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 <asm/uaccess.h>
42 /* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
90 * [10-Sep-98 Alan Modra] Another symlink change.
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 static char *getname_flags(const char __user
*filename
, int flags
, int *empty
)
122 char *result
= __getname(), *err
;
125 if (unlikely(!result
))
126 return ERR_PTR(-ENOMEM
);
128 len
= strncpy_from_user(result
, filename
, PATH_MAX
);
130 if (unlikely(len
< 0))
133 /* The empty path is special. */
134 if (unlikely(!len
)) {
137 err
= ERR_PTR(-ENOENT
);
138 if (!(flags
& LOOKUP_EMPTY
))
142 err
= ERR_PTR(-ENAMETOOLONG
);
143 if (likely(len
< PATH_MAX
)) {
144 audit_getname(result
);
153 char *getname(const char __user
* filename
)
155 return getname_flags(filename
, 0, NULL
);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name
)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname
);
169 static int check_acl(struct inode
*inode
, int mask
)
171 #ifdef CONFIG_FS_POSIX_ACL
172 struct posix_acl
*acl
;
174 if (mask
& MAY_NOT_BLOCK
) {
175 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
178 /* no ->get_acl() calls in RCU mode... */
179 if (acl
== ACL_NOT_CACHED
)
181 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
184 acl
= get_cached_acl(inode
, ACL_TYPE_ACCESS
);
187 * A filesystem can force a ACL callback by just never filling the
188 * ACL cache. But normally you'd fill the cache either at inode
189 * instantiation time, or on the first ->get_acl call.
191 * If the filesystem doesn't have a get_acl() function at all, we'll
192 * just create the negative cache entry.
194 if (acl
== ACL_NOT_CACHED
) {
195 if (inode
->i_op
->get_acl
) {
196 acl
= inode
->i_op
->get_acl(inode
, ACL_TYPE_ACCESS
);
200 set_cached_acl(inode
, ACL_TYPE_ACCESS
, NULL
);
206 int error
= posix_acl_permission(inode
, acl
, mask
);
207 posix_acl_release(acl
);
216 * This does the basic permission checking
218 static int acl_permission_check(struct inode
*inode
, int mask
)
220 unsigned int mode
= inode
->i_mode
;
222 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
225 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
226 int error
= check_acl(inode
, mask
);
227 if (error
!= -EAGAIN
)
231 if (in_group_p(inode
->i_gid
))
236 * If the DACs are ok we don't need any capability check.
238 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
244 * generic_permission - check for access rights on a Posix-like filesystem
245 * @inode: inode to check access rights for
246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
248 * Used to check for read/write/execute permissions on a file.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
251 * are used for other things.
253 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
254 * request cannot be satisfied (eg. requires blocking or too much complexity).
255 * It would then be called again in ref-walk mode.
257 int generic_permission(struct inode
*inode
, int mask
)
262 * Do the basic permission checks.
264 ret
= acl_permission_check(inode
, mask
);
268 if (S_ISDIR(inode
->i_mode
)) {
269 /* DACs are overridable for directories */
270 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
272 if (!(mask
& MAY_WRITE
))
273 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
278 * Read/write DACs are always overridable.
279 * Executable DACs are overridable when there is
280 * at least one exec bit set.
282 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
283 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
287 * Searching includes executable on directories, else just read.
289 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
290 if (mask
== MAY_READ
)
291 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
298 * We _really_ want to just do "generic_permission()" without
299 * even looking at the inode->i_op values. So we keep a cache
300 * flag in inode->i_opflags, that says "this has not special
301 * permission function, use the fast case".
303 static inline int do_inode_permission(struct inode
*inode
, int mask
)
305 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
306 if (likely(inode
->i_op
->permission
))
307 return inode
->i_op
->permission(inode
, mask
);
309 /* This gets set once for the inode lifetime */
310 spin_lock(&inode
->i_lock
);
311 inode
->i_opflags
|= IOP_FASTPERM
;
312 spin_unlock(&inode
->i_lock
);
314 return generic_permission(inode
, mask
);
318 * __inode_permission - Check for access rights to a given inode
319 * @inode: Inode to check permission on
320 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
322 * Check for read/write/execute permissions on an inode.
324 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
326 * This does not check for a read-only file system. You probably want
327 * inode_permission().
329 int __inode_permission(struct inode
*inode
, int mask
)
333 if (unlikely(mask
& MAY_WRITE
)) {
335 * Nobody gets write access to an immutable file.
337 if (IS_IMMUTABLE(inode
))
341 retval
= do_inode_permission(inode
, mask
);
345 retval
= devcgroup_inode_permission(inode
, mask
);
349 return security_inode_permission(inode
, mask
);
353 * sb_permission - Check superblock-level permissions
354 * @sb: Superblock of inode to check permission on
355 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
357 * Separate out file-system wide checks from inode-specific permission checks.
359 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
361 if (unlikely(mask
& MAY_WRITE
)) {
362 umode_t mode
= inode
->i_mode
;
364 /* Nobody gets write access to a read-only fs. */
365 if ((sb
->s_flags
& MS_RDONLY
) &&
366 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
373 * inode_permission - Check for access rights to a given inode
374 * @inode: Inode to check permission on
375 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
377 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
378 * this, letting us set arbitrary permissions for filesystem access without
379 * changing the "normal" UIDs which are used for other things.
381 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
383 int inode_permission(struct inode
*inode
, int mask
)
387 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
390 return __inode_permission(inode
, mask
);
394 * path_get - get a reference to a path
395 * @path: path to get the reference to
397 * Given a path increment the reference count to the dentry and the vfsmount.
399 void path_get(struct path
*path
)
404 EXPORT_SYMBOL(path_get
);
407 * path_put - put a reference to a path
408 * @path: path to put the reference to
410 * Given a path decrement the reference count to the dentry and the vfsmount.
412 void path_put(struct path
*path
)
417 EXPORT_SYMBOL(path_put
);
420 * Path walking has 2 modes, rcu-walk and ref-walk (see
421 * Documentation/filesystems/path-lookup.txt). In situations when we can't
422 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
423 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
424 * mode. Refcounts are grabbed at the last known good point before rcu-walk
425 * got stuck, so ref-walk may continue from there. If this is not successful
426 * (eg. a seqcount has changed), then failure is returned and it's up to caller
427 * to restart the path walk from the beginning in ref-walk mode.
430 static inline void lock_rcu_walk(void)
432 br_read_lock(&vfsmount_lock
);
436 static inline void unlock_rcu_walk(void)
439 br_read_unlock(&vfsmount_lock
);
443 * unlazy_walk - try to switch to ref-walk mode.
444 * @nd: nameidata pathwalk data
445 * @dentry: child of nd->path.dentry or NULL
446 * Returns: 0 on success, -ECHILD on failure
448 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
449 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
450 * @nd or NULL. Must be called from rcu-walk context.
452 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
454 struct fs_struct
*fs
= current
->fs
;
455 struct dentry
*parent
= nd
->path
.dentry
;
458 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
459 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
461 spin_lock(&fs
->lock
);
462 if (nd
->root
.mnt
!= fs
->root
.mnt
||
463 nd
->root
.dentry
!= fs
->root
.dentry
)
466 spin_lock(&parent
->d_lock
);
468 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
470 BUG_ON(nd
->inode
!= parent
->d_inode
);
472 if (dentry
->d_parent
!= parent
)
474 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
475 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
478 * If the sequence check on the child dentry passed, then
479 * the child has not been removed from its parent. This
480 * means the parent dentry must be valid and able to take
481 * a reference at this point.
483 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
484 BUG_ON(!parent
->d_count
);
486 spin_unlock(&dentry
->d_lock
);
488 spin_unlock(&parent
->d_lock
);
491 spin_unlock(&fs
->lock
);
493 mntget(nd
->path
.mnt
);
496 nd
->flags
&= ~LOOKUP_RCU
;
500 spin_unlock(&dentry
->d_lock
);
502 spin_unlock(&parent
->d_lock
);
505 spin_unlock(&fs
->lock
);
509 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
511 return dentry
->d_op
->d_revalidate(dentry
, flags
);
515 * complete_walk - successful completion of path walk
516 * @nd: pointer nameidata
518 * If we had been in RCU mode, drop out of it and legitimize nd->path.
519 * Revalidate the final result, unless we'd already done that during
520 * the path walk or the filesystem doesn't ask for it. Return 0 on
521 * success, -error on failure. In case of failure caller does not
522 * need to drop nd->path.
