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/export.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 <linux/posix_acl.h>
36 #include <asm/uaccess.h>
41 /* [Feb-1997 T. Schoebel-Theuer]
42 * Fundamental changes in the pathname lookup mechanisms (namei)
43 * were necessary because of omirr. The reason is that omirr needs
44 * to know the _real_ pathname, not the user-supplied one, in case
45 * of symlinks (and also when transname replacements occur).
47 * The new code replaces the old recursive symlink resolution with
48 * an iterative one (in case of non-nested symlink chains). It does
49 * this with calls to <fs>_follow_link().
50 * As a side effect, dir_namei(), _namei() and follow_link() are now
51 * replaced with a single function lookup_dentry() that can handle all
52 * the special cases of the former code.
54 * With the new dcache, the pathname is stored at each inode, at least as
55 * long as the refcount of the inode is positive. As a side effect, the
56 * size of the dcache depends on the inode cache and thus is dynamic.
58 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
59 * resolution to correspond with current state of the code.
61 * Note that the symlink resolution is not *completely* iterative.
62 * There is still a significant amount of tail- and mid- recursion in
63 * the algorithm. Also, note that <fs>_readlink() is not used in
64 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
65 * may return different results than <fs>_follow_link(). Many virtual
66 * filesystems (including /proc) exhibit this behavior.
69 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
70 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
71 * and the name already exists in form of a symlink, try to create the new
72 * name indicated by the symlink. The old code always complained that the
73 * name already exists, due to not following the symlink even if its target
74 * is nonexistent. The new semantics affects also mknod() and link() when
75 * the name is a symlink pointing to a non-existent name.
77 * I don't know which semantics is the right one, since I have no access
78 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
79 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
80 * "old" one. Personally, I think the new semantics is much more logical.
81 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
82 * file does succeed in both HP-UX and SunOs, but not in Solaris
83 * and in the old Linux semantics.
86 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
87 * semantics. See the comments in "open_namei" and "do_link" below.
89 * [10-Sep-98 Alan Modra] Another symlink change.
92 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
93 * inside the path - always follow.
94 * in the last component in creation/removal/renaming - never follow.
95 * if LOOKUP_FOLLOW passed - follow.
96 * if the pathname has trailing slashes - follow.
97 * otherwise - don't follow.
98 * (applied in that order).
100 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
101 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
102 * During the 2.4 we need to fix the userland stuff depending on it -
103 * hopefully we will be able to get rid of that wart in 2.5. So far only
104 * XEmacs seems to be relying on it...
107 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
108 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
109 * any extra contention...
112 /* In order to reduce some races, while at the same time doing additional
113 * checking and hopefully speeding things up, we copy filenames to the
114 * kernel data space before using them..
116 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
117 * PATH_MAX includes the nul terminator --RR.
119 static char *getname_flags(const char __user
*filename
, int flags
, int *empty
)
121 char *result
= __getname(), *err
;
124 if (unlikely(!result
))
125 return ERR_PTR(-ENOMEM
);
127 len
= strncpy_from_user(result
, filename
, PATH_MAX
);
129 if (unlikely(len
< 0))
132 /* The empty path is special. */
133 if (unlikely(!len
)) {
136 err
= ERR_PTR(-ENOENT
);
137 if (!(flags
& LOOKUP_EMPTY
))
141 err
= ERR_PTR(-ENAMETOOLONG
);
142 if (likely(len
< PATH_MAX
)) {
143 audit_getname(result
);
152 char *getname(const char __user
* filename
)
154 return getname_flags(filename
, 0, NULL
);
157 #ifdef CONFIG_AUDITSYSCALL
158 void putname(const char *name
)
160 if (unlikely(!audit_dummy_context()))
165 EXPORT_SYMBOL(putname
);
168 static int check_acl(struct inode
*inode
, int mask
)
170 #ifdef CONFIG_FS_POSIX_ACL
171 struct posix_acl
*acl
;
173 if (mask
& MAY_NOT_BLOCK
) {
174 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
177 /* no ->get_acl() calls in RCU mode... */
178 if (acl
== ACL_NOT_CACHED
)
180 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
183 acl
= get_cached_acl(inode
, ACL_TYPE_ACCESS
);
186 * A filesystem can force a ACL callback by just never filling the
187 * ACL cache. But normally you'd fill the cache either at inode
188 * instantiation time, or on the first ->get_acl call.
190 * If the filesystem doesn't have a get_acl() function at all, we'll
191 * just create the negative cache entry.
193 if (acl
== ACL_NOT_CACHED
) {
194 if (inode
->i_op
->get_acl
) {
195 acl
= inode
->i_op
->get_acl(inode
, ACL_TYPE_ACCESS
);
199 set_cached_acl(inode
, ACL_TYPE_ACCESS
, NULL
);
205 int error
= posix_acl_permission(inode
, acl
, mask
);
206 posix_acl_release(acl
);
215 * This does the basic permission checking
217 static int acl_permission_check(struct inode
*inode
, int mask
)
219 unsigned int mode
= inode
->i_mode
;
221 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
224 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
225 int error
= check_acl(inode
, mask
);
226 if (error
!= -EAGAIN
)
230 if (in_group_p(inode
->i_gid
))
235 * If the DACs are ok we don't need any capability check.
237 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
243 * generic_permission - check for access rights on a Posix-like filesystem
244 * @inode: inode to check access rights for
245 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
247 * Used to check for read/write/execute permissions on a file.
248 * We use "fsuid" for this, letting us set arbitrary permissions
249 * for filesystem access without changing the "normal" uids which
250 * are used for other things.
252 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
253 * request cannot be satisfied (eg. requires blocking or too much complexity).
254 * It would then be called again in ref-walk mode.
256 int generic_permission(struct inode
*inode
, int mask
)
261 * Do the basic permission checks.
263 ret
= acl_permission_check(inode
, mask
);
267 if (S_ISDIR(inode
->i_mode
)) {
268 /* DACs are overridable for directories */
269 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
271 if (!(mask
& MAY_WRITE
))
272 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
277 * Read/write DACs are always overridable.
278 * Executable DACs are overridable when there is
279 * at least one exec bit set.
281 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
282 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
286 * Searching includes executable on directories, else just read.
288 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
289 if (mask
== MAY_READ
)
290 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
297 * We _really_ want to just do "generic_permission()" without
298 * even looking at the inode->i_op values. So we keep a cache
299 * flag in inode->i_opflags, that says "this has not special
300 * permission function, use the fast case".
302 static inline int do_inode_permission(struct inode
*inode
, int mask
)
304 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
305 if (likely(inode
->i_op
->permission
))
306 return inode
->i_op
->permission(inode
, mask
);
308 /* This gets set once for the inode lifetime */
309 spin_lock(&inode
->i_lock
);
310 inode
->i_opflags
|= IOP_FASTPERM
;
311 spin_unlock(&inode
->i_lock
);
313 return generic_permission(inode
, mask
);
317 * inode_permission - check for access rights to a given inode
318 * @inode: inode to check permission on
319 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
321 * Used to check for read/write/execute permissions on an inode.
322 * We use "fsuid" for this, letting us set arbitrary permissions
323 * for filesystem access without changing the "normal" uids which
324 * are used for other things.
326 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
328 int inode_permission(struct inode
*inode
, int mask
)
332 if (unlikely(mask
& MAY_WRITE
)) {
333 umode_t mode
= inode
->i_mode
;
336 * Nobody gets write access to a read-only fs.
338 if (IS_RDONLY(inode
) &&
339 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
343 * Nobody gets write access to an immutable file.
345 if (IS_IMMUTABLE(inode
))
349 retval
= do_inode_permission(inode
, mask
);
353 retval
= devcgroup_inode_permission(inode
, mask
);
357 return security_inode_permission(inode
, mask
);
361 * path_get - get a reference to a path
362 * @path: path to get the reference to
364 * Given a path increment the reference count to the dentry and the vfsmount.
366 void path_get(struct path
*path
)
371 EXPORT_SYMBOL(path_get
);
374 * path_put - put a reference to a path
375 * @path: path to put the reference to
377 * Given a path decrement the reference count to the dentry and the vfsmount.
379 void path_put(struct path
*path
)
384 EXPORT_SYMBOL(path_put
);
387 * Path walking has 2 modes, rcu-walk and ref-walk (see
388 * Documentation/filesystems/path-lookup.txt). In situations when we can't
389 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
390 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
391 * mode. Refcounts are grabbed at the last known good point before rcu-walk
392 * got stuck, so ref-walk may continue from there. If this is not successful
393 * (eg. a seqcount has changed), then failure is returned and it's up to caller
394 * to restart the path walk from the beginning in ref-walk mode.
398 * unlazy_walk - try to switch to ref-walk mode.
