4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user
*filename
, char *page
)
120 unsigned long len
= PATH_MAX
;
122 if (!segment_eq(get_fs(), KERNEL_DS
)) {
123 if ((unsigned long) filename
>= TASK_SIZE
)
125 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
126 len
= TASK_SIZE
- (unsigned long) filename
;
129 retval
= strncpy_from_user(page
, filename
, len
);
133 return -ENAMETOOLONG
;
139 char * getname(const char __user
* filename
)
143 result
= ERR_PTR(-ENOMEM
);
146 int retval
= do_getname(filename
, tmp
);
151 result
= ERR_PTR(retval
);
154 audit_getname(result
);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name
)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname
);
170 * This does basic POSIX ACL permission checking
172 static int acl_permission_check(struct inode
*inode
, int mask
, unsigned int flags
,
173 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
175 umode_t mode
= inode
->i_mode
;
177 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
179 if (current_fsuid() == inode
->i_uid
)
182 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
183 int error
= check_acl(inode
, mask
, flags
);
184 if (error
!= -EAGAIN
)
188 if (in_group_p(inode
->i_gid
))
193 * If the DACs are ok we don't need any capability check.
195 if ((mask
& ~mode
) == 0)
201 * generic_permission - check for access rights on a Posix-like filesystem
202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
205 * @flags: IPERM_FLAG_ flags.
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
210 * are used for other things.
212 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
213 * request cannot be satisfied (eg. requires blocking or too much complexity).
214 * It would then be called again in ref-walk mode.
216 int generic_permission(struct inode
*inode
, int mask
, unsigned int flags
,
217 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
222 * Do the basic POSIX ACL permission checks.
224 ret
= acl_permission_check(inode
, mask
, flags
, check_acl
);
229 * Read/write DACs are always overridable.
230 * Executable DACs are overridable if at least one exec bit is set.
232 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
233 if (capable(CAP_DAC_OVERRIDE
))
237 * Searching includes executable on directories, else just read.
239 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
240 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
241 if (capable(CAP_DAC_READ_SEARCH
))
248 * inode_permission - check for access rights to a given inode
249 * @inode: inode to check permission on
250 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
252 * Used to check for read/write/execute permissions on an inode.
253 * We use "fsuid" for this, letting us set arbitrary permissions
254 * for filesystem access without changing the "normal" uids which
255 * are used for other things.
257 int inode_permission(struct inode
*inode
, int mask
)
261 if (mask
& MAY_WRITE
) {
262 umode_t mode
= inode
->i_mode
;
265 * Nobody gets write access to a read-only fs.
267 if (IS_RDONLY(inode
) &&
268 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
272 * Nobody gets write access to an immutable file.
274 if (IS_IMMUTABLE(inode
))
278 if (inode
->i_op
->permission
)
279 retval
= inode
->i_op
->permission(inode
, mask
, 0);
281 retval
= generic_permission(inode
, mask
, 0,
282 inode
->i_op
->check_acl
);
287 retval
= devcgroup_inode_permission(inode
, mask
);
291 return security_inode_permission(inode
, mask
);
295 * file_permission - check for additional access rights to a given file
296 * @file: file to check access rights for
297 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
299 * Used to check for read/write/execute permissions on an already opened
303 * Do not use this function in new code. All access checks should
304 * be done using inode_permission().
306 int file_permission(struct file
*file
, int mask
)
308 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
312 * get_write_access() gets write permission for a file.
313 * put_write_access() releases this write permission.
314 * This is used for regular files.
315 * We cannot support write (and maybe mmap read-write shared) accesses and
316 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
317 * can have the following values:
318 * 0: no writers, no VM_DENYWRITE mappings
319 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
320 * > 0: (i_writecount) users are writing to the file.
322 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
323 * except for the cases where we don't hold i_writecount yet. Then we need to
324 * use {get,deny}_write_access() - these functions check the sign and refuse
325 * to do the change if sign is wrong. Exclusion between them is provided by
326 * the inode->i_lock spinlock.
329 int get_write_access(struct inode
* inode
)
331 spin_lock(&inode
->i_lock
);
332 if (atomic_read(&inode
->i_writecount
) < 0) {
333 spin_unlock(&inode
->i_lock
);
336 atomic_inc(&inode
->i_writecount
);
337 spin_unlock(&inode
->i_lock
);
342 int deny_write_access(struct file
* file
)
344 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
346 spin_lock(&inode
->i_lock
);
347 if (atomic_read(&inode
->i_writecount
) > 0) {
348 spin_unlock(&inode
->i_lock
);
351 atomic_dec(&inode
->i_writecount
);
352 spin_unlock(&inode
->i_lock
);
358 * path_get - get a reference to a path
359 * @path: path to get the reference to
361 * Given a path increment the reference count to the dentry and the vfsmount.
363 void path_get(struct path
*path
)
368 EXPORT_SYMBOL(path_get
);
371 * path_put - put a reference to a path
372 * @path: path to put the reference to
374 * Given a path decrement the reference count to the dentry and the vfsmount.
376 void path_put(struct path
*path
)
381 EXPORT_SYMBOL(path_put
);
384 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
385 * @nd: nameidata pathwalk data to drop
386 * Returns: 0 on success, -ECHILD on failure
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
390 * to drop out of rcu-walk mode and take normal reference counts on dentries
391 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
392 * refcounts at the last known good point before rcu-walk got stuck, so
393 * ref-walk may continue from there. If this is not successful (eg. a seqcount
394 * has changed), then failure is returned and path walk restarts from the
395 * beginning in ref-walk mode.
397 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
398 * ref-walk. Must be called from rcu-walk context.
400 static int nameidata_drop_rcu(struct nameidata
*nd
)
402 struct fs_struct
*fs
= current
->fs
;
403 struct dentry
*dentry
= nd
->path
.dentry
;
405 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
407 spin_lock(&fs
->lock
);
408 if (nd
->root
.mnt
!= fs
->root
.mnt
||
409 nd
->root
.dentry
!= fs
->root
.dentry
)
412 spin_lock(&dentry
->d_lock
);
413 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
415 BUG_ON(nd
->inode
!= dentry
->d_inode
);
416 spin_unlock(&dentry
->d_lock
);
419 spin_unlock(&fs
->lock
);
421 mntget(nd
->path
.mnt
);
424 br_read_unlock(vfsmount_lock
);
425 nd
->flags
&= ~LOOKUP_RCU
;
428 spin_unlock(&dentry
->d_lock
);
431 spin_unlock(&fs
->lock
);
435 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
436 static inline int nameidata_drop_rcu_maybe(struct nameidata
*nd
)
438 if (nd
->flags
& LOOKUP_RCU
)
439 return nameidata_drop_rcu(nd
);
444 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
445 * @nd: nameidata pathwalk data to drop
446 * @dentry: dentry to drop
447 * Returns: 0 on success, -ECHILD on failure
449 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
450 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
451 * @nd. Must be called from rcu-walk context.
453 static int nameidata_dentry_drop_rcu(struct nameidata
*nd
, struct dentry
*dentry
)
455 struct fs_struct
*fs
= current
->fs
;
456 struct dentry
*parent
= nd
->path
.dentry
;
458 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
460 spin_lock(&fs
->lock
);
461 if (nd
->root
.mnt
!= fs
->root
.mnt
||
462 nd
->root
.dentry
!= fs
->root
.dentry
)
465 spin_lock(&parent
->d_lock
);
466 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
467 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
470 * If the sequence check on the child dentry passed, then the child has
471 * not been removed from its parent. This means the parent dentry must
472 * be valid and able to take a reference at this point.
474 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
475 BUG_ON(!parent
->d_count
);
477 spin_unlock(&dentry
->d_lock
);
478 spin_unlock(&parent
->d_lock
);
481 spin_unlock(&fs
->lock
);
483 mntget(nd
->path
.mnt
);
486 br_read_unlock(vfsmount_lock
);
487 nd
->flags
&= ~LOOKUP_RCU
;
490 spin_unlock(&dentry
->d_lock
);
491 spin_unlock(&parent
->d_lock
);
494 spin_unlock(&fs
->lock
);
498 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
499 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata
*nd
, struct dentry
*dentry
)
501 if (nd
->flags
& LOOKUP_RCU
) {
502 if (unlikely(nameidata_dentry_drop_rcu(nd
, dentry
))) {
503 nd
->flags
&= ~LOOKUP_RCU
;
506 br_read_unlock(vfsmount_lock
);
514 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
515 * @nd: nameidata pathwalk data to drop
516 * Returns: 0 on success, -ECHILD on failure
518 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
519 * nd->path should be the final element of the lookup, so nd->root is discarded.
