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 return nameidata_dentry_drop_rcu(nd
, dentry
);
507 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
508 * @nd: nameidata pathwalk data to drop
509 * Returns: 0 on success, -ECHILD on failure
511 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
512 * nd->path should be the final element of the lookup, so nd->root is discarded.
513 * Must be called from rcu-walk context.
515 static int nameidata_drop_rcu_last(struct nameidata
*nd
)
517 struct dentry
*dentry
= nd
->path
.dentry
;
519 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
520 nd
->flags
&= ~LOOKUP_RCU
;
522 spin_lock(&dentry
->d_lock
);
523 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
525 BUG_ON(nd
->inode
!= dentry
->d_inode
);
526 spin_unlock(&dentry
->d_lock
);
528 mntget(nd
->path
.mnt
);
531 br_read_unlock(vfsmount_lock
);
536 spin_unlock(&dentry
->d_lock
);
538 br_read_unlock(vfsmount_lock
);
543 * release_open_intent - free up open intent resources
544 * @nd: pointer to nameidata
546 void release_open_intent(struct nameidata
*nd
)
548 struct file
*file
= nd
->intent
.open
.file
;
550 if (file
&& !IS_ERR(file
)) {
551 if (file
->f_path
.dentry
== NULL
)
558 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
560 return dentry
->d_op
->d_revalidate(dentry
, nd
);
563 static struct dentry
*
564 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
566 int status
= d_revalidate(dentry
, nd
);
567 if (unlikely(status
<= 0)) {
569 * The dentry failed validation.
570 * If d_revalidate returned 0 attempt to invalidate
571 * the dentry otherwise d_revalidate is asking us
572 * to return a fail status.
576 dentry
= ERR_PTR(status
);
577 } else if (!d_invalidate(dentry
)) {
585 static inline struct dentry
*
586 do_revalidate_rcu(struct dentry
*dentry
, struct nameidata
*nd
)
588 int status
= d_revalidate(dentry
, nd
);
589 if (likely(status
> 0))
591 if (status
== -ECHILD
) {
592 if (nameidata_dentry_drop_rcu(nd
, dentry
))
593 return ERR_PTR(-ECHILD
);
594 return do_revalidate(dentry
, nd
);
597 return ERR_PTR(status
);
598 /* Don't d_invalidate in rcu-walk mode */
599 if (nameidata_dentry_drop_rcu(nd
, dentry
))
600 return ERR_PTR(-ECHILD
);
601 if (!d_invalidate(dentry
)) {
609 * handle_reval_path - force revalidation of a dentry
611 * In some situations the path walking code will trust dentries without
612 * revalidating them. This causes problems for filesystems that depend on
613 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
614 * (which indicates that it's possible for the dentry to go stale), force
615 * a d_revalidate call before proceeding.
617 * Returns 0 if the revalidation was successful. If the revalidation fails,
618 * either return the error returned by d_revalidate or -ESTALE if the
619 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
620 * invalidate the dentry. It's up to the caller to handle putting references
621 * to the path if necessary.
623 static inline int handle_reval_path(struct nameidata
*nd
)
625 struct dentry
*dentry
= nd
->path
.dentry
;
628 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
631 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
634 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
637 /* Note: we do not d_invalidate() */
638 status
= d_revalidate(dentry
, nd
);
649 * Short-cut version of permission(), for calling on directories
650 * during pathname resolution. Combines parts of permission()
651 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
653 * If appropriate, check DAC only. If not appropriate, or
654 * short-cut DAC fails, then call ->permission() to do more
655 * complete permission check.
657 static inline int exec_permission(struct inode
*inode
, unsigned int flags
)
661 if (inode
->i_op
->permission
) {
662 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
, flags
);
664 ret
= acl_permission_check(inode
, MAY_EXEC
, flags
,
665 inode
->i_op
->check_acl
);
672 if (capable(CAP_DAC_OVERRIDE
) || capable(CAP_DAC_READ_SEARCH
))
677 return security_inode_exec_permission(inode
, flags
);
680 static __always_inline
void set_root(struct nameidata
*nd
)
683 get_fs_root(current
->fs
, &nd
->root
);
686 static int link_path_walk(const char *, struct nameidata
*);
688 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
691 struct fs_struct
*fs
= current
->fs
;
695 seq
= read_seqcount_begin(&fs
->seq
);
697 } while (read_seqcount_retry(&fs
->seq
, seq
));
701 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
713 nd
->flags
|= LOOKUP_JUMPED
;
715 nd
->inode
= nd
->path
.dentry
->d_inode
;
717 ret
= link_path_walk(link
, nd
);
721 return PTR_ERR(link
);
724 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
727 if (path
->mnt
!= nd
->path
.mnt
)
731 static inline void path_to_nameidata(const struct path
*path
,
732 struct nameidata
*nd
)
734 if (!(nd
->flags
& LOOKUP_RCU
)) {
735 dput(nd
->path
.dentry
);
736 if (nd
->path
.mnt
!= path
->mnt
)
737 mntput(nd
->path
.mnt
);
739 nd
->path
.mnt
= path
->mnt
;
740 nd
->path
.dentry
= path
->dentry
;
743 static __always_inline
int
744 __do_follow_link(const struct path
*link
, struct nameidata
*nd
, void **p
)
747 struct dentry
*dentry
= link
->dentry
;
749 BUG_ON(nd
->flags
& LOOKUP_RCU
);
751 touch_atime(link
->mnt
, dentry
);
752 nd_set_link(nd
, NULL
);
754 if (link
->mnt
== nd
->path
.mnt
)
757 error
= security_inode_follow_link(link
->dentry
, nd
);
759 *p
= ERR_PTR(error
); /* no ->put_link(), please */
764 nd
->last_type
= LAST_BIND
;
765 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
768 char *s
= nd_get_link(nd
);
771 error
= __vfs_follow_link(nd
, s
);
772 else if (nd
->last_type
== LAST_BIND
)
773 nd
->flags
|= LOOKUP_JUMPED
;
779 * This limits recursive symlink follows to 8, while
780 * limiting consecutive symlinks to 40.
782 * Without that kind of total limit, nasty chains of consecutive
783 * symlinks can cause almost arbitrarily long lookups.
785 static inline int do_follow_link(struct inode
*inode
, struct path
*path
, struct nameidata
*nd
)
790 /* We drop rcu-walk here */
791 if (nameidata_dentry_drop_rcu_maybe(nd
, path
->dentry
))
793 BUG_ON(inode
!= path
->dentry
->d_inode
);
795 if (current
->link_count
>= MAX_NESTED_LINKS
)
797 if (current
->total_link_count
>= 40)
799 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
801 current
->link_count
++;
802 current
->total_link_count
++;
804 err
= __do_follow_link(path
, nd
, &cookie
);
805 if (!IS_ERR(cookie
) && path
->dentry
->d_inode
->i_op
->put_link
)
806 path
->dentry
->d_inode
->i_op
->put_link(path
->dentry
, nd
, cookie
);
808 current
->link_count
--;
812 path_put_conditional(path
, nd
);
817 static int follow_up_rcu(struct path
*path
)
819 struct vfsmount
*parent
;
820 struct dentry
*mountpoint
;
822 parent
= path
->mnt
->mnt_parent
;
823 if (parent
== path
->mnt
)
825 mountpoint
= path
->mnt
->mnt_mountpoint
;
826 path
->dentry
= mountpoint
;
831 int follow_up(struct path
*path
)
833 struct vfsmount
*parent
;
834 struct dentry
*mountpoint
;
836 br_read_lock(vfsmount_lock
);
837 parent
= path
->mnt
->mnt_parent
;
838 if (parent
== path
->mnt
) {
839 br_read_unlock(vfsmount_lock
);
843 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
844 br_read_unlock(vfsmount_lock
);
846 path
->dentry
= mountpoint
;
853 * Perform an automount
854 * - return -EISDIR to tell follow_managed() to stop and return the path we
857 static int follow_automount(struct path
*path
, unsigned flags
,
860 struct vfsmount
*mnt
;
863 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
866 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
867 * and this is the terminal part of the path.
