4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user
*filename
, char *page
)
120 unsigned long len
= PATH_MAX
;
122 if (!segment_eq(get_fs(), KERNEL_DS
)) {
123 if ((unsigned long) filename
>= TASK_SIZE
)
125 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
126 len
= TASK_SIZE
- (unsigned long) filename
;
129 retval
= strncpy_from_user(page
, filename
, len
);
133 return -ENAMETOOLONG
;
139 static char *getname_flags(const char __user
* filename
, int flags
)
143 result
= ERR_PTR(-ENOMEM
);
146 int retval
= do_getname(filename
, tmp
);
150 if (retval
!= -ENOENT
|| !(flags
& LOOKUP_EMPTY
)) {
152 result
= ERR_PTR(retval
);
156 audit_getname(result
);
160 char *getname(const char __user
* filename
)
162 return getname_flags(filename
, 0);
165 #ifdef CONFIG_AUDITSYSCALL
166 void putname(const char *name
)
168 if (unlikely(!audit_dummy_context()))
173 EXPORT_SYMBOL(putname
);
177 * This does basic POSIX ACL permission checking
179 static int acl_permission_check(struct inode
*inode
, int mask
, unsigned int flags
,
180 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
182 umode_t mode
= inode
->i_mode
;
184 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
186 if (current_user_ns() != inode_userns(inode
))
189 if (current_fsuid() == inode
->i_uid
)
192 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
193 int error
= check_acl(inode
, mask
, flags
);
194 if (error
!= -EAGAIN
)
198 if (in_group_p(inode
->i_gid
))
204 * If the DACs are ok we don't need any capability check.
206 if ((mask
& ~mode
) == 0)
212 * generic_permission - check for access rights on a Posix-like filesystem
213 * @inode: inode to check access rights for
214 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
215 * @check_acl: optional callback to check for Posix ACLs
216 * @flags: IPERM_FLAG_ flags.
218 * Used to check for read/write/execute permissions on a file.
219 * We use "fsuid" for this, letting us set arbitrary permissions
220 * for filesystem access without changing the "normal" uids which
221 * are used for other things.
223 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
224 * request cannot be satisfied (eg. requires blocking or too much complexity).
225 * It would then be called again in ref-walk mode.
227 int generic_permission(struct inode
*inode
, int mask
, unsigned int flags
,
228 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
233 * Do the basic POSIX ACL permission checks.
235 ret
= acl_permission_check(inode
, mask
, flags
, check_acl
);
240 * Read/write DACs are always overridable.
241 * Executable DACs are overridable if at least one exec bit is set.
243 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
244 if (ns_capable(inode_userns(inode
), CAP_DAC_OVERRIDE
))
248 * Searching includes executable on directories, else just read.
250 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
251 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
252 if (ns_capable(inode_userns(inode
), CAP_DAC_READ_SEARCH
))
259 * inode_permission - check for access rights to a given inode
260 * @inode: inode to check permission on
261 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
263 * Used to check for read/write/execute permissions on an inode.
264 * We use "fsuid" for this, letting us set arbitrary permissions
265 * for filesystem access without changing the "normal" uids which
266 * are used for other things.
268 int inode_permission(struct inode
*inode
, int mask
)
272 if (mask
& MAY_WRITE
) {
273 umode_t mode
= inode
->i_mode
;
276 * Nobody gets write access to a read-only fs.
278 if (IS_RDONLY(inode
) &&
279 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
283 * Nobody gets write access to an immutable file.
285 if (IS_IMMUTABLE(inode
))
289 if (inode
->i_op
->permission
)
290 retval
= inode
->i_op
->permission(inode
, mask
, 0);
292 retval
= generic_permission(inode
, mask
, 0,
293 inode
->i_op
->check_acl
);
298 retval
= devcgroup_inode_permission(inode
, mask
);
302 return security_inode_permission(inode
, mask
);
306 * file_permission - check for additional access rights to a given file
307 * @file: file to check access rights for
308 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
310 * Used to check for read/write/execute permissions on an already opened
314 * Do not use this function in new code. All access checks should
315 * be done using inode_permission().
317 int file_permission(struct file
*file
, int mask
)
319 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
323 * get_write_access() gets write permission for a file.
324 * put_write_access() releases this write permission.
325 * This is used for regular files.
326 * We cannot support write (and maybe mmap read-write shared) accesses and
327 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
328 * can have the following values:
329 * 0: no writers, no VM_DENYWRITE mappings
330 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
331 * > 0: (i_writecount) users are writing to the file.
333 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
334 * except for the cases where we don't hold i_writecount yet. Then we need to
335 * use {get,deny}_write_access() - these functions check the sign and refuse
336 * to do the change if sign is wrong. Exclusion between them is provided by
337 * the inode->i_lock spinlock.
340 int get_write_access(struct inode
* inode
)
342 spin_lock(&inode
->i_lock
);
343 if (atomic_read(&inode
->i_writecount
) < 0) {
344 spin_unlock(&inode
->i_lock
);
347 atomic_inc(&inode
->i_writecount
);
348 spin_unlock(&inode
->i_lock
);
353 int deny_write_access(struct file
* file
)
355 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
357 spin_lock(&inode
->i_lock
);
358 if (atomic_read(&inode
->i_writecount
) > 0) {
359 spin_unlock(&inode
->i_lock
);
362 atomic_dec(&inode
->i_writecount
);
363 spin_unlock(&inode
->i_lock
);
369 * path_get - get a reference to a path
370 * @path: path to get the reference to
372 * Given a path increment the reference count to the dentry and the vfsmount.
374 void path_get(struct path
*path
)
379 EXPORT_SYMBOL(path_get
);
382 * path_put - put a reference to a path
383 * @path: path to put the reference to
385 * Given a path decrement the reference count to the dentry and the vfsmount.
387 void path_put(struct path
*path
)
392 EXPORT_SYMBOL(path_put
);
395 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
396 * @nd: nameidata pathwalk data to drop
397 * Returns: 0 on success, -ECHILD on failure
399 * Path walking has 2 modes, rcu-walk and ref-walk (see
400 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
401 * to drop out of rcu-walk mode and take normal reference counts on dentries
402 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
403 * refcounts at the last known good point before rcu-walk got stuck, so
404 * ref-walk may continue from there. If this is not successful (eg. a seqcount
405 * has changed), then failure is returned and path walk restarts from the
406 * beginning in ref-walk mode.
408 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
409 * ref-walk. Must be called from rcu-walk context.
411 static int nameidata_drop_rcu(struct nameidata
*nd
)
413 struct fs_struct
*fs
= current
->fs
;
414 struct dentry
*dentry
= nd
->path
.dentry
;
417 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
418 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
420 spin_lock(&fs
->lock
);
421 if (nd
->root
.mnt
!= fs
->root
.mnt
||
422 nd
->root
.dentry
!= fs
->root
.dentry
)
425 spin_lock(&dentry
->d_lock
);
426 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
428 BUG_ON(nd
->inode
!= dentry
->d_inode
);
429 spin_unlock(&dentry
->d_lock
);
432 spin_unlock(&fs
->lock
);
434 mntget(nd
->path
.mnt
);
437 br_read_unlock(vfsmount_lock
);
438 nd
->flags
&= ~LOOKUP_RCU
;
441 spin_unlock(&dentry
->d_lock
);
444 spin_unlock(&fs
->lock
);
448 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
449 static inline int nameidata_drop_rcu_maybe(struct nameidata
*nd
)
451 if (nd
->flags
& LOOKUP_RCU
)
452 return nameidata_drop_rcu(nd
);
457 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
458 * @nd: nameidata pathwalk data to drop
459 * @dentry: dentry to drop
460 * Returns: 0 on success, -ECHILD on failure
462 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
463 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
464 * @nd. Must be called from rcu-walk context.
466 static int nameidata_dentry_drop_rcu(struct nameidata
*nd
, struct dentry
*dentry
)
468 struct fs_struct
*fs
= current
->fs
;
469 struct dentry
*parent
= nd
->path
.dentry
;
472 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
473 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
475 spin_lock(&fs
->lock
);
476 if (nd
->root
.mnt
!= fs
->root
.mnt
||
477 nd
->root
.dentry
!= fs
->root
.dentry
)
480 spin_lock(&parent
->d_lock
);
481 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
482 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
485 * If the sequence check on the child dentry passed, then the child has
486 * not been removed from its parent. This means the parent dentry must
487 * be valid and able to take a reference at this point.