399 * @nd: nameidata pathwalk data
400 * @dentry: child of nd->path.dentry or NULL
401 * Returns: 0 on success, -ECHILD on failure
403 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
404 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
405 * @nd or NULL. Must be called from rcu-walk context.
407 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
409 struct fs_struct
*fs
= current
->fs
;
410 struct dentry
*parent
= nd
->path
.dentry
;
413 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
414 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
416 spin_lock(&fs
->lock
);
417 if (nd
->root
.mnt
!= fs
->root
.mnt
||
418 nd
->root
.dentry
!= fs
->root
.dentry
)
421 spin_lock(&parent
->d_lock
);
423 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
425 BUG_ON(nd
->inode
!= parent
->d_inode
);
427 if (dentry
->d_parent
!= parent
)
429 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
430 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
433 * If the sequence check on the child dentry passed, then
434 * the child has not been removed from its parent. This
435 * means the parent dentry must be valid and able to take
436 * a reference at this point.
438 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
439 BUG_ON(!parent
->d_count
);
441 spin_unlock(&dentry
->d_lock
);
443 spin_unlock(&parent
->d_lock
);
446 spin_unlock(&fs
->lock
);
448 mntget(nd
->path
.mnt
);
451 br_read_unlock(vfsmount_lock
);
452 nd
->flags
&= ~LOOKUP_RCU
;
456 spin_unlock(&dentry
->d_lock
);
458 spin_unlock(&parent
->d_lock
);
461 spin_unlock(&fs
->lock
);
466 * release_open_intent - free up open intent resources
467 * @nd: pointer to nameidata
469 void release_open_intent(struct nameidata
*nd
)
471 struct file
*file
= nd
->intent
.open
.file
;
473 if (file
&& !IS_ERR(file
)) {
474 if (file
->f_path
.dentry
== NULL
)
481 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
483 return dentry
->d_op
->d_revalidate(dentry
, nd
);
487 * complete_walk - successful completion of path walk
488 * @nd: pointer nameidata
490 * If we had been in RCU mode, drop out of it and legitimize nd->path.
491 * Revalidate the final result, unless we'd already done that during
492 * the path walk or the filesystem doesn't ask for it. Return 0 on
493 * success, -error on failure. In case of failure caller does not
494 * need to drop nd->path.
496 static int complete_walk(struct nameidata
*nd
)
498 struct dentry
*dentry
= nd
->path
.dentry
;
501 if (nd
->flags
& LOOKUP_RCU
) {
502 nd
->flags
&= ~LOOKUP_RCU
;
503 if (!(nd
->flags
& LOOKUP_ROOT
))
505 spin_lock(&dentry
->d_lock
);
506 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
507 spin_unlock(&dentry
->d_lock
);
509 br_read_unlock(vfsmount_lock
);
512 BUG_ON(nd
->inode
!= dentry
->d_inode
);
513 spin_unlock(&dentry
->d_lock
);
514 mntget(nd
->path
.mnt
);
516 br_read_unlock(vfsmount_lock
);
519 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
522 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
525 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
528 /* Note: we do not d_invalidate() */
529 status
= d_revalidate(dentry
, nd
);
540 static __always_inline
void set_root(struct nameidata
*nd
)
543 get_fs_root(current
->fs
, &nd
->root
);
546 static int link_path_walk(const char *, struct nameidata
*);
548 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
551 struct fs_struct
*fs
= current
->fs
;
555 seq
= read_seqcount_begin(&fs
->seq
);
557 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
558 } while (read_seqcount_retry(&fs
->seq
, seq
));
562 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
574 nd
->flags
|= LOOKUP_JUMPED
;
576 nd
->inode
= nd
->path
.dentry
->d_inode
;
578 ret
= link_path_walk(link
, nd
);
582 return PTR_ERR(link
);
585 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
588 if (path
->mnt
!= nd
->path
.mnt
)
592 static inline void path_to_nameidata(const struct path
*path
,
593 struct nameidata
*nd
)
595 if (!(nd
->flags
& LOOKUP_RCU
)) {
596 dput(nd
->path
.dentry
);
597 if (nd
->path
.mnt
!= path
->mnt
)
598 mntput(nd
->path
.mnt
);
600 nd
->path
.mnt
= path
->mnt
;
601 nd
->path
.dentry
= path
->dentry
;
604 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
606 struct inode
*inode
= link
->dentry
->d_inode
;
607 if (!IS_ERR(cookie
) && inode
->i_op
->put_link
)
608 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
612 static __always_inline
int
613 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
616 struct dentry
*dentry
= link
->dentry
;
618 BUG_ON(nd
->flags
& LOOKUP_RCU
);
620 if (link
->mnt
== nd
->path
.mnt
)
623 if (unlikely(current
->total_link_count
>= 40)) {
624 *p
= ERR_PTR(-ELOOP
); /* no ->put_link(), please */
629 current
->total_link_count
++;
632 nd_set_link(nd
, NULL
);
634 error
= security_inode_follow_link(link
->dentry
, nd
);
636 *p
= ERR_PTR(error
); /* no ->put_link(), please */
641 nd
->last_type
= LAST_BIND
;
642 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
645 char *s
= nd_get_link(nd
);
648 error
= __vfs_follow_link(nd
, s
);
649 else if (nd
->last_type
== LAST_BIND
) {
650 nd
->flags
|= LOOKUP_JUMPED
;
651 nd
->inode
= nd
->path
.dentry
->d_inode
;
652 if (nd
->inode
->i_op
->follow_link
) {
653 /* stepped on a _really_ weird one */
662 static int follow_up_rcu(struct path
*path
)
664 struct mount
*mnt
= real_mount(path
->mnt
);
665 struct mount
*parent
;
666 struct dentry
*mountpoint
;
668 parent
= mnt
->mnt_parent
;
669 if (&parent
->mnt
== path
->mnt
)
671 mountpoint
= mnt
->mnt_mountpoint
;
672 path
->dentry
= mountpoint
;
673 path
->mnt
= &parent
->mnt
;
677 int follow_up(struct path
*path
)
679 struct mount
*mnt
= real_mount(path
->mnt
);
680 struct mount
*parent
;
681 struct dentry
*mountpoint
;
683 br_read_lock(vfsmount_lock
);
684 parent
= mnt
->mnt_parent
;
685 if (&parent
->mnt
== path
->mnt
) {
686 br_read_unlock(vfsmount_lock
);
689 mntget(&parent
->mnt
);
690 mountpoint
= dget(mnt
->mnt_mountpoint
);
691 br_read_unlock(vfsmount_lock
);
693 path
->dentry
= mountpoint
;
695 path
->mnt
= &parent
->mnt
;
700 * Perform an automount
701 * - return -EISDIR to tell follow_managed() to stop and return the path we
704 static int follow_automount(struct path
*path
, unsigned flags
,
707 struct vfsmount
*mnt
;
710 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
713 /* We don't want to mount if someone's just doing a stat -
714 * unless they're stat'ing a directory and appended a '/' to
717 * We do, however, want to mount if someone wants to open or
718 * create a file of any type under the mountpoint, wants to
719 * traverse through the mountpoint or wants to open the
720 * mounted directory. Also, autofs may mark negative dentries
721 * as being automount points. These will need the attentions
722 * of the daemon to instantiate them before they can be used.
724 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
725 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
726 path
->dentry
->d_inode
)
729 current
->total_link_count
++;
730 if (current
->total_link_count
>= 40)
733 mnt
= path
->dentry
->d_op
->d_automount(path
);
736 * The filesystem is allowed to return -EISDIR here to indicate
737 * it doesn't want to automount. For instance, autofs would do
738 * this so that its userspace daemon can mount on this dentry.
740 * However, we can only permit this if it's a terminal point in
741 * the path being looked up; if it wasn't then the remainder of
742 * the path is inaccessible and we should say so.
744 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
749 if (!mnt
) /* mount collision */
753 /* lock_mount() may release path->mnt on error */
757 err
= finish_automount(mnt
, path
);
761 /* Someone else made a mount here whilst we were busy */
766 path
->dentry
= dget(mnt
->mnt_root
);
775 * Handle a dentry that is managed in some way.
776 * - Flagged for transit management (autofs)
777 * - Flagged as mountpoint
778 * - Flagged as automount point
780 * This may only be called in refwalk mode.
