4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user
*filename
, char *page
)
120 unsigned long len
= PATH_MAX
;
122 if (!segment_eq(get_fs(), KERNEL_DS
)) {
123 if ((unsigned long) filename
>= TASK_SIZE
)
125 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
126 len
= TASK_SIZE
- (unsigned long) filename
;
129 retval
= strncpy_from_user(page
, filename
, len
);
133 return -ENAMETOOLONG
;
139 static char *getname_flags(const char __user
* filename
, int flags
)
143 result
= ERR_PTR(-ENOMEM
);
146 int retval
= do_getname(filename
, tmp
);
150 if (retval
!= -ENOENT
|| !(flags
& LOOKUP_EMPTY
)) {
152 result
= ERR_PTR(retval
);
156 audit_getname(result
);
160 char *getname(const char __user
* filename
)
162 return getname_flags(filename
, 0);
165 #ifdef CONFIG_AUDITSYSCALL
166 void putname(const char *name
)
168 if (unlikely(!audit_dummy_context()))
173 EXPORT_SYMBOL(putname
);
177 * This does basic POSIX ACL permission checking
179 static int acl_permission_check(struct inode
*inode
, int mask
, unsigned int flags
,
180 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
182 umode_t mode
= inode
->i_mode
;
184 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
186 if (current_fsuid() == inode
->i_uid
)
189 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
190 int error
= check_acl(inode
, mask
, flags
);
191 if (error
!= -EAGAIN
)
195 if (in_group_p(inode
->i_gid
))
200 * If the DACs are ok we don't need any capability check.
202 if ((mask
& ~mode
) == 0)
208 * generic_permission - check for access rights on a Posix-like filesystem
209 * @inode: inode to check access rights for
210 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
211 * @check_acl: optional callback to check for Posix ACLs
212 * @flags: IPERM_FLAG_ flags.
214 * Used to check for read/write/execute permissions on a file.
215 * We use "fsuid" for this, letting us set arbitrary permissions
216 * for filesystem access without changing the "normal" uids which
217 * are used for other things.
219 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
220 * request cannot be satisfied (eg. requires blocking or too much complexity).
221 * It would then be called again in ref-walk mode.
223 int generic_permission(struct inode
*inode
, int mask
, unsigned int flags
,
224 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
229 * Do the basic POSIX ACL permission checks.
231 ret
= acl_permission_check(inode
, mask
, flags
, check_acl
);
236 * Read/write DACs are always overridable.
237 * Executable DACs are overridable if at least one exec bit is set.
239 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
240 if (capable(CAP_DAC_OVERRIDE
))
244 * Searching includes executable on directories, else just read.
246 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
247 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
248 if (capable(CAP_DAC_READ_SEARCH
))
255 * inode_permission - check for access rights to a given inode
256 * @inode: inode to check permission on
257 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
259 * Used to check for read/write/execute permissions on an inode.
260 * We use "fsuid" for this, letting us set arbitrary permissions
261 * for filesystem access without changing the "normal" uids which
262 * are used for other things.
264 int inode_permission(struct inode
*inode
, int mask
)
268 if (mask
& MAY_WRITE
) {
269 umode_t mode
= inode
->i_mode
;
272 * Nobody gets write access to a read-only fs.
274 if (IS_RDONLY(inode
) &&
275 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
279 * Nobody gets write access to an immutable file.
281 if (IS_IMMUTABLE(inode
))
285 if (inode
->i_op
->permission
)
286 retval
= inode
->i_op
->permission(inode
, mask
, 0);
288 retval
= generic_permission(inode
, mask
, 0,
289 inode
->i_op
->check_acl
);
294 retval
= devcgroup_inode_permission(inode
, mask
);
298 return security_inode_permission(inode
, mask
);
302 * file_permission - check for additional access rights to a given file
303 * @file: file to check access rights for
304 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
306 * Used to check for read/write/execute permissions on an already opened
310 * Do not use this function in new code. All access checks should
311 * be done using inode_permission().
313 int file_permission(struct file
*file
, int mask
)
315 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
319 * get_write_access() gets write permission for a file.
320 * put_write_access() releases this write permission.
321 * This is used for regular files.
322 * We cannot support write (and maybe mmap read-write shared) accesses and
323 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
324 * can have the following values:
325 * 0: no writers, no VM_DENYWRITE mappings
326 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
327 * > 0: (i_writecount) users are writing to the file.
329 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
330 * except for the cases where we don't hold i_writecount yet. Then we need to
331 * use {get,deny}_write_access() - these functions check the sign and refuse
332 * to do the change if sign is wrong. Exclusion between them is provided by
333 * the inode->i_lock spinlock.
336 int get_write_access(struct inode
* inode
)
338 spin_lock(&inode
->i_lock
);
339 if (atomic_read(&inode
->i_writecount
) < 0) {
340 spin_unlock(&inode
->i_lock
);
343 atomic_inc(&inode
->i_writecount
);
344 spin_unlock(&inode
->i_lock
);
349 int deny_write_access(struct file
* file
)
351 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
353 spin_lock(&inode
->i_lock
);
354 if (atomic_read(&inode
->i_writecount
) > 0) {
355 spin_unlock(&inode
->i_lock
);
358 atomic_dec(&inode
->i_writecount
);
359 spin_unlock(&inode
->i_lock
);
365 * path_get - get a reference to a path
366 * @path: path to get the reference to
368 * Given a path increment the reference count to the dentry and the vfsmount.
370 void path_get(struct path
*path
)
375 EXPORT_SYMBOL(path_get
);
378 * path_put - put a reference to a path
379 * @path: path to put the reference to
381 * Given a path decrement the reference count to the dentry and the vfsmount.
383 void path_put(struct path
*path
)
388 EXPORT_SYMBOL(path_put
);
391 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
392 * @nd: nameidata pathwalk data to drop
393 * Returns: 0 on success, -ECHILD on failure
395 * Path walking has 2 modes, rcu-walk and ref-walk (see
396 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
397 * to drop out of rcu-walk mode and take normal reference counts on dentries
398 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
399 * refcounts at the last known good point before rcu-walk got stuck, so
400 * ref-walk may continue from there. If this is not successful (eg. a seqcount
401 * has changed), then failure is returned and path walk restarts from the
402 * beginning in ref-walk mode.
404 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
405 * ref-walk. Must be called from rcu-walk context.
407 static int nameidata_drop_rcu(struct nameidata
*nd
)
409 struct fs_struct
*fs
= current
->fs
;
410 struct dentry
*dentry
= nd
->path
.dentry
;
413 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
414 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
416 spin_lock(&fs
->lock
);
417 if (nd
->root
.mnt
!= fs
->root
.mnt
||
418 nd
->root
.dentry
!= fs
->root
.dentry
)
421 spin_lock(&dentry
->d_lock
);
422 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
424 BUG_ON(nd
->inode
!= dentry
->d_inode
);
425 spin_unlock(&dentry
->d_lock
);
428 spin_unlock(&fs
->lock
);
430 mntget(nd
->path
.mnt
);
433 br_read_unlock(vfsmount_lock
);
434 nd
->flags
&= ~LOOKUP_RCU
;
437 spin_unlock(&dentry
->d_lock
);
440 spin_unlock(&fs
->lock
);
444 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
445 static inline int nameidata_drop_rcu_maybe(struct nameidata
*nd
)
447 if (nd
->flags
& LOOKUP_RCU
)
448 return nameidata_drop_rcu(nd
);
453 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
454 * @nd: nameidata pathwalk data to drop
455 * @dentry: dentry to drop
456 * Returns: 0 on success, -ECHILD on failure
458 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
459 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
460 * @nd. Must be called from rcu-walk context.
462 static int nameidata_dentry_drop_rcu(struct nameidata
*nd
, struct dentry
*dentry
)
464 struct fs_struct
*fs
= current
->fs
;
465 struct dentry
*parent
= nd
->path
.dentry
;
468 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
469 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
471 spin_lock(&fs
->lock
);
472 if (nd
->root
.mnt
!= fs
->root
.mnt
||
473 nd
->root
.dentry
!= fs
->root
.dentry
)
476 spin_lock(&parent
->d_lock
);
477 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
478 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
481 * If the sequence check on the child dentry passed, then the child has
482 * not been removed from its parent. This means the parent dentry must
483 * be valid and able to take a reference at this point.