489 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
490 BUG_ON(!parent
->d_count
);
492 spin_unlock(&dentry
->d_lock
);
493 spin_unlock(&parent
->d_lock
);
496 spin_unlock(&fs
->lock
);
498 mntget(nd
->path
.mnt
);
501 br_read_unlock(vfsmount_lock
);
502 nd
->flags
&= ~LOOKUP_RCU
;
505 spin_unlock(&dentry
->d_lock
);
506 spin_unlock(&parent
->d_lock
);
509 spin_unlock(&fs
->lock
);
513 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
514 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata
*nd
, struct dentry
*dentry
)
516 if (nd
->flags
& LOOKUP_RCU
) {
517 if (unlikely(nameidata_dentry_drop_rcu(nd
, dentry
))) {
518 nd
->flags
&= ~LOOKUP_RCU
;
519 if (!(nd
->flags
& LOOKUP_ROOT
))
522 br_read_unlock(vfsmount_lock
);
530 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
531 * @nd: nameidata pathwalk data to drop
532 * Returns: 0 on success, -ECHILD on failure
534 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
535 * nd->path should be the final element of the lookup, so nd->root is discarded.
536 * Must be called from rcu-walk context.
538 static int nameidata_drop_rcu_last(struct nameidata
*nd
)
540 struct dentry
*dentry
= nd
->path
.dentry
;
542 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
543 nd
->flags
&= ~LOOKUP_RCU
;
544 if (!(nd
->flags
& LOOKUP_ROOT
))
546 spin_lock(&dentry
->d_lock
);
547 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
549 BUG_ON(nd
->inode
!= dentry
->d_inode
);
550 spin_unlock(&dentry
->d_lock
);
552 mntget(nd
->path
.mnt
);
555 br_read_unlock(vfsmount_lock
);
560 spin_unlock(&dentry
->d_lock
);
562 br_read_unlock(vfsmount_lock
);
567 * release_open_intent - free up open intent resources
568 * @nd: pointer to nameidata
570 void release_open_intent(struct nameidata
*nd
)
572 struct file
*file
= nd
->intent
.open
.file
;
574 if (file
&& !IS_ERR(file
)) {
575 if (file
->f_path
.dentry
== NULL
)
582 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
584 return dentry
->d_op
->d_revalidate(dentry
, nd
);
587 static struct dentry
*
588 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
590 int status
= d_revalidate(dentry
, nd
);
591 if (unlikely(status
<= 0)) {
593 * The dentry failed validation.
594 * If d_revalidate returned 0 attempt to invalidate
595 * the dentry otherwise d_revalidate is asking us
596 * to return a fail status.
600 dentry
= ERR_PTR(status
);
601 } else if (!d_invalidate(dentry
)) {
610 * handle_reval_path - force revalidation of a dentry
612 * In some situations the path walking code will trust dentries without
613 * revalidating them. This causes problems for filesystems that depend on
614 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
615 * (which indicates that it's possible for the dentry to go stale), force
616 * a d_revalidate call before proceeding.
618 * Returns 0 if the revalidation was successful. If the revalidation fails,
619 * either return the error returned by d_revalidate or -ESTALE if the
620 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
621 * invalidate the dentry. It's up to the caller to handle putting references
622 * to the path if necessary.
624 static inline int handle_reval_path(struct nameidata
*nd
)
626 struct dentry
*dentry
= nd
->path
.dentry
;
629 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
632 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
635 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
638 /* Note: we do not d_invalidate() */
639 status
= d_revalidate(dentry
, nd
);
650 * Short-cut version of permission(), for calling on directories
651 * during pathname resolution. Combines parts of permission()
652 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
654 * If appropriate, check DAC only. If not appropriate, or
655 * short-cut DAC fails, then call ->permission() to do more
656 * complete permission check.
658 static inline int exec_permission(struct inode
*inode
, unsigned int flags
)
661 struct user_namespace
*ns
= inode_userns(inode
);
663 if (inode
->i_op
->permission
) {
664 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
, flags
);
666 ret
= acl_permission_check(inode
, MAY_EXEC
, flags
,
667 inode
->i_op
->check_acl
);
674 if (ns_capable(ns
, CAP_DAC_OVERRIDE
) ||
675 ns_capable(ns
, CAP_DAC_READ_SEARCH
))
680 return security_inode_exec_permission(inode
, flags
);
683 static __always_inline
void set_root(struct nameidata
*nd
)
686 get_fs_root(current
->fs
, &nd
->root
);
689 static int link_path_walk(const char *, struct nameidata
*);
691 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
694 struct fs_struct
*fs
= current
->fs
;
698 seq
= read_seqcount_begin(&fs
->seq
);
700 } while (read_seqcount_retry(&fs
->seq
, seq
));
704 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
716 nd
->flags
|= LOOKUP_JUMPED
;
718 nd
->inode
= nd
->path
.dentry
->d_inode
;
720 ret
= link_path_walk(link
, nd
);
724 return PTR_ERR(link
);
727 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
730 if (path
->mnt
!= nd
->path
.mnt
)
734 static inline void path_to_nameidata(const struct path
*path
,
735 struct nameidata
*nd
)
737 if (!(nd
->flags
& LOOKUP_RCU
)) {
738 dput(nd
->path
.dentry
);
739 if (nd
->path
.mnt
!= path
->mnt
)
740 mntput(nd
->path
.mnt
);
742 nd
->path
.mnt
= path
->mnt
;
743 nd
->path
.dentry
= path
->dentry
;
746 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
748 struct inode
*inode
= link
->dentry
->d_inode
;
749 if (!IS_ERR(cookie
) && inode
->i_op
->put_link
)
750 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
754 static __always_inline
int
755 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
758 struct dentry
*dentry
= link
->dentry
;
760 BUG_ON(nd
->flags
& LOOKUP_RCU
);
762 if (link
->mnt
== nd
->path
.mnt
)
765 if (unlikely(current
->total_link_count
>= 40)) {
766 *p
= ERR_PTR(-ELOOP
); /* no ->put_link(), please */
771 current
->total_link_count
++;
773 touch_atime(link
->mnt
, dentry
);
774 nd_set_link(nd
, NULL
);
776 error
= security_inode_follow_link(link
->dentry
, nd
);
778 *p
= ERR_PTR(error
); /* no ->put_link(), please */
783 nd
->last_type
= LAST_BIND
;
784 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
787 char *s
= nd_get_link(nd
);
790 error
= __vfs_follow_link(nd
, s
);
791 else if (nd
->last_type
== LAST_BIND
) {
792 nd
->flags
|= LOOKUP_JUMPED
;
793 nd
->inode
= nd
->path
.dentry
->d_inode
;
794 if (nd
->inode
->i_op
->follow_link
) {
795 /* stepped on a _really_ weird one */
804 static int follow_up_rcu(struct path
*path
)
806 struct vfsmount
*parent
;
807 struct dentry
*mountpoint
;
809 parent
= path
->mnt
->mnt_parent
;
810 if (parent
== path
->mnt
)
812 mountpoint
= path
->mnt
->mnt_mountpoint
;
813 path
->dentry
= mountpoint
;
818 int follow_up(struct path
*path
)
820 struct vfsmount
*parent
;
821 struct dentry
*mountpoint
;
823 br_read_lock(vfsmount_lock
);
824 parent
= path
->mnt
->mnt_parent
;
825 if (parent
== path
->mnt
) {
826 br_read_unlock(vfsmount_lock
);
830 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
831 br_read_unlock(vfsmount_lock
);
833 path
->dentry
= mountpoint
;
840 * Perform an automount
841 * - return -EISDIR to tell follow_managed() to stop and return the path we
844 static int follow_automount(struct path
*path
, unsigned flags
,
847 struct vfsmount
*mnt
;
850 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
853 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
854 * and this is the terminal part of the path.
856 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
857 return -EISDIR
; /* we actually want to stop here */
859 /* We want to mount if someone is trying to open/create a file of any
860 * type under the mountpoint, wants to traverse through the mountpoint
861 * or wants to open the mounted directory.
863 * We don't want to mount if someone's just doing a stat and they've
864 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
865 * appended a '/' to the name.
