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/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
121 void final_putname(struct filename
*name
)
123 if (name
->separate
) {
124 __putname(name
->name
);
131 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
133 static struct filename
*
134 getname_flags(const char __user
*filename
, int flags
, int *empty
)
136 struct filename
*result
, *err
;
141 result
= audit_reusename(filename
);
145 result
= __getname();
146 if (unlikely(!result
))
147 return ERR_PTR(-ENOMEM
);
150 * First, try to embed the struct filename inside the names_cache
153 kname
= (char *)result
+ sizeof(*result
);
154 result
->name
= kname
;
155 result
->separate
= false;
156 max
= EMBEDDED_NAME_MAX
;
159 len
= strncpy_from_user(kname
, filename
, max
);
160 if (unlikely(len
< 0)) {
166 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
167 * separate struct filename so we can dedicate the entire
168 * names_cache allocation for the pathname, and re-do the copy from
171 if (len
== EMBEDDED_NAME_MAX
&& max
== EMBEDDED_NAME_MAX
) {
172 kname
= (char *)result
;
174 result
= kzalloc(sizeof(*result
), GFP_KERNEL
);
176 err
= ERR_PTR(-ENOMEM
);
177 result
= (struct filename
*)kname
;
180 result
->name
= kname
;
181 result
->separate
= true;
186 /* The empty path is special. */
187 if (unlikely(!len
)) {
190 err
= ERR_PTR(-ENOENT
);
191 if (!(flags
& LOOKUP_EMPTY
))
195 err
= ERR_PTR(-ENAMETOOLONG
);
196 if (unlikely(len
>= PATH_MAX
))
199 result
->uptr
= filename
;
200 audit_getname(result
);
204 final_putname(result
);
209 getname(const char __user
* filename
)
211 return getname_flags(filename
, 0, NULL
);
213 EXPORT_SYMBOL(getname
);
215 #ifdef CONFIG_AUDITSYSCALL
216 void putname(struct filename
*name
)
218 if (unlikely(!audit_dummy_context()))
219 return audit_putname(name
);
224 static int check_acl(struct inode
*inode
, int mask
)
226 #ifdef CONFIG_FS_POSIX_ACL
227 struct posix_acl
*acl
;
229 if (mask
& MAY_NOT_BLOCK
) {
230 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
233 /* no ->get_acl() calls in RCU mode... */
234 if (acl
== ACL_NOT_CACHED
)
236 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
239 acl
= get_cached_acl(inode
, ACL_TYPE_ACCESS
);
242 * A filesystem can force a ACL callback by just never filling the
243 * ACL cache. But normally you'd fill the cache either at inode
244 * instantiation time, or on the first ->get_acl call.
246 * If the filesystem doesn't have a get_acl() function at all, we'll
247 * just create the negative cache entry.
249 if (acl
== ACL_NOT_CACHED
) {
250 if (inode
->i_op
->get_acl
) {
251 acl
= inode
->i_op
->get_acl(inode
, ACL_TYPE_ACCESS
);
255 set_cached_acl(inode
, ACL_TYPE_ACCESS
, NULL
);
261 int error
= posix_acl_permission(inode
, acl
, mask
);
262 posix_acl_release(acl
);
271 * This does the basic permission checking
273 static int acl_permission_check(struct inode
*inode
, int mask
)
275 unsigned int mode
= inode
->i_mode
;
277 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
280 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
281 int error
= check_acl(inode
, mask
);
282 if (error
!= -EAGAIN
)
286 if (in_group_p(inode
->i_gid
))
291 * If the DACs are ok we don't need any capability check.
293 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
299 * generic_permission - check for access rights on a Posix-like filesystem
300 * @inode: inode to check access rights for
301 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
303 * Used to check for read/write/execute permissions on a file.
304 * We use "fsuid" for this, letting us set arbitrary permissions
305 * for filesystem access without changing the "normal" uids which
306 * are used for other things.
308 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
309 * request cannot be satisfied (eg. requires blocking or too much complexity).
310 * It would then be called again in ref-walk mode.
312 int generic_permission(struct inode
*inode
, int mask
)
317 * Do the basic permission checks.
319 ret
= acl_permission_check(inode
, mask
);
323 if (S_ISDIR(inode
->i_mode
)) {
324 /* DACs are overridable for directories */
325 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
327 if (!(mask
& MAY_WRITE
))
328 if (capable_wrt_inode_uidgid(inode
,
329 CAP_DAC_READ_SEARCH
))
334 * Read/write DACs are always overridable.
335 * Executable DACs are overridable when there is
336 * at least one exec bit set.
338 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
339 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
343 * Searching includes executable on directories, else just read.
345 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
346 if (mask
== MAY_READ
)
347 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_READ_SEARCH
))
354 * We _really_ want to just do "generic_permission()" without
355 * even looking at the inode->i_op values. So we keep a cache
356 * flag in inode->i_opflags, that says "this has not special
357 * permission function, use the fast case".
359 static inline int do_inode_permission(struct inode
*inode
, int mask
)
361 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
362 if (likely(inode
->i_op
->permission
))
363 return inode
->i_op
->permission(inode
, mask
);
365 /* This gets set once for the inode lifetime */
366 spin_lock(&inode
->i_lock
);
367 inode
->i_opflags
|= IOP_FASTPERM
;
368 spin_unlock(&inode
->i_lock
);
370 return generic_permission(inode
, mask
);
374 * __inode_permission - Check for access rights to a given inode
375 * @inode: Inode to check permission on
376 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
378 * Check for read/write/execute permissions on an inode.
380 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
382 * This does not check for a read-only file system. You probably want
383 * inode_permission().
385 int __inode_permission(struct inode
*inode
, int mask
)
389 if (unlikely(mask
& MAY_WRITE
)) {
391 * Nobody gets write access to an immutable file.
393 if (IS_IMMUTABLE(inode
))
397 retval
= do_inode_permission(inode
, mask
);
401 retval
= devcgroup_inode_permission(inode
, mask
);
405 return security_inode_permission(inode
, mask
);
409 * sb_permission - Check superblock-level permissions
410 * @sb: Superblock of inode to check permission on
411 * @inode: Inode to check permission on
412 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
414 * Separate out file-system wide checks from inode-specific permission checks.
416 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
418 if (unlikely(mask
& MAY_WRITE
)) {
419 umode_t mode
= inode
->i_mode
;
421 /* Nobody gets write access to a read-only fs. */
422 if ((sb
->s_flags
& MS_RDONLY
) &&
423 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
430 * inode_permission - Check for access rights to a given inode
431 * @inode: Inode to check permission on
432 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
434 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
435 * this, letting us set arbitrary permissions for filesystem access without
436 * changing the "normal" UIDs which are used for other things.
438 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
440 int inode_permission(struct inode
*inode
, int mask
)
444 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
447 return __inode_permission(inode
, mask
);
451 * path_get - get a reference to a path
452 * @path: path to get the reference to
454 * Given a path increment the reference count to the dentry and the vfsmount.
456 void path_get(const struct path
*path
)
461 EXPORT_SYMBOL(path_get
);
464 * path_put - put a reference to a path
465 * @path: path to put the reference to
467 * Given a path decrement the reference count to the dentry and the vfsmount.
469 void path_put(const struct path
*path
)
474 EXPORT_SYMBOL(path_put
);
477 * Path walking has 2 modes, rcu-walk and ref-walk (see
478 * Documentation/filesystems/path-lookup.txt). In situations when we can't
479 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
480 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
481 * mode. Refcounts are grabbed at the last known good point before rcu-walk
482 * got stuck, so ref-walk may continue from there. If this is not successful
483 * (eg. a seqcount has changed), then failure is returned and it's up to caller
484 * to restart the path walk from the beginning in ref-walk mode.
487 static inline void lock_rcu_walk(void)
489 br_read_lock(&vfsmount_lock
);
493 static inline void unlock_rcu_walk(void)
496 br_read_unlock(&vfsmount_lock
);
500 * unlazy_walk - try to switch to ref-walk mode.
501 * @nd: nameidata pathwalk data
502 * @dentry: child of nd->path.dentry or NULL
503 * Returns: 0 on success, -ECHILD on failure
505 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
506 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
507 * @nd or NULL. Must be called from rcu-walk context.
509 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
511 struct fs_struct
*fs
= current
->fs
;
512 struct dentry
*parent
= nd
->path
.dentry
;
515 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
516 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
518 spin_lock(&fs
->lock
);
519 if (nd
->root
.mnt
!= fs
->root
.mnt
||
520 nd
->root
.dentry
!= fs
->root
.dentry
)
523 spin_lock(&parent
->d_lock
);
525 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
527 BUG_ON(nd
->inode
!= parent
->d_inode
);
529 if (dentry
->d_parent
!= parent
)
531 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
532 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
535 * If the sequence check on the child dentry passed, then
536 * the child has not been removed from its parent. This
537 * means the parent dentry must be valid and able to take
538 * a reference at this point.
540 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
541 BUG_ON(!parent
->d_count
);
543 spin_unlock(&dentry
->d_lock
);
545 spin_unlock(&parent
->d_lock
);
548 spin_unlock(&fs
->lock
);
550 mntget(nd
->path
.mnt
);
553 nd
->flags
&= ~LOOKUP_RCU
;
557 spin_unlock(&dentry
->d_lock
);
559 spin_unlock(&parent
->d_lock
);
562 spin_unlock(&fs
->lock
);
566 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
568 return dentry
->d_op
->d_revalidate(dentry
, flags
);
572 * complete_walk - successful completion of path walk
573 * @nd: pointer nameidata
575 * If we had been in RCU mode, drop out of it and legitimize nd->path.
576 * Revalidate the final result, unless we'd already done that during
577 * the path walk or the filesystem doesn't ask for it. Return 0 on
578 * success, -error on failure. In case of failure caller does not
579 * need to drop nd->path.