782 * Serialization is taken care of in namespace.c
784 static int follow_managed(struct path
*path
, unsigned flags
)
786 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
788 bool need_mntput
= false;
791 /* Given that we're not holding a lock here, we retain the value in a
792 * local variable for each dentry as we look at it so that we don't see
793 * the components of that value change under us */
794 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
795 managed
&= DCACHE_MANAGED_DENTRY
,
796 unlikely(managed
!= 0)) {
797 /* Allow the filesystem to manage the transit without i_mutex
799 if (managed
& DCACHE_MANAGE_TRANSIT
) {
800 BUG_ON(!path
->dentry
->d_op
);
801 BUG_ON(!path
->dentry
->d_op
->d_manage
);
802 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
807 /* Transit to a mounted filesystem. */
808 if (managed
& DCACHE_MOUNTED
) {
809 struct vfsmount
*mounted
= lookup_mnt(path
);
815 path
->dentry
= dget(mounted
->mnt_root
);
820 /* Something is mounted on this dentry in another
821 * namespace and/or whatever was mounted there in this
822 * namespace got unmounted before we managed to get the
826 /* Handle an automount point */
827 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
828 ret
= follow_automount(path
, flags
, &need_mntput
);
834 /* We didn't change the current path point */
838 if (need_mntput
&& path
->mnt
== mnt
)
842 return ret
< 0 ? ret
: need_mntput
;
845 int follow_down_one(struct path
*path
)
847 struct vfsmount
*mounted
;
849 mounted
= lookup_mnt(path
);
854 path
->dentry
= dget(mounted
->mnt_root
);
860 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
862 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
863 dentry
->d_op
->d_manage(dentry
, true) < 0);
867 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
868 * we meet a managed dentry that would need blocking.
870 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
871 struct inode
**inode
)
874 struct mount
*mounted
;
876 * Don't forget we might have a non-mountpoint managed dentry
877 * that wants to block transit.
879 if (unlikely(managed_dentry_might_block(path
->dentry
)))
882 if (!d_mountpoint(path
->dentry
))
885 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
888 path
->mnt
= &mounted
->mnt
;
889 path
->dentry
= mounted
->mnt
.mnt_root
;
890 nd
->flags
|= LOOKUP_JUMPED
;
891 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
893 * Update the inode too. We don't need to re-check the
894 * dentry sequence number here after this d_inode read,
895 * because a mount-point is always pinned.
897 *inode
= path
->dentry
->d_inode
;
902 static void follow_mount_rcu(struct nameidata
*nd
)
904 while (d_mountpoint(nd
->path
.dentry
)) {
905 struct mount
*mounted
;
906 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
909 nd
->path
.mnt
= &mounted
->mnt
;
910 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
911 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
915 static int follow_dotdot_rcu(struct nameidata
*nd
)
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 nd
->path
.dentry
= parent
;
936 if (!follow_up_rcu(&nd
->path
))
938 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
940 follow_mount_rcu(nd
);
941 nd
->inode
= nd
->path
.dentry
->d_inode
;
945 nd
->flags
&= ~LOOKUP_RCU
;
946 if (!(nd
->flags
& LOOKUP_ROOT
))
949 br_read_unlock(vfsmount_lock
);
954 * Follow down to the covering mount currently visible to userspace. At each
955 * point, the filesystem owning that dentry may be queried as to whether the
956 * caller is permitted to proceed or not.
958 int follow_down(struct path
*path
)
963 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
964 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
965 /* Allow the filesystem to manage the transit without i_mutex
968 * We indicate to the filesystem if someone is trying to mount
969 * something here. This gives autofs the chance to deny anyone
970 * other than its daemon the right to mount on its
973 * The filesystem may sleep at this point.
975 if (managed
& DCACHE_MANAGE_TRANSIT
) {
976 BUG_ON(!path
->dentry
->d_op
);
977 BUG_ON(!path
->dentry
->d_op
->d_manage
);
978 ret
= path
->dentry
->d_op
->d_manage(
979 path
->dentry
, false);
981 return ret
== -EISDIR
? 0 : ret
;
984 /* Transit to a mounted filesystem. */
985 if (managed
& DCACHE_MOUNTED
) {
986 struct vfsmount
*mounted
= lookup_mnt(path
);
992 path
->dentry
= dget(mounted
->mnt_root
);
996 /* Don't handle automount points here */
1003 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1005 static void follow_mount(struct path
*path
)
1007 while (d_mountpoint(path
->dentry
)) {
1008 struct vfsmount
*mounted
= lookup_mnt(path
);
1013 path
->mnt
= mounted
;
1014 path
->dentry
= dget(mounted
->mnt_root
);
1018 static void follow_dotdot(struct nameidata
*nd
)
1023 struct dentry
*old
= nd
->path
.dentry
;
1025 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1026 nd
->path
.mnt
== nd
->root
.mnt
) {
1029 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1030 /* rare case of legitimate dget_parent()... */
1031 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1035 if (!follow_up(&nd
->path
))
1038 follow_mount(&nd
->path
);
1039 nd
->inode
= nd
->path
.dentry
->d_inode
;
1043 * This looks up the name in dcache, possibly revalidates the old dentry and
1044 * allocates a new one if not found or not valid. In the need_lookup argument
1045 * returns whether i_op->lookup is necessary.
1047 * dir->d_inode->i_mutex must be held
1049 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1050 struct nameidata
*nd
, bool *need_lookup
)
1052 struct dentry
*dentry
;
1055 *need_lookup
= false;
1056 dentry
= d_lookup(dir
, name
);
1058 if (d_need_lookup(dentry
)) {
1059 *need_lookup
= true;
1060 } else if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1061 error
= d_revalidate(dentry
, nd
);
1062 if (unlikely(error
<= 0)) {
1065 return ERR_PTR(error
);
1066 } else if (!d_invalidate(dentry
)) {
1075 dentry
= d_alloc(dir
, name
);
1076 if (unlikely(!dentry
))
1077 return ERR_PTR(-ENOMEM
);
1079 *need_lookup
= true;
1085 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1086 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1088 * dir->d_inode->i_mutex must be held
1090 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1091 struct nameidata
*nd
)
1095 /* Don't create child dentry for a dead directory. */
1096 if (unlikely(IS_DEADDIR(dir
))) {
1098 return ERR_PTR(-ENOENT
);
1101 old
= dir
->i_op
->lookup(dir
, dentry
, nd
);
1102 if (unlikely(old
)) {
1109 static struct dentry
*__lookup_hash(struct qstr
*name
,
1110 struct dentry
*base
, struct nameidata
*nd
)
1113 struct dentry
*dentry
;
1115 dentry
= lookup_dcache(name
, base
, nd
, &need_lookup
);
1119 return lookup_real(base
->d_inode
, dentry
, nd
);
1123 * It's more convoluted than I'd like it to be, but... it's still fairly
1124 * small and for now I'd prefer to have fast path as straight as possible.
1125 * It _is_ time-critical.
1127 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1128 struct path
*path
, struct inode
**inode
)
1130 struct vfsmount
*mnt
= nd
->path
.mnt
;
1131 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1137 * Rename seqlock is not required here because in the off chance
1138 * of a false negative due to a concurrent rename, we're going to
1139 * do the non-racy lookup, below.
1141 if (nd
->flags
& LOOKUP_RCU
) {
1143 dentry
= __d_lookup_rcu(parent
, name
, &seq
, nd
->inode
);
1148 * This sequence count validates that the inode matches
1149 * the dentry name information from lookup.
1151 *inode
= dentry
->d_inode
;
1152 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1156 * This sequence count validates that the parent had no
1157 * changes while we did the lookup of the dentry above.
1159 * The memory barrier in read_seqcount_begin of child is
1160 * enough, we can use __read_seqcount_retry here.
1162 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1166 if (unlikely(d_need_lookup(dentry
)))
1168 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1169 status
= d_revalidate(dentry
, nd
);
1170 if (unlikely(status
<= 0)) {
1171 if (status
!= -ECHILD
)
1177 path
->dentry
= dentry
;
1178 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1180 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1184 if (unlazy_walk(nd
, dentry
))
1187 dentry
= __d_lookup(parent
, name
);
1190 if (unlikely(!dentry
))
1193 if (unlikely(d_need_lookup(dentry
))) {
1198 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1199 status
= d_revalidate(dentry
, nd
);
1200 if (unlikely(status
<= 0)) {
1205 if (!d_invalidate(dentry
)) {
1212 path
->dentry
= dentry
;
1213 err
= follow_managed(path
, nd
->flags
);
1214 if (unlikely(err
< 0)) {
1215 path_put_conditional(path
, nd
);
1219 nd
->flags
|= LOOKUP_JUMPED
;
1220 *inode
= path
->dentry
->d_inode
;
1224 BUG_ON(nd
->inode
!= parent
->d_inode
);
1226 mutex_lock(&parent
->d_inode
->i_mutex
);
1227 dentry
= __lookup_hash(name
, parent
, nd
);
1228 mutex_unlock(&parent
->d_inode
->i_mutex
);
1230 return PTR_ERR(dentry
);
1234 static inline int may_lookup(struct nameidata
*nd
)
1236 if (nd
->flags
& LOOKUP_RCU
) {
1237 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1240 if (unlazy_walk(nd
, NULL
))
1243 return inode_permission(nd
->inode
, MAY_EXEC
);
1246 static inline int handle_dots(struct nameidata
*nd
, int type
)
1248 if (type
== LAST_DOTDOT
) {
1249 if (nd
->flags
& LOOKUP_RCU
) {
1250 if (follow_dotdot_rcu(nd
))
1258 static void terminate_walk(struct nameidata
*nd
)
1260 if (!(nd
->flags
& LOOKUP_RCU
)) {
1261 path_put(&nd
->path
);
1263 nd
->flags
&= ~LOOKUP_RCU
;
1264 if (!(nd
->flags
& LOOKUP_ROOT
))
1265 nd
->root
.mnt
= NULL
;
1267 br_read_unlock(vfsmount_lock
);
1272 * Do we need to follow links? We _really_ want to be able
1273 * to do this check without having to look at inode->i_op,
1274 * so we keep a cache of "no, this doesn't need follow_link"
1275 * for the common case.