869 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
870 return -EISDIR
; /* we actually want to stop here */
872 /* We want to mount if someone is trying to open/create a file of any
873 * type under the mountpoint, wants to traverse through the mountpoint
874 * or wants to open the mounted directory.
876 * We don't want to mount if someone's just doing a stat and they've
877 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
878 * appended a '/' to the name.
880 if (!(flags
& LOOKUP_FOLLOW
) &&
881 !(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
882 LOOKUP_OPEN
| LOOKUP_CREATE
)))
885 current
->total_link_count
++;
886 if (current
->total_link_count
>= 40)
889 mnt
= path
->dentry
->d_op
->d_automount(path
);
892 * The filesystem is allowed to return -EISDIR here to indicate
893 * it doesn't want to automount. For instance, autofs would do
894 * this so that its userspace daemon can mount on this dentry.
896 * However, we can only permit this if it's a terminal point in
897 * the path being looked up; if it wasn't then the remainder of
898 * the path is inaccessible and we should say so.
900 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_CONTINUE
))
905 if (!mnt
) /* mount collision */
908 err
= finish_automount(mnt
, path
);
912 /* Someone else made a mount here whilst we were busy */
919 path
->dentry
= dget(mnt
->mnt_root
);
929 * Handle a dentry that is managed in some way.
930 * - Flagged for transit management (autofs)
931 * - Flagged as mountpoint
932 * - Flagged as automount point
934 * This may only be called in refwalk mode.
936 * Serialization is taken care of in namespace.c
938 static int follow_managed(struct path
*path
, unsigned flags
)
941 bool need_mntput
= false;
944 /* Given that we're not holding a lock here, we retain the value in a
945 * local variable for each dentry as we look at it so that we don't see
946 * the components of that value change under us */
947 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
948 managed
&= DCACHE_MANAGED_DENTRY
,
949 unlikely(managed
!= 0)) {
950 /* Allow the filesystem to manage the transit without i_mutex
952 if (managed
& DCACHE_MANAGE_TRANSIT
) {
953 BUG_ON(!path
->dentry
->d_op
);
954 BUG_ON(!path
->dentry
->d_op
->d_manage
);
955 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
,
958 return ret
== -EISDIR
? 0 : ret
;
961 /* Transit to a mounted filesystem. */
962 if (managed
& DCACHE_MOUNTED
) {
963 struct vfsmount
*mounted
= lookup_mnt(path
);
969 path
->dentry
= dget(mounted
->mnt_root
);
974 /* Something is mounted on this dentry in another
975 * namespace and/or whatever was mounted there in this
976 * namespace got unmounted before we managed to get the
980 /* Handle an automount point */
981 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
982 ret
= follow_automount(path
, flags
, &need_mntput
);
984 return ret
== -EISDIR
? 0 : ret
;
988 /* We didn't change the current path point */
994 int follow_down_one(struct path
*path
)
996 struct vfsmount
*mounted
;
998 mounted
= lookup_mnt(path
);
1002 path
->mnt
= mounted
;
1003 path
->dentry
= dget(mounted
->mnt_root
);
1010 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
1011 * meet a managed dentry and we're not walking to "..". True is returned to
1012 * continue, false to abort.
1014 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1015 struct inode
**inode
, bool reverse_transit
)
1017 while (d_mountpoint(path
->dentry
)) {
1018 struct vfsmount
*mounted
;
1019 if (unlikely(path
->dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) &&
1021 path
->dentry
->d_op
->d_manage(path
->dentry
, false, true) < 0)
1023 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1026 path
->mnt
= mounted
;
1027 path
->dentry
= mounted
->mnt_root
;
1028 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1029 *inode
= path
->dentry
->d_inode
;
1032 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1033 return reverse_transit
;
1037 static int follow_dotdot_rcu(struct nameidata
*nd
)
1039 struct inode
*inode
= nd
->inode
;
1044 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1045 nd
->path
.mnt
== nd
->root
.mnt
) {
1048 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1049 struct dentry
*old
= nd
->path
.dentry
;
1050 struct dentry
*parent
= old
->d_parent
;
1053 seq
= read_seqcount_begin(&parent
->d_seq
);
1054 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1056 inode
= parent
->d_inode
;
1057 nd
->path
.dentry
= parent
;
1061 if (!follow_up_rcu(&nd
->path
))
1063 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1064 inode
= nd
->path
.dentry
->d_inode
;
1066 __follow_mount_rcu(nd
, &nd
->path
, &inode
, true);
1073 * Follow down to the covering mount currently visible to userspace. At each
1074 * point, the filesystem owning that dentry may be queried as to whether the
1075 * caller is permitted to proceed or not.
1077 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1080 int follow_down(struct path
*path
, bool mounting_here
)
1085 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1086 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1087 /* Allow the filesystem to manage the transit without i_mutex
1090 * We indicate to the filesystem if someone is trying to mount
1091 * something here. This gives autofs the chance to deny anyone
1092 * other than its daemon the right to mount on its
1095 * The filesystem may sleep at this point.
1097 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1098 BUG_ON(!path
->dentry
->d_op
);
1099 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1100 ret
= path
->dentry
->d_op
->d_manage(
1101 path
->dentry
, mounting_here
, false);
1103 return ret
== -EISDIR
? 0 : ret
;
1106 /* Transit to a mounted filesystem. */
1107 if (managed
& DCACHE_MOUNTED
) {
1108 struct vfsmount
*mounted
= lookup_mnt(path
);
1113 path
->mnt
= mounted
;
1114 path
->dentry
= dget(mounted
->mnt_root
);
1118 /* Don't handle automount points here */
1125 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1127 static void follow_mount(struct path
*path
)
1129 while (d_mountpoint(path
->dentry
)) {
1130 struct vfsmount
*mounted
= lookup_mnt(path
);
1135 path
->mnt
= mounted
;
1136 path
->dentry
= dget(mounted
->mnt_root
);
1140 static void follow_dotdot(struct nameidata
*nd
)
1145 struct dentry
*old
= nd
->path
.dentry
;
1147 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1148 nd
->path
.mnt
== nd
->root
.mnt
) {
1151 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1152 /* rare case of legitimate dget_parent()... */
1153 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1157 if (!follow_up(&nd
->path
))
1160 follow_mount(&nd
->path
);
1161 nd
->inode
= nd
->path
.dentry
->d_inode
;
1165 * Allocate a dentry with name and parent, and perform a parent
1166 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1167 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1168 * have verified that no child exists while under i_mutex.
1170 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1171 struct qstr
*name
, struct nameidata
*nd
)
1173 struct inode
*inode
= parent
->d_inode
;
1174 struct dentry
*dentry
;
1177 /* Don't create child dentry for a dead directory. */
1178 if (unlikely(IS_DEADDIR(inode
)))
1179 return ERR_PTR(-ENOENT
);
1181 dentry
= d_alloc(parent
, name
);
1182 if (unlikely(!dentry
))
1183 return ERR_PTR(-ENOMEM
);
1185 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1186 if (unlikely(old
)) {
1194 * It's more convoluted than I'd like it to be, but... it's still fairly
1195 * small and for now I'd prefer to have fast path as straight as possible.