581 static int complete_walk(struct nameidata
*nd
)
583 struct dentry
*dentry
= nd
->path
.dentry
;
586 if (nd
->flags
& LOOKUP_RCU
) {
587 nd
->flags
&= ~LOOKUP_RCU
;
588 if (!(nd
->flags
& LOOKUP_ROOT
))
590 spin_lock(&dentry
->d_lock
);
591 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
592 spin_unlock(&dentry
->d_lock
);
596 BUG_ON(nd
->inode
!= dentry
->d_inode
);
597 spin_unlock(&dentry
->d_lock
);
598 mntget(nd
->path
.mnt
);
602 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
605 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
608 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
619 static __always_inline
void set_root(struct nameidata
*nd
)
622 get_fs_root(current
->fs
, &nd
->root
);
625 static int link_path_walk(const char *, struct nameidata
*);
627 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
630 struct fs_struct
*fs
= current
->fs
;
634 seq
= read_seqcount_begin(&fs
->seq
);
636 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
637 } while (read_seqcount_retry(&fs
->seq
, seq
));
641 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
653 nd
->flags
|= LOOKUP_JUMPED
;
655 nd
->inode
= nd
->path
.dentry
->d_inode
;
657 ret
= link_path_walk(link
, nd
);
661 return PTR_ERR(link
);
664 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
667 if (path
->mnt
!= nd
->path
.mnt
)
671 static inline void path_to_nameidata(const struct path
*path
,
672 struct nameidata
*nd
)
674 if (!(nd
->flags
& LOOKUP_RCU
)) {
675 dput(nd
->path
.dentry
);
676 if (nd
->path
.mnt
!= path
->mnt
)
677 mntput(nd
->path
.mnt
);
679 nd
->path
.mnt
= path
->mnt
;
680 nd
->path
.dentry
= path
->dentry
;
684 * Helper to directly jump to a known parsed path from ->follow_link,
685 * caller must have taken a reference to path beforehand.
687 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
692 nd
->inode
= nd
->path
.dentry
->d_inode
;
693 nd
->flags
|= LOOKUP_JUMPED
;
696 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
698 struct inode
*inode
= link
->dentry
->d_inode
;
699 if (inode
->i_op
->put_link
)
700 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
704 int sysctl_protected_symlinks __read_mostly
= 0;
705 int sysctl_protected_hardlinks __read_mostly
= 0;
708 * may_follow_link - Check symlink following for unsafe situations
709 * @link: The path of the symlink
710 * @nd: nameidata pathwalk data
712 * In the case of the sysctl_protected_symlinks sysctl being enabled,
713 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
714 * in a sticky world-writable directory. This is to protect privileged
715 * processes from failing races against path names that may change out
716 * from under them by way of other users creating malicious symlinks.
717 * It will permit symlinks to be followed only when outside a sticky
718 * world-writable directory, or when the uid of the symlink and follower
719 * match, or when the directory owner matches the symlink's owner.
721 * Returns 0 if following the symlink is allowed, -ve on error.
723 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
725 const struct inode
*inode
;
726 const struct inode
*parent
;
728 if (!sysctl_protected_symlinks
)
731 /* Allowed if owner and follower match. */
732 inode
= link
->dentry
->d_inode
;
733 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
736 /* Allowed if parent directory not sticky and world-writable. */
737 parent
= nd
->path
.dentry
->d_inode
;
738 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
741 /* Allowed if parent directory and link owner match. */
742 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
745 audit_log_link_denied("follow_link", link
);
746 path_put_conditional(link
, nd
);
752 * safe_hardlink_source - Check for safe hardlink conditions
753 * @inode: the source inode to hardlink from
755 * Return false if at least one of the following conditions:
756 * - inode is not a regular file
758 * - inode is setgid and group-exec
759 * - access failure for read and write
761 * Otherwise returns true.
763 static bool safe_hardlink_source(struct inode
*inode
)
765 umode_t mode
= inode
->i_mode
;
767 /* Special files should not get pinned to the filesystem. */
771 /* Setuid files should not get pinned to the filesystem. */
775 /* Executable setgid files should not get pinned to the filesystem. */
776 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
779 /* Hardlinking to unreadable or unwritable sources is dangerous. */
780 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
787 * may_linkat - Check permissions for creating a hardlink
788 * @link: the source to hardlink from
790 * Block hardlink when all of:
791 * - sysctl_protected_hardlinks enabled
792 * - fsuid does not match inode
793 * - hardlink source is unsafe (see safe_hardlink_source() above)
796 * Returns 0 if successful, -ve on error.
798 static int may_linkat(struct path
*link
)
800 const struct cred
*cred
;
803 if (!sysctl_protected_hardlinks
)
806 cred
= current_cred();
807 inode
= link
->dentry
->d_inode
;
809 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
810 * otherwise, it must be a safe source.
812 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
816 audit_log_link_denied("linkat", link
);
820 static __always_inline
int
821 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
823 struct dentry
*dentry
= link
->dentry
;
827 BUG_ON(nd
->flags
& LOOKUP_RCU
);
829 if (link
->mnt
== nd
->path
.mnt
)
833 if (unlikely(current
->total_link_count
>= 40))
834 goto out_put_nd_path
;
837 current
->total_link_count
++;
840 nd_set_link(nd
, NULL
);
842 error
= security_inode_follow_link(link
->dentry
, nd
);
844 goto out_put_nd_path
;
846 nd
->last_type
= LAST_BIND
;
847 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
850 goto out_put_nd_path
;
855 error
= __vfs_follow_link(nd
, s
);
857 put_link(nd
, link
, *p
);
869 static int follow_up_rcu(struct path
*path
)
871 struct mount
*mnt
= real_mount(path
->mnt
);
872 struct mount
*parent
;
873 struct dentry
*mountpoint
;
875 parent
= mnt
->mnt_parent
;
876 if (&parent
->mnt
== path
->mnt
)
878 mountpoint
= mnt
->mnt_mountpoint
;
879 path
->dentry
= mountpoint
;
880 path
->mnt
= &parent
->mnt
;
885 * follow_up - Find the mountpoint of path's vfsmount
887 * Given a path, find the mountpoint of its source file system.
888 * Replace @path with the path of the mountpoint in the parent mount.
891 * Return 1 if we went up a level and 0 if we were already at the
894 int follow_up(struct path
*path
)
896 struct mount
*mnt
= real_mount(path
->mnt
);
897 struct mount
*parent
;
898 struct dentry
*mountpoint
;
900 br_read_lock(&vfsmount_lock
);
901 parent
= mnt
->mnt_parent
;
903 br_read_unlock(&vfsmount_lock
);
906 mntget(&parent
->mnt
);
907 mountpoint
= dget(mnt
->mnt_mountpoint
);
908 br_read_unlock(&vfsmount_lock
);
910 path
->dentry
= mountpoint
;
912 path
->mnt
= &parent
->mnt
;
917 * Perform an automount
918 * - return -EISDIR to tell follow_managed() to stop and return the path we
921 static int follow_automount(struct path
*path
, unsigned flags
,
924 struct vfsmount
*mnt
;
927 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
930 /* We don't want to mount if someone's just doing a stat -
931 * unless they're stat'ing a directory and appended a '/' to
934 * We do, however, want to mount if someone wants to open or
935 * create a file of any type under the mountpoint, wants to
936 * traverse through the mountpoint or wants to open the
937 * mounted directory. Also, autofs may mark negative dentries
938 * as being automount points. These will need the attentions
939 * of the daemon to instantiate them before they can be used.
941 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
942 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
943 path
->dentry
->d_inode
)
946 current
->total_link_count
++;
947 if (current
->total_link_count
>= 40)
950 mnt
= path
->dentry
->d_op
->d_automount(path
);
953 * The filesystem is allowed to return -EISDIR here to indicate
954 * it doesn't want to automount. For instance, autofs would do
955 * this so that its userspace daemon can mount on this dentry.
957 * However, we can only permit this if it's a terminal point in
958 * the path being looked up; if it wasn't then the remainder of
959 * the path is inaccessible and we should say so.
961 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
966 if (!mnt
) /* mount collision */
970 /* lock_mount() may release path->mnt on error */
974 err
= finish_automount(mnt
, path
);
978 /* Someone else made a mount here whilst we were busy */
983 path
->dentry
= dget(mnt
->mnt_root
);
992 * Handle a dentry that is managed in some way.
993 * - Flagged for transit management (autofs)
994 * - Flagged as mountpoint
995 * - Flagged as automount point
997 * This may only be called in refwalk mode.
999 * Serialization is taken care of in namespace.c
1001 static int follow_managed(struct path
*path
, unsigned flags
)
1003 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1005 bool need_mntput
= false;
1008 /* Given that we're not holding a lock here, we retain the value in a
1009 * local variable for each dentry as we look at it so that we don't see
1010 * the components of that value change under us */
1011 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1012 managed
&= DCACHE_MANAGED_DENTRY
,
1013 unlikely(managed
!= 0)) {
1014 /* Allow the filesystem to manage the transit without i_mutex
1016 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1017 BUG_ON(!path
->dentry
->d_op
);
1018 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1019 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1024 /* Transit to a mounted filesystem. */
1025 if (managed
& DCACHE_MOUNTED
) {
1026 struct vfsmount
*mounted
= lookup_mnt(path
);
1031 path
->mnt
= mounted
;
1032 path
->dentry
= dget(mounted
->mnt_root
);
1037 /* Something is mounted on this dentry in another
1038 * namespace and/or whatever was mounted there in this
1039 * namespace got unmounted before we managed to get the
1043 /* Handle an automount point */
1044 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1045 ret
= follow_automount(path
, flags
, &need_mntput
);
1051 /* We didn't change the current path point */
1055 if (need_mntput
&& path
->mnt
== mnt
)
1059 return ret
< 0 ? ret
: need_mntput
;
1062 int follow_down_one(struct path
*path
)
1064 struct vfsmount
*mounted
;
1066 mounted
= lookup_mnt(path
);
1070 path
->mnt
= mounted
;
1071 path
->dentry
= dget(mounted
->mnt_root
);
1077 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
1079 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
1080 dentry
->d_op
->d_manage(dentry
, true) < 0);
1084 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1085 * we meet a managed dentry that would need blocking.