1277 static inline int should_follow_link(struct inode
*inode
, int follow
)
1279 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1280 if (likely(inode
->i_op
->follow_link
))
1283 /* This gets set once for the inode lifetime */
1284 spin_lock(&inode
->i_lock
);
1285 inode
->i_opflags
|= IOP_NOFOLLOW
;
1286 spin_unlock(&inode
->i_lock
);
1291 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1292 struct qstr
*name
, int type
, int follow
)
1294 struct inode
*inode
;
1297 * "." and ".." are special - ".." especially so because it has
1298 * to be able to know about the current root directory and
1299 * parent relationships.
1301 if (unlikely(type
!= LAST_NORM
))
1302 return handle_dots(nd
, type
);
1303 err
= do_lookup(nd
, name
, path
, &inode
);
1304 if (unlikely(err
)) {
1309 path_to_nameidata(path
, nd
);
1313 if (should_follow_link(inode
, follow
)) {
1314 if (nd
->flags
& LOOKUP_RCU
) {
1315 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1320 BUG_ON(inode
!= path
->dentry
->d_inode
);
1323 path_to_nameidata(path
, nd
);
1329 * This limits recursive symlink follows to 8, while
1330 * limiting consecutive symlinks to 40.
1332 * Without that kind of total limit, nasty chains of consecutive
1333 * symlinks can cause almost arbitrarily long lookups.
1335 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1339 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1340 path_put_conditional(path
, nd
);
1341 path_put(&nd
->path
);
1344 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1347 current
->link_count
++;
1350 struct path link
= *path
;
1353 res
= follow_link(&link
, nd
, &cookie
);
1355 res
= walk_component(nd
, path
, &nd
->last
,
1356 nd
->last_type
, LOOKUP_FOLLOW
);
1357 put_link(nd
, &link
, cookie
);
1360 current
->link_count
--;
1366 * We really don't want to look at inode->i_op->lookup
1367 * when we don't have to. So we keep a cache bit in
1368 * the inode ->i_opflags field that says "yes, we can
1369 * do lookup on this inode".
1371 static inline int can_lookup(struct inode
*inode
)
1373 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1375 if (likely(!inode
->i_op
->lookup
))
1378 /* We do this once for the lifetime of the inode */
1379 spin_lock(&inode
->i_lock
);
1380 inode
->i_opflags
|= IOP_LOOKUP
;
1381 spin_unlock(&inode
->i_lock
);
1386 * We can do the critical dentry name comparison and hashing
1387 * operations one word at a time, but we are limited to:
1389 * - Architectures with fast unaligned word accesses. We could
1390 * do a "get_unaligned()" if this helps and is sufficiently
1393 * - Little-endian machines (so that we can generate the mask
1394 * of low bytes efficiently). Again, we *could* do a byte
1395 * swapping load on big-endian architectures if that is not
1396 * expensive enough to make the optimization worthless.
1398 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1399 * do not trap on the (extremely unlikely) case of a page
1400 * crossing operation.
1402 * - Furthermore, we need an efficient 64-bit compile for the
1403 * 64-bit case in order to generate the "number of bytes in
1404 * the final mask". Again, that could be replaced with a
1405 * efficient population count instruction or similar.
1407 #ifdef CONFIG_DCACHE_WORD_ACCESS
1409 #include <asm/word-at-a-time.h>
1413 static inline unsigned int fold_hash(unsigned long hash
)
1415 hash
+= hash
>> (8*sizeof(int));
1419 #else /* 32-bit case */
1421 #define fold_hash(x) (x)
1425 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1427 unsigned long a
, mask
;
1428 unsigned long hash
= 0;
1431 a
= load_unaligned_zeropad(name
);
1432 if (len
< sizeof(unsigned long))
1436 name
+= sizeof(unsigned long);
1437 len
-= sizeof(unsigned long);
1441 mask
= ~(~0ul << len
*8);
1444 return fold_hash(hash
);
1446 EXPORT_SYMBOL(full_name_hash
);
1449 * Calculate the length and hash of the path component, and
1450 * return the length of the component;
1452 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1454 unsigned long a
, mask
, hash
, len
;
1457 len
= -sizeof(unsigned long);
1459 hash
= (hash
+ a
) * 9;
1460 len
+= sizeof(unsigned long);
1461 a
= load_unaligned_zeropad(name
+len
);
1462 /* Do we have any NUL or '/' bytes in this word? */
1463 mask
= has_zero(a
) | has_zero(a
^ REPEAT_BYTE('/'));
1466 /* The mask *below* the first high bit set */
1467 mask
= (mask
- 1) & ~mask
;
1470 *hashp
= fold_hash(hash
);
1472 return len
+ count_masked_bytes(mask
);
1477 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1479 unsigned long hash
= init_name_hash();
1481 hash
= partial_name_hash(*name
++, hash
);
1482 return end_name_hash(hash
);
1484 EXPORT_SYMBOL(full_name_hash
);
1487 * We know there's a real path component here of at least
1490 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1492 unsigned long hash
= init_name_hash();
1493 unsigned long len
= 0, c
;
1495 c
= (unsigned char)*name
;
1498 hash
= partial_name_hash(c
, hash
);
1499 c
= (unsigned char)name
[len
];
1500 } while (c
&& c
!= '/');
1501 *hashp
= end_name_hash(hash
);
1509 * This is the basic name resolution function, turning a pathname into
1510 * the final dentry. We expect 'base' to be positive and a directory.
1512 * Returns 0 and nd will have valid dentry and mnt on success.
1513 * Returns error and drops reference to input namei data on failure.
1515 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1525 /* At this point we know we have a real path component. */
1531 err
= may_lookup(nd
);
1535 len
= hash_name(name
, &this.hash
);
1540 if (name
[0] == '.') switch (len
) {
1542 if (name
[1] == '.') {
1544 nd
->flags
|= LOOKUP_JUMPED
;
1550 if (likely(type
== LAST_NORM
)) {
1551 struct dentry
*parent
= nd
->path
.dentry
;
1552 nd
->flags
&= ~LOOKUP_JUMPED
;
1553 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1554 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1562 goto last_component
;
1564 * If it wasn't NUL, we know it was '/'. Skip that
1565 * slash, and continue until no more slashes.
