4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user
*filename
, char *page
)
120 unsigned long len
= PATH_MAX
;
122 if (!segment_eq(get_fs(), KERNEL_DS
)) {
123 if ((unsigned long) filename
>= TASK_SIZE
)
125 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
126 len
= TASK_SIZE
- (unsigned long) filename
;
129 retval
= strncpy_from_user(page
, filename
, len
);
133 return -ENAMETOOLONG
;
139 char * getname(const char __user
* filename
)
143 result
= ERR_PTR(-ENOMEM
);
146 int retval
= do_getname(filename
, tmp
);
151 result
= ERR_PTR(retval
);
154 audit_getname(result
);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name
)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname
);
170 * This does basic POSIX ACL permission checking
172 static int acl_permission_check(struct inode
*inode
, int mask
, unsigned int flags
,
173 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
175 umode_t mode
= inode
->i_mode
;
177 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
179 if (current_fsuid() == inode
->i_uid
)
182 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
183 int error
= check_acl(inode
, mask
, flags
);
184 if (error
!= -EAGAIN
)
188 if (in_group_p(inode
->i_gid
))
193 * If the DACs are ok we don't need any capability check.
195 if ((mask
& ~mode
) == 0)
201 * generic_permission - check for access rights on a Posix-like filesystem
202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
205 * @flags: IPERM_FLAG_ flags.
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
210 * are used for other things.
212 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
213 * request cannot be satisfied (eg. requires blocking or too much complexity).
214 * It would then be called again in ref-walk mode.
216 int generic_permission(struct inode
*inode
, int mask
, unsigned int flags
,
217 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
222 * Do the basic POSIX ACL permission checks.
224 ret
= acl_permission_check(inode
, mask
, flags
, check_acl
);
229 * Read/write DACs are always overridable.
230 * Executable DACs are overridable if at least one exec bit is set.
232 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
233 if (capable(CAP_DAC_OVERRIDE
))
237 * Searching includes executable on directories, else just read.
239 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
240 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
241 if (capable(CAP_DAC_READ_SEARCH
))
248 * inode_permission - check for access rights to a given inode
249 * @inode: inode to check permission on
250 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
252 * Used to check for read/write/execute permissions on an inode.
253 * We use "fsuid" for this, letting us set arbitrary permissions
254 * for filesystem access without changing the "normal" uids which
255 * are used for other things.
257 int inode_permission(struct inode
*inode
, int mask
)
261 if (mask
& MAY_WRITE
) {
262 umode_t mode
= inode
->i_mode
;
265 * Nobody gets write access to a read-only fs.
267 if (IS_RDONLY(inode
) &&
268 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
272 * Nobody gets write access to an immutable file.
274 if (IS_IMMUTABLE(inode
))
278 if (inode
->i_op
->permission
)
279 retval
= inode
->i_op
->permission(inode
, mask
, 0);
281 retval
= generic_permission(inode
, mask
, 0,
282 inode
->i_op
->check_acl
);
287 retval
= devcgroup_inode_permission(inode
, mask
);
291 return security_inode_permission(inode
, mask
);
295 * file_permission - check for additional access rights to a given file
296 * @file: file to check access rights for
297 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
299 * Used to check for read/write/execute permissions on an already opened
303 * Do not use this function in new code. All access checks should
304 * be done using inode_permission().
306 int file_permission(struct file
*file
, int mask
)
308 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
312 * get_write_access() gets write permission for a file.
313 * put_write_access() releases this write permission.
314 * This is used for regular files.
315 * We cannot support write (and maybe mmap read-write shared) accesses and
316 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
317 * can have the following values:
318 * 0: no writers, no VM_DENYWRITE mappings
319 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
320 * > 0: (i_writecount) users are writing to the file.
322 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
323 * except for the cases where we don't hold i_writecount yet. Then we need to
324 * use {get,deny}_write_access() - these functions check the sign and refuse
325 * to do the change if sign is wrong. Exclusion between them is provided by
326 * the inode->i_lock spinlock.
329 int get_write_access(struct inode
* inode
)
331 spin_lock(&inode
->i_lock
);
332 if (atomic_read(&inode
->i_writecount
) < 0) {
333 spin_unlock(&inode
->i_lock
);
336 atomic_inc(&inode
->i_writecount
);
337 spin_unlock(&inode
->i_lock
);
342 int deny_write_access(struct file
* file
)
344 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
346 spin_lock(&inode
->i_lock
);
347 if (atomic_read(&inode
->i_writecount
) > 0) {
348 spin_unlock(&inode
->i_lock
);
351 atomic_dec(&inode
->i_writecount
);
352 spin_unlock(&inode
->i_lock
);
358 * path_get - get a reference to a path
359 * @path: path to get the reference to
361 * Given a path increment the reference count to the dentry and the vfsmount.
363 void path_get(struct path
*path
)
368 EXPORT_SYMBOL(path_get
);
371 * path_put - put a reference to a path
372 * @path: path to put the reference to
374 * Given a path decrement the reference count to the dentry and the vfsmount.
376 void path_put(struct path
*path
)
381 EXPORT_SYMBOL(path_put
);
384 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
385 * @nd: nameidata pathwalk data to drop
386 * Returns: 0 on success, -ECHILD on failure
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
390 * to drop out of rcu-walk mode and take normal reference counts on dentries
391 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
392 * refcounts at the last known good point before rcu-walk got stuck, so
393 * ref-walk may continue from there. If this is not successful (eg. a seqcount
394 * has changed), then failure is returned and path walk restarts from the
395 * beginning in ref-walk mode.
397 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
398 * ref-walk. Must be called from rcu-walk context.
400 static int nameidata_drop_rcu(struct nameidata
*nd
)
402 struct fs_struct
*fs
= current
->fs
;
403 struct dentry
*dentry
= nd
->path
.dentry
;
405 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
407 spin_lock(&fs
->lock
);
408 if (nd
->root
.mnt
!= fs
->root
.mnt
||
409 nd
->root
.dentry
!= fs
->root
.dentry
)
412 spin_lock(&dentry
->d_lock
);
413 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
415 BUG_ON(nd
->inode
!= dentry
->d_inode
);
416 spin_unlock(&dentry
->d_lock
);
419 spin_unlock(&fs
->lock
);
421 mntget(nd
->path
.mnt
);
424 br_read_unlock(vfsmount_lock
);
425 nd
->flags
&= ~LOOKUP_RCU
;
428 spin_unlock(&dentry
->d_lock
);
431 spin_unlock(&fs
->lock
);
435 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
436 static inline int nameidata_drop_rcu_maybe(struct nameidata
*nd
)
438 if (nd
->flags
& LOOKUP_RCU
)
439 return nameidata_drop_rcu(nd
);
444 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
445 * @nd: nameidata pathwalk data to drop
446 * @dentry: dentry to drop
447 * Returns: 0 on success, -ECHILD on failure
449 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
450 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
451 * @nd. Must be called from rcu-walk context.
453 static int nameidata_dentry_drop_rcu(struct nameidata
*nd
, struct dentry
*dentry
)
455 struct fs_struct
*fs
= current
->fs
;
456 struct dentry
*parent
= nd
->path
.dentry
;
458 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
460 spin_lock(&fs
->lock
);
461 if (nd
->root
.mnt
!= fs
->root
.mnt
||
462 nd
->root
.dentry
!= fs
->root
.dentry
)
465 spin_lock(&parent
->d_lock
);
466 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
467 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
470 * If the sequence check on the child dentry passed, then the child has
471 * not been removed from its parent. This means the parent dentry must
472 * be valid and able to take a reference at this point.
474 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
475 BUG_ON(!parent
->d_count
);
477 spin_unlock(&dentry
->d_lock
);
478 spin_unlock(&parent
->d_lock
);
481 spin_unlock(&fs
->lock
);
483 mntget(nd
->path
.mnt
);
486 br_read_unlock(vfsmount_lock
);
487 nd
->flags
&= ~LOOKUP_RCU
;
490 spin_unlock(&dentry
->d_lock
);
491 spin_unlock(&parent
->d_lock
);
494 spin_unlock(&fs
->lock
);
498 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
499 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata
*nd
, struct dentry
*dentry
)
501 if (nd
->flags
& LOOKUP_RCU
) {
502 if (unlikely(nameidata_dentry_drop_rcu(nd
, dentry
))) {
503 nd
->flags
&= ~LOOKUP_RCU
;
506 br_read_unlock(vfsmount_lock
);
514 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
515 * @nd: nameidata pathwalk data to drop
516 * Returns: 0 on success, -ECHILD on failure
518 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
519 * nd->path should be the final element of the lookup, so nd->root is discarded.
