(nd
, NULL
))
1282 return exec_permission(nd
->inode
, 0);
1285 static inline int handle_dots(struct nameidata
*nd
, int type
)
1287 if (type
== LAST_DOTDOT
) {
1288 if (nd
->flags
& LOOKUP_RCU
) {
1289 if (follow_dotdot_rcu(nd
))
1297 static void terminate_walk(struct nameidata
*nd
)
1299 if (!(nd
->flags
& LOOKUP_RCU
)) {
1300 path_put(&nd
->path
);
1302 nd
->flags
&= ~LOOKUP_RCU
;
1303 if (!(nd
->flags
& LOOKUP_ROOT
))
1304 nd
->root
.mnt
= NULL
;
1306 br_read_unlock(vfsmount_lock
);
1310 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1311 struct qstr
*name
, int type
, int follow
)
1313 struct inode
*inode
;
1316 * "." and ".." are special - ".." especially so because it has
1317 * to be able to know about the current root directory and
1318 * parent relationships.
1320 if (unlikely(type
!= LAST_NORM
))
1321 return handle_dots(nd
, type
);
1322 err
= do_lookup(nd
, name
, path
, &inode
);
1323 if (unlikely(err
)) {
1328 path_to_nameidata(path
, nd
);
1332 if (unlikely(inode
->i_op
->follow_link
) && follow
) {
1333 if (nd
->flags
& LOOKUP_RCU
) {
1334 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1339 BUG_ON(inode
!= path
->dentry
->d_inode
);
1342 path_to_nameidata(path
, nd
);
1348 * This limits recursive symlink follows to 8, while
1349 * limiting consecutive symlinks to 40.
1351 * Without that kind of total limit, nasty chains of consecutive
1352 * symlinks can cause almost arbitrarily long lookups.
1354 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1358 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1359 path_put_conditional(path
, nd
);
1360 path_put(&nd
->path
);
1363 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1366 current
->link_count
++;
1369 struct path link
= *path
;
1372 res
= follow_link(&link
, nd
, &cookie
);
1374 res
= walk_component(nd
, path
, &nd
->last
,
1375 nd
->last_type
, LOOKUP_FOLLOW
);
1376 put_link(nd
, &link
, cookie
);
1379 current
->link_count
--;
1386 * This is the basic name resolution function, turning a pathname into
1387 * the final dentry. We expect 'base' to be positive and a directory.
1389 * Returns 0 and nd will have valid dentry and mnt on success.
1390 * Returns error and drops reference to input namei data on failure.
1392 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1396 unsigned int lookup_flags
= nd
->flags
;
1403 /* At this point we know we have a real path component. */
1410 nd
->flags
|= LOOKUP_CONTINUE
;
1412 err
= may_lookup(nd
);
1417 c
= *(const unsigned char *)name
;
1419 hash
= init_name_hash();
1422 hash
= partial_name_hash(c
, hash
);
1423 c
= *(const unsigned char *)name
;
1424 } while (c
&& (c
!= '/'));
1425 this.len
= name
- (const char *) this.name
;
1426 this.hash
= end_name_hash(hash
);
1429 if (this.name
[0] == '.') switch (this.len
) {
1431 if (this.name
[1] == '.') {
1433 nd
->flags
|= LOOKUP_JUMPED
;
1439 if (likely(type
== LAST_NORM
)) {
1440 struct dentry
*parent
= nd
->path
.dentry
;
1441 nd
->flags
&= ~LOOKUP_JUMPED
;
1442 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1443 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1450 /* remove trailing slashes? */
1452 goto last_component
;
1453 while (*++name
== '/');
1455 goto last_component
;
1457 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1462 err
= nested_symlink(&next
, nd
);
1467 if (!nd
->inode
->i_op
->lookup
)
1470 /* here ends the main loop */
1473 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1474 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1476 nd
->last_type
= type
;
1483 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1484 struct nameidata
*nd
, struct file
**fp
)
1490 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1491 nd
->flags
= flags
| LOOKUP_JUMPED
;
1493 if (flags
& LOOKUP_ROOT
) {
1494 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1496 if (!inode
->i_op
->lookup
)
1498 retval
= inode_permission(inode
, MAY_EXEC
);
1502 nd
->path
= nd
->root
;
1504 if (flags
& LOOKUP_RCU
) {
1505 br_read_lock(vfsmount_lock
);
1507 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1509 path_get(&nd
->path
);
1514 nd
->root
.mnt
= NULL
;
1517 if (flags
& LOOKUP_RCU
) {
1518 br_read_lock(vfsmount_lock
);
1523 path_get(&nd
->root
);
1525 nd
->path
= nd
->root
;
1526 } else if (dfd
== AT_FDCWD
) {
1527 if (flags
& LOOKUP_RCU
) {
1528 struct fs_struct
*fs
= current
->fs
;
1531 br_read_lock(vfsmount_lock
);
1535 seq
= read_seqcount_begin(&fs
->seq
);
1537 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1538 } while (read_seqcount_retry(&fs
->seq
, seq
));
1540 get_fs_pwd(current
->fs
, &nd
->path
);
1543 struct dentry
*dentry
;
1545 file
= fget_raw_light(dfd
, &fput_needed
);
1550 dentry
= file
->f_path
.dentry
;
1554 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1557 retval
= exec_permission(dentry
->d_inode
, 0);
1562 nd
->path
= file
->f_path
;
1563 if (flags
& LOOKUP_RCU
) {
1566 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1567 br_read_lock(vfsmount_lock
);
1570 path_get(&file
->f_path
);
1571 fput_light(file
, fput_needed
);
1575 nd
->inode
= nd
->path
.dentry
->d_inode
;
1579 fput_light(file
, fput_needed
);
1584 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1586 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1587 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1589 nd
->flags
&= ~LOOKUP_PARENT
;
1590 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1591 nd
->flags
& LOOKUP_FOLLOW
);
1594 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1595 static int path_lookupat(int dfd
, const char *name
,
1596 unsigned int flags
, struct nameidata
*nd
)
1598 struct file
*base
= NULL
;
1603 * Path walking is largely split up into 2 different synchronisation
1604 * schemes, rcu-walk and ref-walk (explained in
1605 * Documentation/filesystems/path-lookup.txt). These share much of the
1606 * path walk code, but some things particularly setup, cleanup, and
1607 * following mounts are sufficiently divergent that functions are
1608 * duplicated. Typically there is a function foo(), and its RCU
1609 * analogue, foo_rcu().