1196 * It _is_ time-critical.
1198 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1199 struct path
*path
, struct inode
**inode
)
1201 struct vfsmount
*mnt
= nd
->path
.mnt
;
1202 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1207 * See if the low-level filesystem might want
1208 * to use its own hash..
1210 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1211 err
= parent
->d_op
->d_hash(parent
, nd
->inode
, name
);
1217 * Rename seqlock is not required here because in the off chance
1218 * of a false negative due to a concurrent rename, we're going to
1219 * do the non-racy lookup, below.
1221 if (nd
->flags
& LOOKUP_RCU
) {
1225 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1227 if (nameidata_drop_rcu(nd
))
1231 /* Memory barrier in read_seqcount_begin of child is enough */
1232 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1236 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1237 dentry
= do_revalidate_rcu(dentry
, nd
);
1242 if (!(nd
->flags
& LOOKUP_RCU
))
1246 path
->dentry
= dentry
;
1247 if (likely(__follow_mount_rcu(nd
, path
, inode
, false)))
1249 if (nameidata_drop_rcu(nd
))
1253 dentry
= __d_lookup(parent
, name
);
1257 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1258 dentry
= do_revalidate(dentry
, nd
);
1266 path
->dentry
= dentry
;
1267 err
= follow_managed(path
, nd
->flags
);
1268 if (unlikely(err
< 0)) {
1269 path_put_conditional(path
, nd
);
1272 *inode
= path
->dentry
->d_inode
;
1276 dir
= parent
->d_inode
;
1277 BUG_ON(nd
->inode
!= dir
);
1279 mutex_lock(&dir
->i_mutex
);
1281 * First re-do the cached lookup just in case it was created
1282 * while we waited for the directory semaphore, or the first
1283 * lookup failed due to an unrelated rename.
1285 * This could use version numbering or similar to avoid unnecessary
1286 * cache lookups, but then we'd have to do the first lookup in the
1287 * non-racy way. However in the common case here, everything should
1288 * be hot in cache, so would it be a big win?
1290 dentry
= d_lookup(parent
, name
);
1291 if (likely(!dentry
)) {
1292 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1293 mutex_unlock(&dir
->i_mutex
);
1299 * Uhhuh! Nasty case: the cache was re-populated while
1300 * we waited on the semaphore. Need to revalidate.
1302 mutex_unlock(&dir
->i_mutex
);
1306 return PTR_ERR(dentry
);
1309 static inline int may_lookup(struct nameidata
*nd
)
1311 if (nd
->flags
& LOOKUP_RCU
) {
1312 int err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1315 if (nameidata_drop_rcu(nd
))
1318 return exec_permission(nd
->inode
, 0);
1321 static inline int handle_dots(struct nameidata
*nd
, int type
)
1323 if (type
== LAST_DOTDOT
) {
1324 if (nd
->flags
& LOOKUP_RCU
) {
1325 if (follow_dotdot_rcu(nd
))
1335 * This is the basic name resolution function, turning a pathname into
1336 * the final dentry. We expect 'base' to be positive and a directory.
1338 * Returns 0 and nd will have valid dentry and mnt on success.
1339 * Returns error and drops reference to input namei data on failure.
1341 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1345 unsigned int lookup_flags
= nd
->flags
;
1353 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
1355 /* At this point we know we have a real path component. */
1357 struct inode
*inode
;
1363 nd
->flags
|= LOOKUP_CONTINUE
;
1365 err
= may_lookup(nd
);
1370 c
= *(const unsigned char *)name
;
1372 hash
= init_name_hash();
1375 hash
= partial_name_hash(c
, hash
);
1376 c
= *(const unsigned char *)name
;
1377 } while (c
&& (c
!= '/'));
1378 this.len
= name
- (const char *) this.name
;
1379 this.hash
= end_name_hash(hash
);
1382 if (this.name
[0] == '.') switch (this.len
) {
1384 if (this.name
[1] == '.') {
1386 nd
->flags
|= LOOKUP_JUMPED
;
1392 if (likely(type
== LAST_NORM
))
1393 nd
->flags
&= ~LOOKUP_JUMPED
;
1395 /* remove trailing slashes? */
1397 goto last_component
;
1398 while (*++name
== '/');
1400 goto last_with_slashes
;
1403 * "." and ".." are special - ".." especially so because it has
1404 * to be able to know about the current root directory and
1405 * parent relationships.
1407 if (unlikely(type
!= LAST_NORM
)) {
1408 err
= handle_dots(nd
, type
);
1414 /* This does the actual lookups.. */
1415 err
= do_lookup(nd
, &this, &next
, &inode
);
1419 if (inode
&& inode
->i_op
->follow_link
) {
1420 err
= do_follow_link(inode
, &next
, nd
);
1423 nd
->inode
= nd
->path
.dentry
->d_inode
;
1425 path_to_nameidata(&next
, nd
);
1432 if (!nd
->inode
->i_op
->lookup
)
1435 /* here ends the main loop */
1438 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1440 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1441 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1442 if (lookup_flags
& LOOKUP_PARENT
)
1444 if (unlikely(type
!= LAST_NORM
)) {
1445 err
= handle_dots(nd
, type
);
1450 err
= do_lookup(nd
, &this, &next
, &inode
);
1453 if (inode
&& unlikely(inode
->i_op
->follow_link
) &&
1454 (lookup_flags
& LOOKUP_FOLLOW
)) {
1455 err
= do_follow_link(inode
, &next
, nd
);
1458 nd
->inode
= nd
->path
.dentry
->d_inode
;
1460 path_to_nameidata(&next
, nd
);
1466 if (lookup_flags
& LOOKUP_DIRECTORY
) {
1468 if (!nd
->inode
->i_op
->lookup
)
1474 nd
->last_type
= type
;
1477 if (!(nd
->flags
& LOOKUP_RCU
))
1478 path_put(&nd
->path
);
1480 if (nd
->flags
& LOOKUP_RCU
) {
1481 nd
->flags
&= ~LOOKUP_RCU
;
1482 nd
->root
.mnt
= NULL
;
1484 br_read_unlock(vfsmount_lock
);
1489 static int path_init(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1495 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1496 nd
->flags
= flags
| LOOKUP_JUMPED
;
1498 nd
->root
.mnt
= NULL
;
1502 if (flags
& LOOKUP_RCU
) {
1503 br_read_lock(vfsmount_lock
);
1508 path_get(&nd
->root
);
1510 nd
->path
= nd
->root
;
1511 } else if (dfd
== AT_FDCWD
) {
1512 if (flags
& LOOKUP_RCU
) {
1513 struct fs_struct
*fs
= current
->fs
;
1516 br_read_lock(vfsmount_lock
);
1520 seq
= read_seqcount_begin(&fs
->seq
);
1522 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1523 } while (read_seqcount_retry(&fs
->seq
, seq
));
1525 get_fs_pwd(current
->fs
, &nd
->path
);
1528 struct dentry
*dentry
;
1530 file
= fget_light(dfd
, &fput_needed
);
1535 dentry
= file
->f_path
.dentry
;
1538 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1541 retval
= file_permission(file
, MAY_EXEC
);
1545 nd
->path
= file
->f_path
;
1546 if (flags
& LOOKUP_RCU
) {
1549 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1550 br_read_lock(vfsmount_lock
);
1553 path_get(&file
->f_path
);
1554 fput_light(file
, fput_needed
);
1558 nd
->inode
= nd
->path
.dentry
->d_inode
;
1562 fput_light(file
, fput_needed
);
1567 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1568 static int path_lookupat(int dfd
, const char *name
,
1569 unsigned int flags
, struct nameidata
*nd
)
1574 * Path walking is largely split up into 2 different synchronisation
1575 * schemes, rcu-walk and ref-walk (explained in
1576 * Documentation/filesystems/path-lookup.txt). These share much of the
1577 * path walk code, but some things particularly setup, cleanup, and
1578 * following mounts are sufficiently divergent that functions are
1579 * duplicated. Typically there is a function foo(), and its RCU
1580 * analogue, foo_rcu().