520 * Must be called from rcu-walk context.
522 static int nameidata_drop_rcu_last(struct nameidata
*nd
)
524 struct dentry
*dentry
= nd
->path
.dentry
;
526 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
527 nd
->flags
&= ~LOOKUP_RCU
;
529 spin_lock(&dentry
->d_lock
);
530 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
532 BUG_ON(nd
->inode
!= dentry
->d_inode
);
533 spin_unlock(&dentry
->d_lock
);
535 mntget(nd
->path
.mnt
);
538 br_read_unlock(vfsmount_lock
);
543 spin_unlock(&dentry
->d_lock
);
545 br_read_unlock(vfsmount_lock
);
550 * release_open_intent - free up open intent resources
551 * @nd: pointer to nameidata
553 void release_open_intent(struct nameidata
*nd
)
555 struct file
*file
= nd
->intent
.open
.file
;
557 if (file
&& !IS_ERR(file
)) {
558 if (file
->f_path
.dentry
== NULL
)
565 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
567 return dentry
->d_op
->d_revalidate(dentry
, nd
);
570 static struct dentry
*
571 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
573 int status
= d_revalidate(dentry
, nd
);
574 if (unlikely(status
<= 0)) {
576 * The dentry failed validation.
577 * If d_revalidate returned 0 attempt to invalidate
578 * the dentry otherwise d_revalidate is asking us
579 * to return a fail status.
583 dentry
= ERR_PTR(status
);
584 } else if (!d_invalidate(dentry
)) {
593 * handle_reval_path - force revalidation of a dentry
595 * In some situations the path walking code will trust dentries without
596 * revalidating them. This causes problems for filesystems that depend on
597 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
598 * (which indicates that it's possible for the dentry to go stale), force
599 * a d_revalidate call before proceeding.
601 * Returns 0 if the revalidation was successful. If the revalidation fails,
602 * either return the error returned by d_revalidate or -ESTALE if the
603 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
604 * invalidate the dentry. It's up to the caller to handle putting references
605 * to the path if necessary.
607 static inline int handle_reval_path(struct nameidata
*nd
)
609 struct dentry
*dentry
= nd
->path
.dentry
;
612 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
615 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
618 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
621 /* Note: we do not d_invalidate() */
622 status
= d_revalidate(dentry
, nd
);
633 * Short-cut version of permission(), for calling on directories
634 * during pathname resolution. Combines parts of permission()
635 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
637 * If appropriate, check DAC only. If not appropriate, or
638 * short-cut DAC fails, then call ->permission() to do more
639 * complete permission check.
641 static inline int exec_permission(struct inode
*inode
, unsigned int flags
)
645 if (inode
->i_op
->permission
) {
646 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
, flags
);
648 ret
= acl_permission_check(inode
, MAY_EXEC
, flags
,
649 inode
->i_op
->check_acl
);
656 if (capable(CAP_DAC_OVERRIDE
) || capable(CAP_DAC_READ_SEARCH
))
661 return security_inode_exec_permission(inode
, flags
);
664 static __always_inline
void set_root(struct nameidata
*nd
)
667 get_fs_root(current
->fs
, &nd
->root
);
670 static int link_path_walk(const char *, struct nameidata
*);
672 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
675 struct fs_struct
*fs
= current
->fs
;
679 seq
= read_seqcount_begin(&fs
->seq
);
681 } while (read_seqcount_retry(&fs
->seq
, seq
));
685 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
697 nd
->flags
|= LOOKUP_JUMPED
;
699 nd
->inode
= nd
->path
.dentry
->d_inode
;
701 ret
= link_path_walk(link
, nd
);
705 return PTR_ERR(link
);
708 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
711 if (path
->mnt
!= nd
->path
.mnt
)
715 static inline void path_to_nameidata(const struct path
*path
,
716 struct nameidata
*nd
)
718 if (!(nd
->flags
& LOOKUP_RCU
)) {
719 dput(nd
->path
.dentry
);
720 if (nd
->path
.mnt
!= path
->mnt
)
721 mntput(nd
->path
.mnt
);
723 nd
->path
.mnt
= path
->mnt
;
724 nd
->path
.dentry
= path
->dentry
;
727 static __always_inline
int
728 __do_follow_link(const struct path
*link
, struct nameidata
*nd
, void **p
)
731 struct dentry
*dentry
= link
->dentry
;
733 BUG_ON(nd
->flags
& LOOKUP_RCU
);
735 touch_atime(link
->mnt
, dentry
);
736 nd_set_link(nd
, NULL
);
738 if (link
->mnt
== nd
->path
.mnt
)
741 error
= security_inode_follow_link(link
->dentry
, nd
);
743 *p
= ERR_PTR(error
); /* no ->put_link(), please */
748 nd
->last_type
= LAST_BIND
;
749 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
752 char *s
= nd_get_link(nd
);
755 error
= __vfs_follow_link(nd
, s
);
756 else if (nd
->last_type
== LAST_BIND
)
757 nd
->flags
|= LOOKUP_JUMPED
;
763 * This limits recursive symlink follows to 8, while
764 * limiting consecutive symlinks to 40.
766 * Without that kind of total limit, nasty chains of consecutive
767 * symlinks can cause almost arbitrarily long lookups.
769 static inline int do_follow_link(struct inode
*inode
, struct path
*path
, struct nameidata
*nd
)
774 /* We drop rcu-walk here */
775 if (nameidata_dentry_drop_rcu_maybe(nd
, path
->dentry
))
777 BUG_ON(inode
!= path
->dentry
->d_inode
);
779 if (current
->link_count
>= MAX_NESTED_LINKS
)
781 if (current
->total_link_count
>= 40)
783 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
785 current
->link_count
++;
786 current
->total_link_count
++;
788 err
= __do_follow_link(path
, nd
, &cookie
);
789 if (!IS_ERR(cookie
) && path
->dentry
->d_inode
->i_op
->put_link
)
790 path
->dentry
->d_inode
->i_op
->put_link(path
->dentry
, nd
, cookie
);
792 current
->link_count
--;
796 path_put_conditional(path
, nd
);
801 static int follow_up_rcu(struct path
*path
)
803 struct vfsmount
*parent
;
804 struct dentry
*mountpoint
;
806 parent
= path
->mnt
->mnt_parent
;
807 if (parent
== path
->mnt
)
809 mountpoint
= path
->mnt
->mnt_mountpoint
;
810 path
->dentry
= mountpoint
;
815 int follow_up(struct path
*path
)
817 struct vfsmount
*parent
;
818 struct dentry
*mountpoint
;
820 br_read_lock(vfsmount_lock
);
821 parent
= path
->mnt
->mnt_parent
;
822 if (parent
== path
->mnt
) {
823 br_read_unlock(vfsmount_lock
);
827 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
828 br_read_unlock(vfsmount_lock
);
830 path
->dentry
= mountpoint
;
837 * Perform an automount
838 * - return -EISDIR to tell follow_managed() to stop and return the path we
841 static int follow_automount(struct path
*path
, unsigned flags
,
844 struct vfsmount
*mnt
;
847 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
850 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
851 * and this is the terminal part of the path.
853 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
854 return -EISDIR
; /* we actually want to stop here */
856 /* We want to mount if someone is trying to open/create a file of any
857 * type under the mountpoint, wants to traverse through the mountpoint
858 * or wants to open the mounted directory.
860 * We don't want to mount if someone's just doing a stat and they've
861 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
862 * appended a '/' to the name.
864 if (!(flags
& LOOKUP_FOLLOW
) &&
865 !(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
866 LOOKUP_OPEN
| LOOKUP_CREATE
)))
869 current
->total_link_count
++;
870 if (current
->total_link_count
>= 40)
873 mnt
= path
->dentry
->d_op
->d_automount(path
);
876 * The filesystem is allowed to return -EISDIR here to indicate
877 * it doesn't want to automount. For instance, autofs would do
878 * this so that its userspace daemon can mount on this dentry.
