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-existent 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 unsigned int 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 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
701 } while (read_seqcount_retry(&fs
->seq
, seq
));
705 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
717 nd
->flags
|= LOOKUP_JUMPED
;
719 nd
->inode
= nd
->path
.dentry
->d_inode
;
721 ret
= link_path_walk(link
, nd
);
725 return PTR_ERR(link
);
728 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
731 if (path
->mnt
!= nd
->path
.mnt
)
735 static inline void path_to_nameidata(const struct path
*path
,
736 struct nameidata
*nd
)
738 if (!(nd
->flags
& LOOKUP_RCU
)) {
739 dput(nd
->path
.dentry
);
740 if (nd
->path
.mnt
!= path
->mnt
)
741 mntput(nd
->path
.mnt
);
743 nd
->path
.mnt
= path
->mnt
;
744 nd
->path
.dentry
= path
->dentry
;
747 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
749 struct inode
*inode
= link
->dentry
->d_inode
;
750 if (!IS_ERR(cookie
) && inode
->i_op
->put_link
)
751 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
755 static __always_inline
int
756 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
759 struct dentry
*dentry
= link
->dentry
;
761 BUG_ON(nd
->flags
& LOOKUP_RCU
);
763 if (link
->mnt
== nd
->path
.mnt
)
766 if (unlikely(current
->total_link_count
>= 40)) {
767 *p
= ERR_PTR(-ELOOP
); /* no ->put_link(), please */
772 current
->total_link_count
++;
774 touch_atime(link
->mnt
, dentry
);
775 nd_set_link(nd
, NULL
);
777 error
= security_inode_follow_link(link
->dentry
, nd
);
779 *p
= ERR_PTR(error
); /* no ->put_link(), please */
784 nd
->last_type
= LAST_BIND
;
785 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
788 char *s
= nd_get_link(nd
);
791 error
= __vfs_follow_link(nd
, s
);
792 else if (nd
->last_type
== LAST_BIND
) {
793 nd
->flags
|= LOOKUP_JUMPED
;
794 nd
->inode
= nd
->path
.dentry
->d_inode
;
795 if (nd
->inode
->i_op
->follow_link
) {
796 /* stepped on a _really_ weird one */
805 static int follow_up_rcu(struct path
*path
)
807 struct vfsmount
*parent
;
808 struct dentry
*mountpoint
;
810 parent
= path
->mnt
->mnt_parent
;
811 if (parent
== path
->mnt
)
813 mountpoint
= path
->mnt
->mnt_mountpoint
;
814 path
->dentry
= mountpoint
;
819 int follow_up(struct path
*path
)
821 struct vfsmount
*parent
;
822 struct dentry
*mountpoint
;
824 br_read_lock(vfsmount_lock
);
825 parent
= path
->mnt
->mnt_parent
;
826 if (parent
== path
->mnt
) {
827 br_read_unlock(vfsmount_lock
);
831 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
832 br_read_unlock(vfsmount_lock
);
834 path
->dentry
= mountpoint
;
841 * Perform an automount
842 * - return -EISDIR to tell follow_managed() to stop and return the path we
845 static int follow_automount(struct path
*path
, unsigned flags
,
848 struct vfsmount
*mnt
;
851 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
854 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
855 * and this is the terminal part of the path.
857 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
858 return -EISDIR
; /* we actually want to stop here */
860 /* We want to mount if someone is trying to open/create a file of any
861 * type under the mountpoint, wants to traverse through the mountpoint
862 * or wants to open the mounted directory.
864 * We don't want to mount if someone's just doing a stat and they've
865 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
866 * appended a '/' to the name.
868 if (!(flags
& LOOKUP_FOLLOW
) &&
869 !(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
870 LOOKUP_OPEN
| LOOKUP_CREATE
)))
873 current
->total_link_count
++;
874 if (current
->total_link_count
>= 40)
877 mnt
= path
->dentry
->d_op
->d_automount(path
);
880 * The filesystem is allowed to return -EISDIR here to indicate
881 * it doesn't want to automount. For instance, autofs would do
882 * this so that its userspace daemon can mount on this dentry.
884 * However, we can only permit this if it's a terminal point in
885 * the path being looked up; if it wasn't then the remainder of
886 * the path is inaccessible and we should say so.
888 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_CONTINUE
))
893 if (!mnt
) /* mount collision */
896 err
= finish_automount(mnt
, path
);
900 /* Someone else made a mount here whilst we were busy */
907 path
->dentry
= dget(mnt
->mnt_root
);
917 * Handle a dentry that is managed in some way.
918 * - Flagged for transit management (autofs)
919 * - Flagged as mountpoint
920 * - Flagged as automount point
922 * This may only be called in refwalk mode.
924 * Serialization is taken care of in namespace.c
926 static int follow_managed(struct path
*path
, unsigned flags
)
929 bool need_mntput
= false;
932 /* Given that we're not holding a lock here, we retain the value in a
933 * local variable for each dentry as we look at it so that we don't see
934 * the components of that value change under us */
935 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
936 managed
&= DCACHE_MANAGED_DENTRY
,
937 unlikely(managed
!= 0)) {
938 /* Allow the filesystem to manage the transit without i_mutex
940 if (managed
& DCACHE_MANAGE_TRANSIT
) {
941 BUG_ON(!path
->dentry
->d_op
);
942 BUG_ON(!path
->dentry
->d_op
->d_manage
);
943 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
945 return ret
== -EISDIR
? 0 : ret
;
948 /* Transit to a mounted filesystem. */
949 if (managed
& DCACHE_MOUNTED
) {
950 struct vfsmount
*mounted
= lookup_mnt(path
);
956 path
->dentry
= dget(mounted
->mnt_root
);
961 /* Something is mounted on this dentry in another
962 * namespace and/or whatever was mounted there in this
963 * namespace got unmounted before we managed to get the
967 /* Handle an automount point */
968 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
969 ret
= follow_automount(path
, flags
, &need_mntput
);
971 return ret
== -EISDIR
? 0 : ret
;
975 /* We didn't change the current path point */
981 int follow_down_one(struct path
*path
)
983 struct vfsmount
*mounted
;
985 mounted
= lookup_mnt(path
);
990 path
->dentry
= dget(mounted
->mnt_root
);
996 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
998 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
999 dentry
->d_op
->d_manage(dentry
, true) < 0);
1003 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
1004 * meet a managed dentry and we're not walking to "..". True is returned to
1005 * continue, false to abort.
1007 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1008 struct inode
**inode
, bool reverse_transit
)
1011 struct vfsmount
*mounted
;
1013 * Don't forget we might have a non-mountpoint managed dentry
1014 * that wants to block transit.
