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-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* 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 char * getname(const char __user
* filename
)
143 result
= ERR_PTR(-ENOMEM
);
146 int retval
= do_getname(filename
, tmp
);
151 result
= ERR_PTR(retval
);
154 audit_getname(result
);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name
)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname
);
170 * This does basic POSIX ACL permission checking
172 static inline int __acl_permission_check(struct inode
*inode
, int mask
,
173 int (*check_acl
)(struct inode
*inode
, int mask
), int rcu
)
175 umode_t mode
= inode
->i_mode
;
177 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
179 if (current_fsuid() == inode
->i_uid
)
182 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
186 int error
= check_acl(inode
, mask
);
187 if (error
!= -EAGAIN
)
192 if (in_group_p(inode
->i_gid
))
197 * If the DACs are ok we don't need any capability check.
199 if ((mask
& ~mode
) == 0)
204 static inline int acl_permission_check(struct inode
*inode
, int mask
,
205 int (*check_acl
)(struct inode
*inode
, int mask
))
207 return __acl_permission_check(inode
, mask
, check_acl
, 0);
211 * generic_permission - check for access rights on a Posix-like filesystem
212 * @inode: inode to check access rights for
213 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
214 * @check_acl: optional callback to check for Posix ACLs
216 * Used to check for read/write/execute permissions on a file.
217 * We use "fsuid" for this, letting us set arbitrary permissions
218 * for filesystem access without changing the "normal" uids which
219 * are used for other things..
221 int generic_permission(struct inode
*inode
, int mask
,
222 int (*check_acl
)(struct inode
*inode
, int mask
))
227 * Do the basic POSIX ACL permission checks.
229 ret
= acl_permission_check(inode
, mask
, check_acl
);
234 * Read/write DACs are always overridable.
235 * Executable DACs are overridable if at least one exec bit is set.
237 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
238 if (capable(CAP_DAC_OVERRIDE
))
242 * Searching includes executable on directories, else just read.
244 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
245 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
246 if (capable(CAP_DAC_READ_SEARCH
))
253 * inode_permission - check for access rights to a given inode
254 * @inode: inode to check permission on
255 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
257 * Used to check for read/write/execute permissions on an inode.
258 * We use "fsuid" for this, letting us set arbitrary permissions
259 * for filesystem access without changing the "normal" uids which
260 * are used for other things.
262 int inode_permission(struct inode
*inode
, int mask
)
266 if (mask
& MAY_WRITE
) {
267 umode_t mode
= inode
->i_mode
;
270 * Nobody gets write access to a read-only fs.
272 if (IS_RDONLY(inode
) &&
273 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
277 * Nobody gets write access to an immutable file.
279 if (IS_IMMUTABLE(inode
))
283 if (inode
->i_op
->permission
)
284 retval
= inode
->i_op
->permission(inode
, mask
);
286 retval
= generic_permission(inode
, mask
, inode
->i_op
->check_acl
);
291 retval
= devcgroup_inode_permission(inode
, mask
);
295 return security_inode_permission(inode
, mask
);
299 * file_permission - check for additional access rights to a given file
300 * @file: file to check access rights for
301 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
303 * Used to check for read/write/execute permissions on an already opened
307 * Do not use this function in new code. All access checks should
308 * be done using inode_permission().
310 int file_permission(struct file
*file
, int mask
)
312 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
316 * get_write_access() gets write permission for a file.
317 * put_write_access() releases this write permission.
318 * This is used for regular files.
319 * We cannot support write (and maybe mmap read-write shared) accesses and
320 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
321 * can have the following values:
322 * 0: no writers, no VM_DENYWRITE mappings
323 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
324 * > 0: (i_writecount) users are writing to the file.
326 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
327 * except for the cases where we don't hold i_writecount yet. Then we need to
328 * use {get,deny}_write_access() - these functions check the sign and refuse
329 * to do the change if sign is wrong. Exclusion between them is provided by
330 * the inode->i_lock spinlock.
333 int get_write_access(struct inode
* inode
)
335 spin_lock(&inode
->i_lock
);
336 if (atomic_read(&inode
->i_writecount
) < 0) {
337 spin_unlock(&inode
->i_lock
);
340 atomic_inc(&inode
->i_writecount
);
341 spin_unlock(&inode
->i_lock
);
346 int deny_write_access(struct file
* file
)
348 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
350 spin_lock(&inode
->i_lock
);
351 if (atomic_read(&inode
->i_writecount
) > 0) {
352 spin_unlock(&inode
->i_lock
);
355 atomic_dec(&inode
->i_writecount
);
356 spin_unlock(&inode
->i_lock
);
362 * path_get - get a reference to a path
363 * @path: path to get the reference to
365 * Given a path increment the reference count to the dentry and the vfsmount.
367 void path_get(struct path
*path
)
372 EXPORT_SYMBOL(path_get
);
375 * path_put - put a reference to a path
376 * @path: path to put the reference to
378 * Given a path decrement the reference count to the dentry and the vfsmount.
380 void path_put(struct path
*path
)
385 EXPORT_SYMBOL(path_put
);
388 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
389 * @nd: nameidata pathwalk data to drop
390 * @Returns: 0 on success, -ECHLID on failure
392 * Path walking has 2 modes, rcu-walk and ref-walk (see
393 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
394 * to drop out of rcu-walk mode and take normal reference counts on dentries
395 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
396 * refcounts at the last known good point before rcu-walk got stuck, so
397 * ref-walk may continue from there. If this is not successful (eg. a seqcount
398 * has changed), then failure is returned and path walk restarts from the
399 * beginning in ref-walk mode.
401 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
402 * ref-walk. Must be called from rcu-walk context.
404 static int nameidata_drop_rcu(struct nameidata
*nd
)
406 struct fs_struct
*fs
= current
->fs
;
407 struct dentry
*dentry
= nd
->path
.dentry
;
409 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
411 spin_lock(&fs
->lock
);
412 if (nd
->root
.mnt
!= fs
->root
.mnt
||
413 nd
->root
.dentry
!= fs
->root
.dentry
)
416 spin_lock(&dentry
->d_lock
);
417 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
419 BUG_ON(nd
->inode
!= dentry
->d_inode
);
420 spin_unlock(&dentry
->d_lock
);
423 spin_unlock(&fs
->lock
);
425 mntget(nd
->path
.mnt
);
428 br_read_unlock(vfsmount_lock
);
429 nd
->flags
&= ~LOOKUP_RCU
;
432 spin_unlock(&dentry
->d_lock
);
435 spin_unlock(&fs
->lock
);
439 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
440 static inline int nameidata_drop_rcu_maybe(struct nameidata
*nd
)
442 if (nd
->flags
& LOOKUP_RCU
)
443 return nameidata_drop_rcu(nd
);
448 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
449 * @nd: nameidata pathwalk data to drop
450 * @dentry: dentry to drop
451 * @Returns: 0 on success, -ECHLID on failure
453 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
454 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
455 * @nd. Must be called from rcu-walk context.
457 static int nameidata_dentry_drop_rcu(struct nameidata
*nd
, struct dentry
*dentry
)
459 struct fs_struct
*fs
= current
->fs
;
460 struct dentry
*parent
= nd
->path
.dentry
;
462 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
464 spin_lock(&fs
->lock
);
465 if (nd
->root
.mnt
!= fs
->root
.mnt
||
466 nd
->root
.dentry
!= fs
->root
.dentry
)
469 spin_lock(&parent
->d_lock
);
470 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
471 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
474 * If the sequence check on the child dentry passed, then the child has
475 * not been removed from its parent. This means the parent dentry must
476 * be valid and able to take a reference at this point.
478 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
479 BUG_ON(!parent
->d_count
);
481 spin_unlock(&dentry
->d_lock
);
482 spin_unlock(&parent
->d_lock
);
485 spin_unlock(&fs
->lock
);
487 mntget(nd
->path
.mnt
);
490 br_read_unlock(vfsmount_lock
);
491 nd
->flags
&= ~LOOKUP_RCU
;
494 spin_unlock(&dentry
->d_lock
);
495 spin_unlock(&parent
->d_lock
);
498 spin_unlock(&fs
->lock
);
502 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
503 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata
*nd
, struct dentry
*dentry
)
505 if (nd
->flags
& LOOKUP_RCU
)
506 return nameidata_dentry_drop_rcu(nd
, dentry
);
511 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
512 * @nd: nameidata pathwalk data to drop
513 * @Returns: 0 on success, -ECHLID on failure
515 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
516 * nd->path should be the final element of the lookup, so nd->root is discarded.
