4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existent name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user
*filename
, char *page
)
120 unsigned long len
= PATH_MAX
;
122 if (!segment_eq(get_fs(), KERNEL_DS
)) {
123 if ((unsigned long) filename
>= TASK_SIZE
)
125 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
126 len
= TASK_SIZE
- (unsigned long) filename
;
129 retval
= strncpy_from_user(page
, filename
, len
);
133 return -ENAMETOOLONG
;
139 static char *getname_flags(const char __user
*filename
, int flags
, int *empty
)
143 result
= ERR_PTR(-ENOMEM
);
146 int retval
= do_getname(filename
, tmp
);
150 if (retval
== -ENOENT
&& empty
)
152 if (retval
!= -ENOENT
|| !(flags
& LOOKUP_EMPTY
)) {
154 result
= ERR_PTR(retval
);
158 audit_getname(result
);
162 char *getname(const char __user
* filename
)
164 return getname_flags(filename
, 0, 0);
167 #ifdef CONFIG_AUDITSYSCALL
168 void putname(const char *name
)
170 if (unlikely(!audit_dummy_context()))
175 EXPORT_SYMBOL(putname
);
179 * This does basic POSIX ACL permission checking
181 static int acl_permission_check(struct inode
*inode
, int mask
, unsigned int flags
,
182 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
184 unsigned int mode
= inode
->i_mode
;
186 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
188 if (current_user_ns() != inode_userns(inode
))
191 if (current_fsuid() == inode
->i_uid
)
194 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
195 int error
= check_acl(inode
, mask
, flags
);
196 if (error
!= -EAGAIN
)
200 if (in_group_p(inode
->i_gid
))
206 * If the DACs are ok we don't need any capability check.
208 if ((mask
& ~mode
) == 0)
214 * generic_permission - check for access rights on a Posix-like filesystem
215 * @inode: inode to check access rights for
216 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
217 * @check_acl: optional callback to check for Posix ACLs
218 * @flags: IPERM_FLAG_ flags.
220 * Used to check for read/write/execute permissions on a file.
221 * We use "fsuid" for this, letting us set arbitrary permissions
222 * for filesystem access without changing the "normal" uids which
223 * are used for other things.
225 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
226 * request cannot be satisfied (eg. requires blocking or too much complexity).
227 * It would then be called again in ref-walk mode.
229 int generic_permission(struct inode
*inode
, int mask
, unsigned int flags
,
230 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
235 * Do the basic POSIX ACL permission checks.
237 ret
= acl_permission_check(inode
, mask
, flags
, check_acl
);
242 * Read/write DACs are always overridable.
243 * Executable DACs are overridable for all directories and
244 * for non-directories that have least one exec bit set.
246 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
247 if (ns_capable(inode_userns(inode
), CAP_DAC_OVERRIDE
))
251 * Searching includes executable on directories, else just read.
253 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
254 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
255 if (ns_capable(inode_userns(inode
), CAP_DAC_READ_SEARCH
))
262 * inode_permission - check for access rights to a given inode
263 * @inode: inode to check permission on
264 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
266 * Used to check for read/write/execute permissions on an inode.
267 * We use "fsuid" for this, letting us set arbitrary permissions
268 * for filesystem access without changing the "normal" uids which
269 * are used for other things.
271 int inode_permission(struct inode
*inode
, int mask
)
275 if (mask
& MAY_WRITE
) {
276 umode_t mode
= inode
->i_mode
;
279 * Nobody gets write access to a read-only fs.
281 if (IS_RDONLY(inode
) &&
282 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
286 * Nobody gets write access to an immutable file.
288 if (IS_IMMUTABLE(inode
))
292 if (inode
->i_op
->permission
)
293 retval
= inode
->i_op
->permission(inode
, mask
, 0);
295 retval
= generic_permission(inode
, mask
, 0,
296 inode
->i_op
->check_acl
);
301 retval
= devcgroup_inode_permission(inode
, mask
);
305 return security_inode_permission(inode
, mask
);
309 * file_permission - check for additional access rights to a given file
310 * @file: file to check access rights for
311 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
313 * Used to check for read/write/execute permissions on an already opened
317 * Do not use this function in new code. All access checks should
318 * be done using inode_permission().
320 int file_permission(struct file
*file
, int mask
)
322 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
326 * get_write_access() gets write permission for a file.
327 * put_write_access() releases this write permission.
328 * This is used for regular files.
329 * We cannot support write (and maybe mmap read-write shared) accesses and
330 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
331 * can have the following values:
332 * 0: no writers, no VM_DENYWRITE mappings
333 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
334 * > 0: (i_writecount) users are writing to the file.
336 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
337 * except for the cases where we don't hold i_writecount yet. Then we need to
338 * use {get,deny}_write_access() - these functions check the sign and refuse
339 * to do the change if sign is wrong. Exclusion between them is provided by
340 * the inode->i_lock spinlock.
343 int get_write_access(struct inode
* inode
)
345 spin_lock(&inode
->i_lock
);
346 if (atomic_read(&inode
->i_writecount
) < 0) {
347 spin_unlock(&inode
->i_lock
);
350 atomic_inc(&inode
->i_writecount
);
351 spin_unlock(&inode
->i_lock
);
356 int deny_write_access(struct file
* file
)
358 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
360 spin_lock(&inode
->i_lock
);
361 if (atomic_read(&inode
->i_writecount
) > 0) {
362 spin_unlock(&inode
->i_lock
);
365 atomic_dec(&inode
->i_writecount
);
366 spin_unlock(&inode
->i_lock
);
372 * path_get - get a reference to a path
373 * @path: path to get the reference to
375 * Given a path increment the reference count to the dentry and the vfsmount.
377 void path_get(struct path
*path
)
382 EXPORT_SYMBOL(path_get
);
385 * path_put - put a reference to a path
386 * @path: path to put the reference to
388 * Given a path decrement the reference count to the dentry and the vfsmount.
390 void path_put(struct path
*path
)
395 EXPORT_SYMBOL(path_put
);
398 * Path walking has 2 modes, rcu-walk and ref-walk (see
399 * Documentation/filesystems/path-lookup.txt). In situations when we can't
400 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
401 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
402 * mode. Refcounts are grabbed at the last known good point before rcu-walk
403 * got stuck, so ref-walk may continue from there. If this is not successful
404 * (eg. a seqcount has changed), then failure is returned and it's up to caller
405 * to restart the path walk from the beginning in ref-walk mode.
409 * unlazy_walk - try to switch to ref-walk mode.
410 * @nd: nameidata pathwalk data
411 * @dentry: child of nd->path.dentry or NULL
412 * Returns: 0 on success, -ECHILD on failure
414 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
415 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
416 * @nd or NULL. Must be called from rcu-walk context.
418 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
420 struct fs_struct
*fs
= current
->fs
;
421 struct dentry
*parent
= nd
->path
.dentry
;
424 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
425 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
427 spin_lock(&fs
->lock
);
428 if (nd
->root
.mnt
!= fs
->root
.mnt
||
429 nd
->root
.dentry
!= fs
->root
.dentry
)
432 spin_lock(&parent
->d_lock
);
434 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
436 BUG_ON(nd
->inode
!= parent
->d_inode
);
438 if (dentry
->d_parent
!= parent
)
440 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
441 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
444 * If the sequence check on the child dentry passed, then
445 * the child has not been removed from its parent. This
446 * means the parent dentry must be valid and able to take
447 * a reference at this point.
449 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
450 BUG_ON(!parent
->d_count
);
452 spin_unlock(&dentry
->d_lock
);
454 spin_unlock(&parent
->d_lock
);
457 spin_unlock(&fs
->lock
);
459 mntget(nd
->path
.mnt
);
462 br_read_unlock(vfsmount_lock
);
463 nd
->flags
&= ~LOOKUP_RCU
;
467 spin_unlock(&dentry
->d_lock
);
469 spin_unlock(&parent
->d_lock
);
472 spin_unlock(&fs
->lock
);
477 * release_open_intent - free up open intent resources
478 * @nd: pointer to nameidata
480 void release_open_intent(struct nameidata
*nd
)
482 struct file
*file
= nd
->intent
.open
.file
;
484 if (file
&& !IS_ERR(file
)) {
485 if (file
->f_path
.dentry
== NULL
)
492 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
494 return dentry
->d_op
->d_revalidate(dentry
, nd
);
497 static struct dentry
*
498 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
500 int status
= d_revalidate(dentry
, nd
);
501 if (unlikely(status
<= 0)) {
503 * The dentry failed validation.
