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 if at least one exec bit is set.
243 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
244 if (ns_capable(inode_userns(inode
), CAP_DAC_OVERRIDE
))
248 * Searching includes executable on directories, else just read.
250 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
251 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
252 if (ns_capable(inode_userns(inode
), CAP_DAC_READ_SEARCH
))
259 * inode_permission - check for access rights to a given inode
260 * @inode: inode to check permission on
261 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
263 * Used to check for read/write/execute permissions on an inode.
264 * We use "fsuid" for this, letting us set arbitrary permissions
265 * for filesystem access without changing the "normal" uids which
266 * are used for other things.
268 int inode_permission(struct inode
*inode
, int mask
)
272 if (mask
& MAY_WRITE
) {
273 umode_t mode
= inode
->i_mode
;
276 * Nobody gets write access to a read-only fs.
278 if (IS_RDONLY(inode
) &&
279 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
283 * Nobody gets write access to an immutable file.
285 if (IS_IMMUTABLE(inode
))
289 if (inode
->i_op
->permission
)
290 retval
= inode
->i_op
->permission(inode
, mask
, 0);
292 retval
= generic_permission(inode
, mask
, 0,
293 inode
->i_op
->check_acl
);
298 retval
= devcgroup_inode_permission(inode
, mask
);
302 return security_inode_permission(inode
, mask
);
306 * file_permission - check for additional access rights to a given file
307 * @file: file to check access rights for
308 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
310 * Used to check for read/write/execute permissions on an already opened
314 * Do not use this function in new code. All access checks should
315 * be done using inode_permission().
317 int file_permission(struct file
*file
, int mask
)
319 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
323 * get_write_access() gets write permission for a file.
324 * put_write_access() releases this write permission.
325 * This is used for regular files.
326 * We cannot support write (and maybe mmap read-write shared) accesses and
327 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
328 * can have the following values:
329 * 0: no writers, no VM_DENYWRITE mappings
330 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
331 * > 0: (i_writecount) users are writing to the file.
333 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
334 * except for the cases where we don't hold i_writecount yet. Then we need to
335 * use {get,deny}_write_access() - these functions check the sign and refuse
336 * to do the change if sign is wrong. Exclusion between them is provided by
337 * the inode->i_lock spinlock.
340 int get_write_access(struct inode
* inode
)
342 spin_lock(&inode
->i_lock
);
343 if (atomic_read(&inode
->i_writecount
) < 0) {
344 spin_unlock(&inode
->i_lock
);
347 atomic_inc(&inode
->i_writecount
);
348 spin_unlock(&inode
->i_lock
);
353 int deny_write_access(struct file
* file
)
355 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
357 spin_lock(&inode
->i_lock
);
358 if (atomic_read(&inode
->i_writecount
) > 0) {
359 spin_unlock(&inode
->i_lock
);
362 atomic_dec(&inode
->i_writecount
);
363 spin_unlock(&inode
->i_lock
);
369 * path_get - get a reference to a path
370 * @path: path to get the reference to
372 * Given a path increment the reference count to the dentry and the vfsmount.
374 void path_get(struct path
*path
)
379 EXPORT_SYMBOL(path_get
);
382 * path_put - put a reference to a path
383 * @path: path to put the reference to
385 * Given a path decrement the reference count to the dentry and the vfsmount.
387 void path_put(struct path
*path
)
392 EXPORT_SYMBOL(path_put
);
395 * Path walking has 2 modes, rcu-walk and ref-walk (see
396 * Documentation/filesystems/path-lookup.txt). In situations when we can't
397 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
398 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
399 * mode. Refcounts are grabbed at the last known good point before rcu-walk
400 * got stuck, so ref-walk may continue from there. If this is not successful
401 * (eg. a seqcount has changed), then failure is returned and it's up to caller
402 * to restart the path walk from the beginning in ref-walk mode.
406 * unlazy_walk - try to switch to ref-walk mode.
407 * @nd: nameidata pathwalk data
408 * @dentry: child of nd->path.dentry or NULL
409 * Returns: 0 on success, -ECHILD on failure
411 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
412 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
413 * @nd or NULL. Must be called from rcu-walk context.
415 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
417 struct fs_struct
*fs
= current
->fs
;
418 struct dentry
*parent
= nd
->path
.dentry
;
421 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
422 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
424 spin_lock(&fs
->lock
);
425 if (nd
->root
.mnt
!= fs
->root
.mnt
||
426 nd
->root
.dentry
!= fs
->root
.dentry
)
429 spin_lock(&parent
->d_lock
);
431 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
433 BUG_ON(nd
->inode
!= parent
->d_inode
);
435 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
436 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
439 * If the sequence check on the child dentry passed, then
440 * the child has not been removed from its parent. This
441 * means the parent dentry must be valid and able to take
442 * a reference at this point.
444 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
445 BUG_ON(!parent
->d_count
);
447 spin_unlock(&dentry
->d_lock
);
449 spin_unlock(&parent
->d_lock
);
452 spin_unlock(&fs
->lock
);
454 mntget(nd
->path
.mnt
);
457 br_read_unlock(vfsmount_lock
);
458 nd
->flags
&= ~LOOKUP_RCU
;
462 spin_unlock(&dentry
->d_lock
);
464 spin_unlock(&parent
->d_lock
);
467 spin_unlock(&fs
->lock
);
472 * release_open_intent - free up open intent resources
473 * @nd: pointer to nameidata
475 void release_open_intent(struct nameidata
*nd
)
477 struct file
*file
= nd
->intent
.open
.file
;
479 if (file
&& !IS_ERR(file
)) {
480 if (file
->f_path
.dentry
== NULL
)
487 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
489 return dentry
->d_op
->d_revalidate(dentry
, nd
);
492 static struct dentry
*
493 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
495 int status
= d_revalidate(dentry
, nd
);
496 if (unlikely(status
<= 0)) {
498 * The dentry failed validation.
499 * If d_revalidate returned 0 attempt to invalidate
500 * the dentry otherwise d_revalidate is asking us
501 * to return a fail status.
505 dentry
= ERR_PTR(status
);
506 } else if (!d_invalidate(dentry
)) {
515 * complete_walk - successful completion of path walk
516 * @nd: pointer nameidata
518 * If we had been in RCU mode, drop out of it and legitimize nd->path.
519 * Revalidate the final result, unless we'd already done that during
520 * the path walk or the filesystem doesn't ask for it. Return 0 on
521 * success, -error on failure. In case of failure caller does not
522 * need to drop nd->path.
524 static int complete_walk(struct nameidata
*nd
)
526 struct dentry
*dentry
= nd
->path
.dentry
;
529 if (nd
->flags
& LOOKUP_RCU
) {
530 nd
->flags
&= ~LOOKUP_RCU
;
531 if (!(nd
->flags
& LOOKUP_ROOT
))
533 spin_lock(&dentry
->d_lock
);
534 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
535 spin_unlock(&dentry
->d_lock
);
537 br_read_unlock(vfsmount_lock
);
540 BUG_ON(nd
->inode
!= dentry
->d_inode
);
541 spin_unlock(&dentry
->d_lock
);
542 mntget(nd
->path
.mnt
);
544 br_read_unlock(vfsmount_lock
);
547 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
550 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
553 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
556 /* Note: we do not d_invalidate() */
557 status
= d_revalidate(dentry
, nd
);
569 * Short-cut version of permission(), for calling on directories
570 * during pathname resolution. Combines parts of permission()
571 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
573 * If appropriate, check DAC only. If not appropriate, or
574 * short-cut DAC fails, then call ->permission() to do more
575 * complete permission check.