485 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
486 BUG_ON(!parent
->d_count
);
488 spin_unlock(&dentry
->d_lock
);
489 spin_unlock(&parent
->d_lock
);
492 spin_unlock(&fs
->lock
);
494 mntget(nd
->path
.mnt
);
497 br_read_unlock(vfsmount_lock
);
498 nd
->flags
&= ~LOOKUP_RCU
;
501 spin_unlock(&dentry
->d_lock
);
502 spin_unlock(&parent
->d_lock
);
505 spin_unlock(&fs
->lock
);
509 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
510 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata
*nd
, struct dentry
*dentry
)
512 if (nd
->flags
& LOOKUP_RCU
) {
513 if (unlikely(nameidata_dentry_drop_rcu(nd
, dentry
))) {
514 nd
->flags
&= ~LOOKUP_RCU
;
515 if (!(nd
->flags
& LOOKUP_ROOT
))
518 br_read_unlock(vfsmount_lock
);
526 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
527 * @nd: nameidata pathwalk data to drop
528 * Returns: 0 on success, -ECHILD on failure
530 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
531 * nd->path should be the final element of the lookup, so nd->root is discarded.
532 * Must be called from rcu-walk context.
534 static int nameidata_drop_rcu_last(struct nameidata
*nd
)
536 struct dentry
*dentry
= nd
->path
.dentry
;
538 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
539 nd
->flags
&= ~LOOKUP_RCU
;
540 if (!(nd
->flags
& LOOKUP_ROOT
))
542 spin_lock(&dentry
->d_lock
);
543 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
545 BUG_ON(nd
->inode
!= dentry
->d_inode
);
546 spin_unlock(&dentry
->d_lock
);
548 mntget(nd
->path
.mnt
);
551 br_read_unlock(vfsmount_lock
);
556 spin_unlock(&dentry
->d_lock
);
558 br_read_unlock(vfsmount_lock
);
563 * release_open_intent - free up open intent resources
564 * @nd: pointer to nameidata
566 void release_open_intent(struct nameidata
*nd
)
568 struct file
*file
= nd
->intent
.open
.file
;
570 if (file
&& !IS_ERR(file
)) {
571 if (file
->f_path
.dentry
== NULL
)
578 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
580 return dentry
->d_op
->d_revalidate(dentry
, nd
);
583 static struct dentry
*
584 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
586 int status
= d_revalidate(dentry
, nd
);
587 if (unlikely(status
<= 0)) {
589 * The dentry failed validation.
590 * If d_revalidate returned 0 attempt to invalidate
591 * the dentry otherwise d_revalidate is asking us
592 * to return a fail status.
596 dentry
= ERR_PTR(status
);
597 } else if (!d_invalidate(dentry
)) {
606 * handle_reval_path - force revalidation of a dentry
608 * In some situations the path walking code will trust dentries without
609 * revalidating them. This causes problems for filesystems that depend on
610 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
611 * (which indicates that it's possible for the dentry to go stale), force
612 * a d_revalidate call before proceeding.
614 * Returns 0 if the revalidation was successful. If the revalidation fails,
615 * either return the error returned by d_revalidate or -ESTALE if the
616 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
617 * invalidate the dentry. It's up to the caller to handle putting references
618 * to the path if necessary.
620 static inline int handle_reval_path(struct nameidata
*nd
)
622 struct dentry
*dentry
= nd
->path
.dentry
;
625 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
628 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
631 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
634 /* Note: we do not d_invalidate() */
635 status
= d_revalidate(dentry
, nd
);
646 * Short-cut version of permission(), for calling on directories
647 * during pathname resolution. Combines parts of permission()
648 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
650 * If appropriate, check DAC only. If not appropriate, or
651 * short-cut DAC fails, then call ->permission() to do more
652 * complete permission check.
654 static inline int exec_permission(struct inode
*inode
, unsigned int flags
)
658 if (inode
->i_op
->permission
) {
659 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
, flags
);
661 ret
= acl_permission_check(inode
, MAY_EXEC
, flags
,
662 inode
->i_op
->check_acl
);
669 if (capable(CAP_DAC_OVERRIDE
) || capable(CAP_DAC_READ_SEARCH
))
674 return security_inode_exec_permission(inode
, flags
);
677 static __always_inline
void set_root(struct nameidata
*nd
)
680 get_fs_root(current
->fs
, &nd
->root
);
683 static int link_path_walk(const char *, struct nameidata
*);
685 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
688 struct fs_struct
*fs
= current
->fs
;
692 seq
= read_seqcount_begin(&fs
->seq
);
694 } while (read_seqcount_retry(&fs
->seq
, seq
));
698 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
710 nd
->flags
|= LOOKUP_JUMPED
;
712 nd
->inode
= nd
->path
.dentry
->d_inode
;
714 ret
= link_path_walk(link
, nd
);
718 return PTR_ERR(link
);
721 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
724 if (path
->mnt
!= nd
->path
.mnt
)
728 static inline void path_to_nameidata(const struct path
*path
,
729 struct nameidata
*nd
)
731 if (!(nd
->flags
& LOOKUP_RCU
)) {
732 dput(nd
->path
.dentry
);
733 if (nd
->path
.mnt
!= path
->mnt
)
734 mntput(nd
->path
.mnt
);
736 nd
->path
.mnt
= path
->mnt
;
737 nd
->path
.dentry
= path
->dentry
;
740 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
742 struct inode
*inode
= link
->dentry
->d_inode
;
743 if (!IS_ERR(cookie
) && inode
->i_op
->put_link
)
744 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
748 static __always_inline
int
749 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
752 struct dentry
*dentry
= link
->dentry
;
754 BUG_ON(nd
->flags
& LOOKUP_RCU
);
756 if (link
->mnt
== nd
->path
.mnt
)
759 if (unlikely(current
->total_link_count
>= 40)) {
760 *p
= ERR_PTR(-ELOOP
); /* no ->put_link(), please */
765 current
->total_link_count
++;
767 touch_atime(link
->mnt
, dentry
);
768 nd_set_link(nd
, NULL
);
770 error
= security_inode_follow_link(link
->dentry
, nd
);
772 *p
= ERR_PTR(error
); /* no ->put_link(), please */
777 nd
->last_type
= LAST_BIND
;
778 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
781 char *s
= nd_get_link(nd
);
784 error
= __vfs_follow_link(nd
, s
);
785 else if (nd
->last_type
== LAST_BIND
) {
786 nd
->flags
|= LOOKUP_JUMPED
;
787 nd
->inode
= nd
->path
.dentry
->d_inode
;
788 if (nd
->inode
->i_op
->follow_link
) {
789 /* stepped on a _really_ weird one */
798 static int follow_up_rcu(struct path
*path
)
800 struct vfsmount
*parent
;
801 struct dentry
*mountpoint
;
803 parent
= path
->mnt
->mnt_parent
;
804 if (parent
== path
->mnt
)
806 mountpoint
= path
->mnt
->mnt_mountpoint
;
807 path
->dentry
= mountpoint
;
812 int follow_up(struct path
*path
)
814 struct vfsmount
*parent
;
815 struct dentry
*mountpoint
;
817 br_read_lock(vfsmount_lock
);
818 parent
= path
->mnt
->mnt_parent
;
819 if (parent
== path
->mnt
) {
820 br_read_unlock(vfsmount_lock
);
824 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
825 br_read_unlock(vfsmount_lock
);
827 path
->dentry
= mountpoint
;
834 * Perform an automount
835 * - return -EISDIR to tell follow_managed() to stop and return the path we
838 static int follow_automount(struct path
*path
, unsigned flags
,
841 struct vfsmount
*mnt
;
844 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
847 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
848 * and this is the terminal part of the path.
850 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
851 return -EISDIR
; /* we actually want to stop here */
853 /* We want to mount if someone is trying to open/create a file of any
854 * type under the mountpoint, wants to traverse through the mountpoint
855 * or wants to open the mounted directory.
857 * We don't want to mount if someone's just doing a stat and they've
858 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
859 * appended a '/' to the name.