1087 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1088 struct inode
**inode
)
1091 struct mount
*mounted
;
1093 * Don't forget we might have a non-mountpoint managed dentry
1094 * that wants to block transit.
1096 if (unlikely(managed_dentry_might_block(path
->dentry
)))
1099 if (!d_mountpoint(path
->dentry
))
1102 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1105 path
->mnt
= &mounted
->mnt
;
1106 path
->dentry
= mounted
->mnt
.mnt_root
;
1107 nd
->flags
|= LOOKUP_JUMPED
;
1108 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1110 * Update the inode too. We don't need to re-check the
1111 * dentry sequence number here after this d_inode read,
1112 * because a mount-point is always pinned.
1114 *inode
= path
->dentry
->d_inode
;
1119 static void follow_mount_rcu(struct nameidata
*nd
)
1121 while (d_mountpoint(nd
->path
.dentry
)) {
1122 struct mount
*mounted
;
1123 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
1126 nd
->path
.mnt
= &mounted
->mnt
;
1127 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1128 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1132 static int follow_dotdot_rcu(struct nameidata
*nd
)
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 struct dentry
*old
= nd
->path
.dentry
;
1143 struct dentry
*parent
= old
->d_parent
;
1146 seq
= read_seqcount_begin(&parent
->d_seq
);
1147 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1149 nd
->path
.dentry
= parent
;
1153 if (!follow_up_rcu(&nd
->path
))
1155 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1157 follow_mount_rcu(nd
);
1158 nd
->inode
= nd
->path
.dentry
->d_inode
;
1162 nd
->flags
&= ~LOOKUP_RCU
;
1163 if (!(nd
->flags
& LOOKUP_ROOT
))
1164 nd
->root
.mnt
= NULL
;
1170 * Follow down to the covering mount currently visible to userspace. At each
1171 * point, the filesystem owning that dentry may be queried as to whether the
1172 * caller is permitted to proceed or not.
1174 int follow_down(struct path
*path
)
1179 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1180 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1181 /* Allow the filesystem to manage the transit without i_mutex
1184 * We indicate to the filesystem if someone is trying to mount
1185 * something here. This gives autofs the chance to deny anyone
1186 * other than its daemon the right to mount on its
1189 * The filesystem may sleep at this point.
1191 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1192 BUG_ON(!path
->dentry
->d_op
);
1193 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1194 ret
= path
->dentry
->d_op
->d_manage(
1195 path
->dentry
, false);
1197 return ret
== -EISDIR
? 0 : ret
;
1200 /* Transit to a mounted filesystem. */
1201 if (managed
& DCACHE_MOUNTED
) {
1202 struct vfsmount
*mounted
= lookup_mnt(path
);
1207 path
->mnt
= mounted
;
1208 path
->dentry
= dget(mounted
->mnt_root
);
1212 /* Don't handle automount points here */
1219 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1221 static void follow_mount(struct path
*path
)
1223 while (d_mountpoint(path
->dentry
)) {
1224 struct vfsmount
*mounted
= lookup_mnt(path
);
1229 path
->mnt
= mounted
;
1230 path
->dentry
= dget(mounted
->mnt_root
);
1234 static void follow_dotdot(struct nameidata
*nd
)
1239 struct dentry
*old
= nd
->path
.dentry
;
1241 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1242 nd
->path
.mnt
== nd
->root
.mnt
) {
1245 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1246 /* rare case of legitimate dget_parent()... */
1247 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1251 if (!follow_up(&nd
->path
))
1254 follow_mount(&nd
->path
);
1255 nd
->inode
= nd
->path
.dentry
->d_inode
;
1259 * This looks up the name in dcache, possibly revalidates the old dentry and
1260 * allocates a new one if not found or not valid. In the need_lookup argument
1261 * returns whether i_op->lookup is necessary.
1263 * dir->d_inode->i_mutex must be held
1265 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1266 unsigned int flags
, bool *need_lookup
)
1268 struct dentry
*dentry
;
1271 *need_lookup
= false;
1272 dentry
= d_lookup(dir
, name
);
1274 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1275 error
= d_revalidate(dentry
, flags
);
1276 if (unlikely(error
<= 0)) {
1279 return ERR_PTR(error
);
1280 } else if (!d_invalidate(dentry
)) {
1289 dentry
= d_alloc(dir
, name
);
1290 if (unlikely(!dentry
))
1291 return ERR_PTR(-ENOMEM
);
1293 *need_lookup
= true;
1299 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1300 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1302 * dir->d_inode->i_mutex must be held
1304 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1309 /* Don't create child dentry for a dead directory. */
1310 if (unlikely(IS_DEADDIR(dir
))) {
1312 return ERR_PTR(-ENOENT
);
1315 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1316 if (unlikely(old
)) {
1323 static struct dentry
*__lookup_hash(struct qstr
*name
,
1324 struct dentry
*base
, unsigned int flags
)
1327 struct dentry
*dentry
;
1329 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1333 return lookup_real(base
->d_inode
, dentry
, flags
);
1337 * It's more convoluted than I'd like it to be, but... it's still fairly
1338 * small and for now I'd prefer to have fast path as straight as possible.
1339 * It _is_ time-critical.
1341 static int lookup_fast(struct nameidata
*nd
,
1342 struct path
*path
, struct inode
**inode
)
1344 struct vfsmount
*mnt
= nd
->path
.mnt
;
1345 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1351 * Rename seqlock is not required here because in the off chance
1352 * of a false negative due to a concurrent rename, we're going to
1353 * do the non-racy lookup, below.
1355 if (nd
->flags
& LOOKUP_RCU
) {
1357 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
, nd
->inode
);
1362 * This sequence count validates that the inode matches
1363 * the dentry name information from lookup.
1365 *inode
= dentry
->d_inode
;
1366 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1370 * This sequence count validates that the parent had no
1371 * changes while we did the lookup of the dentry above.
1373 * The memory barrier in read_seqcount_begin of child is
1374 * enough, we can use __read_seqcount_retry here.
1376 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1380 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1381 status
= d_revalidate(dentry
, nd
->flags
);
1382 if (unlikely(status
<= 0)) {
1383 if (status
!= -ECHILD
)
1389 path
->dentry
= dentry
;
1390 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1392 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1396 if (unlazy_walk(nd
, dentry
))
1399 dentry
= __d_lookup(parent
, &nd
->last
);
1402 if (unlikely(!dentry
))
1405 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1406 status
= d_revalidate(dentry
, nd
->flags
);
1407 if (unlikely(status
<= 0)) {
1412 if (!d_invalidate(dentry
)) {
1419 path
->dentry
= dentry
;
1420 err
= follow_managed(path
, nd
->flags
);
1421 if (unlikely(err
< 0)) {
1422 path_put_conditional(path
, nd
);
1426 nd
->flags
|= LOOKUP_JUMPED
;
1427 *inode
= path
->dentry
->d_inode
;
1434 /* Fast lookup failed, do it the slow way */
1435 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1437 struct dentry
*dentry
, *parent
;
1440 parent
= nd
->path
.dentry
;
1441 BUG_ON(nd
->inode
!= parent
->d_inode
);
1443 mutex_lock(&parent
->d_inode
->i_mutex
);
1444 dentry
= __lookup_hash(&nd
->last
, parent
, nd
->flags
);
1445 mutex_unlock(&parent
->d_inode
->i_mutex
);
1447 return PTR_ERR(dentry
);
1448 path
->mnt
= nd
->path
.mnt
;
1449 path
->dentry
= dentry
;
1450 err
= follow_managed(path
, nd
->flags
);
1451 if (unlikely(err
< 0)) {
1452 path_put_conditional(path
, nd
);
1456 nd
->flags
|= LOOKUP_JUMPED
;
1460 static inline int may_lookup(struct nameidata
*nd
)
1462 if (nd
->flags
& LOOKUP_RCU
) {
1463 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1466 if (unlazy_walk(nd
, NULL
))
1469 return inode_permission(nd
->inode
, MAY_EXEC
);
1472 static inline int handle_dots(struct nameidata
*nd
, int type
)
1474 if (type
== LAST_DOTDOT
) {
1475 if (nd
->flags
& LOOKUP_RCU
) {
1476 if (follow_dotdot_rcu(nd
))
1484 static void terminate_walk(struct nameidata
*nd
)
1486 if (!(nd
->flags
& LOOKUP_RCU
)) {
1487 path_put(&nd
->path
);
1489 nd
->flags
&= ~LOOKUP_RCU
;
1490 if (!(nd
->flags
& LOOKUP_ROOT
))
1491 nd
->root
.mnt
= NULL
;
1497 * Do we need to follow links? We _really_ want to be able
1498 * to do this check without having to look at inode->i_op,
1499 * so we keep a cache of "no, this doesn't need follow_link"
1500 * for the common case.
1502 static inline int should_follow_link(struct inode
*inode
, int follow
)
1504 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1505 if (likely(inode
->i_op
->follow_link
))
1508 /* This gets set once for the inode lifetime */
1509 spin_lock(&inode
->i_lock
);
1510 inode
->i_opflags
|= IOP_NOFOLLOW
;
1511 spin_unlock(&inode
->i_lock
);
1516 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1519 struct inode
*inode
;
1522 * "." and ".." are special - ".." especially so because it has
1523 * to be able to know about the current root directory and
1524 * parent relationships.