577 static inline int exec_permission(struct inode
*inode
, unsigned int flags
)
580 struct user_namespace
*ns
= inode_userns(inode
);
582 if (inode
->i_op
->permission
) {
583 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
, flags
);
585 ret
= acl_permission_check(inode
, MAY_EXEC
, flags
,
586 inode
->i_op
->check_acl
);
593 if (ns_capable(ns
, CAP_DAC_OVERRIDE
) ||
594 ns_capable(ns
, CAP_DAC_READ_SEARCH
))
599 return security_inode_exec_permission(inode
, flags
);
602 static __always_inline
void set_root(struct nameidata
*nd
)
605 get_fs_root(current
->fs
, &nd
->root
);
608 static int link_path_walk(const char *, struct nameidata
*);
610 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
613 struct fs_struct
*fs
= current
->fs
;
617 seq
= read_seqcount_begin(&fs
->seq
);
619 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
620 } while (read_seqcount_retry(&fs
->seq
, seq
));
624 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
636 nd
->flags
|= LOOKUP_JUMPED
;
638 nd
->inode
= nd
->path
.dentry
->d_inode
;
640 ret
= link_path_walk(link
, nd
);
644 return PTR_ERR(link
);
647 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
650 if (path
->mnt
!= nd
->path
.mnt
)
654 static inline void path_to_nameidata(const struct path
*path
,
655 struct nameidata
*nd
)
657 if (!(nd
->flags
& LOOKUP_RCU
)) {
658 dput(nd
->path
.dentry
);
659 if (nd
->path
.mnt
!= path
->mnt
)
660 mntput(nd
->path
.mnt
);
662 nd
->path
.mnt
= path
->mnt
;
663 nd
->path
.dentry
= path
->dentry
;
666 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
668 struct inode
*inode
= link
->dentry
->d_inode
;
669 if (!IS_ERR(cookie
) && inode
->i_op
->put_link
)
670 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
674 static __always_inline
int
675 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
678 struct dentry
*dentry
= link
->dentry
;
680 BUG_ON(nd
->flags
& LOOKUP_RCU
);
682 if (link
->mnt
== nd
->path
.mnt
)
685 if (unlikely(current
->total_link_count
>= 40)) {
686 *p
= ERR_PTR(-ELOOP
); /* no ->put_link(), please */
691 current
->total_link_count
++;
693 touch_atime(link
->mnt
, dentry
);
694 nd_set_link(nd
, NULL
);
696 error
= security_inode_follow_link(link
->dentry
, nd
);
698 *p
= ERR_PTR(error
); /* no ->put_link(), please */
703 nd
->last_type
= LAST_BIND
;
704 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
707 char *s
= nd_get_link(nd
);
710 error
= __vfs_follow_link(nd
, s
);
711 else if (nd
->last_type
== LAST_BIND
) {
712 nd
->flags
|= LOOKUP_JUMPED
;
713 nd
->inode
= nd
->path
.dentry
->d_inode
;
714 if (nd
->inode
->i_op
->follow_link
) {
715 /* stepped on a _really_ weird one */
724 static int follow_up_rcu(struct path
*path
)
726 struct vfsmount
*parent
;
727 struct dentry
*mountpoint
;
729 parent
= path
->mnt
->mnt_parent
;
730 if (parent
== path
->mnt
)
732 mountpoint
= path
->mnt
->mnt_mountpoint
;
733 path
->dentry
= mountpoint
;
738 int follow_up(struct path
*path
)
740 struct vfsmount
*parent
;
741 struct dentry
*mountpoint
;
743 br_read_lock(vfsmount_lock
);
744 parent
= path
->mnt
->mnt_parent
;
745 if (parent
== path
->mnt
) {
746 br_read_unlock(vfsmount_lock
);
750 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
751 br_read_unlock(vfsmount_lock
);
753 path
->dentry
= mountpoint
;
760 * Perform an automount
761 * - return -EISDIR to tell follow_managed() to stop and return the path we
764 static int follow_automount(struct path
*path
, unsigned flags
,
767 struct vfsmount
*mnt
;
770 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
773 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
774 * and this is the terminal part of the path.
776 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
777 return -EISDIR
; /* we actually want to stop here */
779 /* We want to mount if someone is trying to open/create a file of any
780 * type under the mountpoint, wants to traverse through the mountpoint
781 * or wants to open the mounted directory.
783 * We don't want to mount if someone's just doing a stat and they've
784 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
785 * appended a '/' to the name.
787 if (!(flags
& LOOKUP_FOLLOW
) &&
788 !(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
789 LOOKUP_OPEN
| LOOKUP_CREATE
)))
792 current
->total_link_count
++;
793 if (current
->total_link_count
>= 40)
796 mnt
= path
->dentry
->d_op
->d_automount(path
);
799 * The filesystem is allowed to return -EISDIR here to indicate
800 * it doesn't want to automount. For instance, autofs would do
801 * this so that its userspace daemon can mount on this dentry.
803 * However, we can only permit this if it's a terminal point in
804 * the path being looked up; if it wasn't then the remainder of
805 * the path is inaccessible and we should say so.
807 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_CONTINUE
))
812 if (!mnt
) /* mount collision */
815 err
= finish_automount(mnt
, path
);
819 /* Someone else made a mount here whilst we were busy */
826 path
->dentry
= dget(mnt
->mnt_root
);
836 * Handle a dentry that is managed in some way.
837 * - Flagged for transit management (autofs)
838 * - Flagged as mountpoint
839 * - Flagged as automount point
841 * This may only be called in refwalk mode.
843 * Serialization is taken care of in namespace.c
845 static int follow_managed(struct path
*path
, unsigned flags
)
848 bool need_mntput
= false;
851 /* Given that we're not holding a lock here, we retain the value in a
852 * local variable for each dentry as we look at it so that we don't see
853 * the components of that value change under us */
854 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
855 managed
&= DCACHE_MANAGED_DENTRY
,
856 unlikely(managed
!= 0)) {
857 /* Allow the filesystem to manage the transit without i_mutex
859 if (managed
& DCACHE_MANAGE_TRANSIT
) {
860 BUG_ON(!path
->dentry
->d_op
);
861 BUG_ON(!path
->dentry
->d_op
->d_manage
);
862 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
864 return ret
== -EISDIR
? 0 : ret
;
867 /* Transit to a mounted filesystem. */
868 if (managed
& DCACHE_MOUNTED
) {
869 struct vfsmount
*mounted
= lookup_mnt(path
);
875 path
->dentry
= dget(mounted
->mnt_root
);
880 /* Something is mounted on this dentry in another
881 * namespace and/or whatever was mounted there in this
882 * namespace got unmounted before we managed to get the
886 /* Handle an automount point */
887 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
888 ret
= follow_automount(path
, flags
, &need_mntput
);
890 return ret
== -EISDIR
? 0 : ret
;
894 /* We didn't change the current path point */
900 int follow_down_one(struct path
*path
)
902 struct vfsmount
*mounted
;
904 mounted
= lookup_mnt(path
);
909 path
->dentry
= dget(mounted
->mnt_root
);
915 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
917 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
918 dentry
->d_op
->d_manage(dentry
, true) < 0);
922 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
923 * meet a managed dentry and we're not walking to "..". True is returned to
924 * continue, false to abort.
926 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
927 struct inode
**inode
, bool reverse_transit
)
930 struct vfsmount
*mounted
;
932 * Don't forget we might have a non-mountpoint managed dentry
933 * that wants to block transit.