861 if (!(flags
& LOOKUP_FOLLOW
) &&
862 !(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
863 LOOKUP_OPEN
| LOOKUP_CREATE
)))
866 current
->total_link_count
++;
867 if (current
->total_link_count
>= 40)
870 mnt
= path
->dentry
->d_op
->d_automount(path
);
873 * The filesystem is allowed to return -EISDIR here to indicate
874 * it doesn't want to automount. For instance, autofs would do
875 * this so that its userspace daemon can mount on this dentry.
877 * However, we can only permit this if it's a terminal point in
878 * the path being looked up; if it wasn't then the remainder of
879 * the path is inaccessible and we should say so.
881 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_CONTINUE
))
886 if (!mnt
) /* mount collision */
889 err
= finish_automount(mnt
, path
);
893 /* Someone else made a mount here whilst we were busy */
900 path
->dentry
= dget(mnt
->mnt_root
);
910 * Handle a dentry that is managed in some way.
911 * - Flagged for transit management (autofs)
912 * - Flagged as mountpoint
913 * - Flagged as automount point
915 * This may only be called in refwalk mode.
917 * Serialization is taken care of in namespace.c
919 static int follow_managed(struct path
*path
, unsigned flags
)
922 bool need_mntput
= false;
925 /* Given that we're not holding a lock here, we retain the value in a
926 * local variable for each dentry as we look at it so that we don't see
927 * the components of that value change under us */
928 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
929 managed
&= DCACHE_MANAGED_DENTRY
,
930 unlikely(managed
!= 0)) {
931 /* Allow the filesystem to manage the transit without i_mutex
933 if (managed
& DCACHE_MANAGE_TRANSIT
) {
934 BUG_ON(!path
->dentry
->d_op
);
935 BUG_ON(!path
->dentry
->d_op
->d_manage
);
936 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
,
939 return ret
== -EISDIR
? 0 : ret
;
942 /* Transit to a mounted filesystem. */
943 if (managed
& DCACHE_MOUNTED
) {
944 struct vfsmount
*mounted
= lookup_mnt(path
);
950 path
->dentry
= dget(mounted
->mnt_root
);
955 /* Something is mounted on this dentry in another
956 * namespace and/or whatever was mounted there in this
957 * namespace got unmounted before we managed to get the
961 /* Handle an automount point */
962 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
963 ret
= follow_automount(path
, flags
, &need_mntput
);
965 return ret
== -EISDIR
? 0 : ret
;
969 /* We didn't change the current path point */
975 int follow_down_one(struct path
*path
)
977 struct vfsmount
*mounted
;
979 mounted
= lookup_mnt(path
);
984 path
->dentry
= dget(mounted
->mnt_root
);
991 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
992 * meet a managed dentry and we're not walking to "..". True is returned to
993 * continue, false to abort.
995 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
996 struct inode
**inode
, bool reverse_transit
)
998 while (d_mountpoint(path
->dentry
)) {
999 struct vfsmount
*mounted
;
1000 if (unlikely(path
->dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) &&
1002 path
->dentry
->d_op
->d_manage(path
->dentry
, false, true) < 0)
1004 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1007 path
->mnt
= mounted
;
1008 path
->dentry
= mounted
->mnt_root
;
1009 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1010 *inode
= path
->dentry
->d_inode
;
1013 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1014 return reverse_transit
;
1018 static int follow_dotdot_rcu(struct nameidata
*nd
)
1020 struct inode
*inode
= nd
->inode
;
1025 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1026 nd
->path
.mnt
== nd
->root
.mnt
) {
1029 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1030 struct dentry
*old
= nd
->path
.dentry
;
1031 struct dentry
*parent
= old
->d_parent
;
1034 seq
= read_seqcount_begin(&parent
->d_seq
);
1035 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1037 inode
= parent
->d_inode
;
1038 nd
->path
.dentry
= parent
;
1042 if (!follow_up_rcu(&nd
->path
))
1044 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1045 inode
= nd
->path
.dentry
->d_inode
;
1047 __follow_mount_rcu(nd
, &nd
->path
, &inode
, true);
1052 nd
->flags
&= ~LOOKUP_RCU
;
1053 if (!(nd
->flags
& LOOKUP_ROOT
))
1054 nd
->root
.mnt
= NULL
;
1056 br_read_unlock(vfsmount_lock
);
1061 * Follow down to the covering mount currently visible to userspace. At each
1062 * point, the filesystem owning that dentry may be queried as to whether the
1063 * caller is permitted to proceed or not.
1065 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1068 int follow_down(struct path
*path
, bool mounting_here
)
1073 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1074 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1075 /* Allow the filesystem to manage the transit without i_mutex
1078 * We indicate to the filesystem if someone is trying to mount
1079 * something here. This gives autofs the chance to deny anyone
1080 * other than its daemon the right to mount on its
1083 * The filesystem may sleep at this point.
1085 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1086 BUG_ON(!path
->dentry
->d_op
);
1087 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1088 ret
= path
->dentry
->d_op
->d_manage(
1089 path
->dentry
, mounting_here
, false);
1091 return ret
== -EISDIR
? 0 : ret
;
1094 /* Transit to a mounted filesystem. */
1095 if (managed
& DCACHE_MOUNTED
) {
1096 struct vfsmount
*mounted
= lookup_mnt(path
);
1101 path
->mnt
= mounted
;
1102 path
->dentry
= dget(mounted
->mnt_root
);
1106 /* Don't handle automount points here */
1113 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1115 static void follow_mount(struct path
*path
)
1117 while (d_mountpoint(path
->dentry
)) {
1118 struct vfsmount
*mounted
= lookup_mnt(path
);
1123 path
->mnt
= mounted
;
1124 path
->dentry
= dget(mounted
->mnt_root
);
1128 static void follow_dotdot(struct nameidata
*nd
)
1133 struct dentry
*old
= nd
->path
.dentry
;
1135 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1136 nd
->path
.mnt
== nd
->root
.mnt
) {
1139 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1140 /* rare case of legitimate dget_parent()... */
1141 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1145 if (!follow_up(&nd
->path
))
1148 follow_mount(&nd
->path
);
1149 nd
->inode
= nd
->path
.dentry
->d_inode
;
1153 * Allocate a dentry with name and parent, and perform a parent
1154 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1155 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1156 * have verified that no child exists while under i_mutex.
1158 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1159 struct qstr
*name
, struct nameidata
*nd
)
1161 struct inode
*inode
= parent
->d_inode
;
1162 struct dentry
*dentry
;
1165 /* Don't create child dentry for a dead directory. */
1166 if (unlikely(IS_DEADDIR(inode
)))
1167 return ERR_PTR(-ENOENT
);
1169 dentry
= d_alloc(parent
, name
);
1170 if (unlikely(!dentry
))
1171 return ERR_PTR(-ENOMEM
);
1173 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1174 if (unlikely(old
)) {
1182 * It's more convoluted than I'd like it to be, but... it's still fairly
1183 * small and for now I'd prefer to have fast path as straight as possible.
1184 * It _is_ time-critical.
1186 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1187 struct path
*path
, struct inode
**inode
)
1189 struct vfsmount
*mnt
= nd
->path
.mnt
;
1190 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1196 * Rename seqlock is not required here because in the off chance
1197 * of a false negative due to a concurrent rename, we're going to
1198 * do the non-racy lookup, below.
