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/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <asm/uaccess.h>
42 /* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
90 * [10-Sep-98 Alan Modra] Another symlink change.
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 static char *getname_flags(const char __user
*filename
, int flags
, int *empty
)
122 char *result
= __getname(), *err
;
125 if (unlikely(!result
))
126 return ERR_PTR(-ENOMEM
);
128 len
= strncpy_from_user(result
, filename
, PATH_MAX
);
130 if (unlikely(len
< 0))
133 /* The empty path is special. */
134 if (unlikely(!len
)) {
137 err
= ERR_PTR(-ENOENT
);
138 if (!(flags
& LOOKUP_EMPTY
))
142 err
= ERR_PTR(-ENAMETOOLONG
);
143 if (likely(len
< PATH_MAX
)) {
144 audit_getname(result
);
153 char *getname(const char __user
* filename
)
155 return getname_flags(filename
, 0, NULL
);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name
)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname
);
169 static int check_acl(struct inode
*inode
, int mask
)
171 #ifdef CONFIG_FS_POSIX_ACL
172 struct posix_acl
*acl
;
174 if (mask
& MAY_NOT_BLOCK
) {
175 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
178 /* no ->get_acl() calls in RCU mode... */
179 if (acl
== ACL_NOT_CACHED
)
181 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
184 acl
= get_cached_acl(inode
, ACL_TYPE_ACCESS
);
187 * A filesystem can force a ACL callback by just never filling the
188 * ACL cache. But normally you'd fill the cache either at inode
189 * instantiation time, or on the first ->get_acl call.
191 * If the filesystem doesn't have a get_acl() function at all, we'll
192 * just create the negative cache entry.
194 if (acl
== ACL_NOT_CACHED
) {
195 if (inode
->i_op
->get_acl
) {
196 acl
= inode
->i_op
->get_acl(inode
, ACL_TYPE_ACCESS
);
200 set_cached_acl(inode
, ACL_TYPE_ACCESS
, NULL
);
206 int error
= posix_acl_permission(inode
, acl
, mask
);
207 posix_acl_release(acl
);
216 * This does the basic permission checking
218 static int acl_permission_check(struct inode
*inode
, int mask
)
220 unsigned int mode
= inode
->i_mode
;
222 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
225 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
226 int error
= check_acl(inode
, mask
);
227 if (error
!= -EAGAIN
)
231 if (in_group_p(inode
->i_gid
))
236 * If the DACs are ok we don't need any capability check.
238 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
244 * generic_permission - check for access rights on a Posix-like filesystem
245 * @inode: inode to check access rights for
246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
248 * Used to check for read/write/execute permissions on a file.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
251 * are used for other things.
253 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
254 * request cannot be satisfied (eg. requires blocking or too much complexity).
255 * It would then be called again in ref-walk mode.
257 int generic_permission(struct inode
*inode
, int mask
)
262 * Do the basic permission checks.
264 ret
= acl_permission_check(inode
, mask
);
268 if (S_ISDIR(inode
->i_mode
)) {
269 /* DACs are overridable for directories */
270 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
272 if (!(mask
& MAY_WRITE
))
273 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
278 * Read/write DACs are always overridable.
279 * Executable DACs are overridable when there is
280 * at least one exec bit set.
282 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
283 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
287 * Searching includes executable on directories, else just read.
289 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
290 if (mask
== MAY_READ
)
291 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
298 * We _really_ want to just do "generic_permission()" without
299 * even looking at the inode->i_op values. So we keep a cache
300 * flag in inode->i_opflags, that says "this has not special
301 * permission function, use the fast case".
303 static inline int do_inode_permission(struct inode
*inode
, int mask
)
305 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
306 if (likely(inode
->i_op
->permission
))
307 return inode
->i_op
->permission(inode
, mask
);
309 /* This gets set once for the inode lifetime */
310 spin_lock(&inode
->i_lock
);
311 inode
->i_opflags
|= IOP_FASTPERM
;
312 spin_unlock(&inode
->i_lock
);
314 return generic_permission(inode
, mask
);
318 * inode_permission - check for access rights to a given inode
319 * @inode: inode to check permission on
320 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
322 * Used to check for read/write/execute permissions on an inode.
323 * We use "fsuid" for this, letting us set arbitrary permissions
324 * for filesystem access without changing the "normal" uids which
325 * are used for other things.
327 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
329 int inode_permission(struct inode
*inode
, int mask
)
333 if (unlikely(mask
& MAY_WRITE
)) {
334 umode_t mode
= inode
->i_mode
;
337 * Nobody gets write access to a read-only fs.
339 if (IS_RDONLY(inode
) &&
340 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
344 * Nobody gets write access to an immutable file.
346 if (IS_IMMUTABLE(inode
))
350 retval
= do_inode_permission(inode
, mask
);
354 retval
= devcgroup_inode_permission(inode
, mask
);
358 return security_inode_permission(inode
, mask
);
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 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). In situations when we can't
390 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
391 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
392 * mode. Refcounts are grabbed at the last known good point before rcu-walk
393 * got stuck, so ref-walk may continue from there. If this is not successful
394 * (eg. a seqcount has changed), then failure is returned and it's up to caller
395 * to restart the path walk from the beginning in ref-walk mode.
399 * unlazy_walk - try to switch to ref-walk mode.
400 * @nd: nameidata pathwalk data
401 * @dentry: child of nd->path.dentry or NULL
402 * Returns: 0 on success, -ECHILD on failure
404 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
405 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
406 * @nd or NULL. Must be called from rcu-walk context.
408 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
410 struct fs_struct
*fs
= current
->fs
;
411 struct dentry
*parent
= nd
->path
.dentry
;
414 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
415 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
417 spin_lock(&fs
->lock
);
418 if (nd
->root
.mnt
!= fs
->root
.mnt
||
419 nd
->root
.dentry
!= fs
->root
.dentry
)
422 spin_lock(&parent
->d_lock
);
424 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
426 BUG_ON(nd
->inode
!= parent
->d_inode
);
428 if (dentry
->d_parent
!= parent
)
430 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
431 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
434 * If the sequence check on the child dentry passed, then
435 * the child has not been removed from its parent. This
436 * means the parent dentry must be valid and able to take
437 * a reference at this point.
439 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
440 BUG_ON(!parent
->d_count
);
442 spin_unlock(&dentry
->d_lock
);
444 spin_unlock(&parent
->d_lock
);
447 spin_unlock(&fs
->lock
);
449 mntget(nd
->path
.mnt
);
452 br_read_unlock(vfsmount_lock
);
453 nd
->flags
&= ~LOOKUP_RCU
;
457 spin_unlock(&dentry
->d_lock
);
459 spin_unlock(&parent
->d_lock
);
462 spin_unlock(&fs
->lock
);
467 * release_open_intent - free up open intent resources
468 * @nd: pointer to nameidata
470 void release_open_intent(struct nameidata
*nd
)
472 struct file
*file
= nd
->intent
.open
.file
;
474 if (file
&& !IS_ERR(file
)) {
475 if (file
->f_path
.dentry
== NULL
)
482 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
484 return dentry
->d_op
->d_revalidate(dentry
, nd
);
488 * complete_walk - successful completion of path walk
489 * @nd: pointer nameidata
491 * If we had been in RCU mode, drop out of it and legitimize nd->path.
492 * Revalidate the final result, unless we'd already done that during
493 * the path walk or the filesystem doesn't ask for it. Return 0 on
494 * success, -error on failure. In case of failure caller does not
495 * need to drop nd->path.