935 *inode
= path
->dentry
->d_inode
;
936 if (!reverse_transit
&&
937 unlikely(managed_dentry_might_block(path
->dentry
)))
940 if (!d_mountpoint(path
->dentry
))
943 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
947 path
->dentry
= mounted
->mnt_root
;
948 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
951 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
952 return reverse_transit
;
956 static int follow_dotdot_rcu(struct nameidata
*nd
)
958 struct inode
*inode
= nd
->inode
;
963 if (nd
->path
.dentry
== nd
->root
.dentry
&&
964 nd
->path
.mnt
== nd
->root
.mnt
) {
967 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
968 struct dentry
*old
= nd
->path
.dentry
;
969 struct dentry
*parent
= old
->d_parent
;
972 seq
= read_seqcount_begin(&parent
->d_seq
);
973 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
975 inode
= parent
->d_inode
;
976 nd
->path
.dentry
= parent
;
980 if (!follow_up_rcu(&nd
->path
))
982 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
983 inode
= nd
->path
.dentry
->d_inode
;
985 __follow_mount_rcu(nd
, &nd
->path
, &inode
, true);
990 nd
->flags
&= ~LOOKUP_RCU
;
991 if (!(nd
->flags
& LOOKUP_ROOT
))
994 br_read_unlock(vfsmount_lock
);
999 * Follow down to the covering mount currently visible to userspace. At each
1000 * point, the filesystem owning that dentry may be queried as to whether the
1001 * caller is permitted to proceed or not.
1003 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1006 int follow_down(struct path
*path
)
1011 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1012 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1013 /* Allow the filesystem to manage the transit without i_mutex
1016 * We indicate to the filesystem if someone is trying to mount
1017 * something here. This gives autofs the chance to deny anyone
1018 * other than its daemon the right to mount on its
1021 * The filesystem may sleep at this point.
1023 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1024 BUG_ON(!path
->dentry
->d_op
);
1025 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1026 ret
= path
->dentry
->d_op
->d_manage(
1027 path
->dentry
, false);
1029 return ret
== -EISDIR
? 0 : ret
;
1032 /* Transit to a mounted filesystem. */
1033 if (managed
& DCACHE_MOUNTED
) {
1034 struct vfsmount
*mounted
= lookup_mnt(path
);
1039 path
->mnt
= mounted
;
1040 path
->dentry
= dget(mounted
->mnt_root
);
1044 /* Don't handle automount points here */
1051 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1053 static void follow_mount(struct path
*path
)
1055 while (d_mountpoint(path
->dentry
)) {
1056 struct vfsmount
*mounted
= lookup_mnt(path
);
1061 path
->mnt
= mounted
;
1062 path
->dentry
= dget(mounted
->mnt_root
);
1066 static void follow_dotdot(struct nameidata
*nd
)
1071 struct dentry
*old
= nd
->path
.dentry
;
1073 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1074 nd
->path
.mnt
== nd
->root
.mnt
) {
1077 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1078 /* rare case of legitimate dget_parent()... */
1079 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1083 if (!follow_up(&nd
->path
))
1086 follow_mount(&nd
->path
);
1087 nd
->inode
= nd
->path
.dentry
->d_inode
;
1091 * Allocate a dentry with name and parent, and perform a parent
1092 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1093 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1094 * have verified that no child exists while under i_mutex.
1096 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1097 struct qstr
*name
, struct nameidata
*nd
)
1099 struct inode
*inode
= parent
->d_inode
;
1100 struct dentry
*dentry
;
1103 /* Don't create child dentry for a dead directory. */
1104 if (unlikely(IS_DEADDIR(inode
)))
1105 return ERR_PTR(-ENOENT
);
1107 dentry
= d_alloc(parent
, name
);
1108 if (unlikely(!dentry
))
1109 return ERR_PTR(-ENOMEM
);
1111 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1112 if (unlikely(old
)) {
1120 * It's more convoluted than I'd like it to be, but... it's still fairly
1121 * small and for now I'd prefer to have fast path as straight as possible.
1122 * It _is_ time-critical.
1124 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1125 struct path
*path
, struct inode
**inode
)
1127 struct vfsmount
*mnt
= nd
->path
.mnt
;
1128 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1134 * Rename seqlock is not required here because in the off chance
1135 * of a false negative due to a concurrent rename, we're going to
1136 * do the non-racy lookup, below.
1138 if (nd
->flags
& LOOKUP_RCU
) {
1141 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1145 /* Memory barrier in read_seqcount_begin of child is enough */
1146 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1150 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1151 status
= d_revalidate(dentry
, nd
);
1152 if (unlikely(status
<= 0)) {
1153 if (status
!= -ECHILD
)
1159 path
->dentry
= dentry
;
1160 if (likely(__follow_mount_rcu(nd
, path
, inode
, false)))
1163 if (unlazy_walk(nd
, dentry
))
1166 dentry
= __d_lookup(parent
, name
);
1170 if (unlikely(!dentry
)) {
1171 struct inode
*dir
= parent
->d_inode
;
1172 BUG_ON(nd
->inode
!= dir
);
1174 mutex_lock(&dir
->i_mutex
);
1175 dentry
= d_lookup(parent
, name
);
1176 if (likely(!dentry
)) {
1177 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1178 if (IS_ERR(dentry
)) {
1179 mutex_unlock(&dir
->i_mutex
);
1180 return PTR_ERR(dentry
);
1186 mutex_unlock(&dir
->i_mutex
);
1188 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1189 status
= d_revalidate(dentry
, nd
);
1190 if (unlikely(status
<= 0)) {
1195 if (!d_invalidate(dentry
)) {
1204 path
->dentry
= dentry
;
1205 err
= follow_managed(path
, nd
->flags
);
1206 if (unlikely(err
< 0)) {
1207 path_put_conditional(path
, nd
);
1210 *inode
= path
->dentry
->d_inode
;
1214 static inline int may_lookup(struct nameidata
*nd
)
1216 if (nd
->flags
& LOOKUP_RCU
) {
1217 int err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1220 if (unlazy_walk(nd
, NULL
))
1223 return exec_permission(nd
->inode
, 0);
1226 static inline int handle_dots(struct nameidata
*nd
, int type
)
1228 if (type
== LAST_DOTDOT
) {
1229 if (nd
->flags
& LOOKUP_RCU
) {
1230 if (follow_dotdot_rcu(nd
))
1238 static void terminate_walk(struct nameidata
*nd
)
1240 if (!(nd
->flags
& LOOKUP_RCU
)) {
1241 path_put(&nd
->path
);
1243 nd
->flags
&= ~LOOKUP_RCU
;
1244 if (!(nd
->flags
& LOOKUP_ROOT
))
1245 nd
->root
.mnt
= NULL
;
1247 br_read_unlock(vfsmount_lock
);
1251 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1252 struct qstr
*name
, int type
, int follow
)
1254 struct inode
*inode
;
1257 * "." and ".." are special - ".." especially so because it has
1258 * to be able to know about the current root directory and
1259 * parent relationships.
1261 if (unlikely(type
!= LAST_NORM
))
1262 return handle_dots(nd
, type
);
1263 err
= do_lookup(nd
, name
, path
, &inode
);
1264 if (unlikely(err
)) {
1269 path_to_nameidata(path
, nd
);
1273 if (unlikely(inode
->i_op
->follow_link
) && follow
) {
1274 if (nd
->flags
& LOOKUP_RCU
) {
1275 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1280 BUG_ON(inode
!= path
->dentry
->d_inode
);
1283 path_to_nameidata(path
, nd
);
1289 * This limits recursive symlink follows to 8, while
1290 * limiting consecutive symlinks to 40.
1292 * Without that kind of total limit, nasty chains of consecutive
1293 * symlinks can cause almost arbitrarily long lookups.