1200 if (nd
->flags
& LOOKUP_RCU
) {
1203 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1207 /* Memory barrier in read_seqcount_begin of child is enough */
1208 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1212 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1213 status
= d_revalidate(dentry
, nd
);
1214 if (unlikely(status
<= 0)) {
1215 if (status
!= -ECHILD
)
1221 path
->dentry
= dentry
;
1222 if (likely(__follow_mount_rcu(nd
, path
, inode
, false)))
1226 if (nameidata_dentry_drop_rcu(nd
, dentry
))
1229 if (nameidata_drop_rcu(nd
))
1233 dentry
= __d_lookup(parent
, name
);
1237 if (unlikely(!dentry
)) {
1238 struct inode
*dir
= parent
->d_inode
;
1239 BUG_ON(nd
->inode
!= dir
);
1241 mutex_lock(&dir
->i_mutex
);
1242 dentry
= d_lookup(parent
, name
);
1243 if (likely(!dentry
)) {
1244 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1245 if (IS_ERR(dentry
)) {
1246 mutex_unlock(&dir
->i_mutex
);
1247 return PTR_ERR(dentry
);
1253 mutex_unlock(&dir
->i_mutex
);
1255 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1256 status
= d_revalidate(dentry
, nd
);
1257 if (unlikely(status
<= 0)) {
1262 if (!d_invalidate(dentry
)) {
1271 path
->dentry
= dentry
;
1272 err
= follow_managed(path
, nd
->flags
);
1273 if (unlikely(err
< 0)) {
1274 path_put_conditional(path
, nd
);
1277 *inode
= path
->dentry
->d_inode
;
1281 static inline int may_lookup(struct nameidata
*nd
)
1283 if (nd
->flags
& LOOKUP_RCU
) {
1284 int err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1287 if (nameidata_drop_rcu(nd
))
1290 return exec_permission(nd
->inode
, 0);
1293 static inline int handle_dots(struct nameidata
*nd
, int type
)
1295 if (type
== LAST_DOTDOT
) {
1296 if (nd
->flags
& LOOKUP_RCU
) {
1297 if (follow_dotdot_rcu(nd
))
1305 static void terminate_walk(struct nameidata
*nd
)
1307 if (!(nd
->flags
& LOOKUP_RCU
)) {
1308 path_put(&nd
->path
);
1310 nd
->flags
&= ~LOOKUP_RCU
;
1311 if (!(nd
->flags
& LOOKUP_ROOT
))
1312 nd
->root
.mnt
= NULL
;
1314 br_read_unlock(vfsmount_lock
);
1318 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1319 struct qstr
*name
, int type
, int follow
)
1321 struct inode
*inode
;
1324 * "." and ".." are special - ".." especially so because it has
1325 * to be able to know about the current root directory and
1326 * parent relationships.
1328 if (unlikely(type
!= LAST_NORM
))
1329 return handle_dots(nd
, type
);
1330 err
= do_lookup(nd
, name
, path
, &inode
);
1331 if (unlikely(err
)) {
1336 path_to_nameidata(path
, nd
);
1340 if (unlikely(inode
->i_op
->follow_link
) && follow
) {
1341 if (nameidata_dentry_drop_rcu_maybe(nd
, path
->dentry
))
1343 BUG_ON(inode
!= path
->dentry
->d_inode
);
1346 path_to_nameidata(path
, nd
);
1352 * This limits recursive symlink follows to 8, while
1353 * limiting consecutive symlinks to 40.
1355 * Without that kind of total limit, nasty chains of consecutive
1356 * symlinks can cause almost arbitrarily long lookups.
1358 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1362 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1363 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1364 path_put_conditional(path
, nd
);
1365 path_put(&nd
->path
);
1370 current
->link_count
++;
1373 struct path link
= *path
;
1376 res
= follow_link(&link
, nd
, &cookie
);
1378 res
= walk_component(nd
, path
, &nd
->last
,
1379 nd
->last_type
, LOOKUP_FOLLOW
);
1380 put_link(nd
, &link
, cookie
);
1383 current
->link_count
--;
1390 * This is the basic name resolution function, turning a pathname into
1391 * the final dentry. We expect 'base' to be positive and a directory.
1393 * Returns 0 and nd will have valid dentry and mnt on success.
1394 * Returns error and drops reference to input namei data on failure.
1396 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1400 unsigned int lookup_flags
= nd
->flags
;
1407 /* At this point we know we have a real path component. */
1414 nd
->flags
|= LOOKUP_CONTINUE
;
1416 err
= may_lookup(nd
);
1421 c
= *(const unsigned char *)name
;
1423 hash
= init_name_hash();
1426 hash
= partial_name_hash(c
, hash
);
1427 c
= *(const unsigned char *)name
;
1428 } while (c
&& (c
!= '/'));
1429 this.len
= name
- (const char *) this.name
;
1430 this.hash
= end_name_hash(hash
);
1433 if (this.name
[0] == '.') switch (this.len
) {
1435 if (this.name
[1] == '.') {
1437 nd
->flags
|= LOOKUP_JUMPED
;
1443 if (likely(type
== LAST_NORM
)) {
1444 struct dentry
*parent
= nd
->path
.dentry
;
1445 nd
->flags
&= ~LOOKUP_JUMPED
;
1446 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1447 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1454 /* remove trailing slashes? */
1456 goto last_component
;
1457 while (*++name
== '/');
1459 goto last_component
;
1461 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1466 err
= nested_symlink(&next
, nd
);
1471 if (!nd
->inode
->i_op
->lookup
)
1474 /* here ends the main loop */
1477 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1478 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1480 nd
->last_type
= type
;
1487 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1488 struct nameidata
*nd
, struct file
**fp
)
1494 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1495 nd
->flags
= flags
| LOOKUP_JUMPED
;
1497 if (flags
& LOOKUP_ROOT
) {
1498 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1500 if (!inode
->i_op
->lookup
)
1502 retval
= inode_permission(inode
, MAY_EXEC
);
1506 nd
->path
= nd
->root
;
1508 if (flags
& LOOKUP_RCU
) {
1509 br_read_lock(vfsmount_lock
);
1511 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1513 path_get(&nd
->path
);
1518 nd
->root
.mnt
= NULL
;
1521 if (flags
& LOOKUP_RCU
) {
1522 br_read_lock(vfsmount_lock
);
1527 path_get(&nd
->root
);
1529 nd
->path
= nd
->root
;
1530 } else if (dfd
== AT_FDCWD
) {
1531 if (flags
& LOOKUP_RCU
) {
1532 struct fs_struct
*fs
= current
->fs
;
1535 br_read_lock(vfsmount_lock
);
1539 seq
= read_seqcount_begin(&fs
->seq
);
1541 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1542 } while (read_seqcount_retry(&fs
->seq
, seq
));
1544 get_fs_pwd(current
->fs
, &nd
->path
);
1547 struct dentry
*dentry
;
1549 file
= fget_raw_light(dfd
, &fput_needed
);
1554 dentry
= file
->f_path
.dentry
;
1558 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1561 retval
= file_permission(file
, MAY_EXEC
);
1566 nd
->path
= file
->f_path
;
1567 if (flags
& LOOKUP_RCU
) {
1570 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1571 br_read_lock(vfsmount_lock
);
1574 path_get(&file
->f_path
);
1575 fput_light(file
, fput_needed
);
1579 nd
->inode
= nd
->path
.dentry
->d_inode
;
1583 fput_light(file
, fput_needed
);
1588 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1590 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1591 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1593 nd
->flags
&= ~LOOKUP_PARENT
;
1594 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1595 nd
->flags
& LOOKUP_FOLLOW
);
1598 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1599 static int path_lookupat(int dfd
, const char *name
,
1600 unsigned int flags
, struct nameidata
*nd
)
1602 struct file
*base
= NULL
;
1607 * Path walking is largely split up into 2 different synchronisation
1608 * schemes, rcu-walk and ref-walk (explained in
1609 * Documentation/filesystems/path-lookup.txt). These share much of the
1610 * path walk code, but some things particularly setup, cleanup, and
1611 * following mounts are sufficiently divergent that functions are
1612 * duplicated. Typically there is a function foo(), and its RCU
1613 * analogue, foo_rcu().
1615 * -ECHILD is the error number of choice (just to avoid clashes) that
1616 * is returned if some aspect of an rcu-walk fails. Such an error must
1617 * be handled by restarting a traditional ref-walk (which will always
1618 * be able to complete).