1582 * -ECHILD is the error number of choice (just to avoid clashes) that
1583 * is returned if some aspect of an rcu-walk fails. Such an error must
1584 * be handled by restarting a traditional ref-walk (which will always
1585 * be able to complete).
1587 retval
= path_init(dfd
, name
, flags
, nd
);
1589 if (unlikely(retval
))
1592 current
->total_link_count
= 0;
1593 retval
= link_path_walk(name
, nd
);
1595 if (nd
->flags
& LOOKUP_RCU
) {
1596 /* went all way through without dropping RCU */
1598 if (nameidata_drop_rcu_last(nd
))
1603 retval
= handle_reval_path(nd
);
1611 path_put(&nd
->root
);
1612 nd
->root
.mnt
= NULL
;
1617 static int do_path_lookup(int dfd
, const char *name
,
1618 unsigned int flags
, struct nameidata
*nd
)
1620 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1621 if (unlikely(retval
== -ECHILD
))
1622 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1623 if (unlikely(retval
== -ESTALE
))
1624 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1626 if (likely(!retval
)) {
1627 if (unlikely(!audit_dummy_context())) {
1628 if (nd
->path
.dentry
&& nd
->inode
)
1629 audit_inode(name
, nd
->path
.dentry
);
1635 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1637 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1640 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1642 struct nameidata nd
;
1643 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1650 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1651 * @dentry: pointer to dentry of the base directory
1652 * @mnt: pointer to vfs mount of the base directory
1653 * @name: pointer to file name
1654 * @flags: lookup flags
1655 * @nd: pointer to nameidata
1657 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1658 const char *name
, unsigned int flags
,
1659 struct nameidata
*nd
)
1663 /* same as do_path_lookup */
1664 nd
->last_type
= LAST_ROOT
;
1665 nd
->flags
= flags
| LOOKUP_JUMPED
;
1668 nd
->path
.dentry
= dentry
;
1670 path_get(&nd
->path
);
1671 nd
->root
= nd
->path
;
1672 path_get(&nd
->root
);
1673 nd
->inode
= nd
->path
.dentry
->d_inode
;
1675 current
->total_link_count
= 0;
1677 result
= link_path_walk(name
, nd
);
1679 result
= handle_reval_path(nd
);
1680 if (result
== -ESTALE
) {
1681 /* nd->path had been dropped */
1682 current
->total_link_count
= 0;
1683 nd
->path
.dentry
= dentry
;
1685 nd
->inode
= dentry
->d_inode
;
1686 path_get(&nd
->path
);
1687 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_REVAL
;
1689 result
= link_path_walk(name
, nd
);
1691 result
= handle_reval_path(nd
);
1693 if (unlikely(!result
&& !audit_dummy_context() && nd
->path
.dentry
&&
1695 audit_inode(name
, nd
->path
.dentry
);
1697 path_put(&nd
->root
);
1698 nd
->root
.mnt
= NULL
;
1703 static struct dentry
*__lookup_hash(struct qstr
*name
,
1704 struct dentry
*base
, struct nameidata
*nd
)
1706 struct inode
*inode
= base
->d_inode
;
1707 struct dentry
*dentry
;
1710 err
= exec_permission(inode
, 0);
1712 return ERR_PTR(err
);
1715 * See if the low-level filesystem might want
1716 * to use its own hash..
1718 if (base
->d_flags
& DCACHE_OP_HASH
) {
1719 err
= base
->d_op
->d_hash(base
, inode
, name
);
1720 dentry
= ERR_PTR(err
);
1726 * Don't bother with __d_lookup: callers are for creat as
1727 * well as unlink, so a lot of the time it would cost
1730 dentry
= d_lookup(base
, name
);
1732 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1733 dentry
= do_revalidate(dentry
, nd
);
1736 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1742 * Restricted form of lookup. Doesn't follow links, single-component only,
1743 * needs parent already locked. Doesn't follow mounts.
1746 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1748 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1751 static int __lookup_one_len(const char *name
, struct qstr
*this,
1752 struct dentry
*base
, int len
)
1762 hash
= init_name_hash();
1764 c
= *(const unsigned char *)name
++;
1765 if (c
== '/' || c
== '\0')
1767 hash
= partial_name_hash(c
, hash
);
1769 this->hash
= end_name_hash(hash
);
1774 * lookup_one_len - filesystem helper to lookup single pathname component
1775 * @name: pathname component to lookup
1776 * @base: base directory to lookup from
1777 * @len: maximum length @len should be interpreted to
1779 * Note that this routine is purely a helper for filesystem usage and should
1780 * not be called by generic code. Also note that by using this function the
1781 * nameidata argument is passed to the filesystem methods and a filesystem
1782 * using this helper needs to be prepared for that.
1784 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1789 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1791 err
= __lookup_one_len(name
, &this, base
, len
);
1793 return ERR_PTR(err
);
1795 return __lookup_hash(&this, base
, NULL
);
1798 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1801 struct nameidata nd
;
1802 char *tmp
= getname(name
);
1803 int err
= PTR_ERR(tmp
);
1806 BUG_ON(flags
& LOOKUP_PARENT
);
1808 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1816 static int user_path_parent(int dfd
, const char __user
*path
,
1817 struct nameidata
*nd
, char **name
)
1819 char *s
= getname(path
);
1825 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1835 * It's inline, so penalty for filesystems that don't use sticky bit is
1838 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1840 uid_t fsuid
= current_fsuid();
1842 if (!(dir
->i_mode
& S_ISVTX
))
1844 if (inode
->i_uid
== fsuid
)
1846 if (dir
->i_uid
== fsuid
)
1848 return !capable(CAP_FOWNER
);
1852 * Check whether we can remove a link victim from directory dir, check
1853 * whether the type of victim is right.
1854 * 1. We can't do it if dir is read-only (done in permission())
1855 * 2. We should have write and exec permissions on dir
1856 * 3. We can't remove anything from append-only dir
1857 * 4. We can't do anything with immutable dir (done in permission())
1858 * 5. If the sticky bit on dir is set we should either
1859 * a. be owner of dir, or
1860 * b. be owner of victim, or
1861 * c. have CAP_FOWNER capability
1862 * 6. If the victim is append-only or immutable we can't do antyhing with
1863 * links pointing to it.
1864 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1865 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1866 * 9. We can't remove a root or mountpoint.
1867 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1868 * nfs_async_unlink().