497 static int complete_walk(struct nameidata
*nd
)
499 struct dentry
*dentry
= nd
->path
.dentry
;
502 if (nd
->flags
& LOOKUP_RCU
) {
503 nd
->flags
&= ~LOOKUP_RCU
;
504 if (!(nd
->flags
& LOOKUP_ROOT
))
506 spin_lock(&dentry
->d_lock
);
507 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
508 spin_unlock(&dentry
->d_lock
);
510 br_read_unlock(vfsmount_lock
);
513 BUG_ON(nd
->inode
!= dentry
->d_inode
);
514 spin_unlock(&dentry
->d_lock
);
515 mntget(nd
->path
.mnt
);
517 br_read_unlock(vfsmount_lock
);
520 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
523 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
526 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
529 /* Note: we do not d_invalidate() */
530 status
= d_revalidate(dentry
, nd
);
541 static __always_inline
void set_root(struct nameidata
*nd
)
544 get_fs_root(current
->fs
, &nd
->root
);
547 static int link_path_walk(const char *, struct nameidata
*);
549 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
552 struct fs_struct
*fs
= current
->fs
;
556 seq
= read_seqcount_begin(&fs
->seq
);
558 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
559 } while (read_seqcount_retry(&fs
->seq
, seq
));
563 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
575 nd
->flags
|= LOOKUP_JUMPED
;
577 nd
->inode
= nd
->path
.dentry
->d_inode
;
579 ret
= link_path_walk(link
, nd
);
583 return PTR_ERR(link
);
586 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
589 if (path
->mnt
!= nd
->path
.mnt
)
593 static inline void path_to_nameidata(const struct path
*path
,
594 struct nameidata
*nd
)
596 if (!(nd
->flags
& LOOKUP_RCU
)) {
597 dput(nd
->path
.dentry
);
598 if (nd
->path
.mnt
!= path
->mnt
)
599 mntput(nd
->path
.mnt
);
601 nd
->path
.mnt
= path
->mnt
;
602 nd
->path
.dentry
= path
->dentry
;
605 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
607 struct inode
*inode
= link
->dentry
->d_inode
;
608 if (!IS_ERR(cookie
) && inode
->i_op
->put_link
)
609 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
613 static __always_inline
int
614 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
617 struct dentry
*dentry
= link
->dentry
;
619 BUG_ON(nd
->flags
& LOOKUP_RCU
);
621 if (link
->mnt
== nd
->path
.mnt
)
624 if (unlikely(current
->total_link_count
>= 40)) {
625 *p
= ERR_PTR(-ELOOP
); /* no ->put_link(), please */
630 current
->total_link_count
++;
633 nd_set_link(nd
, NULL
);
635 error
= security_inode_follow_link(link
->dentry
, nd
);
637 *p
= ERR_PTR(error
); /* no ->put_link(), please */
642 nd
->last_type
= LAST_BIND
;
643 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
646 char *s
= nd_get_link(nd
);
649 error
= __vfs_follow_link(nd
, s
);
650 else if (nd
->last_type
== LAST_BIND
) {
651 nd
->flags
|= LOOKUP_JUMPED
;
652 nd
->inode
= nd
->path
.dentry
->d_inode
;
653 if (nd
->inode
->i_op
->follow_link
) {
654 /* stepped on a _really_ weird one */
663 static int follow_up_rcu(struct path
*path
)
665 struct mount
*mnt
= real_mount(path
->mnt
);
666 struct mount
*parent
;
667 struct dentry
*mountpoint
;
669 parent
= mnt
->mnt_parent
;
670 if (&parent
->mnt
== path
->mnt
)
672 mountpoint
= mnt
->mnt_mountpoint
;
673 path
->dentry
= mountpoint
;
674 path
->mnt
= &parent
->mnt
;
678 int follow_up(struct path
*path
)
680 struct mount
*mnt
= real_mount(path
->mnt
);
681 struct mount
*parent
;
682 struct dentry
*mountpoint
;
684 br_read_lock(vfsmount_lock
);
685 parent
= mnt
->mnt_parent
;
686 if (&parent
->mnt
== path
->mnt
) {
687 br_read_unlock(vfsmount_lock
);
690 mntget(&parent
->mnt
);
691 mountpoint
= dget(mnt
->mnt_mountpoint
);
692 br_read_unlock(vfsmount_lock
);
694 path
->dentry
= mountpoint
;
696 path
->mnt
= &parent
->mnt
;
701 * Perform an automount
702 * - return -EISDIR to tell follow_managed() to stop and return the path we
705 static int follow_automount(struct path
*path
, unsigned flags
,
708 struct vfsmount
*mnt
;
711 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
714 /* We don't want to mount if someone's just doing a stat -
715 * unless they're stat'ing a directory and appended a '/' to
718 * We do, however, want to mount if someone wants to open or
719 * create a file of any type under the mountpoint, wants to
720 * traverse through the mountpoint or wants to open the
721 * mounted directory. Also, autofs may mark negative dentries
722 * as being automount points. These will need the attentions
723 * of the daemon to instantiate them before they can be used.
725 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
726 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
727 path
->dentry
->d_inode
)
730 current
->total_link_count
++;
731 if (current
->total_link_count
>= 40)
734 mnt
= path
->dentry
->d_op
->d_automount(path
);
737 * The filesystem is allowed to return -EISDIR here to indicate
738 * it doesn't want to automount. For instance, autofs would do
739 * this so that its userspace daemon can mount on this dentry.
741 * However, we can only permit this if it's a terminal point in
742 * the path being looked up; if it wasn't then the remainder of
743 * the path is inaccessible and we should say so.
745 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
750 if (!mnt
) /* mount collision */
754 /* lock_mount() may release path->mnt on error */
758 err
= finish_automount(mnt
, path
);
762 /* Someone else made a mount here whilst we were busy */
767 path
->dentry
= dget(mnt
->mnt_root
);
776 * Handle a dentry that is managed in some way.
777 * - Flagged for transit management (autofs)
778 * - Flagged as mountpoint
779 * - Flagged as automount point
781 * This may only be called in refwalk mode.
783 * Serialization is taken care of in namespace.c
785 static int follow_managed(struct path
*path
, unsigned flags
)
787 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
789 bool need_mntput
= false;
792 /* Given that we're not holding a lock here, we retain the value in a
793 * local variable for each dentry as we look at it so that we don't see
794 * the components of that value change under us */
795 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
796 managed
&= DCACHE_MANAGED_DENTRY
,
797 unlikely(managed
!= 0)) {
798 /* Allow the filesystem to manage the transit without i_mutex
800 if (managed
& DCACHE_MANAGE_TRANSIT
) {
801 BUG_ON(!path
->dentry
->d_op
);
802 BUG_ON(!path
->dentry
->d_op
->d_manage
);
803 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
808 /* Transit to a mounted filesystem. */
809 if (managed
& DCACHE_MOUNTED
) {
810 struct vfsmount
*mounted
= lookup_mnt(path
);
816 path
->dentry
= dget(mounted
->mnt_root
);
821 /* Something is mounted on this dentry in another
822 * namespace and/or whatever was mounted there in this
823 * namespace got unmounted before we managed to get the
827 /* Handle an automount point */
828 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
829 ret
= follow_automount(path
, flags
, &need_mntput
);
835 /* We didn't change the current path point */
839 if (need_mntput
&& path
->mnt
== mnt
)
843 return ret
< 0 ? ret
: need_mntput
;
846 int follow_down_one(struct path
*path
)
848 struct vfsmount
*mounted
;
850 mounted
= lookup_mnt(path
);
855 path
->dentry
= dget(mounted
->mnt_root
);
861 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
863 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
864 dentry
->d_op
->d_manage(dentry
, true) < 0);
868 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
869 * we meet a managed dentry that would need blocking.
871 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
872 struct inode
**inode
)
875 struct mount
*mounted
;
877 * Don't forget we might have a non-mountpoint managed dentry
878 * that wants to block transit.
880 if (unlikely(managed_dentry_might_block(path
->dentry
)))
883 if (!d_mountpoint(path
->dentry
))
886 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
889 path
->mnt
= &mounted
->mnt
;
890 path
->dentry
= mounted
->mnt
.mnt_root
;
891 nd
->flags
|= LOOKUP_JUMPED
;
892 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
894 * Update the inode too. We don't need to re-check the
895 * dentry sequence number here after this d_inode read,
896 * because a mount-point is always pinned.
898 *inode
= path
->dentry
->d_inode
;
903 static void follow_mount_rcu(struct nameidata
*nd
)
905 while (d_mountpoint(nd
->path
.dentry
)) {
906 struct mount
*mounted
;
907 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
910 nd
->path
.mnt
= &mounted
->mnt
;
911 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
912 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
916 static int follow_dotdot_rcu(struct nameidata
*nd
)
921 if (nd
->path
.dentry
== nd
->root
.dentry
&&
922 nd
->path
.mnt
== nd
->root
.mnt
) {
925 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
926 struct dentry
*old
= nd
->path
.dentry
;
927 struct dentry
*parent
= old
->d_parent
;
930 seq
= read_seqcount_begin(&parent
->d_seq
);
931 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
933 nd
->path
.dentry
= parent
;
937 if (!follow_up_rcu(&nd
->path
))
939 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
941 follow_mount_rcu(nd
);
942 nd
->inode
= nd
->path
.dentry
->d_inode
;
946 nd
->flags
&= ~LOOKUP_RCU
;
947 if (!(nd
->flags
& LOOKUP_ROOT
))
950 br_read_unlock(vfsmount_lock
);
955 * Follow down to the covering mount currently visible to userspace. At each
956 * point, the filesystem owning that dentry may be queried as to whether the
957 * caller is permitted to proceed or not.