1295 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1299 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1300 path_put_conditional(path
, nd
);
1301 path_put(&nd
->path
);
1304 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1307 current
->link_count
++;
1310 struct path link
= *path
;
1313 res
= follow_link(&link
, nd
, &cookie
);
1315 res
= walk_component(nd
, path
, &nd
->last
,
1316 nd
->last_type
, LOOKUP_FOLLOW
);
1317 put_link(nd
, &link
, cookie
);
1320 current
->link_count
--;
1327 * This is the basic name resolution function, turning a pathname into
1328 * the final dentry. We expect 'base' to be positive and a directory.
1330 * Returns 0 and nd will have valid dentry and mnt on success.
1331 * Returns error and drops reference to input namei data on failure.
1333 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1337 unsigned int lookup_flags
= nd
->flags
;
1344 /* At this point we know we have a real path component. */
1351 nd
->flags
|= LOOKUP_CONTINUE
;
1353 err
= may_lookup(nd
);
1358 c
= *(const unsigned char *)name
;
1360 hash
= init_name_hash();
1363 hash
= partial_name_hash(c
, hash
);
1364 c
= *(const unsigned char *)name
;
1365 } while (c
&& (c
!= '/'));
1366 this.len
= name
- (const char *) this.name
;
1367 this.hash
= end_name_hash(hash
);
1370 if (this.name
[0] == '.') switch (this.len
) {
1372 if (this.name
[1] == '.') {
1374 nd
->flags
|= LOOKUP_JUMPED
;
1380 if (likely(type
== LAST_NORM
)) {
1381 struct dentry
*parent
= nd
->path
.dentry
;
1382 nd
->flags
&= ~LOOKUP_JUMPED
;
1383 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1384 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1391 /* remove trailing slashes? */
1393 goto last_component
;
1394 while (*++name
== '/');
1396 goto last_component
;
1398 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1403 err
= nested_symlink(&next
, nd
);
1408 if (!nd
->inode
->i_op
->lookup
)
1411 /* here ends the main loop */
1414 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1415 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1417 nd
->last_type
= type
;
1424 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1425 struct nameidata
*nd
, struct file
**fp
)
1431 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1432 nd
->flags
= flags
| LOOKUP_JUMPED
;
1434 if (flags
& LOOKUP_ROOT
) {
1435 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1437 if (!inode
->i_op
->lookup
)
1439 retval
= inode_permission(inode
, MAY_EXEC
);
1443 nd
->path
= nd
->root
;
1445 if (flags
& LOOKUP_RCU
) {
1446 br_read_lock(vfsmount_lock
);
1448 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1450 path_get(&nd
->path
);
1455 nd
->root
.mnt
= NULL
;
1458 if (flags
& LOOKUP_RCU
) {
1459 br_read_lock(vfsmount_lock
);
1464 path_get(&nd
->root
);
1466 nd
->path
= nd
->root
;
1467 } else if (dfd
== AT_FDCWD
) {
1468 if (flags
& LOOKUP_RCU
) {
1469 struct fs_struct
*fs
= current
->fs
;
1472 br_read_lock(vfsmount_lock
);
1476 seq
= read_seqcount_begin(&fs
->seq
);
1478 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1479 } while (read_seqcount_retry(&fs
->seq
, seq
));
1481 get_fs_pwd(current
->fs
, &nd
->path
);
1484 struct dentry
*dentry
;
1486 file
= fget_raw_light(dfd
, &fput_needed
);
1491 dentry
= file
->f_path
.dentry
;
1495 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1498 retval
= file_permission(file
, MAY_EXEC
);
1503 nd
->path
= file
->f_path
;
1504 if (flags
& LOOKUP_RCU
) {
1507 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1508 br_read_lock(vfsmount_lock
);
1511 path_get(&file
->f_path
);
1512 fput_light(file
, fput_needed
);
1516 nd
->inode
= nd
->path
.dentry
->d_inode
;
1520 fput_light(file
, fput_needed
);
1525 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1527 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1528 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1530 nd
->flags
&= ~LOOKUP_PARENT
;
1531 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1532 nd
->flags
& LOOKUP_FOLLOW
);
1535 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1536 static int path_lookupat(int dfd
, const char *name
,
1537 unsigned int flags
, struct nameidata
*nd
)
1539 struct file
*base
= NULL
;
1544 * Path walking is largely split up into 2 different synchronisation
1545 * schemes, rcu-walk and ref-walk (explained in
1546 * Documentation/filesystems/path-lookup.txt). These share much of the
1547 * path walk code, but some things particularly setup, cleanup, and
1548 * following mounts are sufficiently divergent that functions are
1549 * duplicated. Typically there is a function foo(), and its RCU
1550 * analogue, foo_rcu().
1552 * -ECHILD is the error number of choice (just to avoid clashes) that
1553 * is returned if some aspect of an rcu-walk fails. Such an error must
1554 * be handled by restarting a traditional ref-walk (which will always
1555 * be able to complete).
1557 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1562 current
->total_link_count
= 0;
1563 err
= link_path_walk(name
, nd
);
1565 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1566 err
= lookup_last(nd
, &path
);
1569 struct path link
= path
;
1570 nd
->flags
|= LOOKUP_PARENT
;
1571 err
= follow_link(&link
, nd
, &cookie
);
1573 err
= lookup_last(nd
, &path
);
1574 put_link(nd
, &link
, cookie
);
1579 err
= complete_walk(nd
);
1581 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1582 if (!nd
->inode
->i_op
->lookup
) {
1583 path_put(&nd
->path
);
1591 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1592 path_put(&nd
->root
);
1593 nd
->root
.mnt
= NULL
;
1598 static int do_path_lookup(int dfd
, const char *name
,
1599 unsigned int flags
, struct nameidata
*nd
)
1601 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1602 if (unlikely(retval
== -ECHILD
))
1603 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1604 if (unlikely(retval
== -ESTALE
))
1605 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1607 if (likely(!retval
)) {
1608 if (unlikely(!audit_dummy_context())) {
1609 if (nd
->path
.dentry
&& nd
->inode
)
1610 audit_inode(name
, nd
->path
.dentry
);
1616 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1618 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1621 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1623 struct nameidata nd
;
1624 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1631 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1632 * @dentry: pointer to dentry of the base directory
1633 * @mnt: pointer to vfs mount of the base directory
1634 * @name: pointer to file name
1635 * @flags: lookup flags
1636 * @nd: pointer to nameidata
1638 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1639 const char *name
, unsigned int flags
,
1640 struct nameidata
*nd
)
1642 nd
->root
.dentry
= dentry
;
1644 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1645 return do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, nd
);
1648 static struct dentry
*__lookup_hash(struct qstr
*name
,
1649 struct dentry
*base
, struct nameidata
*nd
)
1651 struct inode
*inode
= base
->d_inode
;
1652 struct dentry
*dentry
;
1655 err
= exec_permission(inode
, 0);
1657 return ERR_PTR(err
);
1660 * Don't bother with __d_lookup: callers are for creat as
1661 * well as unlink, so a lot of the time it would cost
1664 dentry
= d_lookup(base
, name
);
1666 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1667 dentry
= do_revalidate(dentry
, nd
);
1670 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1676 * Restricted form of lookup. Doesn't follow links, single-component only,
1677 * needs parent already locked. Doesn't follow mounts.
1680 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1682 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1686 * lookup_one_len - filesystem helper to lookup single pathname component
1687 * @name: pathname component to lookup
1688 * @base: base directory to lookup from
1689 * @len: maximum length @len should be interpreted to
1691 * Note that this routine is purely a helper for filesystem usage and should
1692 * not be called by generic code. Also note that by using this function the
1693 * nameidata argument is passed to the filesystem methods and a filesystem
1694 * using this helper needs to be prepared for that.