880 * However, we can only permit this if it's a terminal point in
881 * the path being looked up; if it wasn't then the remainder of
882 * the path is inaccessible and we should say so.
884 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_CONTINUE
))
889 if (!mnt
) /* mount collision */
892 err
= finish_automount(mnt
, path
);
896 /* Someone else made a mount here whilst we were busy */
903 path
->dentry
= dget(mnt
->mnt_root
);
913 * Handle a dentry that is managed in some way.
914 * - Flagged for transit management (autofs)
915 * - Flagged as mountpoint
916 * - Flagged as automount point
918 * This may only be called in refwalk mode.
920 * Serialization is taken care of in namespace.c
922 static int follow_managed(struct path
*path
, unsigned flags
)
925 bool need_mntput
= false;
928 /* Given that we're not holding a lock here, we retain the value in a
929 * local variable for each dentry as we look at it so that we don't see
930 * the components of that value change under us */
931 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
932 managed
&= DCACHE_MANAGED_DENTRY
,
933 unlikely(managed
!= 0)) {
934 /* Allow the filesystem to manage the transit without i_mutex
936 if (managed
& DCACHE_MANAGE_TRANSIT
) {
937 BUG_ON(!path
->dentry
->d_op
);
938 BUG_ON(!path
->dentry
->d_op
->d_manage
);
939 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
,
942 return ret
== -EISDIR
? 0 : ret
;
945 /* Transit to a mounted filesystem. */
946 if (managed
& DCACHE_MOUNTED
) {
947 struct vfsmount
*mounted
= lookup_mnt(path
);
953 path
->dentry
= dget(mounted
->mnt_root
);
958 /* Something is mounted on this dentry in another
959 * namespace and/or whatever was mounted there in this
960 * namespace got unmounted before we managed to get the
964 /* Handle an automount point */
965 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
966 ret
= follow_automount(path
, flags
, &need_mntput
);
968 return ret
== -EISDIR
? 0 : ret
;
972 /* We didn't change the current path point */
978 int follow_down_one(struct path
*path
)
980 struct vfsmount
*mounted
;
982 mounted
= lookup_mnt(path
);
987 path
->dentry
= dget(mounted
->mnt_root
);
994 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
995 * meet a managed dentry and we're not walking to "..". True is returned to
996 * continue, false to abort.
998 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
999 struct inode
**inode
, bool reverse_transit
)
1001 while (d_mountpoint(path
->dentry
)) {
1002 struct vfsmount
*mounted
;
1003 if (unlikely(path
->dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) &&
1005 path
->dentry
->d_op
->d_manage(path
->dentry
, false, true) < 0)
1007 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1010 path
->mnt
= mounted
;
1011 path
->dentry
= mounted
->mnt_root
;
1012 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1013 *inode
= path
->dentry
->d_inode
;
1016 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1017 return reverse_transit
;
1021 static int follow_dotdot_rcu(struct nameidata
*nd
)
1023 struct inode
*inode
= nd
->inode
;
1028 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1029 nd
->path
.mnt
== nd
->root
.mnt
) {
1032 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1033 struct dentry
*old
= nd
->path
.dentry
;
1034 struct dentry
*parent
= old
->d_parent
;
1037 seq
= read_seqcount_begin(&parent
->d_seq
);
1038 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1040 inode
= parent
->d_inode
;
1041 nd
->path
.dentry
= parent
;
1045 if (!follow_up_rcu(&nd
->path
))
1047 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1048 inode
= nd
->path
.dentry
->d_inode
;
1050 __follow_mount_rcu(nd
, &nd
->path
, &inode
, true);
1055 nd
->flags
&= ~LOOKUP_RCU
;
1056 nd
->root
.mnt
= NULL
;
1058 br_read_unlock(vfsmount_lock
);
1063 * Follow down to the covering mount currently visible to userspace. At each
1064 * point, the filesystem owning that dentry may be queried as to whether the
1065 * caller is permitted to proceed or not.
1067 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1070 int follow_down(struct path
*path
, bool mounting_here
)
1075 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1076 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1077 /* Allow the filesystem to manage the transit without i_mutex
1080 * We indicate to the filesystem if someone is trying to mount
1081 * something here. This gives autofs the chance to deny anyone
1082 * other than its daemon the right to mount on its
1085 * The filesystem may sleep at this point.
1087 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1088 BUG_ON(!path
->dentry
->d_op
);
1089 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1090 ret
= path
->dentry
->d_op
->d_manage(
1091 path
->dentry
, mounting_here
, false);
1093 return ret
== -EISDIR
? 0 : ret
;
1096 /* Transit to a mounted filesystem. */
1097 if (managed
& DCACHE_MOUNTED
) {
1098 struct vfsmount
*mounted
= lookup_mnt(path
);
1103 path
->mnt
= mounted
;
1104 path
->dentry
= dget(mounted
->mnt_root
);
1108 /* Don't handle automount points here */
1115 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1117 static void follow_mount(struct path
*path
)
1119 while (d_mountpoint(path
->dentry
)) {
1120 struct vfsmount
*mounted
= lookup_mnt(path
);
1125 path
->mnt
= mounted
;
1126 path
->dentry
= dget(mounted
->mnt_root
);
1130 static void follow_dotdot(struct nameidata
*nd
)
1135 struct dentry
*old
= nd
->path
.dentry
;
1137 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1138 nd
->path
.mnt
== nd
->root
.mnt
) {
1141 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1142 /* rare case of legitimate dget_parent()... */
1143 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1147 if (!follow_up(&nd
->path
))
1150 follow_mount(&nd
->path
);
1151 nd
->inode
= nd
->path
.dentry
->d_inode
;
1155 * Allocate a dentry with name and parent, and perform a parent
1156 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1157 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1158 * have verified that no child exists while under i_mutex.
1160 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1161 struct qstr
*name
, struct nameidata
*nd
)
1163 struct inode
*inode
= parent
->d_inode
;
1164 struct dentry
*dentry
;
1167 /* Don't create child dentry for a dead directory. */
1168 if (unlikely(IS_DEADDIR(inode
)))
1169 return ERR_PTR(-ENOENT
);
1171 dentry
= d_alloc(parent
, name
);
1172 if (unlikely(!dentry
))
1173 return ERR_PTR(-ENOMEM
);
1175 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1176 if (unlikely(old
)) {
1184 * It's more convoluted than I'd like it to be, but... it's still fairly
1185 * small and for now I'd prefer to have fast path as straight as possible.
1186 * It _is_ time-critical.
1188 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1189 struct path
*path
, struct inode
**inode
)
1191 struct vfsmount
*mnt
= nd
->path
.mnt
;
1192 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1198 * Rename seqlock is not required here because in the off chance
1199 * of a false negative due to a concurrent rename, we're going to
1200 * do the non-racy lookup, below.
1202 if (nd
->flags
& LOOKUP_RCU
) {
1205 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1209 /* Memory barrier in read_seqcount_begin of child is enough */
1210 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1214 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1215 status
= d_revalidate(dentry
, nd
);
1216 if (unlikely(status
<= 0)) {
1217 if (status
!= -ECHILD
)
1223 path
->dentry
= dentry
;
1224 if (likely(__follow_mount_rcu(nd
, path
, inode
, false)))
1228 if (nameidata_dentry_drop_rcu(nd
, dentry
))
1231 if (nameidata_drop_rcu(nd
))
1235 dentry
= __d_lookup(parent
, name
);
1239 if (unlikely(!dentry
)) {
1240 struct inode
*dir
= parent
->d_inode
;
1241 BUG_ON(nd
->inode
!= dir
);
1243 mutex_lock(&dir
->i_mutex
);
1244 dentry
= d_lookup(parent
, name
);
1245 if (likely(!dentry
)) {
1246 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1247 if (IS_ERR(dentry
)) {
1248 mutex_unlock(&dir
->i_mutex
);
1249 return PTR_ERR(dentry
);
1255 mutex_unlock(&dir
->i_mutex
);
1257 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1258 status
= d_revalidate(dentry
, nd
);
1259 if (unlikely(status
<= 0)) {
1264 if (!d_invalidate(dentry
)) {
1273 path
->dentry
= dentry
;
1274 err
= follow_managed(path
, nd
->flags
);
1275 if (unlikely(err
< 0)) {
1276 path_put_conditional(path
, nd
);
1279 *inode
= path
->dentry
->d_inode
;
1283 static inline int may_lookup(struct nameidata
*nd
)
1285 if (nd
->flags
& LOOKUP_RCU
) {
1286 int err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1289 if (nameidata_drop_rcu(nd
))
1292 return exec_permission(nd
->inode
, 0);
1295 static inline int handle_dots(struct nameidata
*nd
, int type
)
1297 if (type
== LAST_DOTDOT
) {
1298 if (nd
->flags
& LOOKUP_RCU
) {
1299 if (follow_dotdot_rcu(nd
))
1307 static void terminate_walk(struct nameidata
*nd
)
1309 if (!(nd
->flags
& LOOKUP_RCU
)) {
1310 path_put(&nd
->path
);
1312 nd
->flags
&= ~LOOKUP_RCU
;
1313 nd
->root
.mnt
= NULL
;
1315 br_read_unlock(vfsmount_lock
);
1321 * This is the basic name resolution function, turning a pathname into
1322 * the final dentry. We expect 'base' to be positive and a directory.