517 * Must be called from rcu-walk context.
519 static int nameidata_drop_rcu_last(struct nameidata
*nd
)
521 struct dentry
*dentry
= nd
->path
.dentry
;
523 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
524 nd
->flags
&= ~LOOKUP_RCU
;
526 spin_lock(&dentry
->d_lock
);
527 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
529 BUG_ON(nd
->inode
!= dentry
->d_inode
);
530 spin_unlock(&dentry
->d_lock
);
532 mntget(nd
->path
.mnt
);
535 br_read_unlock(vfsmount_lock
);
540 spin_unlock(&dentry
->d_lock
);
542 br_read_unlock(vfsmount_lock
);
546 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
547 static inline int nameidata_drop_rcu_last_maybe(struct nameidata
*nd
)
549 if (likely(nd
->flags
& LOOKUP_RCU
))
550 return nameidata_drop_rcu_last(nd
);
555 * release_open_intent - free up open intent resources
556 * @nd: pointer to nameidata
558 void release_open_intent(struct nameidata
*nd
)
560 if (nd
->intent
.open
.file
->f_path
.dentry
== NULL
)
561 put_filp(nd
->intent
.open
.file
);
563 fput(nd
->intent
.open
.file
);
566 static inline struct dentry
*
567 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
569 int status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
570 if (unlikely(status
<= 0)) {
572 * The dentry failed validation.
573 * If d_revalidate returned 0 attempt to invalidate
574 * the dentry otherwise d_revalidate is asking us
575 * to return a fail status.
578 if (!d_invalidate(dentry
)) {
584 dentry
= ERR_PTR(status
);
590 static inline int need_reval_dot(struct dentry
*dentry
)
592 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
595 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
602 * force_reval_path - force revalidation of a dentry
604 * In some situations the path walking code will trust dentries without
605 * revalidating them. This causes problems for filesystems that depend on
606 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
607 * (which indicates that it's possible for the dentry to go stale), force
608 * a d_revalidate call before proceeding.
610 * Returns 0 if the revalidation was successful. If the revalidation fails,
611 * either return the error returned by d_revalidate or -ESTALE if the
612 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
613 * invalidate the dentry. It's up to the caller to handle putting references
614 * to the path if necessary.
617 force_reval_path(struct path
*path
, struct nameidata
*nd
)
620 struct dentry
*dentry
= path
->dentry
;
623 * only check on filesystems where it's possible for the dentry to
626 if (!need_reval_dot(dentry
))
629 status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
634 d_invalidate(dentry
);
641 * Short-cut version of permission(), for calling on directories
642 * during pathname resolution. Combines parts of permission()
643 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
645 * If appropriate, check DAC only. If not appropriate, or
646 * short-cut DAC fails, then call ->permission() to do more
647 * complete permission check.
649 static inline int __exec_permission(struct inode
*inode
, int rcu
)
653 if (inode
->i_op
->permission
) {
656 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
);
661 ret
= __acl_permission_check(inode
, MAY_EXEC
, inode
->i_op
->check_acl
, rcu
);
664 if (rcu
&& ret
== -ECHILD
)
667 if (capable(CAP_DAC_OVERRIDE
) || capable(CAP_DAC_READ_SEARCH
))
672 return security_inode_exec_permission(inode
, rcu
);
675 static int exec_permission(struct inode
*inode
)
677 return __exec_permission(inode
, 0);
680 static int exec_permission_rcu(struct inode
*inode
)
682 return __exec_permission(inode
, 1);
685 static __always_inline
void set_root(struct nameidata
*nd
)
688 get_fs_root(current
->fs
, &nd
->root
);
691 static int link_path_walk(const char *, struct nameidata
*);
693 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
696 struct fs_struct
*fs
= current
->fs
;
700 seq
= read_seqcount_begin(&fs
->seq
);
702 } while (read_seqcount_retry(&fs
->seq
, seq
));
706 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
719 nd
->inode
= nd
->path
.dentry
->d_inode
;
721 ret
= link_path_walk(link
, nd
);
725 return PTR_ERR(link
);
728 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
731 if (path
->mnt
!= nd
->path
.mnt
)
735 static inline void path_to_nameidata(struct path
*path
, struct nameidata
*nd
)
737 if (!(nd
->flags
& LOOKUP_RCU
)) {
738 dput(nd
->path
.dentry
);
739 if (nd
->path
.mnt
!= path
->mnt
)
740 mntput(nd
->path
.mnt
);
742 nd
->path
.mnt
= path
->mnt
;
743 nd
->path
.dentry
= path
->dentry
;
746 static __always_inline
int
747 __do_follow_link(struct path
*path
, struct nameidata
*nd
, void **p
)
750 struct dentry
*dentry
= path
->dentry
;
752 touch_atime(path
->mnt
, dentry
);
753 nd_set_link(nd
, NULL
);
755 if (path
->mnt
!= nd
->path
.mnt
) {
756 path_to_nameidata(path
, nd
);
757 nd
->inode
= nd
->path
.dentry
->d_inode
;
762 nd
->last_type
= LAST_BIND
;
763 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
766 char *s
= nd_get_link(nd
);
769 error
= __vfs_follow_link(nd
, s
);
770 else if (nd
->last_type
== LAST_BIND
) {
771 error
= force_reval_path(&nd
->path
, nd
);
780 * This limits recursive symlink follows to 8, while
781 * limiting consecutive symlinks to 40.
783 * Without that kind of total limit, nasty chains of consecutive
784 * symlinks can cause almost arbitrarily long lookups.
786 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
790 if (current
->link_count
>= MAX_NESTED_LINKS
)
792 if (current
->total_link_count
>= 40)
794 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
796 err
= security_inode_follow_link(path
->dentry
, nd
);
799 current
->link_count
++;
800 current
->total_link_count
++;
802 err
= __do_follow_link(path
, nd
, &cookie
);
803 if (!IS_ERR(cookie
) && path
->dentry
->d_inode
->i_op
->put_link
)
804 path
->dentry
->d_inode
->i_op
->put_link(path
->dentry
, nd
, cookie
);
806 current
->link_count
--;
810 path_put_conditional(path
, nd
);
815 static int follow_up_rcu(struct path
*path
)
817 struct vfsmount
*parent
;
818 struct dentry
*mountpoint
;
820 parent
= path
->mnt
->mnt_parent
;
821 if (parent
== path
->mnt
)
823 mountpoint
= path
->mnt
->mnt_mountpoint
;
824 path
->dentry
= mountpoint
;
829 int follow_up(struct path
*path
)
831 struct vfsmount
*parent
;
832 struct dentry
*mountpoint
;
834 br_read_lock(vfsmount_lock
);
835 parent
= path
->mnt
->mnt_parent
;
836 if (parent
== path
->mnt
) {
837 br_read_unlock(vfsmount_lock
);
841 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
842 br_read_unlock(vfsmount_lock
);
844 path
->dentry
= mountpoint
;
851 * serialization is taken care of in namespace.c
853 static void __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
854 struct inode
**inode
)
856 while (d_mountpoint(path
->dentry
)) {
857 struct vfsmount
*mounted
;
858 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
862 path
->dentry
= mounted
->mnt_root
;
863 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
864 *inode
= path
->dentry
->d_inode
;
868 static int __follow_mount(struct path
*path
)
871 while (d_mountpoint(path
->dentry
)) {
872 struct vfsmount
*mounted
= lookup_mnt(path
);
879 path
->dentry
= dget(mounted
->mnt_root
);
885 static void follow_mount(struct path
*path
)
887 while (d_mountpoint(path
->dentry
)) {
888 struct vfsmount
*mounted
= lookup_mnt(path
);
894 path
->dentry
= dget(mounted
->mnt_root
);
898 int follow_down(struct path
*path
)
900 struct vfsmount
*mounted
;
902 mounted
= lookup_mnt(path
);
907 path
->dentry
= dget(mounted
->mnt_root
);
913 static int follow_dotdot_rcu(struct nameidata
*nd
)
915 struct inode
*inode
= nd
->inode
;
920 if (nd
->path
.dentry
== nd
->root
.dentry
&&
921 nd
->path
.mnt
== nd
->root
.mnt
) {
924 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
925 struct dentry
*old
= nd
->path
.dentry
;
926 struct dentry
*parent
= old
->d_parent
;
929 seq
= read_seqcount_begin(&parent
->d_seq
);
930 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
932 inode
= parent
->d_inode
;
933 nd
->path
.dentry
= parent
;
937 if (!follow_up_rcu(&nd
->path
))
939 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
940 inode
= nd
->path
.dentry
->d_inode
;
942 __follow_mount_rcu(nd
, &nd
->path
, &inode
);
948 static void follow_dotdot(struct nameidata
*nd
)
953 struct dentry
*old
= nd
->path
.dentry
;
955 if (nd
->path
.dentry
== nd
->root
.dentry
&&
956 nd
->path
.mnt
== nd
->root
.mnt
) {
959 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
960 /* rare case of legitimate dget_parent()... */
961 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
965 if (!follow_up(&nd
->path
))
968 follow_mount(&nd
->path
);
969 nd
->inode
= nd
->path
.dentry
->d_inode
;
973 * Allocate a dentry with name and parent, and perform a parent
974 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
975 * on error. parent->d_inode->i_mutex must be held. d_lookup must
976 * have verified that no child exists while under i_mutex.