959 int follow_down(struct path
*path
)
964 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
965 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
966 /* Allow the filesystem to manage the transit without i_mutex
969 * We indicate to the filesystem if someone is trying to mount
970 * something here. This gives autofs the chance to deny anyone
971 * other than its daemon the right to mount on its
974 * The filesystem may sleep at this point.
976 if (managed
& DCACHE_MANAGE_TRANSIT
) {
977 BUG_ON(!path
->dentry
->d_op
);
978 BUG_ON(!path
->dentry
->d_op
->d_manage
);
979 ret
= path
->dentry
->d_op
->d_manage(
980 path
->dentry
, false);
982 return ret
== -EISDIR
? 0 : ret
;
985 /* Transit to a mounted filesystem. */
986 if (managed
& DCACHE_MOUNTED
) {
987 struct vfsmount
*mounted
= lookup_mnt(path
);
993 path
->dentry
= dget(mounted
->mnt_root
);
997 /* Don't handle automount points here */
1004 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1006 static void follow_mount(struct path
*path
)
1008 while (d_mountpoint(path
->dentry
)) {
1009 struct vfsmount
*mounted
= lookup_mnt(path
);
1014 path
->mnt
= mounted
;
1015 path
->dentry
= dget(mounted
->mnt_root
);
1019 static void follow_dotdot(struct nameidata
*nd
)
1024 struct dentry
*old
= nd
->path
.dentry
;
1026 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1027 nd
->path
.mnt
== nd
->root
.mnt
) {
1030 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1031 /* rare case of legitimate dget_parent()... */
1032 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1036 if (!follow_up(&nd
->path
))
1039 follow_mount(&nd
->path
);
1040 nd
->inode
= nd
->path
.dentry
->d_inode
;
1044 * This looks up the name in dcache, possibly revalidates the old dentry and
1045 * allocates a new one if not found or not valid. In the need_lookup argument
1046 * returns whether i_op->lookup is necessary.
1048 * dir->d_inode->i_mutex must be held
1050 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1051 struct nameidata
*nd
, bool *need_lookup
)
1053 struct dentry
*dentry
;
1056 *need_lookup
= false;
1057 dentry
= d_lookup(dir
, name
);
1059 if (d_need_lookup(dentry
)) {
1060 *need_lookup
= true;
1061 } else if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1062 error
= d_revalidate(dentry
, nd
);
1063 if (unlikely(error
<= 0)) {
1066 return ERR_PTR(error
);
1067 } else if (!d_invalidate(dentry
)) {
1076 dentry
= d_alloc(dir
, name
);
1077 if (unlikely(!dentry
))
1078 return ERR_PTR(-ENOMEM
);
1080 *need_lookup
= true;
1086 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1087 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1089 * dir->d_inode->i_mutex must be held
1091 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1092 struct nameidata
*nd
)
1096 /* Don't create child dentry for a dead directory. */
1097 if (unlikely(IS_DEADDIR(dir
))) {
1099 return ERR_PTR(-ENOENT
);
1102 old
= dir
->i_op
->lookup(dir
, dentry
, nd
);
1103 if (unlikely(old
)) {
1110 static struct dentry
*__lookup_hash(struct qstr
*name
,
1111 struct dentry
*base
, struct nameidata
*nd
)
1114 struct dentry
*dentry
;
1116 dentry
= lookup_dcache(name
, base
, nd
, &need_lookup
);
1120 return lookup_real(base
->d_inode
, dentry
, nd
);
1124 * It's more convoluted than I'd like it to be, but... it's still fairly
1125 * small and for now I'd prefer to have fast path as straight as possible.
1126 * It _is_ time-critical.
1128 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1129 struct path
*path
, struct inode
**inode
)
1131 struct vfsmount
*mnt
= nd
->path
.mnt
;
1132 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1138 * Rename seqlock is not required here because in the off chance
1139 * of a false negative due to a concurrent rename, we're going to
1140 * do the non-racy lookup, below.
1142 if (nd
->flags
& LOOKUP_RCU
) {
1144 dentry
= __d_lookup_rcu(parent
, name
, &seq
, nd
->inode
);
1149 * This sequence count validates that the inode matches
1150 * the dentry name information from lookup.
1152 *inode
= dentry
->d_inode
;
1153 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1157 * This sequence count validates that the parent had no
1158 * changes while we did the lookup of the dentry above.
1160 * The memory barrier in read_seqcount_begin of child is
1161 * enough, we can use __read_seqcount_retry here.
1163 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1167 if (unlikely(d_need_lookup(dentry
)))
1169 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1170 status
= d_revalidate(dentry
, nd
);
1171 if (unlikely(status
<= 0)) {
1172 if (status
!= -ECHILD
)
1178 path
->dentry
= dentry
;
1179 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1181 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1185 if (unlazy_walk(nd
, dentry
))
1188 dentry
= __d_lookup(parent
, name
);
1191 if (unlikely(!dentry
))
1194 if (unlikely(d_need_lookup(dentry
))) {
1199 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1200 status
= d_revalidate(dentry
, nd
);
1201 if (unlikely(status
<= 0)) {
1206 if (!d_invalidate(dentry
)) {
1213 path
->dentry
= dentry
;
1214 err
= follow_managed(path
, nd
->flags
);
1215 if (unlikely(err
< 0)) {
1216 path_put_conditional(path
, nd
);
1220 nd
->flags
|= LOOKUP_JUMPED
;
1221 *inode
= path
->dentry
->d_inode
;
1225 BUG_ON(nd
->inode
!= parent
->d_inode
);
1227 mutex_lock(&parent
->d_inode
->i_mutex
);
1228 dentry
= __lookup_hash(name
, parent
, nd
);
1229 mutex_unlock(&parent
->d_inode
->i_mutex
);
1231 return PTR_ERR(dentry
);
1235 static inline int may_lookup(struct nameidata
*nd
)
1237 if (nd
->flags
& LOOKUP_RCU
) {
1238 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1241 if (unlazy_walk(nd
, NULL
))
1244 return inode_permission(nd
->inode
, MAY_EXEC
);
1247 static inline int handle_dots(struct nameidata
*nd
, int type
)
1249 if (type
== LAST_DOTDOT
) {
1250 if (nd
->flags
& LOOKUP_RCU
) {
1251 if (follow_dotdot_rcu(nd
))
1259 static void terminate_walk(struct nameidata
*nd
)
1261 if (!(nd
->flags
& LOOKUP_RCU
)) {
1262 path_put(&nd
->path
);
1264 nd
->flags
&= ~LOOKUP_RCU
;
1265 if (!(nd
->flags
& LOOKUP_ROOT
))
1266 nd
->root
.mnt
= NULL
;
1268 br_read_unlock(vfsmount_lock
);
1273 * Do we need to follow links? We _really_ want to be able
1274 * to do this check without having to look at inode->i_op,
1275 * so we keep a cache of "no, this doesn't need follow_link"
1276 * for the common case.
1278 static inline int should_follow_link(struct inode
*inode
, int follow
)
1280 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1281 if (likely(inode
->i_op
->follow_link
))
1284 /* This gets set once for the inode lifetime */
1285 spin_lock(&inode
->i_lock
);
1286 inode
->i_opflags
|= IOP_NOFOLLOW
;
1287 spin_unlock(&inode
->i_lock
);
1292 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1293 struct qstr
*name
, int type
, int follow
)
1295 struct inode
*inode
;
1298 * "." and ".." are special - ".." especially so because it has
1299 * to be able to know about the current root directory and
1300 * parent relationships.