1324 * Returns 0 and nd will have valid dentry and mnt on success.
1325 * Returns error and drops reference to input namei data on failure.
1327 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1331 unsigned int lookup_flags
= nd
->flags
;
1339 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
1341 /* At this point we know we have a real path component. */
1343 struct inode
*inode
;
1349 nd
->flags
|= LOOKUP_CONTINUE
;
1351 err
= may_lookup(nd
);
1356 c
= *(const unsigned char *)name
;
1358 hash
= init_name_hash();
1361 hash
= partial_name_hash(c
, hash
);
1362 c
= *(const unsigned char *)name
;
1363 } while (c
&& (c
!= '/'));
1364 this.len
= name
- (const char *) this.name
;
1365 this.hash
= end_name_hash(hash
);
1368 if (this.name
[0] == '.') switch (this.len
) {
1370 if (this.name
[1] == '.') {
1372 nd
->flags
|= LOOKUP_JUMPED
;
1378 if (likely(type
== LAST_NORM
)) {
1379 struct dentry
*parent
= nd
->path
.dentry
;
1380 nd
->flags
&= ~LOOKUP_JUMPED
;
1381 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1382 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1389 /* remove trailing slashes? */
1391 goto last_component
;
1392 while (*++name
== '/');
1394 goto last_with_slashes
;
1397 * "." and ".." are special - ".." especially so because it has
1398 * to be able to know about the current root directory and
1399 * parent relationships.
1401 if (unlikely(type
!= LAST_NORM
)) {
1402 if (handle_dots(nd
, type
))
1407 /* This does the actual lookups.. */
1408 err
= do_lookup(nd
, &this, &next
, &inode
);
1412 if (inode
&& inode
->i_op
->follow_link
) {
1413 err
= do_follow_link(inode
, &next
, nd
);
1416 nd
->inode
= nd
->path
.dentry
->d_inode
;
1418 path_to_nameidata(&next
, nd
);
1425 if (!nd
->inode
->i_op
->lookup
)
1428 /* here ends the main loop */
1431 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1433 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1434 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1435 if (lookup_flags
& LOOKUP_PARENT
)
1437 if (unlikely(type
!= LAST_NORM
))
1438 return handle_dots(nd
, type
);
1439 err
= do_lookup(nd
, &this, &next
, &inode
);
1442 if (inode
&& unlikely(inode
->i_op
->follow_link
) &&
1443 (lookup_flags
& LOOKUP_FOLLOW
)) {
1444 err
= do_follow_link(inode
, &next
, nd
);
1447 nd
->inode
= nd
->path
.dentry
->d_inode
;
1449 path_to_nameidata(&next
, nd
);
1455 if (lookup_flags
& LOOKUP_DIRECTORY
) {
1457 if (!nd
->inode
->i_op
->lookup
)
1463 nd
->last_type
= type
;
1470 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1471 struct nameidata
*nd
, struct file
**fp
)
1477 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1478 nd
->flags
= flags
| LOOKUP_JUMPED
;
1480 nd
->root
.mnt
= NULL
;
1483 if (flags
& LOOKUP_RCU
) {
1484 br_read_lock(vfsmount_lock
);
1489 path_get(&nd
->root
);
1491 nd
->path
= nd
->root
;
1492 } else if (dfd
== AT_FDCWD
) {
1493 if (flags
& LOOKUP_RCU
) {
1494 struct fs_struct
*fs
= current
->fs
;
1497 br_read_lock(vfsmount_lock
);
1501 seq
= read_seqcount_begin(&fs
->seq
);
1503 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1504 } while (read_seqcount_retry(&fs
->seq
, seq
));
1506 get_fs_pwd(current
->fs
, &nd
->path
);
1509 struct dentry
*dentry
;
1511 file
= fget_light(dfd
, &fput_needed
);
1516 dentry
= file
->f_path
.dentry
;
1519 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1522 retval
= file_permission(file
, MAY_EXEC
);
1526 nd
->path
= file
->f_path
;
1527 if (flags
& LOOKUP_RCU
) {
1530 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1531 br_read_lock(vfsmount_lock
);
1534 path_get(&file
->f_path
);
1535 fput_light(file
, fput_needed
);
1539 nd
->inode
= nd
->path
.dentry
->d_inode
;
1543 fput_light(file
, fput_needed
);
1548 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1549 static int path_lookupat(int dfd
, const char *name
,
1550 unsigned int flags
, struct nameidata
*nd
)
1552 struct file
*base
= NULL
;
1556 * Path walking is largely split up into 2 different synchronisation
1557 * schemes, rcu-walk and ref-walk (explained in
1558 * Documentation/filesystems/path-lookup.txt). These share much of the
1559 * path walk code, but some things particularly setup, cleanup, and
1560 * following mounts are sufficiently divergent that functions are
1561 * duplicated. Typically there is a function foo(), and its RCU
1562 * analogue, foo_rcu().
1564 * -ECHILD is the error number of choice (just to avoid clashes) that
1565 * is returned if some aspect of an rcu-walk fails. Such an error must
1566 * be handled by restarting a traditional ref-walk (which will always
1567 * be able to complete).
1569 retval
= path_init(dfd
, name
, flags
, nd
, &base
);
1571 if (unlikely(retval
))
1574 current
->total_link_count
= 0;
1575 retval
= link_path_walk(name
, nd
);
1577 if (nd
->flags
& LOOKUP_RCU
) {
1578 /* went all way through without dropping RCU */
1580 if (nameidata_drop_rcu_last(nd
))
1585 retval
= handle_reval_path(nd
);
1591 path_put(&nd
->root
);
1592 nd
->root
.mnt
= NULL
;
1597 static int do_path_lookup(int dfd
, const char *name
,
1598 unsigned int flags
, struct nameidata
*nd
)
1600 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1601 if (unlikely(retval
== -ECHILD
))
1602 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1603 if (unlikely(retval
== -ESTALE
))
1604 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1606 if (likely(!retval
)) {
1607 if (unlikely(!audit_dummy_context())) {
1608 if (nd
->path
.dentry
&& nd
->inode
)
1609 audit_inode(name
, nd
->path
.dentry
);
1615 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1617 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1620 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1622 struct nameidata nd
;
1623 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1630 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1631 * @dentry: pointer to dentry of the base directory
1632 * @mnt: pointer to vfs mount of the base directory
1633 * @name: pointer to file name
1634 * @flags: lookup flags
1635 * @nd: pointer to nameidata
1637 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1638 const char *name
, unsigned int flags
,
1639 struct nameidata
*nd
)
1643 /* same as do_path_lookup */
1644 nd
->last_type
= LAST_ROOT
;
1645 nd
->flags
= flags
| LOOKUP_JUMPED
;
1648 nd
->path
.dentry
= dentry
;
1650 path_get(&nd
->path
);
1651 nd
->root
= nd
->path
;
1652 path_get(&nd
->root
);
1653 nd
->inode
= nd
->path
.dentry
->d_inode
;
1655 current
->total_link_count
= 0;
1657 result
= link_path_walk(name
, nd
);
1659 result
= handle_reval_path(nd
);
1660 if (result
== -ESTALE
) {
1661 /* nd->path had been dropped */
1662 current
->total_link_count
= 0;
1663 nd
->path
.dentry
= dentry
;
1665 nd
->inode
= dentry
->d_inode
;
1666 path_get(&nd
->path
);
1667 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_REVAL
;
1669 result
= link_path_walk(name
, nd
);
1671 result
= handle_reval_path(nd
);
1673 if (unlikely(!result
&& !audit_dummy_context() && nd
->path
.dentry
&&
1675 audit_inode(name
, nd
->path
.dentry
);
1677 path_put(&nd
->root
);
1678 nd
->root
.mnt
= NULL
;
1683 static struct dentry
*__lookup_hash(struct qstr
*name
,
1684 struct dentry
*base
, struct nameidata
*nd
)
1686 struct inode
*inode
= base
->d_inode
;
1687 struct dentry
*dentry
;
1690 err
= exec_permission(inode
, 0);
1692 return ERR_PTR(err
);
1695 * Don't bother with __d_lookup: callers are for creat as
1696 * well as unlink, so a lot of the time it would cost
1699 dentry
= d_lookup(base
, name
);
1701 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1702 dentry
= do_revalidate(dentry
, nd
);
1705 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1711 * Restricted form of lookup. Doesn't follow links, single-component only,
1712 * needs parent already locked. Doesn't follow mounts.