978 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
979 struct qstr
*name
, struct nameidata
*nd
)
981 struct inode
*inode
= parent
->d_inode
;
982 struct dentry
*dentry
;
985 /* Don't create child dentry for a dead directory. */
986 if (unlikely(IS_DEADDIR(inode
)))
987 return ERR_PTR(-ENOENT
);
989 dentry
= d_alloc(parent
, name
);
990 if (unlikely(!dentry
))
991 return ERR_PTR(-ENOMEM
);
993 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1002 * It's more convoluted than I'd like it to be, but... it's still fairly
1003 * small and for now I'd prefer to have fast path as straight as possible.
1004 * It _is_ time-critical.
1006 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1007 struct path
*path
, struct inode
**inode
)
1009 struct vfsmount
*mnt
= nd
->path
.mnt
;
1010 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1013 * See if the low-level filesystem might want
1014 * to use its own hash..
1016 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1017 int err
= parent
->d_op
->d_hash(parent
, nd
->inode
, name
);
1023 * Rename seqlock is not required here because in the off chance
1024 * of a false negative due to a concurrent rename, we're going to
1025 * do the non-racy lookup, below.
1027 if (nd
->flags
& LOOKUP_RCU
) {
1031 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1033 if (nameidata_drop_rcu(nd
))
1037 /* Memory barrier in read_seqcount_begin of child is enough */
1038 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1042 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1043 /* We commonly drop rcu-walk here */
1044 if (nameidata_dentry_drop_rcu(nd
, dentry
))
1046 goto need_revalidate
;
1049 path
->dentry
= dentry
;
1050 __follow_mount_rcu(nd
, path
, inode
);
1052 dentry
= __d_lookup(parent
, name
);
1056 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
)
1057 goto need_revalidate
;
1060 path
->dentry
= dentry
;
1061 __follow_mount(path
);
1062 *inode
= path
->dentry
->d_inode
;
1067 dir
= parent
->d_inode
;
1068 BUG_ON(nd
->inode
!= dir
);
1070 mutex_lock(&dir
->i_mutex
);
1072 * First re-do the cached lookup just in case it was created
1073 * while we waited for the directory semaphore, or the first
1074 * lookup failed due to an unrelated rename.
1076 * This could use version numbering or similar to avoid unnecessary
1077 * cache lookups, but then we'd have to do the first lookup in the
1078 * non-racy way. However in the common case here, everything should
1079 * be hot in cache, so would it be a big win?
1081 dentry
= d_lookup(parent
, name
);
1082 if (likely(!dentry
)) {
1083 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1084 mutex_unlock(&dir
->i_mutex
);
1090 * Uhhuh! Nasty case: the cache was re-populated while
1091 * we waited on the semaphore. Need to revalidate.
1093 mutex_unlock(&dir
->i_mutex
);
1097 dentry
= do_revalidate(dentry
, nd
);
1105 return PTR_ERR(dentry
);
1109 * This is a temporary kludge to deal with "automount" symlinks; proper
1110 * solution is to trigger them on follow_mount(), so that do_lookup()
1111 * would DTRT. To be killed before 2.6.34-final.
1113 static inline int follow_on_final(struct inode
*inode
, unsigned lookup_flags
)
1115 return inode
&& unlikely(inode
->i_op
->follow_link
) &&
1116 ((lookup_flags
& LOOKUP_FOLLOW
) || S_ISDIR(inode
->i_mode
));
1121 * This is the basic name resolution function, turning a pathname into
1122 * the final dentry. We expect 'base' to be positive and a directory.
1124 * Returns 0 and nd will have valid dentry and mnt on success.
1125 * Returns error and drops reference to input namei data on failure.
1127 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1131 unsigned int lookup_flags
= nd
->flags
;
1139 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
1141 /* At this point we know we have a real path component. */
1143 struct inode
*inode
;
1148 nd
->flags
|= LOOKUP_CONTINUE
;
1149 if (nd
->flags
& LOOKUP_RCU
) {
1150 err
= exec_permission_rcu(nd
->inode
);
1151 if (err
== -ECHILD
) {
1152 if (nameidata_drop_rcu(nd
))
1158 err
= exec_permission(nd
->inode
);
1164 c
= *(const unsigned char *)name
;
1166 hash
= init_name_hash();
1169 hash
= partial_name_hash(c
, hash
);
1170 c
= *(const unsigned char *)name
;
1171 } while (c
&& (c
!= '/'));
1172 this.len
= name
- (const char *) this.name
;
1173 this.hash
= end_name_hash(hash
);
1175 /* remove trailing slashes? */
1177 goto last_component
;
1178 while (*++name
== '/');
1180 goto last_with_slashes
;
1183 * "." and ".." are special - ".." especially so because it has
1184 * to be able to know about the current root directory and
1185 * parent relationships.
1187 if (this.name
[0] == '.') switch (this.len
) {
1191 if (this.name
[1] != '.')
1193 if (nd
->flags
& LOOKUP_RCU
) {
1194 if (follow_dotdot_rcu(nd
))
1202 /* This does the actual lookups.. */
1203 err
= do_lookup(nd
, &this, &next
, &inode
);
1210 if (inode
->i_op
->follow_link
) {
1211 /* We commonly drop rcu-walk here */
1212 if (nameidata_dentry_drop_rcu_maybe(nd
, next
.dentry
))
1214 BUG_ON(inode
!= next
.dentry
->d_inode
);
1215 err
= do_follow_link(&next
, nd
);
1218 nd
->inode
= nd
->path
.dentry
->d_inode
;
1223 path_to_nameidata(&next
, nd
);
1227 if (!nd
->inode
->i_op
->lookup
)
1230 /* here ends the main loop */
1233 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1235 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1236 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1237 if (lookup_flags
& LOOKUP_PARENT
)
1239 if (this.name
[0] == '.') switch (this.len
) {
1243 if (this.name
[1] != '.')
1245 if (nd
->flags
& LOOKUP_RCU
) {
1246 if (follow_dotdot_rcu(nd
))
1254 err
= do_lookup(nd
, &this, &next
, &inode
);
1257 if (follow_on_final(inode
, lookup_flags
)) {
1258 if (nameidata_dentry_drop_rcu_maybe(nd
, next
.dentry
))
1260 BUG_ON(inode
!= next
.dentry
->d_inode
);
1261 err
= do_follow_link(&next
, nd
);
1264 nd
->inode
= nd
->path
.dentry
->d_inode
;
1266 path_to_nameidata(&next
, nd
);
1272 if (lookup_flags
& LOOKUP_DIRECTORY
) {
1274 if (!nd
->inode
->i_op
->lookup
)
1280 nd
->last_type
= LAST_NORM
;
1281 if (this.name
[0] != '.')
1284 nd
->last_type
= LAST_DOT
;
1285 else if (this.len
== 2 && this.name
[1] == '.')
1286 nd
->last_type
= LAST_DOTDOT
;
1291 * We bypassed the ordinary revalidation routines.
1292 * We may need to check the cached dentry for staleness.