867 if (!(flags
& LOOKUP_FOLLOW
) &&
868 !(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
869 LOOKUP_OPEN
| LOOKUP_CREATE
)))
872 current
->total_link_count
++;
873 if (current
->total_link_count
>= 40)
876 mnt
= path
->dentry
->d_op
->d_automount(path
);
879 * The filesystem is allowed to return -EISDIR here to indicate
880 * it doesn't want to automount. For instance, autofs would do
881 * this so that its userspace daemon can mount on this dentry.
883 * However, we can only permit this if it's a terminal point in
884 * the path being looked up; if it wasn't then the remainder of
885 * the path is inaccessible and we should say so.
887 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_CONTINUE
))
892 if (!mnt
) /* mount collision */
895 err
= finish_automount(mnt
, path
);
899 /* Someone else made a mount here whilst we were busy */
906 path
->dentry
= dget(mnt
->mnt_root
);
916 * Handle a dentry that is managed in some way.
917 * - Flagged for transit management (autofs)
918 * - Flagged as mountpoint
919 * - Flagged as automount point
921 * This may only be called in refwalk mode.
923 * Serialization is taken care of in namespace.c
925 static int follow_managed(struct path
*path
, unsigned flags
)
928 bool need_mntput
= false;
931 /* Given that we're not holding a lock here, we retain the value in a
932 * local variable for each dentry as we look at it so that we don't see
933 * the components of that value change under us */
934 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
935 managed
&= DCACHE_MANAGED_DENTRY
,
936 unlikely(managed
!= 0)) {
937 /* Allow the filesystem to manage the transit without i_mutex
939 if (managed
& DCACHE_MANAGE_TRANSIT
) {
940 BUG_ON(!path
->dentry
->d_op
);
941 BUG_ON(!path
->dentry
->d_op
->d_manage
);
942 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
944 return ret
== -EISDIR
? 0 : ret
;
947 /* Transit to a mounted filesystem. */
948 if (managed
& DCACHE_MOUNTED
) {
949 struct vfsmount
*mounted
= lookup_mnt(path
);
955 path
->dentry
= dget(mounted
->mnt_root
);
960 /* Something is mounted on this dentry in another
961 * namespace and/or whatever was mounted there in this
962 * namespace got unmounted before we managed to get the
966 /* Handle an automount point */
967 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
968 ret
= follow_automount(path
, flags
, &need_mntput
);
970 return ret
== -EISDIR
? 0 : ret
;
974 /* We didn't change the current path point */
980 int follow_down_one(struct path
*path
)
982 struct vfsmount
*mounted
;
984 mounted
= lookup_mnt(path
);
989 path
->dentry
= dget(mounted
->mnt_root
);
996 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
997 * meet a managed dentry and we're not walking to "..". True is returned to
998 * continue, false to abort.
1000 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1001 struct inode
**inode
, bool reverse_transit
)
1003 while (d_mountpoint(path
->dentry
)) {
1004 struct vfsmount
*mounted
;
1005 if (unlikely(path
->dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) &&
1007 path
->dentry
->d_op
->d_manage(path
->dentry
, true) < 0)
1009 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1012 path
->mnt
= mounted
;
1013 path
->dentry
= mounted
->mnt_root
;
1014 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1015 *inode
= path
->dentry
->d_inode
;
1018 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1019 return reverse_transit
;
1023 static int follow_dotdot_rcu(struct nameidata
*nd
)
1025 struct inode
*inode
= nd
->inode
;
1030 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1031 nd
->path
.mnt
== nd
->root
.mnt
) {
1034 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1035 struct dentry
*old
= nd
->path
.dentry
;
1036 struct dentry
*parent
= old
->d_parent
;
1039 seq
= read_seqcount_begin(&parent
->d_seq
);
1040 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1042 inode
= parent
->d_inode
;
1043 nd
->path
.dentry
= parent
;
1047 if (!follow_up_rcu(&nd
->path
))
1049 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1050 inode
= nd
->path
.dentry
->d_inode
;
1052 __follow_mount_rcu(nd
, &nd
->path
, &inode
, true);
1057 nd
->flags
&= ~LOOKUP_RCU
;
1058 if (!(nd
->flags
& LOOKUP_ROOT
))
1059 nd
->root
.mnt
= NULL
;
1061 br_read_unlock(vfsmount_lock
);
1066 * Follow down to the covering mount currently visible to userspace. At each
1067 * point, the filesystem owning that dentry may be queried as to whether the
1068 * caller is permitted to proceed or not.
1070 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1073 int follow_down(struct path
*path
)
1078 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1079 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1080 /* Allow the filesystem to manage the transit without i_mutex
1083 * We indicate to the filesystem if someone is trying to mount
1084 * something here. This gives autofs the chance to deny anyone
1085 * other than its daemon the right to mount on its
1088 * The filesystem may sleep at this point.
1090 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1091 BUG_ON(!path
->dentry
->d_op
);
1092 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1093 ret
= path
->dentry
->d_op
->d_manage(
1094 path
->dentry
, false);
1096 return ret
== -EISDIR
? 0 : ret
;
1099 /* Transit to a mounted filesystem. */
1100 if (managed
& DCACHE_MOUNTED
) {
1101 struct vfsmount
*mounted
= lookup_mnt(path
);
1106 path
->mnt
= mounted
;
1107 path
->dentry
= dget(mounted
->mnt_root
);
1111 /* Don't handle automount points here */
1118 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1120 static void follow_mount(struct path
*path
)
1122 while (d_mountpoint(path
->dentry
)) {
1123 struct vfsmount
*mounted
= lookup_mnt(path
);
1128 path
->mnt
= mounted
;
1129 path
->dentry
= dget(mounted
->mnt_root
);
1133 static void follow_dotdot(struct nameidata
*nd
)
1138 struct dentry
*old
= nd
->path
.dentry
;
1140 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1141 nd
->path
.mnt
== nd
->root
.mnt
) {
1144 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1145 /* rare case of legitimate dget_parent()... */
1146 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1150 if (!follow_up(&nd
->path
))
1153 follow_mount(&nd
->path
);
1154 nd
->inode
= nd
->path
.dentry
->d_inode
;
1158 * Allocate a dentry with name and parent, and perform a parent
1159 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1160 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1161 * have verified that no child exists while under i_mutex.
1163 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1164 struct qstr
*name
, struct nameidata
*nd
)
1166 struct inode
*inode
= parent
->d_inode
;
1167 struct dentry
*dentry
;
1170 /* Don't create child dentry for a dead directory. */
1171 if (unlikely(IS_DEADDIR(inode
)))
1172 return ERR_PTR(-ENOENT
);
1174 dentry
= d_alloc(parent
, name
);
1175 if (unlikely(!dentry
))
1176 return ERR_PTR(-ENOMEM
);
1178 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1179 if (unlikely(old
)) {
1187 * It's more convoluted than I'd like it to be, but... it's still fairly
1188 * small and for now I'd prefer to have fast path as straight as possible.
1189 * It _is_ time-critical.
1191 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1192 struct path
*path
, struct inode
**inode
)
1194 struct vfsmount
*mnt
= nd
->path
.mnt
;
1195 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1201 * Rename seqlock is not required here because in the off chance
1202 * of a false negative due to a concurrent rename, we're going to
1203 * do the non-racy lookup, below.