1715 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1717 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1721 * lookup_one_len - filesystem helper to lookup single pathname component
1722 * @name: pathname component to lookup
1723 * @base: base directory to lookup from
1724 * @len: maximum length @len should be interpreted to
1726 * Note that this routine is purely a helper for filesystem usage and should
1727 * not be called by generic code. Also note that by using this function the
1728 * nameidata argument is passed to the filesystem methods and a filesystem
1729 * using this helper needs to be prepared for that.
1731 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1737 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1742 return ERR_PTR(-EACCES
);
1744 hash
= init_name_hash();
1746 c
= *(const unsigned char *)name
++;
1747 if (c
== '/' || c
== '\0')
1748 return ERR_PTR(-EACCES
);
1749 hash
= partial_name_hash(c
, hash
);
1751 this.hash
= end_name_hash(hash
);
1753 * See if the low-level filesystem might want
1754 * to use its own hash..
1756 if (base
->d_flags
& DCACHE_OP_HASH
) {
1757 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1759 return ERR_PTR(err
);
1762 return __lookup_hash(&this, base
, NULL
);
1765 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1768 struct nameidata nd
;
1769 char *tmp
= getname(name
);
1770 int err
= PTR_ERR(tmp
);
1773 BUG_ON(flags
& LOOKUP_PARENT
);
1775 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1783 static int user_path_parent(int dfd
, const char __user
*path
,
1784 struct nameidata
*nd
, char **name
)
1786 char *s
= getname(path
);
1792 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1802 * It's inline, so penalty for filesystems that don't use sticky bit is
1805 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1807 uid_t fsuid
= current_fsuid();
1809 if (!(dir
->i_mode
& S_ISVTX
))
1811 if (inode
->i_uid
== fsuid
)
1813 if (dir
->i_uid
== fsuid
)
1815 return !capable(CAP_FOWNER
);
1819 * Check whether we can remove a link victim from directory dir, check
1820 * whether the type of victim is right.
1821 * 1. We can't do it if dir is read-only (done in permission())
1822 * 2. We should have write and exec permissions on dir
1823 * 3. We can't remove anything from append-only dir
1824 * 4. We can't do anything with immutable dir (done in permission())
1825 * 5. If the sticky bit on dir is set we should either
1826 * a. be owner of dir, or
1827 * b. be owner of victim, or
1828 * c. have CAP_FOWNER capability
1829 * 6. If the victim is append-only or immutable we can't do antyhing with
1830 * links pointing to it.
1831 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1832 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1833 * 9. We can't remove a root or mountpoint.
1834 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1835 * nfs_async_unlink().
1837 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1841 if (!victim
->d_inode
)
1844 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1845 audit_inode_child(victim
, dir
);
1847 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1852 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1853 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1856 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1858 if (IS_ROOT(victim
))
1860 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1862 if (IS_DEADDIR(dir
))
1864 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1869 /* Check whether we can create an object with dentry child in directory
1871 * 1. We can't do it if child already exists (open has special treatment for
1872 * this case, but since we are inlined it's OK)
1873 * 2. We can't do it if dir is read-only (done in permission())
1874 * 3. We should have write and exec permissions on dir
1875 * 4. We can't do it if dir is immutable (done in permission())
1877 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1881 if (IS_DEADDIR(dir
))
1883 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1887 * p1 and p2 should be directories on the same fs.
1889 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1894 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1898 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1900 p
= d_ancestor(p2
, p1
);
1902 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1903 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1907 p
= d_ancestor(p1
, p2
);
1909 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1910 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1914 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1915 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1919 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1921 mutex_unlock(&p1
->d_inode
->i_mutex
);
1923 mutex_unlock(&p2
->d_inode
->i_mutex
);
1924 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1928 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1929 struct nameidata
*nd
)
1931 int error
= may_create(dir
, dentry
);
1936 if (!dir
->i_op
->create
)
1937 return -EACCES
; /* shouldn't it be ENOSYS? */
1940 error
= security_inode_create(dir
, dentry
, mode
);
1943 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1945 fsnotify_create(dir
, dentry
);
1949 int may_open(struct path
*path
, int acc_mode
, int flag
)
1951 struct dentry
*dentry
= path
->dentry
;
1952 struct inode
*inode
= dentry
->d_inode
;
1958 switch (inode
->i_mode
& S_IFMT
) {
1962 if (acc_mode
& MAY_WRITE
)
1967 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
1976 error
= inode_permission(inode
, acc_mode
);
1981 * An append-only file must be opened in append mode for writing.
1983 if (IS_APPEND(inode
)) {
1984 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
1990 /* O_NOATIME can only be set by the owner or superuser */
1991 if (flag
& O_NOATIME
&& !is_owner_or_cap(inode
))
1995 * Ensure there are no outstanding leases on the file.
1997 return break_lease(inode
, flag
);
2000 static int handle_truncate(struct file
*filp
)
2002 struct path
*path
= &filp
->f_path
;
2003 struct inode
*inode
= path
->dentry
->d_inode
;
2004 int error
= get_write_access(inode
);
2008 * Refuse to truncate files with mandatory locks held on them.
2010 error
= locks_verify_locked(inode
);
2012 error
= security_path_truncate(path
);
2014 error
= do_truncate(path
->dentry
, 0,
2015 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2018 put_write_access(inode
);
2023 * Note that while the flag value (low two bits) for sys_open means:
2028 * it is changed into
2029 * 00 - no permissions needed
2030 * 01 - read-permission
2031 * 10 - write-permission
2033 * for the internal routines (ie open_namei()/follow_link() etc)
2034 * This is more logical, and also allows the 00 "no perm needed"
2035 * to be used for symlinks (where the permissions are checked
2039 static inline int open_to_namei_flags(int flag
)
2041 if ((flag
+1) & O_ACCMODE
)
2047 * Handle the last step of open()
2049 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2050 const struct open_flags
*op
, const char *pathname
)
2052 struct dentry
*dir
= nd
->path
.dentry
;
2053 struct dentry
*dentry
;
2054 int open_flag
= op
->open_flag
;
2055 int will_truncate
= open_flag
& O_TRUNC
;
2059 struct inode
*inode
;
2062 nd
->flags
&= ~LOOKUP_PARENT
;
2063 nd
->flags
|= op
->intent
;
2065 switch (nd
->last_type
) {
2068 error
= handle_dots(nd
, nd
->last_type
);
2070 return ERR_PTR(error
);
2073 if (nd
->flags
& LOOKUP_RCU
) {
2074 if (nameidata_drop_rcu_last(nd
))
2075 return ERR_PTR(-ECHILD
);
2077 error
= handle_reval_path(nd
);
2080 audit_inode(pathname
, nd
->path
.dentry
);
2081 if (open_flag
& O_CREAT
) {
2087 /* can't be RCU mode here */
2088 error
= handle_reval_path(nd
);
2091 audit_inode(pathname
, dir
);
2095 if (!(open_flag
& O_CREAT
)) {
2096 if (nd
->last
.name
[nd
->last
.len
])
2097 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2098 /* we _can_ be in RCU mode here */
2099 error
= do_lookup(nd
, &nd
->last
, path
, &inode
);
2102 return ERR_PTR(error
);
2105 path_to_nameidata(path
, nd
);
2107 return ERR_PTR(-ENOENT
);
2109 if (unlikely(inode
->i_op
->follow_link
)) {
2110 /* We drop rcu-walk here */
2111 if (nameidata_dentry_drop_rcu_maybe(nd
, path
->dentry
))
2112 return ERR_PTR(-ECHILD
);
2115 path_to_nameidata(path
, nd
);
2118 if (nd
->flags
& LOOKUP_RCU
) {
2119 if (nameidata_drop_rcu_last(nd
))
2120 return ERR_PTR(-ECHILD
);
2124 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2125 if (!inode
->i_op
->lookup
)
2128 audit_inode(pathname
, nd
->path
.dentry
);
2132 /* create side of things */
2134 if (nd
->flags
& LOOKUP_RCU
) {
2135 if (nameidata_drop_rcu_last(nd
))
2136 return ERR_PTR(-ECHILD
);
2139 audit_inode(pathname
, dir
);
2141 /* trailing slashes? */
2142 if (nd
->last
.name
[nd
->last
.len
])
2145 mutex_lock(&dir
->d_inode
->i_mutex
);
2147 dentry
= lookup_hash(nd
);
2148 error
= PTR_ERR(dentry
);
2149 if (IS_ERR(dentry
)) {
2150 mutex_unlock(&dir
->d_inode
->i_mutex
);
2154 path
->dentry
= dentry
;
2155 path
->mnt
= nd
->path
.mnt
;
2157 /* Negative dentry, just create the file */
2158 if (!dentry
->d_inode
) {
2159 int mode
= op
->mode
;
2160 if (!IS_POSIXACL(dir
->d_inode
))
2161 mode
&= ~current_umask();
2163 * This write is needed to ensure that a
2164 * rw->ro transition does not occur between
2165 * the time when the file is created and when
2166 * a permanent write count is taken through
2167 * the 'struct file' in nameidata_to_filp().