1526 if (unlikely(nd
->last_type
!= LAST_NORM
))
1527 return handle_dots(nd
, nd
->last_type
);
1528 err
= lookup_fast(nd
, path
, &inode
);
1529 if (unlikely(err
)) {
1533 err
= lookup_slow(nd
, path
);
1537 inode
= path
->dentry
->d_inode
;
1543 if (should_follow_link(inode
, follow
)) {
1544 if (nd
->flags
& LOOKUP_RCU
) {
1545 if (unlikely(nd
->path
.mnt
!= path
->mnt
||
1546 unlazy_walk(nd
, path
->dentry
))) {
1551 BUG_ON(inode
!= path
->dentry
->d_inode
);
1554 path_to_nameidata(path
, nd
);
1559 path_to_nameidata(path
, nd
);
1566 * This limits recursive symlink follows to 8, while
1567 * limiting consecutive symlinks to 40.
1569 * Without that kind of total limit, nasty chains of consecutive
1570 * symlinks can cause almost arbitrarily long lookups.
1572 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1576 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1577 path_put_conditional(path
, nd
);
1578 path_put(&nd
->path
);
1581 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1584 current
->link_count
++;
1587 struct path link
= *path
;
1590 res
= follow_link(&link
, nd
, &cookie
);
1593 res
= walk_component(nd
, path
, LOOKUP_FOLLOW
);
1594 put_link(nd
, &link
, cookie
);
1597 current
->link_count
--;
1603 * We really don't want to look at inode->i_op->lookup
1604 * when we don't have to. So we keep a cache bit in
1605 * the inode ->i_opflags field that says "yes, we can
1606 * do lookup on this inode".
1608 static inline int can_lookup(struct inode
*inode
)
1610 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1612 if (likely(!inode
->i_op
->lookup
))
1615 /* We do this once for the lifetime of the inode */
1616 spin_lock(&inode
->i_lock
);
1617 inode
->i_opflags
|= IOP_LOOKUP
;
1618 spin_unlock(&inode
->i_lock
);
1623 * We can do the critical dentry name comparison and hashing
1624 * operations one word at a time, but we are limited to:
1626 * - Architectures with fast unaligned word accesses. We could
1627 * do a "get_unaligned()" if this helps and is sufficiently
1630 * - Little-endian machines (so that we can generate the mask
1631 * of low bytes efficiently). Again, we *could* do a byte
1632 * swapping load on big-endian architectures if that is not
1633 * expensive enough to make the optimization worthless.
1635 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1636 * do not trap on the (extremely unlikely) case of a page
1637 * crossing operation.
1639 * - Furthermore, we need an efficient 64-bit compile for the
1640 * 64-bit case in order to generate the "number of bytes in
1641 * the final mask". Again, that could be replaced with a
1642 * efficient population count instruction or similar.
1644 #ifdef CONFIG_DCACHE_WORD_ACCESS
1646 #include <asm/word-at-a-time.h>
1650 static inline unsigned int fold_hash(unsigned long hash
)
1652 return hash_64(hash
, 32);
1655 #else /* 32-bit case */
1657 #define fold_hash(x) (x)
1661 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1663 unsigned long a
, mask
;
1664 unsigned long hash
= 0;
1667 a
= load_unaligned_zeropad(name
);
1668 if (len
< sizeof(unsigned long))
1672 name
+= sizeof(unsigned long);
1673 len
-= sizeof(unsigned long);
1677 mask
= ~(~0ul << len
*8);
1680 return fold_hash(hash
);
1682 EXPORT_SYMBOL(full_name_hash
);
1685 * Calculate the length and hash of the path component, and
1686 * return the length of the component;
1688 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1690 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1691 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1694 len
= -sizeof(unsigned long);
1696 hash
= (hash
+ a
) * 9;
1697 len
+= sizeof(unsigned long);
1698 a
= load_unaligned_zeropad(name
+len
);
1699 b
= a
^ REPEAT_BYTE('/');
1700 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1702 adata
= prep_zero_mask(a
, adata
, &constants
);
1703 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1705 mask
= create_zero_mask(adata
| bdata
);
1707 hash
+= a
& zero_bytemask(mask
);
1708 *hashp
= fold_hash(hash
);
1710 return len
+ find_zero(mask
);
1715 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1717 unsigned long hash
= init_name_hash();
1719 hash
= partial_name_hash(*name
++, hash
);
1720 return end_name_hash(hash
);
1722 EXPORT_SYMBOL(full_name_hash
);
1725 * We know there's a real path component here of at least
1728 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1730 unsigned long hash
= init_name_hash();
1731 unsigned long len
= 0, c
;
1733 c
= (unsigned char)*name
;
1736 hash
= partial_name_hash(c
, hash
);
1737 c
= (unsigned char)name
[len
];
1738 } while (c
&& c
!= '/');
1739 *hashp
= end_name_hash(hash
);
1747 * This is the basic name resolution function, turning a pathname into
1748 * the final dentry. We expect 'base' to be positive and a directory.
1750 * Returns 0 and nd will have valid dentry and mnt on success.
1751 * Returns error and drops reference to input namei data on failure.
1753 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1763 /* At this point we know we have a real path component. */
1769 err
= may_lookup(nd
);
1773 len
= hash_name(name
, &this.hash
);
1778 if (name
[0] == '.') switch (len
) {
1780 if (name
[1] == '.') {
1782 nd
->flags
|= LOOKUP_JUMPED
;
1788 if (likely(type
== LAST_NORM
)) {
1789 struct dentry
*parent
= nd
->path
.dentry
;
1790 nd
->flags
&= ~LOOKUP_JUMPED
;
1791 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1792 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1800 nd
->last_type
= type
;
1805 * If it wasn't NUL, we know it was '/'. Skip that
1806 * slash, and continue until no more slashes.
1810 } while (unlikely(name
[len
] == '/'));
1816 err
= walk_component(nd
, &next
, LOOKUP_FOLLOW
);
1821 err
= nested_symlink(&next
, nd
);
1825 if (!can_lookup(nd
->inode
)) {
1834 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1835 struct nameidata
*nd
, struct file
**fp
)
1839 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1840 nd
->flags
= flags
| LOOKUP_JUMPED
;
1842 if (flags
& LOOKUP_ROOT
) {
1843 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1845 if (!can_lookup(inode
))
1847 retval
= inode_permission(inode
, MAY_EXEC
);
1851 nd
->path
= nd
->root
;
1853 if (flags
& LOOKUP_RCU
) {
1855 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1857 path_get(&nd
->path
);
1862 nd
->root
.mnt
= NULL
;
1865 if (flags
& LOOKUP_RCU
) {
1870 path_get(&nd
->root
);
1872 nd
->path
= nd
->root
;
1873 } else if (dfd
== AT_FDCWD
) {
1874 if (flags
& LOOKUP_RCU
) {
1875 struct fs_struct
*fs
= current
->fs
;
1881 seq
= read_seqcount_begin(&fs
->seq
);
1883 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1884 } while (read_seqcount_retry(&fs
->seq
, seq
));
1886 get_fs_pwd(current
->fs
, &nd
->path
);
1889 /* Caller must check execute permissions on the starting path component */
1890 struct fd f
= fdget_raw(dfd
);
1891 struct dentry
*dentry
;
1896 dentry
= f
.file
->f_path
.dentry
;
1899 if (!can_lookup(dentry
->d_inode
)) {
1905 nd
->path
= f
.file
->f_path
;
1906 if (flags
& LOOKUP_RCU
) {
1909 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1912 path_get(&nd
->path
);
1917 nd
->inode
= nd
->path
.dentry
->d_inode
;
1921 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1923 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1924 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1926 nd
->flags
&= ~LOOKUP_PARENT
;
1927 return walk_component(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
);
1930 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1931 static int path_lookupat(int dfd
, const char *name
,
1932 unsigned int flags
, struct nameidata
*nd
)
1934 struct file
*base
= NULL
;
1939 * Path walking is largely split up into 2 different synchronisation
1940 * schemes, rcu-walk and ref-walk (explained in
1941 * Documentation/filesystems/path-lookup.txt). These share much of the
1942 * path walk code, but some things particularly setup, cleanup, and
1943 * following mounts are sufficiently divergent that functions are
1944 * duplicated. Typically there is a function foo(), and its RCU
1945 * analogue, foo_rcu().
1947 * -ECHILD is the error number of choice (just to avoid clashes) that
1948 * is returned if some aspect of an rcu-walk fails. Such an error must
1949 * be handled by restarting a traditional ref-walk (which will always
1950 * be able to complete).
1952 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1957 current
->total_link_count
= 0;
1958 err
= link_path_walk(name
, nd
);
1960 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1961 err
= lookup_last(nd
, &path
);
1964 struct path link
= path
;
1965 err
= may_follow_link(&link
, nd
);
1968 nd
->flags
|= LOOKUP_PARENT
;
1969 err
= follow_link(&link
, nd
, &cookie
);
1972 err
= lookup_last(nd
, &path
);
1973 put_link(nd
, &link
, cookie
);
1978 err
= complete_walk(nd
);
1980 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1981 if (!can_lookup(nd
->inode
)) {
1982 path_put(&nd
->path
);
1990 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1991 path_put(&nd
->root
);
1992 nd
->root
.mnt
= NULL
;
1997 static int filename_lookup(int dfd
, struct filename
*name
,
1998 unsigned int flags
, struct nameidata
*nd
)
2000 int retval
= path_lookupat(dfd
, name
->name
, flags
| LOOKUP_RCU
, nd
);
2001 if (unlikely(retval
== -ECHILD
))
2002 retval
= path_lookupat(dfd
, name
->name
, flags
, nd
);
2003 if (unlikely(retval
== -ESTALE
))
2004 retval
= path_lookupat(dfd
, name
->name
,
2005 flags
| LOOKUP_REVAL
, nd
);
2007 if (likely(!retval
))
2008 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2012 static int do_path_lookup(int dfd
, const char *name
,
2013 unsigned int flags
, struct nameidata
*nd
)
2015 struct filename filename
= { .name
= name
};
2017 return filename_lookup(dfd
, &filename
, flags
, nd
);
2020 /* does lookup, returns the object with parent locked */
2021 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2023 struct nameidata nd
;
2025 int err
= do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, &nd
);
2027 return ERR_PTR(err
);
2028 if (nd
.last_type
!= LAST_NORM
) {
2030 return ERR_PTR(-EINVAL
);
2032 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2033 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2035 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2043 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2045 struct nameidata nd
;
2046 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
2053 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2054 * @dentry: pointer to dentry of the base directory
2055 * @mnt: pointer to vfs mount of the base directory
2056 * @name: pointer to file name
2057 * @flags: lookup flags
2058 * @path: pointer to struct path to fill
2060 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2061 const char *name
, unsigned int flags
,
2064 struct nameidata nd
;
2066 nd
.root
.dentry
= dentry
;
2068 BUG_ON(flags
& LOOKUP_PARENT
);
2069 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2070 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
2077 * Restricted form of lookup. Doesn't follow links, single-component only,
2078 * needs parent already locked. Doesn't follow mounts.