504 * If d_revalidate returned 0 attempt to invalidate
505 * the dentry otherwise d_revalidate is asking us
506 * to return a fail status.
510 dentry
= ERR_PTR(status
);
511 } else if (!d_invalidate(dentry
)) {
520 * complete_walk - successful completion of path walk
521 * @nd: pointer nameidata
523 * If we had been in RCU mode, drop out of it and legitimize nd->path.
524 * Revalidate the final result, unless we'd already done that during
525 * the path walk or the filesystem doesn't ask for it. Return 0 on
526 * success, -error on failure. In case of failure caller does not
527 * need to drop nd->path.
529 static int complete_walk(struct nameidata
*nd
)
531 struct dentry
*dentry
= nd
->path
.dentry
;
534 if (nd
->flags
& LOOKUP_RCU
) {
535 nd
->flags
&= ~LOOKUP_RCU
;
536 if (!(nd
->flags
& LOOKUP_ROOT
))
538 spin_lock(&dentry
->d_lock
);
539 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
540 spin_unlock(&dentry
->d_lock
);
542 br_read_unlock(vfsmount_lock
);
545 BUG_ON(nd
->inode
!= dentry
->d_inode
);
546 spin_unlock(&dentry
->d_lock
);
547 mntget(nd
->path
.mnt
);
549 br_read_unlock(vfsmount_lock
);
552 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
555 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
558 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
561 /* Note: we do not d_invalidate() */
562 status
= d_revalidate(dentry
, nd
);
574 * Short-cut version of permission(), for calling on directories
575 * during pathname resolution. Combines parts of permission()
576 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
578 * If appropriate, check DAC only. If not appropriate, or
579 * short-cut DAC fails, then call ->permission() to do more
580 * complete permission check.
582 static inline int exec_permission(struct inode
*inode
, unsigned int flags
)
585 struct user_namespace
*ns
= inode_userns(inode
);
587 if (inode
->i_op
->permission
) {
588 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
, flags
);
590 ret
= acl_permission_check(inode
, MAY_EXEC
, flags
,
591 inode
->i_op
->check_acl
);
598 if (ns_capable(ns
, CAP_DAC_OVERRIDE
) ||
599 ns_capable(ns
, CAP_DAC_READ_SEARCH
))
604 return security_inode_exec_permission(inode
, flags
);
607 static __always_inline
void set_root(struct nameidata
*nd
)
610 get_fs_root(current
->fs
, &nd
->root
);
613 static int link_path_walk(const char *, struct nameidata
*);
615 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
618 struct fs_struct
*fs
= current
->fs
;
622 seq
= read_seqcount_begin(&fs
->seq
);
624 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
625 } while (read_seqcount_retry(&fs
->seq
, seq
));
629 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
641 nd
->flags
|= LOOKUP_JUMPED
;
643 nd
->inode
= nd
->path
.dentry
->d_inode
;
645 ret
= link_path_walk(link
, nd
);
649 return PTR_ERR(link
);
652 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
655 if (path
->mnt
!= nd
->path
.mnt
)
659 static inline void path_to_nameidata(const struct path
*path
,
660 struct nameidata
*nd
)
662 if (!(nd
->flags
& LOOKUP_RCU
)) {
663 dput(nd
->path
.dentry
);
664 if (nd
->path
.mnt
!= path
->mnt
)
665 mntput(nd
->path
.mnt
);
667 nd
->path
.mnt
= path
->mnt
;
668 nd
->path
.dentry
= path
->dentry
;
671 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
673 struct inode
*inode
= link
->dentry
->d_inode
;
674 if (!IS_ERR(cookie
) && inode
->i_op
->put_link
)
675 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
679 static __always_inline
int
680 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
683 struct dentry
*dentry
= link
->dentry
;
685 BUG_ON(nd
->flags
& LOOKUP_RCU
);
687 if (link
->mnt
== nd
->path
.mnt
)
690 if (unlikely(current
->total_link_count
>= 40)) {
691 *p
= ERR_PTR(-ELOOP
); /* no ->put_link(), please */
696 current
->total_link_count
++;
698 touch_atime(link
->mnt
, dentry
);
699 nd_set_link(nd
, NULL
);
701 error
= security_inode_follow_link(link
->dentry
, nd
);
703 *p
= ERR_PTR(error
); /* no ->put_link(), please */
708 nd
->last_type
= LAST_BIND
;
709 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
712 char *s
= nd_get_link(nd
);
715 error
= __vfs_follow_link(nd
, s
);
716 else if (nd
->last_type
== LAST_BIND
) {
717 nd
->flags
|= LOOKUP_JUMPED
;
718 nd
->inode
= nd
->path
.dentry
->d_inode
;
719 if (nd
->inode
->i_op
->follow_link
) {
720 /* stepped on a _really_ weird one */
729 static int follow_up_rcu(struct path
*path
)
731 struct vfsmount
*parent
;
732 struct dentry
*mountpoint
;
734 parent
= path
->mnt
->mnt_parent
;
735 if (parent
== path
->mnt
)
737 mountpoint
= path
->mnt
->mnt_mountpoint
;
738 path
->dentry
= mountpoint
;
743 int follow_up(struct path
*path
)
745 struct vfsmount
*parent
;
746 struct dentry
*mountpoint
;
748 br_read_lock(vfsmount_lock
);
749 parent
= path
->mnt
->mnt_parent
;
750 if (parent
== path
->mnt
) {
751 br_read_unlock(vfsmount_lock
);
755 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
756 br_read_unlock(vfsmount_lock
);
758 path
->dentry
= mountpoint
;
765 * Perform an automount
766 * - return -EISDIR to tell follow_managed() to stop and return the path we
769 static int follow_automount(struct path
*path
, unsigned flags
,
772 struct vfsmount
*mnt
;
775 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
778 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
779 * and this is the terminal part of the path.
781 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
782 return -EISDIR
; /* we actually want to stop here */
784 /* We don't want to mount if someone's just doing a stat -
785 * unless they're stat'ing a directory and appended a '/' to
788 * We do, however, want to mount if someone wants to open or
789 * create a file of any type under the mountpoint, wants to
790 * traverse through the mountpoint or wants to open the
791 * mounted directory. Also, autofs may mark negative dentries
792 * as being automount points. These will need the attentions
793 * of the daemon to instantiate them before they can be used.
795 if (!(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
796 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
797 path
->dentry
->d_inode
)
800 current
->total_link_count
++;
801 if (current
->total_link_count
>= 40)
804 mnt
= path
->dentry
->d_op
->d_automount(path
);
807 * The filesystem is allowed to return -EISDIR here to indicate
808 * it doesn't want to automount. For instance, autofs would do
809 * this so that its userspace daemon can mount on this dentry.
811 * However, we can only permit this if it's a terminal point in
812 * the path being looked up; if it wasn't then the remainder of
813 * the path is inaccessible and we should say so.
815 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_CONTINUE
))
820 if (!mnt
) /* mount collision */
824 /* lock_mount() may release path->mnt on error */
828 err
= finish_automount(mnt
, path
);
832 /* Someone else made a mount here whilst we were busy */
837 path
->dentry
= dget(mnt
->mnt_root
);
846 * Handle a dentry that is managed in some way.
847 * - Flagged for transit management (autofs)
848 * - Flagged as mountpoint
849 * - Flagged as automount point
851 * This may only be called in refwalk mode.