1294 if (need_reval_dot(nd
->path
.dentry
)) {
1295 if (nameidata_drop_rcu_maybe(nd
))
1298 /* Note: we do not d_invalidate() */
1299 if (!nd
->path
.dentry
->d_op
->d_revalidate(
1300 nd
->path
.dentry
, nd
))
1304 if (nameidata_drop_rcu_last_maybe(nd
))
1308 if (!(nd
->flags
& LOOKUP_RCU
))
1309 path_put_conditional(&next
, nd
);
1312 if (!(nd
->flags
& LOOKUP_RCU
))
1313 path_put(&nd
->path
);
1318 static inline int path_walk_rcu(const char *name
, struct nameidata
*nd
)
1320 current
->total_link_count
= 0;
1322 return link_path_walk(name
, nd
);
1325 static inline int path_walk_simple(const char *name
, struct nameidata
*nd
)
1327 current
->total_link_count
= 0;
1329 return link_path_walk(name
, nd
);
1332 static int path_walk(const char *name
, struct nameidata
*nd
)
1334 struct path save
= nd
->path
;
1337 current
->total_link_count
= 0;
1339 /* make sure the stuff we saved doesn't go away */
1342 result
= link_path_walk(name
, nd
);
1343 if (result
== -ESTALE
) {
1344 /* nd->path had been dropped */
1345 current
->total_link_count
= 0;
1347 path_get(&nd
->path
);
1348 nd
->flags
|= LOOKUP_REVAL
;
1349 result
= link_path_walk(name
, nd
);
1357 static void path_finish_rcu(struct nameidata
*nd
)
1359 if (nd
->flags
& LOOKUP_RCU
) {
1360 /* RCU dangling. Cancel it. */
1361 nd
->flags
&= ~LOOKUP_RCU
;
1362 nd
->root
.mnt
= NULL
;
1364 br_read_unlock(vfsmount_lock
);
1370 static int path_init_rcu(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1376 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1377 nd
->flags
= flags
| LOOKUP_RCU
;
1379 nd
->root
.mnt
= NULL
;
1383 struct fs_struct
*fs
= current
->fs
;
1386 br_read_lock(vfsmount_lock
);
1390 seq
= read_seqcount_begin(&fs
->seq
);
1391 nd
->root
= fs
->root
;
1392 nd
->path
= nd
->root
;
1393 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1394 } while (read_seqcount_retry(&fs
->seq
, seq
));
1396 } else if (dfd
== AT_FDCWD
) {
1397 struct fs_struct
*fs
= current
->fs
;
1400 br_read_lock(vfsmount_lock
);
1404 seq
= read_seqcount_begin(&fs
->seq
);
1406 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1407 } while (read_seqcount_retry(&fs
->seq
, seq
));
1410 struct dentry
*dentry
;
1412 file
= fget_light(dfd
, &fput_needed
);
1417 dentry
= file
->f_path
.dentry
;
1420 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1423 retval
= file_permission(file
, MAY_EXEC
);
1427 nd
->path
= file
->f_path
;
1431 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1432 br_read_lock(vfsmount_lock
);
1435 nd
->inode
= nd
->path
.dentry
->d_inode
;
1439 fput_light(file
, fput_needed
);
1444 static int path_init(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1450 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1453 nd
->root
.mnt
= NULL
;
1457 nd
->path
= nd
->root
;
1458 path_get(&nd
->root
);
1459 } else if (dfd
== AT_FDCWD
) {
1460 get_fs_pwd(current
->fs
, &nd
->path
);
1462 struct dentry
*dentry
;
1464 file
= fget_light(dfd
, &fput_needed
);
1469 dentry
= file
->f_path
.dentry
;
1472 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1475 retval
= file_permission(file
, MAY_EXEC
);
1479 nd
->path
= file
->f_path
;
1480 path_get(&file
->f_path
);
1482 fput_light(file
, fput_needed
);
1484 nd
->inode
= nd
->path
.dentry
->d_inode
;
1488 fput_light(file
, fput_needed
);
1493 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1494 static int do_path_lookup(int dfd
, const char *name
,
1495 unsigned int flags
, struct nameidata
*nd
)
1500 * Path walking is largely split up into 2 different synchronisation
1501 * schemes, rcu-walk and ref-walk (explained in
1502 * Documentation/filesystems/path-lookup.txt). These share much of the
1503 * path walk code, but some things particularly setup, cleanup, and
1504 * following mounts are sufficiently divergent that functions are
1505 * duplicated. Typically there is a function foo(), and its RCU
1506 * analogue, foo_rcu().
1508 * -ECHILD is the error number of choice (just to avoid clashes) that
1509 * is returned if some aspect of an rcu-walk fails. Such an error must
1510 * be handled by restarting a traditional ref-walk (which will always
1511 * be able to complete).
1513 retval
= path_init_rcu(dfd
, name
, flags
, nd
);
1514 if (unlikely(retval
))
1516 retval
= path_walk_rcu(name
, nd
);
1517 path_finish_rcu(nd
);
1519 path_put(&nd
->root
);
1520 nd
->root
.mnt
= NULL
;
1523 if (unlikely(retval
== -ECHILD
|| retval
== -ESTALE
)) {
1524 /* slower, locked walk */
1525 if (retval
== -ESTALE
)
1526 flags
|= LOOKUP_REVAL
;
1527 retval
= path_init(dfd
, name
, flags
, nd
);
1528 if (unlikely(retval
))
1530 retval
= path_walk(name
, nd
);
1532 path_put(&nd
->root
);
1533 nd
->root
.mnt
= NULL
;
1537 if (likely(!retval
)) {
1538 if (unlikely(!audit_dummy_context())) {
1539 if (nd
->path
.dentry
&& nd
->inode
)
1540 audit_inode(name
, nd
->path
.dentry
);
1547 int path_lookup(const char *name
, unsigned int flags
,
1548 struct nameidata
*nd
)
1550 return do_path_lookup(AT_FDCWD
, name
, flags
, nd
);
1553 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1555 struct nameidata nd
;
1556 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1563 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1564 * @dentry: pointer to dentry of the base directory
1565 * @mnt: pointer to vfs mount of the base directory
1566 * @name: pointer to file name
1567 * @flags: lookup flags
1568 * @nd: pointer to nameidata
1570 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1571 const char *name
, unsigned int flags
,
1572 struct nameidata
*nd
)
1576 /* same as do_path_lookup */
1577 nd
->last_type
= LAST_ROOT
;
1581 nd
->path
.dentry
= dentry
;
1583 path_get(&nd
->path
);
1584 nd
->root
= nd
->path
;
1585 path_get(&nd
->root
);
1586 nd
->inode
= nd
->path
.dentry
->d_inode
;
1588 retval
= path_walk(name
, nd
);
1589 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1591 audit_inode(name
, nd
->path
.dentry
);
1593 path_put(&nd
->root
);
1594 nd
->root
.mnt
= NULL
;
1599 static struct dentry
*__lookup_hash(struct qstr
*name
,
1600 struct dentry
*base
, struct nameidata
*nd
)
1602 struct inode
*inode
= base
->d_inode
;
1603 struct dentry
*dentry
;
1606 err
= exec_permission(inode
);
1608 return ERR_PTR(err
);
1611 * See if the low-level filesystem might want
1612 * to use its own hash..
1614 if (base
->d_flags
& DCACHE_OP_HASH
) {
1615 err
= base
->d_op
->d_hash(base
, inode
, name
);
1616 dentry
= ERR_PTR(err
);
1622 * Don't bother with __d_lookup: callers are for creat as
1623 * well as unlink, so a lot of the time it would cost
1626 dentry
= d_lookup(base
, name
);
1628 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1629 dentry
= do_revalidate(dentry
, nd
);
1632 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1638 * Restricted form of lookup. Doesn't follow links, single-component only,
1639 * needs parent already locked. Doesn't follow mounts.
1642 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1644 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1647 static int __lookup_one_len(const char *name
, struct qstr
*this,
1648 struct dentry
*base
, int len
)
1658 hash
= init_name_hash();
1660 c
= *(const unsigned char *)name
++;
1661 if (c
== '/' || c
== '\0')
1663 hash
= partial_name_hash(c
, hash
);
1665 this->hash
= end_name_hash(hash
);
1670 * lookup_one_len - filesystem helper to lookup single pathname component
1671 * @name: pathname component to lookup
1672 * @base: base directory to lookup from
1673 * @len: maximum length @len should be interpreted to
1675 * Note that this routine is purely a helper for filesystem usage and should
1676 * not be called by generic code. Also note that by using this function the
1677 * nameidata argument is passed to the filesystem methods and a filesystem
1678 * using this helper needs to be prepared for that.
1680 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1685 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1687 err
= __lookup_one_len(name
, &this, base
, len
);
1689 return ERR_PTR(err
);
1691 return __lookup_hash(&this, base
, NULL
);
1694 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1697 struct nameidata nd
;
1698 char *tmp
= getname(name
);
1699 int err
= PTR_ERR(tmp
);
1702 BUG_ON(flags
& LOOKUP_PARENT
);
1704 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1712 static int user_path_parent(int dfd
, const char __user
*path
,
1713 struct nameidata
*nd
, char **name
)
1715 char *s
= getname(path
);
1721 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1731 * It's inline, so penalty for filesystems that don't use sticky bit is
1734 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1736 uid_t fsuid
= current_fsuid();
1738 if (!(dir
->i_mode
& S_ISVTX
))
1740 if (inode
->i_uid
== fsuid
)
1742 if (dir
->i_uid
== fsuid
)
1744 return !capable(CAP_FOWNER
);
1748 * Check whether we can remove a link victim from directory dir, check
1749 * whether the type of victim is right.