520 * Must be called from rcu-walk context.
522 static int nameidata_drop_rcu_last(struct nameidata
*nd
)
524 struct dentry
*dentry
= nd
->path
.dentry
;
526 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
527 nd
->flags
&= ~LOOKUP_RCU
;
529 spin_lock(&dentry
->d_lock
);
530 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
532 BUG_ON(nd
->inode
!= dentry
->d_inode
);
533 spin_unlock(&dentry
->d_lock
);
535 mntget(nd
->path
.mnt
);
538 br_read_unlock(vfsmount_lock
);
543 spin_unlock(&dentry
->d_lock
);
545 br_read_unlock(vfsmount_lock
);
550 * release_open_intent - free up open intent resources
551 * @nd: pointer to nameidata
553 void release_open_intent(struct nameidata
*nd
)
555 struct file
*file
= nd
->intent
.open
.file
;
557 if (file
&& !IS_ERR(file
)) {
558 if (file
->f_path
.dentry
== NULL
)
565 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
567 return dentry
->d_op
->d_revalidate(dentry
, nd
);
570 static struct dentry
*
571 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
573 int status
= d_revalidate(dentry
, nd
);
574 if (unlikely(status
<= 0)) {
576 * The dentry failed validation.
577 * If d_revalidate returned 0 attempt to invalidate
578 * the dentry otherwise d_revalidate is asking us
579 * to return a fail status.
583 dentry
= ERR_PTR(status
);
584 } else if (!d_invalidate(dentry
)) {
592 static inline struct dentry
*
593 do_revalidate_rcu(struct dentry
*dentry
, struct nameidata
*nd
)
595 int status
= d_revalidate(dentry
, nd
);
596 if (likely(status
> 0))
598 if (status
== -ECHILD
) {
599 if (nameidata_dentry_drop_rcu(nd
, dentry
))
600 return ERR_PTR(-ECHILD
);
601 return do_revalidate(dentry
, nd
);
604 return ERR_PTR(status
);
605 /* Don't d_invalidate in rcu-walk mode */
606 if (nameidata_dentry_drop_rcu(nd
, dentry
))
607 return ERR_PTR(-ECHILD
);
608 if (!d_invalidate(dentry
)) {
616 * handle_reval_path - force revalidation of a dentry
618 * In some situations the path walking code will trust dentries without
619 * revalidating them. This causes problems for filesystems that depend on
620 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
621 * (which indicates that it's possible for the dentry to go stale), force
622 * a d_revalidate call before proceeding.
624 * Returns 0 if the revalidation was successful. If the revalidation fails,
625 * either return the error returned by d_revalidate or -ESTALE if the
626 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
627 * invalidate the dentry. It's up to the caller to handle putting references
628 * to the path if necessary.
630 static inline int handle_reval_path(struct nameidata
*nd
)
632 struct dentry
*dentry
= nd
->path
.dentry
;
635 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
638 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
641 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
644 /* Note: we do not d_invalidate() */
645 status
= d_revalidate(dentry
, nd
);
656 * Short-cut version of permission(), for calling on directories
657 * during pathname resolution. Combines parts of permission()
658 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
660 * If appropriate, check DAC only. If not appropriate, or
661 * short-cut DAC fails, then call ->permission() to do more
662 * complete permission check.
664 static inline int exec_permission(struct inode
*inode
, unsigned int flags
)
668 if (inode
->i_op
->permission
) {
669 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
, flags
);
671 ret
= acl_permission_check(inode
, MAY_EXEC
, flags
,
672 inode
->i_op
->check_acl
);
679 if (capable(CAP_DAC_OVERRIDE
) || capable(CAP_DAC_READ_SEARCH
))
684 return security_inode_exec_permission(inode
, flags
);
687 static __always_inline
void set_root(struct nameidata
*nd
)
690 get_fs_root(current
->fs
, &nd
->root
);
693 static int link_path_walk(const char *, struct nameidata
*);
695 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
698 struct fs_struct
*fs
= current
->fs
;
702 seq
= read_seqcount_begin(&fs
->seq
);
704 } while (read_seqcount_retry(&fs
->seq
, seq
));
708 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
720 nd
->flags
|= LOOKUP_JUMPED
;
722 nd
->inode
= nd
->path
.dentry
->d_inode
;
724 ret
= link_path_walk(link
, nd
);
728 return PTR_ERR(link
);
731 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
734 if (path
->mnt
!= nd
->path
.mnt
)
738 static inline void path_to_nameidata(const struct path
*path
,
739 struct nameidata
*nd
)
741 if (!(nd
->flags
& LOOKUP_RCU
)) {
742 dput(nd
->path
.dentry
);
743 if (nd
->path
.mnt
!= path
->mnt
)
744 mntput(nd
->path
.mnt
);
746 nd
->path
.mnt
= path
->mnt
;
747 nd
->path
.dentry
= path
->dentry
;
750 static __always_inline
int
751 __do_follow_link(const struct path
*link
, struct nameidata
*nd
, void **p
)
754 struct dentry
*dentry
= link
->dentry
;
756 BUG_ON(nd
->flags
& LOOKUP_RCU
);
758 touch_atime(link
->mnt
, dentry
);
759 nd_set_link(nd
, NULL
);
761 if (link
->mnt
== nd
->path
.mnt
)
764 error
= security_inode_follow_link(link
->dentry
, nd
);
766 *p
= ERR_PTR(error
); /* no ->put_link(), please */
771 nd
->last_type
= LAST_BIND
;
772 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
775 char *s
= nd_get_link(nd
);
778 error
= __vfs_follow_link(nd
, s
);
779 else if (nd
->last_type
== LAST_BIND
)
780 nd
->flags
|= LOOKUP_JUMPED
;
786 * This limits recursive symlink follows to 8, while
787 * limiting consecutive symlinks to 40.
789 * Without that kind of total limit, nasty chains of consecutive
790 * symlinks can cause almost arbitrarily long lookups.
792 static inline int do_follow_link(struct inode
*inode
, struct path
*path
, struct nameidata
*nd
)
797 /* We drop rcu-walk here */
798 if (nameidata_dentry_drop_rcu_maybe(nd
, path
->dentry
))
800 BUG_ON(inode
!= path
->dentry
->d_inode
);
802 if (current
->link_count
>= MAX_NESTED_LINKS
)
804 if (current
->total_link_count
>= 40)
806 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
808 current
->link_count
++;
809 current
->total_link_count
++;
811 err
= __do_follow_link(path
, nd
, &cookie
);
812 if (!IS_ERR(cookie
) && path
->dentry
->d_inode
->i_op
->put_link
)
813 path
->dentry
->d_inode
->i_op
->put_link(path
->dentry
, nd
, cookie
);
815 current
->link_count
--;
819 path_put_conditional(path
, nd
);
824 static int follow_up_rcu(struct path
*path
)
826 struct vfsmount
*parent
;
827 struct dentry
*mountpoint
;
829 parent
= path
->mnt
->mnt_parent
;
830 if (parent
== path
->mnt
)
832 mountpoint
= path
->mnt
->mnt_mountpoint
;
833 path
->dentry
= mountpoint
;
838 int follow_up(struct path
*path
)
840 struct vfsmount
*parent
;
841 struct dentry
*mountpoint
;
843 br_read_lock(vfsmount_lock
);
844 parent
= path
->mnt
->mnt_parent
;
845 if (parent
== path
->mnt
) {
846 br_read_unlock(vfsmount_lock
);
850 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
851 br_read_unlock(vfsmount_lock
);
853 path
->dentry
= mountpoint
;
860 * Perform an automount
861 * - return -EISDIR to tell follow_managed() to stop and return the path we
864 static int follow_automount(struct path
*path
, unsigned flags
,
867 struct vfsmount
*mnt
;
870 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
873 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
874 * and this is the terminal part of the path.
876 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
877 return -EISDIR
; /* we actually want to stop here */
879 /* We want to mount if someone is trying to open/create a file of any
880 * type under the mountpoint, wants to traverse through the mountpoint
881 * or wants to open the mounted directory.
883 * We don't want to mount if someone's just doing a stat and they've
884 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
885 * appended a '/' to the name.
887 if (!(flags
& LOOKUP_FOLLOW
) &&
888 !(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
889 LOOKUP_OPEN
| LOOKUP_CREATE
)))
892 current
->total_link_count
++;
893 if (current
->total_link_count
>= 40)
896 mnt
= path
->dentry
->d_op
->d_automount(path
);
899 * The filesystem is allowed to return -EISDIR here to indicate
900 * it doesn't want to automount. For instance, autofs would do
901 * this so that its userspace daemon can mount on this dentry.