2081 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2083 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2087 * lookup_one_len - filesystem helper to lookup single pathname component
2088 * @name: pathname component to lookup
2089 * @base: base directory to lookup from
2090 * @len: maximum length @len should be interpreted to
2092 * Note that this routine is purely a helper for filesystem usage and should
2093 * not be called by generic code. Also note that by using this function the
2094 * nameidata argument is passed to the filesystem methods and a filesystem
2095 * using this helper needs to be prepared for that.
2097 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2103 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2107 this.hash
= full_name_hash(name
, len
);
2109 return ERR_PTR(-EACCES
);
2111 if (unlikely(name
[0] == '.')) {
2112 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2113 return ERR_PTR(-EACCES
);
2117 c
= *(const unsigned char *)name
++;
2118 if (c
== '/' || c
== '\0')
2119 return ERR_PTR(-EACCES
);
2122 * See if the low-level filesystem might want
2123 * to use its own hash..
2125 if (base
->d_flags
& DCACHE_OP_HASH
) {
2126 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
2128 return ERR_PTR(err
);
2131 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2133 return ERR_PTR(err
);
2135 return __lookup_hash(&this, base
, 0);
2138 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2139 struct path
*path
, int *empty
)
2141 struct nameidata nd
;
2142 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2143 int err
= PTR_ERR(tmp
);
2146 BUG_ON(flags
& LOOKUP_PARENT
);
2148 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2156 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2159 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2163 * NB: most callers don't do anything directly with the reference to the
2164 * to struct filename, but the nd->last pointer points into the name string
2165 * allocated by getname. So we must hold the reference to it until all
2166 * path-walking is complete.
2168 static struct filename
*
2169 user_path_parent(int dfd
, const char __user
*path
, struct nameidata
*nd
,
2172 struct filename
*s
= getname(path
);
2175 /* only LOOKUP_REVAL is allowed in extra flags */
2176 flags
&= LOOKUP_REVAL
;
2181 error
= filename_lookup(dfd
, s
, flags
| LOOKUP_PARENT
, nd
);
2184 return ERR_PTR(error
);
2191 * It's inline, so penalty for filesystems that don't use sticky bit is
2194 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
2196 kuid_t fsuid
= current_fsuid();
2198 if (!(dir
->i_mode
& S_ISVTX
))
2200 if (uid_eq(inode
->i_uid
, fsuid
))
2202 if (uid_eq(dir
->i_uid
, fsuid
))
2204 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2208 * Check whether we can remove a link victim from directory dir, check
2209 * whether the type of victim is right.
2210 * 1. We can't do it if dir is read-only (done in permission())
2211 * 2. We should have write and exec permissions on dir
2212 * 3. We can't remove anything from append-only dir
2213 * 4. We can't do anything with immutable dir (done in permission())
2214 * 5. If the sticky bit on dir is set we should either
2215 * a. be owner of dir, or
2216 * b. be owner of victim, or
2217 * c. have CAP_FOWNER capability
2218 * 6. If the victim is append-only or immutable we can't do antyhing with
2219 * links pointing to it.
2220 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2221 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2222 * 9. We can't remove a root or mountpoint.
2223 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2224 * nfs_async_unlink().
2226 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
2230 if (!victim
->d_inode
)
2233 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2234 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2236 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2241 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
2242 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
2245 if (!S_ISDIR(victim
->d_inode
->i_mode
))
2247 if (IS_ROOT(victim
))
2249 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
2251 if (IS_DEADDIR(dir
))
2253 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2258 /* Check whether we can create an object with dentry child in directory
2260 * 1. We can't do it if child already exists (open has special treatment for
2261 * this case, but since we are inlined it's OK)
2262 * 2. We can't do it if dir is read-only (done in permission())
2263 * 3. We should have write and exec permissions on dir
2264 * 4. We can't do it if dir is immutable (done in permission())
2266 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2268 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2271 if (IS_DEADDIR(dir
))
2273 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2277 * p1 and p2 should be directories on the same fs.
2279 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2284 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2288 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2290 p
= d_ancestor(p2
, p1
);
2292 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2293 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2297 p
= d_ancestor(p1
, p2
);
2299 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2300 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2304 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2305 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2309 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2311 mutex_unlock(&p1
->d_inode
->i_mutex
);
2313 mutex_unlock(&p2
->d_inode
->i_mutex
);
2314 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2318 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2321 int error
= may_create(dir
, dentry
);
2325 if (!dir
->i_op
->create
)
2326 return -EACCES
; /* shouldn't it be ENOSYS? */
2329 error
= security_inode_create(dir
, dentry
, mode
);
2332 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2334 fsnotify_create(dir
, dentry
);
2338 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2340 struct dentry
*dentry
= path
->dentry
;
2341 struct inode
*inode
= dentry
->d_inode
;
2351 switch (inode
->i_mode
& S_IFMT
) {
2355 if (acc_mode
& MAY_WRITE
)
2360 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2369 error
= inode_permission(inode
, acc_mode
);
2374 * An append-only file must be opened in append mode for writing.
2376 if (IS_APPEND(inode
)) {
2377 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2383 /* O_NOATIME can only be set by the owner or superuser */
2384 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2390 static int handle_truncate(struct file
*filp
)
2392 struct path
*path
= &filp
->f_path
;
2393 struct inode
*inode
= path
->dentry
->d_inode
;
2394 int error
= get_write_access(inode
);
2398 * Refuse to truncate files with mandatory locks held on them.
2400 error
= locks_verify_locked(inode
);
2402 error
= security_path_truncate(path
);
2404 error
= do_truncate(path
->dentry
, 0,
2405 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2408 put_write_access(inode
);
2412 static inline int open_to_namei_flags(int flag
)
2414 if ((flag
& O_ACCMODE
) == 3)
2419 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2421 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2425 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2429 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2433 * Attempt to atomically look up, create and open a file from a negative
2436 * Returns 0 if successful. The file will have been created and attached to
2437 * @file by the filesystem calling finish_open().
2439 * Returns 1 if the file was looked up only or didn't need creating. The
2440 * caller will need to perform the open themselves. @path will have been
2441 * updated to point to the new dentry. This may be negative.
2443 * Returns an error code otherwise.
2445 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2446 struct path
*path
, struct file
*file
,
2447 const struct open_flags
*op
,
2448 bool got_write
, bool need_lookup
,
2451 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2452 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2456 int create_error
= 0;
2457 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2459 BUG_ON(dentry
->d_inode
);
2461 /* Don't create child dentry for a dead directory. */
2462 if (unlikely(IS_DEADDIR(dir
))) {
2468 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2469 mode
&= ~current_umask();
2471 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
)) {
2472 open_flag
&= ~O_TRUNC
;
2473 *opened
|= FILE_CREATED
;
2477 * Checking write permission is tricky, bacuse we don't know if we are
2478 * going to actually need it: O_CREAT opens should work as long as the
2479 * file exists. But checking existence breaks atomicity. The trick is
2480 * to check access and if not granted clear O_CREAT from the flags.
2482 * Another problem is returing the "right" error value (e.g. for an
2483 * O_EXCL open we want to return EEXIST not EROFS).
2485 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2486 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2487 if (!(open_flag
& O_CREAT
)) {
2489 * No O_CREATE -> atomicity not a requirement -> fall
2490 * back to lookup + open
2493 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2494 /* Fall back and fail with the right error */
2495 create_error
= -EROFS
;
2498 /* No side effects, safe to clear O_CREAT */
2499 create_error
= -EROFS
;
2500 open_flag
&= ~O_CREAT
;
2504 if (open_flag
& O_CREAT
) {
2505 error
= may_o_create(&nd
->path
, dentry
, mode
);
2507 create_error
= error
;
2508 if (open_flag
& O_EXCL
)
2510 open_flag
&= ~O_CREAT
;
2514 if (nd
->flags
& LOOKUP_DIRECTORY
)
2515 open_flag
|= O_DIRECTORY
;
2517 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2518 file
->f_path
.mnt
= nd
->path
.mnt
;
2519 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2522 if (create_error
&& error
== -ENOENT
)
2523 error
= create_error
;
2527 acc_mode
= op
->acc_mode
;
2528 if (*opened
& FILE_CREATED
) {
2529 fsnotify_create(dir
, dentry
);
2530 acc_mode
= MAY_OPEN
;
2533 if (error
) { /* returned 1, that is */
2534 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2538 if (file
->f_path
.dentry
) {
2540 dentry
= file
->f_path
.dentry
;
2542 if (create_error
&& dentry
->d_inode
== NULL
) {
2543 error
= create_error
;
2550 * We didn't have the inode before the open, so check open permission
2553 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2563 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2565 return PTR_ERR(dentry
);
2568 int open_flag
= op
->open_flag
;
2570 error
= create_error
;
2571 if ((open_flag
& O_EXCL
)) {
2572 if (!dentry
->d_inode
)
2574 } else if (!dentry
->d_inode
) {
2576 } else if ((open_flag
& O_TRUNC
) &&
2577 S_ISREG(dentry
->d_inode
->i_mode
)) {
2580 /* will fail later, go on to get the right error */
2584 path
->dentry
= dentry
;
2585 path
->mnt
= nd
->path
.mnt
;
2590 * Look up and maybe create and open the last component.