1750 * 1. We can't do it if dir is read-only (done in permission())
1751 * 2. We should have write and exec permissions on dir
1752 * 3. We can't remove anything from append-only dir
1753 * 4. We can't do anything with immutable dir (done in permission())
1754 * 5. If the sticky bit on dir is set we should either
1755 * a. be owner of dir, or
1756 * b. be owner of victim, or
1757 * c. have CAP_FOWNER capability
1758 * 6. If the victim is append-only or immutable we can't do antyhing with
1759 * links pointing to it.
1760 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1761 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1762 * 9. We can't remove a root or mountpoint.
1763 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1764 * nfs_async_unlink().
1766 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1770 if (!victim
->d_inode
)
1773 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1774 audit_inode_child(victim
, dir
);
1776 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1781 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1782 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1785 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1787 if (IS_ROOT(victim
))
1789 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1791 if (IS_DEADDIR(dir
))
1793 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1798 /* Check whether we can create an object with dentry child in directory
1800 * 1. We can't do it if child already exists (open has special treatment for
1801 * this case, but since we are inlined it's OK)
1802 * 2. We can't do it if dir is read-only (done in permission())
1803 * 3. We should have write and exec permissions on dir
1804 * 4. We can't do it if dir is immutable (done in permission())
1806 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1810 if (IS_DEADDIR(dir
))
1812 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1816 * p1 and p2 should be directories on the same fs.
1818 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1823 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1827 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1829 p
= d_ancestor(p2
, p1
);
1831 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1832 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1836 p
= d_ancestor(p1
, p2
);
1838 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1839 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1843 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1844 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1848 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1850 mutex_unlock(&p1
->d_inode
->i_mutex
);
1852 mutex_unlock(&p2
->d_inode
->i_mutex
);
1853 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1857 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1858 struct nameidata
*nd
)
1860 int error
= may_create(dir
, dentry
);
1865 if (!dir
->i_op
->create
)
1866 return -EACCES
; /* shouldn't it be ENOSYS? */
1869 error
= security_inode_create(dir
, dentry
, mode
);
1872 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1874 fsnotify_create(dir
, dentry
);
1878 int may_open(struct path
*path
, int acc_mode
, int flag
)
1880 struct dentry
*dentry
= path
->dentry
;
1881 struct inode
*inode
= dentry
->d_inode
;
1887 switch (inode
->i_mode
& S_IFMT
) {
1891 if (acc_mode
& MAY_WRITE
)
1896 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
1905 error
= inode_permission(inode
, acc_mode
);
1910 * An append-only file must be opened in append mode for writing.
1912 if (IS_APPEND(inode
)) {
1913 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
1919 /* O_NOATIME can only be set by the owner or superuser */
1920 if (flag
& O_NOATIME
&& !is_owner_or_cap(inode
))
1924 * Ensure there are no outstanding leases on the file.
1926 return break_lease(inode
, flag
);
1929 static int handle_truncate(struct path
*path
)
1931 struct inode
*inode
= path
->dentry
->d_inode
;
1932 int error
= get_write_access(inode
);
1936 * Refuse to truncate files with mandatory locks held on them.
1938 error
= locks_verify_locked(inode
);
1940 error
= security_path_truncate(path
);
1942 error
= do_truncate(path
->dentry
, 0,
1943 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
1946 put_write_access(inode
);
1951 * Be careful about ever adding any more callers of this
1952 * function. Its flags must be in the namei format, not
1953 * what get passed to sys_open().
1955 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
1956 int open_flag
, int mode
)
1959 struct dentry
*dir
= nd
->path
.dentry
;
1961 if (!IS_POSIXACL(dir
->d_inode
))
1962 mode
&= ~current_umask();
1963 error
= security_path_mknod(&nd
->path
, path
->dentry
, mode
, 0);
1966 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
1968 mutex_unlock(&dir
->d_inode
->i_mutex
);
1969 dput(nd
->path
.dentry
);
1970 nd
->path
.dentry
= path
->dentry
;
1974 /* Don't check for write permission, don't truncate */
1975 return may_open(&nd
->path
, 0, open_flag
& ~O_TRUNC
);
1979 * Note that while the flag value (low two bits) for sys_open means:
1984 * it is changed into
1985 * 00 - no permissions needed
1986 * 01 - read-permission
1987 * 10 - write-permission
1989 * for the internal routines (ie open_namei()/follow_link() etc)
1990 * This is more logical, and also allows the 00 "no perm needed"
1991 * to be used for symlinks (where the permissions are checked
1995 static inline int open_to_namei_flags(int flag
)
1997 if ((flag
+1) & O_ACCMODE
)
2002 static int open_will_truncate(int flag
, struct inode
*inode
)
2005 * We'll never write to the fs underlying
2008 if (special_file(inode
->i_mode
))
2010 return (flag
& O_TRUNC
);
2013 static struct file
*finish_open(struct nameidata
*nd
,
2014 int open_flag
, int acc_mode
)
2020 will_truncate
= open_will_truncate(open_flag
, nd
->path
.dentry
->d_inode
);
2021 if (will_truncate
) {
2022 error
= mnt_want_write(nd
->path
.mnt
);
2026 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2029 mnt_drop_write(nd
->path
.mnt
);
2032 filp
= nameidata_to_filp(nd
);
2033 if (!IS_ERR(filp
)) {
2034 error
= ima_file_check(filp
, acc_mode
);
2037 filp
= ERR_PTR(error
);
2040 if (!IS_ERR(filp
)) {
2041 if (will_truncate
) {
2042 error
= handle_truncate(&nd
->path
);
2045 filp
= ERR_PTR(error
);
2050 * It is now safe to drop the mnt write
2051 * because the filp has had a write taken
2055 mnt_drop_write(nd
->path
.mnt
);
2056 path_put(&nd
->path
);
2060 if (!IS_ERR(nd
->intent
.open
.file
))
2061 release_open_intent(nd
);
2062 path_put(&nd
->path
);
2063 return ERR_PTR(error
);
2067 * Handle O_CREAT case for do_filp_open
2069 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2070 int open_flag
, int acc_mode
,
2071 int mode
, const char *pathname
)
2073 struct dentry
*dir
= nd
->path
.dentry
;
2075 int error
= -EISDIR
;
2077 switch (nd
->last_type
) {
2080 dir
= nd
->path
.dentry
;
2082 if (need_reval_dot(dir
)) {
2083 if (!dir
->d_op
->d_revalidate(dir
, nd
)) {
2092 audit_inode(pathname
, dir
);
2096 /* trailing slashes? */
2097 if (nd
->last
.name
[nd
->last
.len
])
2100 mutex_lock(&dir
->d_inode
->i_mutex
);
2102 path
->dentry
= lookup_hash(nd
);
2103 path
->mnt
= nd
->path
.mnt
;
2105 error
= PTR_ERR(path
->dentry
);
2106 if (IS_ERR(path
->dentry
)) {
2107 mutex_unlock(&dir
->d_inode
->i_mutex
);
2111 if (IS_ERR(nd
->intent
.open
.file
)) {
2112 error
= PTR_ERR(nd
->intent
.open
.file
);
2113 goto exit_mutex_unlock
;
2116 /* Negative dentry, just create the file */
2117 if (!path
->dentry
->d_inode
) {
2119 * This write is needed to ensure that a
2120 * ro->rw transition does not occur between
2121 * the time when the file is created and when
2122 * a permanent write count is taken through
2123 * the 'struct file' in nameidata_to_filp().
2125 error
= mnt_want_write(nd
->path
.mnt
);
2127 goto exit_mutex_unlock
;
2128 error
= __open_namei_create(nd
, path
, open_flag
, mode
);
2130 mnt_drop_write(nd
->path
.mnt
);
2133 filp
= nameidata_to_filp(nd
);
2134 mnt_drop_write(nd
->path
.mnt
);
2135 path_put(&nd
->path
);
2136 if (!IS_ERR(filp
)) {
2137 error
= ima_file_check(filp
, acc_mode
);
2140 filp
= ERR_PTR(error
);
2147 * It already exists.