524 static int complete_walk(struct nameidata
*nd
)
526 struct dentry
*dentry
= nd
->path
.dentry
;
529 if (nd
->flags
& LOOKUP_RCU
) {
530 nd
->flags
&= ~LOOKUP_RCU
;
531 if (!(nd
->flags
& LOOKUP_ROOT
))
533 spin_lock(&dentry
->d_lock
);
534 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
535 spin_unlock(&dentry
->d_lock
);
539 BUG_ON(nd
->inode
!= dentry
->d_inode
);
540 spin_unlock(&dentry
->d_lock
);
541 mntget(nd
->path
.mnt
);
545 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
548 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
551 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
554 /* Note: we do not d_invalidate() */
555 status
= d_revalidate(dentry
, nd
->flags
);
566 static __always_inline
void set_root(struct nameidata
*nd
)
569 get_fs_root(current
->fs
, &nd
->root
);
572 static int link_path_walk(const char *, struct nameidata
*);
574 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
577 struct fs_struct
*fs
= current
->fs
;
581 seq
= read_seqcount_begin(&fs
->seq
);
583 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
584 } while (read_seqcount_retry(&fs
->seq
, seq
));
588 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
600 nd
->flags
|= LOOKUP_JUMPED
;
602 nd
->inode
= nd
->path
.dentry
->d_inode
;
604 ret
= link_path_walk(link
, nd
);
608 return PTR_ERR(link
);
611 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
614 if (path
->mnt
!= nd
->path
.mnt
)
618 static inline void path_to_nameidata(const struct path
*path
,
619 struct nameidata
*nd
)
621 if (!(nd
->flags
& LOOKUP_RCU
)) {
622 dput(nd
->path
.dentry
);
623 if (nd
->path
.mnt
!= path
->mnt
)
624 mntput(nd
->path
.mnt
);
626 nd
->path
.mnt
= path
->mnt
;
627 nd
->path
.dentry
= path
->dentry
;
631 * Helper to directly jump to a known parsed path from ->follow_link,
632 * caller must have taken a reference to path beforehand.
634 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
639 nd
->inode
= nd
->path
.dentry
->d_inode
;
640 nd
->flags
|= LOOKUP_JUMPED
;
642 BUG_ON(nd
->inode
->i_op
->follow_link
);
645 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
647 struct inode
*inode
= link
->dentry
->d_inode
;
648 if (inode
->i_op
->put_link
)
649 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
653 int sysctl_protected_symlinks __read_mostly
= 1;
654 int sysctl_protected_hardlinks __read_mostly
= 1;
657 * may_follow_link - Check symlink following for unsafe situations
658 * @link: The path of the symlink
660 * In the case of the sysctl_protected_symlinks sysctl being enabled,
661 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
662 * in a sticky world-writable directory. This is to protect privileged
663 * processes from failing races against path names that may change out
664 * from under them by way of other users creating malicious symlinks.
665 * It will permit symlinks to be followed only when outside a sticky
666 * world-writable directory, or when the uid of the symlink and follower
667 * match, or when the directory owner matches the symlink's owner.
669 * Returns 0 if following the symlink is allowed, -ve on error.
671 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
673 const struct inode
*inode
;
674 const struct inode
*parent
;
676 if (!sysctl_protected_symlinks
)
679 /* Allowed if owner and follower match. */
680 inode
= link
->dentry
->d_inode
;
681 if (current_cred()->fsuid
== inode
->i_uid
)
684 /* Allowed if parent directory not sticky and world-writable. */
685 parent
= nd
->path
.dentry
->d_inode
;
686 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
689 /* Allowed if parent directory and link owner match. */
690 if (parent
->i_uid
== inode
->i_uid
)
693 path_put_conditional(link
, nd
);
695 audit_log_link_denied("follow_link", link
);
700 * safe_hardlink_source - Check for safe hardlink conditions
701 * @inode: the source inode to hardlink from
703 * Return false if at least one of the following conditions:
704 * - inode is not a regular file
706 * - inode is setgid and group-exec
707 * - access failure for read and write
709 * Otherwise returns true.
711 static bool safe_hardlink_source(struct inode
*inode
)
713 umode_t mode
= inode
->i_mode
;
715 /* Special files should not get pinned to the filesystem. */
719 /* Setuid files should not get pinned to the filesystem. */
723 /* Executable setgid files should not get pinned to the filesystem. */
724 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
727 /* Hardlinking to unreadable or unwritable sources is dangerous. */
728 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
735 * may_linkat - Check permissions for creating a hardlink
736 * @link: the source to hardlink from
738 * Block hardlink when all of:
739 * - sysctl_protected_hardlinks enabled
740 * - fsuid does not match inode
741 * - hardlink source is unsafe (see safe_hardlink_source() above)
744 * Returns 0 if successful, -ve on error.
746 static int may_linkat(struct path
*link
)
748 const struct cred
*cred
;
751 if (!sysctl_protected_hardlinks
)
754 cred
= current_cred();
755 inode
= link
->dentry
->d_inode
;
757 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
758 * otherwise, it must be a safe source.
760 if (cred
->fsuid
== inode
->i_uid
|| safe_hardlink_source(inode
) ||
764 audit_log_link_denied("linkat", link
);
768 static __always_inline
int
769 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
771 struct dentry
*dentry
= link
->dentry
;
775 BUG_ON(nd
->flags
& LOOKUP_RCU
);
777 if (link
->mnt
== nd
->path
.mnt
)
781 if (unlikely(current
->total_link_count
>= 40))
782 goto out_put_nd_path
;
785 current
->total_link_count
++;
788 nd_set_link(nd
, NULL
);
790 error
= security_inode_follow_link(link
->dentry
, nd
);
792 goto out_put_nd_path
;
794 nd
->last_type
= LAST_BIND
;
795 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
798 goto out_put_nd_path
;
803 error
= __vfs_follow_link(nd
, s
);
805 put_link(nd
, link
, *p
);
816 static int follow_up_rcu(struct path
*path
)
818 struct mount
*mnt
= real_mount(path
->mnt
);
819 struct mount
*parent
;
820 struct dentry
*mountpoint
;
822 parent
= mnt
->mnt_parent
;
823 if (&parent
->mnt
== path
->mnt
)
825 mountpoint
= mnt
->mnt_mountpoint
;
826 path
->dentry
= mountpoint
;
827 path
->mnt
= &parent
->mnt
;
832 * follow_up - Find the mountpoint of path's vfsmount
834 * Given a path, find the mountpoint of its source file system.
835 * Replace @path with the path of the mountpoint in the parent mount.
838 * Return 1 if we went up a level and 0 if we were already at the
841 int follow_up(struct path
*path
)
843 struct mount
*mnt
= real_mount(path
->mnt
);
844 struct mount
*parent
;
845 struct dentry
*mountpoint
;
847 br_read_lock(&vfsmount_lock
);
848 parent
= mnt
->mnt_parent
;
850 br_read_unlock(&vfsmount_lock
);
853 mntget(&parent
->mnt
);
854 mountpoint
= dget(mnt
->mnt_mountpoint
);
855 br_read_unlock(&vfsmount_lock
);
857 path
->dentry
= mountpoint
;
859 path
->mnt
= &parent
->mnt
;
864 * Perform an automount
865 * - return -EISDIR to tell follow_managed() to stop and return the path we
868 static int follow_automount(struct path
*path
, unsigned flags
,
871 struct vfsmount
*mnt
;
874 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
877 /* We don't want to mount if someone's just doing a stat -
878 * unless they're stat'ing a directory and appended a '/' to
881 * We do, however, want to mount if someone wants to open or
882 * create a file of any type under the mountpoint, wants to
883 * traverse through the mountpoint or wants to open the
884 * mounted directory. Also, autofs may mark negative dentries
885 * as being automount points. These will need the attentions
886 * of the daemon to instantiate them before they can be used.
888 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
889 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
890 path
->dentry
->d_inode
)
893 current
->total_link_count
++;
894 if (current
->total_link_count
>= 40)
897 mnt
= path
->dentry
->d_op
->d_automount(path
);
900 * The filesystem is allowed to return -EISDIR here to indicate
901 * it doesn't want to automount. For instance, autofs would do
902 * this so that its userspace daemon can mount on this dentry.
904 * However, we can only permit this if it's a terminal point in
905 * the path being looked up; if it wasn't then the remainder of
906 * the path is inaccessible and we should say so.
908 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
913 if (!mnt
) /* mount collision */
917 /* lock_mount() may release path->mnt on error */
921 err
= finish_automount(mnt
, path
);
925 /* Someone else made a mount here whilst we were busy */
930 path
->dentry
= dget(mnt
->mnt_root
);
939 * Handle a dentry that is managed in some way.
940 * - Flagged for transit management (autofs)
941 * - Flagged as mountpoint
942 * - Flagged as automount point
944 * This may only be called in refwalk mode.
946 * Serialization is taken care of in namespace.c
948 static int follow_managed(struct path
*path
, unsigned flags
)
950 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
952 bool need_mntput
= false;
955 /* Given that we're not holding a lock here, we retain the value in a
956 * local variable for each dentry as we look at it so that we don't see
957 * the components of that value change under us */
958 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
959 managed
&= DCACHE_MANAGED_DENTRY
,
960 unlikely(managed
!= 0)) {
961 /* Allow the filesystem to manage the transit without i_mutex
963 if (managed
& DCACHE_MANAGE_TRANSIT
) {
964 BUG_ON(!path
->dentry
->d_op
);
965 BUG_ON(!path
->dentry
->d_op
->d_manage
);
966 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
971 /* Transit to a mounted filesystem. */
972 if (managed
& DCACHE_MOUNTED
) {
973 struct vfsmount
*mounted
= lookup_mnt(path
);
979 path
->dentry
= dget(mounted
->mnt_root
);
984 /* Something is mounted on this dentry in another
985 * namespace and/or whatever was mounted there in this
986 * namespace got unmounted before we managed to get the
990 /* Handle an automount point */
991 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
992 ret
= follow_automount(path
, flags
, &need_mntput
);
998 /* We didn't change the current path point */
1002 if (need_mntput
&& path
->mnt
== mnt
)
1006 return ret
< 0 ? ret
: need_mntput
;
1009 int follow_down_one(struct path
*path
)
1011 struct vfsmount
*mounted
;
1013 mounted
= lookup_mnt(path
);
1017 path
->mnt
= mounted
;
1018 path
->dentry
= dget(mounted
->mnt_root
);
1024 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
1026 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
1027 dentry
->d_op
->d_manage(dentry
, true) < 0);
1031 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1032 * we meet a managed dentry that would need blocking.