1620 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1625 current
->total_link_count
= 0;
1626 err
= link_path_walk(name
, nd
);
1628 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1629 err
= lookup_last(nd
, &path
);
1632 struct path link
= path
;
1633 nd
->flags
|= LOOKUP_PARENT
;
1634 err
= follow_link(&link
, nd
, &cookie
);
1636 err
= lookup_last(nd
, &path
);
1637 put_link(nd
, &link
, cookie
);
1641 if (nd
->flags
& LOOKUP_RCU
) {
1642 /* went all way through without dropping RCU */
1644 if (nameidata_drop_rcu_last(nd
))
1649 err
= handle_reval_path(nd
);
1651 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1652 if (!nd
->inode
->i_op
->lookup
) {
1653 path_put(&nd
->path
);
1661 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1662 path_put(&nd
->root
);
1663 nd
->root
.mnt
= NULL
;
1668 static int do_path_lookup(int dfd
, const char *name
,
1669 unsigned int flags
, struct nameidata
*nd
)
1671 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1672 if (unlikely(retval
== -ECHILD
))
1673 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1674 if (unlikely(retval
== -ESTALE
))
1675 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1677 if (likely(!retval
)) {
1678 if (unlikely(!audit_dummy_context())) {
1679 if (nd
->path
.dentry
&& nd
->inode
)
1680 audit_inode(name
, nd
->path
.dentry
);
1686 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1688 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1691 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1693 struct nameidata nd
;
1694 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1701 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1702 * @dentry: pointer to dentry of the base directory
1703 * @mnt: pointer to vfs mount of the base directory
1704 * @name: pointer to file name
1705 * @flags: lookup flags
1706 * @nd: pointer to nameidata
1708 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1709 const char *name
, unsigned int flags
,
1710 struct nameidata
*nd
)
1712 nd
->root
.dentry
= dentry
;
1714 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1715 return do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, nd
);
1718 static struct dentry
*__lookup_hash(struct qstr
*name
,
1719 struct dentry
*base
, struct nameidata
*nd
)
1721 struct inode
*inode
= base
->d_inode
;
1722 struct dentry
*dentry
;
1725 err
= exec_permission(inode
, 0);
1727 return ERR_PTR(err
);
1730 * Don't bother with __d_lookup: callers are for creat as
1731 * well as unlink, so a lot of the time it would cost
1734 dentry
= d_lookup(base
, name
);
1736 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1737 dentry
= do_revalidate(dentry
, nd
);
1740 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1746 * Restricted form of lookup. Doesn't follow links, single-component only,
1747 * needs parent already locked. Doesn't follow mounts.
1750 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1752 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1756 * lookup_one_len - filesystem helper to lookup single pathname component
1757 * @name: pathname component to lookup
1758 * @base: base directory to lookup from
1759 * @len: maximum length @len should be interpreted to
1761 * Note that this routine is purely a helper for filesystem usage and should
1762 * not be called by generic code. Also note that by using this function the
1763 * nameidata argument is passed to the filesystem methods and a filesystem
1764 * using this helper needs to be prepared for that.
1766 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1772 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1777 return ERR_PTR(-EACCES
);
1779 hash
= init_name_hash();
1781 c
= *(const unsigned char *)name
++;
1782 if (c
== '/' || c
== '\0')
1783 return ERR_PTR(-EACCES
);
1784 hash
= partial_name_hash(c
, hash
);
1786 this.hash
= end_name_hash(hash
);
1788 * See if the low-level filesystem might want
1789 * to use its own hash..
1791 if (base
->d_flags
& DCACHE_OP_HASH
) {
1792 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1794 return ERR_PTR(err
);
1797 return __lookup_hash(&this, base
, NULL
);
1800 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1803 struct nameidata nd
;
1804 char *tmp
= getname_flags(name
, flags
);
1805 int err
= PTR_ERR(tmp
);
1808 BUG_ON(flags
& LOOKUP_PARENT
);
1810 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1818 static int user_path_parent(int dfd
, const char __user
*path
,
1819 struct nameidata
*nd
, char **name
)
1821 char *s
= getname(path
);
1827 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1837 * It's inline, so penalty for filesystems that don't use sticky bit is
1840 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1842 uid_t fsuid
= current_fsuid();
1844 if (!(dir
->i_mode
& S_ISVTX
))
1846 if (inode
->i_uid
== fsuid
)
1848 if (dir
->i_uid
== fsuid
)
1850 return !capable(CAP_FOWNER
);
1854 * Check whether we can remove a link victim from directory dir, check
1855 * whether the type of victim is right.
1856 * 1. We can't do it if dir is read-only (done in permission())
1857 * 2. We should have write and exec permissions on dir
1858 * 3. We can't remove anything from append-only dir
1859 * 4. We can't do anything with immutable dir (done in permission())
1860 * 5. If the sticky bit on dir is set we should either
1861 * a. be owner of dir, or
1862 * b. be owner of victim, or
1863 * c. have CAP_FOWNER capability
1864 * 6. If the victim is append-only or immutable we can't do antyhing with
1865 * links pointing to it.
1866 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1867 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1868 * 9. We can't remove a root or mountpoint.
1869 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1870 * nfs_async_unlink().
1872 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1876 if (!victim
->d_inode
)
1879 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1880 audit_inode_child(victim
, dir
);
1882 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1887 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1888 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1891 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1893 if (IS_ROOT(victim
))
1895 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1897 if (IS_DEADDIR(dir
))
1899 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1904 /* Check whether we can create an object with dentry child in directory
1906 * 1. We can't do it if child already exists (open has special treatment for
1907 * this case, but since we are inlined it's OK)
1908 * 2. We can't do it if dir is read-only (done in permission())
1909 * 3. We should have write and exec permissions on dir
1910 * 4. We can't do it if dir is immutable (done in permission())
1912 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1916 if (IS_DEADDIR(dir
))
1918 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1922 * p1 and p2 should be directories on the same fs.
1924 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1929 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1933 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1935 p
= d_ancestor(p2
, p1
);
1937 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1938 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1942 p
= d_ancestor(p1
, p2
);
1944 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1945 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1949 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1950 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1954 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1956 mutex_unlock(&p1
->d_inode
->i_mutex
);
1958 mutex_unlock(&p2
->d_inode
->i_mutex
);
1959 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1963 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1964 struct nameidata
*nd
)
1966 int error
= may_create(dir
, dentry
);
1971 if (!dir
->i_op
->create
)
1972 return -EACCES
; /* shouldn't it be ENOSYS? */
1975 error
= security_inode_create(dir
, dentry
, mode
);
1978 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1980 fsnotify_create(dir
, dentry
);
1984 static int may_open(struct path
*path
, int acc_mode
, int flag
)
1986 struct dentry
*dentry
= path
->dentry
;
1987 struct inode
*inode
= dentry
->d_inode
;
1997 switch (inode
->i_mode
& S_IFMT
) {
2001 if (acc_mode
& MAY_WRITE
)
2006 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2015 error
= inode_permission(inode
, acc_mode
);
2020 * An append-only file must be opened in append mode for writing.
2022 if (IS_APPEND(inode
)) {
2023 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2029 /* O_NOATIME can only be set by the owner or superuser */
2030 if (flag
& O_NOATIME
&& !is_owner_or_cap(inode
))
2034 * Ensure there are no outstanding leases on the file.
2036 return break_lease(inode
, flag
);
2039 static int handle_truncate(struct file
*filp
)
2041 struct path
*path
= &filp
->f_path
;
2042 struct inode
*inode
= path
->dentry
->d_inode
;
2043 int error
= get_write_access(inode
);
2047 * Refuse to truncate files with mandatory locks held on them.