1569 } while (unlikely(name
[len
] == '/'));
1571 goto last_component
;
1574 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1579 err
= nested_symlink(&next
, nd
);
1583 if (can_lookup(nd
->inode
))
1587 /* here ends the main loop */
1591 nd
->last_type
= type
;
1598 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1599 struct nameidata
*nd
, struct file
**fp
)
1605 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1606 nd
->flags
= flags
| LOOKUP_JUMPED
;
1608 if (flags
& LOOKUP_ROOT
) {
1609 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1611 if (!inode
->i_op
->lookup
)
1613 retval
= inode_permission(inode
, MAY_EXEC
);
1617 nd
->path
= nd
->root
;
1619 if (flags
& LOOKUP_RCU
) {
1620 br_read_lock(vfsmount_lock
);
1622 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1624 path_get(&nd
->path
);
1629 nd
->root
.mnt
= NULL
;
1632 if (flags
& LOOKUP_RCU
) {
1633 br_read_lock(vfsmount_lock
);
1638 path_get(&nd
->root
);
1640 nd
->path
= nd
->root
;
1641 } else if (dfd
== AT_FDCWD
) {
1642 if (flags
& LOOKUP_RCU
) {
1643 struct fs_struct
*fs
= current
->fs
;
1646 br_read_lock(vfsmount_lock
);
1650 seq
= read_seqcount_begin(&fs
->seq
);
1652 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1653 } while (read_seqcount_retry(&fs
->seq
, seq
));
1655 get_fs_pwd(current
->fs
, &nd
->path
);
1658 struct dentry
*dentry
;
1660 file
= fget_raw_light(dfd
, &fput_needed
);
1665 dentry
= file
->f_path
.dentry
;
1669 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1672 retval
= inode_permission(dentry
->d_inode
, MAY_EXEC
);
1677 nd
->path
= file
->f_path
;
1678 if (flags
& LOOKUP_RCU
) {
1681 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1682 br_read_lock(vfsmount_lock
);
1685 path_get(&file
->f_path
);
1686 fput_light(file
, fput_needed
);
1690 nd
->inode
= nd
->path
.dentry
->d_inode
;
1694 fput_light(file
, fput_needed
);
1699 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1701 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1702 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1704 nd
->flags
&= ~LOOKUP_PARENT
;
1705 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1706 nd
->flags
& LOOKUP_FOLLOW
);
1709 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1710 static int path_lookupat(int dfd
, const char *name
,
1711 unsigned int flags
, struct nameidata
*nd
)
1713 struct file
*base
= NULL
;
1718 * Path walking is largely split up into 2 different synchronisation
1719 * schemes, rcu-walk and ref-walk (explained in
1720 * Documentation/filesystems/path-lookup.txt). These share much of the
1721 * path walk code, but some things particularly setup, cleanup, and
1722 * following mounts are sufficiently divergent that functions are
1723 * duplicated. Typically there is a function foo(), and its RCU
1724 * analogue, foo_rcu().
1726 * -ECHILD is the error number of choice (just to avoid clashes) that
1727 * is returned if some aspect of an rcu-walk fails. Such an error must
1728 * be handled by restarting a traditional ref-walk (which will always
1729 * be able to complete).
1731 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1736 current
->total_link_count
= 0;
1737 err
= link_path_walk(name
, nd
);
1739 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1740 err
= lookup_last(nd
, &path
);
1743 struct path link
= path
;
1744 nd
->flags
|= LOOKUP_PARENT
;
1745 err
= follow_link(&link
, nd
, &cookie
);
1747 err
= lookup_last(nd
, &path
);
1748 put_link(nd
, &link
, cookie
);
1753 err
= complete_walk(nd
);
1755 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1756 if (!nd
->inode
->i_op
->lookup
) {
1757 path_put(&nd
->path
);
1765 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1766 path_put(&nd
->root
);
1767 nd
->root
.mnt
= NULL
;
1772 static int do_path_lookup(int dfd
, const char *name
,
1773 unsigned int flags
, struct nameidata
*nd
)
1775 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1776 if (unlikely(retval
== -ECHILD
))
1777 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1778 if (unlikely(retval
== -ESTALE
))
1779 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1781 if (likely(!retval
)) {
1782 if (unlikely(!audit_dummy_context())) {
1783 if (nd
->path
.dentry
&& nd
->inode
)
1784 audit_inode(name
, nd
->path
.dentry
);
1790 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1792 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1795 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1797 struct nameidata nd
;
1798 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1805 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1806 * @dentry: pointer to dentry of the base directory
1807 * @mnt: pointer to vfs mount of the base directory
1808 * @name: pointer to file name
1809 * @flags: lookup flags
1810 * @path: pointer to struct path to fill
1812 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1813 const char *name
, unsigned int flags
,
1816 struct nameidata nd
;
1818 nd
.root
.dentry
= dentry
;
1820 BUG_ON(flags
& LOOKUP_PARENT
);
1821 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1822 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
1829 * Restricted form of lookup. Doesn't follow links, single-component only,
1830 * needs parent already locked. Doesn't follow mounts.
1833 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1835 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1839 * lookup_one_len - filesystem helper to lookup single pathname component
1840 * @name: pathname component to lookup
1841 * @base: base directory to lookup from
1842 * @len: maximum length @len should be interpreted to
1844 * Note that this routine is purely a helper for filesystem usage and should
1845 * not be called by generic code. Also note that by using this function the
1846 * nameidata argument is passed to the filesystem methods and a filesystem
1847 * using this helper needs to be prepared for that.
1849 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1855 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1859 this.hash
= full_name_hash(name
, len
);
1861 return ERR_PTR(-EACCES
);
1864 c
= *(const unsigned char *)name
++;
1865 if (c
== '/' || c
== '\0')
1866 return ERR_PTR(-EACCES
);
1869 * See if the low-level filesystem might want
1870 * to use its own hash..
1872 if (base
->d_flags
& DCACHE_OP_HASH
) {
1873 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1875 return ERR_PTR(err
);
1878 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
1880 return ERR_PTR(err
);
1882 return __lookup_hash(&this, base
, NULL
);
1885 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
1886 struct path
*path
, int *empty
)
1888 struct nameidata nd
;
1889 char *tmp
= getname_flags(name
, flags
, empty
);
1890 int err
= PTR_ERR(tmp
);
1893 BUG_ON(flags
& LOOKUP_PARENT
);
1895 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1903 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1906 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
1909 static int user_path_parent(int dfd
, const char __user
*path
,
1910 struct nameidata
*nd
, char **name
)
1912 char *s
= getname(path
);
1918 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1928 * It's inline, so penalty for filesystems that don't use sticky bit is
1931 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1933 kuid_t fsuid
= current_fsuid();
1935 if (!(dir
->i_mode
& S_ISVTX
))
1937 if (uid_eq(inode
->i_uid
, fsuid
))
1939 if (uid_eq(dir
->i_uid
, fsuid
))
1941 return !inode_capable(inode
, CAP_FOWNER
);
1945 * Check whether we can remove a link victim from directory dir, check
1946 * whether the type of victim is right.
1947 * 1. We can't do it if dir is read-only (done in permission())
1948 * 2. We should have write and exec permissions on dir
1949 * 3. We can't remove anything from append-only dir
1950 * 4. We can't do anything with immutable dir (done in permission())
1951 * 5. If the sticky bit on dir is set we should either
1952 * a. be owner of dir, or
1953 * b. be owner of victim, or
1954 * c. have CAP_FOWNER capability
1955 * 6. If the victim is append-only or immutable we can't do antyhing with
1956 * links pointing to it.
1957 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1958 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1959 * 9. We can't remove a root or mountpoint.
1960 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1961 * nfs_async_unlink().
1963 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1967 if (!victim
->d_inode
)
1970 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1971 audit_inode_child(victim
, dir
);
1973 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1978 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1979 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1982 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1984 if (IS_ROOT(victim
))
1986 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1988 if (IS_DEADDIR(dir
))
1990 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1995 /* Check whether we can create an object with dentry child in directory
1997 * 1. We can't do it if child already exists (open has special treatment for
1998 * this case, but since we are inlined it's OK)
1999 * 2. We can't do it if dir is read-only (done in permission())
2000 * 3. We should have write and exec permissions on dir
2001 * 4. We can't do it if dir is immutable (done in permission())
2003 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2007 if (IS_DEADDIR(dir
))
2009 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2013 * p1 and p2 should be directories on the same fs.
2015 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2020 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2024 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2026 p
= d_ancestor(p2
, p1
);
2028 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2029 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2033 p
= d_ancestor(p1
, p2
);
2035 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2036 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2040 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2041 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2045 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2047 mutex_unlock(&p1
->d_inode
->i_mutex
);
2049 mutex_unlock(&p2
->d_inode
->i_mutex
);
2050 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2054 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2055 struct nameidata
*nd
)
2057 int error
= may_create(dir
, dentry
);
2062 if (!dir
->i_op
->create
)
2063 return -EACCES
; /* shouldn't it be ENOSYS? */
2066 error
= security_inode_create(dir
, dentry
, mode
);
2069 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
2071 fsnotify_create(dir
, dentry
);
2075 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2077 struct dentry
*dentry
= path
->dentry
;
2078 struct inode
*inode
= dentry
->d_inode
;
2088 switch (inode
->i_mode
& S_IFMT
) {
2092 if (acc_mode
& MAY_WRITE
)
2097 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2106 error
= inode_permission(inode
, acc_mode
);
2111 * An append-only file must be opened in append mode for writing.
2113 if (IS_APPEND(inode
)) {
2114 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2120 /* O_NOATIME can only be set by the owner or superuser */
2121 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2127 static int handle_truncate(struct file
*filp
)
2129 struct path
*path
= &filp
->f_path
;
2130 struct inode
*inode
= path
->dentry
->d_inode
;
2131 int error
= get_write_access(inode
);
2135 * Refuse to truncate files with mandatory locks held on them.