1302 if (unlikely(type
!= LAST_NORM
))
1303 return handle_dots(nd
, type
);
1304 err
= do_lookup(nd
, name
, path
, &inode
);
1305 if (unlikely(err
)) {
1310 path_to_nameidata(path
, nd
);
1314 if (should_follow_link(inode
, follow
)) {
1315 if (nd
->flags
& LOOKUP_RCU
) {
1316 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1321 BUG_ON(inode
!= path
->dentry
->d_inode
);
1324 path_to_nameidata(path
, nd
);
1330 * This limits recursive symlink follows to 8, while
1331 * limiting consecutive symlinks to 40.
1333 * Without that kind of total limit, nasty chains of consecutive
1334 * symlinks can cause almost arbitrarily long lookups.
1336 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1340 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1341 path_put_conditional(path
, nd
);
1342 path_put(&nd
->path
);
1345 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1348 current
->link_count
++;
1351 struct path link
= *path
;
1354 res
= follow_link(&link
, nd
, &cookie
);
1356 res
= walk_component(nd
, path
, &nd
->last
,
1357 nd
->last_type
, LOOKUP_FOLLOW
);
1358 put_link(nd
, &link
, cookie
);
1361 current
->link_count
--;
1367 * We really don't want to look at inode->i_op->lookup
1368 * when we don't have to. So we keep a cache bit in
1369 * the inode ->i_opflags field that says "yes, we can
1370 * do lookup on this inode".
1372 static inline int can_lookup(struct inode
*inode
)
1374 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1376 if (likely(!inode
->i_op
->lookup
))
1379 /* We do this once for the lifetime of the inode */
1380 spin_lock(&inode
->i_lock
);
1381 inode
->i_opflags
|= IOP_LOOKUP
;
1382 spin_unlock(&inode
->i_lock
);
1387 * We can do the critical dentry name comparison and hashing
1388 * operations one word at a time, but we are limited to:
1390 * - Architectures with fast unaligned word accesses. We could
1391 * do a "get_unaligned()" if this helps and is sufficiently
1394 * - Little-endian machines (so that we can generate the mask
1395 * of low bytes efficiently). Again, we *could* do a byte
1396 * swapping load on big-endian architectures if that is not
1397 * expensive enough to make the optimization worthless.
1399 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1400 * do not trap on the (extremely unlikely) case of a page
1401 * crossing operation.
1403 * - Furthermore, we need an efficient 64-bit compile for the
1404 * 64-bit case in order to generate the "number of bytes in
1405 * the final mask". Again, that could be replaced with a
1406 * efficient population count instruction or similar.
1408 #ifdef CONFIG_DCACHE_WORD_ACCESS
1410 #include <asm/word-at-a-time.h>
1414 static inline unsigned int fold_hash(unsigned long hash
)
1416 hash
+= hash
>> (8*sizeof(int));
1420 #else /* 32-bit case */
1422 #define fold_hash(x) (x)
1426 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1428 unsigned long a
, mask
;
1429 unsigned long hash
= 0;
1432 a
= load_unaligned_zeropad(name
);
1433 if (len
< sizeof(unsigned long))
1437 name
+= sizeof(unsigned long);
1438 len
-= sizeof(unsigned long);
1442 mask
= ~(~0ul << len
*8);
1445 return fold_hash(hash
);
1447 EXPORT_SYMBOL(full_name_hash
);
1450 * Calculate the length and hash of the path component, and
1451 * return the length of the component;
1453 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1455 unsigned long a
, mask
, hash
, len
;
1458 len
= -sizeof(unsigned long);
1460 hash
= (hash
+ a
) * 9;
1461 len
+= sizeof(unsigned long);
1462 a
= load_unaligned_zeropad(name
+len
);
1463 /* Do we have any NUL or '/' bytes in this word? */
1464 mask
= has_zero(a
) | has_zero(a
^ REPEAT_BYTE('/'));
1467 /* The mask *below* the first high bit set */
1468 mask
= (mask
- 1) & ~mask
;
1471 *hashp
= fold_hash(hash
);
1473 return len
+ count_masked_bytes(mask
);
1478 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1480 unsigned long hash
= init_name_hash();
1482 hash
= partial_name_hash(*name
++, hash
);
1483 return end_name_hash(hash
);
1485 EXPORT_SYMBOL(full_name_hash
);
1488 * We know there's a real path component here of at least
1491 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1493 unsigned long hash
= init_name_hash();
1494 unsigned long len
= 0, c
;
1496 c
= (unsigned char)*name
;
1499 hash
= partial_name_hash(c
, hash
);
1500 c
= (unsigned char)name
[len
];
1501 } while (c
&& c
!= '/');
1502 *hashp
= end_name_hash(hash
);
1510 * This is the basic name resolution function, turning a pathname into
1511 * the final dentry. We expect 'base' to be positive and a directory.
1513 * Returns 0 and nd will have valid dentry and mnt on success.
1514 * Returns error and drops reference to input namei data on failure.
1516 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1526 /* At this point we know we have a real path component. */
1532 err
= may_lookup(nd
);
1536 len
= hash_name(name
, &this.hash
);
1541 if (name
[0] == '.') switch (len
) {
1543 if (name
[1] == '.') {
1545 nd
->flags
|= LOOKUP_JUMPED
;
1551 if (likely(type
== LAST_NORM
)) {
1552 struct dentry
*parent
= nd
->path
.dentry
;
1553 nd
->flags
&= ~LOOKUP_JUMPED
;
1554 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1555 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1563 goto last_component
;
1565 * If it wasn't NUL, we know it was '/'. Skip that
1566 * slash, and continue until no more slashes.
1570 } while (unlikely(name
[len
] == '/'));
1572 goto last_component
;
1575 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1580 err
= nested_symlink(&next
, nd
);
1584 if (can_lookup(nd
->inode
))
1588 /* here ends the main loop */
1592 nd
->last_type
= type
;
1599 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1600 struct nameidata
*nd
, struct file
**fp
)
1606 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1607 nd
->flags
= flags
| LOOKUP_JUMPED
;
1609 if (flags
& LOOKUP_ROOT
) {
1610 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1612 if (!inode
->i_op
->lookup
)
1614 retval
= inode_permission(inode
, MAY_EXEC
);
1618 nd
->path
= nd
->root
;
1620 if (flags
& LOOKUP_RCU
) {
1621 br_read_lock(vfsmount_lock
);
1623 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1625 path_get(&nd
->path
);
1630 nd
->root
.mnt
= NULL
;
1633 if (flags
& LOOKUP_RCU
) {
1634 br_read_lock(vfsmount_lock
);
1639 path_get(&nd
->root
);
1641 nd
->path
= nd
->root
;
1642 } else if (dfd
== AT_FDCWD
) {
1643 if (flags
& LOOKUP_RCU
) {
1644 struct fs_struct
*fs
= current
->fs
;
1647 br_read_lock(vfsmount_lock
);
1651 seq
= read_seqcount_begin(&fs
->seq
);
1653 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1654 } while (read_seqcount_retry(&fs
->seq
, seq
));
1656 get_fs_pwd(current
->fs
, &nd
->path
);
1659 struct dentry
*dentry
;
1661 file
= fget_raw_light(dfd
, &fput_needed
);
1666 dentry
= file
->f_path
.dentry
;
1670 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1673 retval
= inode_permission(dentry
->d_inode
, MAY_EXEC
);
1678 nd
->path
= file
->f_path
;
1679 if (flags
& LOOKUP_RCU
) {
1682 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1683 br_read_lock(vfsmount_lock
);
1686 path_get(&file
->f_path
);
1687 fput_light(file
, fput_needed
);
1691 nd
->inode
= nd
->path
.dentry
->d_inode
;
1695 fput_light(file
, fput_needed
);
1700 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1702 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1703 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1705 nd
->flags
&= ~LOOKUP_PARENT
;
1706 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1707 nd
->flags
& LOOKUP_FOLLOW
);
1710 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1711 static int path_lookupat(int dfd
, const char *name
,
1712 unsigned int flags
, struct nameidata
*nd
)
1714 struct file
*base
= NULL
;
1719 * Path walking is largely split up into 2 different synchronisation
1720 * schemes, rcu-walk and ref-walk (explained in
1721 * Documentation/filesystems/path-lookup.txt). These share much of the
1722 * path walk code, but some things particularly setup, cleanup, and
1723 * following mounts are sufficiently divergent that functions are
1724 * duplicated. Typically there is a function foo(), and its RCU
1725 * analogue, foo_rcu().