2049 error
= locks_verify_locked(inode
);
2051 error
= security_path_truncate(path
);
2053 error
= do_truncate(path
->dentry
, 0,
2054 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2057 put_write_access(inode
);
2062 * Note that while the flag value (low two bits) for sys_open means:
2067 * it is changed into
2068 * 00 - no permissions needed
2069 * 01 - read-permission
2070 * 10 - write-permission
2072 * for the internal routines (ie open_namei()/follow_link() etc)
2073 * This is more logical, and also allows the 00 "no perm needed"
2074 * to be used for symlinks (where the permissions are checked
2078 static inline int open_to_namei_flags(int flag
)
2080 if ((flag
+1) & O_ACCMODE
)
2086 * Handle the last step of open()
2088 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2089 const struct open_flags
*op
, const char *pathname
)
2091 struct dentry
*dir
= nd
->path
.dentry
;
2092 struct dentry
*dentry
;
2093 int open_flag
= op
->open_flag
;
2094 int will_truncate
= open_flag
& O_TRUNC
;
2096 int acc_mode
= op
->acc_mode
;
2100 nd
->flags
&= ~LOOKUP_PARENT
;
2101 nd
->flags
|= op
->intent
;
2103 switch (nd
->last_type
) {
2106 error
= handle_dots(nd
, nd
->last_type
);
2108 return ERR_PTR(error
);
2111 if (nd
->flags
& LOOKUP_RCU
) {
2112 if (nameidata_drop_rcu_last(nd
))
2113 return ERR_PTR(-ECHILD
);
2115 error
= handle_reval_path(nd
);
2118 audit_inode(pathname
, nd
->path
.dentry
);
2119 if (open_flag
& O_CREAT
) {
2125 /* can't be RCU mode here */
2126 error
= handle_reval_path(nd
);
2129 audit_inode(pathname
, dir
);
2133 if (!(open_flag
& O_CREAT
)) {
2135 if (nd
->last
.name
[nd
->last
.len
])
2136 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2137 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2139 /* we _can_ be in RCU mode here */
2140 error
= walk_component(nd
, path
, &nd
->last
, LAST_NORM
,
2143 return ERR_PTR(error
);
2144 if (error
) /* symlink */
2147 if (nd
->flags
& LOOKUP_RCU
) {
2148 if (nameidata_drop_rcu_last(nd
))
2149 return ERR_PTR(-ECHILD
);
2153 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2154 if (!nd
->inode
->i_op
->lookup
)
2157 audit_inode(pathname
, nd
->path
.dentry
);
2161 /* create side of things */
2163 if (nd
->flags
& LOOKUP_RCU
) {
2164 if (nameidata_drop_rcu_last(nd
))
2165 return ERR_PTR(-ECHILD
);
2168 audit_inode(pathname
, dir
);
2170 /* trailing slashes? */
2171 if (nd
->last
.name
[nd
->last
.len
])
2174 mutex_lock(&dir
->d_inode
->i_mutex
);
2176 dentry
= lookup_hash(nd
);
2177 error
= PTR_ERR(dentry
);
2178 if (IS_ERR(dentry
)) {
2179 mutex_unlock(&dir
->d_inode
->i_mutex
);
2183 path
->dentry
= dentry
;
2184 path
->mnt
= nd
->path
.mnt
;
2186 /* Negative dentry, just create the file */
2187 if (!dentry
->d_inode
) {
2188 int mode
= op
->mode
;
2189 if (!IS_POSIXACL(dir
->d_inode
))
2190 mode
&= ~current_umask();
2192 * This write is needed to ensure that a
2193 * rw->ro transition does not occur between
2194 * the time when the file is created and when
2195 * a permanent write count is taken through
2196 * the 'struct file' in nameidata_to_filp().
2198 error
= mnt_want_write(nd
->path
.mnt
);
2200 goto exit_mutex_unlock
;
2202 /* Don't check for write permission, don't truncate */
2203 open_flag
&= ~O_TRUNC
;
2205 acc_mode
= MAY_OPEN
;
2206 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2208 goto exit_mutex_unlock
;
2209 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2211 goto exit_mutex_unlock
;
2212 mutex_unlock(&dir
->d_inode
->i_mutex
);
2213 dput(nd
->path
.dentry
);
2214 nd
->path
.dentry
= dentry
;
2219 * It already exists.
2221 mutex_unlock(&dir
->d_inode
->i_mutex
);
2222 audit_inode(pathname
, path
->dentry
);
2225 if (open_flag
& O_EXCL
)
2228 error
= follow_managed(path
, nd
->flags
);
2233 if (!path
->dentry
->d_inode
)
2236 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2239 path_to_nameidata(path
, nd
);
2240 nd
->inode
= path
->dentry
->d_inode
;
2242 if (S_ISDIR(nd
->inode
->i_mode
))
2245 if (!S_ISREG(nd
->inode
->i_mode
))
2248 if (will_truncate
) {
2249 error
= mnt_want_write(nd
->path
.mnt
);
2255 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2258 filp
= nameidata_to_filp(nd
);
2259 if (!IS_ERR(filp
)) {
2260 error
= ima_file_check(filp
, op
->acc_mode
);
2263 filp
= ERR_PTR(error
);
2266 if (!IS_ERR(filp
)) {
2267 if (will_truncate
) {
2268 error
= handle_truncate(filp
);
2271 filp
= ERR_PTR(error
);
2277 mnt_drop_write(nd
->path
.mnt
);
2278 path_put(&nd
->path
);
2282 mutex_unlock(&dir
->d_inode
->i_mutex
);
2284 path_put_conditional(path
, nd
);
2286 filp
= ERR_PTR(error
);
2290 static struct file
*path_openat(int dfd
, const char *pathname
,
2291 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2293 struct file
*base
= NULL
;
2298 filp
= get_empty_filp();
2300 return ERR_PTR(-ENFILE
);
2302 filp
->f_flags
= op
->open_flag
;
2303 nd
->intent
.open
.file
= filp
;
2304 nd
->intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2305 nd
->intent
.open
.create_mode
= op
->mode
;
2307 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2308 if (unlikely(error
))
2311 current
->total_link_count
= 0;
2312 error
= link_path_walk(pathname
, nd
);
2313 if (unlikely(error
))
2316 filp
= do_last(nd
, &path
, op
, pathname
);
2317 while (unlikely(!filp
)) { /* trailing symlink */
2318 struct path link
= path
;
2320 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2321 path_put_conditional(&path
, nd
);
2322 path_put(&nd
->path
);
2323 filp
= ERR_PTR(-ELOOP
);
2326 nd
->flags
|= LOOKUP_PARENT
;
2327 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2328 error
= follow_link(&link
, nd
, &cookie
);
2329 if (unlikely(error
))
2330 filp
= ERR_PTR(error
);
2332 filp
= do_last(nd
, &path
, op
, pathname
);
2333 put_link(nd
, &link
, cookie
);
2336 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2337 path_put(&nd
->root
);
2340 release_open_intent(nd
);
2344 filp
= ERR_PTR(error
);
2348 struct file
*do_filp_open(int dfd
, const char *pathname
,
2349 const struct open_flags
*op
, int flags
)
2351 struct nameidata nd
;
2354 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2355 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2356 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2357 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2358 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2362 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2363 const char *name
, const struct open_flags
*op
, int flags
)
2365 struct nameidata nd
;
2369 nd
.root
.dentry
= dentry
;
2371 flags
|= LOOKUP_ROOT
;
2373 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2374 return ERR_PTR(-ELOOP
);
2376 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2377 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2378 file
= path_openat(-1, name
, &nd
, op
, flags
);
2379 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2380 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2385 * lookup_create - lookup a dentry, creating it if it doesn't exist
2386 * @nd: nameidata info
2387 * @is_dir: directory flag
2389 * Simple function to lookup and return a dentry and create it
2390 * if it doesn't exist. Is SMP-safe.
2392 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2394 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2396 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2398 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2400 * Yucky last component or no last component at all?
2401 * (foo/., foo/.., /////)
2403 if (nd
->last_type
!= LAST_NORM
)
2405 nd
->flags
&= ~LOOKUP_PARENT
;
2406 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2407 nd
->intent
.open
.flags
= O_EXCL
;
2410 * Do the final lookup.
2412 dentry
= lookup_hash(nd
);
2416 if (dentry
->d_inode
)
2419 * Special case - lookup gave negative, but... we had foo/bar/
2420 * From the vfs_mknod() POV we just have a negative dentry -
2421 * all is fine. Let's be bastards - you had / on the end, you've
2422 * been asking for (non-existent) directory. -ENOENT for you.
2424 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2426 dentry
= ERR_PTR(-ENOENT
);
2431 dentry
= ERR_PTR(-EEXIST
);
2435 EXPORT_SYMBOL_GPL(lookup_create
);
2437 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2439 int error
= may_create(dir
, dentry
);
2444 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
2447 if (!dir
->i_op
->mknod
)
2450 error
= devcgroup_inode_mknod(mode
, dev
);
2454 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2458 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2460 fsnotify_create(dir
, dentry
);
2464 static int may_mknod(mode_t mode
)
2466 switch (mode
& S_IFMT
) {
2472 case 0: /* zero mode translates to S_IFREG */
2481 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2486 struct dentry
*dentry
;
2487 struct nameidata nd
;
2492 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2496 dentry
= lookup_create(&nd
, 0);
2497 if (IS_ERR(dentry
)) {
2498 error
= PTR_ERR(dentry
);
2501 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2502 mode
&= ~current_umask();
2503 error
= may_mknod(mode
);
2506 error
= mnt_want_write(nd
.path
.mnt
);
2509 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2511 goto out_drop_write
;
2512 switch (mode
& S_IFMT
) {
2513 case 0: case S_IFREG
:
2514 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2516 case S_IFCHR
: case S_IFBLK
:
2517 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2518 new_decode_dev(dev
));
2520 case S_IFIFO
: case S_IFSOCK
:
2521 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2525 mnt_drop_write(nd
.path
.mnt
);
2529 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2536 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2538 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2541 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2543 int error
= may_create(dir
, dentry
);
2548 if (!dir
->i_op
->mkdir
)
2551 mode
&= (S_IRWXUGO
|S_ISVTX
);
2552 error
= security_inode_mkdir(dir
, dentry
, mode
);
2556 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2558 fsnotify_mkdir(dir
, dentry
);
2562 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2566 struct dentry
*dentry
;
2567 struct nameidata nd
;
2569 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2573 dentry
= lookup_create(&nd
, 1);
2574 error
= PTR_ERR(dentry
);
2578 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2579 mode
&= ~current_umask();
2580 error
= mnt_want_write(nd
.path
.mnt
);
2583 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2585 goto out_drop_write
;
2586 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2588 mnt_drop_write(nd
.path
.mnt
);
2592 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2599 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2601 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2605 * We try to drop the dentry early: we should have
2606 * a usage count of 2 if we're the only user of this
2607 * dentry, and if that is true (possibly after pruning
2608 * the dcache), then we drop the dentry now.