1016 if (!reverse_transit
&&
1017 unlikely(managed_dentry_might_block(path
->dentry
)))
1020 if (!d_mountpoint(path
->dentry
))
1023 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1026 path
->mnt
= mounted
;
1027 path
->dentry
= mounted
->mnt_root
;
1028 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1030 * Update the inode too. We don't need to re-check the
1031 * dentry sequence number here after this d_inode read,
1032 * because a mount-point is always pinned.
1034 *inode
= path
->dentry
->d_inode
;
1037 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1038 return reverse_transit
;
1042 static int follow_dotdot_rcu(struct nameidata
*nd
)
1044 struct inode
*inode
= nd
->inode
;
1049 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1050 nd
->path
.mnt
== nd
->root
.mnt
) {
1053 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1054 struct dentry
*old
= nd
->path
.dentry
;
1055 struct dentry
*parent
= old
->d_parent
;
1058 seq
= read_seqcount_begin(&parent
->d_seq
);
1059 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1061 inode
= parent
->d_inode
;
1062 nd
->path
.dentry
= parent
;
1066 if (!follow_up_rcu(&nd
->path
))
1068 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1069 inode
= nd
->path
.dentry
->d_inode
;
1071 __follow_mount_rcu(nd
, &nd
->path
, &inode
, true);
1076 nd
->flags
&= ~LOOKUP_RCU
;
1077 if (!(nd
->flags
& LOOKUP_ROOT
))
1078 nd
->root
.mnt
= NULL
;
1080 br_read_unlock(vfsmount_lock
);
1085 * Follow down to the covering mount currently visible to userspace. At each
1086 * point, the filesystem owning that dentry may be queried as to whether the
1087 * caller is permitted to proceed or not.
1089 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1092 int follow_down(struct path
*path
)
1097 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1098 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1099 /* Allow the filesystem to manage the transit without i_mutex
1102 * We indicate to the filesystem if someone is trying to mount
1103 * something here. This gives autofs the chance to deny anyone
1104 * other than its daemon the right to mount on its
1107 * The filesystem may sleep at this point.
1109 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1110 BUG_ON(!path
->dentry
->d_op
);
1111 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1112 ret
= path
->dentry
->d_op
->d_manage(
1113 path
->dentry
, false);
1115 return ret
== -EISDIR
? 0 : ret
;
1118 /* Transit to a mounted filesystem. */
1119 if (managed
& DCACHE_MOUNTED
) {
1120 struct vfsmount
*mounted
= lookup_mnt(path
);
1125 path
->mnt
= mounted
;
1126 path
->dentry
= dget(mounted
->mnt_root
);
1130 /* Don't handle automount points here */
1137 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1139 static void follow_mount(struct path
*path
)
1141 while (d_mountpoint(path
->dentry
)) {
1142 struct vfsmount
*mounted
= lookup_mnt(path
);
1147 path
->mnt
= mounted
;
1148 path
->dentry
= dget(mounted
->mnt_root
);
1152 static void follow_dotdot(struct nameidata
*nd
)
1157 struct dentry
*old
= nd
->path
.dentry
;
1159 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1160 nd
->path
.mnt
== nd
->root
.mnt
) {
1163 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1164 /* rare case of legitimate dget_parent()... */
1165 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1169 if (!follow_up(&nd
->path
))
1172 follow_mount(&nd
->path
);
1173 nd
->inode
= nd
->path
.dentry
->d_inode
;
1177 * Allocate a dentry with name and parent, and perform a parent
1178 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1179 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1180 * have verified that no child exists while under i_mutex.
1182 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1183 struct qstr
*name
, struct nameidata
*nd
)
1185 struct inode
*inode
= parent
->d_inode
;
1186 struct dentry
*dentry
;
1189 /* Don't create child dentry for a dead directory. */
1190 if (unlikely(IS_DEADDIR(inode
)))
1191 return ERR_PTR(-ENOENT
);
1193 dentry
= d_alloc(parent
, name
);
1194 if (unlikely(!dentry
))
1195 return ERR_PTR(-ENOMEM
);
1197 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1198 if (unlikely(old
)) {
1206 * It's more convoluted than I'd like it to be, but... it's still fairly
1207 * small and for now I'd prefer to have fast path as straight as possible.
1208 * It _is_ time-critical.
1210 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1211 struct path
*path
, struct inode
**inode
)
1213 struct vfsmount
*mnt
= nd
->path
.mnt
;
1214 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1220 * Rename seqlock is not required here because in the off chance
1221 * of a false negative due to a concurrent rename, we're going to
1222 * do the non-racy lookup, below.
1224 if (nd
->flags
& LOOKUP_RCU
) {
1227 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1231 /* Memory barrier in read_seqcount_begin of child is enough */
1232 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1236 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1237 status
= d_revalidate(dentry
, nd
);
1238 if (unlikely(status
<= 0)) {
1239 if (status
!= -ECHILD
)
1245 path
->dentry
= dentry
;
1246 if (likely(__follow_mount_rcu(nd
, path
, inode
, false)))
1250 if (nameidata_dentry_drop_rcu(nd
, dentry
))
1253 if (nameidata_drop_rcu(nd
))
1257 dentry
= __d_lookup(parent
, name
);
1261 if (unlikely(!dentry
)) {
1262 struct inode
*dir
= parent
->d_inode
;
1263 BUG_ON(nd
->inode
!= dir
);
1265 mutex_lock(&dir
->i_mutex
);
1266 dentry
= d_lookup(parent
, name
);
1267 if (likely(!dentry
)) {
1268 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1269 if (IS_ERR(dentry
)) {
1270 mutex_unlock(&dir
->i_mutex
);
1271 return PTR_ERR(dentry
);
1277 mutex_unlock(&dir
->i_mutex
);
1279 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1280 status
= d_revalidate(dentry
, nd
);
1281 if (unlikely(status
<= 0)) {
1286 if (!d_invalidate(dentry
)) {
1295 path
->dentry
= dentry
;
1296 err
= follow_managed(path
, nd
->flags
);
1297 if (unlikely(err
< 0)) {
1298 path_put_conditional(path
, nd
);
1301 *inode
= path
->dentry
->d_inode
;
1305 static inline int may_lookup(struct nameidata
*nd
)
1307 if (nd
->flags
& LOOKUP_RCU
) {
1308 int err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1311 if (nameidata_drop_rcu(nd
))
1314 return exec_permission(nd
->inode
, 0);
1317 static inline int handle_dots(struct nameidata
*nd
, int type
)
1319 if (type
== LAST_DOTDOT
) {
1320 if (nd
->flags
& LOOKUP_RCU
) {
1321 if (follow_dotdot_rcu(nd
))
1329 static void terminate_walk(struct nameidata
*nd
)
1331 if (!(nd
->flags
& LOOKUP_RCU
)) {
1332 path_put(&nd
->path
);
1334 nd
->flags
&= ~LOOKUP_RCU
;
1335 if (!(nd
->flags
& LOOKUP_ROOT
))
1336 nd
->root
.mnt
= NULL
;
1338 br_read_unlock(vfsmount_lock
);
1342 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1343 struct qstr
*name
, int type
, int follow
)
1345 struct inode
*inode
;
1348 * "." and ".." are special - ".." especially so because it has
1349 * to be able to know about the current root directory and
1350 * parent relationships.