1611 * -ECHILD is the error number of choice (just to avoid clashes) that
1612 * is returned if some aspect of an rcu-walk fails. Such an error must
1613 * be handled by restarting a traditional ref-walk (which will always
1614 * be able to complete).
1616 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1621 current
->total_link_count
= 0;
1622 err
= link_path_walk(name
, nd
);
1624 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1625 err
= lookup_last(nd
, &path
);
1628 struct path link
= path
;
1629 nd
->flags
|= LOOKUP_PARENT
;
1630 err
= follow_link(&link
, nd
, &cookie
);
1632 err
= lookup_last(nd
, &path
);
1633 put_link(nd
, &link
, cookie
);
1638 err
= complete_walk(nd
);
1640 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1641 if (!nd
->inode
->i_op
->lookup
) {
1642 path_put(&nd
->path
);
1650 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1651 path_put(&nd
->root
);
1652 nd
->root
.mnt
= NULL
;
1657 static int do_path_lookup(int dfd
, const char *name
,
1658 unsigned int flags
, struct nameidata
*nd
)
1660 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1661 if (unlikely(retval
== -ECHILD
))
1662 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1663 if (unlikely(retval
== -ESTALE
))
1664 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1666 if (likely(!retval
)) {
1667 if (unlikely(!audit_dummy_context())) {
1668 if (nd
->path
.dentry
&& nd
->inode
)
1669 audit_inode(name
, nd
->path
.dentry
);
1675 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1677 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1680 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1682 struct nameidata nd
;
1683 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1690 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1691 * @dentry: pointer to dentry of the base directory
1692 * @mnt: pointer to vfs mount of the base directory
1693 * @name: pointer to file name
1694 * @flags: lookup flags
1695 * @nd: pointer to nameidata
1697 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1698 const char *name
, unsigned int flags
,
1699 struct nameidata
*nd
)
1701 nd
->root
.dentry
= dentry
;
1703 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1704 return do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, nd
);
1707 static struct dentry
*__lookup_hash(struct qstr
*name
,
1708 struct dentry
*base
, struct nameidata
*nd
)
1710 struct inode
*inode
= base
->d_inode
;
1711 struct dentry
*dentry
;
1714 err
= exec_permission(inode
, 0);
1716 return ERR_PTR(err
);
1719 * Don't bother with __d_lookup: callers are for creat as
1720 * well as unlink, so a lot of the time it would cost
1723 dentry
= d_lookup(base
, name
);
1725 if (dentry
&& d_need_lookup(dentry
)) {
1727 * __lookup_hash is called with the parent dir's i_mutex already
1728 * held, so we are good to go here.
1730 dentry
= d_inode_lookup(base
, dentry
, nd
);
1735 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1736 dentry
= do_revalidate(dentry
, nd
);
1739 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1745 * Restricted form of lookup. Doesn't follow links, single-component only,
1746 * needs parent already locked. Doesn't follow mounts.
1749 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1751 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1755 * lookup_one_len - filesystem helper to lookup single pathname component
1756 * @name: pathname component to lookup
1757 * @base: base directory to lookup from
1758 * @len: maximum length @len should be interpreted to
1760 * Note that this routine is purely a helper for filesystem usage and should
1761 * not be called by generic code. Also note that by using this function the
1762 * nameidata argument is passed to the filesystem methods and a filesystem
1763 * using this helper needs to be prepared for that.
1765 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1771 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1776 return ERR_PTR(-EACCES
);
1778 hash
= init_name_hash();
1780 c
= *(const unsigned char *)name
++;
1781 if (c
== '/' || c
== '\0')
1782 return ERR_PTR(-EACCES
);
1783 hash
= partial_name_hash(c
, hash
);
1785 this.hash
= end_name_hash(hash
);
1787 * See if the low-level filesystem might want
1788 * to use its own hash..