2592 * Must be called with i_mutex held on parent.
2594 * Returns 0 if the file was successfully atomically created (if necessary) and
2595 * opened. In this case the file will be returned attached to @file.
2597 * Returns 1 if the file was not completely opened at this time, though lookups
2598 * and creations will have been performed and the dentry returned in @path will
2599 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2600 * specified then a negative dentry may be returned.
2602 * An error code is returned otherwise.
2604 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2605 * cleared otherwise prior to returning.
2607 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2609 const struct open_flags
*op
,
2610 bool got_write
, int *opened
)
2612 struct dentry
*dir
= nd
->path
.dentry
;
2613 struct inode
*dir_inode
= dir
->d_inode
;
2614 struct dentry
*dentry
;
2618 *opened
&= ~FILE_CREATED
;
2619 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2621 return PTR_ERR(dentry
);
2623 /* Cached positive dentry: will open in f_op->open */
2624 if (!need_lookup
&& dentry
->d_inode
)
2627 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2628 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2629 need_lookup
, opened
);
2633 BUG_ON(dentry
->d_inode
);
2635 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2637 return PTR_ERR(dentry
);
2640 /* Negative dentry, just create the file */
2641 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2642 umode_t mode
= op
->mode
;
2643 if (!IS_POSIXACL(dir
->d_inode
))
2644 mode
&= ~current_umask();
2646 * This write is needed to ensure that a
2647 * rw->ro transition does not occur between
2648 * the time when the file is created and when
2649 * a permanent write count is taken through
2650 * the 'struct file' in finish_open().
2656 *opened
|= FILE_CREATED
;
2657 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2660 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2661 nd
->flags
& LOOKUP_EXCL
);
2666 path
->dentry
= dentry
;
2667 path
->mnt
= nd
->path
.mnt
;
2676 * Handle the last step of open()
2678 static int do_last(struct nameidata
*nd
, struct path
*path
,
2679 struct file
*file
, const struct open_flags
*op
,
2680 int *opened
, struct filename
*name
)
2682 struct dentry
*dir
= nd
->path
.dentry
;
2683 int open_flag
= op
->open_flag
;
2684 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2685 bool got_write
= false;
2686 int acc_mode
= op
->acc_mode
;
2687 struct inode
*inode
;
2688 bool symlink_ok
= false;
2689 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2690 bool retried
= false;
2693 nd
->flags
&= ~LOOKUP_PARENT
;
2694 nd
->flags
|= op
->intent
;
2696 switch (nd
->last_type
) {
2699 error
= handle_dots(nd
, nd
->last_type
);
2704 error
= complete_walk(nd
);
2707 audit_inode(name
, nd
->path
.dentry
, 0);
2708 if (open_flag
& O_CREAT
) {
2714 error
= complete_walk(nd
);
2717 audit_inode(name
, dir
, 0);
2721 if (!(open_flag
& O_CREAT
)) {
2722 if (nd
->last
.name
[nd
->last
.len
])
2723 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2724 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2726 /* we _can_ be in RCU mode here */
2727 error
= lookup_fast(nd
, path
, &inode
);
2734 BUG_ON(nd
->inode
!= dir
->d_inode
);
2736 /* create side of things */
2738 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2739 * has been cleared when we got to the last component we are
2742 error
= complete_walk(nd
);
2746 audit_inode(name
, dir
, LOOKUP_PARENT
);
2748 /* trailing slashes? */
2749 if (nd
->last
.name
[nd
->last
.len
])
2754 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2755 error
= mnt_want_write(nd
->path
.mnt
);
2759 * do _not_ fail yet - we might not need that or fail with
2760 * a different error; let lookup_open() decide; we'll be
2761 * dropping this one anyway.
2764 mutex_lock(&dir
->d_inode
->i_mutex
);
2765 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2766 mutex_unlock(&dir
->d_inode
->i_mutex
);
2772 if ((*opened
& FILE_CREATED
) ||
2773 !S_ISREG(file_inode(file
)->i_mode
))
2774 will_truncate
= false;
2776 audit_inode(name
, file
->f_path
.dentry
, 0);
2780 if (*opened
& FILE_CREATED
) {
2781 /* Don't check for write permission, don't truncate */
2782 open_flag
&= ~O_TRUNC
;
2783 will_truncate
= false;
2784 acc_mode
= MAY_OPEN
;
2785 path_to_nameidata(path
, nd
);
2786 goto finish_open_created
;
2790 * create/update audit record if it already exists.
2792 if (path
->dentry
->d_inode
)
2793 audit_inode(name
, path
->dentry
, 0);
2796 * If atomic_open() acquired write access it is dropped now due to
2797 * possible mount and symlink following (this might be optimized away if
2801 mnt_drop_write(nd
->path
.mnt
);
2806 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
2809 error
= follow_managed(path
, nd
->flags
);
2814 nd
->flags
|= LOOKUP_JUMPED
;
2816 BUG_ON(nd
->flags
& LOOKUP_RCU
);
2817 inode
= path
->dentry
->d_inode
;
2819 /* we _can_ be in RCU mode here */
2822 path_to_nameidata(path
, nd
);
2826 if (should_follow_link(inode
, !symlink_ok
)) {
2827 if (nd
->flags
& LOOKUP_RCU
) {
2828 if (unlikely(nd
->path
.mnt
!= path
->mnt
||
2829 unlazy_walk(nd
, path
->dentry
))) {
2834 BUG_ON(inode
!= path
->dentry
->d_inode
);
2838 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
2839 path_to_nameidata(path
, nd
);
2841 save_parent
.dentry
= nd
->path
.dentry
;
2842 save_parent
.mnt
= mntget(path
->mnt
);
2843 nd
->path
.dentry
= path
->dentry
;
2847 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2848 error
= complete_walk(nd
);
2850 path_put(&save_parent
);
2854 if ((open_flag
& O_CREAT
) && S_ISDIR(nd
->inode
->i_mode
))
2857 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !can_lookup(nd
->inode
))
2859 audit_inode(name
, nd
->path
.dentry
, 0);
2861 if (!S_ISREG(nd
->inode
->i_mode
))
2862 will_truncate
= false;
2864 if (will_truncate
) {
2865 error
= mnt_want_write(nd
->path
.mnt
);
2870 finish_open_created
:
2871 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2874 file
->f_path
.mnt
= nd
->path
.mnt
;
2875 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
2877 if (error
== -EOPENSTALE
)
2882 error
= open_check_o_direct(file
);
2885 error
= ima_file_check(file
, op
->acc_mode
);
2889 if (will_truncate
) {
2890 error
= handle_truncate(file
);
2896 mnt_drop_write(nd
->path
.mnt
);
2897 path_put(&save_parent
);
2902 path_put_conditional(path
, nd
);
2909 /* If no saved parent or already retried then can't retry */
2910 if (!save_parent
.dentry
|| retried
)
2913 BUG_ON(save_parent
.dentry
!= dir
);
2914 path_put(&nd
->path
);
2915 nd
->path
= save_parent
;
2916 nd
->inode
= dir
->d_inode
;
2917 save_parent
.mnt
= NULL
;
2918 save_parent
.dentry
= NULL
;
2920 mnt_drop_write(nd
->path
.mnt
);
2927 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
2928 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2930 struct file
*base
= NULL
;
2936 file
= get_empty_filp();
2940 file
->f_flags
= op
->open_flag
;
2942 error
= path_init(dfd
, pathname
->name
, flags
| LOOKUP_PARENT
, nd
, &base
);
2943 if (unlikely(error
))
2946 current
->total_link_count
= 0;
2947 error
= link_path_walk(pathname
->name
, nd
);
2948 if (unlikely(error
))
2951 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2952 while (unlikely(error
> 0)) { /* trailing symlink */
2953 struct path link
= path
;
2955 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2956 path_put_conditional(&path
, nd
);
2957 path_put(&nd
->path
);
2961 error
= may_follow_link(&link
, nd
);
2962 if (unlikely(error
))
2964 nd
->flags
|= LOOKUP_PARENT
;
2965 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2966 error
= follow_link(&link
, nd
, &cookie
);
2967 if (unlikely(error
))
2969 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2970 put_link(nd
, &link
, cookie
);
2973 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2974 path_put(&nd
->root
);
2977 if (!(opened
& FILE_OPENED
)) {
2981 if (unlikely(error
)) {
2982 if (error
== -EOPENSTALE
) {
2983 if (flags
& LOOKUP_RCU
)
2988 file
= ERR_PTR(error
);
2993 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
2994 const struct open_flags
*op
, int flags
)
2996 struct nameidata nd
;
2999 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3000 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3001 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3002 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3003 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3007 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3008 const char *name
, const struct open_flags
*op
, int flags
)
3010 struct nameidata nd
;
3012 struct filename filename
= { .name
= name
};
3015 nd
.root
.dentry
= dentry
;
3017 flags
|= LOOKUP_ROOT
;
3019 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
3020 return ERR_PTR(-ELOOP
);
3022 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3023 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3024 file
= path_openat(-1, &filename
, &nd
, op
, flags
);
3025 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3026 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3030 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3031 struct path
*path
, unsigned int lookup_flags
)
3033 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3034 struct nameidata nd
;
3037 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3040 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3041 * other flags passed in are ignored!
3043 lookup_flags
&= LOOKUP_REVAL
;
3045 error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
|lookup_flags
, &nd
);
3047 return ERR_PTR(error
);
3050 * Yucky last component or no last component at all?
3051 * (foo/., foo/.., /////)
3053 if (nd
.last_type
!= LAST_NORM
)
3055 nd
.flags
&= ~LOOKUP_PARENT
;
3056 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3058 /* don't fail immediately if it's r/o, at least try to report other errors */
3059 err2
= mnt_want_write(nd
.path
.mnt
);
3061 * Do the final lookup.