903 * However, we can only permit this if it's a terminal point in
904 * the path being looked up; if it wasn't then the remainder of
905 * the path is inaccessible and we should say so.
907 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_CONTINUE
))
912 if (!mnt
) /* mount collision */
915 err
= finish_automount(mnt
, path
);
919 /* Someone else made a mount here whilst we were busy */
926 path
->dentry
= dget(mnt
->mnt_root
);
936 * Handle a dentry that is managed in some way.
937 * - Flagged for transit management (autofs)
938 * - Flagged as mountpoint
939 * - Flagged as automount point
941 * This may only be called in refwalk mode.
943 * Serialization is taken care of in namespace.c
945 static int follow_managed(struct path
*path
, unsigned flags
)
948 bool need_mntput
= false;
951 /* Given that we're not holding a lock here, we retain the value in a
952 * local variable for each dentry as we look at it so that we don't see
953 * the components of that value change under us */
954 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
955 managed
&= DCACHE_MANAGED_DENTRY
,
956 unlikely(managed
!= 0)) {
957 /* Allow the filesystem to manage the transit without i_mutex
959 if (managed
& DCACHE_MANAGE_TRANSIT
) {
960 BUG_ON(!path
->dentry
->d_op
);
961 BUG_ON(!path
->dentry
->d_op
->d_manage
);
962 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
,
965 return ret
== -EISDIR
? 0 : ret
;
968 /* Transit to a mounted filesystem. */
969 if (managed
& DCACHE_MOUNTED
) {
970 struct vfsmount
*mounted
= lookup_mnt(path
);
976 path
->dentry
= dget(mounted
->mnt_root
);
981 /* Something is mounted on this dentry in another
982 * namespace and/or whatever was mounted there in this
983 * namespace got unmounted before we managed to get the
987 /* Handle an automount point */
988 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
989 ret
= follow_automount(path
, flags
, &need_mntput
);
991 return ret
== -EISDIR
? 0 : ret
;
995 /* We didn't change the current path point */
1001 int follow_down_one(struct path
*path
)
1003 struct vfsmount
*mounted
;
1005 mounted
= lookup_mnt(path
);
1009 path
->mnt
= mounted
;
1010 path
->dentry
= dget(mounted
->mnt_root
);
1017 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
1018 * meet a managed dentry and we're not walking to "..". True is returned to
1019 * continue, false to abort.
1021 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1022 struct inode
**inode
, bool reverse_transit
)
1024 while (d_mountpoint(path
->dentry
)) {
1025 struct vfsmount
*mounted
;
1026 if (unlikely(path
->dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) &&
1028 path
->dentry
->d_op
->d_manage(path
->dentry
, false, true) < 0)
1030 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1033 path
->mnt
= mounted
;
1034 path
->dentry
= mounted
->mnt_root
;
1035 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1036 *inode
= path
->dentry
->d_inode
;
1039 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1040 return reverse_transit
;
1044 static int follow_dotdot_rcu(struct nameidata
*nd
)
1046 struct inode
*inode
= nd
->inode
;
1051 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1052 nd
->path
.mnt
== nd
->root
.mnt
) {
1055 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1056 struct dentry
*old
= nd
->path
.dentry
;
1057 struct dentry
*parent
= old
->d_parent
;
1060 seq
= read_seqcount_begin(&parent
->d_seq
);
1061 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1063 inode
= parent
->d_inode
;
1064 nd
->path
.dentry
= parent
;
1068 if (!follow_up_rcu(&nd
->path
))
1070 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1071 inode
= nd
->path
.dentry
->d_inode
;
1073 __follow_mount_rcu(nd
, &nd
->path
, &inode
, true);
1078 nd
->flags
&= ~LOOKUP_RCU
;
1079 nd
->root
.mnt
= NULL
;
1081 br_read_unlock(vfsmount_lock
);
1086 * Follow down to the covering mount currently visible to userspace. At each
1087 * point, the filesystem owning that dentry may be queried as to whether the
1088 * caller is permitted to proceed or not.
1090 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1093 int follow_down(struct path
*path
, bool mounting_here
)
1098 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1099 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1100 /* Allow the filesystem to manage the transit without i_mutex
1103 * We indicate to the filesystem if someone is trying to mount
1104 * something here. This gives autofs the chance to deny anyone
1105 * other than its daemon the right to mount on its
1108 * The filesystem may sleep at this point.
1110 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1111 BUG_ON(!path
->dentry
->d_op
);
1112 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1113 ret
= path
->dentry
->d_op
->d_manage(
1114 path
->dentry
, mounting_here
, false);
1116 return ret
== -EISDIR
? 0 : ret
;
1119 /* Transit to a mounted filesystem. */
1120 if (managed
& DCACHE_MOUNTED
) {
1121 struct vfsmount
*mounted
= lookup_mnt(path
);
1126 path
->mnt
= mounted
;
1127 path
->dentry
= dget(mounted
->mnt_root
);
1131 /* Don't handle automount points here */
1138 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1140 static void follow_mount(struct path
*path
)
1142 while (d_mountpoint(path
->dentry
)) {
1143 struct vfsmount
*mounted
= lookup_mnt(path
);
1148 path
->mnt
= mounted
;
1149 path
->dentry
= dget(mounted
->mnt_root
);
1153 static void follow_dotdot(struct nameidata
*nd
)
1158 struct dentry
*old
= nd
->path
.dentry
;
1160 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1161 nd
->path
.mnt
== nd
->root
.mnt
) {
1164 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1165 /* rare case of legitimate dget_parent()... */
1166 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1170 if (!follow_up(&nd
->path
))
1173 follow_mount(&nd
->path
);
1174 nd
->inode
= nd
->path
.dentry
->d_inode
;
1178 * Allocate a dentry with name and parent, and perform a parent
1179 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1180 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1181 * have verified that no child exists while under i_mutex.
1183 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1184 struct qstr
*name
, struct nameidata
*nd
)
1186 struct inode
*inode
= parent
->d_inode
;
1187 struct dentry
*dentry
;
1190 /* Don't create child dentry for a dead directory. */
1191 if (unlikely(IS_DEADDIR(inode
)))
1192 return ERR_PTR(-ENOENT
);
1194 dentry
= d_alloc(parent
, name
);
1195 if (unlikely(!dentry
))
1196 return ERR_PTR(-ENOMEM
);
1198 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1199 if (unlikely(old
)) {
1207 * It's more convoluted than I'd like it to be, but... it's still fairly
1208 * small and for now I'd prefer to have fast path as straight as possible.
1209 * It _is_ time-critical.
1211 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1212 struct path
*path
, struct inode
**inode
)
1214 struct vfsmount
*mnt
= nd
->path
.mnt
;
1215 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1220 * Rename seqlock is not required here because in the off chance
1221 * of a false negative due to a concurrent rename, we're going to
1222 * do the non-racy lookup, below.
1224 if (nd
->flags
& LOOKUP_RCU
) {
1228 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1230 if (nameidata_drop_rcu(nd
))
1234 /* Memory barrier in read_seqcount_begin of child is enough */
1235 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1239 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1240 dentry
= do_revalidate_rcu(dentry
, nd
);
1245 if (!(nd
->flags
& LOOKUP_RCU
))
1249 path
->dentry
= dentry
;
1250 if (likely(__follow_mount_rcu(nd
, path
, inode
, false)))
1252 if (nameidata_drop_rcu(nd
))
1256 dentry
= __d_lookup(parent
, name
);
1260 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1261 dentry
= do_revalidate(dentry
, nd
);
1269 path
->dentry
= dentry
;
1270 err
= follow_managed(path
, nd
->flags
);
1271 if (unlikely(err
< 0)) {
1272 path_put_conditional(path
, nd
);
1275 *inode
= path
->dentry
->d_inode
;
1279 dir
= parent
->d_inode
;
1280 BUG_ON(nd
->inode
!= dir
);
1282 mutex_lock(&dir
->i_mutex
);
1284 * First re-do the cached lookup just in case it was created
1285 * while we waited for the directory semaphore, or the first
1286 * lookup failed due to an unrelated rename.
1288 * This could use version numbering or similar to avoid unnecessary
1289 * cache lookups, but then we'd have to do the first lookup in the
1290 * non-racy way. However in the common case here, everything should
1291 * be hot in cache, so would it be a big win?
1293 dentry
= d_lookup(parent
, name
);
1294 if (likely(!dentry
)) {
1295 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1296 mutex_unlock(&dir
->i_mutex
);
1302 * Uhhuh! Nasty case: the cache was re-populated while
1303 * we waited on the semaphore. Need to revalidate.