1790 if (base
->d_flags
& DCACHE_OP_HASH
) {
1791 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1793 return ERR_PTR(err
);
1796 return __lookup_hash(&this, base
, NULL
);
1799 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1802 struct nameidata nd
;
1803 char *tmp
= getname_flags(name
, flags
);
1804 int err
= PTR_ERR(tmp
);
1807 BUG_ON(flags
& LOOKUP_PARENT
);
1809 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1817 static int user_path_parent(int dfd
, const char __user
*path
,
1818 struct nameidata
*nd
, char **name
)
1820 char *s
= getname(path
);
1826 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1836 * It's inline, so penalty for filesystems that don't use sticky bit is
1839 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1841 uid_t fsuid
= current_fsuid();
1843 if (!(dir
->i_mode
& S_ISVTX
))
1845 if (current_user_ns() != inode_userns(inode
))
1847 if (inode
->i_uid
== fsuid
)
1849 if (dir
->i_uid
== fsuid
)
1853 return !ns_capable(inode_userns(inode
), CAP_FOWNER
);
1857 * Check whether we can remove a link victim from directory dir, check
1858 * whether the type of victim is right.
1859 * 1. We can't do it if dir is read-only (done in permission())
1860 * 2. We should have write and exec permissions on dir
1861 * 3. We can't remove anything from append-only dir
1862 * 4. We can't do anything with immutable dir (done in permission())
1863 * 5. If the sticky bit on dir is set we should either
1864 * a. be owner of dir, or
1865 * b. be owner of victim, or
1866 * c. have CAP_FOWNER capability
1867 * 6. If the victim is append-only or immutable we can't do antyhing with
1868 * links pointing to it.
1869 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1870 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1871 * 9. We can't remove a root or mountpoint.
1872 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1873 * nfs_async_unlink().
1875 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1879 if (!victim
->d_inode
)
1882 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1883 audit_inode_child(victim
, dir
);
1885 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1890 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1891 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1894 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1896 if (IS_ROOT(victim
))
1898 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1900 if (IS_DEADDIR(dir
))
1902 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1907 /* Check whether we can create an object with dentry child in directory
1909 * 1. We can't do it if child already exists (open has special treatment for
1910 * this case, but since we are inlined it's OK)
1911 * 2. We can't do it if dir is read-only (done in permission())
1912 * 3. We should have write and exec permissions on dir
1913 * 4. We can't do it if dir is immutable (done in permission())
1915 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1919 if (IS_DEADDIR(dir
))
1921 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1925 * p1 and p2 should be directories on the same fs.
1927 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1932 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1936 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1938 p
= d_ancestor(p2
, p1
);
1940 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1941 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1945 p
= d_ancestor(p1
, p2
);
1947 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1948 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1952 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1953 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1957 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1959 mutex_unlock(&p1
->d_inode
->i_mutex
);
1961 mutex_unlock(&p2
->d_inode
->i_mutex
);
1962 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1966 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1967 struct nameidata
*nd
)
1969 int error
= may_create(dir
, dentry
);
1974 if (!dir
->i_op
->create
)
1975 return -EACCES
; /* shouldn't it be ENOSYS? */
1978 error
= security_inode_create(dir
, dentry
, mode
);
1981 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1983 fsnotify_create(dir
, dentry
);
1987 static int may_open(struct path
*path
, int acc_mode
, int flag
)
1989 struct dentry
*dentry
= path
->dentry
;
1990 struct inode
*inode
= dentry
->d_inode
;
2000 switch (inode
->i_mode
& S_IFMT
) {
2004 if (acc_mode
& MAY_WRITE
)
2009 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2018 error
= inode_permission(inode
, acc_mode
);
2023 * An append-only file must be opened in append mode for writing.
2025 if (IS_APPEND(inode
)) {
2026 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2032 /* O_NOATIME can only be set by the owner or superuser */
2033 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2037 * Ensure there are no outstanding leases on the file.
2039 return break_lease(inode
, flag
);
2042 static int handle_truncate(struct file
*filp
)
2044 struct path
*path
= &filp
->f_path
;
2045 struct inode
*inode
= path
->dentry
->d_inode
;
2046 int error
= get_write_access(inode
);
2050 * Refuse to truncate files with mandatory locks held on them.