853 * Serialization is taken care of in namespace.c
855 static int follow_managed(struct path
*path
, unsigned flags
)
857 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
859 bool need_mntput
= false;
862 /* Given that we're not holding a lock here, we retain the value in a
863 * local variable for each dentry as we look at it so that we don't see
864 * the components of that value change under us */
865 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
866 managed
&= DCACHE_MANAGED_DENTRY
,
867 unlikely(managed
!= 0)) {
868 /* Allow the filesystem to manage the transit without i_mutex
870 if (managed
& DCACHE_MANAGE_TRANSIT
) {
871 BUG_ON(!path
->dentry
->d_op
);
872 BUG_ON(!path
->dentry
->d_op
->d_manage
);
873 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
878 /* Transit to a mounted filesystem. */
879 if (managed
& DCACHE_MOUNTED
) {
880 struct vfsmount
*mounted
= lookup_mnt(path
);
886 path
->dentry
= dget(mounted
->mnt_root
);
891 /* Something is mounted on this dentry in another
892 * namespace and/or whatever was mounted there in this
893 * namespace got unmounted before we managed to get the
897 /* Handle an automount point */
898 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
899 ret
= follow_automount(path
, flags
, &need_mntput
);
905 /* We didn't change the current path point */
909 if (need_mntput
&& path
->mnt
== mnt
)
916 int follow_down_one(struct path
*path
)
918 struct vfsmount
*mounted
;
920 mounted
= lookup_mnt(path
);
925 path
->dentry
= dget(mounted
->mnt_root
);
931 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
933 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
934 dentry
->d_op
->d_manage(dentry
, true) < 0);
938 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
939 * we meet a managed dentry that would need blocking.
941 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
942 struct inode
**inode
)
945 struct vfsmount
*mounted
;
947 * Don't forget we might have a non-mountpoint managed dentry
948 * that wants to block transit.
950 if (unlikely(managed_dentry_might_block(path
->dentry
)))
953 if (!d_mountpoint(path
->dentry
))
956 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
960 path
->dentry
= mounted
->mnt_root
;
961 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
963 * Update the inode too. We don't need to re-check the
964 * dentry sequence number here after this d_inode read,
965 * because a mount-point is always pinned.
967 *inode
= path
->dentry
->d_inode
;
972 static void follow_mount_rcu(struct nameidata
*nd
)
974 while (d_mountpoint(nd
->path
.dentry
)) {
975 struct vfsmount
*mounted
;
976 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
979 nd
->path
.mnt
= mounted
;
980 nd
->path
.dentry
= mounted
->mnt_root
;
981 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
985 static int follow_dotdot_rcu(struct nameidata
*nd
)
990 if (nd
->path
.dentry
== nd
->root
.dentry
&&
991 nd
->path
.mnt
== nd
->root
.mnt
) {
994 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
995 struct dentry
*old
= nd
->path
.dentry
;
996 struct dentry
*parent
= old
->d_parent
;
999 seq
= read_seqcount_begin(&parent
->d_seq
);
1000 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1002 nd
->path
.dentry
= parent
;
1006 if (!follow_up_rcu(&nd
->path
))
1008 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1010 follow_mount_rcu(nd
);
1011 nd
->inode
= nd
->path
.dentry
->d_inode
;
1015 nd
->flags
&= ~LOOKUP_RCU
;
1016 if (!(nd
->flags
& LOOKUP_ROOT
))
1017 nd
->root
.mnt
= NULL
;
1019 br_read_unlock(vfsmount_lock
);
1024 * Follow down to the covering mount currently visible to userspace. At each
1025 * point, the filesystem owning that dentry may be queried as to whether the
1026 * caller is permitted to proceed or not.
1028 int follow_down(struct path
*path
)
1033 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1034 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1035 /* Allow the filesystem to manage the transit without i_mutex
1038 * We indicate to the filesystem if someone is trying to mount
1039 * something here. This gives autofs the chance to deny anyone
1040 * other than its daemon the right to mount on its
1043 * The filesystem may sleep at this point.
1045 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1046 BUG_ON(!path
->dentry
->d_op
);
1047 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1048 ret
= path
->dentry
->d_op
->d_manage(
1049 path
->dentry
, false);
1051 return ret
== -EISDIR
? 0 : ret
;
1054 /* Transit to a mounted filesystem. */
1055 if (managed
& DCACHE_MOUNTED
) {
1056 struct vfsmount
*mounted
= lookup_mnt(path
);
1061 path
->mnt
= mounted
;
1062 path
->dentry
= dget(mounted
->mnt_root
);
1066 /* Don't handle automount points here */
1073 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1075 static void follow_mount(struct path
*path
)
1077 while (d_mountpoint(path
->dentry
)) {
1078 struct vfsmount
*mounted
= lookup_mnt(path
);
1083 path
->mnt
= mounted
;
1084 path
->dentry
= dget(mounted
->mnt_root
);
1088 static void follow_dotdot(struct nameidata
*nd
)
1093 struct dentry
*old
= nd
->path
.dentry
;
1095 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1096 nd
->path
.mnt
== nd
->root
.mnt
) {
1099 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1100 /* rare case of legitimate dget_parent()... */
1101 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1105 if (!follow_up(&nd
->path
))
1108 follow_mount(&nd
->path
);
1109 nd
->inode
= nd
->path
.dentry
->d_inode
;
1113 * Allocate a dentry with name and parent, and perform a parent
1114 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1115 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1116 * have verified that no child exists while under i_mutex.
1118 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1119 struct qstr
*name
, struct nameidata
*nd
)
1121 struct inode
*inode
= parent
->d_inode
;
1122 struct dentry
*dentry
;
1125 /* Don't create child dentry for a dead directory. */
1126 if (unlikely(IS_DEADDIR(inode
)))
1127 return ERR_PTR(-ENOENT
);
1129 dentry
= d_alloc(parent
, name
);
1130 if (unlikely(!dentry
))
1131 return ERR_PTR(-ENOMEM
);
1133 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1134 if (unlikely(old
)) {
1142 * It's more convoluted than I'd like it to be, but... it's still fairly
1143 * small and for now I'd prefer to have fast path as straight as possible.
1144 * It _is_ time-critical.
1146 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1147 struct path
*path
, struct inode
**inode
)
1149 struct vfsmount
*mnt
= nd
->path
.mnt
;
1150 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1156 * Rename seqlock is not required here because in the off chance
1157 * of a false negative due to a concurrent rename, we're going to
1158 * do the non-racy lookup, below.
1160 if (nd
->flags
& LOOKUP_RCU
) {
1163 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1167 /* Memory barrier in read_seqcount_begin of child is enough */
1168 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1172 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1173 status
= d_revalidate(dentry
, nd
);
1174 if (unlikely(status
<= 0)) {
1175 if (status
!= -ECHILD
)
1181 path
->dentry
= dentry
;
1182 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1184 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1188 if (unlazy_walk(nd
, dentry
))
1191 dentry
= __d_lookup(parent
, name
);
1195 if (unlikely(!dentry
)) {
1196 struct inode
*dir
= parent
->d_inode
;
1197 BUG_ON(nd
->inode
!= dir
);
1199 mutex_lock(&dir
->i_mutex
);
1200 dentry
= d_lookup(parent
, name
);
1201 if (likely(!dentry
)) {
1202 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1203 if (IS_ERR(dentry
)) {
1204 mutex_unlock(&dir
->i_mutex
);
1205 return PTR_ERR(dentry
);
1211 mutex_unlock(&dir
->i_mutex
);
1213 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1214 status
= d_revalidate(dentry
, nd
);
1215 if (unlikely(status
<= 0)) {
1220 if (!d_invalidate(dentry
)) {
1229 path
->dentry
= dentry
;
1230 err
= follow_managed(path
, nd
->flags
);
1231 if (unlikely(err
< 0)) {
1232 path_put_conditional(path
, nd
);
1235 *inode
= path
->dentry
->d_inode
;
1239 static inline int may_lookup(struct nameidata
*nd
)
1241 if (nd
->flags
& LOOKUP_RCU
) {
1242 int err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1245 if (unlazy_walk(nd
, NULL
))
1248 return exec_permission(nd
->inode
, 0);
1251 static inline int handle_dots(struct nameidata
*nd
, int type
)
1253 if (type
== LAST_DOTDOT
) {
1254 if (nd
->flags
& LOOKUP_RCU
) {
1255 if (follow_dotdot_rcu(nd
))
1263 static void terminate_walk(struct nameidata
*nd
)
1265 if (!(nd
->flags
& LOOKUP_RCU
)) {
1266 path_put(&nd
->path
);
1268 nd
->flags
&= ~LOOKUP_RCU
;
1269 if (!(nd
->flags
& LOOKUP_ROOT
))
1270 nd
->root
.mnt
= NULL
;
1272 br_read_unlock(vfsmount_lock
);
1276 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1277 struct qstr
*name
, int type
, int follow
)
1279 struct inode
*inode
;
1282 * "." and ".." are special - ".." especially so because it has
1283 * to be able to know about the current root directory and
1284 * parent relationships.