1727 * -ECHILD is the error number of choice (just to avoid clashes) that
1728 * is returned if some aspect of an rcu-walk fails. Such an error must
1729 * be handled by restarting a traditional ref-walk (which will always
1730 * be able to complete).
1732 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1737 current
->total_link_count
= 0;
1738 err
= link_path_walk(name
, nd
);
1740 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1741 err
= lookup_last(nd
, &path
);
1744 struct path link
= path
;
1745 nd
->flags
|= LOOKUP_PARENT
;
1746 err
= follow_link(&link
, nd
, &cookie
);
1748 err
= lookup_last(nd
, &path
);
1749 put_link(nd
, &link
, cookie
);
1754 err
= complete_walk(nd
);
1756 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1757 if (!nd
->inode
->i_op
->lookup
) {
1758 path_put(&nd
->path
);
1766 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1767 path_put(&nd
->root
);
1768 nd
->root
.mnt
= NULL
;
1773 static int do_path_lookup(int dfd
, const char *name
,
1774 unsigned int flags
, struct nameidata
*nd
)
1776 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1777 if (unlikely(retval
== -ECHILD
))
1778 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1779 if (unlikely(retval
== -ESTALE
))
1780 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1782 if (likely(!retval
)) {
1783 if (unlikely(!audit_dummy_context())) {
1784 if (nd
->path
.dentry
&& nd
->inode
)
1785 audit_inode(name
, nd
->path
.dentry
);
1791 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1793 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1796 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1798 struct nameidata nd
;
1799 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1806 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1807 * @dentry: pointer to dentry of the base directory
1808 * @mnt: pointer to vfs mount of the base directory
1809 * @name: pointer to file name
1810 * @flags: lookup flags
1811 * @path: pointer to struct path to fill
1813 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1814 const char *name
, unsigned int flags
,
1817 struct nameidata nd
;
1819 nd
.root
.dentry
= dentry
;
1821 BUG_ON(flags
& LOOKUP_PARENT
);
1822 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1823 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
1830 * Restricted form of lookup. Doesn't follow links, single-component only,
1831 * needs parent already locked. Doesn't follow mounts.
1834 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1836 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1840 * lookup_one_len - filesystem helper to lookup single pathname component
1841 * @name: pathname component to lookup
1842 * @base: base directory to lookup from
1843 * @len: maximum length @len should be interpreted to
1845 * Note that this routine is purely a helper for filesystem usage and should
1846 * not be called by generic code. Also note that by using this function the
1847 * nameidata argument is passed to the filesystem methods and a filesystem
1848 * using this helper needs to be prepared for that.
1850 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1856 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1860 this.hash
= full_name_hash(name
, len
);
1862 return ERR_PTR(-EACCES
);
1865 c
= *(const unsigned char *)name
++;
1866 if (c
== '/' || c
== '\0')
1867 return ERR_PTR(-EACCES
);
1870 * See if the low-level filesystem might want
1871 * to use its own hash..
1873 if (base
->d_flags
& DCACHE_OP_HASH
) {
1874 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1876 return ERR_PTR(err
);
1879 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
1881 return ERR_PTR(err
);
1883 return __lookup_hash(&this, base
, NULL
);
1886 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
1887 struct path
*path
, int *empty
)
1889 struct nameidata nd
;
1890 char *tmp
= getname_flags(name
, flags
, empty
);
1891 int err
= PTR_ERR(tmp
);
1894 BUG_ON(flags
& LOOKUP_PARENT
);
1896 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1904 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1907 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
1910 static int user_path_parent(int dfd
, const char __user
*path
,
1911 struct nameidata
*nd
, char **name
)
1913 char *s
= getname(path
);
1919 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1929 * It's inline, so penalty for filesystems that don't use sticky bit is
1932 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1934 kuid_t fsuid
= current_fsuid();
1936 if (!(dir
->i_mode
& S_ISVTX
))
1938 if (uid_eq(inode
->i_uid
, fsuid
))
1940 if (uid_eq(dir
->i_uid
, fsuid
))
1942 return !inode_capable(inode
, CAP_FOWNER
);
1946 * Check whether we can remove a link victim from directory dir, check
1947 * whether the type of victim is right.
1948 * 1. We can't do it if dir is read-only (done in permission())
1949 * 2. We should have write and exec permissions on dir
1950 * 3. We can't remove anything from append-only dir
1951 * 4. We can't do anything with immutable dir (done in permission())
1952 * 5. If the sticky bit on dir is set we should either
1953 * a. be owner of dir, or
1954 * b. be owner of victim, or
1955 * c. have CAP_FOWNER capability
1956 * 6. If the victim is append-only or immutable we can't do antyhing with
1957 * links pointing to it.
1958 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1959 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1960 * 9. We can't remove a root or mountpoint.
1961 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1962 * nfs_async_unlink().
1964 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1968 if (!victim
->d_inode
)
1971 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1972 audit_inode_child(victim
, dir
);
1974 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1979 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1980 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1983 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1985 if (IS_ROOT(victim
))
1987 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1989 if (IS_DEADDIR(dir
))
1991 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1996 /* Check whether we can create an object with dentry child in directory
1998 * 1. We can't do it if child already exists (open has special treatment for
1999 * this case, but since we are inlined it's OK)
2000 * 2. We can't do it if dir is read-only (done in permission())
2001 * 3. We should have write and exec permissions on dir
2002 * 4. We can't do it if dir is immutable (done in permission())
2004 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2008 if (IS_DEADDIR(dir
))
2010 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2014 * p1 and p2 should be directories on the same fs.
2016 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2021 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2025 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2027 p
= d_ancestor(p2
, p1
);
2029 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2030 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2034 p
= d_ancestor(p1
, p2
);
2036 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2037 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2041 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2042 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2046 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2048 mutex_unlock(&p1
->d_inode
->i_mutex
);
2050 mutex_unlock(&p2
->d_inode
->i_mutex
);
2051 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2055 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2056 struct nameidata
*nd
)
2058 int error
= may_create(dir
, dentry
);
2063 if (!dir
->i_op
->create
)
2064 return -EACCES
; /* shouldn't it be ENOSYS? */
2067 error
= security_inode_create(dir
, dentry
, mode
);
2070 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
2072 fsnotify_create(dir
, dentry
);
2076 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2078 struct dentry
*dentry
= path
->dentry
;
2079 struct inode
*inode
= dentry
->d_inode
;
2089 switch (inode
->i_mode
& S_IFMT
) {
2093 if (acc_mode
& MAY_WRITE
)
2098 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2107 error
= inode_permission(inode
, acc_mode
);
2112 * An append-only file must be opened in append mode for writing.
2114 if (IS_APPEND(inode
)) {
2115 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2121 /* O_NOATIME can only be set by the owner or superuser */
2122 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2128 static int handle_truncate(struct file
*filp
)
2130 struct path
*path
= &filp
->f_path
;
2131 struct inode
*inode
= path
->dentry
->d_inode
;
2132 int error
= get_write_access(inode
);
2136 * Refuse to truncate files with mandatory locks held on them.