1352 if (unlikely(type
!= LAST_NORM
))
1353 return handle_dots(nd
, type
);
1354 err
= do_lookup(nd
, name
, path
, &inode
);
1355 if (unlikely(err
)) {
1360 path_to_nameidata(path
, nd
);
1364 if (unlikely(inode
->i_op
->follow_link
) && follow
) {
1365 if (nameidata_dentry_drop_rcu_maybe(nd
, path
->dentry
))
1367 BUG_ON(inode
!= path
->dentry
->d_inode
);
1370 path_to_nameidata(path
, nd
);
1376 * This limits recursive symlink follows to 8, while
1377 * limiting consecutive symlinks to 40.
1379 * Without that kind of total limit, nasty chains of consecutive
1380 * symlinks can cause almost arbitrarily long lookups.
1382 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1386 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1387 path_put_conditional(path
, nd
);
1388 path_put(&nd
->path
);
1391 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1394 current
->link_count
++;
1397 struct path link
= *path
;
1400 res
= follow_link(&link
, nd
, &cookie
);
1402 res
= walk_component(nd
, path
, &nd
->last
,
1403 nd
->last_type
, LOOKUP_FOLLOW
);
1404 put_link(nd
, &link
, cookie
);
1407 current
->link_count
--;
1414 * This is the basic name resolution function, turning a pathname into
1415 * the final dentry. We expect 'base' to be positive and a directory.
1417 * Returns 0 and nd will have valid dentry and mnt on success.
1418 * Returns error and drops reference to input namei data on failure.
1420 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1424 unsigned int lookup_flags
= nd
->flags
;
1431 /* At this point we know we have a real path component. */
1438 nd
->flags
|= LOOKUP_CONTINUE
;
1440 err
= may_lookup(nd
);
1445 c
= *(const unsigned char *)name
;
1447 hash
= init_name_hash();
1450 hash
= partial_name_hash(c
, hash
);
1451 c
= *(const unsigned char *)name
;
1452 } while (c
&& (c
!= '/'));
1453 this.len
= name
- (const char *) this.name
;
1454 this.hash
= end_name_hash(hash
);
1457 if (this.name
[0] == '.') switch (this.len
) {
1459 if (this.name
[1] == '.') {
1461 nd
->flags
|= LOOKUP_JUMPED
;
1467 if (likely(type
== LAST_NORM
)) {
1468 struct dentry
*parent
= nd
->path
.dentry
;
1469 nd
->flags
&= ~LOOKUP_JUMPED
;
1470 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1471 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1478 /* remove trailing slashes? */
1480 goto last_component
;
1481 while (*++name
== '/');
1483 goto last_component
;
1485 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1490 err
= nested_symlink(&next
, nd
);
1495 if (!nd
->inode
->i_op
->lookup
)
1498 /* here ends the main loop */
1501 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1502 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1504 nd
->last_type
= type
;
1511 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1512 struct nameidata
*nd
, struct file
**fp
)
1518 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1519 nd
->flags
= flags
| LOOKUP_JUMPED
;
1521 if (flags
& LOOKUP_ROOT
) {
1522 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1524 if (!inode
->i_op
->lookup
)
1526 retval
= inode_permission(inode
, MAY_EXEC
);
1530 nd
->path
= nd
->root
;
1532 if (flags
& LOOKUP_RCU
) {
1533 br_read_lock(vfsmount_lock
);
1535 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1537 path_get(&nd
->path
);
1542 nd
->root
.mnt
= NULL
;
1545 if (flags
& LOOKUP_RCU
) {
1546 br_read_lock(vfsmount_lock
);
1551 path_get(&nd
->root
);
1553 nd
->path
= nd
->root
;
1554 } else if (dfd
== AT_FDCWD
) {
1555 if (flags
& LOOKUP_RCU
) {
1556 struct fs_struct
*fs
= current
->fs
;
1559 br_read_lock(vfsmount_lock
);
1563 seq
= read_seqcount_begin(&fs
->seq
);
1565 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1566 } while (read_seqcount_retry(&fs
->seq
, seq
));
1568 get_fs_pwd(current
->fs
, &nd
->path
);
1571 struct dentry
*dentry
;
1573 file
= fget_raw_light(dfd
, &fput_needed
);
1578 dentry
= file
->f_path
.dentry
;
1582 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1585 retval
= file_permission(file
, MAY_EXEC
);
1590 nd
->path
= file
->f_path
;
1591 if (flags
& LOOKUP_RCU
) {
1594 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1595 br_read_lock(vfsmount_lock
);
1598 path_get(&file
->f_path
);
1599 fput_light(file
, fput_needed
);
1603 nd
->inode
= nd
->path
.dentry
->d_inode
;
1607 fput_light(file
, fput_needed
);
1612 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1614 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1615 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1617 nd
->flags
&= ~LOOKUP_PARENT
;
1618 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1619 nd
->flags
& LOOKUP_FOLLOW
);
1622 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1623 static int path_lookupat(int dfd
, const char *name
,
1624 unsigned int flags
, struct nameidata
*nd
)
1626 struct file
*base
= NULL
;
1631 * Path walking is largely split up into 2 different synchronisation
1632 * schemes, rcu-walk and ref-walk (explained in
1633 * Documentation/filesystems/path-lookup.txt). These share much of the
1634 * path walk code, but some things particularly setup, cleanup, and
1635 * following mounts are sufficiently divergent that functions are
1636 * duplicated. Typically there is a function foo(), and its RCU
1637 * analogue, foo_rcu().
1639 * -ECHILD is the error number of choice (just to avoid clashes) that
1640 * is returned if some aspect of an rcu-walk fails. Such an error must
1641 * be handled by restarting a traditional ref-walk (which will always
1642 * be able to complete).