1286 if (unlikely(type
!= LAST_NORM
))
1287 return handle_dots(nd
, type
);
1288 err
= do_lookup(nd
, name
, path
, &inode
);
1289 if (unlikely(err
)) {
1294 path_to_nameidata(path
, nd
);
1298 if (unlikely(inode
->i_op
->follow_link
) && follow
) {
1299 if (nd
->flags
& LOOKUP_RCU
) {
1300 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1305 BUG_ON(inode
!= path
->dentry
->d_inode
);
1308 path_to_nameidata(path
, nd
);
1314 * This limits recursive symlink follows to 8, while
1315 * limiting consecutive symlinks to 40.
1317 * Without that kind of total limit, nasty chains of consecutive
1318 * symlinks can cause almost arbitrarily long lookups.
1320 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1324 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1325 path_put_conditional(path
, nd
);
1326 path_put(&nd
->path
);
1329 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1332 current
->link_count
++;
1335 struct path link
= *path
;
1338 res
= follow_link(&link
, nd
, &cookie
);
1340 res
= walk_component(nd
, path
, &nd
->last
,
1341 nd
->last_type
, LOOKUP_FOLLOW
);
1342 put_link(nd
, &link
, cookie
);
1345 current
->link_count
--;
1352 * This is the basic name resolution function, turning a pathname into
1353 * the final dentry. We expect 'base' to be positive and a directory.
1355 * Returns 0 and nd will have valid dentry and mnt on success.
1356 * Returns error and drops reference to input namei data on failure.
1358 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1362 unsigned int lookup_flags
= nd
->flags
;
1369 /* At this point we know we have a real path component. */
1376 nd
->flags
|= LOOKUP_CONTINUE
;
1378 err
= may_lookup(nd
);
1383 c
= *(const unsigned char *)name
;
1385 hash
= init_name_hash();
1388 hash
= partial_name_hash(c
, hash
);
1389 c
= *(const unsigned char *)name
;
1390 } while (c
&& (c
!= '/'));
1391 this.len
= name
- (const char *) this.name
;
1392 this.hash
= end_name_hash(hash
);
1395 if (this.name
[0] == '.') switch (this.len
) {
1397 if (this.name
[1] == '.') {
1399 nd
->flags
|= LOOKUP_JUMPED
;
1405 if (likely(type
== LAST_NORM
)) {
1406 struct dentry
*parent
= nd
->path
.dentry
;
1407 nd
->flags
&= ~LOOKUP_JUMPED
;
1408 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1409 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1416 /* remove trailing slashes? */
1418 goto last_component
;
1419 while (*++name
== '/');
1421 goto last_component
;
1423 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1428 err
= nested_symlink(&next
, nd
);
1433 if (!nd
->inode
->i_op
->lookup
)
1436 /* here ends the main loop */
1439 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1440 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1442 nd
->last_type
= type
;
1449 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1450 struct nameidata
*nd
, struct file
**fp
)
1456 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1457 nd
->flags
= flags
| LOOKUP_JUMPED
;
1459 if (flags
& LOOKUP_ROOT
) {
1460 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1462 if (!inode
->i_op
->lookup
)
1464 retval
= inode_permission(inode
, MAY_EXEC
);
1468 nd
->path
= nd
->root
;
1470 if (flags
& LOOKUP_RCU
) {
1471 br_read_lock(vfsmount_lock
);
1473 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1475 path_get(&nd
->path
);
1480 nd
->root
.mnt
= NULL
;
1483 if (flags
& LOOKUP_RCU
) {
1484 br_read_lock(vfsmount_lock
);
1489 path_get(&nd
->root
);
1491 nd
->path
= nd
->root
;
1492 } else if (dfd
== AT_FDCWD
) {
1493 if (flags
& LOOKUP_RCU
) {
1494 struct fs_struct
*fs
= current
->fs
;
1497 br_read_lock(vfsmount_lock
);
1501 seq
= read_seqcount_begin(&fs
->seq
);
1503 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1504 } while (read_seqcount_retry(&fs
->seq
, seq
));
1506 get_fs_pwd(current
->fs
, &nd
->path
);
1509 struct dentry
*dentry
;
1511 file
= fget_raw_light(dfd
, &fput_needed
);
1516 dentry
= file
->f_path
.dentry
;
1520 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1523 retval
= file_permission(file
, MAY_EXEC
);
1528 nd
->path
= file
->f_path
;
1529 if (flags
& LOOKUP_RCU
) {
1532 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1533 br_read_lock(vfsmount_lock
);
1536 path_get(&file
->f_path
);
1537 fput_light(file
, fput_needed
);
1541 nd
->inode
= nd
->path
.dentry
->d_inode
;
1545 fput_light(file
, fput_needed
);
1550 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1552 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1553 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1555 nd
->flags
&= ~LOOKUP_PARENT
;
1556 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1557 nd
->flags
& LOOKUP_FOLLOW
);
1560 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1561 static int path_lookupat(int dfd
, const char *name
,
1562 unsigned int flags
, struct nameidata
*nd
)
1564 struct file
*base
= NULL
;
1569 * Path walking is largely split up into 2 different synchronisation
1570 * schemes, rcu-walk and ref-walk (explained in
1571 * Documentation/filesystems/path-lookup.txt). These share much of the
1572 * path walk code, but some things particularly setup, cleanup, and
1573 * following mounts are sufficiently divergent that functions are
1574 * duplicated. Typically there is a function foo(), and its RCU
1575 * analogue, foo_rcu().
1577 * -ECHILD is the error number of choice (just to avoid clashes) that
1578 * is returned if some aspect of an rcu-walk fails. Such an error must
1579 * be handled by restarting a traditional ref-walk (which will always
1580 * be able to complete).
1582 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1587 current
->total_link_count
= 0;
1588 err
= link_path_walk(name
, nd
);
1590 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1591 err
= lookup_last(nd
, &path
);
1594 struct path link
= path
;
1595 nd
->flags
|= LOOKUP_PARENT
;
1596 err
= follow_link(&link
, nd
, &cookie
);
1598 err
= lookup_last(nd
, &path
);
1599 put_link(nd
, &link
, cookie
);
1604 err
= complete_walk(nd
);
1606 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1607 if (!nd
->inode
->i_op
->lookup
) {
1608 path_put(&nd
->path
);
1616 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1617 path_put(&nd
->root
);
1618 nd
->root
.mnt
= NULL
;
1623 static int do_path_lookup(int dfd
, const char *name
,
1624 unsigned int flags
, struct nameidata
*nd
)
1626 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1627 if (unlikely(retval
== -ECHILD
))
1628 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1629 if (unlikely(retval
== -ESTALE
))
1630 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1632 if (likely(!retval
)) {
1633 if (unlikely(!audit_dummy_context())) {
1634 if (nd
->path
.dentry
&& nd
->inode
)
1635 audit_inode(name
, nd
->path
.dentry
);
1641 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1643 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1646 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1648 struct nameidata nd
;
1649 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1656 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1657 * @dentry: pointer to dentry of the base directory
1658 * @mnt: pointer to vfs mount of the base directory
1659 * @name: pointer to file name
1660 * @flags: lookup flags
1661 * @nd: pointer to nameidata
1663 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1664 const char *name
, unsigned int flags
,
1665 struct nameidata
*nd
)
1667 nd
->root
.dentry
= dentry
;
1669 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1670 return do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, nd
);
1673 static struct dentry
*__lookup_hash(struct qstr
*name
,
1674 struct dentry
*base
, struct nameidata
*nd
)
1676 struct inode
*inode
= base
->d_inode
;
1677 struct dentry
*dentry
;
1680 err
= exec_permission(inode
, 0);
1682 return ERR_PTR(err
);
1685 * Don't bother with __d_lookup: callers are for creat as
1686 * well as unlink, so a lot of the time it would cost
1689 dentry
= d_lookup(base
, name
);
1691 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1692 dentry
= do_revalidate(dentry
, nd
);
1695 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1701 * Restricted form of lookup. Doesn't follow links, single-component only,
1702 * needs parent already locked. Doesn't follow mounts.