2138 error
= locks_verify_locked(inode
);
2140 error
= security_path_truncate(path
);
2142 error
= do_truncate(path
->dentry
, 0,
2143 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2146 put_write_access(inode
);
2150 static inline int open_to_namei_flags(int flag
)
2152 if ((flag
& O_ACCMODE
) == 3)
2158 * Handle the last step of open()
2160 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2161 const struct open_flags
*op
, const char *pathname
)
2163 struct dentry
*dir
= nd
->path
.dentry
;
2164 struct dentry
*dentry
;
2165 int open_flag
= op
->open_flag
;
2166 int will_truncate
= open_flag
& O_TRUNC
;
2168 int acc_mode
= op
->acc_mode
;
2172 nd
->flags
&= ~LOOKUP_PARENT
;
2173 nd
->flags
|= op
->intent
;
2175 switch (nd
->last_type
) {
2178 error
= handle_dots(nd
, nd
->last_type
);
2180 return ERR_PTR(error
);
2183 error
= complete_walk(nd
);
2185 return ERR_PTR(error
);
2186 audit_inode(pathname
, nd
->path
.dentry
);
2187 if (open_flag
& O_CREAT
) {
2193 error
= complete_walk(nd
);
2195 return ERR_PTR(error
);
2196 audit_inode(pathname
, dir
);
2200 if (!(open_flag
& O_CREAT
)) {
2202 if (nd
->last
.name
[nd
->last
.len
])
2203 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2204 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2206 /* we _can_ be in RCU mode here */
2207 error
= walk_component(nd
, path
, &nd
->last
, LAST_NORM
,
2210 return ERR_PTR(error
);
2211 if (error
) /* symlink */
2214 error
= complete_walk(nd
);
2216 return ERR_PTR(error
);
2219 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2220 if (!nd
->inode
->i_op
->lookup
)
2223 audit_inode(pathname
, nd
->path
.dentry
);
2227 /* create side of things */
2229 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2230 * cleared when we got to the last component we are about to look up
2232 error
= complete_walk(nd
);
2234 return ERR_PTR(error
);
2236 audit_inode(pathname
, dir
);
2238 /* trailing slashes? */
2239 if (nd
->last
.name
[nd
->last
.len
])
2242 mutex_lock(&dir
->d_inode
->i_mutex
);
2244 dentry
= lookup_hash(nd
);
2245 error
= PTR_ERR(dentry
);
2246 if (IS_ERR(dentry
)) {
2247 mutex_unlock(&dir
->d_inode
->i_mutex
);
2251 path
->dentry
= dentry
;
2252 path
->mnt
= nd
->path
.mnt
;
2254 /* Negative dentry, just create the file */
2255 if (!dentry
->d_inode
) {
2256 umode_t mode
= op
->mode
;
2257 if (!IS_POSIXACL(dir
->d_inode
))
2258 mode
&= ~current_umask();
2260 * This write is needed to ensure that a
2261 * rw->ro transition does not occur between
2262 * the time when the file is created and when
2263 * a permanent write count is taken through
2264 * the 'struct file' in nameidata_to_filp().
2266 error
= mnt_want_write(nd
->path
.mnt
);
2268 goto exit_mutex_unlock
;
2270 /* Don't check for write permission, don't truncate */
2271 open_flag
&= ~O_TRUNC
;
2273 acc_mode
= MAY_OPEN
;
2274 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2276 goto exit_mutex_unlock
;
2277 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2279 goto exit_mutex_unlock
;
2280 mutex_unlock(&dir
->d_inode
->i_mutex
);
2281 dput(nd
->path
.dentry
);
2282 nd
->path
.dentry
= dentry
;
2287 * It already exists.
2289 mutex_unlock(&dir
->d_inode
->i_mutex
);
2290 audit_inode(pathname
, path
->dentry
);
2293 if (open_flag
& O_EXCL
)
2296 error
= follow_managed(path
, nd
->flags
);
2301 nd
->flags
|= LOOKUP_JUMPED
;
2304 if (!path
->dentry
->d_inode
)
2307 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2310 path_to_nameidata(path
, nd
);
2311 nd
->inode
= path
->dentry
->d_inode
;
2312 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2313 error
= complete_walk(nd
);
2315 return ERR_PTR(error
);
2317 if (S_ISDIR(nd
->inode
->i_mode
))
2320 if (!S_ISREG(nd
->inode
->i_mode
))
2323 if (will_truncate
) {
2324 error
= mnt_want_write(nd
->path
.mnt
);
2330 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2333 filp
= nameidata_to_filp(nd
);
2334 if (!IS_ERR(filp
)) {
2335 error
= ima_file_check(filp
, op
->acc_mode
);
2338 filp
= ERR_PTR(error
);
2341 if (!IS_ERR(filp
)) {
2342 if (will_truncate
) {
2343 error
= handle_truncate(filp
);
2346 filp
= ERR_PTR(error
);
2352 mnt_drop_write(nd
->path
.mnt
);
2353 path_put(&nd
->path
);
2357 mutex_unlock(&dir
->d_inode
->i_mutex
);
2359 path_put_conditional(path
, nd
);
2361 filp
= ERR_PTR(error
);
2365 static struct file
*path_openat(int dfd
, const char *pathname
,
2366 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2368 struct file
*base
= NULL
;
2373 filp
= get_empty_filp();
2375 return ERR_PTR(-ENFILE
);
2377 filp
->f_flags
= op
->open_flag
;
2378 nd
->intent
.open
.file
= filp
;
2379 nd
->intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2380 nd
->intent
.open
.create_mode
= op
->mode
;
2382 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2383 if (unlikely(error
))
2386 current
->total_link_count
= 0;
2387 error
= link_path_walk(pathname
, nd
);
2388 if (unlikely(error
))
2391 filp
= do_last(nd
, &path
, op
, pathname
);
2392 while (unlikely(!filp
)) { /* trailing symlink */
2393 struct path link
= path
;
2395 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2396 path_put_conditional(&path
, nd
);
2397 path_put(&nd
->path
);
2398 filp
= ERR_PTR(-ELOOP
);
2401 nd
->flags
|= LOOKUP_PARENT
;
2402 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2403 error
= follow_link(&link
, nd
, &cookie
);
2404 if (unlikely(error
))
2405 filp
= ERR_PTR(error
);
2407 filp
= do_last(nd
, &path
, op
, pathname
);
2408 put_link(nd
, &link
, cookie
);
2411 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2412 path_put(&nd
->root
);
2415 release_open_intent(nd
);
2419 filp
= ERR_PTR(error
);
2423 struct file
*do_filp_open(int dfd
, const char *pathname
,
2424 const struct open_flags
*op
, int flags
)
2426 struct nameidata nd
;
2429 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2430 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2431 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2432 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2433 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2437 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2438 const char *name
, const struct open_flags
*op
, int flags
)
2440 struct nameidata nd
;
2444 nd
.root
.dentry
= dentry
;
2446 flags
|= LOOKUP_ROOT
;
2448 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2449 return ERR_PTR(-ELOOP
);
2451 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2452 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2453 file
= path_openat(-1, name
, &nd
, op
, flags
);
2454 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2455 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2459 struct dentry
*kern_path_create(int dfd
, const char *pathname
, struct path
*path
, int is_dir
)
2461 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2462 struct nameidata nd
;
2463 int error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
, &nd
);
2465 return ERR_PTR(error
);
2468 * Yucky last component or no last component at all?
2469 * (foo/., foo/.., /////)
2471 if (nd
.last_type
!= LAST_NORM
)
2473 nd
.flags
&= ~LOOKUP_PARENT
;
2474 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2475 nd
.intent
.open
.flags
= O_EXCL
;
2478 * Do the final lookup.
2480 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2481 dentry
= lookup_hash(&nd
);
2485 if (dentry
->d_inode
)
2488 * Special case - lookup gave negative, but... we had foo/bar/
2489 * From the vfs_mknod() POV we just have a negative dentry -
2490 * all is fine. Let's be bastards - you had / on the end, you've
2491 * been asking for (non-existent) directory. -ENOENT for you.
2493 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
2495 dentry
= ERR_PTR(-ENOENT
);
2502 dentry
= ERR_PTR(-EEXIST
);
2504 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2509 EXPORT_SYMBOL(kern_path_create
);
2511 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
, struct path
*path
, int is_dir
)
2513 char *tmp
= getname(pathname
);
2516 return ERR_CAST(tmp
);
2517 res
= kern_path_create(dfd
, tmp
, path
, is_dir
);
2521 EXPORT_SYMBOL(user_path_create
);
2523 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2525 int error
= may_create(dir
, dentry
);
2530 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
2533 if (!dir
->i_op
->mknod
)
2536 error
= devcgroup_inode_mknod(mode
, dev
);
2540 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2544 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2546 fsnotify_create(dir
, dentry
);
2550 static int may_mknod(umode_t mode
)
2552 switch (mode
& S_IFMT
) {
2558 case 0: /* zero mode translates to S_IFREG */
2567 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
2570 struct dentry
*dentry
;
2577 dentry
= user_path_create(dfd
, filename
, &path
, 0);
2579 return PTR_ERR(dentry
);
2581 if (!IS_POSIXACL(path
.dentry
->d_inode
))
2582 mode
&= ~current_umask();
2583 error
= may_mknod(mode
);
2586 error
= mnt_want_write(path
.mnt
);
2589 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
2591 goto out_drop_write
;
2592 switch (mode
& S_IFMT
) {
2593 case 0: case S_IFREG
:
2594 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,NULL
);
2596 case S_IFCHR
: case S_IFBLK
:
2597 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
2598 new_decode_dev(dev
));
2600 case S_IFIFO
: case S_IFSOCK
:
2601 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
2605 mnt_drop_write(path
.mnt
);
2608 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2614 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
2616 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2619 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2621 int error
= may_create(dir
, dentry
);
2622 unsigned max_links
= dir
->i_sb
->s_max_links
;
2627 if (!dir
->i_op
->mkdir
)
2630 mode
&= (S_IRWXUGO
|S_ISVTX
);
2631 error
= security_inode_mkdir(dir
, dentry
, mode
);
2635 if (max_links
&& dir
->i_nlink
>= max_links
)
2638 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2640 fsnotify_mkdir(dir
, dentry
);
2644 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
2646 struct dentry
*dentry
;
2650 dentry
= user_path_create(dfd
, pathname
, &path
, 1);
2652 return PTR_ERR(dentry
);
2654 if (!IS_POSIXACL(path
.dentry
->d_inode
))
2655 mode
&= ~current_umask();
2656 error
= mnt_want_write(path
.mnt
);
2659 error
= security_path_mkdir(&path
, dentry
, mode
);
2661 goto out_drop_write
;
2662 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
2664 mnt_drop_write(path
.mnt
);
2667 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2672 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
2674 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2678 * The dentry_unhash() helper will try to drop the dentry early: we
2679 * should have a usage count of 1 if we're the only user of this
2680 * dentry, and if that is true (possibly after pruning the dcache),
2681 * then we drop the dentry now.