2169 error
= mnt_want_write(nd
->path
.mnt
);
2171 goto exit_mutex_unlock
;
2173 /* Don't check for write permission, don't truncate */
2174 open_flag
&= ~O_TRUNC
;
2177 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2179 goto exit_mutex_unlock
;
2180 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2182 goto exit_mutex_unlock
;
2183 mutex_unlock(&dir
->d_inode
->i_mutex
);
2184 dput(nd
->path
.dentry
);
2185 nd
->path
.dentry
= dentry
;
2190 * It already exists.
2192 mutex_unlock(&dir
->d_inode
->i_mutex
);
2193 audit_inode(pathname
, path
->dentry
);
2196 if (open_flag
& O_EXCL
)
2199 error
= follow_managed(path
, nd
->flags
);
2204 if (!path
->dentry
->d_inode
)
2207 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2210 path_to_nameidata(path
, nd
);
2211 nd
->inode
= path
->dentry
->d_inode
;
2213 if (S_ISDIR(nd
->inode
->i_mode
))
2216 if (!S_ISREG(nd
->inode
->i_mode
))
2219 if (will_truncate
) {
2220 error
= mnt_want_write(nd
->path
.mnt
);
2226 error
= may_open(&nd
->path
, skip_perm
? 0 : op
->acc_mode
, open_flag
);
2229 filp
= nameidata_to_filp(nd
);
2230 if (!IS_ERR(filp
)) {
2231 error
= ima_file_check(filp
, op
->acc_mode
);
2234 filp
= ERR_PTR(error
);
2237 if (!IS_ERR(filp
)) {
2238 if (will_truncate
) {
2239 error
= handle_truncate(filp
);
2242 filp
= ERR_PTR(error
);
2248 mnt_drop_write(nd
->path
.mnt
);
2249 path_put(&nd
->path
);
2253 mutex_unlock(&dir
->d_inode
->i_mutex
);
2255 path_put_conditional(path
, nd
);
2257 filp
= ERR_PTR(error
);
2261 static struct file
*path_openat(int dfd
, const char *pathname
,
2262 const struct open_flags
*op
, int flags
)
2264 struct file
*base
= NULL
;
2266 struct nameidata nd
;
2271 filp
= get_empty_filp();
2273 return ERR_PTR(-ENFILE
);
2275 filp
->f_flags
= op
->open_flag
;
2276 nd
.intent
.open
.file
= filp
;
2277 nd
.intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2278 nd
.intent
.open
.create_mode
= op
->mode
;
2280 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, &nd
, &base
);
2281 if (unlikely(error
))
2284 current
->total_link_count
= 0;
2285 error
= link_path_walk(pathname
, &nd
);
2286 if (unlikely(error
))
2289 filp
= do_last(&nd
, &path
, op
, pathname
);
2290 while (unlikely(!filp
)) { /* trailing symlink */
2291 struct path link
= path
;
2292 struct inode
*linki
= link
.dentry
->d_inode
;
2294 if (!(nd
.flags
& LOOKUP_FOLLOW
) || count
++ == 32) {
2295 path_put_conditional(&path
, &nd
);
2297 filp
= ERR_PTR(-ELOOP
);
2301 * This is subtle. Instead of calling do_follow_link() we do
2302 * the thing by hands. The reason is that this way we have zero
2303 * link_count and path_walk() (called from ->follow_link)
2304 * honoring LOOKUP_PARENT. After that we have the parent and
2305 * last component, i.e. we are in the same situation as after
2306 * the first path_walk(). Well, almost - if the last component
2307 * is normal we get its copy stored in nd->last.name and we will
2308 * have to putname() it when we are done. Procfs-like symlinks
2309 * just set LAST_BIND.
2311 nd
.flags
|= LOOKUP_PARENT
;
2312 nd
.flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2313 error
= __do_follow_link(&link
, &nd
, &cookie
);
2314 if (unlikely(error
))
2315 filp
= ERR_PTR(error
);
2317 filp
= do_last(&nd
, &path
, op
, pathname
);
2318 if (!IS_ERR(cookie
) && linki
->i_op
->put_link
)
2319 linki
->i_op
->put_link(link
.dentry
, &nd
, cookie
);
2327 release_open_intent(&nd
);
2331 filp
= ERR_PTR(error
);
2335 struct file
*do_filp_open(int dfd
, const char *pathname
,
2336 const struct open_flags
*op
, int flags
)
2340 filp
= path_openat(dfd
, pathname
, op
, flags
| LOOKUP_RCU
);
2341 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2342 filp
= path_openat(dfd
, pathname
, op
, flags
);
2343 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2344 filp
= path_openat(dfd
, pathname
, op
, flags
| LOOKUP_REVAL
);
2349 * lookup_create - lookup a dentry, creating it if it doesn't exist
2350 * @nd: nameidata info
2351 * @is_dir: directory flag
2353 * Simple function to lookup and return a dentry and create it
2354 * if it doesn't exist. Is SMP-safe.
2356 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2358 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2360 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2362 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2364 * Yucky last component or no last component at all?
2365 * (foo/., foo/.., /////)
2367 if (nd
->last_type
!= LAST_NORM
)
2369 nd
->flags
&= ~LOOKUP_PARENT
;
2370 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2371 nd
->intent
.open
.flags
= O_EXCL
;
2374 * Do the final lookup.
2376 dentry
= lookup_hash(nd
);
2380 if (dentry
->d_inode
)
2383 * Special case - lookup gave negative, but... we had foo/bar/
2384 * From the vfs_mknod() POV we just have a negative dentry -
2385 * all is fine. Let's be bastards - you had / on the end, you've
2386 * been asking for (non-existent) directory. -ENOENT for you.