1696 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1702 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1707 return ERR_PTR(-EACCES
);
1709 hash
= init_name_hash();
1711 c
= *(const unsigned char *)name
++;
1712 if (c
== '/' || c
== '\0')
1713 return ERR_PTR(-EACCES
);
1714 hash
= partial_name_hash(c
, hash
);
1716 this.hash
= end_name_hash(hash
);
1718 * See if the low-level filesystem might want
1719 * to use its own hash..
1721 if (base
->d_flags
& DCACHE_OP_HASH
) {
1722 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1724 return ERR_PTR(err
);
1727 return __lookup_hash(&this, base
, NULL
);
1730 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1733 struct nameidata nd
;
1734 char *tmp
= getname_flags(name
, flags
);
1735 int err
= PTR_ERR(tmp
);
1738 BUG_ON(flags
& LOOKUP_PARENT
);
1740 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1748 static int user_path_parent(int dfd
, const char __user
*path
,
1749 struct nameidata
*nd
, char **name
)
1751 char *s
= getname(path
);
1757 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1767 * It's inline, so penalty for filesystems that don't use sticky bit is
1770 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1772 uid_t fsuid
= current_fsuid();
1774 if (!(dir
->i_mode
& S_ISVTX
))
1776 if (current_user_ns() != inode_userns(inode
))
1778 if (inode
->i_uid
== fsuid
)
1780 if (dir
->i_uid
== fsuid
)
1784 return !ns_capable(inode_userns(inode
), CAP_FOWNER
);
1788 * Check whether we can remove a link victim from directory dir, check
1789 * whether the type of victim is right.
1790 * 1. We can't do it if dir is read-only (done in permission())
1791 * 2. We should have write and exec permissions on dir
1792 * 3. We can't remove anything from append-only dir
1793 * 4. We can't do anything with immutable dir (done in permission())
1794 * 5. If the sticky bit on dir is set we should either
1795 * a. be owner of dir, or
1796 * b. be owner of victim, or
1797 * c. have CAP_FOWNER capability
1798 * 6. If the victim is append-only or immutable we can't do antyhing with
1799 * links pointing to it.
1800 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1801 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1802 * 9. We can't remove a root or mountpoint.
1803 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1804 * nfs_async_unlink().
1806 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1810 if (!victim
->d_inode
)
1813 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1814 audit_inode_child(victim
, dir
);
1816 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1821 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1822 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1825 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1827 if (IS_ROOT(victim
))
1829 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1831 if (IS_DEADDIR(dir
))
1833 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1838 /* Check whether we can create an object with dentry child in directory
1840 * 1. We can't do it if child already exists (open has special treatment for
1841 * this case, but since we are inlined it's OK)
1842 * 2. We can't do it if dir is read-only (done in permission())
1843 * 3. We should have write and exec permissions on dir
1844 * 4. We can't do it if dir is immutable (done in permission())
1846 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1850 if (IS_DEADDIR(dir
))
1852 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1856 * p1 and p2 should be directories on the same fs.
1858 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1863 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1867 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1869 p
= d_ancestor(p2
, p1
);
1871 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1872 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1876 p
= d_ancestor(p1
, p2
);
1878 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1879 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1883 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1884 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1888 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1890 mutex_unlock(&p1
->d_inode
->i_mutex
);
1892 mutex_unlock(&p2
->d_inode
->i_mutex
);
1893 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1897 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1898 struct nameidata
*nd
)
1900 int error
= may_create(dir
, dentry
);
1905 if (!dir
->i_op
->create
)
1906 return -EACCES
; /* shouldn't it be ENOSYS? */
1909 error
= security_inode_create(dir
, dentry
, mode
);
1912 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1914 fsnotify_create(dir
, dentry
);
1918 static int may_open(struct path
*path
, int acc_mode
, int flag
)
1920 struct dentry
*dentry
= path
->dentry
;
1921 struct inode
*inode
= dentry
->d_inode
;
1931 switch (inode
->i_mode
& S_IFMT
) {
1935 if (acc_mode
& MAY_WRITE
)
1940 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
1949 error
= inode_permission(inode
, acc_mode
);
1954 * An append-only file must be opened in append mode for writing.
1956 if (IS_APPEND(inode
)) {
1957 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
1963 /* O_NOATIME can only be set by the owner or superuser */
1964 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
1968 * Ensure there are no outstanding leases on the file.
1970 return break_lease(inode
, flag
);
1973 static int handle_truncate(struct file
*filp
)
1975 struct path
*path
= &filp
->f_path
;
1976 struct inode
*inode
= path
->dentry
->d_inode
;
1977 int error
= get_write_access(inode
);
1981 * Refuse to truncate files with mandatory locks held on them.
1983 error
= locks_verify_locked(inode
);
1985 error
= security_path_truncate(path
);
1987 error
= do_truncate(path
->dentry
, 0,
1988 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
1991 put_write_access(inode
);
1996 * Note that while the flag value (low two bits) for sys_open means:
2001 * it is changed into
2002 * 00 - no permissions needed
2003 * 01 - read-permission
2004 * 10 - write-permission
2006 * for the internal routines (ie open_namei()/follow_link() etc)
2007 * This is more logical, and also allows the 00 "no perm needed"
2008 * to be used for symlinks (where the permissions are checked
2012 static inline int open_to_namei_flags(int flag
)
2014 if ((flag
+1) & O_ACCMODE
)
2020 * Handle the last step of open()
2022 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2023 const struct open_flags
*op
, const char *pathname
)
2025 struct dentry
*dir
= nd
->path
.dentry
;
2026 struct dentry
*dentry
;
2027 int open_flag
= op
->open_flag
;
2028 int will_truncate
= open_flag
& O_TRUNC
;
2030 int acc_mode
= op
->acc_mode
;
2034 nd
->flags
&= ~LOOKUP_PARENT
;
2035 nd
->flags
|= op
->intent
;
2037 switch (nd
->last_type
) {
2040 error
= handle_dots(nd
, nd
->last_type
);
2042 return ERR_PTR(error
);
2045 error
= complete_walk(nd
);
2047 return ERR_PTR(error
);
2048 audit_inode(pathname
, nd
->path
.dentry
);
2049 if (open_flag
& O_CREAT
) {
2055 error
= complete_walk(nd
);
2057 return ERR_PTR(error
);
2058 audit_inode(pathname
, dir
);
2062 if (!(open_flag
& O_CREAT
)) {
2064 if (nd
->last
.name
[nd
->last
.len
])
2065 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2066 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2068 /* we _can_ be in RCU mode here */
2069 error
= walk_component(nd
, path
, &nd
->last
, LAST_NORM
,
2072 return ERR_PTR(error
);
2073 if (error
) /* symlink */
2076 error
= complete_walk(nd
);
2078 return ERR_PTR(-ECHILD
);
2081 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2082 if (!nd
->inode
->i_op
->lookup
)
2085 audit_inode(pathname
, nd
->path
.dentry
);
2089 /* create side of things */
2090 error
= complete_walk(nd
);
2092 return ERR_PTR(error
);
2094 audit_inode(pathname
, dir
);
2096 /* trailing slashes? */
2097 if (nd
->last
.name
[nd
->last
.len
])
2100 mutex_lock(&dir
->d_inode
->i_mutex
);
2102 dentry
= lookup_hash(nd
);
2103 error
= PTR_ERR(dentry
);
2104 if (IS_ERR(dentry
)) {
2105 mutex_unlock(&dir
->d_inode
->i_mutex
);
2109 path
->dentry
= dentry
;
2110 path
->mnt
= nd
->path
.mnt
;
2112 /* Negative dentry, just create the file */
2113 if (!dentry
->d_inode
) {
2114 int mode
= op
->mode
;
2115 if (!IS_POSIXACL(dir
->d_inode
))
2116 mode
&= ~current_umask();
2118 * This write is needed to ensure that a
2119 * rw->ro transition does not occur between
2120 * the time when the file is created and when
2121 * a permanent write count is taken through
2122 * the 'struct file' in nameidata_to_filp().