2683 * A low-level filesystem can, if it choses, legally
2686 * if (!d_unhashed(dentry))
2689 * if it cannot handle the case of removing a directory
2690 * that is still in use by something else..
2692 void dentry_unhash(struct dentry
*dentry
)
2694 shrink_dcache_parent(dentry
);
2695 spin_lock(&dentry
->d_lock
);
2696 if (dentry
->d_count
== 1)
2698 spin_unlock(&dentry
->d_lock
);
2701 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2703 int error
= may_delete(dir
, dentry
, 1);
2708 if (!dir
->i_op
->rmdir
)
2712 mutex_lock(&dentry
->d_inode
->i_mutex
);
2715 if (d_mountpoint(dentry
))
2718 error
= security_inode_rmdir(dir
, dentry
);
2722 shrink_dcache_parent(dentry
);
2723 error
= dir
->i_op
->rmdir(dir
, dentry
);
2727 dentry
->d_inode
->i_flags
|= S_DEAD
;
2731 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2738 static long do_rmdir(int dfd
, const char __user
*pathname
)
2742 struct dentry
*dentry
;
2743 struct nameidata nd
;
2745 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2749 switch(nd
.last_type
) {
2761 nd
.flags
&= ~LOOKUP_PARENT
;
2763 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2764 dentry
= lookup_hash(&nd
);
2765 error
= PTR_ERR(dentry
);
2768 if (!dentry
->d_inode
) {
2772 error
= mnt_want_write(nd
.path
.mnt
);
2775 error
= security_path_rmdir(&nd
.path
, dentry
);
2778 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2780 mnt_drop_write(nd
.path
.mnt
);
2784 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2791 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2793 return do_rmdir(AT_FDCWD
, pathname
);
2796 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2798 int error
= may_delete(dir
, dentry
, 0);
2803 if (!dir
->i_op
->unlink
)
2806 mutex_lock(&dentry
->d_inode
->i_mutex
);
2807 if (d_mountpoint(dentry
))
2810 error
= security_inode_unlink(dir
, dentry
);
2812 error
= dir
->i_op
->unlink(dir
, dentry
);
2817 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2819 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2820 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2821 fsnotify_link_count(dentry
->d_inode
);
2829 * Make sure that the actual truncation of the file will occur outside its
2830 * directory's i_mutex. Truncate can take a long time if there is a lot of
2831 * writeout happening, and we don't want to prevent access to the directory
2832 * while waiting on the I/O.
2834 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2838 struct dentry
*dentry
;
2839 struct nameidata nd
;
2840 struct inode
*inode
= NULL
;
2842 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2847 if (nd
.last_type
!= LAST_NORM
)
2850 nd
.flags
&= ~LOOKUP_PARENT
;
2852 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2853 dentry
= lookup_hash(&nd
);
2854 error
= PTR_ERR(dentry
);
2855 if (!IS_ERR(dentry
)) {
2856 /* Why not before? Because we want correct error value */
2857 if (nd
.last
.name
[nd
.last
.len
])
2859 inode
= dentry
->d_inode
;
2863 error
= mnt_want_write(nd
.path
.mnt
);
2866 error
= security_path_unlink(&nd
.path
, dentry
);
2869 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2871 mnt_drop_write(nd
.path
.mnt
);
2875 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2877 iput(inode
); /* truncate the inode here */
2884 error
= !dentry
->d_inode
? -ENOENT
:
2885 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2889 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2891 if ((flag
& ~AT_REMOVEDIR
) != 0)
2894 if (flag
& AT_REMOVEDIR
)
2895 return do_rmdir(dfd
, pathname
);
2897 return do_unlinkat(dfd
, pathname
);
2900 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2902 return do_unlinkat(AT_FDCWD
, pathname
);
2905 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2907 int error
= may_create(dir
, dentry
);
2912 if (!dir
->i_op
->symlink
)
2915 error
= security_inode_symlink(dir
, dentry
, oldname
);
2919 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2921 fsnotify_create(dir
, dentry
);
2925 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2926 int, newdfd
, const char __user
*, newname
)
2930 struct dentry
*dentry
;
2933 from
= getname(oldname
);
2935 return PTR_ERR(from
);
2937 dentry
= user_path_create(newdfd
, newname
, &path
, 0);
2938 error
= PTR_ERR(dentry
);
2942 error
= mnt_want_write(path
.mnt
);
2945 error
= security_path_symlink(&path
, dentry
, from
);
2947 goto out_drop_write
;
2948 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
);
2950 mnt_drop_write(path
.mnt
);
2953 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
2960 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2962 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2965 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2967 struct inode
*inode
= old_dentry
->d_inode
;
2968 unsigned max_links
= dir
->i_sb
->s_max_links
;
2974 error
= may_create(dir
, new_dentry
);
2978 if (dir
->i_sb
!= inode
->i_sb
)
2982 * A link to an append-only or immutable file cannot be created.
2984 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2986 if (!dir
->i_op
->link
)
2988 if (S_ISDIR(inode
->i_mode
))
2991 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2995 mutex_lock(&inode
->i_mutex
);
2996 /* Make sure we don't allow creating hardlink to an unlinked file */
2997 if (inode
->i_nlink
== 0)
2999 else if (max_links
&& inode
->i_nlink
>= max_links
)
3002 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3003 mutex_unlock(&inode
->i_mutex
);
3005 fsnotify_link(dir
, inode
, new_dentry
);
3010 * Hardlinks are often used in delicate situations. We avoid
3011 * security-related surprises by not following symlinks on the
3014 * We don't follow them on the oldname either to be compatible
3015 * with linux 2.0, and to avoid hard-linking to directories
3016 * and other special files. --ADM
3018 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3019 int, newdfd
, const char __user
*, newname
, int, flags
)
3021 struct dentry
*new_dentry
;
3022 struct path old_path
, new_path
;
3026 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3029 * To use null names we require CAP_DAC_READ_SEARCH
3030 * This ensures that not everyone will be able to create
3031 * handlink using the passed filedescriptor.
3033 if (flags
& AT_EMPTY_PATH
) {
3034 if (!capable(CAP_DAC_READ_SEARCH
))
3039 if (flags
& AT_SYMLINK_FOLLOW
)
3040 how
|= LOOKUP_FOLLOW
;
3042 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3046 new_dentry
= user_path_create(newdfd
, newname
, &new_path
, 0);
3047 error
= PTR_ERR(new_dentry
);
3048 if (IS_ERR(new_dentry
))
3052 if (old_path
.mnt
!= new_path
.mnt
)
3054 error
= mnt_want_write(new_path
.mnt
);
3057 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3059 goto out_drop_write
;
3060 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3062 mnt_drop_write(new_path
.mnt
);
3065 mutex_unlock(&new_path
.dentry
->d_inode
->i_mutex
);
3066 path_put(&new_path
);
3068 path_put(&old_path
);
3073 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3075 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3079 * The worst of all namespace operations - renaming directory. "Perverted"
3080 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3082 * a) we can get into loop creation. Check is done in is_subdir().