1644 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1649 current
->total_link_count
= 0;
1650 err
= link_path_walk(name
, nd
);
1652 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1653 err
= lookup_last(nd
, &path
);
1656 struct path link
= path
;
1657 nd
->flags
|= LOOKUP_PARENT
;
1658 err
= follow_link(&link
, nd
, &cookie
);
1660 err
= lookup_last(nd
, &path
);
1661 put_link(nd
, &link
, cookie
);
1665 if (nd
->flags
& LOOKUP_RCU
) {
1666 /* went all way through without dropping RCU */
1668 if (nameidata_drop_rcu_last(nd
))
1673 err
= handle_reval_path(nd
);
1675 path_put(&nd
->path
);
1678 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1679 if (!nd
->inode
->i_op
->lookup
) {
1680 path_put(&nd
->path
);
1688 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1689 path_put(&nd
->root
);
1690 nd
->root
.mnt
= NULL
;
1695 static int do_path_lookup(int dfd
, const char *name
,
1696 unsigned int flags
, struct nameidata
*nd
)
1698 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1699 if (unlikely(retval
== -ECHILD
))
1700 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1701 if (unlikely(retval
== -ESTALE
))
1702 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1704 if (likely(!retval
)) {
1705 if (unlikely(!audit_dummy_context())) {
1706 if (nd
->path
.dentry
&& nd
->inode
)
1707 audit_inode(name
, nd
->path
.dentry
);
1713 int kern_path_parent(const char *name
, struct nameidata
*nd
)
1715 return do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, nd
);
1718 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1720 struct nameidata nd
;
1721 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1728 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1729 * @dentry: pointer to dentry of the base directory
1730 * @mnt: pointer to vfs mount of the base directory
1731 * @name: pointer to file name
1732 * @flags: lookup flags
1733 * @nd: pointer to nameidata
1735 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1736 const char *name
, unsigned int flags
,
1737 struct nameidata
*nd
)
1739 nd
->root
.dentry
= dentry
;
1741 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1742 return do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, nd
);
1745 static struct dentry
*__lookup_hash(struct qstr
*name
,
1746 struct dentry
*base
, struct nameidata
*nd
)
1748 struct inode
*inode
= base
->d_inode
;
1749 struct dentry
*dentry
;
1752 err
= exec_permission(inode
, 0);
1754 return ERR_PTR(err
);
1757 * Don't bother with __d_lookup: callers are for creat as
1758 * well as unlink, so a lot of the time it would cost
1761 dentry
= d_lookup(base
, name
);
1763 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1764 dentry
= do_revalidate(dentry
, nd
);
1767 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1773 * Restricted form of lookup. Doesn't follow links, single-component only,
1774 * needs parent already locked. Doesn't follow mounts.
1777 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1779 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1783 * lookup_one_len - filesystem helper to lookup single pathname component
1784 * @name: pathname component to lookup
1785 * @base: base directory to lookup from
1786 * @len: maximum length @len should be interpreted to
1788 * Note that this routine is purely a helper for filesystem usage and should
1789 * not be called by generic code. Also note that by using this function the
1790 * nameidata argument is passed to the filesystem methods and a filesystem
1791 * using this helper needs to be prepared for that.
1793 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1799 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1804 return ERR_PTR(-EACCES
);
1806 hash
= init_name_hash();
1808 c
= *(const unsigned char *)name
++;
1809 if (c
== '/' || c
== '\0')
1810 return ERR_PTR(-EACCES
);
1811 hash
= partial_name_hash(c
, hash
);
1813 this.hash
= end_name_hash(hash
);
1815 * See if the low-level filesystem might want
1816 * to use its own hash..
1818 if (base
->d_flags
& DCACHE_OP_HASH
) {
1819 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
1821 return ERR_PTR(err
);
1824 return __lookup_hash(&this, base
, NULL
);
1827 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1830 struct nameidata nd
;
1831 char *tmp
= getname_flags(name
, flags
);
1832 int err
= PTR_ERR(tmp
);
1835 BUG_ON(flags
& LOOKUP_PARENT
);
1837 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1845 static int user_path_parent(int dfd
, const char __user
*path
,
1846 struct nameidata
*nd
, char **name
)
1848 char *s
= getname(path
);
1854 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1864 * It's inline, so penalty for filesystems that don't use sticky bit is
1867 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1869 uid_t fsuid
= current_fsuid();
1871 if (!(dir
->i_mode
& S_ISVTX
))
1873 if (current_user_ns() != inode_userns(inode
))
1875 if (inode
->i_uid
== fsuid
)
1877 if (dir
->i_uid
== fsuid
)
1881 return !ns_capable(inode_userns(inode
), CAP_FOWNER
);
1885 * Check whether we can remove a link victim from directory dir, check
1886 * whether the type of victim is right.
1887 * 1. We can't do it if dir is read-only (done in permission())
1888 * 2. We should have write and exec permissions on dir
1889 * 3. We can't remove anything from append-only dir
1890 * 4. We can't do anything with immutable dir (done in permission())
1891 * 5. If the sticky bit on dir is set we should either
1892 * a. be owner of dir, or
1893 * b. be owner of victim, or
1894 * c. have CAP_FOWNER capability
1895 * 6. If the victim is append-only or immutable we can't do antyhing with
1896 * links pointing to it.
1897 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1898 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1899 * 9. We can't remove a root or mountpoint.
1900 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1901 * nfs_async_unlink().
1903 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1907 if (!victim
->d_inode
)
1910 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1911 audit_inode_child(victim
, dir
);
1913 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1918 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1919 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1922 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1924 if (IS_ROOT(victim
))
1926 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1928 if (IS_DEADDIR(dir
))
1930 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1935 /* Check whether we can create an object with dentry child in directory
1937 * 1. We can't do it if child already exists (open has special treatment for
1938 * this case, but since we are inlined it's OK)
1939 * 2. We can't do it if dir is read-only (done in permission())
1940 * 3. We should have write and exec permissions on dir
1941 * 4. We can't do it if dir is immutable (done in permission())
1943 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1947 if (IS_DEADDIR(dir
))
1949 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1953 * p1 and p2 should be directories on the same fs.
1955 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1960 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1964 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1966 p
= d_ancestor(p2
, p1
);
1968 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1969 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1973 p
= d_ancestor(p1
, p2
);
1975 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1976 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1980 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1981 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1985 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1987 mutex_unlock(&p1
->d_inode
->i_mutex
);
1989 mutex_unlock(&p2
->d_inode
->i_mutex
);
1990 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1994 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1995 struct nameidata
*nd
)
1997 int error
= may_create(dir
, dentry
);
2002 if (!dir
->i_op
->create
)
2003 return -EACCES
; /* shouldn't it be ENOSYS? */
2006 error
= security_inode_create(dir
, dentry
, mode
);
2009 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
2011 fsnotify_create(dir
, dentry
);
2015 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2017 struct dentry
*dentry
= path
->dentry
;
2018 struct inode
*inode
= dentry
->d_inode
;
2028 switch (inode
->i_mode
& S_IFMT
) {
2032 if (acc_mode
& MAY_WRITE
)
2037 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2046 error
= inode_permission(inode
, acc_mode
);
2051 * An append-only file must be opened in append mode for writing.
2053 if (IS_APPEND(inode
)) {
2054 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2060 /* O_NOATIME can only be set by the owner or superuser */
2061 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2065 * Ensure there are no outstanding leases on the file.