1034 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1035 struct inode
**inode
)
1038 struct mount
*mounted
;
1040 * Don't forget we might have a non-mountpoint managed dentry
1041 * that wants to block transit.
1043 if (unlikely(managed_dentry_might_block(path
->dentry
)))
1046 if (!d_mountpoint(path
->dentry
))
1049 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1052 path
->mnt
= &mounted
->mnt
;
1053 path
->dentry
= mounted
->mnt
.mnt_root
;
1054 nd
->flags
|= LOOKUP_JUMPED
;
1055 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1057 * Update the inode too. We don't need to re-check the
1058 * dentry sequence number here after this d_inode read,
1059 * because a mount-point is always pinned.
1061 *inode
= path
->dentry
->d_inode
;
1066 static void follow_mount_rcu(struct nameidata
*nd
)
1068 while (d_mountpoint(nd
->path
.dentry
)) {
1069 struct mount
*mounted
;
1070 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
1073 nd
->path
.mnt
= &mounted
->mnt
;
1074 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1075 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1079 static int follow_dotdot_rcu(struct nameidata
*nd
)
1084 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1085 nd
->path
.mnt
== nd
->root
.mnt
) {
1088 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1089 struct dentry
*old
= nd
->path
.dentry
;
1090 struct dentry
*parent
= old
->d_parent
;
1093 seq
= read_seqcount_begin(&parent
->d_seq
);
1094 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1096 nd
->path
.dentry
= parent
;
1100 if (!follow_up_rcu(&nd
->path
))
1102 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1104 follow_mount_rcu(nd
);
1105 nd
->inode
= nd
->path
.dentry
->d_inode
;
1109 nd
->flags
&= ~LOOKUP_RCU
;
1110 if (!(nd
->flags
& LOOKUP_ROOT
))
1111 nd
->root
.mnt
= NULL
;
1117 * Follow down to the covering mount currently visible to userspace. At each
1118 * point, the filesystem owning that dentry may be queried as to whether the
1119 * caller is permitted to proceed or not.
1121 int follow_down(struct path
*path
)
1126 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1127 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1128 /* Allow the filesystem to manage the transit without i_mutex
1131 * We indicate to the filesystem if someone is trying to mount
1132 * something here. This gives autofs the chance to deny anyone
1133 * other than its daemon the right to mount on its
1136 * The filesystem may sleep at this point.
1138 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1139 BUG_ON(!path
->dentry
->d_op
);
1140 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1141 ret
= path
->dentry
->d_op
->d_manage(
1142 path
->dentry
, false);
1144 return ret
== -EISDIR
? 0 : ret
;
1147 /* Transit to a mounted filesystem. */
1148 if (managed
& DCACHE_MOUNTED
) {
1149 struct vfsmount
*mounted
= lookup_mnt(path
);
1154 path
->mnt
= mounted
;
1155 path
->dentry
= dget(mounted
->mnt_root
);
1159 /* Don't handle automount points here */
1166 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1168 static void follow_mount(struct path
*path
)
1170 while (d_mountpoint(path
->dentry
)) {
1171 struct vfsmount
*mounted
= lookup_mnt(path
);
1176 path
->mnt
= mounted
;
1177 path
->dentry
= dget(mounted
->mnt_root
);
1181 static void follow_dotdot(struct nameidata
*nd
)
1186 struct dentry
*old
= nd
->path
.dentry
;
1188 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1189 nd
->path
.mnt
== nd
->root
.mnt
) {
1192 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1193 /* rare case of legitimate dget_parent()... */
1194 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1198 if (!follow_up(&nd
->path
))
1201 follow_mount(&nd
->path
);
1202 nd
->inode
= nd
->path
.dentry
->d_inode
;
1206 * This looks up the name in dcache, possibly revalidates the old dentry and
1207 * allocates a new one if not found or not valid. In the need_lookup argument
1208 * returns whether i_op->lookup is necessary.
1210 * dir->d_inode->i_mutex must be held
1212 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1213 unsigned int flags
, bool *need_lookup
)
1215 struct dentry
*dentry
;
1218 *need_lookup
= false;
1219 dentry
= d_lookup(dir
, name
);
1221 if (d_need_lookup(dentry
)) {
1222 *need_lookup
= true;
1223 } else if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1224 error
= d_revalidate(dentry
, flags
);
1225 if (unlikely(error
<= 0)) {
1228 return ERR_PTR(error
);
1229 } else if (!d_invalidate(dentry
)) {
1238 dentry
= d_alloc(dir
, name
);
1239 if (unlikely(!dentry
))
1240 return ERR_PTR(-ENOMEM
);
1242 *need_lookup
= true;
1248 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1249 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1251 * dir->d_inode->i_mutex must be held
1253 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1258 /* Don't create child dentry for a dead directory. */
1259 if (unlikely(IS_DEADDIR(dir
))) {
1261 return ERR_PTR(-ENOENT
);
1264 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1265 if (unlikely(old
)) {
1272 static struct dentry
*__lookup_hash(struct qstr
*name
,
1273 struct dentry
*base
, unsigned int flags
)
1276 struct dentry
*dentry
;
1278 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1282 return lookup_real(base
->d_inode
, dentry
, flags
);
1286 * It's more convoluted than I'd like it to be, but... it's still fairly
1287 * small and for now I'd prefer to have fast path as straight as possible.
1288 * It _is_ time-critical.
1290 static int lookup_fast(struct nameidata
*nd
, struct qstr
*name
,
1291 struct path
*path
, struct inode
**inode
)
1293 struct vfsmount
*mnt
= nd
->path
.mnt
;
1294 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1300 * Rename seqlock is not required here because in the off chance
1301 * of a false negative due to a concurrent rename, we're going to
1302 * do the non-racy lookup, below.
1304 if (nd
->flags
& LOOKUP_RCU
) {
1306 dentry
= __d_lookup_rcu(parent
, name
, &seq
, nd
->inode
);
1311 * This sequence count validates that the inode matches
1312 * the dentry name information from lookup.
1314 *inode
= dentry
->d_inode
;
1315 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1319 * This sequence count validates that the parent had no
1320 * changes while we did the lookup of the dentry above.
1322 * The memory barrier in read_seqcount_begin of child is
1323 * enough, we can use __read_seqcount_retry here.
1325 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1329 if (unlikely(d_need_lookup(dentry
)))
1331 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1332 status
= d_revalidate(dentry
, nd
->flags
);
1333 if (unlikely(status
<= 0)) {
1334 if (status
!= -ECHILD
)
1340 path
->dentry
= dentry
;
1341 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1343 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1347 if (unlazy_walk(nd
, dentry
))
1350 dentry
= __d_lookup(parent
, name
);
1353 if (unlikely(!dentry
))
1356 if (unlikely(d_need_lookup(dentry
))) {
1361 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1362 status
= d_revalidate(dentry
, nd
->flags
);
1363 if (unlikely(status
<= 0)) {
1368 if (!d_invalidate(dentry
)) {
1375 path
->dentry
= dentry
;
1376 err
= follow_managed(path
, nd
->flags
);
1377 if (unlikely(err
< 0)) {
1378 path_put_conditional(path
, nd
);
1382 nd
->flags
|= LOOKUP_JUMPED
;
1383 *inode
= path
->dentry
->d_inode
;
1390 /* Fast lookup failed, do it the slow way */
1391 static int lookup_slow(struct nameidata
*nd
, struct qstr
*name
,
1394 struct dentry
*dentry
, *parent
;
1397 parent
= nd
->path
.dentry
;
1398 BUG_ON(nd
->inode
!= parent
->d_inode
);
1400 mutex_lock(&parent
->d_inode
->i_mutex
);
1401 dentry
= __lookup_hash(name
, parent
, nd
->flags
);
1402 mutex_unlock(&parent
->d_inode
->i_mutex
);
1404 return PTR_ERR(dentry
);
1405 path
->mnt
= nd
->path
.mnt
;
1406 path
->dentry
= dentry
;
1407 err
= follow_managed(path
, nd
->flags
);
1408 if (unlikely(err
< 0)) {
1409 path_put_conditional(path
, nd
);
1413 nd
->flags
|= LOOKUP_JUMPED
;
1417 static inline int may_lookup(struct nameidata
*nd
)
1419 if (nd
->flags
& LOOKUP_RCU
) {
1420 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1423 if (unlazy_walk(nd
, NULL
))
1426 return inode_permission(nd
->inode
, MAY_EXEC
);
1429 static inline int handle_dots(struct nameidata
*nd
, int type
)
1431 if (type
== LAST_DOTDOT
) {
1432 if (nd
->flags
& LOOKUP_RCU
) {
1433 if (follow_dotdot_rcu(nd
))
1441 static void terminate_walk(struct nameidata
*nd
)
1443 if (!(nd
->flags
& LOOKUP_RCU
)) {
1444 path_put(&nd
->path
);
1446 nd
->flags
&= ~LOOKUP_RCU
;
1447 if (!(nd
->flags
& LOOKUP_ROOT
))
1448 nd
->root
.mnt
= NULL
;
1454 * Do we need to follow links? We _really_ want to be able
1455 * to do this check without having to look at inode->i_op,
1456 * so we keep a cache of "no, this doesn't need follow_link"
1457 * for the common case.