2388 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2390 dentry
= ERR_PTR(-ENOENT
);
2395 dentry
= ERR_PTR(-EEXIST
);
2399 EXPORT_SYMBOL_GPL(lookup_create
);
2401 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2403 int error
= may_create(dir
, dentry
);
2408 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
2411 if (!dir
->i_op
->mknod
)
2414 error
= devcgroup_inode_mknod(mode
, dev
);
2418 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2422 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2424 fsnotify_create(dir
, dentry
);
2428 static int may_mknod(mode_t mode
)
2430 switch (mode
& S_IFMT
) {
2436 case 0: /* zero mode translates to S_IFREG */
2445 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2450 struct dentry
*dentry
;
2451 struct nameidata nd
;
2456 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2460 dentry
= lookup_create(&nd
, 0);
2461 if (IS_ERR(dentry
)) {
2462 error
= PTR_ERR(dentry
);
2465 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2466 mode
&= ~current_umask();
2467 error
= may_mknod(mode
);
2470 error
= mnt_want_write(nd
.path
.mnt
);
2473 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2475 goto out_drop_write
;
2476 switch (mode
& S_IFMT
) {
2477 case 0: case S_IFREG
:
2478 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2480 case S_IFCHR
: case S_IFBLK
:
2481 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2482 new_decode_dev(dev
));
2484 case S_IFIFO
: case S_IFSOCK
:
2485 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2489 mnt_drop_write(nd
.path
.mnt
);
2493 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2500 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2502 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2505 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2507 int error
= may_create(dir
, dentry
);
2512 if (!dir
->i_op
->mkdir
)
2515 mode
&= (S_IRWXUGO
|S_ISVTX
);
2516 error
= security_inode_mkdir(dir
, dentry
, mode
);
2520 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2522 fsnotify_mkdir(dir
, dentry
);
2526 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2530 struct dentry
*dentry
;
2531 struct nameidata nd
;
2533 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2537 dentry
= lookup_create(&nd
, 1);
2538 error
= PTR_ERR(dentry
);
2542 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2543 mode
&= ~current_umask();
2544 error
= mnt_want_write(nd
.path
.mnt
);
2547 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2549 goto out_drop_write
;
2550 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2552 mnt_drop_write(nd
.path
.mnt
);
2556 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2563 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2565 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2569 * We try to drop the dentry early: we should have
2570 * a usage count of 2 if we're the only user of this
2571 * dentry, and if that is true (possibly after pruning
2572 * the dcache), then we drop the dentry now.
2574 * A low-level filesystem can, if it choses, legally
2577 * if (!d_unhashed(dentry))
2580 * if it cannot handle the case of removing a directory
2581 * that is still in use by something else..
2583 void dentry_unhash(struct dentry
*dentry
)
2586 shrink_dcache_parent(dentry
);
2587 spin_lock(&dentry
->d_lock
);
2588 if (dentry
->d_count
== 2)
2590 spin_unlock(&dentry
->d_lock
);
2593 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2595 int error
= may_delete(dir
, dentry
, 1);
2600 if (!dir
->i_op
->rmdir
)
2603 mutex_lock(&dentry
->d_inode
->i_mutex
);
2604 dentry_unhash(dentry
);
2605 if (d_mountpoint(dentry
))
2608 error
= security_inode_rmdir(dir
, dentry
);
2610 error
= dir
->i_op
->rmdir(dir
, dentry
);
2612 dentry
->d_inode
->i_flags
|= S_DEAD
;
2617 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2626 static long do_rmdir(int dfd
, const char __user
*pathname
)
2630 struct dentry
*dentry
;
2631 struct nameidata nd
;
2633 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2637 switch(nd
.last_type
) {
2649 nd
.flags
&= ~LOOKUP_PARENT
;
2651 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2652 dentry
= lookup_hash(&nd
);
2653 error
= PTR_ERR(dentry
);
2656 error
= mnt_want_write(nd
.path
.mnt
);
2659 error
= security_path_rmdir(&nd
.path
, dentry
);
2662 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2664 mnt_drop_write(nd
.path
.mnt
);
2668 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2675 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2677 return do_rmdir(AT_FDCWD
, pathname
);
2680 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2682 int error
= may_delete(dir
, dentry
, 0);
2687 if (!dir
->i_op
->unlink
)
2690 mutex_lock(&dentry
->d_inode
->i_mutex
);
2691 if (d_mountpoint(dentry
))
2694 error
= security_inode_unlink(dir
, dentry
);
2696 error
= dir
->i_op
->unlink(dir
, dentry
);
2701 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2703 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2704 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2705 fsnotify_link_count(dentry
->d_inode
);
2713 * Make sure that the actual truncation of the file will occur outside its
2714 * directory's i_mutex. Truncate can take a long time if there is a lot of
2715 * writeout happening, and we don't want to prevent access to the directory
2716 * while waiting on the I/O.
2718 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2722 struct dentry
*dentry
;
2723 struct nameidata nd
;
2724 struct inode
*inode
= NULL
;
2726 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2731 if (nd
.last_type
!= LAST_NORM
)
2734 nd
.flags
&= ~LOOKUP_PARENT
;
2736 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2737 dentry
= lookup_hash(&nd
);
2738 error
= PTR_ERR(dentry
);
2739 if (!IS_ERR(dentry
)) {
2740 /* Why not before? Because we want correct error value */
2741 if (nd
.last
.name
[nd
.last
.len
])
2743 inode
= dentry
->d_inode
;
2746 error
= mnt_want_write(nd
.path
.mnt
);
2749 error
= security_path_unlink(&nd
.path
, dentry
);
2752 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2754 mnt_drop_write(nd
.path
.mnt
);
2758 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2760 iput(inode
); /* truncate the inode here */
2767 error
= !dentry
->d_inode
? -ENOENT
:
2768 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2772 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2774 if ((flag
& ~AT_REMOVEDIR
) != 0)
2777 if (flag
& AT_REMOVEDIR
)
2778 return do_rmdir(dfd
, pathname
);
2780 return do_unlinkat(dfd
, pathname
);
2783 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2785 return do_unlinkat(AT_FDCWD
, pathname
);
2788 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2790 int error
= may_create(dir
, dentry
);
2795 if (!dir
->i_op
->symlink
)
2798 error
= security_inode_symlink(dir
, dentry
, oldname
);
2802 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2804 fsnotify_create(dir
, dentry
);
2808 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2809 int, newdfd
, const char __user
*, newname
)
2814 struct dentry
*dentry
;
2815 struct nameidata nd
;
2817 from
= getname(oldname
);
2819 return PTR_ERR(from
);
2821 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2825 dentry
= lookup_create(&nd
, 0);
2826 error
= PTR_ERR(dentry
);
2830 error
= mnt_want_write(nd
.path
.mnt
);
2833 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2835 goto out_drop_write
;
2836 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2838 mnt_drop_write(nd
.path
.mnt
);
2842 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2850 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2852 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2855 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2857 struct inode
*inode
= old_dentry
->d_inode
;
2863 error
= may_create(dir
, new_dentry
);
2867 if (dir
->i_sb
!= inode
->i_sb
)
2871 * A link to an append-only or immutable file cannot be created.
2873 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2875 if (!dir
->i_op
->link
)
2877 if (S_ISDIR(inode
->i_mode
))
2880 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2884 mutex_lock(&inode
->i_mutex
);
2885 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2886 mutex_unlock(&inode
->i_mutex
);
2888 fsnotify_link(dir
, inode
, new_dentry
);
2893 * Hardlinks are often used in delicate situations. We avoid
2894 * security-related surprises by not following symlinks on the
2897 * We don't follow them on the oldname either to be compatible
2898 * with linux 2.0, and to avoid hard-linking to directories
2899 * and other special files. --ADM
2901 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2902 int, newdfd
, const char __user
*, newname
, int, flags
)
2904 struct dentry
*new_dentry
;
2905 struct nameidata nd
;
2906 struct path old_path
;
2910 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
2913 error
= user_path_at(olddfd
, oldname
,
2914 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
2919 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2923 if (old_path
.mnt
!= nd
.path
.mnt
)
2925 new_dentry
= lookup_create(&nd
, 0);
2926 error
= PTR_ERR(new_dentry
);
2927 if (IS_ERR(new_dentry
))
2929 error
= mnt_want_write(nd
.path
.mnt
);
2932 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2934 goto out_drop_write
;
2935 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2937 mnt_drop_write(nd
.path
.mnt
);
2941 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2946 path_put(&old_path
);
2951 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
2953 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2957 * The worst of all namespace operations - renaming directory. "Perverted"
2958 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2960 * a) we can get into loop creation. Check is done in is_subdir().
2961 * b) race potential - two innocent renames can create a loop together.
2962 * That's where 4.4 screws up. Current fix: serialization on
2963 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2965 * c) we have to lock _three_ objects - parents and victim (if it exists).
2966 * And that - after we got ->i_mutex on parents (until then we don't know
2967 * whether the target exists). Solution: try to be smart with locking
2968 * order for inodes. We rely on the fact that tree topology may change
2969 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2970 * move will be locked. Thus we can rank directories by the tree
2971 * (ancestors first) and rank all non-directories after them.