2067 return break_lease(inode
, flag
);
2070 static int handle_truncate(struct file
*filp
)
2072 struct path
*path
= &filp
->f_path
;
2073 struct inode
*inode
= path
->dentry
->d_inode
;
2074 int error
= get_write_access(inode
);
2078 * Refuse to truncate files with mandatory locks held on them.
2080 error
= locks_verify_locked(inode
);
2082 error
= security_path_truncate(path
);
2084 error
= do_truncate(path
->dentry
, 0,
2085 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2088 put_write_access(inode
);
2093 * Note that while the flag value (low two bits) for sys_open means:
2098 * it is changed into
2099 * 00 - no permissions needed
2100 * 01 - read-permission
2101 * 10 - write-permission
2103 * for the internal routines (ie open_namei()/follow_link() etc)
2104 * This is more logical, and also allows the 00 "no perm needed"
2105 * to be used for symlinks (where the permissions are checked
2109 static inline int open_to_namei_flags(int flag
)
2111 if ((flag
+1) & O_ACCMODE
)
2117 * Handle the last step of open()
2119 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2120 const struct open_flags
*op
, const char *pathname
)
2122 struct dentry
*dir
= nd
->path
.dentry
;
2123 struct dentry
*dentry
;
2124 int open_flag
= op
->open_flag
;
2125 int will_truncate
= open_flag
& O_TRUNC
;
2127 int acc_mode
= op
->acc_mode
;
2131 nd
->flags
&= ~LOOKUP_PARENT
;
2132 nd
->flags
|= op
->intent
;
2134 switch (nd
->last_type
) {
2137 error
= handle_dots(nd
, nd
->last_type
);
2139 return ERR_PTR(error
);
2142 if (nd
->flags
& LOOKUP_RCU
) {
2143 if (nameidata_drop_rcu_last(nd
))
2144 return ERR_PTR(-ECHILD
);
2146 error
= handle_reval_path(nd
);
2149 audit_inode(pathname
, nd
->path
.dentry
);
2150 if (open_flag
& O_CREAT
) {
2156 /* can't be RCU mode here */
2157 error
= handle_reval_path(nd
);
2160 audit_inode(pathname
, dir
);
2164 if (!(open_flag
& O_CREAT
)) {
2166 if (nd
->last
.name
[nd
->last
.len
])
2167 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2168 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2170 /* we _can_ be in RCU mode here */
2171 error
= walk_component(nd
, path
, &nd
->last
, LAST_NORM
,
2174 return ERR_PTR(error
);
2175 if (error
) /* symlink */
2178 if (nd
->flags
& LOOKUP_RCU
) {
2179 if (nameidata_drop_rcu_last(nd
))
2180 return ERR_PTR(-ECHILD
);
2184 if (nd
->flags
& LOOKUP_DIRECTORY
) {
2185 if (!nd
->inode
->i_op
->lookup
)
2188 audit_inode(pathname
, nd
->path
.dentry
);
2192 /* create side of things */
2194 if (nd
->flags
& LOOKUP_RCU
) {
2195 if (nameidata_drop_rcu_last(nd
))
2196 return ERR_PTR(-ECHILD
);
2199 audit_inode(pathname
, dir
);
2201 /* trailing slashes? */
2202 if (nd
->last
.name
[nd
->last
.len
])
2205 mutex_lock(&dir
->d_inode
->i_mutex
);
2207 dentry
= lookup_hash(nd
);
2208 error
= PTR_ERR(dentry
);
2209 if (IS_ERR(dentry
)) {
2210 mutex_unlock(&dir
->d_inode
->i_mutex
);
2214 path
->dentry
= dentry
;
2215 path
->mnt
= nd
->path
.mnt
;
2217 /* Negative dentry, just create the file */
2218 if (!dentry
->d_inode
) {
2219 int mode
= op
->mode
;
2220 if (!IS_POSIXACL(dir
->d_inode
))
2221 mode
&= ~current_umask();
2223 * This write is needed to ensure that a
2224 * rw->ro transition does not occur between
2225 * the time when the file is created and when
2226 * a permanent write count is taken through
2227 * the 'struct file' in nameidata_to_filp().
2229 error
= mnt_want_write(nd
->path
.mnt
);
2231 goto exit_mutex_unlock
;
2233 /* Don't check for write permission, don't truncate */
2234 open_flag
&= ~O_TRUNC
;
2236 acc_mode
= MAY_OPEN
;
2237 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2239 goto exit_mutex_unlock
;
2240 error
= vfs_create(dir
->d_inode
, dentry
, mode
, nd
);
2242 goto exit_mutex_unlock
;
2243 mutex_unlock(&dir
->d_inode
->i_mutex
);
2244 dput(nd
->path
.dentry
);
2245 nd
->path
.dentry
= dentry
;
2250 * It already exists.
2252 mutex_unlock(&dir
->d_inode
->i_mutex
);
2253 audit_inode(pathname
, path
->dentry
);
2256 if (open_flag
& O_EXCL
)
2259 error
= follow_managed(path
, nd
->flags
);
2264 if (!path
->dentry
->d_inode
)
2267 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2270 path_to_nameidata(path
, nd
);
2271 nd
->inode
= path
->dentry
->d_inode
;
2273 if (S_ISDIR(nd
->inode
->i_mode
))
2276 if (!S_ISREG(nd
->inode
->i_mode
))
2279 if (will_truncate
) {
2280 error
= mnt_want_write(nd
->path
.mnt
);
2286 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2289 filp
= nameidata_to_filp(nd
);
2290 if (!IS_ERR(filp
)) {
2291 error
= ima_file_check(filp
, op
->acc_mode
);
2294 filp
= ERR_PTR(error
);
2297 if (!IS_ERR(filp
)) {
2298 if (will_truncate
) {
2299 error
= handle_truncate(filp
);
2302 filp
= ERR_PTR(error
);
2308 mnt_drop_write(nd
->path
.mnt
);
2309 path_put(&nd
->path
);
2313 mutex_unlock(&dir
->d_inode
->i_mutex
);
2315 path_put_conditional(path
, nd
);
2317 filp
= ERR_PTR(error
);
2321 static struct file
*path_openat(int dfd
, const char *pathname
,
2322 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2324 struct file
*base
= NULL
;
2329 filp
= get_empty_filp();
2331 return ERR_PTR(-ENFILE
);
2333 filp
->f_flags
= op
->open_flag
;
2334 nd
->intent
.open
.file
= filp
;
2335 nd
->intent
.open
.flags
= open_to_namei_flags(op
->open_flag
);
2336 nd
->intent
.open
.create_mode
= op
->mode
;
2338 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2339 if (unlikely(error
))
2342 current
->total_link_count
= 0;
2343 error
= link_path_walk(pathname
, nd
);
2344 if (unlikely(error
))
2347 filp
= do_last(nd
, &path
, op
, pathname
);
2348 while (unlikely(!filp
)) { /* trailing symlink */
2349 struct path link
= path
;
2351 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2352 path_put_conditional(&path
, nd
);
2353 path_put(&nd
->path
);
2354 filp
= ERR_PTR(-ELOOP
);
2357 nd
->flags
|= LOOKUP_PARENT
;
2358 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2359 error
= follow_link(&link
, nd
, &cookie
);
2360 if (unlikely(error
))
2361 filp
= ERR_PTR(error
);
2363 filp
= do_last(nd
, &path
, op
, pathname
);
2364 put_link(nd
, &link
, cookie
);
2367 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2368 path_put(&nd
->root
);
2371 release_open_intent(nd
);
2375 filp
= ERR_PTR(error
);
2379 struct file
*do_filp_open(int dfd
, const char *pathname
,
2380 const struct open_flags
*op
, int flags
)
2382 struct nameidata nd
;
2385 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2386 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2387 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2388 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2389 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2393 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2394 const char *name
, const struct open_flags
*op
, int flags
)
2396 struct nameidata nd
;
2400 nd
.root
.dentry
= dentry
;
2402 flags
|= LOOKUP_ROOT
;
2404 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2405 return ERR_PTR(-ELOOP
);
2407 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2408 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2409 file
= path_openat(-1, name
, &nd
, op
, flags
);
2410 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2411 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2416 * lookup_create - lookup a dentry, creating it if it doesn't exist
2417 * @nd: nameidata info
2418 * @is_dir: directory flag
2420 * Simple function to lookup and return a dentry and create it
2421 * if it doesn't exist. Is SMP-safe.