1459 static inline int should_follow_link(struct inode
*inode
, int follow
)
1461 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1462 if (likely(inode
->i_op
->follow_link
))
1465 /* This gets set once for the inode lifetime */
1466 spin_lock(&inode
->i_lock
);
1467 inode
->i_opflags
|= IOP_NOFOLLOW
;
1468 spin_unlock(&inode
->i_lock
);
1473 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1474 struct qstr
*name
, int type
, int follow
)
1476 struct inode
*inode
;
1479 * "." and ".." are special - ".." especially so because it has
1480 * to be able to know about the current root directory and
1481 * parent relationships.
1483 if (unlikely(type
!= LAST_NORM
))
1484 return handle_dots(nd
, type
);
1485 err
= lookup_fast(nd
, name
, path
, &inode
);
1486 if (unlikely(err
)) {
1490 err
= lookup_slow(nd
, name
, path
);
1494 inode
= path
->dentry
->d_inode
;
1500 if (should_follow_link(inode
, follow
)) {
1501 if (nd
->flags
& LOOKUP_RCU
) {
1502 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1507 BUG_ON(inode
!= path
->dentry
->d_inode
);
1510 path_to_nameidata(path
, nd
);
1515 path_to_nameidata(path
, nd
);
1522 * This limits recursive symlink follows to 8, while
1523 * limiting consecutive symlinks to 40.
1525 * Without that kind of total limit, nasty chains of consecutive
1526 * symlinks can cause almost arbitrarily long lookups.
1528 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1532 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1533 path_put_conditional(path
, nd
);
1534 path_put(&nd
->path
);
1537 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1540 current
->link_count
++;
1543 struct path link
= *path
;
1546 res
= follow_link(&link
, nd
, &cookie
);
1549 res
= walk_component(nd
, path
, &nd
->last
,
1550 nd
->last_type
, LOOKUP_FOLLOW
);
1551 put_link(nd
, &link
, cookie
);
1554 current
->link_count
--;
1560 * We really don't want to look at inode->i_op->lookup
1561 * when we don't have to. So we keep a cache bit in
1562 * the inode ->i_opflags field that says "yes, we can
1563 * do lookup on this inode".
1565 static inline int can_lookup(struct inode
*inode
)
1567 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1569 if (likely(!inode
->i_op
->lookup
))
1572 /* We do this once for the lifetime of the inode */
1573 spin_lock(&inode
->i_lock
);
1574 inode
->i_opflags
|= IOP_LOOKUP
;
1575 spin_unlock(&inode
->i_lock
);
1580 * We can do the critical dentry name comparison and hashing
1581 * operations one word at a time, but we are limited to:
1583 * - Architectures with fast unaligned word accesses. We could
1584 * do a "get_unaligned()" if this helps and is sufficiently
1587 * - Little-endian machines (so that we can generate the mask
1588 * of low bytes efficiently). Again, we *could* do a byte
1589 * swapping load on big-endian architectures if that is not
1590 * expensive enough to make the optimization worthless.
1592 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1593 * do not trap on the (extremely unlikely) case of a page
1594 * crossing operation.
1596 * - Furthermore, we need an efficient 64-bit compile for the
1597 * 64-bit case in order to generate the "number of bytes in
1598 * the final mask". Again, that could be replaced with a
1599 * efficient population count instruction or similar.
1601 #ifdef CONFIG_DCACHE_WORD_ACCESS
1603 #include <asm/word-at-a-time.h>
1607 static inline unsigned int fold_hash(unsigned long hash
)
1609 hash
+= hash
>> (8*sizeof(int));
1613 #else /* 32-bit case */
1615 #define fold_hash(x) (x)
1619 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1621 unsigned long a
, mask
;
1622 unsigned long hash
= 0;
1625 a
= load_unaligned_zeropad(name
);
1626 if (len
< sizeof(unsigned long))
1630 name
+= sizeof(unsigned long);
1631 len
-= sizeof(unsigned long);
1635 mask
= ~(~0ul << len
*8);
1638 return fold_hash(hash
);
1640 EXPORT_SYMBOL(full_name_hash
);
1643 * Calculate the length and hash of the path component, and
1644 * return the length of the component;
1646 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1648 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1649 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1652 len
= -sizeof(unsigned long);
1654 hash
= (hash
+ a
) * 9;
1655 len
+= sizeof(unsigned long);
1656 a
= load_unaligned_zeropad(name
+len
);
1657 b
= a
^ REPEAT_BYTE('/');
1658 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1660 adata
= prep_zero_mask(a
, adata
, &constants
);
1661 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1663 mask
= create_zero_mask(adata
| bdata
);
1665 hash
+= a
& zero_bytemask(mask
);
1666 *hashp
= fold_hash(hash
);
1668 return len
+ find_zero(mask
);
1673 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1675 unsigned long hash
= init_name_hash();
1677 hash
= partial_name_hash(*name
++, hash
);
1678 return end_name_hash(hash
);
1680 EXPORT_SYMBOL(full_name_hash
);
1683 * We know there's a real path component here of at least
1686 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1688 unsigned long hash
= init_name_hash();
1689 unsigned long len
= 0, c
;
1691 c
= (unsigned char)*name
;
1694 hash
= partial_name_hash(c
, hash
);
1695 c
= (unsigned char)name
[len
];
1696 } while (c
&& c
!= '/');
1697 *hashp
= end_name_hash(hash
);
1705 * This is the basic name resolution function, turning a pathname into
1706 * the final dentry. We expect 'base' to be positive and a directory.
1708 * Returns 0 and nd will have valid dentry and mnt on success.
1709 * Returns error and drops reference to input namei data on failure.
1711 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1721 /* At this point we know we have a real path component. */
1727 err
= may_lookup(nd
);
1731 len
= hash_name(name
, &this.hash
);
1736 if (name
[0] == '.') switch (len
) {
1738 if (name
[1] == '.') {
1740 nd
->flags
|= LOOKUP_JUMPED
;
1746 if (likely(type
== LAST_NORM
)) {
1747 struct dentry
*parent
= nd
->path
.dentry
;
1748 nd
->flags
&= ~LOOKUP_JUMPED
;
1749 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1750 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1758 goto last_component
;
1760 * If it wasn't NUL, we know it was '/'. Skip that
1761 * slash, and continue until no more slashes.
1765 } while (unlikely(name
[len
] == '/'));
1767 goto last_component
;
1770 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1775 err
= nested_symlink(&next
, nd
);
1779 if (can_lookup(nd
->inode
))
1783 /* here ends the main loop */
1787 nd
->last_type
= type
;
1794 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1795 struct nameidata
*nd
, struct file
**fp
)
1801 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1802 nd
->flags
= flags
| LOOKUP_JUMPED
;
1804 if (flags
& LOOKUP_ROOT
) {
1805 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1807 if (!inode
->i_op
->lookup
)
1809 retval
= inode_permission(inode
, MAY_EXEC
);
1813 nd
->path
= nd
->root
;
1815 if (flags
& LOOKUP_RCU
) {
1817 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1819 path_get(&nd
->path
);
1824 nd
->root
.mnt
= NULL
;
1827 if (flags
& LOOKUP_RCU
) {
1832 path_get(&nd
->root
);
1834 nd
->path
= nd
->root
;
1835 } else if (dfd
== AT_FDCWD
) {
1836 if (flags
& LOOKUP_RCU
) {
1837 struct fs_struct
*fs
= current
->fs
;
1843 seq
= read_seqcount_begin(&fs
->seq
);
1845 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1846 } while (read_seqcount_retry(&fs
->seq
, seq
));
1848 get_fs_pwd(current
->fs
, &nd
->path
);
1851 struct dentry
*dentry
;
1853 file
= fget_raw_light(dfd
, &fput_needed
);
1858 dentry
= file
->f_path
.dentry
;
1862 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1865 retval
= inode_permission(dentry
->d_inode
, MAY_EXEC
);
1870 nd
->path
= file
->f_path
;
1871 if (flags
& LOOKUP_RCU
) {
1874 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1877 path_get(&file
->f_path
);
1878 fput_light(file
, fput_needed
);
1882 nd
->inode
= nd
->path
.dentry
->d_inode
;
1886 fput_light(file
, fput_needed
);
1891 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1893 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1894 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1896 nd
->flags
&= ~LOOKUP_PARENT
;
1897 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1898 nd
->flags
& LOOKUP_FOLLOW
);
1901 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1902 static int path_lookupat(int dfd
, const char *name
,
1903 unsigned int flags
, struct nameidata
*nd
)
1905 struct file
*base
= NULL
;
1910 * Path walking is largely split up into 2 different synchronisation
1911 * schemes, rcu-walk and ref-walk (explained in
1912 * Documentation/filesystems/path-lookup.txt). These share much of the
1913 * path walk code, but some things particularly setup, cleanup, and
1914 * following mounts are sufficiently divergent that functions are
1915 * duplicated. Typically there is a function foo(), and its RCU
1916 * analogue, foo_rcu().