1705 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1707 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1711 * lookup_one_len - filesystem helper to lookup single pathname component
1712 * @name: pathname component to lookup
1713 * @base: base directory to lookup from
1714 * @len: maximum length @len should be interpreted to
1716 * Note that this routine is purely a helper for filesystem usage and should
1717 * not be called by generic code. Also note that by using this function the
1718 * nameidata argument is passed to the filesystem methods and a filesystem
1719 * using this helper needs to be prepared for that.
1721 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1727 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1732 return ERR_PTR(-EACCES
);
1734 hash
= init_name_hash();
1736 c
= *(const unsigned char *)name
++;
1737 if (c
== '/' || c
== '\0')
1738 return ERR_PTR(-EACCES
);
1739 hash
= partial_name_hash(c
, hash
);
1741 this.hash
= end_name_hash(hash
);
1743 * See if the low-level filesystem might want
1744 * to use its own hash..
1746 if (base
->d_flags
& DCACHE_OP_HASH
) {
1747 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1749 return ERR_PTR(err
);
1752 return __lookup_hash(&this, base
, NULL
);
1755 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
1756 struct path
*path
, int *empty
)
1758 struct nameidata nd
;
1759 char *tmp
= getname_flags(name
, flags
, empty
);
1760 int err
= PTR_ERR(tmp
);
1763 BUG_ON(flags
& LOOKUP_PARENT
);
1765 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1773 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1776 return user_path_at_empty(dfd
, name
, flags
, path
, 0);
1779 static int user_path_parent(int dfd
, const char __user
*path
,
1780 struct nameidata
*nd
, char **name
)
1782 char *s
= getname(path
);
1788 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1798 * It's inline, so penalty for filesystems that don't use sticky bit is
1801 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1803 uid_t fsuid
= current_fsuid();
1805 if (!(dir
->i_mode
& S_ISVTX
))
1807 if (current_user_ns() != inode_userns(inode
))
1809 if (inode
->i_uid
== fsuid
)
1811 if (dir
->i_uid
== fsuid
)
1815 return !ns_capable(inode_userns(inode
), CAP_FOWNER
);
1819 * Check whether we can remove a link victim from directory dir, check
1820 * whether the type of victim is right.
1821 * 1. We can't do it if dir is read-only (done in permission())
1822 * 2. We should have write and exec permissions on dir
1823 * 3. We can't remove anything from append-only dir
1824 * 4. We can't do anything with immutable dir (done in permission())
1825 * 5. If the sticky bit on dir is set we should either
1826 * a. be owner of dir, or
1827 * b. be owner of victim, or
1828 * c. have CAP_FOWNER capability
1829 * 6. If the victim is append-only or immutable we can't do antyhing with
1830 * links pointing to it.
1831 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1832 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1833 * 9. We can't remove a root or mountpoint.
1834 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1835 * nfs_async_unlink().
1837 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1841 if (!victim
->d_inode
)
1844 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1845 audit_inode_child(victim
, dir
);
1847 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1852 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1853 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1856 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1858 if (IS_ROOT(victim
))
1860 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1862 if (IS_DEADDIR(dir
))
1864 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1869 /* Check whether we can create an object with dentry child in directory
1871 * 1. We can't do it if child already exists (open has special treatment for
1872 * this case, but since we are inlined it's OK)
1873 * 2. We can't do it if dir is read-only (done in permission())
1874 * 3. We should have write and exec permissions on dir
1875 * 4. We can't do it if dir is immutable (done in permission())
1877 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1881 if (IS_DEADDIR(dir
))
1883 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1887 * p1 and p2 should be directories on the same fs.
1889 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1894 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1898 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1900 p
= d_ancestor(p2
, p1
);
1902 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1903 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1907 p
= d_ancestor(p1
, p2
);
1909 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1910 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1914 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1915 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1919 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1921 mutex_unlock(&p1
->d_inode
->i_mutex
);
1923 mutex_unlock(&p2
->d_inode
->i_mutex
);
1924 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1928 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1929 struct nameidata
*nd
)
1931 int error
= may_create(dir
, dentry
);
1936 if (!dir
->i_op
->create
)
1937 return -EACCES
; /* shouldn't it be ENOSYS? */
1940 error
= security_inode_create(dir
, dentry
, mode
);
1943 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1945 fsnotify_create(dir
, dentry
);
1949 static int may_open(struct path
*path
, int acc_mode
, int flag
)
1951 struct dentry
*dentry
= path
->dentry
;
1952 struct inode
*inode
= dentry
->d_inode
;
1962 switch (inode
->i_mode
& S_IFMT
) {
1966 if (acc_mode
& MAY_WRITE
)
1971 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
1980 error
= inode_permission(inode
, acc_mode
);
1985 * An append-only file must be opened in append mode for writing.
1987 if (IS_APPEND(inode
)) {
1988 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
1994 /* O_NOATIME can only be set by the owner or superuser */
1995 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
1999 * Ensure there are no outstanding leases on the file.
2001 return break_lease(inode
, flag
);
2004 static int handle_truncate(struct file
*filp
)
2006 struct path
*path
= &filp
->f_path
;
2007 struct inode
*inode
= path
->dentry
->d_inode
;
2008 int error
= get_write_access(inode
);
2012 * Refuse to truncate files with mandatory locks held on them.
2014 error
= locks_verify_locked(inode
);
2016 error
= security_path_truncate(path
);
2018 error
= do_truncate(path
->dentry
, 0,
2019 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2022 put_write_access(inode
);
2027 * Note that while the flag value (low two bits) for sys_open means:
2032 * it is changed into
2033 * 00 - no permissions needed
2034 * 01 - read-permission
2035 * 10 - write-permission
2037 * for the internal routines (ie open_namei()/follow_link() etc)
2038 * This is more logical, and also allows the 00 "no perm needed"
2039 * to be used for symlinks (where the permissions are checked
2043 static inline int open_to_namei_flags(int flag
)
2045 if ((flag
+1) & O_ACCMODE
)
2051 * Handle the last step of open()
2053 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2054 const struct open_flags
*op
, const char *pathname
)
2056 struct dentry
*dir
= nd
->path
.dentry
;
2057 struct dentry
*dentry
;
2058 int open_flag
= op
->open_flag
;
2059 int will_truncate
= open_flag
& O_TRUNC
;
2061 int acc_mode
= op
->acc_mode
;
2065 nd
->flags
&= ~LOOKUP_PARENT
;
2066 nd
->flags
|= op
->intent
;
2068 switch (nd
->last_type
) {
2071 error
= handle_dots(nd
, nd
->last_type
);
2073 return ERR_PTR(error
);
2076 error
= complete_walk(nd
);
2078 return ERR_PTR(error
);
2079 audit_inode(pathname
, nd
->path
.dentry
);
2080 if (open_flag
& O_CREAT
) {
2086 error
= complete_walk(nd
);
2088 return ERR_PTR(error
);
2089 audit_inode(pathname
, dir
);
2093 if (!(open_flag
& O_CREAT
)) {
2095 if (nd
->last
.name
[nd
->last
.len
])
2096 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2097 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2099 /* we _can_ be in RCU mode here */
2100 error
= walk_component(nd
, path
, &nd
->last
, LAST_NORM
,
2103 return ERR_PTR(error
);
2104 if (error
) /* symlink */
2107 error
= complete_walk(nd
);
2109 return ERR_PTR(-ECHILD
);
2112 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2113 if (!nd
->inode
->i_op
->lookup
)
2116 audit_inode(pathname
, nd
->path
.dentry
);
2120 /* create side of things */
2121 error
= complete_walk(nd
);
2123 return ERR_PTR(error
);
2125 audit_inode(pathname
, dir
);
2127 /* trailing slashes? */
2128 if (nd
->last
.name
[nd
->last
.len
])
2131 mutex_lock(&dir
->d_inode
->i_mutex
);
2133 dentry
= lookup_hash(nd
);
2134 error
= PTR_ERR(dentry
);
2135 if (IS_ERR(dentry
)) {
2136 mutex_unlock(&dir
->d_inode
->i_mutex
);
2140 path
->dentry
= dentry
;
2141 path
->mnt
= nd
->path
.mnt
;
2143 /* Negative dentry, just create the file */
2144 if (!dentry
->d_inode
) {
2145 int mode
= op
->mode
;
2146 if (!IS_POSIXACL(dir
->d_inode
))
2147 mode
&= ~current_umask();
2149 * This write is needed to ensure that a
2150 * rw->ro transition does not occur between
2151 * the time when the file is created and when
2152 * a permanent write count is taken through
2153 * the 'struct file' in nameidata_to_filp().