2423 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2425 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2427 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2429 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2431 * Yucky last component or no last component at all?
2432 * (foo/., foo/.., /////)
2434 if (nd
->last_type
!= LAST_NORM
)
2436 nd
->flags
&= ~LOOKUP_PARENT
;
2437 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2438 nd
->intent
.open
.flags
= O_EXCL
;
2441 * Do the final lookup.
2443 dentry
= lookup_hash(nd
);
2447 if (dentry
->d_inode
)
2450 * Special case - lookup gave negative, but... we had foo/bar/
2451 * From the vfs_mknod() POV we just have a negative dentry -
2452 * all is fine. Let's be bastards - you had / on the end, you've
2453 * been asking for (non-existent) directory. -ENOENT for you.
2455 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2457 dentry
= ERR_PTR(-ENOENT
);
2462 dentry
= ERR_PTR(-EEXIST
);
2466 EXPORT_SYMBOL_GPL(lookup_create
);
2468 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2470 int error
= may_create(dir
, dentry
);
2475 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) &&
2476 !ns_capable(inode_userns(dir
), CAP_MKNOD
))
2479 if (!dir
->i_op
->mknod
)
2482 error
= devcgroup_inode_mknod(mode
, dev
);
2486 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2490 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2492 fsnotify_create(dir
, dentry
);
2496 static int may_mknod(mode_t mode
)
2498 switch (mode
& S_IFMT
) {
2504 case 0: /* zero mode translates to S_IFREG */
2513 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2518 struct dentry
*dentry
;
2519 struct nameidata nd
;
2524 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2528 dentry
= lookup_create(&nd
, 0);
2529 if (IS_ERR(dentry
)) {
2530 error
= PTR_ERR(dentry
);
2533 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2534 mode
&= ~current_umask();
2535 error
= may_mknod(mode
);
2538 error
= mnt_want_write(nd
.path
.mnt
);
2541 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2543 goto out_drop_write
;
2544 switch (mode
& S_IFMT
) {
2545 case 0: case S_IFREG
:
2546 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2548 case S_IFCHR
: case S_IFBLK
:
2549 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2550 new_decode_dev(dev
));
2552 case S_IFIFO
: case S_IFSOCK
:
2553 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2557 mnt_drop_write(nd
.path
.mnt
);
2561 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2568 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2570 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2573 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2575 int error
= may_create(dir
, dentry
);
2580 if (!dir
->i_op
->mkdir
)
2583 mode
&= (S_IRWXUGO
|S_ISVTX
);
2584 error
= security_inode_mkdir(dir
, dentry
, mode
);
2588 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2590 fsnotify_mkdir(dir
, dentry
);
2594 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2598 struct dentry
*dentry
;
2599 struct nameidata nd
;
2601 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2605 dentry
= lookup_create(&nd
, 1);
2606 error
= PTR_ERR(dentry
);
2610 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2611 mode
&= ~current_umask();
2612 error
= mnt_want_write(nd
.path
.mnt
);
2615 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2617 goto out_drop_write
;
2618 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2620 mnt_drop_write(nd
.path
.mnt
);
2624 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2631 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2633 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2637 * We try to drop the dentry early: we should have
2638 * a usage count of 2 if we're the only user of this
2639 * dentry, and if that is true (possibly after pruning
2640 * the dcache), then we drop the dentry now.
2642 * A low-level filesystem can, if it choses, legally
2645 * if (!d_unhashed(dentry))
2648 * if it cannot handle the case of removing a directory
2649 * that is still in use by something else..
2651 void dentry_unhash(struct dentry
*dentry
)
2654 shrink_dcache_parent(dentry
);
2655 spin_lock(&dentry
->d_lock
);
2656 if (dentry
->d_count
== 2)
2658 spin_unlock(&dentry
->d_lock
);
2661 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2663 int error
= may_delete(dir
, dentry
, 1);
2668 if (!dir
->i_op
->rmdir
)
2671 mutex_lock(&dentry
->d_inode
->i_mutex
);
2672 dentry_unhash(dentry
);
2673 if (d_mountpoint(dentry
))
2676 error
= security_inode_rmdir(dir
, dentry
);
2678 error
= dir
->i_op
->rmdir(dir
, dentry
);
2680 dentry
->d_inode
->i_flags
|= S_DEAD
;
2685 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2694 static long do_rmdir(int dfd
, const char __user
*pathname
)
2698 struct dentry
*dentry
;
2699 struct nameidata nd
;
2701 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2705 switch(nd
.last_type
) {
2717 nd
.flags
&= ~LOOKUP_PARENT
;
2719 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2720 dentry
= lookup_hash(&nd
);
2721 error
= PTR_ERR(dentry
);
2724 error
= mnt_want_write(nd
.path
.mnt
);
2727 error
= security_path_rmdir(&nd
.path
, dentry
);
2730 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2732 mnt_drop_write(nd
.path
.mnt
);
2736 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2743 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2745 return do_rmdir(AT_FDCWD
, pathname
);
2748 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2750 int error
= may_delete(dir
, dentry
, 0);
2755 if (!dir
->i_op
->unlink
)
2758 mutex_lock(&dentry
->d_inode
->i_mutex
);
2759 if (d_mountpoint(dentry
))
2762 error
= security_inode_unlink(dir
, dentry
);
2764 error
= dir
->i_op
->unlink(dir
, dentry
);
2769 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2771 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2772 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2773 fsnotify_link_count(dentry
->d_inode
);
2781 * Make sure that the actual truncation of the file will occur outside its
2782 * directory's i_mutex. Truncate can take a long time if there is a lot of
2783 * writeout happening, and we don't want to prevent access to the directory
2784 * while waiting on the I/O.