1918 * -ECHILD is the error number of choice (just to avoid clashes) that
1919 * is returned if some aspect of an rcu-walk fails. Such an error must
1920 * be handled by restarting a traditional ref-walk (which will always
1921 * be able to complete).
1923 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1928 current
->total_link_count
= 0;
1929 err
= link_path_walk(name
, nd
);
1931 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1932 err
= lookup_last(nd
, &path
);
1935 struct path link
= path
;
1936 err
= may_follow_link(&link
, nd
);
1939 nd
->flags
|= LOOKUP_PARENT
;
1940 err
= follow_link(&link
, nd
, &cookie
);
1943 err
= lookup_last(nd
, &path
);
1944 put_link(nd
, &link
, cookie
);
1949 err
= complete_walk(nd
);
1951 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1952 if (!nd
->inode
->i_op
->lookup
) {
1953 path_put(&nd
->path
);
1961 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1962 path_put(&nd
->root
);
1963 nd
->root
.mnt
= NULL
;
1968 static int do_path_lookup(int dfd
, const char *name
,
1969 unsigned int flags
, struct nameidata
*nd
)
1971 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1972 if (unlikely(retval
== -ECHILD
))
1973 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1974 if (unlikely(retval
== -ESTALE
))
1975 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1977 if (likely(!retval
)) {
1978 if (unlikely(!audit_dummy_context())) {
1979 if (nd
->path
.dentry
&& nd
->inode
)
1980 audit_inode(name
, nd
->path
.dentry
);
1986 /* does lookup, returns the object with parent locked */
1987 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
1989 struct nameidata nd
;
1991 int err
= do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, &nd
);
1993 return ERR_PTR(err
);
1994 if (nd
.last_type
!= LAST_NORM
) {
1996 return ERR_PTR(-EINVAL
);
1998 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1999 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2001 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2009 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2011 struct nameidata nd
;
2012 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
2019 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2020 * @dentry: pointer to dentry of the base directory
2021 * @mnt: pointer to vfs mount of the base directory
2022 * @name: pointer to file name
2023 * @flags: lookup flags
2024 * @path: pointer to struct path to fill
2026 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2027 const char *name
, unsigned int flags
,
2030 struct nameidata nd
;
2032 nd
.root
.dentry
= dentry
;
2034 BUG_ON(flags
& LOOKUP_PARENT
);
2035 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2036 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
2043 * Restricted form of lookup. Doesn't follow links, single-component only,
2044 * needs parent already locked. Doesn't follow mounts.
2047 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2049 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2053 * lookup_one_len - filesystem helper to lookup single pathname component
2054 * @name: pathname component to lookup
2055 * @base: base directory to lookup from
2056 * @len: maximum length @len should be interpreted to
2058 * Note that this routine is purely a helper for filesystem usage and should
2059 * not be called by generic code. Also note that by using this function the
2060 * nameidata argument is passed to the filesystem methods and a filesystem
2061 * using this helper needs to be prepared for that.
2063 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2069 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2073 this.hash
= full_name_hash(name
, len
);
2075 return ERR_PTR(-EACCES
);
2078 c
= *(const unsigned char *)name
++;
2079 if (c
== '/' || c
== '\0')
2080 return ERR_PTR(-EACCES
);
2083 * See if the low-level filesystem might want
2084 * to use its own hash..
2086 if (base
->d_flags
& DCACHE_OP_HASH
) {
2087 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
2089 return ERR_PTR(err
);
2092 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2094 return ERR_PTR(err
);
2096 return __lookup_hash(&this, base
, 0);
2099 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2100 struct path
*path
, int *empty
)
2102 struct nameidata nd
;
2103 char *tmp
= getname_flags(name
, flags
, empty
);
2104 int err
= PTR_ERR(tmp
);
2107 BUG_ON(flags
& LOOKUP_PARENT
);
2109 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
2117 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2120 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2123 static int user_path_parent(int dfd
, const char __user
*path
,
2124 struct nameidata
*nd
, char **name
)
2126 char *s
= getname(path
);
2132 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
2142 * It's inline, so penalty for filesystems that don't use sticky bit is
2145 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
2147 kuid_t fsuid
= current_fsuid();
2149 if (!(dir
->i_mode
& S_ISVTX
))
2151 if (uid_eq(inode
->i_uid
, fsuid
))
2153 if (uid_eq(dir
->i_uid
, fsuid
))
2155 return !inode_capable(inode
, CAP_FOWNER
);
2159 * Check whether we can remove a link victim from directory dir, check
2160 * whether the type of victim is right.
2161 * 1. We can't do it if dir is read-only (done in permission())
2162 * 2. We should have write and exec permissions on dir
2163 * 3. We can't remove anything from append-only dir
2164 * 4. We can't do anything with immutable dir (done in permission())
2165 * 5. If the sticky bit on dir is set we should either
2166 * a. be owner of dir, or
2167 * b. be owner of victim, or
2168 * c. have CAP_FOWNER capability
2169 * 6. If the victim is append-only or immutable we can't do antyhing with
2170 * links pointing to it.
2171 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2172 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2173 * 9. We can't remove a root or mountpoint.
2174 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2175 * nfs_async_unlink().
2177 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
2181 if (!victim
->d_inode
)
2184 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2185 audit_inode_child(victim
, dir
);
2187 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2192 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
2193 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
2196 if (!S_ISDIR(victim
->d_inode
->i_mode
))
2198 if (IS_ROOT(victim
))
2200 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
2202 if (IS_DEADDIR(dir
))
2204 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2209 /* Check whether we can create an object with dentry child in directory
2211 * 1. We can't do it if child already exists (open has special treatment for
2212 * this case, but since we are inlined it's OK)
2213 * 2. We can't do it if dir is read-only (done in permission())
2214 * 3. We should have write and exec permissions on dir
2215 * 4. We can't do it if dir is immutable (done in permission())
2217 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2221 if (IS_DEADDIR(dir
))
2223 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2227 * p1 and p2 should be directories on the same fs.
2229 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2234 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2238 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2240 p
= d_ancestor(p2
, p1
);
2242 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2243 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2247 p
= d_ancestor(p1
, p2
);
2249 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2250 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2254 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2255 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2259 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2261 mutex_unlock(&p1
->d_inode
->i_mutex
);
2263 mutex_unlock(&p2
->d_inode
->i_mutex
);
2264 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2268 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2271 int error
= may_create(dir
, dentry
);
2275 if (!dir
->i_op
->create
)
2276 return -EACCES
; /* shouldn't it be ENOSYS? */
2279 error
= security_inode_create(dir
, dentry
, mode
);
2282 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2284 fsnotify_create(dir
, dentry
);
2288 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2290 struct dentry
*dentry
= path
->dentry
;
2291 struct inode
*inode
= dentry
->d_inode
;
2301 switch (inode
->i_mode
& S_IFMT
) {
2305 if (acc_mode
& MAY_WRITE
)
2310 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2319 error
= inode_permission(inode
, acc_mode
);
2324 * An append-only file must be opened in append mode for writing.
2326 if (IS_APPEND(inode
)) {
2327 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2333 /* O_NOATIME can only be set by the owner or superuser */
2334 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2340 static int handle_truncate(struct file
*filp
)
2342 struct path
*path
= &filp
->f_path
;
2343 struct inode
*inode
= path
->dentry
->d_inode
;
2344 int error
= get_write_access(inode
);
2348 * Refuse to truncate files with mandatory locks held on them.
2350 error
= locks_verify_locked(inode
);
2352 error
= security_path_truncate(path
);
2354 error
= do_truncate(path
->dentry
, 0,
2355 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2358 put_write_access(inode
);
2362 static inline int open_to_namei_flags(int flag
)
2364 if ((flag
& O_ACCMODE
) == 3)
2369 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2371 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2375 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2379 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2383 * Attempt to atomically look up, create and open a file from a negative
2386 * Returns 0 if successful. The file will have been created and attached to
2387 * @file by the filesystem calling finish_open().
2389 * Returns 1 if the file was looked up only or didn't need creating. The
2390 * caller will need to perform the open themselves. @path will have been
2391 * updated to point to the new dentry. This may be negative.
2393 * Returns an error code otherwise.