2155 error
= mnt_want_write(nd
->path
.mnt
);
2157 goto exit_mutex_unlock
;
2159 /* Don't check for write permission, don't truncate */
2160 open_flag
&= ~O_TRUNC
;
2162 acc_mode
= MAY_OPEN
;
2163 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2165 goto exit_mutex_unlock
;
2166 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2168 goto exit_mutex_unlock
;
2169 mutex_unlock(&dir
->d_inode
->i_mutex
);
2170 dput(nd
->path
.dentry
);
2171 nd
->path
.dentry
= dentry
;
2176 * It already exists.
2178 mutex_unlock(&dir
->d_inode
->i_mutex
);
2179 audit_inode(pathname
, path
->dentry
);
2182 if (open_flag
& O_EXCL
)
2185 error
= follow_managed(path
, nd
->flags
);
2190 if (!path
->dentry
->d_inode
)
2193 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2196 path_to_nameidata(path
, nd
);
2197 nd
->inode
= path
->dentry
->d_inode
;
2199 if (S_ISDIR(nd
->inode
->i_mode
))
2202 if (!S_ISREG(nd
->inode
->i_mode
))
2205 if (will_truncate
) {
2206 error
= mnt_want_write(nd
->path
.mnt
);
2212 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2215 filp
= nameidata_to_filp(nd
);
2216 if (!IS_ERR(filp
)) {
2217 error
= ima_file_check(filp
, op
->acc_mode
);
2220 filp
= ERR_PTR(error
);
2223 if (!IS_ERR(filp
)) {
2224 if (will_truncate
) {
2225 error
= handle_truncate(filp
);
2228 filp
= ERR_PTR(error
);
2234 mnt_drop_write(nd
->path
.mnt
);
2235 path_put(&nd
->path
);
2239 mutex_unlock(&dir
->d_inode
->i_mutex
);
2241 path_put_conditional(path
, nd
);
2243 filp
= ERR_PTR(error
);
2247 static struct file
*path_openat(int dfd
, const char *pathname
,
2248 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2250 struct file
*base
= NULL
;
2255 filp
= get_empty_filp();
2257 return ERR_PTR(-ENFILE
);
2259 filp
->f_flags
= op
->open_flag
;
2260 nd
->intent
.open
.file
= filp
;
2261 nd
->intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2262 nd
->intent
.open
.create_mode
= op
->mode
;
2264 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2265 if (unlikely(error
))
2268 current
->total_link_count
= 0;
2269 error
= link_path_walk(pathname
, nd
);
2270 if (unlikely(error
))
2273 filp
= do_last(nd
, &path
, op
, pathname
);
2274 while (unlikely(!filp
)) { /* trailing symlink */
2275 struct path link
= path
;
2277 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2278 path_put_conditional(&path
, nd
);
2279 path_put(&nd
->path
);
2280 filp
= ERR_PTR(-ELOOP
);
2283 nd
->flags
|= LOOKUP_PARENT
;
2284 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2285 error
= follow_link(&link
, nd
, &cookie
);
2286 if (unlikely(error
))
2287 filp
= ERR_PTR(error
);
2289 filp
= do_last(nd
, &path
, op
, pathname
);
2290 put_link(nd
, &link
, cookie
);
2293 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2294 path_put(&nd
->root
);
2297 release_open_intent(nd
);
2301 filp
= ERR_PTR(error
);
2305 struct file
*do_filp_open(int dfd
, const char *pathname
,
2306 const struct open_flags
*op
, int flags
)
2308 struct nameidata nd
;
2311 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2312 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2313 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2314 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2315 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2319 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2320 const char *name
, const struct open_flags
*op
, int flags
)
2322 struct nameidata nd
;
2326 nd
.root
.dentry
= dentry
;
2328 flags
|= LOOKUP_ROOT
;
2330 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2331 return ERR_PTR(-ELOOP
);
2333 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2334 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2335 file
= path_openat(-1, name
, &nd
, op
, flags
);
2336 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2337 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2342 * lookup_create - lookup a dentry, creating it if it doesn't exist
2343 * @nd: nameidata info
2344 * @is_dir: directory flag
2346 * Simple function to lookup and return a dentry and create it
2347 * if it doesn't exist. Is SMP-safe.
2349 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2351 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2353 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2355 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2357 * Yucky last component or no last component at all?
2358 * (foo/., foo/.., /////)
2360 if (nd
->last_type
!= LAST_NORM
)
2362 nd
->flags
&= ~LOOKUP_PARENT
;
2363 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2364 nd
->intent
.open
.flags
= O_EXCL
;
2367 * Do the final lookup.
2369 dentry
= lookup_hash(nd
);
2373 if (dentry
->d_inode
)
2376 * Special case - lookup gave negative, but... we had foo/bar/
2377 * From the vfs_mknod() POV we just have a negative dentry -
2378 * all is fine. Let's be bastards - you had / on the end, you've
2379 * been asking for (non-existent) directory. -ENOENT for you.
2381 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2383 dentry
= ERR_PTR(-ENOENT
);
2388 dentry
= ERR_PTR(-EEXIST
);
2392 EXPORT_SYMBOL_GPL(lookup_create
);
2394 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2396 int error
= may_create(dir
, dentry
);
2401 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) &&
2402 !ns_capable(inode_userns(dir
), CAP_MKNOD
))
2405 if (!dir
->i_op
->mknod
)
2408 error
= devcgroup_inode_mknod(mode
, dev
);
2412 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2416 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2418 fsnotify_create(dir
, dentry
);
2422 static int may_mknod(mode_t mode
)
2424 switch (mode
& S_IFMT
) {
2430 case 0: /* zero mode translates to S_IFREG */
2439 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2444 struct dentry
*dentry
;
2445 struct nameidata nd
;
2450 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2454 dentry
= lookup_create(&nd
, 0);
2455 if (IS_ERR(dentry
)) {
2456 error
= PTR_ERR(dentry
);
2459 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2460 mode
&= ~current_umask();
2461 error
= may_mknod(mode
);
2464 error
= mnt_want_write(nd
.path
.mnt
);
2467 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2469 goto out_drop_write
;
2470 switch (mode
& S_IFMT
) {
2471 case 0: case S_IFREG
:
2472 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2474 case S_IFCHR
: case S_IFBLK
:
2475 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2476 new_decode_dev(dev
));
2478 case S_IFIFO
: case S_IFSOCK
:
2479 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2483 mnt_drop_write(nd
.path
.mnt
);
2487 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2494 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2496 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2499 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2501 int error
= may_create(dir
, dentry
);
2506 if (!dir
->i_op
->mkdir
)
2509 mode
&= (S_IRWXUGO
|S_ISVTX
);
2510 error
= security_inode_mkdir(dir
, dentry
, mode
);
2514 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2516 fsnotify_mkdir(dir
, dentry
);
2520 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2524 struct dentry
*dentry
;
2525 struct nameidata nd
;
2527 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2531 dentry
= lookup_create(&nd
, 1);
2532 error
= PTR_ERR(dentry
);
2536 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2537 mode
&= ~current_umask();
2538 error
= mnt_want_write(nd
.path
.mnt
);
2541 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2543 goto out_drop_write
;
2544 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2546 mnt_drop_write(nd
.path
.mnt
);
2550 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2557 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2559 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2563 * The dentry_unhash() helper will try to drop the dentry early: we
2564 * should have a usage count of 2 if we're the only user of this
2565 * dentry, and if that is true (possibly after pruning the dcache),
2566 * then we drop the dentry now.