1870 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1874 if (!victim
->d_inode
)
1877 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1878 audit_inode_child(victim
, dir
);
1880 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1885 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1886 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1889 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1891 if (IS_ROOT(victim
))
1893 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1895 if (IS_DEADDIR(dir
))
1897 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1902 /* Check whether we can create an object with dentry child in directory
1904 * 1. We can't do it if child already exists (open has special treatment for
1905 * this case, but since we are inlined it's OK)
1906 * 2. We can't do it if dir is read-only (done in permission())
1907 * 3. We should have write and exec permissions on dir
1908 * 4. We can't do it if dir is immutable (done in permission())
1910 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1914 if (IS_DEADDIR(dir
))
1916 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1920 * p1 and p2 should be directories on the same fs.
1922 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1927 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1931 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1933 p
= d_ancestor(p2
, p1
);
1935 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1936 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1940 p
= d_ancestor(p1
, p2
);
1942 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1943 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1947 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1948 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1952 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1954 mutex_unlock(&p1
->d_inode
->i_mutex
);
1956 mutex_unlock(&p2
->d_inode
->i_mutex
);
1957 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1961 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1962 struct nameidata
*nd
)
1964 int error
= may_create(dir
, dentry
);
1969 if (!dir
->i_op
->create
)
1970 return -EACCES
; /* shouldn't it be ENOSYS? */
1973 error
= security_inode_create(dir
, dentry
, mode
);
1976 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1978 fsnotify_create(dir
, dentry
);
1982 int may_open(struct path
*path
, int acc_mode
, int flag
)
1984 struct dentry
*dentry
= path
->dentry
;
1985 struct inode
*inode
= dentry
->d_inode
;
1991 switch (inode
->i_mode
& S_IFMT
) {
1995 if (acc_mode
& MAY_WRITE
)
2000 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2009 error
= inode_permission(inode
, acc_mode
);
2014 * An append-only file must be opened in append mode for writing.
2016 if (IS_APPEND(inode
)) {
2017 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2023 /* O_NOATIME can only be set by the owner or superuser */
2024 if (flag
& O_NOATIME
&& !is_owner_or_cap(inode
))
2028 * Ensure there are no outstanding leases on the file.
2030 return break_lease(inode
, flag
);
2033 static int handle_truncate(struct file
*filp
)
2035 struct path
*path
= &filp
->f_path
;
2036 struct inode
*inode
= path
->dentry
->d_inode
;
2037 int error
= get_write_access(inode
);
2041 * Refuse to truncate files with mandatory locks held on them.
2043 error
= locks_verify_locked(inode
);
2045 error
= security_path_truncate(path
);
2047 error
= do_truncate(path
->dentry
, 0,
2048 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2051 put_write_access(inode
);
2056 * Be careful about ever adding any more callers of this
2057 * function. Its flags must be in the namei format, not
2058 * what get passed to sys_open().
2060 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
2061 int open_flag
, int mode
)
2064 struct dentry
*dir
= nd
->path
.dentry
;
2066 if (!IS_POSIXACL(dir
->d_inode
))
2067 mode
&= ~current_umask();
2068 error
= security_path_mknod(&nd
->path
, path
->dentry
, mode
, 0);
2071 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
2073 mutex_unlock(&dir
->d_inode
->i_mutex
);
2074 dput(nd
->path
.dentry
);
2075 nd
->path
.dentry
= path
->dentry
;
2079 /* Don't check for write permission, don't truncate */
2080 return may_open(&nd
->path
, 0, open_flag
& ~O_TRUNC
);
2084 * Note that while the flag value (low two bits) for sys_open means:
2089 * it is changed into
2090 * 00 - no permissions needed
2091 * 01 - read-permission
2092 * 10 - write-permission
2094 * for the internal routines (ie open_namei()/follow_link() etc)
2095 * This is more logical, and also allows the 00 "no perm needed"
2096 * to be used for symlinks (where the permissions are checked
2100 static inline int open_to_namei_flags(int flag
)
2102 if ((flag
+1) & O_ACCMODE
)
2107 static int open_will_truncate(int flag
, struct inode
*inode
)
2110 * We'll never write to the fs underlying
2113 if (special_file(inode
->i_mode
))
2115 return (flag
& O_TRUNC
);
2118 static struct file
*finish_open(struct nameidata
*nd
,
2119 int open_flag
, int acc_mode
)
2125 will_truncate
= open_will_truncate(open_flag
, nd
->path
.dentry
->d_inode
);
2126 if (will_truncate
) {
2127 error
= mnt_want_write(nd
->path
.mnt
);
2131 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2134 mnt_drop_write(nd
->path
.mnt
);
2137 filp
= nameidata_to_filp(nd
);
2138 if (!IS_ERR(filp
)) {
2139 error
= ima_file_check(filp
, acc_mode
);
2142 filp
= ERR_PTR(error
);
2145 if (!IS_ERR(filp
)) {
2146 if (will_truncate
) {
2147 error
= handle_truncate(filp
);
2150 filp
= ERR_PTR(error
);
2155 * It is now safe to drop the mnt write
2156 * because the filp has had a write taken
2160 mnt_drop_write(nd
->path
.mnt
);
2161 path_put(&nd
->path
);
2165 path_put(&nd
->path
);
2166 return ERR_PTR(error
);
2170 * Handle O_CREAT case for do_filp_open
2172 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2173 const struct open_flags
*op
, const char *pathname
)
2175 struct dentry
*dir
= nd
->path
.dentry
;
2179 nd
->flags
&= ~LOOKUP_PARENT
;
2180 nd
->flags
|= op
->intent
;
2182 switch (nd
->last_type
) {
2185 dir
= nd
->path
.dentry
;
2189 error
= handle_reval_path(nd
);
2195 error
= handle_reval_path(nd
);
2198 audit_inode(pathname
, dir
);
2203 /* trailing slashes? */
2204 if (nd
->last
.name
[nd
->last
.len
])
2207 mutex_lock(&dir
->d_inode
->i_mutex
);
2209 path
->dentry
= lookup_hash(nd
);
2210 path
->mnt
= nd
->path
.mnt
;
2212 error
= PTR_ERR(path
->dentry
);
2213 if (IS_ERR(path
->dentry
)) {
2214 mutex_unlock(&dir
->d_inode
->i_mutex
);
2218 if (IS_ERR(nd
->intent
.open
.file
)) {
2219 error
= PTR_ERR(nd
->intent
.open
.file
);
2220 goto exit_mutex_unlock
;
2223 /* Negative dentry, just create the file */
2224 if (!path
->dentry
->d_inode
) {
2226 * This write is needed to ensure that a
2227 * ro->rw transition does not occur between
2228 * the time when the file is created and when
2229 * a permanent write count is taken through
2230 * the 'struct file' in nameidata_to_filp().
2232 error
= mnt_want_write(nd
->path
.mnt
);
2234 goto exit_mutex_unlock
;
2235 error
= __open_namei_create(nd
, path
, op
->open_flag
, op
->mode
);
2237 mnt_drop_write(nd
->path
.mnt
);
2240 filp
= nameidata_to_filp(nd
);
2241 mnt_drop_write(nd
->path
.mnt
);
2242 path_put(&nd
->path
);
2243 if (!IS_ERR(filp
)) {
2244 error
= ima_file_check(filp
, op
->acc_mode
);
2247 filp
= ERR_PTR(error
);
2254 * It already exists.