2124 error
= mnt_want_write(nd
->path
.mnt
);
2126 goto exit_mutex_unlock
;
2128 /* Don't check for write permission, don't truncate */
2129 open_flag
&= ~O_TRUNC
;
2131 acc_mode
= MAY_OPEN
;
2132 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2134 goto exit_mutex_unlock
;
2135 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2137 goto exit_mutex_unlock
;
2138 mutex_unlock(&dir
->d_inode
->i_mutex
);
2139 dput(nd
->path
.dentry
);
2140 nd
->path
.dentry
= dentry
;
2145 * It already exists.
2147 mutex_unlock(&dir
->d_inode
->i_mutex
);
2148 audit_inode(pathname
, path
->dentry
);
2151 if (open_flag
& O_EXCL
)
2154 error
= follow_managed(path
, nd
->flags
);
2159 if (!path
->dentry
->d_inode
)
2162 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2165 path_to_nameidata(path
, nd
);
2166 nd
->inode
= path
->dentry
->d_inode
;
2168 if (S_ISDIR(nd
->inode
->i_mode
))
2171 if (!S_ISREG(nd
->inode
->i_mode
))
2174 if (will_truncate
) {
2175 error
= mnt_want_write(nd
->path
.mnt
);
2181 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2184 filp
= nameidata_to_filp(nd
);
2185 if (!IS_ERR(filp
)) {
2186 error
= ima_file_check(filp
, op
->acc_mode
);
2189 filp
= ERR_PTR(error
);
2192 if (!IS_ERR(filp
)) {
2193 if (will_truncate
) {
2194 error
= handle_truncate(filp
);
2197 filp
= ERR_PTR(error
);
2203 mnt_drop_write(nd
->path
.mnt
);
2204 path_put(&nd
->path
);
2208 mutex_unlock(&dir
->d_inode
->i_mutex
);
2210 path_put_conditional(path
, nd
);
2212 filp
= ERR_PTR(error
);
2216 static struct file
*path_openat(int dfd
, const char *pathname
,
2217 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2219 struct file
*base
= NULL
;
2224 filp
= get_empty_filp();
2226 return ERR_PTR(-ENFILE
);
2228 filp
->f_flags
= op
->open_flag
;
2229 nd
->intent
.open
.file
= filp
;
2230 nd
->intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2231 nd
->intent
.open
.create_mode
= op
->mode
;
2233 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2234 if (unlikely(error
))
2237 current
->total_link_count
= 0;
2238 error
= link_path_walk(pathname
, nd
);
2239 if (unlikely(error
))
2242 filp
= do_last(nd
, &path
, op
, pathname
);
2243 while (unlikely(!filp
)) { /* trailing symlink */
2244 struct path link
= path
;
2246 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2247 path_put_conditional(&path
, nd
);
2248 path_put(&nd
->path
);
2249 filp
= ERR_PTR(-ELOOP
);
2252 nd
->flags
|= LOOKUP_PARENT
;
2253 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2254 error
= follow_link(&link
, nd
, &cookie
);
2255 if (unlikely(error
))
2256 filp
= ERR_PTR(error
);
2258 filp
= do_last(nd
, &path
, op
, pathname
);
2259 put_link(nd
, &link
, cookie
);
2262 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2263 path_put(&nd
->root
);
2266 release_open_intent(nd
);
2270 filp
= ERR_PTR(error
);
2274 struct file
*do_filp_open(int dfd
, const char *pathname
,
2275 const struct open_flags
*op
, int flags
)
2277 struct nameidata nd
;
2280 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2281 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2282 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2283 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2284 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2288 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2289 const char *name
, const struct open_flags
*op
, int flags
)
2291 struct nameidata nd
;
2295 nd
.root
.dentry
= dentry
;
2297 flags
|= LOOKUP_ROOT
;
2299 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2300 return ERR_PTR(-ELOOP
);
2302 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2303 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2304 file
= path_openat(-1, name
, &nd
, op
, flags
);
2305 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2306 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2311 * lookup_create - lookup a dentry, creating it if it doesn't exist
2312 * @nd: nameidata info
2313 * @is_dir: directory flag
2315 * Simple function to lookup and return a dentry and create it
2316 * if it doesn't exist. Is SMP-safe.
2318 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2320 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2322 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2324 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2326 * Yucky last component or no last component at all?
2327 * (foo/., foo/.., /////)
2329 if (nd
->last_type
!= LAST_NORM
)
2331 nd
->flags
&= ~LOOKUP_PARENT
;
2332 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2333 nd
->intent
.open
.flags
= O_EXCL
;
2336 * Do the final lookup.
2338 dentry
= lookup_hash(nd
);
2342 if (dentry
->d_inode
)
2345 * Special case - lookup gave negative, but... we had foo/bar/
2346 * From the vfs_mknod() POV we just have a negative dentry -
2347 * all is fine. Let's be bastards - you had / on the end, you've
2348 * been asking for (non-existent) directory. -ENOENT for you.
2350 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2352 dentry
= ERR_PTR(-ENOENT
);
2357 dentry
= ERR_PTR(-EEXIST
);
2361 EXPORT_SYMBOL_GPL(lookup_create
);
2363 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2365 int error
= may_create(dir
, dentry
);
2370 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) &&
2371 !ns_capable(inode_userns(dir
), CAP_MKNOD
))
2374 if (!dir
->i_op
->mknod
)
2377 error
= devcgroup_inode_mknod(mode
, dev
);
2381 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2385 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2387 fsnotify_create(dir
, dentry
);
2391 static int may_mknod(mode_t mode
)
2393 switch (mode
& S_IFMT
) {
2399 case 0: /* zero mode translates to S_IFREG */
2408 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2413 struct dentry
*dentry
;
2414 struct nameidata nd
;
2419 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2423 dentry
= lookup_create(&nd
, 0);
2424 if (IS_ERR(dentry
)) {
2425 error
= PTR_ERR(dentry
);
2428 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2429 mode
&= ~current_umask();
2430 error
= may_mknod(mode
);
2433 error
= mnt_want_write(nd
.path
.mnt
);
2436 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2438 goto out_drop_write
;
2439 switch (mode
& S_IFMT
) {
2440 case 0: case S_IFREG
:
2441 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2443 case S_IFCHR
: case S_IFBLK
:
2444 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2445 new_decode_dev(dev
));
2447 case S_IFIFO
: case S_IFSOCK
:
2448 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2452 mnt_drop_write(nd
.path
.mnt
);
2456 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2463 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2465 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2468 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2470 int error
= may_create(dir
, dentry
);
2475 if (!dir
->i_op
->mkdir
)
2478 mode
&= (S_IRWXUGO
|S_ISVTX
);
2479 error
= security_inode_mkdir(dir
, dentry
, mode
);
2483 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2485 fsnotify_mkdir(dir
, dentry
);
2489 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2493 struct dentry
*dentry
;
2494 struct nameidata nd
;
2496 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2500 dentry
= lookup_create(&nd
, 1);
2501 error
= PTR_ERR(dentry
);
2505 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2506 mode
&= ~current_umask();
2507 error
= mnt_want_write(nd
.path
.mnt
);
2510 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2512 goto out_drop_write
;
2513 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2515 mnt_drop_write(nd
.path
.mnt
);
2519 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2526 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2528 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2532 * The dentry_unhash() helper will try to drop the dentry early: we
2533 * should have a usage count of 2 if we're the only user of this
2534 * dentry, and if that is true (possibly after pruning the dcache),
2535 * then we drop the dentry now.