2568 * A low-level filesystem can, if it choses, legally
2571 * if (!d_unhashed(dentry))
2574 * if it cannot handle the case of removing a directory
2575 * that is still in use by something else..
2577 void dentry_unhash(struct dentry
*dentry
)
2579 shrink_dcache_parent(dentry
);
2580 spin_lock(&dentry
->d_lock
);
2581 if (dentry
->d_count
== 1)
2583 spin_unlock(&dentry
->d_lock
);
2586 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2588 int error
= may_delete(dir
, dentry
, 1);
2593 if (!dir
->i_op
->rmdir
)
2597 mutex_lock(&dentry
->d_inode
->i_mutex
);
2600 if (d_mountpoint(dentry
))
2603 error
= security_inode_rmdir(dir
, dentry
);
2607 shrink_dcache_parent(dentry
);
2608 error
= dir
->i_op
->rmdir(dir
, dentry
);
2612 dentry
->d_inode
->i_flags
|= S_DEAD
;
2616 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2623 static long do_rmdir(int dfd
, const char __user
*pathname
)
2627 struct dentry
*dentry
;
2628 struct nameidata nd
;
2630 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2634 switch(nd
.last_type
) {
2646 nd
.flags
&= ~LOOKUP_PARENT
;
2648 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2649 dentry
= lookup_hash(&nd
);
2650 error
= PTR_ERR(dentry
);
2653 if (!dentry
->d_inode
) {
2657 error
= mnt_want_write(nd
.path
.mnt
);
2660 error
= security_path_rmdir(&nd
.path
, dentry
);
2663 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2665 mnt_drop_write(nd
.path
.mnt
);
2669 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2676 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2678 return do_rmdir(AT_FDCWD
, pathname
);
2681 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2683 int error
= may_delete(dir
, dentry
, 0);
2688 if (!dir
->i_op
->unlink
)
2691 mutex_lock(&dentry
->d_inode
->i_mutex
);
2692 if (d_mountpoint(dentry
))
2695 error
= security_inode_unlink(dir
, dentry
);
2697 error
= dir
->i_op
->unlink(dir
, dentry
);
2702 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2704 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2705 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2706 fsnotify_link_count(dentry
->d_inode
);
2714 * Make sure that the actual truncation of the file will occur outside its
2715 * directory's i_mutex. Truncate can take a long time if there is a lot of
2716 * writeout happening, and we don't want to prevent access to the directory
2717 * while waiting on the I/O.
2719 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2723 struct dentry
*dentry
;
2724 struct nameidata nd
;
2725 struct inode
*inode
= NULL
;
2727 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2732 if (nd
.last_type
!= LAST_NORM
)
2735 nd
.flags
&= ~LOOKUP_PARENT
;
2737 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2738 dentry
= lookup_hash(&nd
);
2739 error
= PTR_ERR(dentry
);
2740 if (!IS_ERR(dentry
)) {
2741 /* Why not before? Because we want correct error value */
2742 if (nd
.last
.name
[nd
.last
.len
])
2744 inode
= dentry
->d_inode
;
2748 error
= mnt_want_write(nd
.path
.mnt
);
2751 error
= security_path_unlink(&nd
.path
, dentry
);
2754 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2756 mnt_drop_write(nd
.path
.mnt
);
2760 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2762 iput(inode
); /* truncate the inode here */
2769 error
= !dentry
->d_inode
? -ENOENT
:
2770 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2774 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2776 if ((flag
& ~AT_REMOVEDIR
) != 0)
2779 if (flag
& AT_REMOVEDIR
)
2780 return do_rmdir(dfd
, pathname
);
2782 return do_unlinkat(dfd
, pathname
);
2785 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2787 return do_unlinkat(AT_FDCWD
, pathname
);
2790 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2792 int error
= may_create(dir
, dentry
);
2797 if (!dir
->i_op
->symlink
)
2800 error
= security_inode_symlink(dir
, dentry
, oldname
);
2804 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2806 fsnotify_create(dir
, dentry
);
2810 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2811 int, newdfd
, const char __user
*, newname
)
2816 struct dentry
*dentry
;
2817 struct nameidata nd
;
2819 from
= getname(oldname
);
2821 return PTR_ERR(from
);
2823 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2827 dentry
= lookup_create(&nd
, 0);
2828 error
= PTR_ERR(dentry
);
2832 error
= mnt_want_write(nd
.path
.mnt
);
2835 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2837 goto out_drop_write
;
2838 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2840 mnt_drop_write(nd
.path
.mnt
);
2844 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2852 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2854 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2857 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2859 struct inode
*inode
= old_dentry
->d_inode
;
2865 error
= may_create(dir
, new_dentry
);
2869 if (dir
->i_sb
!= inode
->i_sb
)
2873 * A link to an append-only or immutable file cannot be created.
2875 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2877 if (!dir
->i_op
->link
)
2879 if (S_ISDIR(inode
->i_mode
))
2882 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2886 mutex_lock(&inode
->i_mutex
);
2887 /* Make sure we don't allow creating hardlink to an unlinked file */
2888 if (inode
->i_nlink
== 0)
2891 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2892 mutex_unlock(&inode
->i_mutex
);
2894 fsnotify_link(dir
, inode
, new_dentry
);
2899 * Hardlinks are often used in delicate situations. We avoid
2900 * security-related surprises by not following symlinks on the
2903 * We don't follow them on the oldname either to be compatible
2904 * with linux 2.0, and to avoid hard-linking to directories
2905 * and other special files. --ADM
2907 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2908 int, newdfd
, const char __user
*, newname
, int, flags
)
2910 struct dentry
*new_dentry
;
2911 struct nameidata nd
;
2912 struct path old_path
;
2917 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
2920 * To use null names we require CAP_DAC_READ_SEARCH
2921 * This ensures that not everyone will be able to create
2922 * handlink using the passed filedescriptor.
2924 if (flags
& AT_EMPTY_PATH
) {
2925 if (!capable(CAP_DAC_READ_SEARCH
))
2930 if (flags
& AT_SYMLINK_FOLLOW
)
2931 how
|= LOOKUP_FOLLOW
;
2933 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
2937 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2941 if (old_path
.mnt
!= nd
.path
.mnt
)
2943 new_dentry
= lookup_create(&nd
, 0);
2944 error
= PTR_ERR(new_dentry
);
2945 if (IS_ERR(new_dentry
))
2947 error
= mnt_want_write(nd
.path
.mnt
);
2950 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2952 goto out_drop_write
;
2953 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2955 mnt_drop_write(nd
.path
.mnt
);
2959 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2964 path_put(&old_path
);
2969 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
2971 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2975 * The worst of all namespace operations - renaming directory. "Perverted"
2976 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2978 * a) we can get into loop creation. Check is done in is_subdir().
2979 * b) race potential - two innocent renames can create a loop together.
2980 * That's where 4.4 screws up. Current fix: serialization on
2981 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2983 * c) we have to lock _three_ objects - parents and victim (if it exists).