1205 if (nd
->flags
& LOOKUP_RCU
) {
1208 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1212 /* Memory barrier in read_seqcount_begin of child is enough */
1213 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1217 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1218 status
= d_revalidate(dentry
, nd
);
1219 if (unlikely(status
<= 0)) {
1220 if (status
!= -ECHILD
)
1226 path
->dentry
= dentry
;
1227 if (likely(__follow_mount_rcu(nd
, path
, inode
, false)))
1231 if (nameidata_dentry_drop_rcu(nd
, dentry
))
1234 if (nameidata_drop_rcu(nd
))
1238 dentry
= __d_lookup(parent
, name
);
1242 if (unlikely(!dentry
)) {
1243 struct inode
*dir
= parent
->d_inode
;
1244 BUG_ON(nd
->inode
!= dir
);
1246 mutex_lock(&dir
->i_mutex
);
1247 dentry
= d_lookup(parent
, name
);
1248 if (likely(!dentry
)) {
1249 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1250 if (IS_ERR(dentry
)) {
1251 mutex_unlock(&dir
->i_mutex
);
1252 return PTR_ERR(dentry
);
1258 mutex_unlock(&dir
->i_mutex
);
1260 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1261 status
= d_revalidate(dentry
, nd
);
1262 if (unlikely(status
<= 0)) {
1267 if (!d_invalidate(dentry
)) {
1276 path
->dentry
= dentry
;
1277 err
= follow_managed(path
, nd
->flags
);
1278 if (unlikely(err
< 0)) {
1279 path_put_conditional(path
, nd
);
1282 *inode
= path
->dentry
->d_inode
;
1286 static inline int may_lookup(struct nameidata
*nd
)
1288 if (nd
->flags
& LOOKUP_RCU
) {
1289 int err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1292 if (nameidata_drop_rcu(nd
))
1295 return exec_permission(nd
->inode
, 0);
1298 static inline int handle_dots(struct nameidata
*nd
, int type
)
1300 if (type
== LAST_DOTDOT
) {
1301 if (nd
->flags
& LOOKUP_RCU
) {
1302 if (follow_dotdot_rcu(nd
))
1310 static void terminate_walk(struct nameidata
*nd
)
1312 if (!(nd
->flags
& LOOKUP_RCU
)) {
1313 path_put(&nd
->path
);
1315 nd
->flags
&= ~LOOKUP_RCU
;
1316 if (!(nd
->flags
& LOOKUP_ROOT
))
1317 nd
->root
.mnt
= NULL
;
1319 br_read_unlock(vfsmount_lock
);
1323 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1324 struct qstr
*name
, int type
, int follow
)
1326 struct inode
*inode
;
1329 * "." and ".." are special - ".." especially so because it has
1330 * to be able to know about the current root directory and
1331 * parent relationships.
1333 if (unlikely(type
!= LAST_NORM
))
1334 return handle_dots(nd
, type
);
1335 err
= do_lookup(nd
, name
, path
, &inode
);
1336 if (unlikely(err
)) {
1341 path_to_nameidata(path
, nd
);
1345 if (unlikely(inode
->i_op
->follow_link
) && follow
) {
1346 if (nameidata_dentry_drop_rcu_maybe(nd
, path
->dentry
))
1348 BUG_ON(inode
!= path
->dentry
->d_inode
);
1351 path_to_nameidata(path
, nd
);
1357 * This limits recursive symlink follows to 8, while
1358 * limiting consecutive symlinks to 40.
1360 * Without that kind of total limit, nasty chains of consecutive
1361 * symlinks can cause almost arbitrarily long lookups.
1363 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1367 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1368 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1369 path_put_conditional(path
, nd
);
1370 path_put(&nd
->path
);
1375 current
->link_count
++;
1378 struct path link
= *path
;
1381 res
= follow_link(&link
, nd
, &cookie
);
1383 res
= walk_component(nd
, path
, &nd
->last
,
1384 nd
->last_type
, LOOKUP_FOLLOW
);
1385 put_link(nd
, &link
, cookie
);
1388 current
->link_count
--;
1395 * This is the basic name resolution function, turning a pathname into
1396 * the final dentry. We expect 'base' to be positive and a directory.
1398 * Returns 0 and nd will have valid dentry and mnt on success.
1399 * Returns error and drops reference to input namei data on failure.
1401 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1405 unsigned int lookup_flags
= nd
->flags
;
1412 /* At this point we know we have a real path component. */
1419 nd
->flags
|= LOOKUP_CONTINUE
;
1421 err
= may_lookup(nd
);
1426 c
= *(const unsigned char *)name
;
1428 hash
= init_name_hash();
1431 hash
= partial_name_hash(c
, hash
);
1432 c
= *(const unsigned char *)name
;
1433 } while (c
&& (c
!= '/'));
1434 this.len
= name
- (const char *) this.name
;
1435 this.hash
= end_name_hash(hash
);
1438 if (this.name
[0] == '.') switch (this.len
) {
1440 if (this.name
[1] == '.') {
1442 nd
->flags
|= LOOKUP_JUMPED
;
1448 if (likely(type
== LAST_NORM
)) {
1449 struct dentry
*parent
= nd
->path
.dentry
;
1450 nd
->flags
&= ~LOOKUP_JUMPED
;
1451 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1452 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1459 /* remove trailing slashes? */
1461 goto last_component
;
1462 while (*++name
== '/');
1464 goto last_component
;
1466 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1471 err
= nested_symlink(&next
, nd
);
1476 if (!nd
->inode
->i_op
->lookup
)
1479 /* here ends the main loop */
1482 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1483 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1485 nd
->last_type
= type
;
1492 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1493 struct nameidata
*nd
, struct file
**fp
)
1499 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1500 nd
->flags
= flags
| LOOKUP_JUMPED
;
1502 if (flags
& LOOKUP_ROOT
) {
1503 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1505 if (!inode
->i_op
->lookup
)
1507 retval
= inode_permission(inode
, MAY_EXEC
);
1511 nd
->path
= nd
->root
;
1513 if (flags
& LOOKUP_RCU
) {
1514 br_read_lock(vfsmount_lock
);
1516 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1518 path_get(&nd
->path
);
1523 nd
->root
.mnt
= NULL
;
1526 if (flags
& LOOKUP_RCU
) {
1527 br_read_lock(vfsmount_lock
);
1532 path_get(&nd
->root
);
1534 nd
->path
= nd
->root
;
1535 } else if (dfd
== AT_FDCWD
) {
1536 if (flags
& LOOKUP_RCU
) {
1537 struct fs_struct
*fs
= current
->fs
;
1540 br_read_lock(vfsmount_lock
);
1544 seq
= read_seqcount_begin(&fs
->seq
);
1546 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1547 } while (read_seqcount_retry(&fs
->seq
, seq
));
1549 get_fs_pwd(current
->fs
, &nd
->path
);
1552 struct dentry
*dentry
;
1554 file
= fget_raw_light(dfd
, &fput_needed
);
1559 dentry
= file
->f_path
.dentry
;
1563 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1566 retval
= file_permission(file
, MAY_EXEC
);
1571 nd
->path
= file
->f_path
;
1572 if (flags
& LOOKUP_RCU
) {
1575 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1576 br_read_lock(vfsmount_lock
);
1579 path_get(&file
->f_path
);
1580 fput_light(file
, fput_needed
);
1584 nd
->inode
= nd
->path
.dentry
->d_inode
;
1588 fput_light(file
, fput_needed
);
1593 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1595 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1596 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1598 nd
->flags
&= ~LOOKUP_PARENT
;
1599 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1600 nd
->flags
& LOOKUP_FOLLOW
);
1603 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1604 static int path_lookupat(int dfd
, const char *name
,
1605 unsigned int flags
, struct nameidata
*nd
)
1607 struct file
*base
= NULL
;
1612 * Path walking is largely split up into 2 different synchronisation
1613 * schemes, rcu-walk and ref-walk (explained in
1614 * Documentation/filesystems/path-lookup.txt). These share much of the
1615 * path walk code, but some things particularly setup, cleanup, and
1616 * following mounts are sufficiently divergent that functions are
1617 * duplicated. Typically there is a function foo(), and its RCU
1618 * analogue, foo_rcu().
1620 * -ECHILD is the error number of choice (just to avoid clashes) that
1621 * is returned if some aspect of an rcu-walk fails. Such an error must
1622 * be handled by restarting a traditional ref-walk (which will always
1623 * be able to complete).