3083 * b) race potential - two innocent renames can create a loop together.
3084 * That's where 4.4 screws up. Current fix: serialization on
3085 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3087 * c) we have to lock _three_ objects - parents and victim (if it exists).
3088 * And that - after we got ->i_mutex on parents (until then we don't know
3089 * whether the target exists). Solution: try to be smart with locking
3090 * order for inodes. We rely on the fact that tree topology may change
3091 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3092 * move will be locked. Thus we can rank directories by the tree
3093 * (ancestors first) and rank all non-directories after them.
3094 * That works since everybody except rename does "lock parent, lookup,
3095 * lock child" and rename is under ->s_vfs_rename_mutex.
3096 * HOWEVER, it relies on the assumption that any object with ->lookup()
3097 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3098 * we'd better make sure that there's no link(2) for them.
3099 * d) conversion from fhandle to dentry may come in the wrong moment - when
3100 * we are removing the target. Solution: we will have to grab ->i_mutex
3101 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3102 * ->i_mutex on parents, which works but leads to some truly excessive
3105 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3106 struct inode
*new_dir
, struct dentry
*new_dentry
)
3109 struct inode
*target
= new_dentry
->d_inode
;
3110 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3113 * If we are going to change the parent - check write permissions,
3114 * we'll need to flip '..'.
3116 if (new_dir
!= old_dir
) {
3117 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3122 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3128 mutex_lock(&target
->i_mutex
);
3131 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3135 if (max_links
&& !target
&& new_dir
!= old_dir
&&
3136 new_dir
->i_nlink
>= max_links
)
3140 shrink_dcache_parent(new_dentry
);
3141 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3146 target
->i_flags
|= S_DEAD
;
3147 dont_mount(new_dentry
);
3151 mutex_unlock(&target
->i_mutex
);
3154 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3155 d_move(old_dentry
,new_dentry
);
3159 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3160 struct inode
*new_dir
, struct dentry
*new_dentry
)
3162 struct inode
*target
= new_dentry
->d_inode
;
3165 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3171 mutex_lock(&target
->i_mutex
);
3174 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3177 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3182 dont_mount(new_dentry
);
3183 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3184 d_move(old_dentry
, new_dentry
);
3187 mutex_unlock(&target
->i_mutex
);
3192 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3193 struct inode
*new_dir
, struct dentry
*new_dentry
)
3196 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3197 const unsigned char *old_name
;
3199 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3202 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3206 if (!new_dentry
->d_inode
)
3207 error
= may_create(new_dir
, new_dentry
);
3209 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3213 if (!old_dir
->i_op
->rename
)
3216 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3219 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3221 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3223 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3224 new_dentry
->d_inode
, old_dentry
);
3225 fsnotify_oldname_free(old_name
);
3230 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3231 int, newdfd
, const char __user
*, newname
)
3233 struct dentry
*old_dir
, *new_dir
;
3234 struct dentry
*old_dentry
, *new_dentry
;
3235 struct dentry
*trap
;
3236 struct nameidata oldnd
, newnd
;
3241 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3245 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3250 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3253 old_dir
= oldnd
.path
.dentry
;
3255 if (oldnd
.last_type
!= LAST_NORM
)
3258 new_dir
= newnd
.path
.dentry
;
3259 if (newnd
.last_type
!= LAST_NORM
)
3262 oldnd
.flags
&= ~LOOKUP_PARENT
;
3263 newnd
.flags
&= ~LOOKUP_PARENT
;
3264 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3266 trap
= lock_rename(new_dir
, old_dir
);
3268 old_dentry
= lookup_hash(&oldnd
);
3269 error
= PTR_ERR(old_dentry
);
3270 if (IS_ERR(old_dentry
))
3272 /* source must exist */
3274 if (!old_dentry
->d_inode
)
3276 /* unless the source is a directory trailing slashes give -ENOTDIR */
3277 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3279 if (oldnd
.last
.name
[oldnd
.last
.len
])
3281 if (newnd
.last
.name
[newnd
.last
.len
])
3284 /* source should not be ancestor of target */
3286 if (old_dentry
== trap
)
3288 new_dentry
= lookup_hash(&newnd
);
3289 error
= PTR_ERR(new_dentry
);
3290 if (IS_ERR(new_dentry
))
3292 /* target should not be an ancestor of source */
3294 if (new_dentry
== trap
)
3297 error
= mnt_want_write(oldnd
.path
.mnt
);
3300 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3301 &newnd
.path
, new_dentry
);
3304 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3305 new_dir
->d_inode
, new_dentry
);
3307 mnt_drop_write(oldnd
.path
.mnt
);
3313 unlock_rename(new_dir
, old_dir
);
3315 path_put(&newnd
.path
);
3318 path_put(&oldnd
.path
);
3324 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3326 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3329 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3333 len
= PTR_ERR(link
);
3338 if (len
> (unsigned) buflen
)
3340 if (copy_to_user(buffer
, link
, len
))
3347 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3348 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3349 * using) it for any given inode is up to filesystem.
3351 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3353 struct nameidata nd
;
3358 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3360 return PTR_ERR(cookie
);
3362 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3363 if (dentry
->d_inode
->i_op
->put_link
)
3364 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3368 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3370 return __vfs_follow_link(nd
, link
);
3373 /* get the link contents into pagecache */
3374 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3378 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3379 page
= read_mapping_page(mapping
, 0, NULL
);
3384 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3388 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3390 struct page
*page
= NULL
;
3391 char *s
= page_getlink(dentry
, &page
);
3392 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3395 page_cache_release(page
);
3400 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3402 struct page
*page
= NULL
;
3403 nd_set_link(nd
, page_getlink(dentry
, &page
));
3407 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3409 struct page
*page
= cookie
;
3413 page_cache_release(page
);
3418 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3420 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3422 struct address_space
*mapping
= inode
->i_mapping
;
3427 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3429 flags
|= AOP_FLAG_NOFS
;
3432 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3433 flags
, &page
, &fsdata
);
3437 kaddr
= kmap_atomic(page
);
3438 memcpy(kaddr
, symname
, len
-1);
3439 kunmap_atomic(kaddr
);
3441 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3448 mark_inode_dirty(inode
);
3454 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3456 return __page_symlink(inode
, symname
, len
,
3457 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3460 const struct inode_operations page_symlink_inode_operations
= {
3461 .readlink
= generic_readlink
,
3462 .follow_link
= page_follow_link_light
,
3463 .put_link
= page_put_link
,
3466 EXPORT_SYMBOL(user_path_at
);
3467 EXPORT_SYMBOL(follow_down_one
);
3468 EXPORT_SYMBOL(follow_down
);
3469 EXPORT_SYMBOL(follow_up
);
3470 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3471 EXPORT_SYMBOL(getname
);
3472 EXPORT_SYMBOL(lock_rename
);
3473 EXPORT_SYMBOL(lookup_one_len
);
3474 EXPORT_SYMBOL(page_follow_link_light
);
3475 EXPORT_SYMBOL(page_put_link
);
3476 EXPORT_SYMBOL(page_readlink
);
3477 EXPORT_SYMBOL(__page_symlink
);
3478 EXPORT_SYMBOL(page_symlink
);
3479 EXPORT_SYMBOL(page_symlink_inode_operations
);
3480 EXPORT_SYMBOL(kern_path
);
3481 EXPORT_SYMBOL(vfs_path_lookup
);
3482 EXPORT_SYMBOL(inode_permission
);
3483 EXPORT_SYMBOL(unlock_rename
);
3484 EXPORT_SYMBOL(vfs_create
);
3485 EXPORT_SYMBOL(vfs_follow_link
);
3486 EXPORT_SYMBOL(vfs_link
);
3487 EXPORT_SYMBOL(vfs_mkdir
);
3488 EXPORT_SYMBOL(vfs_mknod
);
3489 EXPORT_SYMBOL(generic_permission
);
3490 EXPORT_SYMBOL(vfs_readlink
);
3491 EXPORT_SYMBOL(vfs_rename
);
3492 EXPORT_SYMBOL(vfs_rmdir
);
3493 EXPORT_SYMBOL(vfs_symlink
);
3494 EXPORT_SYMBOL(vfs_unlink
);
3495 EXPORT_SYMBOL(dentry_unhash
);
3496 EXPORT_SYMBOL(generic_readlink
);