1305 mutex_unlock(&dir
->i_mutex
);
1309 return PTR_ERR(dentry
);
1312 static inline int may_lookup(struct nameidata
*nd
)
1314 if (nd
->flags
& LOOKUP_RCU
) {
1315 int err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1318 if (nameidata_drop_rcu(nd
))
1321 return exec_permission(nd
->inode
, 0);
1324 static inline int handle_dots(struct nameidata
*nd
, int type
)
1326 if (type
== LAST_DOTDOT
) {
1327 if (nd
->flags
& LOOKUP_RCU
) {
1328 if (follow_dotdot_rcu(nd
))
1336 static void terminate_walk(struct nameidata
*nd
)
1338 if (!(nd
->flags
& LOOKUP_RCU
)) {
1339 path_put(&nd
->path
);
1341 nd
->flags
&= ~LOOKUP_RCU
;
1342 nd
->root
.mnt
= NULL
;
1344 br_read_unlock(vfsmount_lock
);
1350 * This is the basic name resolution function, turning a pathname into
1351 * the final dentry. We expect 'base' to be positive and a directory.
1353 * Returns 0 and nd will have valid dentry and mnt on success.
1354 * Returns error and drops reference to input namei data on failure.
1356 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1360 unsigned int lookup_flags
= nd
->flags
;
1368 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
1370 /* At this point we know we have a real path component. */
1372 struct inode
*inode
;
1378 nd
->flags
|= LOOKUP_CONTINUE
;
1380 err
= may_lookup(nd
);
1385 c
= *(const unsigned char *)name
;
1387 hash
= init_name_hash();
1390 hash
= partial_name_hash(c
, hash
);
1391 c
= *(const unsigned char *)name
;
1392 } while (c
&& (c
!= '/'));
1393 this.len
= name
- (const char *) this.name
;
1394 this.hash
= end_name_hash(hash
);
1397 if (this.name
[0] == '.') switch (this.len
) {
1399 if (this.name
[1] == '.') {
1401 nd
->flags
|= LOOKUP_JUMPED
;
1407 if (likely(type
== LAST_NORM
)) {
1408 struct dentry
*parent
= nd
->path
.dentry
;
1409 nd
->flags
&= ~LOOKUP_JUMPED
;
1410 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1411 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1418 /* remove trailing slashes? */
1420 goto last_component
;
1421 while (*++name
== '/');
1423 goto last_with_slashes
;
1426 * "." and ".." are special - ".." especially so because it has
1427 * to be able to know about the current root directory and
1428 * parent relationships.
1430 if (unlikely(type
!= LAST_NORM
)) {
1431 if (handle_dots(nd
, type
))
1436 /* This does the actual lookups.. */
1437 err
= do_lookup(nd
, &this, &next
, &inode
);
1441 if (inode
&& inode
->i_op
->follow_link
) {
1442 err
= do_follow_link(inode
, &next
, nd
);
1445 nd
->inode
= nd
->path
.dentry
->d_inode
;
1447 path_to_nameidata(&next
, nd
);
1454 if (!nd
->inode
->i_op
->lookup
)
1457 /* here ends the main loop */
1460 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1462 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1463 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1464 if (lookup_flags
& LOOKUP_PARENT
)
1466 if (unlikely(type
!= LAST_NORM
))
1467 return handle_dots(nd
, type
);
1468 err
= do_lookup(nd
, &this, &next
, &inode
);
1471 if (inode
&& unlikely(inode
->i_op
->follow_link
) &&
1472 (lookup_flags
& LOOKUP_FOLLOW
)) {
1473 err
= do_follow_link(inode
, &next
, nd
);
1476 nd
->inode
= nd
->path
.dentry
->d_inode
;
1478 path_to_nameidata(&next
, nd
);
1484 if (lookup_flags
& LOOKUP_DIRECTORY
) {
1486 if (!nd
->inode
->i_op
->lookup
)
1492 nd
->last_type
= type
;
1499 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1500 struct nameidata
*nd
, struct file
**fp
)
1506 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1507 nd
->flags
= flags
| LOOKUP_JUMPED
;
1509 nd
->root
.mnt
= NULL
;
1512 if (flags
& LOOKUP_RCU
) {
1513 br_read_lock(vfsmount_lock
);
1518 path_get(&nd
->root
);
1520 nd
->path
= nd
->root
;
1521 } else if (dfd
== AT_FDCWD
) {
1522 if (flags
& LOOKUP_RCU
) {
1523 struct fs_struct
*fs
= current
->fs
;
1526 br_read_lock(vfsmount_lock
);
1530 seq
= read_seqcount_begin(&fs
->seq
);
1532 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1533 } while (read_seqcount_retry(&fs
->seq
, seq
));
1535 get_fs_pwd(current
->fs
, &nd
->path
);
1538 struct dentry
*dentry
;
1540 file
= fget_light(dfd
, &fput_needed
);
1545 dentry
= file
->f_path
.dentry
;
1548 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1551 retval
= file_permission(file
, MAY_EXEC
);
1555 nd
->path
= file
->f_path
;
1556 if (flags
& LOOKUP_RCU
) {
1559 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1560 br_read_lock(vfsmount_lock
);
1563 path_get(&file
->f_path
);
1564 fput_light(file
, fput_needed
);
1568 nd
->inode
= nd
->path
.dentry
->d_inode
;
1572 fput_light(file
, fput_needed
);
1577 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1578 static int path_lookupat(int dfd
, const char *name
,
1579 unsigned int flags
, struct nameidata
*nd
)
1581 struct file
*base
= NULL
;
1585 * Path walking is largely split up into 2 different synchronisation
1586 * schemes, rcu-walk and ref-walk (explained in
1587 * Documentation/filesystems/path-lookup.txt). These share much of the
1588 * path walk code, but some things particularly setup, cleanup, and
1589 * following mounts are sufficiently divergent that functions are
1590 * duplicated. Typically there is a function foo(), and its RCU
1591 * analogue, foo_rcu().
1593 * -ECHILD is the error number of choice (just to avoid clashes) that
1594 * is returned if some aspect of an rcu-walk fails. Such an error must
1595 * be handled by restarting a traditional ref-walk (which will always
1596 * be able to complete).
1598 retval
= path_init(dfd
, name
, flags
, nd
, &base
);
1600 if (unlikely(retval
))
1603 current
->total_link_count
= 0;
1604 retval
= link_path_walk(name
, nd
);
1606 if (nd
->flags
& LOOKUP_RCU
) {
1607 /* went all way through without dropping RCU */
1609 if (nameidata_drop_rcu_last(nd
))
1614 retval
= handle_reval_path(nd
);
1620 path_put(&nd
->root
);
1621 nd
->root
.mnt
= NULL
;
1626 static int do_path_lookup(int dfd
, const char *name
,
1627 unsigned int flags
, struct nameidata
*nd
)
1629 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1630 if (unlikely(retval
== -ECHILD
))
1631 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1632 if (unlikely(retval
== -ESTALE
))
1633 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1635 if (likely(!retval
)) {
1636 if (unlikely(!audit_dummy_context())) {
1637 if (nd
->path
.dentry
&& nd
->inode
)
1638 audit_inode(name
, nd
->path
.dentry
);
1644 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1646 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1649 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1651 struct nameidata nd
;
1652 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1659 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1660 * @dentry: pointer to dentry of the base directory
1661 * @mnt: pointer to vfs mount of the base directory
1662 * @name: pointer to file name
1663 * @flags: lookup flags
1664 * @nd: pointer to nameidata
1666 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1667 const char *name
, unsigned int flags
,
1668 struct nameidata
*nd
)
1672 /* same as do_path_lookup */
1673 nd
->last_type
= LAST_ROOT
;
1674 nd
->flags
= flags
| LOOKUP_JUMPED
;
1677 nd
->path
.dentry
= dentry
;
1679 path_get(&nd
->path
);
1680 nd
->root
= nd
->path
;
1681 path_get(&nd
->root
);
1682 nd
->inode
= nd
->path
.dentry
->d_inode
;
1684 current
->total_link_count
= 0;
1686 result
= link_path_walk(name
, nd
);
1688 result
= handle_reval_path(nd
);
1689 if (result
== -ESTALE
) {
1690 /* nd->path had been dropped */
1691 current
->total_link_count
= 0;
1692 nd
->path
.dentry
= dentry
;
1694 nd
->inode
= dentry
->d_inode
;
1695 path_get(&nd
->path
);
1696 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_REVAL
;
1698 result
= link_path_walk(name
, nd
);
1700 result
= handle_reval_path(nd
);
1702 if (unlikely(!result
&& !audit_dummy_context() && nd
->path
.dentry
&&
1704 audit_inode(name
, nd
->path
.dentry
);
1706 path_put(&nd
->root
);
1707 nd
->root
.mnt
= NULL
;
1712 static struct dentry
*__lookup_hash(struct qstr
*name
,
1713 struct dentry
*base
, struct nameidata
*nd
)
1715 struct inode
*inode
= base
->d_inode
;
1716 struct dentry
*dentry
;
1719 err
= exec_permission(inode
, 0);
1721 return ERR_PTR(err
);
1724 * Don't bother with __d_lookup: callers are for creat as
1725 * well as unlink, so a lot of the time it would cost
1728 dentry
= d_lookup(base
, name
);
1730 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1731 dentry
= do_revalidate(dentry
, nd
);
1734 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1740 * Restricted form of lookup. Doesn't follow links, single-component only,
1741 * needs parent already locked. Doesn't follow mounts.