2984 * And that - after we got ->i_mutex on parents (until then we don't know
2985 * whether the target exists). Solution: try to be smart with locking
2986 * order for inodes. We rely on the fact that tree topology may change
2987 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2988 * move will be locked. Thus we can rank directories by the tree
2989 * (ancestors first) and rank all non-directories after them.
2990 * That works since everybody except rename does "lock parent, lookup,
2991 * lock child" and rename is under ->s_vfs_rename_mutex.
2992 * HOWEVER, it relies on the assumption that any object with ->lookup()
2993 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2994 * we'd better make sure that there's no link(2) for them.
2995 * d) conversion from fhandle to dentry may come in the wrong moment - when
2996 * we are removing the target. Solution: we will have to grab ->i_mutex
2997 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2998 * ->i_mutex on parents, which works but leads to some truly excessive
3001 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3002 struct inode
*new_dir
, struct dentry
*new_dentry
)
3005 struct inode
*target
= new_dentry
->d_inode
;
3008 * If we are going to change the parent - check write permissions,
3009 * we'll need to flip '..'.
3011 if (new_dir
!= old_dir
) {
3012 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3017 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3023 mutex_lock(&target
->i_mutex
);
3026 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3030 shrink_dcache_parent(new_dentry
);
3031 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3036 target
->i_flags
|= S_DEAD
;
3037 dont_mount(new_dentry
);
3041 mutex_unlock(&target
->i_mutex
);
3044 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3045 d_move(old_dentry
,new_dentry
);
3049 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3050 struct inode
*new_dir
, struct dentry
*new_dentry
)
3052 struct inode
*target
= new_dentry
->d_inode
;
3055 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3061 mutex_lock(&target
->i_mutex
);
3064 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3067 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3072 dont_mount(new_dentry
);
3073 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3074 d_move(old_dentry
, new_dentry
);
3077 mutex_unlock(&target
->i_mutex
);
3082 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3083 struct inode
*new_dir
, struct dentry
*new_dentry
)
3086 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3087 const unsigned char *old_name
;
3089 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3092 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3096 if (!new_dentry
->d_inode
)
3097 error
= may_create(new_dir
, new_dentry
);
3099 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3103 if (!old_dir
->i_op
->rename
)
3106 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3109 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3111 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3113 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3114 new_dentry
->d_inode
, old_dentry
);
3115 fsnotify_oldname_free(old_name
);
3120 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3121 int, newdfd
, const char __user
*, newname
)
3123 struct dentry
*old_dir
, *new_dir
;
3124 struct dentry
*old_dentry
, *new_dentry
;
3125 struct dentry
*trap
;
3126 struct nameidata oldnd
, newnd
;
3131 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3135 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3140 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3143 old_dir
= oldnd
.path
.dentry
;
3145 if (oldnd
.last_type
!= LAST_NORM
)
3148 new_dir
= newnd
.path
.dentry
;
3149 if (newnd
.last_type
!= LAST_NORM
)
3152 oldnd
.flags
&= ~LOOKUP_PARENT
;
3153 newnd
.flags
&= ~LOOKUP_PARENT
;
3154 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3156 trap
= lock_rename(new_dir
, old_dir
);
3158 old_dentry
= lookup_hash(&oldnd
);
3159 error
= PTR_ERR(old_dentry
);
3160 if (IS_ERR(old_dentry
))
3162 /* source must exist */
3164 if (!old_dentry
->d_inode
)
3166 /* unless the source is a directory trailing slashes give -ENOTDIR */
3167 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3169 if (oldnd
.last
.name
[oldnd
.last
.len
])
3171 if (newnd
.last
.name
[newnd
.last
.len
])
3174 /* source should not be ancestor of target */
3176 if (old_dentry
== trap
)
3178 new_dentry
= lookup_hash(&newnd
);
3179 error
= PTR_ERR(new_dentry
);
3180 if (IS_ERR(new_dentry
))
3182 /* target should not be an ancestor of source */
3184 if (new_dentry
== trap
)
3187 error
= mnt_want_write(oldnd
.path
.mnt
);
3190 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3191 &newnd
.path
, new_dentry
);
3194 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3195 new_dir
->d_inode
, new_dentry
);
3197 mnt_drop_write(oldnd
.path
.mnt
);
3203 unlock_rename(new_dir
, old_dir
);
3205 path_put(&newnd
.path
);
3208 path_put(&oldnd
.path
);
3214 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3216 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3219 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3223 len
= PTR_ERR(link
);
3228 if (len
> (unsigned) buflen
)
3230 if (copy_to_user(buffer
, link
, len
))
3237 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3238 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3239 * using) it for any given inode is up to filesystem.
3241 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3243 struct nameidata nd
;
3248 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3250 return PTR_ERR(cookie
);
3252 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3253 if (dentry
->d_inode
->i_op
->put_link
)
3254 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3258 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3260 return __vfs_follow_link(nd
, link
);
3263 /* get the link contents into pagecache */
3264 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3268 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3269 page
= read_mapping_page(mapping
, 0, NULL
);
3274 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3278 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3280 struct page
*page
= NULL
;
3281 char *s
= page_getlink(dentry
, &page
);
3282 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3285 page_cache_release(page
);
3290 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3292 struct page
*page
= NULL
;
3293 nd_set_link(nd
, page_getlink(dentry
, &page
));
3297 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3299 struct page
*page
= cookie
;
3303 page_cache_release(page
);
3308 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3310 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3312 struct address_space
*mapping
= inode
->i_mapping
;
3317 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3319 flags
|= AOP_FLAG_NOFS
;
3322 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3323 flags
, &page
, &fsdata
);
3327 kaddr
= kmap_atomic(page
, KM_USER0
);
3328 memcpy(kaddr
, symname
, len
-1);
3329 kunmap_atomic(kaddr
, KM_USER0
);
3331 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3338 mark_inode_dirty(inode
);
3344 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3346 return __page_symlink(inode
, symname
, len
,
3347 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3350 const struct inode_operations page_symlink_inode_operations
= {
3351 .readlink
= generic_readlink
,
3352 .follow_link
= page_follow_link_light
,
3353 .put_link
= page_put_link
,
3356 EXPORT_SYMBOL(user_path_at
);
3357 EXPORT_SYMBOL(follow_down_one
);
3358 EXPORT_SYMBOL(follow_down
);
3359 EXPORT_SYMBOL(follow_up
);
3360 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3361 EXPORT_SYMBOL(getname
);
3362 EXPORT_SYMBOL(lock_rename
);
3363 EXPORT_SYMBOL(lookup_one_len
);
3364 EXPORT_SYMBOL(page_follow_link_light
);
3365 EXPORT_SYMBOL(page_put_link
);
3366 EXPORT_SYMBOL(page_readlink
);
3367 EXPORT_SYMBOL(__page_symlink
);
3368 EXPORT_SYMBOL(page_symlink
);
3369 EXPORT_SYMBOL(page_symlink_inode_operations
);
3370 EXPORT_SYMBOL(kern_path_parent
);
3371 EXPORT_SYMBOL(kern_path
);
3372 EXPORT_SYMBOL(vfs_path_lookup
);
3373 EXPORT_SYMBOL(inode_permission
);
3374 EXPORT_SYMBOL(file_permission
);
3375 EXPORT_SYMBOL(unlock_rename
);
3376 EXPORT_SYMBOL(vfs_create
);
3377 EXPORT_SYMBOL(vfs_follow_link
);
3378 EXPORT_SYMBOL(vfs_link
);
3379 EXPORT_SYMBOL(vfs_mkdir
);
3380 EXPORT_SYMBOL(vfs_mknod
);
3381 EXPORT_SYMBOL(generic_permission
);
3382 EXPORT_SYMBOL(vfs_readlink
);
3383 EXPORT_SYMBOL(vfs_rename
);
3384 EXPORT_SYMBOL(vfs_rmdir
);
3385 EXPORT_SYMBOL(vfs_symlink
);
3386 EXPORT_SYMBOL(vfs_unlink
);
3387 EXPORT_SYMBOL(dentry_unhash
);
3388 EXPORT_SYMBOL(generic_readlink
);