2149 mutex_unlock(&dir
->d_inode
->i_mutex
);
2150 audit_inode(pathname
, path
->dentry
);
2153 if (open_flag
& O_EXCL
)
2156 if (__follow_mount(path
)) {
2158 if (open_flag
& O_NOFOLLOW
)
2163 if (!path
->dentry
->d_inode
)
2166 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2169 path_to_nameidata(path
, nd
);
2170 nd
->inode
= path
->dentry
->d_inode
;
2172 if (S_ISDIR(nd
->inode
->i_mode
))
2175 filp
= finish_open(nd
, open_flag
, acc_mode
);
2179 mutex_unlock(&dir
->d_inode
->i_mutex
);
2181 path_put_conditional(path
, nd
);
2183 if (!IS_ERR(nd
->intent
.open
.file
))
2184 release_open_intent(nd
);
2185 path_put(&nd
->path
);
2186 return ERR_PTR(error
);
2190 * Note that the low bits of the passed in "open_flag"
2191 * are not the same as in the local variable "flag". See
2192 * open_to_namei_flags() for more details.
2194 struct file
*do_filp_open(int dfd
, const char *pathname
,
2195 int open_flag
, int mode
, int acc_mode
)
2198 struct nameidata nd
;
2202 int flag
= open_to_namei_flags(open_flag
);
2205 if (!(open_flag
& O_CREAT
))
2208 /* Must never be set by userspace */
2209 open_flag
&= ~FMODE_NONOTIFY
;
2212 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2213 * check for O_DSYNC if the need any syncing at all we enforce it's
2214 * always set instead of having to deal with possibly weird behaviour
2215 * for malicious applications setting only __O_SYNC.
2217 if (open_flag
& __O_SYNC
)
2218 open_flag
|= O_DSYNC
;
2221 acc_mode
= MAY_OPEN
| ACC_MODE(open_flag
);
2223 /* O_TRUNC implies we need access checks for write permissions */
2224 if (open_flag
& O_TRUNC
)
2225 acc_mode
|= MAY_WRITE
;
2227 /* Allow the LSM permission hook to distinguish append
2228 access from general write access. */
2229 if (open_flag
& O_APPEND
)
2230 acc_mode
|= MAY_APPEND
;
2232 flags
= LOOKUP_OPEN
;
2233 if (open_flag
& O_CREAT
) {
2234 flags
|= LOOKUP_CREATE
;
2235 if (open_flag
& O_EXCL
)
2236 flags
|= LOOKUP_EXCL
;
2238 if (open_flag
& O_DIRECTORY
)
2239 flags
|= LOOKUP_DIRECTORY
;
2240 if (!(open_flag
& O_NOFOLLOW
))
2241 flags
|= LOOKUP_FOLLOW
;
2243 filp
= get_empty_filp();
2245 return ERR_PTR(-ENFILE
);
2247 filp
->f_flags
= open_flag
;
2248 nd
.intent
.open
.file
= filp
;
2249 nd
.intent
.open
.flags
= flag
;
2250 nd
.intent
.open
.create_mode
= mode
;
2252 if (open_flag
& O_CREAT
)
2255 /* !O_CREAT, simple open */
2256 error
= do_path_lookup(dfd
, pathname
, flags
, &nd
);
2257 if (unlikely(error
))
2260 if (!(nd
.flags
& LOOKUP_FOLLOW
)) {
2261 if (nd
.inode
->i_op
->follow_link
)
2265 if (nd
.flags
& LOOKUP_DIRECTORY
) {
2266 if (!nd
.inode
->i_op
->lookup
)
2269 audit_inode(pathname
, nd
.path
.dentry
);
2270 filp
= finish_open(&nd
, open_flag
, acc_mode
);
2274 /* OK, have to create the file. Find the parent. */
2275 error
= path_init_rcu(dfd
, pathname
,
2276 LOOKUP_PARENT
| (flags
& LOOKUP_REVAL
), &nd
);
2279 error
= path_walk_rcu(pathname
, &nd
);
2280 path_finish_rcu(&nd
);
2281 if (unlikely(error
== -ECHILD
|| error
== -ESTALE
)) {
2282 /* slower, locked walk */
2283 if (error
== -ESTALE
) {
2285 flags
|= LOOKUP_REVAL
;
2287 error
= path_init(dfd
, pathname
,
2288 LOOKUP_PARENT
| (flags
& LOOKUP_REVAL
), &nd
);
2292 error
= path_walk_simple(pathname
, &nd
);
2294 if (unlikely(error
))
2296 if (unlikely(!audit_dummy_context()))
2297 audit_inode(pathname
, nd
.path
.dentry
);
2300 * We have the parent and last component.
2303 filp
= do_last(&nd
, &path
, open_flag
, acc_mode
, mode
, pathname
);
2304 while (unlikely(!filp
)) { /* trailing symlink */
2308 /* S_ISDIR part is a temporary automount kludge */
2309 if (!(nd
.flags
& LOOKUP_FOLLOW
) && !S_ISDIR(nd
.inode
->i_mode
))
2314 * This is subtle. Instead of calling do_follow_link() we do
2315 * the thing by hands. The reason is that this way we have zero
2316 * link_count and path_walk() (called from ->follow_link)
2317 * honoring LOOKUP_PARENT. After that we have the parent and
2318 * last component, i.e. we are in the same situation as after
2319 * the first path_walk(). Well, almost - if the last component
2320 * is normal we get its copy stored in nd->last.name and we will
2321 * have to putname() it when we are done. Procfs-like symlinks
2322 * just set LAST_BIND.
2324 nd
.flags
|= LOOKUP_PARENT
;
2325 error
= security_inode_follow_link(path
.dentry
, &nd
);
2328 error
= __do_follow_link(&path
, &nd
, &cookie
);
2329 if (unlikely(error
)) {
2330 if (!IS_ERR(cookie
) && nd
.inode
->i_op
->put_link
)
2331 nd
.inode
->i_op
->put_link(path
.dentry
, &nd
, cookie
);
2332 /* nd.path had been dropped */
2337 nd
.flags
&= ~LOOKUP_PARENT
;
2338 filp
= do_last(&nd
, &path
, open_flag
, acc_mode
, mode
, pathname
);
2339 if (nd
.inode
->i_op
->put_link
)
2340 nd
.inode
->i_op
->put_link(holder
.dentry
, &nd
, cookie
);
2346 if (filp
== ERR_PTR(-ESTALE
) && !(flags
& LOOKUP_REVAL
))
2351 path_put_conditional(&path
, &nd
);
2355 if (!IS_ERR(nd
.intent
.open
.file
))
2356 release_open_intent(&nd
);
2357 filp
= ERR_PTR(error
);
2362 * filp_open - open file and return file pointer
2364 * @filename: path to open
2365 * @flags: open flags as per the open(2) second argument
2366 * @mode: mode for the new file if O_CREAT is set, else ignored
2368 * This is the helper to open a file from kernelspace if you really
2369 * have to. But in generally you should not do this, so please move
2370 * along, nothing to see here..
2372 struct file
*filp_open(const char *filename
, int flags
, int mode
)
2374 return do_filp_open(AT_FDCWD
, filename
, flags
, mode
, 0);
2376 EXPORT_SYMBOL(filp_open
);
2379 * lookup_create - lookup a dentry, creating it if it doesn't exist
2380 * @nd: nameidata info
2381 * @is_dir: directory flag
2383 * Simple function to lookup and return a dentry and create it
2384 * if it doesn't exist. Is SMP-safe.
2386 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2388 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2390 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2392 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2394 * Yucky last component or no last component at all?
2395 * (foo/., foo/.., /////)
2397 if (nd
->last_type
!= LAST_NORM
)
2399 nd
->flags
&= ~LOOKUP_PARENT
;
2400 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2401 nd
->intent
.open
.flags
= O_EXCL
;
2404 * Do the final lookup.
2406 dentry
= lookup_hash(nd
);
2410 if (dentry
->d_inode
)
2413 * Special case - lookup gave negative, but... we had foo/bar/
2414 * From the vfs_mknod() POV we just have a negative dentry -
2415 * all is fine. Let's be bastards - you had / on the end, you've
2416 * been asking for (non-existent) directory. -ENOENT for you.