2610 * A low-level filesystem can, if it choses, legally
2613 * if (!d_unhashed(dentry))
2616 * if it cannot handle the case of removing a directory
2617 * that is still in use by something else..
2619 void dentry_unhash(struct dentry
*dentry
)
2622 shrink_dcache_parent(dentry
);
2623 spin_lock(&dentry
->d_lock
);
2624 if (dentry
->d_count
== 2)
2626 spin_unlock(&dentry
->d_lock
);
2629 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2631 int error
= may_delete(dir
, dentry
, 1);
2636 if (!dir
->i_op
->rmdir
)
2639 mutex_lock(&dentry
->d_inode
->i_mutex
);
2640 dentry_unhash(dentry
);
2641 if (d_mountpoint(dentry
))
2644 error
= security_inode_rmdir(dir
, dentry
);
2646 error
= dir
->i_op
->rmdir(dir
, dentry
);
2648 dentry
->d_inode
->i_flags
|= S_DEAD
;
2653 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2662 static long do_rmdir(int dfd
, const char __user
*pathname
)
2666 struct dentry
*dentry
;
2667 struct nameidata nd
;
2669 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2673 switch(nd
.last_type
) {
2685 nd
.flags
&= ~LOOKUP_PARENT
;
2687 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2688 dentry
= lookup_hash(&nd
);
2689 error
= PTR_ERR(dentry
);
2692 error
= mnt_want_write(nd
.path
.mnt
);
2695 error
= security_path_rmdir(&nd
.path
, dentry
);
2698 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2700 mnt_drop_write(nd
.path
.mnt
);
2704 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2711 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2713 return do_rmdir(AT_FDCWD
, pathname
);
2716 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2718 int error
= may_delete(dir
, dentry
, 0);
2723 if (!dir
->i_op
->unlink
)
2726 mutex_lock(&dentry
->d_inode
->i_mutex
);
2727 if (d_mountpoint(dentry
))
2730 error
= security_inode_unlink(dir
, dentry
);
2732 error
= dir
->i_op
->unlink(dir
, dentry
);
2737 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2739 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2740 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2741 fsnotify_link_count(dentry
->d_inode
);
2749 * Make sure that the actual truncation of the file will occur outside its
2750 * directory's i_mutex. Truncate can take a long time if there is a lot of
2751 * writeout happening, and we don't want to prevent access to the directory
2752 * while waiting on the I/O.
2754 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2758 struct dentry
*dentry
;
2759 struct nameidata nd
;
2760 struct inode
*inode
= NULL
;
2762 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2767 if (nd
.last_type
!= LAST_NORM
)
2770 nd
.flags
&= ~LOOKUP_PARENT
;
2772 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2773 dentry
= lookup_hash(&nd
);
2774 error
= PTR_ERR(dentry
);
2775 if (!IS_ERR(dentry
)) {
2776 /* Why not before? Because we want correct error value */
2777 if (nd
.last
.name
[nd
.last
.len
])
2779 inode
= dentry
->d_inode
;
2782 error
= mnt_want_write(nd
.path
.mnt
);
2785 error
= security_path_unlink(&nd
.path
, dentry
);
2788 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2790 mnt_drop_write(nd
.path
.mnt
);
2794 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2796 iput(inode
); /* truncate the inode here */
2803 error
= !dentry
->d_inode
? -ENOENT
:
2804 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2808 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2810 if ((flag
& ~AT_REMOVEDIR
) != 0)
2813 if (flag
& AT_REMOVEDIR
)
2814 return do_rmdir(dfd
, pathname
);
2816 return do_unlinkat(dfd
, pathname
);
2819 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2821 return do_unlinkat(AT_FDCWD
, pathname
);
2824 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2826 int error
= may_create(dir
, dentry
);
2831 if (!dir
->i_op
->symlink
)
2834 error
= security_inode_symlink(dir
, dentry
, oldname
);
2838 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2840 fsnotify_create(dir
, dentry
);
2844 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2845 int, newdfd
, const char __user
*, newname
)
2850 struct dentry
*dentry
;
2851 struct nameidata nd
;
2853 from
= getname(oldname
);
2855 return PTR_ERR(from
);
2857 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2861 dentry
= lookup_create(&nd
, 0);
2862 error
= PTR_ERR(dentry
);
2866 error
= mnt_want_write(nd
.path
.mnt
);
2869 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2871 goto out_drop_write
;
2872 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2874 mnt_drop_write(nd
.path
.mnt
);
2878 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2886 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2888 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2891 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2893 struct inode
*inode
= old_dentry
->d_inode
;
2899 error
= may_create(dir
, new_dentry
);
2903 if (dir
->i_sb
!= inode
->i_sb
)
2907 * A link to an append-only or immutable file cannot be created.
2909 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2911 if (!dir
->i_op
->link
)
2913 if (S_ISDIR(inode
->i_mode
))
2916 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2920 mutex_lock(&inode
->i_mutex
);
2921 /* Make sure we don't allow creating hardlink to an unlinked file */
2922 if (inode
->i_nlink
== 0)
2925 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2926 mutex_unlock(&inode
->i_mutex
);
2928 fsnotify_link(dir
, inode
, new_dentry
);
2933 * Hardlinks are often used in delicate situations. We avoid
2934 * security-related surprises by not following symlinks on the
2937 * We don't follow them on the oldname either to be compatible
2938 * with linux 2.0, and to avoid hard-linking to directories
2939 * and other special files. --ADM
2941 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2942 int, newdfd
, const char __user
*, newname
, int, flags
)
2944 struct dentry
*new_dentry
;
2945 struct nameidata nd
;
2946 struct path old_path
;
2951 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
2954 * To use null names we require CAP_DAC_READ_SEARCH
2955 * This ensures that not everyone will be able to create
2956 * handlink using the passed filedescriptor.
2958 if (flags
& AT_EMPTY_PATH
) {
2959 if (!capable(CAP_DAC_READ_SEARCH
))
2964 if (flags
& AT_SYMLINK_FOLLOW
)
2965 how
|= LOOKUP_FOLLOW
;
2967 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
2971 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2975 if (old_path
.mnt
!= nd
.path
.mnt
)
2977 new_dentry
= lookup_create(&nd
, 0);
2978 error
= PTR_ERR(new_dentry
);
2979 if (IS_ERR(new_dentry
))
2981 error
= mnt_want_write(nd
.path
.mnt
);
2984 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2986 goto out_drop_write
;
2987 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2989 mnt_drop_write(nd
.path
.mnt
);
2993 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2998 path_put(&old_path
);
3003 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3005 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3009 * The worst of all namespace operations - renaming directory. "Perverted"
3010 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3012 * a) we can get into loop creation. Check is done in is_subdir().
3013 * b) race potential - two innocent renames can create a loop together.
3014 * That's where 4.4 screws up. Current fix: serialization on
3015 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3017 * c) we have to lock _three_ objects - parents and victim (if it exists).
3018 * And that - after we got ->i_mutex on parents (until then we don't know
3019 * whether the target exists). Solution: try to be smart with locking
3020 * order for inodes. We rely on the fact that tree topology may change
3021 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3022 * move will be locked. Thus we can rank directories by the tree
3023 * (ancestors first) and rank all non-directories after them.