2052 error
= locks_verify_locked(inode
);
2054 error
= security_path_truncate(path
);
2056 error
= do_truncate(path
->dentry
, 0,
2057 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2060 put_write_access(inode
);
2065 * Note that while the flag value (low two bits) for sys_open means:
2070 * it is changed into
2071 * 00 - no permissions needed
2072 * 01 - read-permission
2073 * 10 - write-permission
2075 * for the internal routines (ie open_namei()/follow_link() etc)
2076 * This is more logical, and also allows the 00 "no perm needed"
2077 * to be used for symlinks (where the permissions are checked
2081 static inline int open_to_namei_flags(int flag
)
2083 if ((flag
+1) & O_ACCMODE
)
2089 * Handle the last step of open()
2091 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2092 const struct open_flags
*op
, const char *pathname
)
2094 struct dentry
*dir
= nd
->path
.dentry
;
2095 struct dentry
*dentry
;
2096 int open_flag
= op
->open_flag
;
2097 int will_truncate
= open_flag
& O_TRUNC
;
2099 int acc_mode
= op
->acc_mode
;
2103 nd
->flags
&= ~LOOKUP_PARENT
;
2104 nd
->flags
|= op
->intent
;
2106 switch (nd
->last_type
) {
2109 error
= handle_dots(nd
, nd
->last_type
);
2111 return ERR_PTR(error
);
2114 error
= complete_walk(nd
);
2116 return ERR_PTR(error
);
2117 audit_inode(pathname
, nd
->path
.dentry
);
2118 if (open_flag
& O_CREAT
) {
2124 error
= complete_walk(nd
);
2126 return ERR_PTR(error
);
2127 audit_inode(pathname
, dir
);
2131 if (!(open_flag
& O_CREAT
)) {
2133 if (nd
->last
.name
[nd
->last
.len
])
2134 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2135 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2137 /* we _can_ be in RCU mode here */
2138 error
= walk_component(nd
, path
, &nd
->last
, LAST_NORM
,
2141 return ERR_PTR(error
);
2142 if (error
) /* symlink */
2145 error
= complete_walk(nd
);
2147 return ERR_PTR(-ECHILD
);
2150 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2151 if (!nd
->inode
->i_op
->lookup
)
2154 audit_inode(pathname
, nd
->path
.dentry
);
2158 /* create side of things */
2159 error
= complete_walk(nd
);
2161 return ERR_PTR(error
);
2163 audit_inode(pathname
, dir
);
2165 /* trailing slashes? */
2166 if (nd
->last
.name
[nd
->last
.len
])
2169 mutex_lock(&dir
->d_inode
->i_mutex
);
2171 dentry
= lookup_hash(nd
);
2172 error
= PTR_ERR(dentry
);
2173 if (IS_ERR(dentry
)) {
2174 mutex_unlock(&dir
->d_inode
->i_mutex
);
2178 path
->dentry
= dentry
;
2179 path
->mnt
= nd
->path
.mnt
;
2181 /* Negative dentry, just create the file */
2182 if (!dentry
->d_inode
) {
2183 int mode
= op
->mode
;
2184 if (!IS_POSIXACL(dir
->d_inode
))
2185 mode
&= ~current_umask();
2187 * This write is needed to ensure that a
2188 * rw->ro transition does not occur between
2189 * the time when the file is created and when
2190 * a permanent write count is taken through
2191 * the 'struct file' in nameidata_to_filp().
2193 error
= mnt_want_write(nd
->path
.mnt
);
2195 goto exit_mutex_unlock
;
2197 /* Don't check for write permission, don't truncate */
2198 open_flag
&= ~O_TRUNC
;
2200 acc_mode
= MAY_OPEN
;
2201 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2203 goto exit_mutex_unlock
;
2204 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2206 goto exit_mutex_unlock
;
2207 mutex_unlock(&dir
->d_inode
->i_mutex
);
2208 dput(nd
->path
.dentry
);
2209 nd
->path
.dentry
= dentry
;
2214 * It already exists.
2216 mutex_unlock(&dir
->d_inode
->i_mutex
);
2217 audit_inode(pathname
, path
->dentry
);
2220 if (open_flag
& O_EXCL
)
2223 error
= follow_managed(path
, nd
->flags
);
2228 if (!path
->dentry
->d_inode
)
2231 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2234 path_to_nameidata(path
, nd
);
2235 nd
->inode
= path
->dentry
->d_inode
;
2237 if (S_ISDIR(nd
->inode
->i_mode
))
2240 if (!S_ISREG(nd
->inode
->i_mode
))
2243 if (will_truncate
) {
2244 error
= mnt_want_write(nd
->path
.mnt
);
2250 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2253 filp
= nameidata_to_filp(nd
);
2254 if (!IS_ERR(filp
)) {
2255 error
= ima_file_check(filp
, op
->acc_mode
);
2258 filp
= ERR_PTR(error
);
2261 if (!IS_ERR(filp
)) {
2262 if (will_truncate
) {
2263 error
= handle_truncate(filp
);
2266 filp
= ERR_PTR(error
);
2272 mnt_drop_write(nd
->path
.mnt
);
2273 path_put(&nd
->path
);
2277 mutex_unlock(&dir
->d_inode
->i_mutex
);
2279 path_put_conditional(path
, nd
);
2281 filp
= ERR_PTR(error
);
2285 static struct file
*path_openat(int dfd
, const char *pathname
,
2286 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2288 struct file
*base
= NULL
;
2293 filp
= get_empty_filp();
2295 return ERR_PTR(-ENFILE
);
2297 filp
->f_flags
= op
->open_flag
;
2298 nd
->intent
.open
.file
= filp
;
2299 nd
->intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2300 nd
->intent
.open
.create_mode
= op
->mode
;
2302 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2303 if (unlikely(error
))
2306 current
->total_link_count
= 0;
2307 error
= link_path_walk(pathname
, nd
);
2308 if (unlikely(error
))
2311 filp
= do_last(nd
, &path
, op
, pathname
);
2312 while (unlikely(!filp
)) { /* trailing symlink */
2313 struct path link
= path
;
2315 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2316 path_put_conditional(&path
, nd
);
2317 path_put(&nd
->path
);
2318 filp
= ERR_PTR(-ELOOP
);
2321 nd
->flags
|= LOOKUP_PARENT
;
2322 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2323 error
= follow_link(&link
, nd
, &cookie
);
2324 if (unlikely(error