2418 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2420 dentry
= ERR_PTR(-ENOENT
);
2425 dentry
= ERR_PTR(-EEXIST
);
2429 EXPORT_SYMBOL_GPL(lookup_create
);
2431 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2433 int error
= may_create(dir
, dentry
);
2438 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
2441 if (!dir
->i_op
->mknod
)
2444 error
= devcgroup_inode_mknod(mode
, dev
);
2448 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2452 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2454 fsnotify_create(dir
, dentry
);
2458 static int may_mknod(mode_t mode
)
2460 switch (mode
& S_IFMT
) {
2466 case 0: /* zero mode translates to S_IFREG */
2475 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2480 struct dentry
*dentry
;
2481 struct nameidata nd
;
2486 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2490 dentry
= lookup_create(&nd
, 0);
2491 if (IS_ERR(dentry
)) {
2492 error
= PTR_ERR(dentry
);
2495 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2496 mode
&= ~current_umask();
2497 error
= may_mknod(mode
);
2500 error
= mnt_want_write(nd
.path
.mnt
);
2503 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2505 goto out_drop_write
;
2506 switch (mode
& S_IFMT
) {
2507 case 0: case S_IFREG
:
2508 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2510 case S_IFCHR
: case S_IFBLK
:
2511 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2512 new_decode_dev(dev
));
2514 case S_IFIFO
: case S_IFSOCK
:
2515 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2519 mnt_drop_write(nd
.path
.mnt
);
2523 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2530 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2532 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2535 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2537 int error
= may_create(dir
, dentry
);
2542 if (!dir
->i_op
->mkdir
)
2545 mode
&= (S_IRWXUGO
|S_ISVTX
);
2546 error
= security_inode_mkdir(dir
, dentry
, mode
);
2550 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2552 fsnotify_mkdir(dir
, dentry
);
2556 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2560 struct dentry
*dentry
;
2561 struct nameidata nd
;
2563 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2567 dentry
= lookup_create(&nd
, 1);
2568 error
= PTR_ERR(dentry
);
2572 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2573 mode
&= ~current_umask();
2574 error
= mnt_want_write(nd
.path
.mnt
);
2577 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2579 goto out_drop_write
;
2580 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2582 mnt_drop_write(nd
.path
.mnt
);
2586 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2593 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2595 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2599 * We try to drop the dentry early: we should have
2600 * a usage count of 2 if we're the only user of this
2601 * dentry, and if that is true (possibly after pruning
2602 * the dcache), then we drop the dentry now.
2604 * A low-level filesystem can, if it choses, legally
2607 * if (!d_unhashed(dentry))
2610 * if it cannot handle the case of removing a directory
2611 * that is still in use by something else..
2613 void dentry_unhash(struct dentry
*dentry
)
2616 shrink_dcache_parent(dentry
);
2617 spin_lock(&dentry
->d_lock
);
2618 if (dentry
->d_count
== 2)
2620 spin_unlock(&dentry
->d_lock
);
2623 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2625 int error
= may_delete(dir
, dentry
, 1);
2630 if (!dir
->i_op
->rmdir
)
2633 mutex_lock(&dentry
->d_inode
->i_mutex
);
2634 dentry_unhash(dentry
);
2635 if (d_mountpoint(dentry
))
2638 error
= security_inode_rmdir(dir
, dentry
);
2640 error
= dir
->i_op
->rmdir(dir
, dentry
);
2642 dentry
->d_inode
->i_flags
|= S_DEAD
;
2647 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2656 static long do_rmdir(int dfd
, const char __user
*pathname
)
2660 struct dentry
*dentry
;
2661 struct nameidata nd
;
2663 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2667 switch(nd
.last_type
) {
2679 nd
.flags
&= ~LOOKUP_PARENT
;
2681 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2682 dentry
= lookup_hash(&nd
);
2683 error
= PTR_ERR(dentry
);
2686 error
= mnt_want_write(nd
.path
.mnt
);
2689 error
= security_path_rmdir(&nd
.path
, dentry
);
2692 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2694 mnt_drop_write(nd
.path
.mnt
);
2698 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2705 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2707 return do_rmdir(AT_FDCWD
, pathname
);
2710 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2712 int error
= may_delete(dir
, dentry
, 0);
2717 if (!dir
->i_op
->unlink
)
2720 mutex_lock(&dentry
->d_inode
->i_mutex
);
2721 if (d_mountpoint(dentry
))
2724 error
= security_inode_unlink(dir
, dentry
);
2726 error
= dir
->i_op
->unlink(dir
, dentry
);
2731 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2733 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2734 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2735 fsnotify_link_count(dentry
->d_inode
);
2743 * Make sure that the actual truncation of the file will occur outside its
2744 * directory's i_mutex. Truncate can take a long time if there is a lot of
2745 * writeout happening, and we don't want to prevent access to the directory
2746 * while waiting on the I/O.
2748 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2752 struct dentry
*dentry
;
2753 struct nameidata nd
;
2754 struct inode
*inode
= NULL
;
2756 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2761 if (nd
.last_type
!= LAST_NORM
)
2764 nd
.flags
&= ~LOOKUP_PARENT
;
2766 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2767 dentry
= lookup_hash(&nd
);
2768 error
= PTR_ERR(dentry
);
2769 if (!IS_ERR(dentry
)) {
2770 /* Why not before? Because we want correct error value */
2771 if (nd
.last
.name
[nd
.last
.len
])
2773 inode
= dentry
->d_inode
;
2776 error
= mnt_want_write(nd
.path
.mnt
);
2779 error
= security_path_unlink(&nd
.path
, dentry
);
2782 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2784 mnt_drop_write(nd
.path
.mnt
);
2788 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2790 iput(inode
); /* truncate the inode here */
2797 error
= !dentry
->d_inode
? -ENOENT
:
2798 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2802 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2804 if ((flag
& ~AT_REMOVEDIR
) != 0)
2807 if (flag
& AT_REMOVEDIR
)
2808 return do_rmdir(dfd
, pathname
);
2810 return do_unlinkat(dfd
, pathname
);
2813 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2815 return do_unlinkat(AT_FDCWD
, pathname
);
2818 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2820 int error
= may_create(dir
, dentry
);
2825 if (!dir
->i_op
->symlink
)
2828 error
= security_inode_symlink(dir
, dentry
, oldname
);
2832 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2834 fsnotify_create(dir
, dentry
);
2838 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2839 int, newdfd
, const char __user
*, newname
)
2844 struct dentry
*dentry
;
2845 struct nameidata nd
;
2847 from
= getname(oldname
);
2849 return PTR_ERR(from
);
2851 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2855 dentry
= lookup_create(&nd
, 0);
2856 error
= PTR_ERR(dentry
);
2860 error
= mnt_want_write(nd
.path
.mnt
);
2863 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2865 goto out_drop_write
;
2866 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2868 mnt_drop_write(nd
.path
.mnt
);
2872 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2880 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2882 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2885 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2887 struct inode
*inode
= old_dentry
->d_inode
;
2893 error
= may_create(dir
, new_dentry
);
2897 if (dir
->i_sb
!= inode
->i_sb
)
2901 * A link to an append-only or immutable file cannot be created.
2903 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2905 if (!dir
->i_op
->link
)
2907 if (S_ISDIR(inode
->i_mode
))
2910 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2914 mutex_lock(&inode
->i_mutex
);
2915 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2916 mutex_unlock(&inode
->i_mutex
);
2918 fsnotify_link(dir
, inode
, new_dentry
);
2923 * Hardlinks are often used in delicate situations. We avoid
2924 * security-related surprises by not following symlinks on the
2927 * We don't follow them on the oldname either to be compatible
2928 * with linux 2.0, and to avoid hard-linking to directories
2929 * and other special files. --ADM
2931 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2932 int, newdfd
, const char __user
*, newname
, int, flags
)
2934 struct dentry
*new_dentry
;
2935 struct nameidata nd
;
2936 struct path old_path
;
2940 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
2943 error
= user_path_at(olddfd
, oldname
,
2944 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
2949 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2953 if (old_path
.mnt
!= nd
.path
.mnt
)
2955 new_dentry
= lookup_create(&nd
, 0);
2956 error
= PTR_ERR(new_dentry
);
2957 if (IS_ERR(new_dentry
))
2959 error
= mnt_want_write(nd
.path
.mnt
);
2962 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2964 goto out_drop_write
;
2965 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2967 mnt_drop_write(nd
.path
.mnt
);
2971 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2976 path_put(&old_path
);
2981 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
2983 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2987 * The worst of all namespace operations - renaming directory. "Perverted"
2988 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2990 * a) we can get into loop creation. Check is done in is_subdir().
2991 * b) race potential - two innocent renames can create a loop together.
2992 * That's where 4.4 screws up. Current fix: serialization on
2993 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2995 * c) we have to lock _three_ objects - parents and victim (if it exists).