2537 * A low-level filesystem can, if it choses, legally
2540 * if (!d_unhashed(dentry))
2543 * if it cannot handle the case of removing a directory
2544 * that is still in use by something else..
2546 void dentry_unhash(struct dentry
*dentry
)
2548 shrink_dcache_parent(dentry
);
2549 spin_lock(&dentry
->d_lock
);
2550 if (dentry
->d_count
== 1)
2552 spin_unlock(&dentry
->d_lock
);
2555 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2557 int error
= may_delete(dir
, dentry
, 1);
2562 if (!dir
->i_op
->rmdir
)
2565 mutex_lock(&dentry
->d_inode
->i_mutex
);
2568 if (d_mountpoint(dentry
))
2571 error
= security_inode_rmdir(dir
, dentry
);
2575 error
= dir
->i_op
->rmdir(dir
, dentry
);
2579 dentry
->d_inode
->i_flags
|= S_DEAD
;
2583 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2589 static long do_rmdir(int dfd
, const char __user
*pathname
)
2593 struct dentry
*dentry
;
2594 struct nameidata nd
;
2596 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2600 switch(nd
.last_type
) {
2612 nd
.flags
&= ~LOOKUP_PARENT
;
2614 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2615 dentry
= lookup_hash(&nd
);
2616 error
= PTR_ERR(dentry
);
2619 error
= mnt_want_write(nd
.path
.mnt
);
2622 error
= security_path_rmdir(&nd
.path
, dentry
);
2625 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2627 mnt_drop_write(nd
.path
.mnt
);
2631 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2638 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2640 return do_rmdir(AT_FDCWD
, pathname
);
2643 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2645 int error
= may_delete(dir
, dentry
, 0);
2650 if (!dir
->i_op
->unlink
)
2653 mutex_lock(&dentry
->d_inode
->i_mutex
);
2654 if (d_mountpoint(dentry
))
2657 error
= security_inode_unlink(dir
, dentry
);
2659 error
= dir
->i_op
->unlink(dir
, dentry
);
2664 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2666 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2667 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2668 fsnotify_link_count(dentry
->d_inode
);
2676 * Make sure that the actual truncation of the file will occur outside its
2677 * directory's i_mutex. Truncate can take a long time if there is a lot of
2678 * writeout happening, and we don't want to prevent access to the directory
2679 * while waiting on the I/O.
2681 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2685 struct dentry
*dentry
;
2686 struct nameidata nd
;
2687 struct inode
*inode
= NULL
;
2689 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2694 if (nd
.last_type
!= LAST_NORM
)
2697 nd
.flags
&= ~LOOKUP_PARENT
;
2699 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2700 dentry
= lookup_hash(&nd
);
2701 error
= PTR_ERR(dentry
);
2702 if (!IS_ERR(dentry
)) {
2703 /* Why not before? Because we want correct error value */
2704 if (nd
.last
.name
[nd
.last
.len
])
2706 inode
= dentry
->d_inode
;
2709 error
= mnt_want_write(nd
.path
.mnt
);
2712 error
= security_path_unlink(&nd
.path
, dentry
);
2715 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2717 mnt_drop_write(nd
.path
.mnt
);
2721 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2723 iput(inode
); /* truncate the inode here */
2730 error
= !dentry
->d_inode
? -ENOENT
:
2731 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2735 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2737 if ((flag
& ~AT_REMOVEDIR
) != 0)
2740 if (flag
& AT_REMOVEDIR
)
2741 return do_rmdir(dfd
, pathname
);
2743 return do_unlinkat(dfd
, pathname
);
2746 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2748 return do_unlinkat(AT_FDCWD
, pathname
);
2751 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2753 int error
= may_create(dir
, dentry
);
2758 if (!dir
->i_op
->symlink
)
2761 error
= security_inode_symlink(dir
, dentry
, oldname
);
2765 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2767 fsnotify_create(dir
, dentry
);
2771 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2772 int, newdfd
, const char __user
*, newname
)
2777 struct dentry
*dentry
;
2778 struct nameidata nd
;
2780 from
= getname(oldname
);
2782 return PTR_ERR(from
);
2784 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2788 dentry
= lookup_create(&nd
, 0);
2789 error
= PTR_ERR(dentry
);
2793 error
= mnt_want_write(nd
.path
.mnt
);
2796 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2798 goto out_drop_write
;
2799 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2801 mnt_drop_write(nd
.path
.mnt
);
2805 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2813 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2815 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2818 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2820 struct inode
*inode
= old_dentry
->d_inode
;
2826 error
= may_create(dir
, new_dentry
);
2830 if (dir
->i_sb
!= inode
->i_sb
)
2834 * A link to an append-only or immutable file cannot be created.
2836 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2838 if (!dir
->i_op
->link
)
2840 if (S_ISDIR(inode
->i_mode
))
2843 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2847 mutex_lock(&inode
->i_mutex
);
2848 /* Make sure we don't allow creating hardlink to an unlinked file */
2849 if (inode
->i_nlink
== 0)
2852 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2853 mutex_unlock(&inode
->i_mutex
);
2855 fsnotify_link(dir
, inode
, new_dentry
);
2860 * Hardlinks are often used in delicate situations. We avoid
2861 * security-related surprises by not following symlinks on the
2864 * We don't follow them on the oldname either to be compatible
2865 * with linux 2.0, and to avoid hard-linking to directories
2866 * and other special files. --ADM
2868 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2869 int, newdfd
, const char __user
*, newname
, int, flags
)
2871 struct dentry
*new_dentry
;
2872 struct nameidata nd
;
2873 struct path old_path
;
2878 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
2881 * To use null names we require CAP_DAC_READ_SEARCH
2882 * This ensures that not everyone will be able to create
2883 * handlink using the passed filedescriptor.
2885 if (flags
& AT_EMPTY_PATH
) {
2886 if (!capable(CAP_DAC_READ_SEARCH
))
2891 if (flags
& AT_SYMLINK_FOLLOW
)
2892 how
|= LOOKUP_FOLLOW
;
2894 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
2898 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2902 if (old_path
.mnt
!= nd
.path
.mnt
)
2904 new_dentry
= lookup_create(&nd
, 0);
2905 error
= PTR_ERR(new_dentry
);
2906 if (IS_ERR(new_dentry
))
2908 error
= mnt_want_write(nd
.path
.mnt
);
2911 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2913 goto out_drop_write
;
2914 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2916 mnt_drop_write(nd
.path
.mnt
);
2920 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2925 path_put(&old_path
);
2930 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
2932 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2936 * The worst of all namespace operations - renaming directory. "Perverted"
2937 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2939 * a) we can get into loop creation. Check is done in is_subdir().
2940 * b) race potential - two innocent renames can create a loop together.
2941 * That's where 4.4 screws up. Current fix: serialization on
2942 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2944 * c) we have to lock _three_ objects - parents and victim (if it exists).
2945 * And that - after we got ->i_mutex on parents (until then we don't know
2946 * whether the target exists). Solution: try to be smart with locking
2947 * order for inodes. We rely on the fact that tree topology may change
2948 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2949 * move will be locked. Thus we can rank directories by the tree
2950 * (ancestors first) and rank all non-directories after them.
2951 * That works since everybody except rename does "lock parent, lookup,
2952 * lock child" and rename is under ->s_vfs_rename_mutex.