))
2325 filp
= ERR_PTR(error
);
2327 filp
= do_last(nd
, &path
, op
, pathname
);
2328 put_link(nd
, &link
, cookie
);
2331 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2332 path_put(&nd
->root
);
2335 release_open_intent(nd
);
2339 filp
= ERR_PTR(error
);
2343 struct file
*do_filp_open(int dfd
, const char *pathname
,
2344 const struct open_flags
*op
, int flags
)
2346 struct nameidata nd
;
2349 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2350 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2351 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2352 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2353 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2357 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2358 const char *name
, const struct open_flags
*op
, int flags
)
2360 struct nameidata nd
;
2364 nd
.root
.dentry
= dentry
;
2366 flags
|= LOOKUP_ROOT
;
2368 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2369 return ERR_PTR(-ELOOP
);
2371 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2372 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2373 file
= path_openat(-1, name
, &nd
, op
, flags
);
2374 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2375 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2380 * lookup_create - lookup a dentry, creating it if it doesn't exist
2381 * @nd: nameidata info
2382 * @is_dir: directory flag
2384 * Simple function to lookup and return a dentry and create it
2385 * if it doesn't exist. Is SMP-safe.
2387 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2389 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2391 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2393 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2395 * Yucky last component or no last component at all?
2396 * (foo/., foo/.., /////)
2398 if (nd
->last_type
!= LAST_NORM
)
2400 nd
->flags
&= ~LOOKUP_PARENT
;
2401 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2402 nd
->intent
.open
.flags
= O_EXCL
;
2405 * Do the final lookup.
2407 dentry
= lookup_hash(nd
);
2411 if (dentry
->d_inode
)
2414 * Special case - lookup gave negative, but... we had foo/bar/
2415 * From the vfs_mknod() POV we just have a negative dentry -
2416 * all is fine. Let's be bastards - you had / on the end, you've
2417 * been asking for (non-existent) directory. -ENOENT for you.
2419 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2421 dentry
= ERR_PTR(-ENOENT
);
2426 dentry
= ERR_PTR(-EEXIST
);
2430 EXPORT_SYMBOL_GPL(lookup_create
);
2432 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2434 int error
= may_create(dir
, dentry
);
2439 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) &&
2440 !ns_capable(inode_userns(dir
), CAP_MKNOD
))
2443 if (!dir
->i_op
->mknod
)
2446 error
= devcgroup_inode_mknod(mode
, dev
);
2450 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2454 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2456 fsnotify_create(dir
, dentry
);
2460 static int may_mknod(mode_t mode
)
2462 switch (mode
& S_IFMT
) {
2468 case 0: /* zero mode translates to S_IFREG */
2477 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2482 struct dentry
*dentry
;
2483 struct nameidata nd
;
2488 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2492 dentry
= lookup_create(&nd
, 0);
2493 if (IS_ERR(dentry
)) {
2494 error
= PTR_ERR(dentry
);
2497 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2498 mode
&= ~current_umask();
2499 error
= may_mknod(mode
);
2502 error
= mnt_want_write(nd
.path
.mnt
);
2505 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2507 goto out_drop_write
;
2508 switch (mode
& S_IFMT
) {
2509 case 0: case S_IFREG
:
2510 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2512 case S_IFCHR
: case S_IFBLK
:
2513 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2514 new_decode_dev(dev
));
2516 case S_IFIFO
: case S_IFSOCK
:
2517 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2521 mnt_drop_write(nd
.path
.mnt
);
2525 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2532 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2534 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2537 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2539 int error
= may_create(dir
, dentry
);
2544 if (!dir
->i_op
->mkdir
)
2547 mode
&= (S_IRWXUGO
|S_ISVTX
);
2548 error
= security_inode_mkdir(dir
, dentry
, mode
);
2552 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2554 fsnotify_mkdir(dir
, dentry
);
2558 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2562 struct dentry
*dentry
;
2563 struct nameidata nd
;
2565 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2569 dentry
= lookup_create(&nd
, 1);
2570 error
= PTR_ERR(dentry
);
2574 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2575 mode
&= ~current_umask();
2576 error
= mnt_want_write(nd
.path
.mnt
);
2579 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2581 goto out_drop_write
;
2582 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2584 mnt_drop_write(nd
.path
.mnt
);
2588 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2595 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2597 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2601 * The dentry_unhash() helper will try to drop the dentry early: we
2602 * should have a usage count of 2 if we're the only user of this
2603 * dentry, and if that is true (possibly after pruning the dcache),
2604 * then we drop the dentry now.