3063 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3064 dentry
= lookup_hash(&nd
);
3069 if (dentry
->d_inode
)
3072 * Special case - lookup gave negative, but... we had foo/bar/
3073 * From the vfs_mknod() POV we just have a negative dentry -
3074 * all is fine. Let's be bastards - you had / on the end, you've
3075 * been asking for (non-existent) directory. -ENOENT for you.
3077 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3081 if (unlikely(err2
)) {
3089 dentry
= ERR_PTR(error
);
3091 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3093 mnt_drop_write(nd
.path
.mnt
);
3098 EXPORT_SYMBOL(kern_path_create
);
3100 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3103 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3104 mnt_drop_write(path
->mnt
);
3107 EXPORT_SYMBOL(done_path_create
);
3109 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3110 struct path
*path
, unsigned int lookup_flags
)
3112 struct filename
*tmp
= getname(pathname
);
3115 return ERR_CAST(tmp
);
3116 res
= kern_path_create(dfd
, tmp
->name
, path
, lookup_flags
);
3120 EXPORT_SYMBOL(user_path_create
);
3122 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3124 int error
= may_create(dir
, dentry
);
3129 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3132 if (!dir
->i_op
->mknod
)
3135 error
= devcgroup_inode_mknod(mode
, dev
);
3139 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3143 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3145 fsnotify_create(dir
, dentry
);
3149 static int may_mknod(umode_t mode
)
3151 switch (mode
& S_IFMT
) {
3157 case 0: /* zero mode translates to S_IFREG */
3166 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3169 struct dentry
*dentry
;
3172 unsigned int lookup_flags
= 0;
3174 error
= may_mknod(mode
);
3178 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3180 return PTR_ERR(dentry
);
3182 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3183 mode
&= ~current_umask();
3184 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3187 switch (mode
& S_IFMT
) {
3188 case 0: case S_IFREG
:
3189 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3191 case S_IFCHR
: case S_IFBLK
:
3192 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3193 new_decode_dev(dev
));
3195 case S_IFIFO
: case S_IFSOCK
:
3196 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3200 done_path_create(&path
, dentry
);
3201 if (retry_estale(error
, lookup_flags
)) {
3202 lookup_flags
|= LOOKUP_REVAL
;
3208 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3210 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3213 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3215 int error
= may_create(dir
, dentry
);
3216 unsigned max_links
= dir
->i_sb
->s_max_links
;
3221 if (!dir
->i_op
->mkdir
)
3224 mode
&= (S_IRWXUGO
|S_ISVTX
);
3225 error
= security_inode_mkdir(dir
, dentry
, mode
);
3229 if (max_links
&& dir
->i_nlink
>= max_links
)
3232 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3234 fsnotify_mkdir(dir
, dentry
);
3238 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3240 struct dentry
*dentry
;
3243 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3246 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3248 return PTR_ERR(dentry
);
3250 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3251 mode
&= ~current_umask();
3252 error
= security_path_mkdir(&path
, dentry
, mode
);
3254 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3255 done_path_create(&path
, dentry
);
3256 if (retry_estale(error
, lookup_flags
)) {
3257 lookup_flags
|= LOOKUP_REVAL
;
3263 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3265 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3269 * The dentry_unhash() helper will try to drop the dentry early: we
3270 * should have a usage count of 1 if we're the only user of this
3271 * dentry, and if that is true (possibly after pruning the dcache),
3272 * then we drop the dentry now.
3274 * A low-level filesystem can, if it choses, legally
3277 * if (!d_unhashed(dentry))
3280 * if it cannot handle the case of removing a directory
3281 * that is still in use by something else..
3283 void dentry_unhash(struct dentry
*dentry
)
3285 shrink_dcache_parent(dentry
);
3286 spin_lock(&dentry
->d_lock
);
3287 if (dentry
->d_count
== 1)
3289 spin_unlock(&dentry
->d_lock
);
3292 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3294 int error
= may_delete(dir
, dentry
, 1);
3299 if (!dir
->i_op
->rmdir
)
3303 mutex_lock(&dentry
->d_inode
->i_mutex
);
3306 if (d_mountpoint(dentry
))
3309 error
= security_inode_rmdir(dir
, dentry
);
3313 shrink_dcache_parent(dentry
);
3314 error
= dir
->i_op
->rmdir(dir
, dentry
);
3318 dentry
->d_inode
->i_flags
|= S_DEAD
;
3322 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3329 static long do_rmdir(int dfd
, const char __user
*pathname
)
3332 struct filename
*name
;
3333 struct dentry
*dentry
;
3334 struct nameidata nd
;
3335 unsigned int lookup_flags
= 0;
3337 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3339 return PTR_ERR(name
);
3341 switch(nd
.last_type
) {
3353 nd
.flags
&= ~LOOKUP_PARENT
;
3354 error
= mnt_want_write(nd
.path
.mnt
);
3358 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3359 dentry
= lookup_hash(&nd
);
3360 error
= PTR_ERR(dentry
);
3363 if (!dentry
->d_inode
) {
3367 error
= security_path_rmdir(&nd
.path
, dentry
);
3370 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3374 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3375 mnt_drop_write(nd
.path
.mnt
);
3379 if (retry_estale(error
, lookup_flags
)) {
3380 lookup_flags
|= LOOKUP_REVAL
;
3386 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3388 return do_rmdir(AT_FDCWD
, pathname
);
3391 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3393 int error
= may_delete(dir
, dentry
, 0);
3398 if (!dir
->i_op
->unlink
)
3401 mutex_lock(&dentry
->d_inode
->i_mutex
);
3402 if (d_mountpoint(dentry
))
3405 error
= security_inode_unlink(dir
, dentry
);
3407 error
= dir
->i_op
->unlink(dir
, dentry
);
3412 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3414 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3415 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3416 fsnotify_link_count(dentry
->d_inode
);
3424 * Make sure that the actual truncation of the file will occur outside its
3425 * directory's i_mutex. Truncate can take a long time if there is a lot of
3426 * writeout happening, and we don't want to prevent access to the directory
3427 * while waiting on the I/O.
3429 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3432 struct filename
*name
;
3433 struct dentry
*dentry
;
3434 struct nameidata nd
;
3435 struct inode
*inode
= NULL
;
3436 unsigned int lookup_flags
= 0;
3438 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3440 return PTR_ERR(name
);
3443 if (nd
.last_type
!= LAST_NORM
)
3446 nd
.flags
&= ~LOOKUP_PARENT
;
3447 error
= mnt_want_write(nd
.path
.mnt
);
3451 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3452 dentry
= lookup_hash(&nd
);
3453 error
= PTR_ERR(dentry
);
3454 if (!IS_ERR(dentry
)) {
3455 /* Why not before? Because we want correct error value */
3456 if (nd
.last
.name
[nd
.last
.len
])
3458 inode
= dentry
->d_inode
;
3462 error
= security_path_unlink(&nd
.path
, dentry
);
3465 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
3469 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3471 iput(inode
); /* truncate the inode here */
3472 mnt_drop_write(nd
.path
.mnt
);
3476 if (retry_estale(error
, lookup_flags
)) {
3477 lookup_flags
|= LOOKUP_REVAL
;
3484 error
= !dentry
->d_inode
? -ENOENT
:
3485 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
3489 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3491 if ((flag
& ~AT_REMOVEDIR
) != 0)
3494 if (flag
& AT_REMOVEDIR
)
3495 return do_rmdir(dfd
, pathname
);
3497 return do_unlinkat(dfd
, pathname
);
3500 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3502 return do_unlinkat(AT_FDCWD
, pathname
);
3505 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3507 int error
= may_create(dir
, dentry
);
3512 if (!dir
->i_op
->symlink
)
3515 error
= security_inode_symlink(dir
, dentry
, oldname
);
3519 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3521 fsnotify_create(dir
, dentry
);
3525 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3526 int, newdfd
, const char __user
*, newname
)
3529 struct filename
*from
;
3530 struct dentry
*dentry
;
3532 unsigned int lookup_flags
= 0;
3534 from
= getname(oldname
);
3536 return PTR_ERR(from
);
3538 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3539 error
= PTR_ERR(dentry
);
3543 error
= security_path_symlink(&path
, dentry
, from
->name
);
3545 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3546 done_path_create(&path
, dentry
);
3547 if (retry_estale(error
, lookup_flags
)) {
3548 lookup_flags
|= LOOKUP_REVAL
;
3556 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3558 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3561 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3563 struct inode
*inode
= old_dentry
->d_inode
;
3564 unsigned max_links
= dir
->i_sb
->s_max_links
;
3570 error
= may_create(dir
, new_dentry
);
3574 if (dir
->i_sb
!= inode
->i_sb
)
3578 * A link to an append-only or immutable file cannot be created.
3580 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3582 if (!dir
->i_op
->link
)
3584 if (S_ISDIR(inode
->i_mode
))
3587 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3591 mutex_lock(&inode
->i_mutex
);
3592 /* Make sure we don't allow creating hardlink to an unlinked file */
3593 if (inode
->i_nlink
== 0)
3595 else if (max_links
&& inode
->i_nlink
>= max_links
)
3598 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3599 mutex_unlock(&inode
->i_mutex
);
3601 fsnotify_link(dir
, inode
, new_dentry
);
3606 * Hardlinks are often used in delicate situations. We avoid
3607 * security-related surprises by not following symlinks on the
3610 * We don't follow them on the oldname either to be compatible
3611 * with linux 2.0, and to avoid hard-linking to directories
3612 * and other special files. --ADM
3614 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3615 int, newdfd
, const char __user
*, newname
, int, flags
)
3617 struct dentry
*new_dentry
;
3618 struct path old_path
, new_path
;
3622 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3625 * To use null names we require CAP_DAC_READ_SEARCH
3626 * This ensures that not everyone will be able to create
3627 * handlink using the passed filedescriptor.