2137 error
= locks_verify_locked(inode
);
2139 error
= security_path_truncate(path
);
2141 error
= do_truncate(path
->dentry
, 0,
2142 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2145 put_write_access(inode
);
2149 static inline int open_to_namei_flags(int flag
)
2151 if ((flag
& O_ACCMODE
) == 3)
2157 * Handle the last step of open()
2159 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2160 const struct open_flags
*op
, const char *pathname
)
2162 struct dentry
*dir
= nd
->path
.dentry
;
2163 struct dentry
*dentry
;
2164 int open_flag
= op
->open_flag
;
2165 int will_truncate
= open_flag
& O_TRUNC
;
2167 int acc_mode
= op
->acc_mode
;
2171 nd
->flags
&= ~LOOKUP_PARENT
;
2172 nd
->flags
|= op
->intent
;
2174 switch (nd
->last_type
) {
2177 error
= handle_dots(nd
, nd
->last_type
);
2179 return ERR_PTR(error
);
2182 error
= complete_walk(nd
);
2184 return ERR_PTR(error
);
2185 audit_inode(pathname
, nd
->path
.dentry
);
2186 if (open_flag
& O_CREAT
) {
2192 error
= complete_walk(nd
);
2194 return ERR_PTR(error
);
2195 audit_inode(pathname
, dir
);
2199 if (!(open_flag
& O_CREAT
)) {
2201 if (nd
->last
.name
[nd
->last
.len
])
2202 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2203 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2205 /* we _can_ be in RCU mode here */
2206 error
= walk_component(nd
, path
, &nd
->last
, LAST_NORM
,
2209 return ERR_PTR(error
);
2210 if (error
) /* symlink */
2213 error
= complete_walk(nd
);
2215 return ERR_PTR(error
);
2218 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2219 if (!nd
->inode
->i_op
->lookup
)
2222 audit_inode(pathname
, nd
->path
.dentry
);
2226 /* create side of things */
2228 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2229 * cleared when we got to the last component we are about to look up
2231 error
= complete_walk(nd
);
2233 return ERR_PTR(error
);
2235 audit_inode(pathname
, dir
);
2237 /* trailing slashes? */
2238 if (nd
->last
.name
[nd
->last
.len
])
2241 mutex_lock(&dir
->d_inode
->i_mutex
);
2243 dentry
= lookup_hash(nd
);
2244 error
= PTR_ERR(dentry
);
2245 if (IS_ERR(dentry
)) {
2246 mutex_unlock(&dir
->d_inode
->i_mutex
);
2250 path
->dentry
= dentry
;
2251 path
->mnt
= nd
->path
.mnt
;
2253 /* Negative dentry, just create the file */
2254 if (!dentry
->d_inode
) {
2255 umode_t mode
= op
->mode
;
2256 if (!IS_POSIXACL(dir
->d_inode
))
2257 mode
&= ~current_umask();
2259 * This write is needed to ensure that a
2260 * rw->ro transition does not occur between
2261 * the time when the file is created and when
2262 * a permanent write count is taken through
2263 * the 'struct file' in nameidata_to_filp().
2265 error
= mnt_want_write(nd
->path
.mnt
);
2267 goto exit_mutex_unlock
;
2269 /* Don't check for write permission, don't truncate */
2270 open_flag
&= ~O_TRUNC
;
2272 acc_mode
= MAY_OPEN
;
2273 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2275 goto exit_mutex_unlock
;
2276 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2278 goto exit_mutex_unlock
;
2279 mutex_unlock(&dir
->d_inode
->i_mutex
);
2280 dput(nd
->path
.dentry
);
2281 nd
->path
.dentry
= dentry
;
2286 * It already exists.
2288 mutex_unlock(&dir
->d_inode
->i_mutex
);
2289 audit_inode(pathname
, path
->dentry
);
2292 if (open_flag
& O_EXCL
)
2295 error
= follow_managed(path
, nd
->flags
);
2300 nd
->flags
|= LOOKUP_JUMPED
;
2303 if (!path
->dentry
->d_inode
)
2306 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2309 path_to_nameidata(path
, nd
);
2310 nd
->inode
= path
->dentry
->d_inode
;
2311 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2312 error
= complete_walk(nd
);
2314 return ERR_PTR(error
);
2316 if (S_ISDIR(nd
->inode
->i_mode
))
2319 if (!S_ISREG(nd
->inode
->i_mode
))
2322 if (will_truncate
) {
2323 error
= mnt_want_write(nd
->path
.mnt
);
2329 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2332 filp
= nameidata_to_filp(nd
);
2333 if (!IS_ERR(filp
)) {
2334 error
= ima_file_check(filp
, op
->acc_mode
);
2337 filp
= ERR_PTR(error
);
2340 if (!IS_ERR(filp
)) {
2341 if (will_truncate
) {
2342 error
= handle_truncate(filp
);
2345 filp
= ERR_PTR(error
);
2351 mnt_drop_write(nd
->path
.mnt
);
2352 path_put(&nd
->path
);
2356 mutex_unlock(&dir
->d_inode
->i_mutex
);
2358 path_put_conditional(path
, nd
);
2360 filp
= ERR_PTR(error
);
2364 static struct file
*path_openat(int dfd
, const char *pathname
,
2365 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2367 struct file
*base
= NULL
;
2372 filp
= get_empty_filp();
2374 return ERR_PTR(-ENFILE
);
2376 filp
->f_flags
= op
->open_flag
;
2377 nd
->intent
.open
.file
= filp
;
2378 nd
->intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2379 nd
->intent
.open
.create_mode
= op
->mode
;
2381 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2382 if (unlikely(error
))
2385 current
->total_link_count
= 0;
2386 error
= link_path_walk(pathname
, nd
);
2387 if (unlikely(error
))
2390 filp
= do_last(nd
, &path
, op
, pathname
);
2391 while (unlikely(!filp
)) { /* trailing symlink */
2392 struct path link
= path
;
2394 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2395 path_put_conditional(&path
, nd
);
2396 path_put(&nd
->path
);
2397 filp
= ERR_PTR(-ELOOP
);
2400 nd
->flags
|= LOOKUP_PARENT
;
2401 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2402 error
= follow_link(&link
, nd
, &cookie
);
2403 if (unlikely(error
))
2404 filp
= ERR_PTR(error
);
2406 filp
= do_last(nd
, &path
, op
, pathname
);
2407 put_link(nd
, &link
, cookie
);
2410 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2411 path_put(&nd
->root
);
2414 release_open_intent(nd
);
2418 filp
= ERR_PTR(error
);
2422 struct file
*do_filp_open(int dfd
, const char *pathname
,
2423 const struct open_flags
*op
, int flags
)
2425 struct nameidata nd
;
2428 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2429 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2430 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2431 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2432 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2436 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2437 const char *name
, const struct open_flags
*op
, int flags
)
2439 struct nameidata nd
;
2443 nd
.root
.dentry
= dentry
;
2445 flags
|= LOOKUP_ROOT
;
2447 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2448 return ERR_PTR(-ELOOP
);
2450 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2451 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2452 file
= path_openat(-1, name
, &nd
, op
, flags
);
2453 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2454 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2458 struct dentry
*kern_path_create(int dfd
, const char *pathname
, struct path
*path
, int is_dir
)
2460 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2461 struct nameidata nd
;
2462 int error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
, &nd
);
2464 return ERR_PTR(error
);
2467 * Yucky last component or no last component at all?
2468 * (foo/., foo/.., /////)
2470 if (nd
.last_type
!= LAST_NORM
)
2472 nd
.flags
&= ~LOOKUP_PARENT
;
2473 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2474 nd
.intent
.open
.flags
= O_EXCL
;
2477 * Do the final lookup.
2479 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2480 dentry
= lookup_hash(&nd
);
2484 if (dentry
->d_inode
)
2487 * Special case - lookup gave negative, but... we had foo/bar/
2488 * From the vfs_mknod() POV we just have a negative dentry -
2489 * all is fine. Let's be bastards - you had / on the end, you've
2490 * been asking for (non-existent) directory. -ENOENT for you.