2606 * A low-level filesystem can, if it choses, legally
2609 * if (!d_unhashed(dentry))
2612 * if it cannot handle the case of removing a directory
2613 * that is still in use by something else..
2615 void dentry_unhash(struct dentry
*dentry
)
2617 shrink_dcache_parent(dentry
);
2618 spin_lock(&dentry
->d_lock
);
2619 if (dentry
->d_count
== 1)
2621 spin_unlock(&dentry
->d_lock
);
2624 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2626 int error
= may_delete(dir
, dentry
, 1);
2631 if (!dir
->i_op
->rmdir
)
2634 mutex_lock(&dentry
->d_inode
->i_mutex
);
2637 if (d_mountpoint(dentry
))
2640 error
= security_inode_rmdir(dir
, dentry
);
2644 shrink_dcache_parent(dentry
);
2645 error
= dir
->i_op
->rmdir(dir
, dentry
);
2649 dentry
->d_inode
->i_flags
|= S_DEAD
;
2653 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2659 static long do_rmdir(int dfd
, const char __user
*pathname
)
2663 struct dentry
*dentry
;
2664 struct nameidata nd
;
2666 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2670 switch(nd
.last_type
) {
2682 nd
.flags
&= ~LOOKUP_PARENT
;
2684 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2685 dentry
= lookup_hash(&nd
);
2686 error
= PTR_ERR(dentry
);
2689 if (!dentry
->d_inode
) {
2693 error
= mnt_want_write(nd
.path
.mnt
);
2696 error
= security_path_rmdir(&nd
.path
, dentry
);
2699 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2701 mnt_drop_write(nd
.path
.mnt
);
2705 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2712 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2714 return do_rmdir(AT_FDCWD
, pathname
);
2717 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2719 int error
= may_delete(dir
, dentry
, 0);
2724 if (!dir
->i_op
->unlink
)
2727 mutex_lock(&dentry
->d_inode
->i_mutex
);
2728 if (d_mountpoint(dentry
))
2731 error
= security_inode_unlink(dir
, dentry
);
2733 error
= dir
->i_op
->unlink(dir
, dentry
);
2738 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2740 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2741 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2742 fsnotify_link_count(dentry
->d_inode
);
2750 * Make sure that the actual truncation of the file will occur outside its
2751 * directory's i_mutex. Truncate can take a long time if there is a lot of
2752 * writeout happening, and we don't want to prevent access to the directory
2753 * while waiting on the I/O.
2755 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2759 struct dentry
*dentry
;
2760 struct nameidata nd
;
2761 struct inode
*inode
= NULL
;
2763 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2768 if (nd
.last_type
!= LAST_NORM
)
2771 nd
.flags
&= ~LOOKUP_PARENT
;
2773 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2774 dentry
= lookup_hash(&nd
);
2775 error
= PTR_ERR(dentry
);
2776 if (!IS_ERR(dentry
)) {
2777 /* Why not before? Because we want correct error value */
2778 if (nd
.last
.name
[nd
.last
.len
])
2780 inode
= dentry
->d_inode
;
2784 error
= mnt_want_write(nd
.path
.mnt
);
2787 error
= security_path_unlink(&nd
.path
, dentry
);
2790 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2792 mnt_drop_write(nd
.path
.mnt
);
2796 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2798 iput(inode
); /* truncate the inode here */
2805 error
= !dentry
->d_inode
? -ENOENT
:
2806 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2810 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2812 if ((flag
& ~AT_REMOVEDIR
) != 0)
2815 if (flag
& AT_REMOVEDIR
)
2816 return do_rmdir(dfd
, pathname
);
2818 return do_unlinkat(dfd
, pathname
);
2821 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2823 return do_unlinkat(AT_FDCWD
, pathname
);
2826 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2828 int error
= may_create(dir
, dentry
);
2833 if (!dir
->i_op
->symlink
)
2836 error
= security_inode_symlink(dir
, dentry
, oldname
);
2840 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2842 fsnotify_create(dir
, dentry
);
2846 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2847 int, newdfd
, const char __user
*, newname
)
2852 struct dentry
*dentry
;
2853 struct nameidata nd
;
2855 from
= getname(oldname
);
2857 return PTR_ERR(from
);
2859 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2863 dentry
= lookup_create(&nd
, 0);
2864 error
= PTR_ERR(dentry
);
2868 error
= mnt_want_write(nd
.path
.mnt
);
2871 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2873 goto out_drop_write
;
2874 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2876 mnt_drop_write(nd
.path
.mnt
);
2880 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2888 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2890 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2893 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2895 struct inode
*inode
= old_dentry
->d_inode
;
2901 error
= may_create(dir
, new_dentry
);
2905 if (dir
->i_sb
!= inode
->i_sb
)
2909 * A link to an append-only or immutable file cannot be created.