2395 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2396 struct path
*path
, struct file
*file
,
2397 const struct open_flags
*op
,
2398 bool got_write
, bool need_lookup
,
2401 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2402 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2406 int create_error
= 0;
2407 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2409 BUG_ON(dentry
->d_inode
);
2411 /* Don't create child dentry for a dead directory. */
2412 if (unlikely(IS_DEADDIR(dir
))) {
2418 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2419 mode
&= ~current_umask();
2421 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
)) {
2422 open_flag
&= ~O_TRUNC
;
2423 *opened
|= FILE_CREATED
;
2427 * Checking write permission is tricky, bacuse we don't know if we are
2428 * going to actually need it: O_CREAT opens should work as long as the
2429 * file exists. But checking existence breaks atomicity. The trick is
2430 * to check access and if not granted clear O_CREAT from the flags.
2432 * Another problem is returing the "right" error value (e.g. for an
2433 * O_EXCL open we want to return EEXIST not EROFS).
2435 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2436 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2437 if (!(open_flag
& O_CREAT
)) {
2439 * No O_CREATE -> atomicity not a requirement -> fall
2440 * back to lookup + open
2443 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2444 /* Fall back and fail with the right error */
2445 create_error
= -EROFS
;
2448 /* No side effects, safe to clear O_CREAT */
2449 create_error
= -EROFS
;
2450 open_flag
&= ~O_CREAT
;
2454 if (open_flag
& O_CREAT
) {
2455 error
= may_o_create(&nd
->path
, dentry
, mode
);
2457 create_error
= error
;
2458 if (open_flag
& O_EXCL
)
2460 open_flag
&= ~O_CREAT
;
2464 if (nd
->flags
& LOOKUP_DIRECTORY
)
2465 open_flag
|= O_DIRECTORY
;
2467 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2468 file
->f_path
.mnt
= nd
->path
.mnt
;
2469 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2472 if (create_error
&& error
== -ENOENT
)
2473 error
= create_error
;
2477 acc_mode
= op
->acc_mode
;
2478 if (*opened
& FILE_CREATED
) {
2479 fsnotify_create(dir
, dentry
);
2480 acc_mode
= MAY_OPEN
;
2483 if (error
) { /* returned 1, that is */
2484 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2488 if (file
->f_path
.dentry
) {
2490 dentry
= file
->f_path
.dentry
;
2492 if (create_error
&& dentry
->d_inode
== NULL
) {
2493 error
= create_error
;
2500 * We didn't have the inode before the open, so check open permission
2503 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2513 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2515 return PTR_ERR(dentry
);
2518 int open_flag
= op
->open_flag
;
2520 error
= create_error
;
2521 if ((open_flag
& O_EXCL
)) {
2522 if (!dentry
->d_inode
)
2524 } else if (!dentry
->d_inode
) {
2526 } else if ((open_flag
& O_TRUNC
) &&
2527 S_ISREG(dentry
->d_inode
->i_mode
)) {
2530 /* will fail later, go on to get the right error */
2534 path
->dentry
= dentry
;
2535 path
->mnt
= nd
->path
.mnt
;
2540 * Look up and maybe create and open the last component.
2542 * Must be called with i_mutex held on parent.
2544 * Returns 0 if the file was successfully atomically created (if necessary) and
2545 * opened. In this case the file will be returned attached to @file.
2547 * Returns 1 if the file was not completely opened at this time, though lookups
2548 * and creations will have been performed and the dentry returned in @path will
2549 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2550 * specified then a negative dentry may be returned.
2552 * An error code is returned otherwise.
2554 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2555 * cleared otherwise prior to returning.
2557 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2559 const struct open_flags
*op
,
2560 bool got_write
, int *opened
)
2562 struct dentry
*dir
= nd
->path
.dentry
;
2563 struct inode
*dir_inode
= dir
->d_inode
;
2564 struct dentry
*dentry
;
2568 *opened
&= ~FILE_CREATED
;
2569 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2571 return PTR_ERR(dentry
);
2573 /* Cached positive dentry: will open in f_op->open */
2574 if (!need_lookup
&& dentry
->d_inode
)
2577 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2578 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2579 need_lookup
, opened
);
2583 BUG_ON(dentry
->d_inode
);
2585 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2587 return PTR_ERR(dentry
);
2590 /* Negative dentry, just create the file */
2591 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2592 umode_t mode
= op
->mode
;
2593 if (!IS_POSIXACL(dir
->d_inode
))
2594 mode
&= ~current_umask();
2596 * This write is needed to ensure that a
2597 * rw->ro transition does not occur between
2598 * the time when the file is created and when
2599 * a permanent write count is taken through
2600 * the 'struct file' in finish_open().
2606 *opened
|= FILE_CREATED
;
2607 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2610 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2611 nd
->flags
& LOOKUP_EXCL
);
2616 path
->dentry
= dentry
;
2617 path
->mnt
= nd
->path
.mnt
;
2626 * Handle the last step of open()
2628 static int do_last(struct nameidata
*nd
, struct path
*path
,
2629 struct file
*file
, const struct open_flags
*op
,
2630 int *opened
, const char *pathname
)
2632 struct dentry
*dir
= nd
->path
.dentry
;
2633 int open_flag
= op
->open_flag
;
2634 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2635 bool got_write
= false;
2636 int acc_mode
= op
->acc_mode
;
2637 struct inode
*inode
;
2638 bool symlink_ok
= false;
2639 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2640 bool retried
= false;
2643 nd
->flags
&= ~LOOKUP_PARENT
;
2644 nd
->flags
|= op
->intent
;
2646 switch (nd
->last_type
) {
2649 error
= handle_dots(nd
, nd
->last_type
);
2654 error
= complete_walk(nd
);
2657 audit_inode(pathname
, nd
->path
.dentry
);
2658 if (open_flag
& O_CREAT
) {
2664 error
= complete_walk(nd
);
2667 audit_inode(pathname
, dir
);
2671 if (!(open_flag
& O_CREAT
)) {
2672 if (nd
->last
.name
[nd
->last
.len
])
2673 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2674 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2676 /* we _can_ be in RCU mode here */
2677 error
= lookup_fast(nd
, &nd
->last
, path
, &inode
);
2684 BUG_ON(nd
->inode
!= dir
->d_inode
);
2686 /* create side of things */
2688 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2689 * has been cleared when we got to the last component we are
2692 error
= complete_walk(nd
);
2696 audit_inode(pathname
, dir
);
2698 /* trailing slashes? */
2699 if (nd
->last
.name
[nd
->last
.len
])
2704 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2705 error
= mnt_want_write(nd
->path
.mnt
);
2709 * do _not_ fail yet - we might not need that or fail with
2710 * a different error; let lookup_open() decide; we'll be
2711 * dropping this one anyway.
2714 mutex_lock(&dir
->d_inode
->i_mutex
);
2715 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2716 mutex_unlock(&dir
->d_inode
->i_mutex
);
2722 if ((*opened
& FILE_CREATED
) ||
2723 !S_ISREG(file
->f_path
.dentry
->d_inode
->i_mode
))
2724 will_truncate
= false;
2726 audit_inode(pathname
, file
->f_path
.dentry
);
2730 if (*opened
& FILE_CREATED
) {
2731 /* Don't check for write permission, don't truncate */
2732 open_flag
&= ~O_TRUNC
;
2733 will_truncate
= false;
2734 acc_mode
= MAY_OPEN
;
2735 path_to_nameidata(path
, nd
);
2736 goto finish_open_created
;
2740 * create/update audit record if it already exists.