1625 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1630 current
->total_link_count
= 0;
1631 err
= link_path_walk(name
, nd
);
1633 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1634 err
= lookup_last(nd
, &path
);
1637 struct path link
= path
;
1638 nd
->flags
|= LOOKUP_PARENT
;
1639 err
= follow_link(&link
, nd
, &cookie
);
1641 err
= lookup_last(nd
, &path
);
1642 put_link(nd
, &link
, cookie
);
1646 if (nd
->flags
& LOOKUP_RCU
) {
1647 /* went all way through without dropping RCU */
1649 if (nameidata_drop_rcu_last(nd
))
1654 err
= handle_reval_path(nd
);
1656 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1657 if (!nd
->inode
->i_op
->lookup
) {
1658 path_put(&nd
->path
);
1666 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1667 path_put(&nd
->root
);
1668 nd
->root
.mnt
= NULL
;
1673 static int do_path_lookup(int dfd
, const char *name
,
1674 unsigned int flags
, struct nameidata
*nd
)
1676 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1677 if (unlikely(retval
== -ECHILD
))
1678 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1679 if (unlikely(retval
== -ESTALE
))
1680 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1682 if (likely(!retval
)) {
1683 if (unlikely(!audit_dummy_context())) {
1684 if (nd
->path
.dentry
&& nd
->inode
)
1685 audit_inode(name
, nd
->path
.dentry
);
1691 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1693 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1696 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1698 struct nameidata nd
;
1699 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1706 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1707 * @dentry: pointer to dentry of the base directory
1708 * @mnt: pointer to vfs mount of the base directory
1709 * @name: pointer to file name
1710 * @flags: lookup flags
1711 * @nd: pointer to nameidata
1713 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1714 const char *name
, unsigned int flags
,
1715 struct nameidata
*nd
)
1717 nd
->root
.dentry
= dentry
;
1719 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1720 return do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, nd
);
1723 static struct dentry
*__lookup_hash(struct qstr
*name
,
1724 struct dentry
*base
, struct nameidata
*nd
)
1726 struct inode
*inode
= base
->d_inode
;
1727 struct dentry
*dentry
;
1730 err
= exec_permission(inode
, 0);
1732 return ERR_PTR(err
);
1735 * Don't bother with __d_lookup: callers are for creat as
1736 * well as unlink, so a lot of the time it would cost
1739 dentry
= d_lookup(base
, name
);
1741 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1742 dentry
= do_revalidate(dentry
, nd
);
1745 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1751 * Restricted form of lookup. Doesn't follow links, single-component only,
1752 * needs parent already locked. Doesn't follow mounts.
1755 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1757 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1761 * lookup_one_len - filesystem helper to lookup single pathname component
1762 * @name: pathname component to lookup
1763 * @base: base directory to lookup from
1764 * @len: maximum length @len should be interpreted to
1766 * Note that this routine is purely a helper for filesystem usage and should
1767 * not be called by generic code. Also note that by using this function the
1768 * nameidata argument is passed to the filesystem methods and a filesystem
1769 * using this helper needs to be prepared for that.
1771 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1777 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1782 return ERR_PTR(-EACCES
);
1784 hash
= init_name_hash();
1786 c
= *(const unsigned char *)name
++;
1787 if (c
== '/' || c
== '\0')
1788 return ERR_PTR(-EACCES
);
1789 hash
= partial_name_hash(c
, hash
);
1791 this.hash
= end_name_hash(hash
);
1793 * See if the low-level filesystem might want
1794 * to use its own hash..
1796 if (base
->d_flags
& DCACHE_OP_HASH
) {
1797 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1799 return ERR_PTR(err
);
1802 return __lookup_hash(&this, base
, NULL
);
1805 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1808 struct nameidata nd
;
1809 char *tmp
= getname_flags(name
, flags
);
1810 int err
= PTR_ERR(tmp
);
1813 BUG_ON(flags
& LOOKUP_PARENT
);
1815 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1823 static int user_path_parent(int dfd
, const char __user
*path
,
1824 struct nameidata
*nd
, char **name
)
1826 char *s
= getname(path
);
1832 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1842 * It's inline, so penalty for filesystems that don't use sticky bit is
1845 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1847 uid_t fsuid
= current_fsuid();
1849 if (!(dir
->i_mode
& S_ISVTX
))
1851 if (current_user_ns() != inode_userns(inode
))
1853 if (inode
->i_uid
== fsuid
)
1855 if (dir
->i_uid
== fsuid
)
1859 return !ns_capable(inode_userns(inode
), CAP_FOWNER
);
1863 * Check whether we can remove a link victim from directory dir, check
1864 * whether the type of victim is right.
1865 * 1. We can't do it if dir is read-only (done in permission())
1866 * 2. We should have write and exec permissions on dir
1867 * 3. We can't remove anything from append-only dir
1868 * 4. We can't do anything with immutable dir (done in permission())
1869 * 5. If the sticky bit on dir is set we should either
1870 * a. be owner of dir, or
1871 * b. be owner of victim, or
1872 * c. have CAP_FOWNER capability
1873 * 6. If the victim is append-only or immutable we can't do antyhing with
1874 * links pointing to it.
1875 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1876 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1877 * 9. We can't remove a root or mountpoint.
1878 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1879 * nfs_async_unlink().
1881 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1885 if (!victim
->d_inode
)
1888 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1889 audit_inode_child(victim
, dir
);
1891 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1896 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1897 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1900 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1902 if (IS_ROOT(victim
))
1904 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1906 if (IS_DEADDIR(dir
))
1908 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1913 /* Check whether we can create an object with dentry child in directory
1915 * 1. We can't do it if child already exists (open has special treatment for
1916 * this case, but since we are inlined it's OK)
1917 * 2. We can't do it if dir is read-only (done in permission())
1918 * 3. We should have write and exec permissions on dir
1919 * 4. We can't do it if dir is immutable (done in permission())
1921 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1925 if (IS_DEADDIR(dir
))
1927 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1931 * p1 and p2 should be directories on the same fs.
1933 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1938 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1942 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1944 p
= d_ancestor(p2
, p1
);
1946 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1947 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1951 p
= d_ancestor(p1
, p2
);
1953 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1954 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1958 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1959 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1963 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1965 mutex_unlock(&p1
->d_inode
->i_mutex
);
1967 mutex_unlock(&p2
->d_inode
->i_mutex
);
1968 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1972 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1973 struct nameidata
*nd
)
1975 int error
= may_create(dir
, dentry
);
1980 if (!dir
->i_op
->create
)
1981 return -EACCES
; /* shouldn't it be ENOSYS? */
1984 error
= security_inode_create(dir
, dentry
, mode
);
1987 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1989 fsnotify_create(dir
, dentry
);
1993 static int may_open(struct path
*path
, int acc_mode
, int flag
)
1995 struct dentry
*dentry
= path
->dentry
;
1996 struct inode
*inode
= dentry
->d_inode
;
2006 switch (inode
->i_mode
& S_IFMT
) {
2010 if (acc_mode
& MAY_WRITE
)
2015 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2024 error
= inode_permission(inode
, acc_mode
);
2029 * An append-only file must be opened in append mode for writing.
2031 if (IS_APPEND(inode
)) {
2032 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2038 /* O_NOATIME can only be set by the owner or superuser */
2039 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2043 * Ensure there are no outstanding leases on the file.
2045 return break_lease(inode
, flag
);
2048 static int handle_truncate(struct file
*filp
)
2050 struct path
*path
= &filp
->f_path
;
2051 struct inode
*inode
= path
->dentry
->d_inode
;
2052 int error
= get_write_access(inode
);
2056 * Refuse to truncate files with mandatory locks held on them.