1744 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1746 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1750 * lookup_one_len - filesystem helper to lookup single pathname component
1751 * @name: pathname component to lookup
1752 * @base: base directory to lookup from
1753 * @len: maximum length @len should be interpreted to
1755 * Note that this routine is purely a helper for filesystem usage and should
1756 * not be called by generic code. Also note that by using this function the
1757 * nameidata argument is passed to the filesystem methods and a filesystem
1758 * using this helper needs to be prepared for that.
1760 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1766 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1771 return ERR_PTR(-EACCES
);
1773 hash
= init_name_hash();
1775 c
= *(const unsigned char *)name
++;
1776 if (c
== '/' || c
== '\0')
1777 return ERR_PTR(-EACCES
);
1778 hash
= partial_name_hash(c
, hash
);
1780 this.hash
= end_name_hash(hash
);
1782 * See if the low-level filesystem might want
1783 * to use its own hash..
1785 if (base
->d_flags
& DCACHE_OP_HASH
) {
1786 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1788 return ERR_PTR(err
);
1791 return __lookup_hash(&this, base
, NULL
);
1794 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1797 struct nameidata nd
;
1798 char *tmp
= getname(name
);
1799 int err
= PTR_ERR(tmp
);
1802 BUG_ON(flags
& LOOKUP_PARENT
);
1804 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1812 static int user_path_parent(int dfd
, const char __user
*path
,
1813 struct nameidata
*nd
, char **name
)
1815 char *s
= getname(path
);
1821 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1831 * It's inline, so penalty for filesystems that don't use sticky bit is
1834 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1836 uid_t fsuid
= current_fsuid();
1838 if (!(dir
->i_mode
& S_ISVTX
))
1840 if (inode
->i_uid
== fsuid
)
1842 if (dir
->i_uid
== fsuid
)
1844 return !capable(CAP_FOWNER
);
1848 * Check whether we can remove a link victim from directory dir, check
1849 * whether the type of victim is right.
1850 * 1. We can't do it if dir is read-only (done in permission())
1851 * 2. We should have write and exec permissions on dir
1852 * 3. We can't remove anything from append-only dir
1853 * 4. We can't do anything with immutable dir (done in permission())
1854 * 5. If the sticky bit on dir is set we should either
1855 * a. be owner of dir, or
1856 * b. be owner of victim, or
1857 * c. have CAP_FOWNER capability
1858 * 6. If the victim is append-only or immutable we can't do antyhing with
1859 * links pointing to it.
1860 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1861 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1862 * 9. We can't remove a root or mountpoint.
1863 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1864 * nfs_async_unlink().
1866 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1870 if (!victim
->d_inode
)
1873 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1874 audit_inode_child(victim
, dir
);
1876 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1881 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1882 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1885 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1887 if (IS_ROOT(victim
))
1889 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1891 if (IS_DEADDIR(dir
))
1893 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1898 /* Check whether we can create an object with dentry child in directory
1900 * 1. We can't do it if child already exists (open has special treatment for
1901 * this case, but since we are inlined it's OK)
1902 * 2. We can't do it if dir is read-only (done in permission())
1903 * 3. We should have write and exec permissions on dir
1904 * 4. We can't do it if dir is immutable (done in permission())
1906 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1910 if (IS_DEADDIR(dir
))
1912 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1916 * p1 and p2 should be directories on the same fs.
1918 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1923 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1927 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1929 p
= d_ancestor(p2
, p1
);
1931 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1932 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1936 p
= d_ancestor(p1
, p2
);
1938 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1939 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1943 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1944 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1948 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1950 mutex_unlock(&p1
->d_inode
->i_mutex
);
1952 mutex_unlock(&p2
->d_inode
->i_mutex
);
1953 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1957 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1958 struct nameidata
*nd
)
1960 int error
= may_create(dir
, dentry
);
1965 if (!dir
->i_op
->create
)
1966 return -EACCES
; /* shouldn't it be ENOSYS? */
1969 error
= security_inode_create(dir
, dentry
, mode
);
1972 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1974 fsnotify_create(dir
, dentry
);
1978 int may_open(struct path
*path
, int acc_mode
, int flag
)
1980 struct dentry
*dentry
= path
->dentry
;
1981 struct inode
*inode
= dentry
->d_inode
;
1987 switch (inode
->i_mode
& S_IFMT
) {
1991 if (acc_mode
& MAY_WRITE
)
1996 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2005 error
= inode_permission(inode
, acc_mode
);
2010 * An append-only file must be opened in append mode for writing.
2012 if (IS_APPEND(inode
)) {
2013 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2019 /* O_NOATIME can only be set by the owner or superuser */
2020 if (flag
& O_NOATIME
&& !is_owner_or_cap(inode
))
2024 * Ensure there are no outstanding leases on the file.
2026 return break_lease(inode
, flag
);
2029 static int handle_truncate(struct file
*filp
)
2031 struct path
*path
= &filp
->f_path
;
2032 struct inode
*inode
= path
->dentry
->d_inode
;
2033 int error
= get_write_access(inode
);
2037 * Refuse to truncate files with mandatory locks held on them.
2039 error
= locks_verify_locked(inode
);
2041 error
= security_path_truncate(path
);
2043 error
= do_truncate(path
->dentry
, 0,
2044 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2047 put_write_access(inode
);
2052 * Be careful about ever adding any more callers of this
2053 * function. Its flags must be in the namei format, not
2054 * what get passed to sys_open().