3629 if (flags
& AT_EMPTY_PATH
) {
3630 if (!capable(CAP_DAC_READ_SEARCH
))
3635 if (flags
& AT_SYMLINK_FOLLOW
)
3636 how
|= LOOKUP_FOLLOW
;
3638 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3642 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
3643 (how
& LOOKUP_REVAL
));
3644 error
= PTR_ERR(new_dentry
);
3645 if (IS_ERR(new_dentry
))
3649 if (old_path
.mnt
!= new_path
.mnt
)
3651 error
= may_linkat(&old_path
);
3652 if (unlikely(error
))
3654 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3657 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3659 done_path_create(&new_path
, new_dentry
);
3660 if (retry_estale(error
, how
)) {
3661 path_put(&old_path
);
3662 how
|= LOOKUP_REVAL
;
3666 path_put(&old_path
);
3671 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3673 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3677 * The worst of all namespace operations - renaming directory. "Perverted"
3678 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3680 * a) we can get into loop creation. Check is done in is_subdir().
3681 * b) race potential - two innocent renames can create a loop together.
3682 * That's where 4.4 screws up. Current fix: serialization on
3683 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3685 * c) we have to lock _three_ objects - parents and victim (if it exists).
3686 * And that - after we got ->i_mutex on parents (until then we don't know
3687 * whether the target exists). Solution: try to be smart with locking
3688 * order for inodes. We rely on the fact that tree topology may change
3689 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3690 * move will be locked. Thus we can rank directories by the tree
3691 * (ancestors first) and rank all non-directories after them.
3692 * That works since everybody except rename does "lock parent, lookup,
3693 * lock child" and rename is under ->s_vfs_rename_mutex.
3694 * HOWEVER, it relies on the assumption that any object with ->lookup()
3695 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3696 * we'd better make sure that there's no link(2) for them.
3697 * d) conversion from fhandle to dentry may come in the wrong moment - when
3698 * we are removing the target. Solution: we will have to grab ->i_mutex
3699 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3700 * ->i_mutex on parents, which works but leads to some truly excessive
3703 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3704 struct inode
*new_dir
, struct dentry
*new_dentry
)
3707 struct inode
*target
= new_dentry
->d_inode
;
3708 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3711 * If we are going to change the parent - check write permissions,
3712 * we'll need to flip '..'.
3714 if (new_dir
!= old_dir
) {
3715 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3720 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3726 mutex_lock(&target
->i_mutex
);
3729 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3733 if (max_links
&& !target
&& new_dir
!= old_dir
&&
3734 new_dir
->i_nlink
>= max_links
)
3738 shrink_dcache_parent(new_dentry
);
3739 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3744 target
->i_flags
|= S_DEAD
;
3745 dont_mount(new_dentry
);
3749 mutex_unlock(&target
->i_mutex
);
3752 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3753 d_move(old_dentry
,new_dentry
);
3757 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3758 struct inode
*new_dir
, struct dentry
*new_dentry
)
3760 struct inode
*target
= new_dentry
->d_inode
;
3763 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3769 mutex_lock(&target
->i_mutex
);
3772 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3775 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3780 dont_mount(new_dentry
);
3781 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3782 d_move(old_dentry
, new_dentry
);
3785 mutex_unlock(&target
->i_mutex
);
3790 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3791 struct inode
*new_dir
, struct dentry
*new_dentry
)
3794 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3795 const unsigned char *old_name
;
3797 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3800 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3804 if (!new_dentry
->d_inode
)
3805 error
= may_create(new_dir
, new_dentry
);
3807 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3811 if (!old_dir
->i_op
->rename
)
3814 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3817 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3819 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3821 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3822 new_dentry
->d_inode
, old_dentry
);
3823 fsnotify_oldname_free(old_name
);
3828 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3829 int, newdfd
, const char __user
*, newname
)
3831 struct dentry
*old_dir
, *new_dir
;
3832 struct dentry
*old_dentry
, *new_dentry
;
3833 struct dentry
*trap
;
3834 struct nameidata oldnd
, newnd
;
3835 struct filename
*from
;
3836 struct filename
*to
;
3837 unsigned int lookup_flags
= 0;
3838 bool should_retry
= false;
3841 from
= user_path_parent(olddfd
, oldname
, &oldnd
, lookup_flags
);
3843 error
= PTR_ERR(from
);
3847 to
= user_path_parent(newdfd
, newname
, &newnd
, lookup_flags
);
3849 error
= PTR_ERR(to
);
3854 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3857 old_dir
= oldnd
.path
.dentry
;
3859 if (oldnd
.last_type
!= LAST_NORM
)
3862 new_dir
= newnd
.path
.dentry
;
3863 if (newnd
.last_type
!= LAST_NORM
)
3866 error
= mnt_want_write(oldnd
.path
.mnt
);
3870 oldnd
.flags
&= ~LOOKUP_PARENT
;
3871 newnd
.flags
&= ~LOOKUP_PARENT
;
3872 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3874 trap
= lock_rename(new_dir
, old_dir
);
3876 old_dentry
= lookup_hash(&oldnd
);
3877 error
= PTR_ERR(old_dentry
);
3878 if (IS_ERR(old_dentry
))
3880 /* source must exist */
3882 if (!old_dentry
->d_inode
)
3884 /* unless the source is a directory trailing slashes give -ENOTDIR */
3885 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3887 if (oldnd
.last
.name
[oldnd
.last
.len
])
3889 if (newnd
.last
.name
[newnd
.last
.len
])
3892 /* source should not be ancestor of target */
3894 if (old_dentry
== trap
)
3896 new_dentry
= lookup_hash(&newnd
);
3897 error
= PTR_ERR(new_dentry
);
3898 if (IS_ERR(new_dentry
))
3900 /* target should not be an ancestor of source */
3902 if (new_dentry
== trap
)
3905 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3906 &newnd
.path
, new_dentry
);
3909 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3910 new_dir
->d_inode
, new_dentry
);
3916 unlock_rename(new_dir
, old_dir
);
3917 mnt_drop_write(oldnd
.path
.mnt
);
3919 if (retry_estale(error
, lookup_flags
))
3920 should_retry
= true;
3921 path_put(&newnd
.path
);
3924 path_put(&oldnd
.path
);
3927 should_retry
= false;
3928 lookup_flags
|= LOOKUP_REVAL
;
3935 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3937 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3940 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3944 len
= PTR_ERR(link
);
3949 if (len
> (unsigned) buflen
)
3951 if (copy_to_user(buffer
, link
, len
))
3958 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3959 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3960 * using) it for any given inode is up to filesystem.
3962 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3964 struct nameidata nd
;
3969 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3971 return PTR_ERR(cookie
);
3973 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3974 if (dentry
->d_inode
->i_op
->put_link
)
3975 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3979 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3981 return __vfs_follow_link(nd
, link
);
3984 /* get the link contents into pagecache */
3985 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3989 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3990 page
= read_mapping_page(mapping
, 0, NULL
);
3995 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3999 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4001 struct page
*page
= NULL
;
4002 char *s
= page_getlink(dentry
, &page
);
4003 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
4006 page_cache_release(page
);
4011 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
4013 struct page
*page
= NULL
;
4014 nd_set_link(nd
, page_getlink(dentry
, &page
));
4018 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
4020 struct page
*page
= cookie
;
4024 page_cache_release(page
);
4029 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4031 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4033 struct address_space
*mapping
= inode
->i_mapping
;
4038 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4040 flags
|= AOP_FLAG_NOFS
;
4043 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4044 flags
, &page
, &fsdata
);
4048 kaddr
= kmap_atomic(page
);
4049 memcpy(kaddr
, symname
, len
-1);
4050 kunmap_atomic(kaddr
);
4052 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4059 mark_inode_dirty(inode
);
4065 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4067 return __page_symlink(inode
, symname
, len
,
4068 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4071 const struct inode_operations page_symlink_inode_operations
= {
4072 .readlink
= generic_readlink
,
4073 .follow_link
= page_follow_link_light
,
4074 .put_link
= page_put_link
,
4077 EXPORT_SYMBOL(user_path_at
);
4078 EXPORT_SYMBOL(follow_down_one
);
4079 EXPORT_SYMBOL(follow_down
);
4080 EXPORT_SYMBOL(follow_up
);
4081 EXPORT_SYMBOL(get_write_access
); /* nfsd */
4082 EXPORT_SYMBOL(lock_rename
);
4083 EXPORT_SYMBOL(lookup_one_len
);
4084 EXPORT_SYMBOL(page_follow_link_light
);
4085 EXPORT_SYMBOL(page_put_link
);
4086 EXPORT_SYMBOL(page_readlink
);
4087 EXPORT_SYMBOL(__page_symlink
);
4088 EXPORT_SYMBOL(page_symlink
);
4089 EXPORT_SYMBOL(page_symlink_inode_operations
);
4090 EXPORT_SYMBOL(kern_path
);
4091 EXPORT_SYMBOL(vfs_path_lookup
);
4092 EXPORT_SYMBOL(inode_permission
);
4093 EXPORT_SYMBOL(unlock_rename
);
4094 EXPORT_SYMBOL(vfs_create
);
4095 EXPORT_SYMBOL(vfs_follow_link
);
4096 EXPORT_SYMBOL(vfs_link
);
4097 EXPORT_SYMBOL(vfs_mkdir
);
4098 EXPORT_SYMBOL(vfs_mknod
);
4099 EXPORT_SYMBOL(generic_permission
);
4100 EXPORT_SYMBOL(vfs_readlink
);
4101 EXPORT_SYMBOL(vfs_rename
);
4102 EXPORT_SYMBOL(vfs_rmdir
);
4103 EXPORT_SYMBOL(vfs_symlink
);
4104 EXPORT_SYMBOL(vfs_unlink
);
4105 EXPORT_SYMBOL(dentry_unhash
);
4106 EXPORT_SYMBOL(generic_readlink
);