2058 error
= locks_verify_locked(inode
);
2060 error
= security_path_truncate(path
);
2062 error
= do_truncate(path
->dentry
, 0,
2063 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2066 put_write_access(inode
);
2071 * Note that while the flag value (low two bits) for sys_open means:
2076 * it is changed into
2077 * 00 - no permissions needed
2078 * 01 - read-permission
2079 * 10 - write-permission
2081 * for the internal routines (ie open_namei()/follow_link() etc)
2082 * This is more logical, and also allows the 00 "no perm needed"
2083 * to be used for symlinks (where the permissions are checked
2087 static inline int open_to_namei_flags(int flag
)
2089 if ((flag
+1) & O_ACCMODE
)
2095 * Handle the last step of open()
2097 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2098 const struct open_flags
*op
, const char *pathname
)
2100 struct dentry
*dir
= nd
->path
.dentry
;
2101 struct dentry
*dentry
;
2102 int open_flag
= op
->open_flag
;
2103 int will_truncate
= open_flag
& O_TRUNC
;
2105 int acc_mode
= op
->acc_mode
;
2109 nd
->flags
&= ~LOOKUP_PARENT
;
2110 nd
->flags
|= op
->intent
;
2112 switch (nd
->last_type
) {
2115 error
= handle_dots(nd
, nd
->last_type
);
2117 return ERR_PTR(error
);
2120 if (nd
->flags
& LOOKUP_RCU
) {
2121 if (nameidata_drop_rcu_last(nd
))
2122 return ERR_PTR(-ECHILD
);
2124 error
= handle_reval_path(nd
);
2127 audit_inode(pathname
, nd
->path
.dentry
);
2128 if (open_flag
& O_CREAT
) {
2134 /* can't be RCU mode here */
2135 error
= handle_reval_path(nd
);
2138 audit_inode(pathname
, dir
);
2142 if (!(open_flag
& O_CREAT
)) {
2144 if (nd
->last
.name
[nd
->last
.len
])
2145 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2146 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2148 /* we _can_ be in RCU mode here */
2149 error
= walk_component(nd
, path
, &nd
->last
, LAST_NORM
,
2152 return ERR_PTR(error
);
2153 if (error
) /* symlink */
2156 if (nd
->flags
& LOOKUP_RCU
) {
2157 if (nameidata_drop_rcu_last(nd
))
2158 return ERR_PTR(-ECHILD
);
2162 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2163 if (!nd
->inode
->i_op
->lookup
)
2166 audit_inode(pathname
, nd
->path
.dentry
);
2170 /* create side of things */
2172 if (nd
->flags
& LOOKUP_RCU
) {
2173 if (nameidata_drop_rcu_last(nd
))
2174 return ERR_PTR(-ECHILD
);
2177 audit_inode(pathname
, dir
);
2179 /* trailing slashes? */
2180 if (nd
->last
.name
[nd
->last
.len
])
2183 mutex_lock(&dir
->d_inode
->i_mutex
);
2185 dentry
= lookup_hash(nd
);
2186 error
= PTR_ERR(dentry
);
2187 if (IS_ERR(dentry
)) {
2188 mutex_unlock(&dir
->d_inode
->i_mutex
);
2192 path
->dentry
= dentry
;
2193 path
->mnt
= nd
->path
.mnt
;
2195 /* Negative dentry, just create the file */
2196 if (!dentry
->d_inode
) {
2197 int mode
= op
->mode
;
2198 if (!IS_POSIXACL(dir
->d_inode
))
2199 mode
&= ~current_umask();
2201 * This write is needed to ensure that a
2202 * rw->ro transition does not occur between
2203 * the time when the file is created and when
2204 * a permanent write count is taken through
2205 * the 'struct file' in nameidata_to_filp().
2207 error
= mnt_want_write(nd
->path
.mnt
);
2209 goto exit_mutex_unlock
;
2211 /* Don't check for write permission, don't truncate */
2212 open_flag
&= ~O_TRUNC
;
2214 acc_mode
= MAY_OPEN
;
2215 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2217 goto exit_mutex_unlock
;
2218 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2220 goto exit_mutex_unlock
;
2221 mutex_unlock(&dir
->d_inode
->i_mutex
);
2222 dput(nd
->path
.dentry
);
2223 nd
->path
.dentry
= dentry
;
2228 * It already exists.
2230 mutex_unlock(&dir
->d_inode
->i_mutex
);
2231 audit_inode(pathname
, path
->dentry
);
2234 if (open_flag
& O_EXCL
)
2237 error
= follow_managed(path
, nd
->flags
);
2242 if (!path
->dentry
->d_inode
)
2245 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2248 path_to_nameidata(path
, nd
);
2249 nd
->inode
= path
->dentry
->d_inode
;
2251 if (S_ISDIR(nd
->inode
->i_mode
))
2254 if (!S_ISREG(nd
->inode
->i_mode
))
2257 if (will_truncate
) {
2258 error
= mnt_want_write(nd
->path
.mnt
);
2264 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2267 filp
= nameidata_to_filp(nd
);
2268 if (!IS_ERR(filp
)) {
2269 error
= ima_file_check(filp
, op
->acc_mode
);
2272 filp
= ERR_PTR(error
);
2275 if (!IS_ERR(filp
)) {
2276 if (will_truncate
) {
2277 error
= handle_truncate(filp
);
2280 filp
= ERR_PTR(error
);
2286 mnt_drop_write(nd
->path
.mnt
);
2287 path_put(&nd
->path
);
2291 mutex_unlock(&dir
->d_inode
->i_mutex
);
2293 path_put_conditional(path
, nd
);
2295 filp
= ERR_PTR(error
);
2299 static struct file
*path_openat(int dfd
, const char *pathname
,
2300 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2302 struct file
*base
= NULL
;
2307 filp
= get_empty_filp();
2309 return ERR_PTR(-ENFILE
);
2311 filp
->f_flags
= op
->open_flag
;
2312 nd
->intent
.open
.file
= filp
;
2313 nd
->intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2314 nd
->intent
.open
.create_mode
= op
->mode
;
2316 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2317 if (unlikely(error
))
2320 current
->total_link_count
= 0;
2321 error
= link_path_walk(pathname
, nd
);
2322 if (unlikely(error
))
2325 filp
= do_last(nd
, &path
, op
, pathname
);
2326 while (unlikely(!filp
)) { /* trailing symlink */
2327 struct path link
= path
;
2329 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2330 path_put_conditional(&path
, nd
);
2331 path_put(&nd
->path
);
2332 filp
= ERR_PTR(-ELOOP
);
2335 nd
->flags
|= LOOKUP_PARENT
;
2336 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2337 error
= follow_link(&link
, nd
, &cookie
);
2338 if (unlikely(error
))
2339 filp
= ERR_PTR(error
);
2341 filp
= do_last(nd
, &path
, op
, pathname
);
2342 put_link(nd
, &link
, cookie
);
2345 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2346 path_put(&nd
->root
);
2349 release_open_intent(nd
);
2353 filp
= ERR_PTR(error
);
2357 struct file
*do_filp_open(int dfd
, const char *pathname
,
2358 const struct open_flags
*op
, int flags
)
2360 struct nameidata nd
;
2363 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2364 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2365 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2366 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2367 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2371 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2372 const char *name
, const struct open_flags
*op
, int flags
)
2374 struct nameidata nd
;
2378 nd
.root
.dentry
= dentry
;
2380 flags
|= LOOKUP_ROOT
;
2382 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2383 return ERR_PTR(-ELOOP
);
2385 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2386 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2387 file
= path_openat(-1, name
, &nd
, op
, flags
);
2388 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2389 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2394 * lookup_create - lookup a dentry, creating it if it doesn't exist
2395 * @nd: nameidata info
2396 * @is_dir: directory flag
2398 * Simple function to lookup and return a dentry and create it
2399 * if it doesn't exist. Is SMP-safe.
2401 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2403 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2405 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2407 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2409 * Yucky last component or no last component at all?
2410 * (foo/., foo/.., /////)
2412 if (nd
->last_type
!= LAST_NORM
)
2414 nd
->flags
&= ~LOOKUP_PARENT
;
2415 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2416 nd
->intent
.open
.flags
= O_EXCL
;
2419 * Do the final lookup.
2421 dentry
= lookup_hash(nd
);
2425 if (dentry
->d_inode
)
2428 * Special case - lookup gave negative, but... we had foo/bar/
2429 * From the vfs_mknod() POV we just have a negative dentry -
2430 * all is fine. Let's be bastards - you had / on the end, you've
2431 * been asking for (non-existent) directory. -ENOENT for you.
2433 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2435 dentry
= ERR_PTR(-ENOENT
);
2440 dentry
= ERR_PTR(-EEXIST
);
2444 EXPORT_SYMBOL_GPL(lookup_create
);
2446 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2448 int error
= may_create(dir
, dentry
);
2453 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) &&
2454 !ns_capable(inode_userns(dir
), CAP_MKNOD
))
2457 if (!dir
->i_op
->mknod
)
2460 error
= devcgroup_inode_mknod(mode
, dev
);
2464 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2468 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2470 fsnotify_create(dir
, dentry
);
2474 static int may_mknod(mode_t mode
)
2476 switch (mode
& S_IFMT
) {
2482 case 0: /* zero mode translates to S_IFREG */
2491 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2496 struct dentry
*dentry
;
2497 struct nameidata nd
;
2502 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2506 dentry
= lookup_create(&nd
, 0);
2507 if (IS_ERR(dentry
)) {
2508 error
= PTR_ERR(dentry
);
2511 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2512 mode
&= ~current_umask();
2513 error
= may_mknod(mode
);
2516 error
= mnt_want_write(nd
.path
.mnt
);
2519 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2521 goto out_drop_write
;
2522 switch (mode
& S_IFMT
) {
2523 case 0: case S_IFREG
:
2524 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2526 case S_IFCHR
: case S_IFBLK
:
2527 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2528 new_decode_dev(dev
));
2530 case S_IFIFO
: case S_IFSOCK
:
2531 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2535 mnt_drop_write(nd
.path
.mnt
);
2539 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2546 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2548 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2551 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2553 int error
= may_create(dir
, dentry
);
2558 if (!dir
->i_op
->mkdir
)
2561 mode
&= (S_IRWXUGO
|S_ISVTX
);
2562 error
= security_inode_mkdir(dir
, dentry
, mode
);
2566 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2568 fsnotify_mkdir(dir
, dentry
);
2572 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2576 struct dentry
*dentry
;
2577 struct nameidata nd
;
2579 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2583 dentry
= lookup_create(&nd
, 1);
2584 error
= PTR_ERR(dentry
);
2588 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2589 mode
&= ~current_umask();
2590 error
= mnt_want_write(nd
.path
.mnt
);
2593 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2595 goto out_drop_write
;
2596 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2598 mnt_drop_write(nd
.path
.mnt
);
2602 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2609 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2611 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2615 * We try to drop the dentry early: we should have
2616 * a usage count of 2 if we're the only user of this
2617 * dentry, and if that is true (possibly after pruning
2618 * the dcache), then we drop the dentry now.