2492 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
2494 dentry
= ERR_PTR(-ENOENT
);
2501 dentry
= ERR_PTR(-EEXIST
);
2503 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2508 EXPORT_SYMBOL(kern_path_create
);
2510 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
, struct path
*path
, int is_dir
)
2512 char *tmp
= getname(pathname
);
2515 return ERR_CAST(tmp
);
2516 res
= kern_path_create(dfd
, tmp
, path
, is_dir
);
2520 EXPORT_SYMBOL(user_path_create
);
2522 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2524 int error
= may_create(dir
, dentry
);
2529 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
2532 if (!dir
->i_op
->mknod
)
2535 error
= devcgroup_inode_mknod(mode
, dev
);
2539 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2543 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2545 fsnotify_create(dir
, dentry
);
2549 static int may_mknod(umode_t mode
)
2551 switch (mode
& S_IFMT
) {
2557 case 0: /* zero mode translates to S_IFREG */
2566 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
2569 struct dentry
*dentry
;
2576 dentry
= user_path_create(dfd
, filename
, &path
, 0);
2578 return PTR_ERR(dentry
);
2580 if (!IS_POSIXACL(path
.dentry
->d_inode
))
2581 mode
&= ~current_umask();
2582 error
= may_mknod(mode
);
2585 error
= mnt_want_write(path
.mnt
);
2588 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
2590 goto out_drop_write
;
2591 switch (mode
& S_IFMT
) {
2592 case 0: case S_IFREG
:
2593 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,NULL
);
2595 case S_IFCHR
: case S_IFBLK
:
2596 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
2597 new_decode_dev(dev
));
2599 case S_IFIFO
: case S_IFSOCK
:
2600 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
2604 mnt_drop_write(path
.mnt
);
2607 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2613 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
2615 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2618 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2620 int error
= may_create(dir
, dentry
);
2621 unsigned max_links
= dir
->i_sb
->s_max_links
;
2626 if (!dir
->i_op
->mkdir
)
2629 mode
&= (S_IRWXUGO
|S_ISVTX
);
2630 error
= security_inode_mkdir(dir
, dentry
, mode
);
2634 if (max_links
&& dir
->i_nlink
>= max_links
)
2637 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2639 fsnotify_mkdir(dir
, dentry
);
2643 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
2645 struct dentry
*dentry
;
2649 dentry
= user_path_create(dfd
, pathname
, &path
, 1);
2651 return PTR_ERR(dentry
);
2653 if (!IS_POSIXACL(path
.dentry
->d_inode
))
2654 mode
&= ~current_umask();
2655 error
= mnt_want_write(path
.mnt
);
2658 error
= security_path_mkdir(&path
, dentry
, mode
);
2660 goto out_drop_write
;
2661 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
2663 mnt_drop_write(path
.mnt
);
2666 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2671 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
2673 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2677 * The dentry_unhash() helper will try to drop the dentry early: we
2678 * should have a usage count of 1 if we're the only user of this
2679 * dentry, and if that is true (possibly after pruning the dcache),
2680 * then we drop the dentry now.
2682 * A low-level filesystem can, if it choses, legally
2685 * if (!d_unhashed(dentry))
2688 * if it cannot handle the case of removing a directory
2689 * that is still in use by something else..
2691 void dentry_unhash(struct dentry
*dentry
)
2693 shrink_dcache_parent(dentry
);
2694 spin_lock(&dentry
->d_lock
);
2695 if (dentry
->d_count
== 1)
2697 spin_unlock(&dentry
->d_lock
);
2700 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2702 int error
= may_delete(dir
, dentry
, 1);
2707 if (!dir
->i_op
->rmdir
)
2711 mutex_lock(&dentry
->d_inode
->i_mutex
);
2714 if (d_mountpoint(dentry
))
2717 error
= security_inode_rmdir(dir
, dentry
);
2721 shrink_dcache_parent(dentry
);
2722 error
= dir
->i_op
->rmdir(dir
, dentry
);
2726 dentry
->d_inode
->i_flags
|= S_DEAD
;
2730 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2737 static long do_rmdir(int dfd
, const char __user
*pathname
)
2741 struct dentry
*dentry
;
2742 struct nameidata nd
;
2744 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2748 switch(nd
.last_type
) {
2760 nd
.flags
&= ~LOOKUP_PARENT
;
2762 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2763 dentry
= lookup_hash(&nd
);
2764 error
= PTR_ERR(dentry
);
2767 if (!dentry
->d_inode
) {
2771 error
= mnt_want_write(nd
.path
.mnt
);
2774 error
= security_path_rmdir(&nd
.path
, dentry
);
2777 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2779 mnt_drop_write(nd
.path
.mnt
);
2783 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2790 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2792 return do_rmdir(AT_FDCWD
, pathname
);
2795 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2797 int error
= may_delete(dir
, dentry
, 0);
2802 if (!dir
->i_op
->unlink
)
2805 mutex_lock(&dentry
->d_inode
->i_mutex
);
2806 if (d_mountpoint(dentry
))
2809 error
= security_inode_unlink(dir
, dentry
);
2811 error
= dir
->i_op
->unlink(dir
, dentry
);
2816 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2818 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2819 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2820 fsnotify_link_count(dentry
->d_inode
);
2828 * Make sure that the actual truncation of the file will occur outside its
2829 * directory's i_mutex. Truncate can take a long time if there is a lot of
2830 * writeout happening, and we don't want to prevent access to the directory
2831 * while waiting on the I/O.
2833 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2837 struct dentry
*dentry
;
2838 struct nameidata nd
;
2839 struct inode
*inode
= NULL
;
2841 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2846 if (nd
.last_type
!= LAST_NORM
)
2849 nd
.flags
&= ~LOOKUP_PARENT
;
2851 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2852 dentry
= lookup_hash(&nd
);
2853 error
= PTR_ERR(dentry
);
2854 if (!IS_ERR(dentry
)) {
2855 /* Why not before? Because we want correct error value */
2856 if (nd
.last
.name
[nd
.last
.len
])
2858 inode
= dentry
->d_inode
;
2862 error
= mnt_want_write(nd
.path
.mnt
);
2865 error
= security_path_unlink(&nd
.path
, dentry
);
2868 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2870 mnt_drop_write(nd
.path
.mnt
);
2874 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2876 iput(inode
); /* truncate the inode here */
2883 error
= !dentry
->d_inode
? -ENOENT
:
2884 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2888 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2890 if ((flag
& ~AT_REMOVEDIR
) != 0)
2893 if (flag
& AT_REMOVEDIR
)
2894 return do_rmdir(dfd
, pathname
);
2896 return do_unlinkat(dfd
, pathname
);
2899 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2901 return do_unlinkat(AT_FDCWD
, pathname
);
2904 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2906 int error
= may_create(dir
, dentry
);
2911 if (!dir
->i_op
->symlink
)
2914 error
= security_inode_symlink(dir
, dentry
, oldname
);
2918 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2920 fsnotify_create(dir
, dentry
);
2924 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2925 int, newdfd
, const char __user
*, newname
)
2929 struct dentry
*dentry
;
2932 from
= getname(oldname
);
2934 return PTR_ERR(from
);
2936 dentry
= user_path_create(newdfd
, newname
, &path
, 0);
2937 error
= PTR_ERR(dentry
);
2941 error
= mnt_want_write(path
.mnt
);
2944 error
= security_path_symlink(&path
, dentry
, from
);
2946 goto out_drop_write
;
2947 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
);
2949 mnt_drop_write(path
.mnt
);
2952 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2959 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2961 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2964 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2966 struct inode
*inode
= old_dentry
->d_inode
;
2967 unsigned max_links
= dir
->i_sb
->s_max_links
;
2973 error
= may_create(dir
, new_dentry
);
2977 if (dir
->i_sb
!= inode
->i_sb
)
2981 * A link to an append-only or immutable file cannot be created.
2983 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2985 if (!dir
->i_op
->link
)
2987 if (S_ISDIR(inode
->i_mode
))
2990 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2994 mutex_lock(&inode
->i_mutex
);
2995 /* Make sure we don't allow creating hardlink to an unlinked file */
2996 if (inode
->i_nlink
== 0)
2998 else if (max_links
&& inode
->i_nlink
>= max_links
)
3001 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3002 mutex_unlock(&inode
->i_mutex
);
3004 fsnotify_link(dir
, inode
, new_dentry
);
3009 * Hardlinks are often used in delicate situations. We avoid
3010 * security-related surprises by not following symlinks on the
3013 * We don't follow them on the oldname either to be compatible
3014 * with linux 2.0, and to avoid hard-linking to directories
3015 * and other special files. --ADM
3017 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3018 int, newdfd
, const char __user
*, newname
, int, flags
)
3020 struct dentry
*new_dentry
;
3021 struct path old_path
, new_path
;
3025 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3028 * To use null names we require CAP_DAC_READ_SEARCH
3029 * This ensures that not everyone will be able to create
3030 * handlink using the passed filedescriptor.
3032 if (flags
& AT_EMPTY_PATH
) {
3033 if (!capable(CAP_DAC_READ_SEARCH
))
3038 if (flags
& AT_SYMLINK_FOLLOW
)
3039 how
|= LOOKUP_FOLLOW
;
3041 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3045 new_dentry
= user_path_create(newdfd
, newname
, &new_path
, 0);
3046 error
= PTR_ERR(new_dentry
);
3047 if (IS_ERR(new_dentry
))
3051 if (old_path
.mnt
!= new_path
.mnt
)
3053 error
= mnt_want_write(new_path
.mnt
);
3056 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3058 goto out_drop_write
;
3059 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3061 mnt_drop_write(new_path
.mnt
);
3064 mutex_unlock(&new_path
.dentry
->d_inode
->i_mutex
);
3065 path_put(&new_path
);
3067 path_put(&old_path
);
3072 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3074 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3078 * The worst of all namespace operations - renaming directory. "Perverted"
3079 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3081 * a) we can get into loop creation. Check is done in is_subdir().