2911 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2913 if (!dir
->i_op
->link
)
2915 if (S_ISDIR(inode
->i_mode
))
2918 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2922 mutex_lock(&inode
->i_mutex
);
2923 /* Make sure we don't allow creating hardlink to an unlinked file */
2924 if (inode
->i_nlink
== 0)
2927 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2928 mutex_unlock(&inode
->i_mutex
);
2930 fsnotify_link(dir
, inode
, new_dentry
);
2935 * Hardlinks are often used in delicate situations. We avoid
2936 * security-related surprises by not following symlinks on the
2939 * We don't follow them on the oldname either to be compatible
2940 * with linux 2.0, and to avoid hard-linking to directories
2941 * and other special files. --ADM
2943 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2944 int, newdfd
, const char __user
*, newname
, int, flags
)
2946 struct dentry
*new_dentry
;
2947 struct nameidata nd
;
2948 struct path old_path
;
2953 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
2956 * To use null names we require CAP_DAC_READ_SEARCH
2957 * This ensures that not everyone will be able to create
2958 * handlink using the passed filedescriptor.
2960 if (flags
& AT_EMPTY_PATH
) {
2961 if (!capable(CAP_DAC_READ_SEARCH
))
2966 if (flags
& AT_SYMLINK_FOLLOW
)
2967 how
|= LOOKUP_FOLLOW
;
2969 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
2973 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2977 if (old_path
.mnt
!= nd
.path
.mnt
)
2979 new_dentry
= lookup_create(&nd
, 0);
2980 error
= PTR_ERR(new_dentry
);
2981 if (IS_ERR(new_dentry
))
2983 error
= mnt_want_write(nd
.path
.mnt
);
2986 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2988 goto out_drop_write
;
2989 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2991 mnt_drop_write(nd
.path
.mnt
);
2995 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3000 path_put(&old_path
);
3005 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3007 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3011 * The worst of all namespace operations - renaming directory. "Perverted"
3012 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3014 * a) we can get into loop creation. Check is done in is_subdir().
3015 * b) race potential - two innocent renames can create a loop together.
3016 * That's where 4.4 screws up. Current fix: serialization on
3017 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3019 * c) we have to lock _three_ objects - parents and victim (if it exists).
3020 * And that - after we got ->i_mutex on parents (until then we don't know
3021 * whether the target exists). Solution: try to be smart with locking
3022 * order for inodes. We rely on the fact that tree topology may change
3023 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3024 * move will be locked. Thus we can rank directories by the tree
3025 * (ancestors first) and rank all non-directories after them.
3026 * That works since everybody except rename does "lock parent, lookup,
3027 * lock child" and rename is under ->s_vfs_rename_mutex.
3028 * HOWEVER, it relies on the assumption that any object with ->lookup()
3029 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3030 * we'd better make sure that there's no link(2) for them.
3031 * d) conversion from fhandle to dentry may come in the wrong moment - when
3032 * we are removing the target. Solution: we will have to grab ->i_mutex
3033 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3034 * ->i_mutex on parents, which works but leads to some truly excessive
3037 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3038 struct inode
*new_dir
, struct dentry
*new_dentry
)
3041 struct inode
*target
= new_dentry
->d_inode
;
3044 * If we are going to change the parent - check write permissions,
3045 * we'll need to flip '..'.
3047 if (new_dir
!= old_dir
) {
3048 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3053 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3058 mutex_lock(&target
->i_mutex
);
3061 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3065 shrink_dcache_parent(new_dentry
);
3066 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3071 target
->i_flags
|= S_DEAD
;
3072 dont_mount(new_dentry
);
3076 mutex_unlock(&target
->i_mutex
);
3078 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3079 d_move(old_dentry
,new_dentry
);
3083 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3084 struct inode
*new_dir
, struct dentry
*new_dentry
)
3086 struct inode
*target
= new_dentry
->d_inode
;
3089 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3095 mutex_lock(&target
->i_mutex
);
3098 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3101 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3106 dont_mount(new_dentry
);
3107 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3108 d_move(old_dentry
, new_dentry
);
3111 mutex_unlock(&target
->i_mutex
);
3116 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3117 struct inode
*new_dir
, struct dentry
*new_dentry
)
3120 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3121 const unsigned char *old_name
;
3123 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3126 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3130 if (!new_dentry
->d_inode
)
3131 error
= may_create(new_dir
, new_dentry
);
3133 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3137 if (!old_dir
->i_op
->rename
)
3140 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3143 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3145 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3147 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3148 new_dentry