2620 * A low-level filesystem can, if it choses, legally
2623 * if (!d_unhashed(dentry))
2626 * if it cannot handle the case of removing a directory
2627 * that is still in use by something else..
2629 void dentry_unhash(struct dentry
*dentry
)
2632 shrink_dcache_parent(dentry
);
2633 spin_lock(&dentry
->d_lock
);
2634 if (dentry
->d_count
== 2)
2636 spin_unlock(&dentry
->d_lock
);
2639 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2641 int error
= may_delete(dir
, dentry
, 1);
2646 if (!dir
->i_op
->rmdir
)
2649 mutex_lock(&dentry
->d_inode
->i_mutex
);
2650 dentry_unhash(dentry
);
2651 if (d_mountpoint(dentry
))
2654 error
= security_inode_rmdir(dir
, dentry
);
2656 error
= dir
->i_op
->rmdir(dir
, dentry
);
2658 dentry
->d_inode
->i_flags
|= S_DEAD
;
2663 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2672 static long do_rmdir(int dfd
, const char __user
*pathname
)
2676 struct dentry
*dentry
;
2677 struct nameidata nd
;
2679 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2683 switch(nd
.last_type
) {
2695 nd
.flags
&= ~LOOKUP_PARENT
;
2697 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2698 dentry
= lookup_hash(&nd
);
2699 error
= PTR_ERR(dentry
);
2702 error
= mnt_want_write(nd
.path
.mnt
);
2705 error
= security_path_rmdir(&nd
.path
, dentry
);
2708 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2710 mnt_drop_write(nd
.path
.mnt
);
2714 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2721 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2723 return do_rmdir(AT_FDCWD
, pathname
);
2726 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2728 int error
= may_delete(dir
, dentry
, 0);
2733 if (!dir
->i_op
->unlink
)
2736 mutex_lock(&dentry
->d_inode
->i_mutex
);
2737 if (d_mountpoint(dentry
))
2740 error
= security_inode_unlink(dir
, dentry
);
2742 error
= dir
->i_op
->unlink(dir
, dentry
);
2747 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2749 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2750 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2751 fsnotify_link_count(dentry
->d_inode
);
2759 * Make sure that the actual truncation of the file will occur outside its
2760 * directory's i_mutex. Truncate can take a long time if there is a lot of
2761 * writeout happening, and we don't want to prevent access to the directory
2762 * while waiting on the I/O.
2764 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2768 struct dentry
*dentry
;
2769 struct nameidata nd
;
2770 struct inode
*inode
= NULL
;
2772 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2777 if (nd
.last_type
!= LAST_NORM
)
2780 nd
.flags
&= ~LOOKUP_PARENT
;
2782 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2783 dentry
= lookup_hash(&nd
);
2784 error
= PTR_ERR(dentry
);
2785 if (!IS_ERR(dentry
)) {
2786 /* Why not before? Because we want correct error value */
2787 if (nd
.last
.name
[nd
.last
.len
])
2789 inode
= dentry
->d_inode
;
2792 error
= mnt_want_write(nd
.path
.mnt
);
2795 error
= security_path_unlink(&nd
.path
, dentry
);
2798 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2800 mnt_drop_write(nd
.path
.mnt
);
2804 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2806 iput(inode
); /* truncate the inode here */
2813 error
= !dentry
->d_inode
? -ENOENT
:
2814 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2818 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2820 if ((flag
& ~AT_REMOVEDIR
) != 0)
2823 if (flag
& AT_REMOVEDIR
)
2824 return do_rmdir(dfd
, pathname
);
2826 return do_unlinkat(dfd
, pathname
);
2829 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2831 return do_unlinkat(AT_FDCWD
, pathname
);
2834 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2836 int error
= may_create(dir
, dentry
);
2841 if (!dir
->i_op
->symlink
)
2844 error
= security_inode_symlink(dir
, dentry
, oldname
);
2848 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2850 fsnotify_create(dir
, dentry
);
2854 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2855 int, newdfd
, const char __user
*, newname
)
2860 struct dentry
*dentry
;
2861 struct nameidata nd
;
2863 from
= getname(oldname
);
2865 return PTR_ERR(from
);
2867 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2871 dentry
= lookup_create(&nd
, 0);
2872 error
= PTR_ERR(dentry
);
2876 error
= mnt_want_write(nd
.path
.mnt
);
2879 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2881 goto out_drop_write
;
2882 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2884 mnt_drop_write(nd
.path
.mnt
);
2888 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2896 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2898 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2901 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2903 struct inode
*inode
= old_dentry
->d_inode
;
2909 error
= may_create(dir
, new_dentry
);
2913 if (dir
->i_sb
!= inode
->i_sb
)
2917 * A link to an append-only or immutable file cannot be created.
2919 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2921 if (!dir
->i_op
->link
)
2923 if (S_ISDIR(inode
->i_mode
))
2926 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2930 mutex_lock(&inode
->i_mutex
);
2931 /* Make sure we don't allow creating hardlink to an unlinked file */
2932 if (inode
->i_nlink
== 0)
2935 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2936 mutex_unlock(&inode
->i_mutex
);
2938 fsnotify_link(dir
, inode
, new_dentry
);
2943 * Hardlinks are often used in delicate situations. We avoid
2944 * security-related surprises by not following symlinks on the
2947 * We don't follow them on the oldname either to be compatible
2948 * with linux 2.0, and to avoid hard-linking to directories
2949 * and other special files. --ADM
2951 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2952 int, newdfd
, const char __user
*, newname
, int, flags
)
2954 struct dentry
*new_dentry
;
2955 struct nameidata nd
;
2956 struct path old_path
;
2961 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
2964 * To use null names we require CAP_DAC_READ_SEARCH
2965 * This ensures that not everyone will be able to create
2966 * handlink using the passed filedescriptor.
2968 if (flags
& AT_EMPTY_PATH
) {
2969 if (!capable(CAP_DAC_READ_SEARCH
))
2974 if (flags
& AT_SYMLINK_FOLLOW
)
2975 how
|= LOOKUP_FOLLOW
;
2977 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
2981 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2985 if (old_path
.mnt
!= nd
.path
.mnt
)
2987 new_dentry
= lookup_create(&nd
, 0);
2988 error
= PTR_ERR(new_dentry
);
2989 if (IS_ERR(new_dentry
))
2991 error
= mnt_want_write(nd
.path
.mnt
);
2994 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2996 goto out_drop_write
;
2997 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2999 mnt_drop_write(nd
.path
.mnt
);
3003 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3008 path_put(&old_path
);
3013 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3015 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3019 * The worst of all namespace operations - renaming directory. "Perverted"
3020 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3022 * a) we can get into loop creation. Check is done in is_subdir().
3023 * b) race potential - two innocent renames can create a loop together.
3024 * That's where 4.4 screws up. Current fix: serialization on
3025 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3027 * c) we have to lock _three_ objects - parents and victim (if it exists).
3028 * And that - after we got ->i_mutex on parents (until then we don't know
3029 * whether the target exists). Solution: try to be smart with locking
3030 * order for inodes. We rely on the fact that tree topology may change
3031 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3032 * move will be locked. Thus we can rank directories by the tree
3033 * (ancestors first) and rank all non-directories after them.