2256 mutex_unlock(&dir
->d_inode
->i_mutex
);
2257 audit_inode(pathname
, path
->dentry
);
2260 if (op
->open_flag
& O_EXCL
)
2263 error
= follow_managed(path
, nd
->flags
);
2268 if (!path
->dentry
->d_inode
)
2271 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2274 path_to_nameidata(path
, nd
);
2275 nd
->inode
= path
->dentry
->d_inode
;
2277 if (S_ISDIR(nd
->inode
->i_mode
))
2280 filp
= finish_open(nd
, op
->open_flag
, op
->acc_mode
);
2284 mutex_unlock(&dir
->d_inode
->i_mutex
);
2286 path_put_conditional(path
, nd
);
2288 path_put(&nd
->path
);
2289 return ERR_PTR(error
);
2292 static struct file
*path_openat(int dfd
, const char *pathname
,
2293 const struct open_flags
*op
, int flags
)
2296 struct nameidata nd
;
2301 filp
= get_empty_filp();
2303 return ERR_PTR(-ENFILE
);
2305 filp
->f_flags
= op
->open_flag
;
2306 nd
.intent
.open
.file
= filp
;
2307 nd
.intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2308 nd
.intent
.open
.create_mode
= op
->mode
;
2310 if (op
->open_flag
& O_CREAT
)
2313 /* !O_CREAT, simple open */
2314 error
= path_lookupat(dfd
, pathname
, flags
| op
->intent
, &nd
);
2315 if (unlikely(error
))
2318 if (!(nd
.flags
& LOOKUP_FOLLOW
)) {
2319 if (nd
.inode
->i_op
->follow_link
)
2323 if (nd
.flags
& LOOKUP_DIRECTORY
) {
2324 if (!nd
.inode
->i_op
->lookup
)
2327 audit_inode(pathname
, nd
.path
.dentry
);
2328 filp
= finish_open(&nd
, op
->open_flag
, op
->acc_mode
);
2329 release_open_intent(&nd
);
2333 /* OK, have to create the file. Find the parent. */
2334 error
= path_lookupat(dfd
, pathname
, LOOKUP_PARENT
| flags
, &nd
);
2335 if (unlikely(error
))
2337 if (unlikely(!audit_dummy_context()))
2338 audit_inode(pathname
, nd
.path
.dentry
);
2341 * We have the parent and last component.
2343 filp
= do_last(&nd
, &path
, op
, pathname
);
2344 while (unlikely(!filp
)) { /* trailing symlink */
2345 struct path link
= path
;
2346 struct inode
*linki
= link
.dentry
->d_inode
;
2349 if (!(nd
.flags
& LOOKUP_FOLLOW
))
2354 * This is subtle. Instead of calling do_follow_link() we do
2355 * the thing by hands. The reason is that this way we have zero
2356 * link_count and path_walk() (called from ->follow_link)
2357 * honoring LOOKUP_PARENT. After that we have the parent and
2358 * last component, i.e. we are in the same situation as after
2359 * the first path_walk(). Well, almost - if the last component
2360 * is normal we get its copy stored in nd->last.name and we will
2361 * have to putname() it when we are done. Procfs-like symlinks
2362 * just set LAST_BIND.
2364 nd
.flags
|= LOOKUP_PARENT
;
2365 nd
.flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2366 error
= __do_follow_link(&link
, &nd
, &cookie
);
2367 if (unlikely(error
))
2368 filp
= ERR_PTR(error
);
2370 filp
= do_last(&nd
, &path
, op
, pathname
);
2371 if (!IS_ERR(cookie
) && linki
->i_op
->put_link
)
2372 linki
->i_op
->put_link(link
.dentry
, &nd
, cookie
);
2378 release_open_intent(&nd
);
2382 path_put_conditional(&path
, &nd
);
2386 filp
= ERR_PTR(error
);
2390 struct file
*do_filp_open(int dfd
, const char *pathname
,
2391 const struct open_flags
*op
, int flags
)
2395 filp
= path_openat(dfd
, pathname
, op
, flags
| LOOKUP_RCU
);
2396 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2397 filp
= path_openat(dfd
, pathname
, op
, flags
);
2398 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2399 filp
= path_openat(dfd
, pathname
, op
, flags
| LOOKUP_REVAL
);
2404 * lookup_create - lookup a dentry, creating it if it doesn't exist
2405 * @nd: nameidata info
2406 * @is_dir: directory flag
2408 * Simple function to lookup and return a dentry and create it
2409 * if it doesn't exist. Is SMP-safe.
2411 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2413 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2415 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2417 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2419 * Yucky last component or no last component at all?
2420 * (foo/., foo/.., /////)
2422 if (nd
->last_type
!= LAST_NORM
)
2424 nd
->flags
&= ~LOOKUP_PARENT
;
2425 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2426 nd
->intent
.open
.flags
= O_EXCL
;
2429 * Do the final lookup.
2431 dentry
= lookup_hash(nd
);
2435 if (dentry
->d_inode
)
2438 * Special case - lookup gave negative, but... we had foo/bar/
2439 * From the vfs_mknod() POV we just have a negative dentry -
2440 * all is fine. Let's be bastards - you had / on the end, you've
2441 * been asking for (non-existent) directory. -ENOENT for you.
2443 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2445 dentry
= ERR_PTR(-ENOENT
);
2450 dentry
= ERR_PTR(-EEXIST
);
2454 EXPORT_SYMBOL_GPL(lookup_create
);
2456 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2458 int error
= may_create(dir
, dentry
);
2463 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
2466 if (!dir
->i_op
->mknod
)
2469 error
= devcgroup_inode_mknod(mode
, dev
);
2473 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2477 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2479 fsnotify_create(dir
, dentry
);
2483 static int may_mknod(mode_t mode
)
2485 switch (mode
& S_IFMT
) {
2491 case 0: /* zero mode translates to S_IFREG */
2500 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2505 struct dentry
*dentry
;
2506 struct nameidata nd
;
2511 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2515 dentry
= lookup_create(&nd
, 0);
2516 if (IS_ERR(dentry
)) {
2517 error
= PTR_ERR(dentry
);
2520 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2521 mode
&= ~current_umask();
2522 error
= may_mknod(mode
);
2525 error
= mnt_want_write(nd
.path
.mnt
);
2528 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2530 goto out_drop_write
;
2531 switch (mode
& S_IFMT
) {
2532 case 0: case S_IFREG
:
2533 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2535 case S_IFCHR
: case S_IFBLK
:
2536 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2537 new_decode_dev(dev
));
2539 case S_IFIFO
: case S_IFSOCK
:
2540 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2544 mnt_drop_write(nd
.path
.mnt
);
2548 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2555 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2557 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2560 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2562 int error
= may_create(dir
, dentry
);
2567 if (!dir
->i_op
->mkdir
)
2570 mode
&= (S_IRWXUGO
|S_ISVTX
);
2571 error
= security_inode_mkdir(dir
, dentry
, mode
);
2575 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2577 fsnotify_mkdir(dir
, dentry
);
2581 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2585 struct dentry
*dentry
;
2586 struct nameidata nd
;
2588 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2592 dentry
= lookup_create(&nd
, 1);
2593 error
= PTR_ERR(dentry
);
2597 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2598 mode
&= ~current_umask();
2599 error
= mnt_want_write(nd
.path
.mnt
);
2602 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2604 goto out_drop_write
;
2605 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2607 mnt_drop_write(nd
.path
.mnt
);
2611 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2618 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2620 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2624 * We try to drop the dentry early: we should have
2625 * a usage count of 2 if we're the only user of this
2626 * dentry, and if that is true (possibly after pruning
2627 * the dcache), then we drop the dentry now.