2742 if (path
->dentry
->d_inode
)
2743 audit_inode(pathname
, path
->dentry
);
2746 * If atomic_open() acquired write access it is dropped now due to
2747 * possible mount and symlink following (this might be optimized away if
2751 mnt_drop_write(nd
->path
.mnt
);
2756 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
2759 error
= follow_managed(path
, nd
->flags
);
2764 nd
->flags
|= LOOKUP_JUMPED
;
2766 BUG_ON(nd
->flags
& LOOKUP_RCU
);
2767 inode
= path
->dentry
->d_inode
;
2769 /* we _can_ be in RCU mode here */
2772 path_to_nameidata(path
, nd
);
2776 if (should_follow_link(inode
, !symlink_ok
)) {
2777 if (nd
->flags
& LOOKUP_RCU
) {
2778 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
2783 BUG_ON(inode
!= path
->dentry
->d_inode
);
2787 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
2788 path_to_nameidata(path
, nd
);
2790 save_parent
.dentry
= nd
->path
.dentry
;
2791 save_parent
.mnt
= mntget(path
->mnt
);
2792 nd
->path
.dentry
= path
->dentry
;
2796 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2797 error
= complete_walk(nd
);
2799 path_put(&save_parent
);
2803 if ((open_flag
& O_CREAT
) && S_ISDIR(nd
->inode
->i_mode
))
2806 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !nd
->inode
->i_op
->lookup
)
2808 audit_inode(pathname
, nd
->path
.dentry
);
2810 if (!S_ISREG(nd
->inode
->i_mode
))
2811 will_truncate
= false;
2813 if (will_truncate
) {
2814 error
= mnt_want_write(nd
->path
.mnt
);
2819 finish_open_created
:
2820 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2823 file
->f_path
.mnt
= nd
->path
.mnt
;
2824 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
2826 if (error
== -EOPENSTALE
)
2831 error
= open_check_o_direct(file
);
2834 error
= ima_file_check(file
, op
->acc_mode
);
2838 if (will_truncate
) {
2839 error
= handle_truncate(file
);
2845 mnt_drop_write(nd
->path
.mnt
);
2846 path_put(&save_parent
);
2851 path_put_conditional(path
, nd
);
2858 /* If no saved parent or already retried then can't retry */
2859 if (!save_parent
.dentry
|| retried
)
2862 BUG_ON(save_parent
.dentry
!= dir
);
2863 path_put(&nd
->path
);
2864 nd
->path
= save_parent
;
2865 nd
->inode
= dir
->d_inode
;
2866 save_parent
.mnt
= NULL
;
2867 save_parent
.dentry
= NULL
;
2869 mnt_drop_write(nd
->path
.mnt
);
2876 static struct file
*path_openat(int dfd
, const char *pathname
,
2877 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2879 struct file
*base
= NULL
;
2885 file
= get_empty_filp();
2887 return ERR_PTR(-ENFILE
);
2889 file
->f_flags
= op
->open_flag
;
2891 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2892 if (unlikely(error
))
2895 current
->total_link_count
= 0;
2896 error
= link_path_walk(pathname
, nd
);
2897 if (unlikely(error
))
2900 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2901 while (unlikely(error
> 0)) { /* trailing symlink */
2902 struct path link
= path
;
2904 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2905 path_put_conditional(&path
, nd
);
2906 path_put(&nd
->path
);
2910 error
= may_follow_link(&link
, nd
);
2911 if (unlikely(error
))
2913 nd
->flags
|= LOOKUP_PARENT
;
2914 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2915 error
= follow_link(&link
, nd
, &cookie
);
2916 if (unlikely(error
))
2918 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2919 put_link(nd
, &link
, cookie
);
2922 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2923 path_put(&nd
->root
);
2926 if (!(opened
& FILE_OPENED
)) {
2930 if (unlikely(error
)) {
2931 if (error
== -EOPENSTALE
) {
2932 if (flags
& LOOKUP_RCU
)
2937 file
= ERR_PTR(error
);
2942 struct file
*do_filp_open(int dfd
, const char *pathname
,
2943 const struct open_flags
*op
, int flags
)
2945 struct nameidata nd
;
2948 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2949 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2950 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2951 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2952 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2956 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2957 const char *name
, const struct open_flags
*op
, int flags
)
2959 struct nameidata nd
;
2963 nd
.root
.dentry
= dentry
;
2965 flags
|= LOOKUP_ROOT
;
2967 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2968 return ERR_PTR(-ELOOP
);
2970 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2971 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2972 file
= path_openat(-1, name
, &nd
, op
, flags
);
2973 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2974 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2978 struct dentry
*kern_path_create(int dfd
, const char *pathname
, struct path
*path
, int is_dir
)
2980 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2981 struct nameidata nd
;
2983 int error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
, &nd
);
2985 return ERR_PTR(error
);
2988 * Yucky last component or no last component at all?
2989 * (foo/., foo/.., /////)
2991 if (nd
.last_type
!= LAST_NORM
)
2993 nd
.flags
&= ~LOOKUP_PARENT
;
2994 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2996 /* don't fail immediately if it's r/o, at least try to report other errors */
2997 err2
= mnt_want_write(nd
.path
.mnt
);
2999 * Do the final lookup.
3001 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3002 dentry
= lookup_hash(&nd
);
3007 if (dentry
->d_inode
)
3010 * Special case - lookup gave negative, but... we had foo/bar/
3011 * From the vfs_mknod() POV we just have a negative dentry -
3012 * all is fine. Let's be bastards - you had / on the end, you've
3013 * been asking for (non-existent) directory. -ENOENT for you.
3015 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3019 if (unlikely(err2
)) {
3027 dentry
= ERR_PTR(error
);
3029 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3031 mnt_drop_write(nd
.path
.mnt
);
3036 EXPORT_SYMBOL(kern_path_create
);
3038 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3041 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3042 mnt_drop_write(path
->mnt
);
3045 EXPORT_SYMBOL(done_path_create
);
3047 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
, struct path
*path
, int is_dir
)
3049 char *tmp
= getname(pathname
);
3052 return ERR_CAST(tmp
);
3053 res
= kern_path_create(dfd
, tmp
, path
, is_dir
);
3057 EXPORT_SYMBOL(user_path_create
);
3059 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3061 int error
= may_create(dir
, dentry
);
3066 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3069 if (!dir
->i_op
->mknod
)
3072 error
= devcgroup_inode_mknod(mode
, dev
);
3076 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3080 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3082 fsnotify_create(dir
, dentry
);
3086 static int may_mknod(umode_t mode
)
3088 switch (mode
& S_IFMT
) {
3094 case 0: /* zero mode translates to S_IFREG */
3103 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3106 struct dentry
*dentry
;
3110 error
= may_mknod(mode
);
3114 dentry
= user_path_create(dfd
, filename
, &path
, 0);
3116 return PTR_ERR(dentry
);
3118 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3119 mode
&= ~current_umask();
3120 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3123 switch (mode
& S_IFMT
) {
3124 case 0: case S_IFREG
:
3125 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3127 case S_IFCHR
: case S_IFBLK
:
3128 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3129 new_decode_dev(dev
));
3131 case S_IFIFO
: case S_IFSOCK
:
3132 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3136 done_path_create(&path
, dentry
);
3140 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3142 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3145 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3147 int error
= may_create(dir
, dentry
);
3148 unsigned max_links
= dir
->i_sb
->s_max_links
;
3153 if (!dir
->i_op
->mkdir
)
3156 mode
&= (S_IRWXUGO
|S_ISVTX
);
3157 error
= security_inode_mkdir(dir
, dentry
, mode
);
3161 if (max_links
&& dir
->i_nlink
>= max_links
)
3164 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3166 fsnotify_mkdir(dir
, dentry
);
3170 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3172 struct dentry
*dentry
;
3176 dentry
= user_path_create(dfd
, pathname
, &path
, 1);
3178 return PTR_ERR(dentry
);
3180 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3181 mode
&= ~current_umask();
3182 error
= security_path_mkdir(&path
, dentry
, mode
);
3184 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3185 done_path_create(&path
, dentry
);
3189 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3191 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3195 * The dentry_unhash() helper will try to drop the dentry early: we
3196 * should have a usage count of 1 if we're the only user of this
3197 * dentry, and if that is true (possibly after pruning the dcache),
3198 * then we drop the dentry now.
3200 * A low-level filesystem can, if it choses, legally
3203 * if (!d_unhashed(dentry))
3206 * if it cannot handle the case of removing a directory
3207 * that is still in use by something else..
3209 void dentry_unhash(struct dentry
*dentry
)
3211 shrink_dcache_parent(dentry
);
3212 spin_lock(&dentry
->d_lock
);
3213 if (dentry
->d_count
== 1)
3215 spin_unlock(&dentry
->d_lock
);
3218 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3220 int error
= may_delete(dir
, dentry
, 1);
3225 if (!dir
->i_op
->rmdir
)
3229 mutex_lock(&dentry
->d_inode
->i_mutex
);
3232 if (d_mountpoint(dentry
))
3235 error
= security_inode_rmdir(dir
, dentry
);
3239 shrink_dcache_parent(dentry
);
3240 error
= dir
->i_op
->rmdir(dir
, dentry
);
3244 dentry
->d_inode
->i_flags
|= S_DEAD
;
3248 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3255 static long do_rmdir(int dfd
, const char __user
*pathname
)
3259 struct dentry
*dentry
;
3260 struct nameidata nd
;
3262 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
3266 switch(nd
.last_type
) {
3278 nd
.flags
&= ~LOOKUP_PARENT
;
3279 error
= mnt_want_write(nd
.path
.mnt
);
3283 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3284 dentry
= lookup_hash(&nd
);
3285 error
= PTR_ERR(dentry
);
3288 if (!dentry
->d_inode
) {
3292 error
= security_path_rmdir(&nd
.path
, dentry
);
3295 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3299 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3300 mnt_drop_write(nd
.path
.mnt
);
3307 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3309 return do_rmdir(AT_FDCWD
, pathname
);
3312 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3314 int error
= may_delete(dir
, dentry
, 0);
3319 if (!dir
->i_op
->unlink
)
3322 mutex_lock(&dentry
->d_inode
->i_mutex
);
3323 if (d_mountpoint(dentry
))
3326 error
= security_inode_unlink(dir
, dentry
);
3328 error
= dir
->i_op
->unlink(dir
, dentry
);
3333 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3335 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3336 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3337 fsnotify_link_count(dentry
->d_inode
);
3345 * Make sure that the actual truncation of the file will occur outside its
3346 * directory's i_mutex. Truncate can take a long time if there is a lot of
3347 * writeout happening, and we don't want to prevent access to the directory
3348 * while waiting on the I/O.