2056 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
2057 int open_flag
, int mode
)
2060 struct dentry
*dir
= nd
->path
.dentry
;
2062 if (!IS_POSIXACL(dir
->d_inode
))
2063 mode
&= ~current_umask();
2064 error
= security_path_mknod(&nd
->path
, path
->dentry
, mode
, 0);
2067 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
2069 mutex_unlock(&dir
->d_inode
->i_mutex
);
2070 dput(nd
->path
.dentry
);
2071 nd
->path
.dentry
= path
->dentry
;
2075 /* Don't check for write permission, don't truncate */
2076 return may_open(&nd
->path
, 0, open_flag
& ~O_TRUNC
);
2080 * Note that while the flag value (low two bits) for sys_open means:
2085 * it is changed into
2086 * 00 - no permissions needed
2087 * 01 - read-permission
2088 * 10 - write-permission
2090 * for the internal routines (ie open_namei()/follow_link() etc)
2091 * This is more logical, and also allows the 00 "no perm needed"
2092 * to be used for symlinks (where the permissions are checked
2096 static inline int open_to_namei_flags(int flag
)
2098 if ((flag
+1) & O_ACCMODE
)
2103 static int open_will_truncate(int flag
, struct inode
*inode
)
2106 * We'll never write to the fs underlying
2109 if (special_file(inode
->i_mode
))
2111 return (flag
& O_TRUNC
);
2115 * Handle the last step of open()
2117 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2118 const struct open_flags
*op
, const char *pathname
)
2120 struct dentry
*dir
= nd
->path
.dentry
;
2123 struct inode
*inode
;
2126 nd
->flags
&= ~LOOKUP_PARENT
;
2127 nd
->flags
|= op
->intent
;
2129 switch (nd
->last_type
) {
2132 error
= handle_dots(nd
, nd
->last_type
);
2134 return ERR_PTR(error
);
2137 if (nd
->flags
& LOOKUP_RCU
) {
2138 if (nameidata_drop_rcu_last(nd
))
2139 return ERR_PTR(-ECHILD
);
2141 error
= handle_reval_path(nd
);
2144 audit_inode(pathname
, nd
->path
.dentry
);
2145 if (op
->open_flag
& O_CREAT
) {
2151 /* can't be RCU mode here */
2152 error
= handle_reval_path(nd
);
2155 audit_inode(pathname
, dir
);
2159 if (!(op
->open_flag
& O_CREAT
)) {
2160 if (nd
->last
.name
[nd
->last
.len
])
2161 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2162 /* we _can_ be in RCU mode here */
2163 error
= do_lookup(nd
, &nd
->last
, path
, &inode
);
2166 return ERR_PTR(error
);
2169 path_to_nameidata(path
, nd
);
2171 return ERR_PTR(-ENOENT
);
2173 if (unlikely(inode
->i_op
->follow_link
)) {
2174 /* We drop rcu-walk here */
2175 if (nameidata_dentry_drop_rcu_maybe(nd
, path
->dentry
))
2176 return ERR_PTR(-ECHILD
);
2179 path_to_nameidata(path
, nd
);
2182 if (nd
->flags
& LOOKUP_RCU
) {
2183 if (nameidata_drop_rcu_last(nd
))
2184 return ERR_PTR(-ECHILD
);
2188 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2189 if (!inode
->i_op
->lookup
)
2192 audit_inode(pathname
, nd
->path
.dentry
);
2196 /* create side of things */
2198 if (nd
->flags
& LOOKUP_RCU
) {
2199 if (nameidata_drop_rcu_last(nd
))
2200 return ERR_PTR(-ECHILD
);
2203 audit_inode(pathname
, dir
);
2205 /* trailing slashes? */
2206 if (nd
->last
.name
[nd
->last
.len
])
2209 mutex_lock(&dir
->d_inode
->i_mutex
);
2211 path
->dentry
= lookup_hash(nd
);
2212 path
->mnt
= nd
->path
.mnt
;
2214 error
= PTR_ERR(path
->dentry
);
2215 if (IS_ERR(path
->dentry
)) {
2216 mutex_unlock(&dir
->d_inode
->i_mutex
);
2220 if (IS_ERR(nd
->intent
.open
.file
)) {
2221 error
= PTR_ERR(nd
->intent
.open
.file
);
2222 goto exit_mutex_unlock
;
2225 /* Negative dentry, just create the file */
2226 if (!path
->dentry
->d_inode
) {
2228 * This write is needed to ensure that a
2229 * ro->rw transition does not occur between
2230 * the time when the file is created and when
2231 * a permanent write count is taken through
2232 * the 'struct file' in nameidata_to_filp().
2234 error
= mnt_want_write(nd
->path
.mnt
);
2236 goto exit_mutex_unlock
;
2237 error
= __open_namei_create(nd
, path
, op
->open_flag
, op
->mode
);
2239 mnt_drop_write(nd
->path
.mnt
);
2242 filp
= nameidata_to_filp(nd
);
2243 mnt_drop_write(nd
->path
.mnt
);
2244 path_put(&nd
->path
);
2245 if (!IS_ERR(filp
)) {
2246 error
= ima_file_check(filp
, op
->acc_mode
);
2249 filp
= ERR_PTR(error
);
2256 * It already exists.
2258 mutex_unlock(&dir
->d_inode
->i_mutex
);
2259 audit_inode(pathname
, path
->dentry
);
2262 if (op
->open_flag
& O_EXCL
)
2265 error
= follow_managed(path
, nd
->flags
);
2270 if (!path
->dentry
->d_inode
)
2273 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2276 path_to_nameidata(path
, nd
);
2277 nd
->inode
= path
->dentry
->d_inode
;
2279 if (S_ISDIR(nd
->inode
->i_mode
))
2282 will_truncate
= open_will_truncate(op
->open_flag
, nd
->path
.dentry
->d_inode
);
2283 if (will_truncate
) {
2284 error
= mnt_want_write(nd
->path
.mnt
);
2288 error
= may_open(&nd
->path
, op
->acc_mode
, op
->open_flag
);
2291 mnt_drop_write(nd
->path
.mnt
);
2294 filp
= nameidata_to_filp(nd
);
2295 if (!IS_ERR(filp
)) {
2296 error
= ima_file_check(filp
, op
->acc_mode
);
2299 filp
= ERR_PTR(error
);
2302 if (!IS_ERR(filp
)) {
2303 if (will_truncate
) {
2304 error
= handle_truncate(filp
);
2307 filp
= ERR_PTR(error
);
2312 * It is now safe to drop the mnt write
2313 * because the filp has had a write taken
2317 mnt_drop_write(nd
->path
.mnt
);
2318 path_put(&nd
->path
);
2322 mutex_unlock(&dir
->d_inode
->i_mutex
);
2324 path_put_conditional(path
, nd
);
2326 path_put(&nd
->path
);
2327 return ERR_PTR(error
);
2330 static struct file
*path_openat(int dfd
, const char *pathname
,
2331 const struct open_flags
*op
, int flags
)
2333 struct file
*base
= NULL
;
2335 struct nameidata nd
;
2340 filp
= get_empty_filp();
2342 return ERR_PTR(-ENFILE
);
2344 filp
->f_flags
= op
->open_flag
;
2345 nd
.intent
.open
.file
= filp
;
2346 nd
.intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2347 nd
.intent
.open
.create_mode
= op
->mode
;
2349 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, &nd
, &base
);
2350 if (unlikely(error
))
2353 current
->total_link_count
= 0;
2354 error
= link_path_walk(pathname
, &nd
);
2355 if (unlikely(error
))
2358 filp
= do_last(&nd
, &path
, op
, pathname
);
2359 while (unlikely(!filp
)) { /* trailing symlink */
2360 struct path link
= path
;
2361 struct inode
*linki
= link
.dentry
->d_inode
;
2364 if (!(nd
.flags
& LOOKUP_FOLLOW
))
2369 * This is subtle. Instead of calling do_follow_link() we do
2370 * the thing by hands. The reason is that this way we have zero
2371 * link_count and path_walk() (called from ->follow_link)
2372 * honoring LOOKUP_PARENT. After that we have the parent and
2373 * last component, i.e. we are in the same situation as after
2374 * the first path_walk(). Well, almost - if the last component
2375 * is normal we get its copy stored in nd->last.name and we will
2376 * have to putname() it when we are done. Procfs-like symlinks
2377 * just set LAST_BIND.
2379 nd
.flags
|= LOOKUP_PARENT
;
2380 nd
.flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2381 error
= __do_follow_link(&link
, &nd
, &cookie
);
2382 if (unlikely(error
))
2383 filp
= ERR_PTR(error
);
2385 filp
= do_last(&nd
, &path
, op
, pathname
);
2386 if (!IS_ERR(cookie
) && linki
->i_op
->put_link
)
2387 linki
->i_op
->put_link(link
.dentry
, &nd
, cookie
);
2395 release_open_intent(&nd
);
2399 path_put_conditional(&path
, &nd
);
2402 filp
= ERR_PTR(error
);
2406 struct file
*do_filp_open(int dfd
, const char *pathname
,
2407 const struct open_flags
*op
, int flags
)
2411 filp
= path_openat(dfd
, pathname
, op
, flags
| LOOKUP_RCU
);
2412 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2413 filp
= path_openat(dfd
, pathname
, op
, flags
);
2414 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2415 filp
= path_openat(dfd
, pathname
, op
, flags
| LOOKUP_REVAL
);
2420 * lookup_create - lookup a dentry, creating it if it doesn't exist
2421 * @nd: nameidata info
2422 * @is_dir: directory flag
2424 * Simple function to lookup and return a dentry and create it
2425 * if it doesn't exist. Is SMP-safe.
2427 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2429 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2431 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2433 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2435 * Yucky last component or no last component at all?
2436 * (foo/., foo/.., /////)
2438 if (nd
->last_type
!= LAST_NORM
)
2440 nd
->flags
&= ~LOOKUP_PARENT
;
2441 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2442 nd
->intent
.open
.flags
= O_EXCL
;
2445 * Do the final lookup.
2447 dentry
= lookup_hash(nd
);
2451 if (dentry
->d_inode
)
2454 * Special case - lookup gave negative, but... we had foo/bar/
2455 * From the vfs_mknod() POV we just have a negative dentry -
2456 * all is fine. Let's be bastards - you had / on the end, you've
2457 * been asking for (non-existent) directory. -ENOENT for you.