2629 * A low-level filesystem can, if it choses, legally
2632 * if (!d_unhashed(dentry))
2635 * if it cannot handle the case of removing a directory
2636 * that is still in use by something else..
2638 void dentry_unhash(struct dentry
*dentry
)
2641 shrink_dcache_parent(dentry
);
2642 spin_lock(&dentry
->d_lock
);
2643 if (dentry
->d_count
== 2)
2645 spin_unlock(&dentry
->d_lock
);
2648 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2650 int error
= may_delete(dir
, dentry
, 1);
2655 if (!dir
->i_op
->rmdir
)
2658 mutex_lock(&dentry
->d_inode
->i_mutex
);
2659 dentry_unhash(dentry
);
2660 if (d_mountpoint(dentry
))
2663 error
= security_inode_rmdir(dir
, dentry
);
2665 error
= dir
->i_op
->rmdir(dir
, dentry
);
2667 dentry
->d_inode
->i_flags
|= S_DEAD
;
2672 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2681 static long do_rmdir(int dfd
, const char __user
*pathname
)
2685 struct dentry
*dentry
;
2686 struct nameidata nd
;
2688 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2692 switch(nd
.last_type
) {
2704 nd
.flags
&= ~LOOKUP_PARENT
;
2706 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2707 dentry
= lookup_hash(&nd
);
2708 error
= PTR_ERR(dentry
);
2711 error
= mnt_want_write(nd
.path
.mnt
);
2714 error
= security_path_rmdir(&nd
.path
, dentry
);
2717 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2719 mnt_drop_write(nd
.path
.mnt
);
2723 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2730 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2732 return do_rmdir(AT_FDCWD
, pathname
);
2735 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2737 int error
= may_delete(dir
, dentry
, 0);
2742 if (!dir
->i_op
->unlink
)
2745 mutex_lock(&dentry
->d_inode
->i_mutex
);
2746 if (d_mountpoint(dentry
))
2749 error
= security_inode_unlink(dir
, dentry
);
2751 error
= dir
->i_op
->unlink(dir
, dentry
);
2756 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2758 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2759 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2760 fsnotify_link_count(dentry
->d_inode
);
2768 * Make sure that the actual truncation of the file will occur outside its
2769 * directory's i_mutex. Truncate can take a long time if there is a lot of
2770 * writeout happening, and we don't want to prevent access to the directory
2771 * while waiting on the I/O.
2773 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2777 struct dentry
*dentry
;
2778 struct nameidata nd
;
2779 struct inode
*inode
= NULL
;
2781 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2786 if (nd
.last_type
!= LAST_NORM
)
2789 nd
.flags
&= ~LOOKUP_PARENT
;
2791 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2792 dentry
= lookup_hash(&nd
);
2793 error
= PTR_ERR(dentry
);
2794 if (!IS_ERR(dentry
)) {
2795 /* Why not before? Because we want correct error value */
2796 if (nd
.last
.name
[nd
.last
.len
])
2798 inode
= dentry
->d_inode
;
2801 error
= mnt_want_write(nd
.path
.mnt
);
2804 error
= security_path_unlink(&nd
.path
, dentry
);
2807 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2809 mnt_drop_write(nd
.path
.mnt
);
2813 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2815 iput(inode
); /* truncate the inode here */
2822 error
= !dentry
->d_inode
? -ENOENT
:
2823 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2827 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2829 if ((flag
& ~AT_REMOVEDIR
) != 0)
2832 if (flag
& AT_REMOVEDIR
)
2833 return do_rmdir(dfd
, pathname
);
2835 return do_unlinkat(dfd
, pathname
);
2838 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2840 return do_unlinkat(AT_FDCWD
, pathname
);
2843 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2845 int error
= may_create(dir
, dentry
);
2850 if (!dir
->i_op
->symlink
)
2853 error
= security_inode_symlink(dir
, dentry
, oldname
);
2857 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2859 fsnotify_create(dir
, dentry
);
2863 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2864 int, newdfd
, const char __user
*, newname
)
2869 struct dentry
*dentry
;
2870 struct nameidata nd
;
2872 from
= getname(oldname
);
2874 return PTR_ERR(from
);
2876 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2880 dentry
= lookup_create(&nd
, 0);
2881 error
= PTR_ERR(dentry
);
2885 error
= mnt_want_write(nd
.path
.mnt
);
2888 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2890 goto out_drop_write
;
2891 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2893 mnt_drop_write(nd
.path
.mnt
);
2897 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2905 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2907 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2910 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2912 struct inode
*inode
= old_dentry
->d_inode
;
2918 error
= may_create(dir
, new_dentry
);
2922 if (dir
->i_sb
!= inode
->i_sb
)
2926 * A link to an append-only or immutable file cannot be created.
2928 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2930 if (!dir
->i_op
->link
)
2932 if (S_ISDIR(inode
->i_mode
))
2935 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2939 mutex_lock(&inode
->i_mutex
);
2940 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2941 mutex_unlock(&inode
->i_mutex
);
2943 fsnotify_link(dir
, inode
, new_dentry
);
2948 * Hardlinks are often used in delicate situations. We avoid
2949 * security-related surprises by not following symlinks on the
2952 * We don't follow them on the oldname either to be compatible
2953 * with linux 2.0, and to avoid hard-linking to directories
2954 * and other special files. --ADM
2956 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2957 int, newdfd
, const char __user
*, newname
, int, flags
)
2959 struct dentry
*new_dentry
;
2960 struct nameidata nd
;
2961 struct path old_path
;
2965 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
2968 error
= user_path_at(olddfd
, oldname
,
2969 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
2974 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2978 if (old_path
.mnt
!= nd
.path
.mnt
)
2980 new_dentry
= lookup_create(&nd
, 0);
2981 error
= PTR_ERR(new_dentry
);
2982 if (IS_ERR(new_dentry
))
2984 error
= mnt_want_write(nd
.path
.mnt
);
2987 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2989 goto out_drop_write
;
2990 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2992 mnt_drop_write(nd
.path
.mnt
);
2996 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3001 path_put(&old_path
);
3006 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3008 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3012 * The worst of all namespace operations - renaming directory. "Perverted"
3013 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3015 * a) we can get into loop creation. Check is done in is_subdir().
3016 * b) race potential - two innocent renames can create a loop together.
3017 * That's where 4.4 screws up. Current fix: serialization on
3018 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3020 * c) we have to lock _three_ objects - parents and victim (if it exists).
3021 * And that - after we got ->i_mutex on parents (until then we don't know
3022 * whether the target exists). Solution: try to be smart with locking
3023 * order for inodes. We rely on the fact that tree topology may change
3024 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3025 * move will be locked. Thus we can rank directories by the tree
3026 * (ancestors first) and rank all non-directories after them.
3027 * That works since everybody except rename does "lock parent, lookup,
3028 * lock child" and rename is under ->s_vfs_rename_mutex.
3029 * HOWEVER, it relies on the assumption that any object with ->lookup()
3030 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3031 * we'd better make sure that there's no link(2) for them.
3032 * d) some filesystems don't support opened-but-unlinked directories,
3033 * either because of layout or because they are not ready to deal with
3034 * all cases correctly. The latter will be fixed (taking this sort of
3035 * stuff into VFS), but the former is not going away. Solution: the same
3036 * trick as in rmdir().