3034 * That works since everybody except rename does "lock parent, lookup,
3035 * lock child" and rename is under ->s_vfs_rename_mutex.
3036 * HOWEVER, it relies on the assumption that any object with ->lookup()
3037 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3038 * we'd better make sure that there's no link(2) for them.
3039 * d) some filesystems don't support opened-but-unlinked directories,
3040 * either because of layout or because they are not ready to deal with
3041 * all cases correctly. The latter will be fixed (taking this sort of
3042 * stuff into VFS), but the former is not going away. Solution: the same
3043 * trick as in rmdir().
3044 * e) conversion from fhandle to dentry may come in the wrong moment - when
3045 * we are removing the target. Solution: we will have to grab ->i_mutex
3046 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3047 * ->i_mutex on parents, which works but leads to some truly excessive
3050 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3051 struct inode
*new_dir
, struct dentry
*new_dentry
)
3054 struct inode
*target
;
3057 * If we are going to change the parent - check write permissions,
3058 * we'll need to flip '..'.
3060 if (new_dir
!= old_dir
) {
3061 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3066 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3070 target
= new_dentry
->d_inode
;
3072 mutex_lock(&target
->i_mutex
);
3073 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3077 dentry_unhash(new_dentry
);
3078 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3082 target
->i_flags
|= S_DEAD
;
3083 dont_mount(new_dentry
);
3085 mutex_unlock(&target
->i_mutex
);
3086 if (d_unhashed(new_dentry
))
3087 d_rehash(new_dentry
);
3091 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3092 d_move(old_dentry
,new_dentry
);
3096 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3097 struct inode
*new_dir
, struct dentry
*new_dentry
)
3099 struct inode
*target
;
3102 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3107 target
= new_dentry
->d_inode
;
3109 mutex_lock(&target
->i_mutex
);
3110 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3113 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3116 dont_mount(new_dentry
);
3117 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3118 d_move(old_dentry
, new_dentry
);
3121 mutex_unlock(&target
->i_mutex
);
3126 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3127 struct inode
*new_dir
, struct dentry
*new_dentry
)
3130 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3131 const unsigned char *old_name
;
3133 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3136 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3140 if (!new_dentry
->d_inode
)
3141 error
= may_create(new_dir
, new_dentry
);
3143 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3147 if (!old_dir
->i_op
->rename
)
3150 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3153 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3155 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3157 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3158 new_dentry
->d_inode
, old_dentry
);
3159 fsnotify_oldname_free(old_name
);
3164 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3165 int, newdfd
, const char __user
*, newname
)
3167 struct dentry
*old_dir
, *new_dir
;
3168 struct dentry
*old_dentry
, *new_dentry
;
3169 struct dentry
*trap
;
3170 struct nameidata oldnd
, newnd
;
3175 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3179 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3184 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3187 old_dir
= oldnd
.path
.dentry
;
3189 if (oldnd
.last_type
!= LAST_NORM
)
3192 new_dir
= newnd
.path
.dentry
;
3193 if (newnd
.last_type
!= LAST_NORM
)
3196 oldnd
.flags
&= ~LOOKUP_PARENT
;
3197 newnd
.flags
&= ~LOOKUP_PARENT
;
3198 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3200 trap
= lock_rename(new_dir
, old_dir
);
3202 old_dentry
= lookup_hash(&oldnd
);
3203 error
= PTR_ERR(old_dentry
);
3204 if (IS_ERR(old_dentry
))
3206 /* source must exist */
3208 if (!old_dentry
->d_inode
)
3210 /* unless the source is a directory trailing slashes give -ENOTDIR */
3211 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3213 if (oldnd
.last
.name
[oldnd
.last
.len
])
3215 if (newnd
.last
.name
[newnd
.last
.len
])
3218 /* source should not be ancestor of target */
3220 if (old_dentry
== trap
)
3222 new_dentry
= lookup_hash(&newnd
);
3223 error
= PTR_ERR(new_dentry
);
3224 if (IS_ERR(new_dentry
))
3226 /* target should not be an ancestor of source */
3228 if (new_dentry
== trap
)
3231 error
= mnt_want_write(oldnd
.path
.mnt
);
3234 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3235 &newnd
.path
, new_dentry
);
3238 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3239 new_dir
->d_inode
, new_dentry
);
3241 mnt_drop_write(oldnd
.path
.mnt
);
3247 unlock_rename(new_dir
, old_dir
);
3249 path_put(&newnd
.path
);
3252 path_put(&oldnd
.path
);
3258 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3260 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3263 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3267 len
= PTR_ERR(link
);
3272 if (len
> (unsigned) buflen
)
3274 if (copy_to_user(buffer
, link
, len
))
3281 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3282 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3283 * using) it for any given inode is up to filesystem.
3285 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3287 struct nameidata nd
;
3292 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3294 return PTR_ERR(cookie
);
3296 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3297 if (dentry
->d_inode
->i_op
->put_link
)
3298 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3302 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3304 return __vfs_follow_link(nd
, link
);
3307 /* get the link contents into pagecache */
3308 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3312 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3313 page
= read_mapping_page(mapping
, 0, NULL
);
3318 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3322 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3324 struct page
*page
= NULL
;
3325 char *s
= page_getlink(dentry
, &page
);
3326 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3329 page_cache_release(page
);
3334 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3336 struct page
*page
= NULL
;
3337 nd_set_link(nd
, page_getlink(dentry
, &page
));
3341 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3343 struct page
*page
= cookie
;
3347 page_cache_release(page
);
3352 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3354 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3356 struct address_space
*mapping
= inode
->i_mapping
;
3361 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3363 flags
|= AOP_FLAG_NOFS
;
3366 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3367 flags
, &page
, &fsdata
);
3371 kaddr
= kmap_atomic(page
, KM_USER0
);
3372 memcpy(kaddr
, symname
, len
-1);
3373 kunmap_atomic(kaddr
, KM_USER0
);
3375 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3382 mark_inode_dirty(inode
);
3388 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3390 return __page_symlink(inode
, symname
, len
,
3391 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3394 const struct inode_operations page_symlink_inode_operations
= {
3395 .readlink
= generic_readlink
,
3396 .follow_link
= page_follow_link_light
,
3397 .put_link
= page_put_link
,
3400 EXPORT_SYMBOL(user_path_at
);
3401 EXPORT_SYMBOL(follow_down_one
);
3402 EXPORT_SYMBOL(follow_down
);
3403 EXPORT_SYMBOL(follow_up
);
3404 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3405 EXPORT_SYMBOL(getname
);
3406 EXPORT_SYMBOL(lock_rename
);
3407 EXPORT_SYMBOL(lookup_one_len
);
3408 EXPORT_SYMBOL(page_follow_link_light
);
3409 EXPORT_SYMBOL(page_put_link
);
3410 EXPORT_SYMBOL(page_readlink
);
3411 EXPORT_SYMBOL(__page_symlink
);
3412 EXPORT_SYMBOL(page_symlink
);
3413 EXPORT_SYMBOL(page_symlink_inode_operations
);
3414 EXPORT_SYMBOL(kern_path_parent
);
3415 EXPORT_SYMBOL(kern_path
);
3416 EXPORT_SYMBOL(vfs_path_lookup
);
3417 EXPORT_SYMBOL(inode_permission
);
3418 EXPORT_SYMBOL(file_permission
);
3419 EXPORT_SYMBOL(unlock_rename
);
3420 EXPORT_SYMBOL(vfs_create
);
3421 EXPORT_SYMBOL(vfs_follow_link
);
3422 EXPORT_SYMBOL(vfs_link
);
3423 EXPORT_SYMBOL(vfs_mkdir
);
3424 EXPORT_SYMBOL(vfs_mknod
);
3425 EXPORT_SYMBOL(generic_permission
);
3426 EXPORT_SYMBOL(vfs_readlink
);
3427 EXPORT_SYMBOL(vfs_rename
);
3428 EXPORT_SYMBOL(vfs_rmdir
);
3429 EXPORT_SYMBOL(vfs_symlink
);
3430 EXPORT_SYMBOL(vfs_unlink
);
3431 EXPORT_SYMBOL(dentry_unhash
);
3432 EXPORT_SYMBOL(generic_readlink
);