2953 * HOWEVER, it relies on the assumption that any object with ->lookup()
2954 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2955 * we'd better make sure that there's no link(2) for them.
2956 * d) conversion from fhandle to dentry may come in the wrong moment - when
2957 * we are removing the target. Solution: we will have to grab ->i_mutex
2958 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2959 * ->i_mutex on parents, which works but leads to some truly excessive
2962 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2963 struct inode
*new_dir
, struct dentry
*new_dentry
)
2966 struct inode
*target
= new_dentry
->d_inode
;
2969 * If we are going to change the parent - check write permissions,
2970 * we'll need to flip '..'.
2972 if (new_dir
!= old_dir
) {
2973 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
2978 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2983 mutex_lock(&target
->i_mutex
);
2986 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
2989 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2994 target
->i_flags
|= S_DEAD
;
2995 dont_mount(new_dentry
);
2999 mutex_unlock(&target
->i_mutex
);
3001 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3002 d_move(old_dentry
,new_dentry
);
3006 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3007 struct inode
*new_dir
, struct dentry
*new_dentry
)
3009 struct inode
*target
= new_dentry
->d_inode
;
3012 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3018 mutex_lock(&target
->i_mutex
);
3021 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3024 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3029 dont_mount(new_dentry
);
3030 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3031 d_move(old_dentry
, new_dentry
);
3034 mutex_unlock(&target
->i_mutex
);
3039 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3040 struct inode
*new_dir
, struct dentry
*new_dentry
)
3043 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3044 const unsigned char *old_name
;
3046 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3049 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3053 if (!new_dentry
->d_inode
)
3054 error
= may_create(new_dir
, new_dentry
);
3056 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3060 if (!old_dir
->i_op
->rename
)
3063 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3066 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3068 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3070 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3071 new_dentry
->d_inode
, old_dentry
);
3072 fsnotify_oldname_free(old_name
);
3077 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3078 int, newdfd
, const char __user
*, newname
)
3080 struct dentry
*old_dir
, *new_dir
;
3081 struct dentry
*old_dentry
, *new_dentry
;
3082 struct dentry
*trap
;
3083 struct nameidata oldnd
, newnd
;
3088 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3092 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3097 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3100 old_dir
= oldnd
.path
.dentry
;
3102 if (oldnd
.last_type
!= LAST_NORM
)
3105 new_dir
= newnd
.path
.dentry
;
3106 if (newnd
.last_type
!= LAST_NORM
)
3109 oldnd
.flags
&= ~LOOKUP_PARENT
;
3110 newnd
.flags
&= ~LOOKUP_PARENT
;
3111 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3113 trap
= lock_rename(new_dir
, old_dir
);
3115 old_dentry
= lookup_hash(&oldnd
);
3116 error
= PTR_ERR(old_dentry
);
3117 if (IS_ERR(old_dentry
))
3119 /* source must exist */
3121 if (!old_dentry
->d_inode
)
3123 /* unless the source is a directory trailing slashes give -ENOTDIR */
3124 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3126 if (oldnd
.last
.name
[oldnd
.last
.len
])
3128 if (newnd
.last
.name
[newnd
.last
.len
])
3131 /* source should not be ancestor of target */
3133 if (old_dentry
== trap
)
3135 new_dentry
= lookup_hash(&newnd
);
3136 error
= PTR_ERR(new_dentry
);
3137 if (IS_ERR(new_dentry
))
3139 /* target should not be an ancestor of source */
3141 if (new_dentry
== trap
)
3144 error
= mnt_want_write(oldnd
.path
.mnt
);
3147 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3148 &newnd
.path
, new_dentry
);
3151 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3152 new_dir
->d_inode
, new_dentry
);
3154 mnt_drop_write(oldnd
.path
.mnt
);
3160 unlock_rename(new_dir
, old_dir
);
3162 path_put(&newnd
.path
);
3165 path_put(&oldnd
.path
);
3171 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3173 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3176 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3180 len
= PTR_ERR(link
);
3185 if (len
> (unsigned) buflen
)
3187 if (copy_to_user(buffer
, link
, len
))
3194 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3195 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3196 * using) it for any given inode is up to filesystem.
3198 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3200 struct nameidata nd
;
3205 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3207 return PTR_ERR(cookie
);
3209 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3210 if (dentry
->d_inode
->i_op
->put_link
)
3211 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3215 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3217 return __vfs_follow_link(nd
, link
);
3220 /* get the link contents into pagecache */
3221 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3225 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3226 page
= read_mapping_page(mapping
, 0, NULL
);
3231 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3235 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3237 struct page
*page
= NULL
;
3238 char *s
= page_getlink(dentry
, &page
);
3239 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3242 page_cache_release(page
);
3247 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3249 struct page
*page
= NULL
;
3250 nd_set_link(nd
, page_getlink(dentry
, &page
));
3254 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3256 struct page
*page
= cookie
;
3260 page_cache_release(page
);
3265 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3267 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3269 struct address_space
*mapping
= inode
->i_mapping
;
3274 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3276 flags
|= AOP_FLAG_NOFS
;
3279 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3280 flags
, &page
, &fsdata
);
3284 kaddr
= kmap_atomic(page
, KM_USER0
);
3285 memcpy(kaddr
, symname
, len
-1);
3286 kunmap_atomic(kaddr
, KM_USER0
);
3288 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3295 mark_inode_dirty(inode
);
3301 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3303 return __page_symlink(inode
, symname
, len
,
3304 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3307 const struct inode_operations page_symlink_inode_operations
= {
3308 .readlink
= generic_readlink
,
3309 .follow_link
= page_follow_link_light
,
3310 .put_link
= page_put_link
,
3313 EXPORT_SYMBOL(user_path_at
);
3314 EXPORT_SYMBOL(follow_down_one
);
3315 EXPORT_SYMBOL(follow_down
);
3316 EXPORT_SYMBOL(follow_up
);
3317 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3318 EXPORT_SYMBOL(getname
);
3319 EXPORT_SYMBOL(lock_rename
);
3320 EXPORT_SYMBOL(lookup_one_len
);
3321 EXPORT_SYMBOL(page_follow_link_light
);
3322 EXPORT_SYMBOL(page_put_link
);
3323 EXPORT_SYMBOL(page_readlink
);
3324 EXPORT_SYMBOL(__page_symlink
);
3325 EXPORT_SYMBOL(page_symlink
);
3326 EXPORT_SYMBOL(page_symlink_inode_operations
);
3327 EXPORT_SYMBOL(kern_path_parent
);
3328 EXPORT_SYMBOL(kern_path
);
3329 EXPORT_SYMBOL(vfs_path_lookup
);
3330 EXPORT_SYMBOL(inode_permission
);
3331 EXPORT_SYMBOL(file_permission
);
3332 EXPORT_SYMBOL(unlock_rename
);
3333 EXPORT_SYMBOL(vfs_create
);
3334 EXPORT_SYMBOL(vfs_follow_link
);
3335 EXPORT_SYMBOL(vfs_link
);
3336 EXPORT_SYMBOL(vfs_mkdir
);
3337 EXPORT_SYMBOL(vfs_mknod
);
3338 EXPORT_SYMBOL(generic_permission
);
3339 EXPORT_SYMBOL(vfs_readlink
);
3340 EXPORT_SYMBOL(vfs_rename
);
3341 EXPORT_SYMBOL(vfs_rmdir
);
3342 EXPORT_SYMBOL(vfs_symlink
);
3343 EXPORT_SYMBOL(vfs_unlink
);
3344 EXPORT_SYMBOL(dentry_unhash
);
3345 EXPORT_SYMBOL(generic_readlink
);