->d_inode
, old_dentry
);
3149 fsnotify_oldname_free(old_name
);
3154 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3155 int, newdfd
, const char __user
*, newname
)
3157 struct dentry
*old_dir
, *new_dir
;
3158 struct dentry
*old_dentry
, *new_dentry
;
3159 struct dentry
*trap
;
3160 struct nameidata oldnd
, newnd
;
3165 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3169 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3174 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3177 old_dir
= oldnd
.path
.dentry
;
3179 if (oldnd
.last_type
!= LAST_NORM
)
3182 new_dir
= newnd
.path
.dentry
;
3183 if (newnd
.last_type
!= LAST_NORM
)
3186 oldnd
.flags
&= ~LOOKUP_PARENT
;
3187 newnd
.flags
&= ~LOOKUP_PARENT
;
3188 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3190 trap
= lock_rename(new_dir
, old_dir
);
3192 old_dentry
= lookup_hash(&oldnd
);
3193 error
= PTR_ERR(old_dentry
);
3194 if (IS_ERR(old_dentry
))
3196 /* source must exist */
3198 if (!old_dentry
->d_inode
)
3200 /* unless the source is a directory trailing slashes give -ENOTDIR */
3201 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3203 if (oldnd
.last
.name
[oldnd
.last
.len
])
3205 if (newnd
.last
.name
[newnd
.last
.len
])
3208 /* source should not be ancestor of target */
3210 if (old_dentry
== trap
)
3212 new_dentry
= lookup_hash(&newnd
);
3213 error
= PTR_ERR(new_dentry
);
3214 if (IS_ERR(new_dentry
))
3216 /* target should not be an ancestor of source */
3218 if (new_dentry
== trap
)
3221 error
= mnt_want_write(oldnd
.path
.mnt
);
3224 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3225 &newnd
.path
, new_dentry
);
3228 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3229 new_dir
->d_inode
, new_dentry
);
3231 mnt_drop_write(oldnd
.path
.mnt
);
3237 unlock_rename(new_dir
, old_dir
);
3239 path_put(&newnd
.path
);
3242 path_put(&oldnd
.path
);
3248 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3250 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3253 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3257 len
= PTR_ERR(link
);
3262 if (len
> (unsigned) buflen
)
3264 if (copy_to_user(buffer
, link
, len
))
3271 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3272 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3273 * using) it for any given inode is up to filesystem.
3275 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3277 struct nameidata nd
;
3282 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3284 return PTR_ERR(cookie
);
3286 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3287 if (dentry
->d_inode
->i_op
->put_link
)
3288 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3292 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3294 return __vfs_follow_link(nd
, link
);
3297 /* get the link contents into pagecache */
3298 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3302 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3303 page
= read_mapping_page(mapping
, 0, NULL
);
3308 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3312 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3314 struct page
*page
= NULL
;
3315 char *s
= page_getlink(dentry
, &page
);
3316 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3319 page_cache_release(page
);
3324 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3326 struct page
*page
= NULL
;
3327 nd_set_link(nd
, page_getlink(dentry
, &page
));
3331 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3333 struct page
*page
= cookie
;
3337 page_cache_release(page
);
3342 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3344 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3346 struct address_space
*mapping
= inode
->i_mapping
;
3351 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3353 flags
|= AOP_FLAG_NOFS
;
3356 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3357 flags
, &page
, &fsdata
);
3361 kaddr
= kmap_atomic(page
, KM_USER0
);
3362 memcpy(kaddr
, symname
, len
-1);
3363 kunmap_atomic(kaddr
, KM_USER0
);
3365 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3372 mark_inode_dirty(inode
);
3378 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3380 return __page_symlink(inode
, symname
, len
,
3381 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3384 const struct inode_operations page_symlink_inode_operations
= {
3385 .readlink
= generic_readlink
,
3386 .follow_link
= page_follow_link_light
,
3387 .put_link
= page_put_link
,
3390 EXPORT_SYMBOL(user_path_at
);
3391 EXPORT_SYMBOL(follow_down_one
);
3392 EXPORT_SYMBOL(follow_down
);
3393 EXPORT_SYMBOL(follow_up
);
3394 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3395 EXPORT_SYMBOL(getname
);
3396 EXPORT_SYMBOL(lock_rename
);
3397 EXPORT_SYMBOL(lookup_one_len
);
3398 EXPORT_SYMBOL(page_follow_link_light
);
3399 EXPORT_SYMBOL(page_put_link
);
3400 EXPORT_SYMBOL(page_readlink
);
3401 EXPORT_SYMBOL(__page_symlink
);
3402 EXPORT_SYMBOL(page_symlink
);
3403 EXPORT_SYMBOL(page_symlink_inode_operations
);
3404 EXPORT_SYMBOL(kern_path_parent
);
3405 EXPORT_SYMBOL(kern_path
);
3406 EXPORT_SYMBOL(vfs_path_lookup
);
3407 EXPORT_SYMBOL(inode_permission
);
3408 EXPORT_SYMBOL(file_permission
);
3409 EXPORT_SYMBOL(unlock_rename
);
3410 EXPORT_SYMBOL(vfs_create
);
3411 EXPORT_SYMBOL(vfs_follow_link
);
3412 EXPORT_SYMBOL(vfs_link
);
3413 EXPORT_SYMBOL(vfs_mkdir
);
3414 EXPORT_SYMBOL(vfs_mknod
);
3415 EXPORT_SYMBOL(generic_permission
);
3416 EXPORT_SYMBOL(vfs_readlink
);
3417 EXPORT_SYMBOL(vfs_rename
);
3418 EXPORT_SYMBOL(vfs_rmdir
);
3419 EXPORT_SYMBOL(vfs_symlink
);
3420 EXPORT_SYMBOL(vfs_unlink
);
3421 EXPORT_SYMBOL(dentry_unhash
);
3422 EXPORT_SYMBOL(generic_readlink
);