2459 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2461 dentry
= ERR_PTR(-ENOENT
);
2466 dentry
= ERR_PTR(-EEXIST
);
2470 EXPORT_SYMBOL_GPL(lookup_create
);
2472 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2474 int error
= may_create(dir
, dentry
);
2479 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
2482 if (!dir
->i_op
->mknod
)
2485 error
= devcgroup_inode_mknod(mode
, dev
);
2489 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2493 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2495 fsnotify_create(dir
, dentry
);
2499 static int may_mknod(mode_t mode
)
2501 switch (mode
& S_IFMT
) {
2507 case 0: /* zero mode translates to S_IFREG */
2516 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2521 struct dentry
*dentry
;
2522 struct nameidata nd
;
2527 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2531 dentry
= lookup_create(&nd
, 0);
2532 if (IS_ERR(dentry
)) {
2533 error
= PTR_ERR(dentry
);
2536 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2537 mode
&= ~current_umask();
2538 error
= may_mknod(mode
);
2541 error
= mnt_want_write(nd
.path
.mnt
);
2544 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2546 goto out_drop_write
;
2547 switch (mode
& S_IFMT
) {
2548 case 0: case S_IFREG
:
2549 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2551 case S_IFCHR
: case S_IFBLK
:
2552 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2553 new_decode_dev(dev
));
2555 case S_IFIFO
: case S_IFSOCK
:
2556 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2560 mnt_drop_write(nd
.path
.mnt
);
2564 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2571 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2573 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2576 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2578 int error
= may_create(dir
, dentry
);
2583 if (!dir
->i_op
->mkdir
)
2586 mode
&= (S_IRWXUGO
|S_ISVTX
);
2587 error
= security_inode_mkdir(dir
, dentry
, mode
);
2591 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2593 fsnotify_mkdir(dir
, dentry
);
2597 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2601 struct dentry
*dentry
;
2602 struct nameidata nd
;
2604 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2608 dentry
= lookup_create(&nd
, 1);
2609 error
= PTR_ERR(dentry
);
2613 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2614 mode
&= ~current_umask();
2615 error
= mnt_want_write(nd
.path
.mnt
);
2618 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2620 goto out_drop_write
;
2621 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2623 mnt_drop_write(nd
.path
.mnt
);
2627 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2634 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2636 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2640 * We try to drop the dentry early: we should have
2641 * a usage count of 2 if we're the only user of this
2642 * dentry, and if that is true (possibly after pruning
2643 * the dcache), then we drop the dentry now.
2645 * A low-level filesystem can, if it choses, legally
2648 * if (!d_unhashed(dentry))
2651 * if it cannot handle the case of removing a directory
2652 * that is still in use by something else..
2654 void dentry_unhash(struct dentry
*dentry
)
2657 shrink_dcache_parent(dentry
);
2658 spin_lock(&dentry
->d_lock
);
2659 if (dentry
->d_count
== 2)
2661 spin_unlock(&dentry
->d_lock
);
2664 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2666 int error
= may_delete(dir
, dentry
, 1);
2671 if (!dir
->i_op
->rmdir
)
2674 mutex_lock(&dentry
->d_inode
->i_mutex
);
2675 dentry_unhash(dentry
);
2676 if (d_mountpoint(dentry
))
2679 error
= security_inode_rmdir(dir
, dentry
);
2681 error
= dir
->i_op
->rmdir(dir
, dentry
);
2683 dentry
->d_inode
->i_flags
|= S_DEAD
;
2688 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2697 static long do_rmdir(int dfd
, const char __user
*pathname
)
2701 struct dentry
*dentry
;
2702 struct nameidata nd
;
2704 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2708 switch(nd
.last_type
) {
2720 nd
.flags
&= ~LOOKUP_PARENT
;
2722 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2723 dentry
= lookup_hash(&nd
);
2724 error
= PTR_ERR(dentry
);
2727 error
= mnt_want_write(nd
.path
.mnt
);
2730 error
= security_path_rmdir(&nd
.path
, dentry
);
2733 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2735 mnt_drop_write(nd
.path
.mnt
);
2739 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2746 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2748 return do_rmdir(AT_FDCWD
, pathname
);
2751 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2753 int error
= may_delete(dir
, dentry
, 0);
2758 if (!dir
->i_op
->unlink
)
2761 mutex_lock(&dentry
->d_inode
->i_mutex
);
2762 if (d_mountpoint(dentry
))
2765 error
= security_inode_unlink(dir
, dentry
);
2767 error
= dir
->i_op
->unlink(dir
, dentry
);
2772 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2774 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2775 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2776 fsnotify_link_count(dentry
->d_inode
);
2784 * Make sure that the actual truncation of the file will occur outside its
2785 * directory's i_mutex. Truncate can take a long time if there is a lot of
2786 * writeout happening, and we don't want to prevent access to the directory
2787 * while waiting on the I/O.
2789 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2793 struct dentry
*dentry
;
2794 struct nameidata nd
;
2795 struct inode
*inode
= NULL
;
2797 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2802 if (nd
.last_type
!= LAST_NORM
)
2805 nd
.flags
&= ~LOOKUP_PARENT
;
2807 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2808 dentry
= lookup_hash(&nd
);
2809 error
= PTR_ERR(dentry
);
2810 if (!IS_ERR(dentry
)) {
2811 /* Why not before? Because we want correct error value */
2812 if (nd
.last
.name
[nd
.last
.len
])
2814 inode
= dentry
->d_inode
;
2817 error
= mnt_want_write(nd
.path
.mnt
);
2820 error
= security_path_unlink(&nd
.path
, dentry
);
2823 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2825 mnt_drop_write(nd
.path
.mnt
);
2829 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2831 iput(inode
); /* truncate the inode here */
2838 error
= !dentry
->d_inode
? -ENOENT
:
2839 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2843 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2845 if ((flag
& ~AT_REMOVEDIR
) != 0)
2848 if (flag
& AT_REMOVEDIR
)
2849 return do_rmdir(dfd
, pathname
);
2851 return do_unlinkat(dfd
, pathname
);
2854 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2856 return do_unlinkat(AT_FDCWD
, pathname
);
2859 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2861 int error
= may_create(dir
, dentry
);
2866 if (!dir
->i_op
->symlink
)
2869 error
= security_inode_symlink(dir
, dentry
, oldname
);
2873 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2875 fsnotify_create(dir
, dentry
);
2879 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2880 int, newdfd
, const char __user
*, newname
)
2885 struct dentry
*dentry
;
2886 struct nameidata nd
;
2888 from
= getname(oldname
);
2890 return PTR_ERR(from
);
2892 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2896 dentry
= lookup_create(&nd
, 0);
2897 error
= PTR_ERR(dentry
);
2901 error
= mnt_want_write(nd
.path
.mnt
);
2904 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2906 goto out_drop_write
;
2907 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2909 mnt_drop_write(nd
.path
.mnt
);
2913 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2921 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2923 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2926 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2928 struct inode
*inode
= old_dentry
->d_inode
;
2934 error
= may_create(dir
, new_dentry
);
2938 if (dir
->i_sb
!= inode
->i_sb
)
2942 * A link to an append-only or immutable file cannot be created.
2944 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2946 if (!dir
->i_op
->link
)
2948 if (S_ISDIR(inode
->i_mode
))
2951 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2955 mutex_lock(&inode
->i_mutex
);
2956 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2957 mutex_unlock(&inode
->i_mutex
);
2959 fsnotify_link(dir
, inode
, new_dentry
);
2964 * Hardlinks are often used in delicate situations. We avoid
2965 * security-related surprises by not following symlinks on the
2968 * We don't follow them on the oldname either to be compatible
2969 * with linux 2.0, and to avoid hard-linking to directories
2970 * and other special files. --ADM
2972 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2973 int, newdfd
, const char __user
*, newname
, int, flags
)
2975 struct dentry
*new_dentry
;
2976 struct nameidata nd
;
2977 struct path old_path
;
2981 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
2984 error
= user_path_at(olddfd
, oldname
,
2985 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
2990 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2994 if (old_path
.mnt
!= nd
.path
.mnt
)
2996 new_dentry
= lookup_create(&nd
, 0);
2997 error
= PTR_ERR(new_dentry
);
2998 if (IS_ERR(new_dentry
))
3000 error
= mnt_want_write(nd
.path
.mnt
);
3003 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
3005 goto out_drop_write
;
3006 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
3008 mnt_drop_write(nd
.path
.mnt
);
3012 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3017 path_put(&old_path
);
3022 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3024 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3028 * The worst of all namespace operations - renaming directory. "Perverted"
3029 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3031 * a) we can get into loop creation. Check is done in is_subdir().
3032 * b) race potential - two innocent renames can create a loop together.
3033 * That's where 4.4 screws up. Current fix: serialization on
3034 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3036 * c) we have to lock _three_ objects - parents and victim (if it exists).