3024 * That works since everybody except rename does "lock parent, lookup,
3025 * lock child" and rename is under ->s_vfs_rename_mutex.
3026 * HOWEVER, it relies on the assumption that any object with ->lookup()
3027 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3028 * we'd better make sure that there's no link(2) for them.
3029 * d) some filesystems don't support opened-but-unlinked directories,
3030 * either because of layout or because they are not ready to deal with
3031 * all cases correctly. The latter will be fixed (taking this sort of
3032 * stuff into VFS), but the former is not going away. Solution: the same
3033 * trick as in rmdir().
3034 * e) conversion from fhandle to dentry may come in the wrong moment - when
3035 * we are removing the target. Solution: we will have to grab ->i_mutex
3036 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3037 * ->i_mutex on parents, which works but leads to some truly excessive
3040 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3041 struct inode
*new_dir
, struct dentry
*new_dentry
)
3044 struct inode
*target
;
3047 * If we are going to change the parent - check write permissions,
3048 * we'll need to flip '..'.
3050 if (new_dir
!= old_dir
) {
3051 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3056 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3060 target
= new_dentry
->d_inode
;
3062 mutex_lock(&target
->i_mutex
);
3063 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3067 dentry_unhash(new_dentry
);
3068 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3072 target
->i_flags
|= S_DEAD
;
3073 dont_mount(new_dentry
);
3075 mutex_unlock(&target
->i_mutex
);
3076 if (d_unhashed(new_dentry
))
3077 d_rehash(new_dentry
);
3081 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3082 d_move(old_dentry
,new_dentry
);
3086 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3087 struct inode
*new_dir
, struct dentry
*new_dentry
)
3089 struct inode
*target
;
3092 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3097 target
= new_dentry
->d_inode
;
3099 mutex_lock(&target
->i_mutex
);
3100 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3103 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3106 dont_mount(new_dentry
);
3107 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3108 d_move(old_dentry
, new_dentry
);
3111 mutex_unlock(&target
->i_mutex
);
3116 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3117 struct inode
*new_dir
, struct dentry
*new_dentry
)
3120 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3121 const unsigned char *old_name
;
3123 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3126 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3130 if (!new_dentry
->d_inode
)
3131 error
= may_create(new_dir
, new_dentry
);
3133 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3137 if (!old_dir
->i_op
->rename
)
3140 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3143 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3145 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3147 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3148 new_dentry
->d_inode
, old_dentry
);
3149 fsnotify_oldname_free(old_name
);
3154 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3155 int, newdfd
, const char __user
*, newname
)
3157 struct dentry
*old_dir
, *new_dir
;
3158 struct dentry
*old_dentry
, *new_dentry
;
3159 struct dentry
*trap
;
3160 struct nameidata oldnd
, newnd
;
3165 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3169 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3174 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3177 old_dir
= oldnd
.path
.dentry
;
3179 if (oldnd
.last_type
!= LAST_NORM
)
3182 new_dir
= newnd
.path
.dentry
;
3183 if (newnd
.last_type
!= LAST_NORM
)
3186 oldnd
.flags
&= ~LOOKUP_PARENT
;
3187 newnd
.flags
&= ~LOOKUP_PARENT
;
3188 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3190 trap
= lock_rename(new_dir
, old_dir
);
3192 old_dentry
= lookup_hash(&oldnd
);
3193 error
= PTR_ERR(old_dentry
);
3194 if (IS_ERR(old_dentry
))
3196 /* source must exist */
3198 if (!old_dentry
->d_inode
)
3200 /* unless the source is a directory trailing slashes give -ENOTDIR */
3201 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3203 if (oldnd
.last
.name
[oldnd
.last
.len
])
3205 if (newnd
.last
.name
[newnd
.last
.len
])
3208 /* source should not be ancestor of target */
3210 if (old_dentry
== trap
)
3212 new_dentry
= lookup_hash(&newnd
);
3213 error
= PTR_ERR(new_dentry
);
3214 if (IS_ERR(new_dentry
))
3216 /* target should not be an ancestor of source */
3218 if (new_dentry
== trap
)
3221 error
= mnt_want_write(oldnd
.path
.mnt
);
3224 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3225 &newnd
.path
, new_dentry
);
3228 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3229 new_dir
->d_inode
, new_dentry
);
3231 mnt_drop_write(oldnd
.path
.mnt
);
3237 unlock_rename(new_dir
, old_dir
);
3239 path_put(&newnd
.path
);
3242 path_put(&oldnd
.path
);
3248 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3250 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3253 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3257 len
= PTR_ERR(link
);
3262 if (len
> (unsigned) buflen
)
3264 if (copy_to_user(buffer
, link
, len
))
3271 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3272 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3273 * using) it for any given inode is up to filesystem.
3275 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3277 struct nameidata nd
;
3282 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3284 return PTR_ERR(cookie
);
3286 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3287 if (dentry
->d_inode
->i_op
->put_link
)
3288 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3292 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3294 return __vfs_follow_link(nd
, link
);
3297 /* get the link contents into pagecache */
3298 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3302 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3303 page
= read_mapping_page(mapping
, 0, NULL
);
3308 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3312 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3314 struct page
*page
= NULL
;
3315 char *s
= page_getlink(dentry
, &page
);
3316 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3319 page_cache_release(page
);
3324 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3326 struct page
*page
= NULL
;
3327 nd_set_link(nd
, page_getlink(dentry
, &page
));
3331 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3333 struct page
*page
= cookie
;
3337 page_cache_release(page
);
3342 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3344 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3346 struct address_space
*mapping
= inode
->i_mapping
;
3351 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3353 flags
|= AOP_FLAG_NOFS
;
3356 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3357 flags
, &page
, &fsdata
);
3361 kaddr
= kmap_atomic(page
, KM_USER0
);
3362 memcpy(kaddr
, symname
, len
-1);
3363 kunmap_atomic(kaddr
, KM_USER0
);
3365 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3372 mark_inode_dirty(inode
);
3378 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3380 return __page_symlink(inode
, symname
, len
,
3381 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3384 const struct inode_operations page_symlink_inode_operations
= {
3385 .readlink
= generic_readlink
,
3386 .follow_link
= page_follow_link_light
,
3387 .put_link
= page_put_link
,
3390 EXPORT_SYMBOL(user_path_at
);
3391 EXPORT_SYMBOL(follow_down_one
);
3392 EXPORT_SYMBOL(follow_down
);
3393 EXPORT_SYMBOL(follow_up
);
3394 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3395 EXPORT_SYMBOL(getname
);
3396 EXPORT_SYMBOL(lock_rename
);
3397 EXPORT_SYMBOL(lookup_one_len
);
3398 EXPORT_SYMBOL(page_follow_link_light
);
3399 EXPORT_SYMBOL(page_put_link
);
3400 EXPORT_SYMBOL(page_readlink
);
3401 EXPORT_SYMBOL(__page_symlink
);
3402 EXPORT_SYMBOL(page_symlink
);
3403 EXPORT_SYMBOL(page_symlink_inode_operations
);
3404 EXPORT_SYMBOL(kern_path_parent
);
3405 EXPORT_SYMBOL(kern_path
);
3406 EXPORT_SYMBOL(vfs_path_lookup
);
3407 EXPORT_SYMBOL(inode_permission
);
3408 EXPORT_SYMBOL(file_permission
);
3409 EXPORT_SYMBOL(unlock_rename
);
3410 EXPORT_SYMBOL(vfs_create
);
3411 EXPORT_SYMBOL(vfs_follow_link
);
3412 EXPORT_SYMBOL(vfs_link
);
3413 EXPORT_SYMBOL(vfs_mkdir
);
3414 EXPORT_SYMBOL(vfs_mknod
);
3415 EXPORT_SYMBOL(generic_permission
);
3416 EXPORT_SYMBOL(vfs_readlink
);
3417 EXPORT_SYMBOL(vfs_rename
);
3418 EXPORT_SYMBOL(vfs_rmdir
);
3419 EXPORT_SYMBOL(vfs_symlink
);
3420 EXPORT_SYMBOL(vfs_unlink
);
3421 EXPORT_SYMBOL(dentry_unhash
);
3422 EXPORT_SYMBOL(generic_readlink
);