2972 * That works since everybody except rename does "lock parent, lookup,
2973 * lock child" and rename is under ->s_vfs_rename_mutex.
2974 * HOWEVER, it relies on the assumption that any object with ->lookup()
2975 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2976 * we'd better make sure that there's no link(2) for them.
2977 * d) some filesystems don't support opened-but-unlinked directories,
2978 * either because of layout or because they are not ready to deal with
2979 * all cases correctly. The latter will be fixed (taking this sort of
2980 * stuff into VFS), but the former is not going away. Solution: the same
2981 * trick as in rmdir().
2982 * e) conversion from fhandle to dentry may come in the wrong moment - when
2983 * we are removing the target. Solution: we will have to grab ->i_mutex
2984 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2985 * ->i_mutex on parents, which works but leads to some truly excessive
2988 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2989 struct inode
*new_dir
, struct dentry
*new_dentry
)
2992 struct inode
*target
;
2995 * If we are going to change the parent - check write permissions,
2996 * we'll need to flip '..'.
2998 if (new_dir
!= old_dir
) {
2999 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3004 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3008 target
= new_dentry
->d_inode
;
3010 mutex_lock(&target
->i_mutex
);
3011 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3015 dentry_unhash(new_dentry
);
3016 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3020 target
->i_flags
|= S_DEAD
;
3021 dont_mount(new_dentry
);
3023 mutex_unlock(&target
->i_mutex
);
3024 if (d_unhashed(new_dentry
))
3025 d_rehash(new_dentry
);
3029 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3030 d_move(old_dentry
,new_dentry
);
3034 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3035 struct inode
*new_dir
, struct dentry
*new_dentry
)
3037 struct inode
*target
;
3040 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3045 target
= new_dentry
->d_inode
;
3047 mutex_lock(&target
->i_mutex
);
3048 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3051 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3054 dont_mount(new_dentry
);
3055 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3056 d_move(old_dentry
, new_dentry
);
3059 mutex_unlock(&target
->i_mutex
);
3064 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3065 struct inode
*new_dir
, struct dentry
*new_dentry
)
3068 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3069 const unsigned char *old_name
;
3071 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3074 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3078 if (!new_dentry
->d_inode
)
3079 error
= may_create(new_dir
, new_dentry
);
3081 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3085 if (!old_dir
->i_op
->rename
)
3088 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3091 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3093 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3095 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3096 new_dentry
->d_inode
, old_dentry
);
3097 fsnotify_oldname_free(old_name
);
3102 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3103 int, newdfd
, const char __user
*, newname
)
3105 struct dentry
*old_dir
, *new_dir
;
3106 struct dentry
*old_dentry
, *new_dentry
;
3107 struct dentry
*trap
;
3108 struct nameidata oldnd
, newnd
;
3113 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3117 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3122 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3125 old_dir
= oldnd
.path
.dentry
;
3127 if (oldnd
.last_type
!= LAST_NORM
)
3130 new_dir
= newnd
.path
.dentry
;
3131 if (newnd
.last_type
!= LAST_NORM
)
3134 oldnd
.flags
&= ~LOOKUP_PARENT
;
3135 newnd
.flags
&= ~LOOKUP_PARENT
;
3136 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3138 trap
= lock_rename(new_dir
, old_dir
);
3140 old_dentry
= lookup_hash(&oldnd
);
3141 error
= PTR_ERR(old_dentry
);
3142 if (IS_ERR(old_dentry
))
3144 /* source must exist */
3146 if (!old_dentry
->d_inode
)
3148 /* unless the source is a directory trailing slashes give -ENOTDIR */
3149 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3151 if (oldnd
.last
.name
[oldnd
.last
.len
])
3153 if (newnd
.last
.name
[newnd
.last
.len
])
3156 /* source should not be ancestor of target */
3158 if (old_dentry
== trap
)
3160 new_dentry
= lookup_hash(&newnd
);
3161 error
= PTR_ERR(new_dentry
);
3162 if (IS_ERR(new_dentry
))
3164 /* target should not be an ancestor of source */
3166 if (new_dentry
== trap
)
3169 error
= mnt_want_write(oldnd
.path
.mnt
);
3172 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3173 &newnd
.path
, new_dentry
);
3176 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3177 new_dir
->d_inode
, new_dentry
);
3179 mnt_drop_write(oldnd
.path
.mnt
);
3185 unlock_rename(new_dir
, old_dir
);
3187 path_put(&newnd
.path
);
3190 path_put(&oldnd
.path
);
3196 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3198 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3201 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3205 len
= PTR_ERR(link
);
3210 if (len
> (unsigned) buflen
)
3212 if (copy_to_user(buffer
, link
, len
))
3219 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3220 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3221 * using) it for any given inode is up to filesystem.
3223 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3225 struct nameidata nd
;
3230 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3232 return PTR_ERR(cookie
);
3234 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3235 if (dentry
->d_inode
->i_op
->put_link
)
3236 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3240 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3242 return __vfs_follow_link(nd
, link
);
3245 /* get the link contents into pagecache */
3246 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3250 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3251 page
= read_mapping_page(mapping
, 0, NULL
);
3256 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3260 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3262 struct page
*page
= NULL
;
3263 char *s
= page_getlink(dentry
, &page
);
3264 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3267 page_cache_release(page
);
3272 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3274 struct page
*page
= NULL
;
3275 nd_set_link(nd
, page_getlink(dentry
, &page
));
3279 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3281 struct page
*page
= cookie
;
3285 page_cache_release(page
);
3290 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3292 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3294 struct address_space
*mapping
= inode
->i_mapping
;
3299 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3301 flags
|= AOP_FLAG_NOFS
;
3304 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3305 flags
, &page
, &fsdata
);
3309 kaddr
= kmap_atomic(page
, KM_USER0
);
3310 memcpy(kaddr
, symname
, len
-1);
3311 kunmap_atomic(kaddr
, KM_USER0
);
3313 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3320 mark_inode_dirty(inode
);
3326 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3328 return __page_symlink(inode
, symname
, len
,
3329 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3332 const struct inode_operations page_symlink_inode_operations
= {
3333 .readlink
= generic_readlink
,
3334 .follow_link
= page_follow_link_light
,
3335 .put_link
= page_put_link
,
3338 EXPORT_SYMBOL(user_path_at
);
3339 EXPORT_SYMBOL(follow_down_one
);
3340 EXPORT_SYMBOL(follow_down
);
3341 EXPORT_SYMBOL(follow_up
);
3342 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3343 EXPORT_SYMBOL(getname
);
3344 EXPORT_SYMBOL(lock_rename
);
3345 EXPORT_SYMBOL(lookup_one_len
);
3346 EXPORT_SYMBOL(page_follow_link_light
);
3347 EXPORT_SYMBOL(page_put_link
);
3348 EXPORT_SYMBOL(page_readlink
);
3349 EXPORT_SYMBOL(__page_symlink
);
3350 EXPORT_SYMBOL(page_symlink
);
3351 EXPORT_SYMBOL(page_symlink_inode_operations
);
3352 EXPORT_SYMBOL(kern_path_parent
);
3353 EXPORT_SYMBOL(kern_path
);
3354 EXPORT_SYMBOL(vfs_path_lookup
);
3355 EXPORT_SYMBOL(inode_permission
);
3356 EXPORT_SYMBOL(file_permission
);
3357 EXPORT_SYMBOL(unlock_rename
);
3358 EXPORT_SYMBOL(vfs_create
);
3359 EXPORT_SYMBOL(vfs_follow_link
);
3360 EXPORT_SYMBOL(vfs_link
);
3361 EXPORT_SYMBOL(vfs_mkdir
);
3362 EXPORT_SYMBOL(vfs_mknod
);
3363 EXPORT_SYMBOL(generic_permission
);
3364 EXPORT_SYMBOL(vfs_readlink
);
3365 EXPORT_SYMBOL(vfs_rename
);
3366 EXPORT_SYMBOL(vfs_rmdir
);
3367 EXPORT_SYMBOL(vfs_symlink
);
3368 EXPORT_SYMBOL(vfs_unlink
);
3369 EXPORT_SYMBOL(dentry_unhash
);
3370 EXPORT_SYMBOL(generic_readlink
);