3037 * And that - after we got ->i_mutex on parents (until then we don't know
3038 * whether the target exists). Solution: try to be smart with locking
3039 * order for inodes. We rely on the fact that tree topology may change
3040 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3041 * move will be locked. Thus we can rank directories by the tree
3042 * (ancestors first) and rank all non-directories after them.
3043 * That works since everybody except rename does "lock parent, lookup,
3044 * lock child" and rename is under ->s_vfs_rename_mutex.
3045 * HOWEVER, it relies on the assumption that any object with ->lookup()
3046 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3047 * we'd better make sure that there's no link(2) for them.
3048 * d) some filesystems don't support opened-but-unlinked directories,
3049 * either because of layout or because they are not ready to deal with
3050 * all cases correctly. The latter will be fixed (taking this sort of
3051 * stuff into VFS), but the former is not going away. Solution: the same
3052 * trick as in rmdir().
3053 * e) conversion from fhandle to dentry may come in the wrong moment - when
3054 * we are removing the target. Solution: we will have to grab ->i_mutex
3055 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3056 * ->i_mutex on parents, which works but leads to some truly excessive
3059 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3060 struct inode
*new_dir
, struct dentry
*new_dentry
)
3063 struct inode
*target
;
3066 * If we are going to change the parent - check write permissions,
3067 * we'll need to flip '..'.
3069 if (new_dir
!= old_dir
) {
3070 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3075 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3079 target
= new_dentry
->d_inode
;
3081 mutex_lock(&target
->i_mutex
);
3082 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3086 dentry_unhash(new_dentry
);
3087 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3091 target
->i_flags
|= S_DEAD
;
3092 dont_mount(new_dentry
);
3094 mutex_unlock(&target
->i_mutex
);
3095 if (d_unhashed(new_dentry
))
3096 d_rehash(new_dentry
);
3100 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3101 d_move(old_dentry
,new_dentry
);
3105 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3106 struct inode
*new_dir
, struct dentry
*new_dentry
)
3108 struct inode
*target
;
3111 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3116 target
= new_dentry
->d_inode
;
3118 mutex_lock(&target
->i_mutex
);
3119 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3122 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3125 dont_mount(new_dentry
);
3126 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3127 d_move(old_dentry
, new_dentry
);
3130 mutex_unlock(&target
->i_mutex
);
3135 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3136 struct inode
*new_dir
, struct dentry
*new_dentry
)
3139 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3140 const unsigned char *old_name
;
3142 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3145 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3149 if (!new_dentry
->d_inode
)
3150 error
= may_create(new_dir
, new_dentry
);
3152 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3156 if (!old_dir
->i_op
->rename
)
3159 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3162 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3164 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3166 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3167 new_dentry
->d_inode
, old_dentry
);
3168 fsnotify_oldname_free(old_name
);
3173 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3174 int, newdfd
, const char __user
*, newname
)
3176 struct dentry
*old_dir
, *new_dir
;
3177 struct dentry
*old_dentry
, *new_dentry
;
3178 struct dentry
*trap
;
3179 struct nameidata oldnd
, newnd
;
3184 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3188 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3193 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3196 old_dir
= oldnd
.path
.dentry
;
3198 if (oldnd
.last_type
!= LAST_NORM
)
3201 new_dir
= newnd
.path
.dentry
;
3202 if (newnd
.last_type
!= LAST_NORM
)
3205 oldnd
.flags
&= ~LOOKUP_PARENT
;
3206 newnd
.flags
&= ~LOOKUP_PARENT
;
3207 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3209 trap
= lock_rename(new_dir
, old_dir
);
3211 old_dentry
= lookup_hash(&oldnd
);
3212 error
= PTR_ERR(old_dentry
);
3213 if (IS_ERR(old_dentry
))
3215 /* source must exist */
3217 if (!old_dentry
->d_inode
)
3219 /* unless the source is a directory trailing slashes give -ENOTDIR */
3220 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3222 if (oldnd
.last
.name
[oldnd
.last
.len
])
3224 if (newnd
.last
.name
[newnd
.last
.len
])
3227 /* source should not be ancestor of target */
3229 if (old_dentry
== trap
)
3231 new_dentry
= lookup_hash(&newnd
);
3232 error
= PTR_ERR(new_dentry
);
3233 if (IS_ERR(new_dentry
))
3235 /* target should not be an ancestor of source */
3237 if (new_dentry
== trap
)
3240 error
= mnt_want_write(oldnd
.path
.mnt
);
3243 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3244 &newnd
.path
, new_dentry
);
3247 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3248 new_dir
->d_inode
, new_dentry
);
3250 mnt_drop_write(oldnd
.path
.mnt
);
3256 unlock_rename(new_dir
, old_dir
);
3258 path_put(&newnd
.path
);
3261 path_put(&oldnd
.path
);
3267 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3269 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3272 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3276 len
= PTR_ERR(link
);
3281 if (len
> (unsigned) buflen
)
3283 if (copy_to_user(buffer
, link
, len
))
3290 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3291 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3292 * using) it for any given inode is up to filesystem.
3294 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3296 struct nameidata nd
;
3301 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3303 return PTR_ERR(cookie
);
3305 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3306 if (dentry
->d_inode
->i_op
->put_link
)
3307 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3311 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3313 return __vfs_follow_link(nd
, link
);
3316 /* get the link contents into pagecache */
3317 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3321 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3322 page
= read_mapping_page(mapping
, 0, NULL
);
3327 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3331 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3333 struct page
*page
= NULL
;
3334 char *s
= page_getlink(dentry
, &page
);
3335 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3338 page_cache_release(page
);
3343 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3345 struct page
*page
= NULL
;
3346 nd_set_link(nd
, page_getlink(dentry
, &page
));
3350 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3352 struct page
*page
= cookie
;
3356 page_cache_release(page
);
3361 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3363 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3365 struct address_space
*mapping
= inode
->i_mapping
;
3370 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3372 flags
|= AOP_FLAG_NOFS
;
3375 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3376 flags
, &page
, &fsdata
);
3380 kaddr
= kmap_atomic(page
, KM_USER0
);
3381 memcpy(kaddr
, symname
, len
-1);
3382 kunmap_atomic(kaddr
, KM_USER0
);
3384 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3391 mark_inode_dirty(inode
);
3397 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3399 return __page_symlink(inode
, symname
, len
,
3400 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3403 const struct inode_operations page_symlink_inode_operations
= {
3404 .readlink
= generic_readlink
,
3405 .follow_link
= page_follow_link_light
,
3406 .put_link
= page_put_link
,
3409 EXPORT_SYMBOL(user_path_at
);
3410 EXPORT_SYMBOL(follow_down_one
);
3411 EXPORT_SYMBOL(follow_down
);
3412 EXPORT_SYMBOL(follow_up
);
3413 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3414 EXPORT_SYMBOL(getname
);
3415 EXPORT_SYMBOL(lock_rename
);
3416 EXPORT_SYMBOL(lookup_one_len
);
3417 EXPORT_SYMBOL(page_follow_link_light
);
3418 EXPORT_SYMBOL(page_put_link
);
3419 EXPORT_SYMBOL(page_readlink
);
3420 EXPORT_SYMBOL(__page_symlink
);
3421 EXPORT_SYMBOL(page_symlink
);
3422 EXPORT_SYMBOL(page_symlink_inode_operations
);
3423 EXPORT_SYMBOL(kern_path_parent
);
3424 EXPORT_SYMBOL(kern_path
);
3425 EXPORT_SYMBOL(vfs_path_lookup
);
3426 EXPORT_SYMBOL(inode_permission
);
3427 EXPORT_SYMBOL(file_permission
);
3428 EXPORT_SYMBOL(unlock_rename
);
3429 EXPORT_SYMBOL(vfs_create
);
3430 EXPORT_SYMBOL(vfs_follow_link
);
3431 EXPORT_SYMBOL(vfs_link
);
3432 EXPORT_SYMBOL(vfs_mkdir
);
3433 EXPORT_SYMBOL(vfs_mknod
);
3434 EXPORT_SYMBOL(generic_permission
);
3435 EXPORT_SYMBOL(vfs_readlink
);
3436 EXPORT_SYMBOL(vfs_rename
);
3437 EXPORT_SYMBOL(vfs_rmdir
);
3438 EXPORT_SYMBOL(vfs_symlink
);
3439 EXPORT_SYMBOL(vfs_unlink
);
3440 EXPORT_SYMBOL(dentry_unhash
);
3441 EXPORT_SYMBOL(generic_readlink
);