2786 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2790 struct dentry
*dentry
;
2791 struct nameidata nd
;
2792 struct inode
*inode
= NULL
;
2794 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2799 if (nd
.last_type
!= LAST_NORM
)
2802 nd
.flags
&= ~LOOKUP_PARENT
;
2804 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2805 dentry
= lookup_hash(&nd
);
2806 error
= PTR_ERR(dentry
);
2807 if (!IS_ERR(dentry
)) {
2808 /* Why not before? Because we want correct error value */
2809 if (nd
.last
.name
[nd
.last
.len
])
2811 inode
= dentry
->d_inode
;
2814 error
= mnt_want_write(nd
.path
.mnt
);
2817 error
= security_path_unlink(&nd
.path
, dentry
);
2820 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2822 mnt_drop_write(nd
.path
.mnt
);
2826 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2828 iput(inode
); /* truncate the inode here */
2835 error
= !dentry
->d_inode
? -ENOENT
:
2836 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2840 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2842 if ((flag
& ~AT_REMOVEDIR
) != 0)
2845 if (flag
& AT_REMOVEDIR
)
2846 return do_rmdir(dfd
, pathname
);
2848 return do_unlinkat(dfd
, pathname
);
2851 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2853 return do_unlinkat(AT_FDCWD
, pathname
);
2856 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2858 int error
= may_create(dir
, dentry
);
2863 if (!dir
->i_op
->symlink
)
2866 error
= security_inode_symlink(dir
, dentry
, oldname
);
2870 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2872 fsnotify_create(dir
, dentry
);
2876 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2877 int, newdfd
, const char __user
*, newname
)
2882 struct dentry
*dentry
;
2883 struct nameidata nd
;
2885 from
= getname(oldname
);
2887 return PTR_ERR(from
);
2889 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2893 dentry
= lookup_create(&nd
, 0);
2894 error
= PTR_ERR(dentry
);
2898 error
= mnt_want_write(nd
.path
.mnt
);
2901 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2903 goto out_drop_write
;
2904 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2906 mnt_drop_write(nd
.path
.mnt
);
2910 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2918 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2920 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2923 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2925 struct inode
*inode
= old_dentry
->d_inode
;
2931 error
= may_create(dir
, new_dentry
);
2935 if (dir
->i_sb
!= inode
->i_sb
)
2939 * A link to an append-only or immutable file cannot be created.
2941 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2943 if (!dir
->i_op
->link
)
2945 if (S_ISDIR(inode
->i_mode
))
2948 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2952 mutex_lock(&inode
->i_mutex
);
2953 /* Make sure we don't allow creating hardlink to an unlinked file */
2954 if (inode
->i_nlink
== 0)
2957 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2958 mutex_unlock(&inode
->i_mutex
);
2960 fsnotify_link(dir
, inode
, new_dentry
);
2965 * Hardlinks are often used in delicate situations. We avoid
2966 * security-related surprises by not following symlinks on the
2969 * We don't follow them on the oldname either to be compatible
2970 * with linux 2.0, and to avoid hard-linking to directories
2971 * and other special files. --ADM
2973 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2974 int, newdfd
, const char __user
*, newname
, int, flags
)
2976 struct dentry
*new_dentry
;
2977 struct nameidata nd
;
2978 struct path old_path
;
2983 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
2986 * To use null names we require CAP_DAC_READ_SEARCH
2987 * This ensures that not everyone will be able to create
2988 * handlink using the passed filedescriptor.
2990 if (flags
& AT_EMPTY_PATH
) {
2991 if (!capable(CAP_DAC_READ_SEARCH
))
2996 if (flags
& AT_SYMLINK_FOLLOW
)
2997 how
|= LOOKUP_FOLLOW
;
2999 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3003 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
3007 if (old_path
.mnt
!= nd
.path
.mnt
)
3009 new_dentry
= lookup_create(&nd
, 0);
3010 error
= PTR_ERR(new_dentry
);
3011 if (IS_ERR(new_dentry
))
3013 error
= mnt_want_write(nd
.path
.mnt
);
3016 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
3018 goto out_drop_write
;
3019 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
3021 mnt_drop_write(nd
.path
.mnt
);
3025 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3030 path_put(&old_path
);
3035 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3037 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3041 * The worst of all namespace operations - renaming directory. "Perverted"
3042 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3044 * a) we can get into loop creation. Check is done in is_subdir().
3045 * b) race potential - two innocent renames can create a loop together.
3046 * That's where 4.4 screws up. Current fix: serialization on
3047 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3049 * c) we have to lock _three_ objects - parents and victim (if it exists).
3050 * And that - after we got ->i_mutex on parents (until then we don't know
3051 * whether the target exists). Solution: try to be smart with locking
3052 * order for inodes. We rely on the fact that tree topology may change
3053 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3054 * move will be locked. Thus we can rank directories by the tree
3055 * (ancestors first) and rank all non-directories after them.
3056 * That works since everybody except rename does "lock parent, lookup,
3057 * lock child" and rename is under ->s_vfs_rename_mutex.
3058 * HOWEVER, it relies on the assumption that any object with ->lookup()
3059 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3060 * we'd better make sure that there's no link(2) for them.
3061 * d) some filesystems don't support opened-but-unlinked directories,
3062 * either because of layout or because they are not ready to deal with
3063 * all cases correctly. The latter will be fixed (taking this sort of
3064 * stuff into VFS), but the former is not going away. Solution: the same
3065 * trick as in rmdir().
3066 * e) conversion from fhandle to dentry may come in the wrong moment - when
3067 * we are removing the target. Solution: we will have to grab ->i_mutex
3068 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3069 * ->i_mutex on parents, which works but leads to some truly excessive
3072 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3073 struct inode
*new_dir
, struct dentry
*new_dentry
)
3076 struct inode
*target
;
3079 * If we are going to change the parent - check write permissions,
3080 * we'll need to flip '..'.