2996 * And that - after we got ->i_mutex on parents (until then we don't know
2997 * whether the target exists). Solution: try to be smart with locking
2998 * order for inodes. We rely on the fact that tree topology may change
2999 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3000 * move will be locked. Thus we can rank directories by the tree
3001 * (ancestors first) and rank all non-directories after them.
3002 * That works since everybody except rename does "lock parent, lookup,
3003 * lock child" and rename is under ->s_vfs_rename_mutex.
3004 * HOWEVER, it relies on the assumption that any object with ->lookup()
3005 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3006 * we'd better make sure that there's no link(2) for them.
3007 * d) some filesystems don't support opened-but-unlinked directories,
3008 * either because of layout or because they are not ready to deal with
3009 * all cases correctly. The latter will be fixed (taking this sort of
3010 * stuff into VFS), but the former is not going away. Solution: the same
3011 * trick as in rmdir().
3012 * e) conversion from fhandle to dentry may come in the wrong moment - when
3013 * we are removing the target. Solution: we will have to grab ->i_mutex
3014 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3015 * ->i_mutex on parents, which works but leads to some truly excessive
3018 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3019 struct inode
*new_dir
, struct dentry
*new_dentry
)
3022 struct inode
*target
;
3025 * If we are going to change the parent - check write permissions,
3026 * we'll need to flip '..'.
3028 if (new_dir
!= old_dir
) {
3029 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3034 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3038 target
= new_dentry
->d_inode
;
3040 mutex_lock(&target
->i_mutex
);
3041 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3045 dentry_unhash(new_dentry
);
3046 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3050 target
->i_flags
|= S_DEAD
;
3051 dont_mount(new_dentry
);
3053 mutex_unlock(&target
->i_mutex
);
3054 if (d_unhashed(new_dentry
))
3055 d_rehash(new_dentry
);
3059 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3060 d_move(old_dentry
,new_dentry
);
3064 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3065 struct inode
*new_dir
, struct dentry
*new_dentry
)
3067 struct inode
*target
;
3070 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3075 target
= new_dentry
->d_inode
;
3077 mutex_lock(&target
->i_mutex
);
3078 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3081 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3084 dont_mount(new_dentry
);
3085 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3086 d_move(old_dentry
, new_dentry
);
3089 mutex_unlock(&target
->i_mutex
);
3094 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3095 struct inode
*new_dir
, struct dentry
*new_dentry
)
3098 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3099 const unsigned char *old_name
;
3101 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3104 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3108 if (!new_dentry
->d_inode
)
3109 error
= may_create(new_dir
, new_dentry
);
3111 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3115 if (!old_dir
->i_op
->rename
)
3118 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3121 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3123 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3125 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3126 new_dentry
->d_inode
, old_dentry
);
3127 fsnotify_oldname_free(old_name
);
3132 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3133 int, newdfd
, const char __user
*, newname
)
3135 struct dentry
*old_dir
, *new_dir
;
3136 struct dentry
*old_dentry
, *new_dentry
;
3137 struct dentry
*trap
;
3138 struct nameidata oldnd
, newnd
;
3143 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3147 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3152 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3155 old_dir
= oldnd
.path
.dentry
;
3157 if (oldnd
.last_type
!= LAST_NORM
)
3160 new_dir
= newnd
.path
.dentry
;
3161 if (newnd
.last_type
!= LAST_NORM
)
3164 oldnd
.flags
&= ~LOOKUP_PARENT
;
3165 newnd
.flags
&= ~LOOKUP_PARENT
;
3166 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3168 trap
= lock_rename(new_dir
, old_dir
);
3170 old_dentry
= lookup_hash(&oldnd
);
3171 error
= PTR_ERR(old_dentry
);
3172 if (IS_ERR(old_dentry
))
3174 /* source must exist */
3176 if (!old_dentry
->d_inode
)
3178 /* unless the source is a directory trailing slashes give -ENOTDIR */
3179 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3181 if (oldnd
.last
.name
[oldnd
.last
.len
])
3183 if (newnd
.last
.name
[newnd
.last
.len
])
3186 /* source should not be ancestor of target */
3188 if (old_dentry
== trap
)
3190 new_dentry
= lookup_hash(&newnd
);
3191 error
= PTR_ERR(new_dentry
);
3192 if (IS_ERR(new_dentry
))
3194 /* target should not be an ancestor of source */
3196 if (new_dentry
== trap
)
3199 error
= mnt_want_write(oldnd
.path
.mnt
);
3202 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3203 &newnd
.path
, new_dentry
);
3206 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3207 new_dir
->d_inode
, new_dentry
);
3209 mnt_drop_write(oldnd
.path
.mnt
);
3215 unlock_rename(new_dir
, old_dir
);
3217 path_put(&newnd
.path
);
3220 path_put(&oldnd
.path
);
3226 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3228 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3231 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3235 len
= PTR_ERR(link
);
3240 if (len
> (unsigned) buflen
)
3242 if (copy_to_user(buffer
, link
, len
))
3249 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3250 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3251 * using) it for any given inode is up to filesystem.
3253 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3255 struct nameidata nd
;
3260 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3262 return PTR_ERR(cookie
);
3264 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3265 if (dentry
->d_inode
->i_op
->put_link
)
3266 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3270 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3272 return __vfs_follow_link(nd
, link
);
3275 /* get the link contents into pagecache */
3276 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3280 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3281 page
= read_mapping_page(mapping
, 0, NULL
);
3286 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3290 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3292 struct page
*page
= NULL
;
3293 char *s
= page_getlink(dentry
, &page
);
3294 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3297 page_cache_release(page
);
3302 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3304 struct page
*page
= NULL
;
3305 nd_set_link(nd
, page_getlink(dentry
, &page
));
3309 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3311 struct page
*page
= cookie
;
3315 page_cache_release(page
);
3320 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3322 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3324 struct address_space
*mapping
= inode
->i_mapping
;
3329 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3331 flags
|= AOP_FLAG_NOFS
;
3334 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3335 flags
, &page
, &fsdata
);
3339 kaddr
= kmap_atomic(page
, KM_USER0
);
3340 memcpy(kaddr
, symname
, len
-1);
3341 kunmap_atomic(kaddr
, KM_USER0
);
3343 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3350 mark_inode_dirty(inode
);
3356 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3358 return __page_symlink(inode
, symname
, len
,
3359 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3362 const struct inode_operations page_symlink_inode_operations
= {
3363 .readlink
= generic_readlink
,
3364 .follow_link
= page_follow_link_light
,
3365 .put_link
= page_put_link
,
3368 EXPORT_SYMBOL(user_path_at
);
3369 EXPORT_SYMBOL(follow_down
);
3370 EXPORT_SYMBOL(follow_up
);
3371 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3372 EXPORT_SYMBOL(getname
);
3373 EXPORT_SYMBOL(lock_rename
);
3374 EXPORT_SYMBOL(lookup_one_len
);
3375 EXPORT_SYMBOL(page_follow_link_light
);
3376 EXPORT_SYMBOL(page_put_link
);
3377 EXPORT_SYMBOL(page_readlink
);
3378 EXPORT_SYMBOL(__page_symlink
);
3379 EXPORT_SYMBOL(page_symlink
);
3380 EXPORT_SYMBOL(page_symlink_inode_operations
);
3381 EXPORT_SYMBOL(path_lookup
);
3382 EXPORT_SYMBOL(kern_path
);
3383 EXPORT_SYMBOL(vfs_path_lookup
);
3384 EXPORT_SYMBOL(inode_permission
);
3385 EXPORT_SYMBOL(file_permission
);
3386 EXPORT_SYMBOL(unlock_rename
);
3387 EXPORT_SYMBOL(vfs_create
);
3388 EXPORT_SYMBOL(vfs_follow_link
);
3389 EXPORT_SYMBOL(vfs_link
);
3390 EXPORT_SYMBOL(vfs_mkdir
);
3391 EXPORT_SYMBOL(vfs_mknod
);
3392 EXPORT_SYMBOL(generic_permission
);
3393 EXPORT_SYMBOL(vfs_readlink
);
3394 EXPORT_SYMBOL(vfs_rename
);
3395 EXPORT_SYMBOL(vfs_rmdir
);
3396 EXPORT_SYMBOL(vfs_symlink
);
3397 EXPORT_SYMBOL(vfs_unlink
);
3398 EXPORT_SYMBOL(dentry_unhash
);
3399 EXPORT_SYMBOL(generic_readlink
);