3082 * b) race potential - two innocent renames can create a loop together.
3083 * That's where 4.4 screws up. Current fix: serialization on
3084 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3086 * c) we have to lock _three_ objects - parents and victim (if it exists).
3087 * And that - after we got ->i_mutex on parents (until then we don't know
3088 * whether the target exists). Solution: try to be smart with locking
3089 * order for inodes. We rely on the fact that tree topology may change
3090 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3091 * move will be locked. Thus we can rank directories by the tree
3092 * (ancestors first) and rank all non-directories after them.
3093 * That works since everybody except rename does "lock parent, lookup,
3094 * lock child" and rename is under ->s_vfs_rename_mutex.
3095 * HOWEVER, it relies on the assumption that any object with ->lookup()
3096 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3097 * we'd better make sure that there's no link(2) for them.
3098 * d) conversion from fhandle to dentry may come in the wrong moment - when
3099 * we are removing the target. Solution: we will have to grab ->i_mutex
3100 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3101 * ->i_mutex on parents, which works but leads to some truly excessive
3104 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3105 struct inode
*new_dir
, struct dentry
*new_dentry
)
3108 struct inode
*target
= new_dentry
->d_inode
;
3109 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3112 * If we are going to change the parent - check write permissions,
3113 * we'll need to flip '..'.
3115 if (new_dir
!= old_dir
) {
3116 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3121 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3127 mutex_lock(&target
->i_mutex
);
3130 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3134 if (max_links
&& !target
&& new_dir
!= old_dir
&&
3135 new_dir
->i_nlink
>= max_links
)
3139 shrink_dcache_parent(new_dentry
);
3140 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3145 target
->i_flags
|= S_DEAD
;
3146 dont_mount(new_dentry
);
3150 mutex_unlock(&target
->i_mutex
);
3153 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3154 d_move(old_dentry
,new_dentry
);
3158 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3159 struct inode
*new_dir
, struct dentry
*new_dentry
)
3161 struct inode
*target
= new_dentry
->d_inode
;
3164 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3170 mutex_lock(&target
->i_mutex
);
3173 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3176 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3181 dont_mount(new_dentry
);
3182 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3183 d_move(old_dentry
, new_dentry
);
3186 mutex_unlock(&target
->i_mutex
);
3191 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3192 struct inode
*new_dir
, struct dentry
*new_dentry
)
3195 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3196 const unsigned char *old_name
;
3198 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3201 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3205 if (!new_dentry
->d_inode
)
3206 error
= may_create(new_dir
, new_dentry
);
3208 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3212 if (!old_dir
->i_op
->rename
)
3215 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3218 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3220 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3222 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3223 new_dentry
->d_inode
, old_dentry
);
3224 fsnotify_oldname_free(old_name
);
3229 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3230 int, newdfd
, const char __user
*, newname
)
3232 struct dentry
*old_dir
, *new_dir
;
3233 struct dentry
*old_dentry
, *new_dentry
;
3234 struct dentry
*trap
;
3235 struct nameidata oldnd
, newnd
;
3240 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3244 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3249 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3252 old_dir
= oldnd
.path
.dentry
;
3254 if (oldnd
.last_type
!= LAST_NORM
)
3257 new_dir
= newnd
.path
.dentry
;
3258 if (newnd
.last_type
!= LAST_NORM
)
3261 oldnd
.flags
&= ~LOOKUP_PARENT
;
3262 newnd
.flags
&= ~LOOKUP_PARENT
;
3263 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3265 trap
= lock_rename(new_dir
, old_dir
);
3267 old_dentry
= lookup_hash(&oldnd
);
3268 error
= PTR_ERR(old_dentry
);
3269 if (IS_ERR(old_dentry
))
3271 /* source must exist */
3273 if (!old_dentry
->d_inode
)
3275 /* unless the source is a directory trailing slashes give -ENOTDIR */
3276 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3278 if (oldnd
.last
.name
[oldnd
.last
.len
])
3280 if (newnd
.last
.name
[newnd
.last
.len
])
3283 /* source should not be ancestor of target */
3285 if (old_dentry
== trap
)
3287 new_dentry
= lookup_hash(&newnd
);
3288 error
= PTR_ERR(new_dentry
);
3289 if (IS_ERR(new_dentry
))
3291 /* target should not be an ancestor of source */
3293 if (new_dentry
== trap
)
3296 error
= mnt_want_write(oldnd
.path
.mnt
);
3299 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3300 &newnd
.path
, new_dentry
);
3303 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3304 new_dir
->d_inode
, new_dentry
);
3306 mnt_drop_write(oldnd
.path
.mnt
);
3312 unlock_rename(new_dir
, old_dir
);
3314 path_put(&newnd
.path
);
3317 path_put(&oldnd
.path
);
3323 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3325 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3328 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3332 len
= PTR_ERR(link
);
3337 if (len
> (unsigned) buflen
)
3339 if (copy_to_user(buffer
, link
, len
))
3346 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3347 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3348 * using) it for any given inode is up to filesystem.
3350 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3352 struct nameidata nd
;
3357 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3359 return PTR_ERR(cookie
);
3361 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3362 if (dentry
->d_inode
->i_op
->put_link
)
3363 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3367 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3369 return __vfs_follow_link(nd
, link
);
3372 /* get the link contents into pagecache */
3373 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3377 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3378 page
= read_mapping_page(mapping
, 0, NULL
);
3383 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3387 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3389 struct page
*page
= NULL
;
3390 char *s
= page_getlink(dentry
, &page
);
3391 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3394 page_cache_release(page
);
3399 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3401 struct page
*page
= NULL
;
3402 nd_set_link(nd
, page_getlink(dentry
, &page
));
3406 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3408 struct page
*page
= cookie
;
3412 page_cache_release(page
);
3417 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3419 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3421 struct address_space
*mapping
= inode
->i_mapping
;
3426 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3428 flags
|= AOP_FLAG_NOFS
;
3431 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3432 flags
, &page
, &fsdata
);
3436 kaddr
= kmap_atomic(page
);
3437 memcpy(kaddr
, symname
, len
-1);
3438 kunmap_atomic(kaddr
);
3440 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3447 mark_inode_dirty(inode
);
3453 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3455 return __page_symlink(inode
, symname
, len
,
3456 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3459 const struct inode_operations page_symlink_inode_operations
= {
3460 .readlink
= generic_readlink
,
3461 .follow_link
= page_follow_link_light
,
3462 .put_link
= page_put_link
,
3465 EXPORT_SYMBOL(user_path_at
);
3466 EXPORT_SYMBOL(follow_down_one
);
3467 EXPORT_SYMBOL(follow_down
);
3468 EXPORT_SYMBOL(follow_up
);
3469 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3470 EXPORT_SYMBOL(getname
);
3471 EXPORT_SYMBOL(lock_rename
);
3472 EXPORT_SYMBOL(lookup_one_len
);
3473 EXPORT_SYMBOL(page_follow_link_light
);
3474 EXPORT_SYMBOL(page_put_link
);
3475 EXPORT_SYMBOL(page_readlink
);
3476 EXPORT_SYMBOL(__page_symlink
);
3477 EXPORT_SYMBOL(page_symlink
);
3478 EXPORT_SYMBOL(page_symlink_inode_operations
);
3479 EXPORT_SYMBOL(kern_path
);
3480 EXPORT_SYMBOL(vfs_path_lookup
);
3481 EXPORT_SYMBOL(inode_permission
);
3482 EXPORT_SYMBOL(unlock_rename
);
3483 EXPORT_SYMBOL(vfs_create
);
3484 EXPORT_SYMBOL(vfs_follow_link
);
3485 EXPORT_SYMBOL(vfs_link
);
3486 EXPORT_SYMBOL(vfs_mkdir
);
3487 EXPORT_SYMBOL(vfs_mknod
);
3488 EXPORT_SYMBOL(generic_permission
);
3489 EXPORT_SYMBOL(vfs_readlink
);
3490 EXPORT_SYMBOL(vfs_rename
);
3491 EXPORT_SYMBOL(vfs_rmdir
);
3492 EXPORT_SYMBOL(vfs_symlink
);
3493 EXPORT_SYMBOL(vfs_unlink
);
3494 EXPORT_SYMBOL(dentry_unhash
);
3495 EXPORT_SYMBOL(generic_readlink
);