3082 if (new_dir
!= old_dir
) {
3083 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3088 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3092 target
= new_dentry
->d_inode
;
3094 mutex_lock(&target
->i_mutex
);
3095 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3099 dentry_unhash(new_dentry
);
3100 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3104 target
->i_flags
|= S_DEAD
;
3105 dont_mount(new_dentry
);
3107 mutex_unlock(&target
->i_mutex
);
3108 if (d_unhashed(new_dentry
))
3109 d_rehash(new_dentry
);
3113 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3114 d_move(old_dentry
,new_dentry
);
3118 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3119 struct inode
*new_dir
, struct dentry
*new_dentry
)
3121 struct inode
*target
;
3124 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3129 target
= new_dentry
->d_inode
;
3131 mutex_lock(&target
->i_mutex
);
3132 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3135 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3138 dont_mount(new_dentry
);
3139 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3140 d_move(old_dentry
, new_dentry
);
3143 mutex_unlock(&target
->i_mutex
);
3148 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3149 struct inode
*new_dir
, struct dentry
*new_dentry
)
3152 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3153 const unsigned char *old_name
;
3155 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3158 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3162 if (!new_dentry
->d_inode
)
3163 error
= may_create(new_dir
, new_dentry
);
3165 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3169 if (!old_dir
->i_op
->rename
)
3172 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3175 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3177 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3179 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3180 new_dentry
->d_inode
, old_dentry
);
3181 fsnotify_oldname_free(old_name
);
3186 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3187 int, newdfd
, const char __user
*, newname
)
3189 struct dentry
*old_dir
, *new_dir
;
3190 struct dentry
*old_dentry
, *new_dentry
;
3191 struct dentry
*trap
;
3192 struct nameidata oldnd
, newnd
;
3197 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3201 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3206 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3209 old_dir
= oldnd
.path
.dentry
;
3211 if (oldnd
.last_type
!= LAST_NORM
)
3214 new_dir
= newnd
.path
.dentry
;
3215 if (newnd
.last_type
!= LAST_NORM
)
3218 oldnd
.flags
&= ~LOOKUP_PARENT
;
3219 newnd
.flags
&= ~LOOKUP_PARENT
;
3220 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3222 trap
= lock_rename(new_dir
, old_dir
);
3224 old_dentry
= lookup_hash(&oldnd
);
3225 error
= PTR_ERR(old_dentry
);
3226 if (IS_ERR(old_dentry
))
3228 /* source must exist */
3230 if (!old_dentry
->d_inode
)
3232 /* unless the source is a directory trailing slashes give -ENOTDIR */
3233 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3235 if (oldnd
.last
.name
[oldnd
.last
.len
])
3237 if (newnd
.last
.name
[newnd
.last
.len
])
3240 /* source should not be ancestor of target */
3242 if (old_dentry
== trap
)
3244 new_dentry
= lookup_hash(&newnd
);
3245 error
= PTR_ERR(new_dentry
);
3246 if (IS_ERR(new_dentry
))
3248 /* target should not be an ancestor of source */
3250 if (new_dentry
== trap
)
3253 error
= mnt_want_write(oldnd
.path
.mnt
);
3256 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3257 &newnd
.path
, new_dentry
);
3260 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3261 new_dir
->d_inode
, new_dentry
);
3263 mnt_drop_write(oldnd
.path
.mnt
);
3269 unlock_rename(new_dir
, old_dir
);
3271 path_put(&newnd
.path
);
3274 path_put(&oldnd
.path
);
3280 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3282 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3285 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3289 len
= PTR_ERR(link
);
3294 if (len
> (unsigned) buflen
)
3296 if (copy_to_user(buffer
, link
, len
))
3303 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3304 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3305 * using) it for any given inode is up to filesystem.
3307 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3309 struct nameidata nd
;
3314 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3316 return PTR_ERR(cookie
);
3318 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3319 if (dentry
->d_inode
->i_op
->put_link
)
3320 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3324 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3326 return __vfs_follow_link(nd
, link
);
3329 /* get the link contents into pagecache */
3330 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3334 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3335 page
= read_mapping_page(mapping
, 0, NULL
);
3340 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3344 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3346 struct page
*page
= NULL
;
3347 char *s
= page_getlink(dentry
, &page
);
3348 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3351 page_cache_release(page
);
3356 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3358 struct page
*page
= NULL
;
3359 nd_set_link(nd
, page_getlink(dentry
, &page
));
3363 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3365 struct page
*page
= cookie
;
3369 page_cache_release(page
);
3374 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3376 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3378 struct address_space
*mapping
= inode
->i_mapping
;
3383 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3385 flags
|= AOP_FLAG_NOFS
;
3388 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3389 flags
, &page
, &fsdata
);
3393 kaddr
= kmap_atomic(page
, KM_USER0
);
3394 memcpy(kaddr
, symname
, len
-1);
3395 kunmap_atomic(kaddr
, KM_USER0
);
3397 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3404 mark_inode_dirty(inode
);
3410 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3412 return __page_symlink(inode
, symname
, len
,
3413 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3416 const struct inode_operations page_symlink_inode_operations
= {
3417 .readlink
= generic_readlink
,
3418 .follow_link
= page_follow_link_light
,
3419 .put_link
= page_put_link
,
3422 EXPORT_SYMBOL(user_path_at
);
3423 EXPORT_SYMBOL(follow_down_one
);
3424 EXPORT_SYMBOL(follow_down
);
3425 EXPORT_SYMBOL(follow_up
);
3426 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3427 EXPORT_SYMBOL(getname
);
3428 EXPORT_SYMBOL(lock_rename
);
3429 EXPORT_SYMBOL(lookup_one_len
);
3430 EXPORT_SYMBOL(page_follow_link_light
);
3431 EXPORT_SYMBOL(page_put_link
);
3432 EXPORT_SYMBOL(page_readlink
);
3433 EXPORT_SYMBOL(__page_symlink
);
3434 EXPORT_SYMBOL(page_symlink
);
3435 EXPORT_SYMBOL(page_symlink_inode_operations
);
3436 EXPORT_SYMBOL(kern_path_parent
);
3437 EXPORT_SYMBOL(kern_path
);
3438 EXPORT_SYMBOL(vfs_path_lookup
);
3439 EXPORT_SYMBOL(inode_permission
);
3440 EXPORT_SYMBOL(file_permission
);
3441 EXPORT_SYMBOL(unlock_rename
);
3442 EXPORT_SYMBOL(vfs_create
);
3443 EXPORT_SYMBOL(vfs_follow_link
);
3444 EXPORT_SYMBOL(vfs_link
);
3445 EXPORT_SYMBOL(vfs_mkdir
);
3446 EXPORT_SYMBOL(vfs_mknod
);
3447 EXPORT_SYMBOL(generic_permission
);
3448 EXPORT_SYMBOL(vfs_readlink
);
3449 EXPORT_SYMBOL(vfs_rename
);
3450 EXPORT_SYMBOL(vfs_rmdir
);
3451 EXPORT_SYMBOL(vfs_symlink
);
3452 EXPORT_SYMBOL(vfs_unlink
);
3453 EXPORT_SYMBOL(dentry_unhash
);
3454 EXPORT_SYMBOL(generic_readlink
);