3037 * e) conversion from fhandle to dentry may come in the wrong moment - when
3038 * we are removing the target. Solution: we will have to grab ->i_mutex
3039 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3040 * ->i_mutex on parents, which works but leads to some truly excessive
3043 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3044 struct inode
*new_dir
, struct dentry
*new_dentry
)
3047 struct inode
*target
;
3050 * If we are going to change the parent - check write permissions,
3051 * we'll need to flip '..'.
3053 if (new_dir
!= old_dir
) {
3054 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3059 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3063 target
= new_dentry
->d_inode
;
3065 mutex_lock(&target
->i_mutex
);
3066 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3070 dentry_unhash(new_dentry
);
3071 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3075 target
->i_flags
|= S_DEAD
;
3076 dont_mount(new_dentry
);
3078 mutex_unlock(&target
->i_mutex
);
3079 if (d_unhashed(new_dentry
))
3080 d_rehash(new_dentry
);
3084 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3085 d_move(old_dentry
,new_dentry
);
3089 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3090 struct inode
*new_dir
, struct dentry
*new_dentry
)
3092 struct inode
*target
;
3095 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3100 target
= new_dentry
->d_inode
;
3102 mutex_lock(&target
->i_mutex
);
3103 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3106 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3109 dont_mount(new_dentry
);
3110 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3111 d_move(old_dentry
, new_dentry
);
3114 mutex_unlock(&target
->i_mutex
);
3119 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3120 struct inode
*new_dir
, struct dentry
*new_dentry
)
3123 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3124 const unsigned char *old_name
;
3126 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3129 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3133 if (!new_dentry
->d_inode
)
3134 error
= may_create(new_dir
, new_dentry
);
3136 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3140 if (!old_dir
->i_op
->rename
)
3143 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3146 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3148 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3150 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3151 new_dentry
->d_inode
, old_dentry
);
3152 fsnotify_oldname_free(old_name
);
3157 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3158 int, newdfd
, const char __user
*, newname
)
3160 struct dentry
*old_dir
, *new_dir
;
3161 struct dentry
*old_dentry
, *new_dentry
;
3162 struct dentry
*trap
;
3163 struct nameidata oldnd
, newnd
;
3168 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3172 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3177 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3180 old_dir
= oldnd
.path
.dentry
;
3182 if (oldnd
.last_type
!= LAST_NORM
)
3185 new_dir
= newnd
.path
.dentry
;
3186 if (newnd
.last_type
!= LAST_NORM
)
3189 oldnd
.flags
&= ~LOOKUP_PARENT
;
3190 newnd
.flags
&= ~LOOKUP_PARENT
;
3191 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3193 trap
= lock_rename(new_dir
, old_dir
);
3195 old_dentry
= lookup_hash(&oldnd
);
3196 error
= PTR_ERR(old_dentry
);
3197 if (IS_ERR(old_dentry
))
3199 /* source must exist */
3201 if (!old_dentry
->d_inode
)
3203 /* unless the source is a directory trailing slashes give -ENOTDIR */
3204 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3206 if (oldnd
.last
.name
[oldnd
.last
.len
])
3208 if (newnd
.last
.name
[newnd
.last
.len
])
3211 /* source should not be ancestor of target */
3213 if (old_dentry
== trap
)
3215 new_dentry
= lookup_hash(&newnd
);
3216 error
= PTR_ERR(new_dentry
);
3217 if (IS_ERR(new_dentry
))
3219 /* target should not be an ancestor of source */
3221 if (new_dentry
== trap
)
3224 error
= mnt_want_write(oldnd
.path
.mnt
);
3227 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3228 &newnd
.path
, new_dentry
);
3231 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3232 new_dir
->d_inode
, new_dentry
);
3234 mnt_drop_write(oldnd
.path
.mnt
);
3240 unlock_rename(new_dir
, old_dir
);
3242 path_put(&newnd
.path
);
3245 path_put(&oldnd
.path
);
3251 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3253 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3256 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3260 len
= PTR_ERR(link
);
3265 if (len
> (unsigned) buflen
)
3267 if (copy_to_user(buffer
, link
, len
))
3274 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3275 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3276 * using) it for any given inode is up to filesystem.
3278 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3280 struct nameidata nd
;
3285 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3287 return PTR_ERR(cookie
);
3289 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3290 if (dentry
->d_inode
->i_op
->put_link
)
3291 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3295 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3297 return __vfs_follow_link(nd
, link
);
3300 /* get the link contents into pagecache */
3301 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3305 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3306 page
= read_mapping_page(mapping
, 0, NULL
);
3311 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3315 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3317 struct page
*page
= NULL
;
3318 char *s
= page_getlink(dentry
, &page
);
3319 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3322 page_cache_release(page
);
3327 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3329 struct page
*page
= NULL
;
3330 nd_set_link(nd
, page_getlink(dentry
, &page
));
3334 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3336 struct page
*page
= cookie
;
3340 page_cache_release(page
);
3345 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3347 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3349 struct address_space
*mapping
= inode
->i_mapping
;
3354 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3356 flags
|= AOP_FLAG_NOFS
;
3359 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3360 flags
, &page
, &fsdata
);
3364 kaddr
= kmap_atomic(page
, KM_USER0
);
3365 memcpy(kaddr
, symname
, len
-1);
3366 kunmap_atomic(kaddr
, KM_USER0
);
3368 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3375 mark_inode_dirty(inode
);
3381 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3383 return __page_symlink(inode
, symname
, len
,
3384 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3387 const struct inode_operations page_symlink_inode_operations
= {
3388 .readlink
= generic_readlink
,
3389 .follow_link
= page_follow_link_light
,
3390 .put_link
= page_put_link
,
3393 EXPORT_SYMBOL(user_path_at
);
3394 EXPORT_SYMBOL(follow_down_one
);
3395 EXPORT_SYMBOL(follow_down
);
3396 EXPORT_SYMBOL(follow_up
);
3397 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3398 EXPORT_SYMBOL(getname
);
3399 EXPORT_SYMBOL(lock_rename
);
3400 EXPORT_SYMBOL(lookup_one_len
);
3401 EXPORT_SYMBOL(page_follow_link_light
);
3402 EXPORT_SYMBOL(page_put_link
);
3403 EXPORT_SYMBOL(page_readlink
);
3404 EXPORT_SYMBOL(__page_symlink
);
3405 EXPORT_SYMBOL(page_symlink
);
3406 EXPORT_SYMBOL(page_symlink_inode_operations
);
3407 EXPORT_SYMBOL(kern_path_parent
);
3408 EXPORT_SYMBOL(kern_path
);
3409 EXPORT_SYMBOL(vfs_path_lookup
);
3410 EXPORT_SYMBOL(inode_permission
);
3411 EXPORT_SYMBOL(file_permission
);
3412 EXPORT_SYMBOL(unlock_rename
);
3413 EXPORT_SYMBOL(vfs_create
);
3414 EXPORT_SYMBOL(vfs_follow_link
);
3415 EXPORT_SYMBOL(vfs_link
);
3416 EXPORT_SYMBOL(vfs_mkdir
);
3417 EXPORT_SYMBOL(vfs_mknod
);
3418 EXPORT_SYMBOL(generic_permission
);
3419 EXPORT_SYMBOL(vfs_readlink
);
3420 EXPORT_SYMBOL(vfs_rename
);
3421 EXPORT_SYMBOL(vfs_rmdir
);
3422 EXPORT_SYMBOL(vfs_symlink
);
3423 EXPORT_SYMBOL(vfs_unlink
);
3424 EXPORT_SYMBOL(dentry_unhash
);
3425 EXPORT_SYMBOL(generic_readlink
);