3350 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3354 struct dentry
*dentry
;
3355 struct nameidata nd
;
3356 struct inode
*inode
= NULL
;
3358 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
3363 if (nd
.last_type
!= LAST_NORM
)
3366 nd
.flags
&= ~LOOKUP_PARENT
;
3367 error
= mnt_want_write(nd
.path
.mnt
);
3371 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3372 dentry
= lookup_hash(&nd
);
3373 error
= PTR_ERR(dentry
);
3374 if (!IS_ERR(dentry
)) {
3375 /* Why not before? Because we want correct error value */
3376 if (nd
.last
.name
[nd
.last
.len
])
3378 inode
= dentry
->d_inode
;
3382 error
= security_path_unlink(&nd
.path
, dentry
);
3385 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
3389 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3391 iput(inode
); /* truncate the inode here */
3392 mnt_drop_write(nd
.path
.mnt
);
3399 error
= !dentry
->d_inode
? -ENOENT
:
3400 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
3404 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3406 if ((flag
& ~AT_REMOVEDIR
) != 0)
3409 if (flag
& AT_REMOVEDIR
)
3410 return do_rmdir(dfd
, pathname
);
3412 return do_unlinkat(dfd
, pathname
);
3415 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3417 return do_unlinkat(AT_FDCWD
, pathname
);
3420 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3422 int error
= may_create(dir
, dentry
);
3427 if (!dir
->i_op
->symlink
)
3430 error
= security_inode_symlink(dir
, dentry
, oldname
);
3434 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3436 fsnotify_create(dir
, dentry
);
3440 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3441 int, newdfd
, const char __user
*, newname
)
3445 struct dentry
*dentry
;
3448 from
= getname(oldname
);
3450 return PTR_ERR(from
);
3452 dentry
= user_path_create(newdfd
, newname
, &path
, 0);
3453 error
= PTR_ERR(dentry
);
3457 error
= security_path_symlink(&path
, dentry
, from
);
3459 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
);
3460 done_path_create(&path
, dentry
);
3466 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3468 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3471 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3473 struct inode
*inode
= old_dentry
->d_inode
;
3474 unsigned max_links
= dir
->i_sb
->s_max_links
;
3480 error
= may_create(dir
, new_dentry
);
3484 if (dir
->i_sb
!= inode
->i_sb
)
3488 * A link to an append-only or immutable file cannot be created.
3490 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3492 if (!dir
->i_op
->link
)
3494 if (S_ISDIR(inode
->i_mode
))
3497 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3501 mutex_lock(&inode
->i_mutex
);
3502 /* Make sure we don't allow creating hardlink to an unlinked file */
3503 if (inode
->i_nlink
== 0)
3505 else if (max_links
&& inode
->i_nlink
>= max_links
)
3508 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3509 mutex_unlock(&inode
->i_mutex
);
3511 fsnotify_link(dir
, inode
, new_dentry
);
3516 * Hardlinks are often used in delicate situations. We avoid
3517 * security-related surprises by not following symlinks on the
3520 * We don't follow them on the oldname either to be compatible
3521 * with linux 2.0, and to avoid hard-linking to directories
3522 * and other special files. --ADM
3524 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3525 int, newdfd
, const char __user
*, newname
, int, flags
)
3527 struct dentry
*new_dentry
;
3528 struct path old_path
, new_path
;
3532 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3535 * To use null names we require CAP_DAC_READ_SEARCH
3536 * This ensures that not everyone will be able to create
3537 * handlink using the passed filedescriptor.
3539 if (flags
& AT_EMPTY_PATH
) {
3540 if (!capable(CAP_DAC_READ_SEARCH
))
3545 if (flags
& AT_SYMLINK_FOLLOW
)
3546 how
|= LOOKUP_FOLLOW
;
3548 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3552 new_dentry
= user_path_create(newdfd
, newname
, &new_path
, 0);
3553 error
= PTR_ERR(new_dentry
);
3554 if (IS_ERR(new_dentry
))
3558 if (old_path
.mnt
!= new_path
.mnt
)
3560 error
= may_linkat(&old_path
);
3561 if (unlikely(error
))
3563 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3566 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3568 done_path_create(&new_path
, new_dentry
);
3570 path_put(&old_path
);
3575 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3577 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3581 * The worst of all namespace operations - renaming directory. "Perverted"
3582 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3584 * a) we can get into loop creation. Check is done in is_subdir().
3585 * b) race potential - two innocent renames can create a loop together.
3586 * That's where 4.4 screws up. Current fix: serialization on
3587 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3589 * c) we have to lock _three_ objects - parents and victim (if it exists).
3590 * And that - after we got ->i_mutex on parents (until then we don't know
3591 * whether the target exists). Solution: try to be smart with locking
3592 * order for inodes. We rely on the fact that tree topology may change
3593 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3594 * move will be locked. Thus we can rank directories by the tree
3595 * (ancestors first) and rank all non-directories after them.
3596 * That works since everybody except rename does "lock parent, lookup,
3597 * lock child" and rename is under ->s_vfs_rename_mutex.
3598 * HOWEVER, it relies on the assumption that any object with ->lookup()
3599 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3600 * we'd better make sure that there's no link(2) for them.
3601 * d) conversion from fhandle to dentry may come in the wrong moment - when
3602 * we are removing the target. Solution: we will have to grab ->i_mutex
3603 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3604 * ->i_mutex on parents, which works but leads to some truly excessive
3607 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3608 struct inode
*new_dir
, struct dentry
*new_dentry
)
3611 struct inode
*target
= new_dentry
->d_inode
;
3612 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3615 * If we are going to change the parent - check write permissions,
3616 * we'll need to flip '..'.
3618 if (new_dir
!= old_dir
) {
3619 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3624 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3630 mutex_lock(&target
->i_mutex
);
3633 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3637 if (max_links
&& !target
&& new_dir
!= old_dir
&&
3638 new_dir
->i_nlink
>= max_links
)
3642 shrink_dcache_parent(new_dentry
);
3643 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3648 target
->i_flags
|= S_DEAD
;
3649 dont_mount(new_dentry
);
3653 mutex_unlock(&target
->i_mutex
);
3656 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3657 d_move(old_dentry
,new_dentry
);
3661 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3662 struct inode
*new_dir
, struct dentry
*new_dentry
)
3664 struct inode
*target
= new_dentry
->d_inode
;
3667 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3673 mutex_lock(&target
->i_mutex
);
3676 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3679 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3684 dont_mount(new_dentry
);
3685 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3686 d_move(old_dentry
, new_dentry
);
3689 mutex_unlock(&target
->i_mutex
);
3694 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3695 struct inode
*new_dir
, struct dentry
*new_dentry
)
3698 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3699 const unsigned char *old_name
;
3701 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3704 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3708 if (!new_dentry
->d_inode
)
3709 error
= may_create(new_dir
, new_dentry
);
3711 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3715 if (!old_dir
->i_op
->rename
)
3718 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3721 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3723 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3725 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3726 new_dentry
->d_inode
, old_dentry
);
3727 fsnotify_oldname_free(old_name
);
3732 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3733 int, newdfd
, const char __user
*, newname
)
3735 struct dentry
*old_dir
, *new_dir
;
3736 struct dentry
*old_dentry
, *new_dentry
;
3737 struct dentry
*trap
;
3738 struct nameidata oldnd
, newnd
;
3743 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3747 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3752 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3755 old_dir
= oldnd
.path
.dentry
;
3757 if (oldnd
.last_type
!= LAST_NORM
)
3760 new_dir
= newnd
.path
.dentry
;
3761 if (newnd
.last_type
!= LAST_NORM
)
3764 error
= mnt_want_write(oldnd
.path
.mnt
);
3768 oldnd
.flags
&= ~LOOKUP_PARENT
;
3769 newnd
.flags
&= ~LOOKUP_PARENT
;
3770 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3772 trap
= lock_rename(new_dir
, old_dir
);
3774 old_dentry
= lookup_hash(&oldnd
);
3775 error
= PTR_ERR(old_dentry
);
3776 if (IS_ERR(old_dentry
))
3778 /* source must exist */
3780 if (!old_dentry
->d_inode
)
3782 /* unless the source is a directory trailing slashes give -ENOTDIR */
3783 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3785 if (oldnd
.last
.name
[oldnd
.last
.len
])
3787 if (newnd
.last
.name
[newnd
.last
.len
])
3790 /* source should not be ancestor of target */
3792 if (old_dentry
== trap
)
3794 new_dentry
= lookup_hash(&newnd
);
3795 error
= PTR_ERR(new_dentry
);
3796 if (IS_ERR(new_dentry
))
3798 /* target should not be an ancestor of source */
3800 if (new_dentry
== trap
)
3803 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3804 &newnd
.path
, new_dentry
);
3807 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3808 new_dir
->d_inode
, new_dentry
);
3814 unlock_rename(new_dir
, old_dir
);
3815 mnt_drop_write(oldnd
.path
.mnt
);
3817 path_put(&newnd
.path
);
3820 path_put(&oldnd
.path
);
3826 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3828 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3831 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3835 len
= PTR_ERR(link
);
3840 if (len
> (unsigned) buflen
)
3842 if (copy_to_user(buffer
, link
, len
))
3849 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3850 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3851 * using) it for any given inode is up to filesystem.
3853 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3855 struct nameidata nd
;
3860 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3862 return PTR_ERR(cookie
);
3864 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3865 if (dentry
->d_inode
->i_op
->put_link
)
3866 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3870 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3872 return __vfs_follow_link(nd
, link
);
3875 /* get the link contents into pagecache */
3876 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3880 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3881 page
= read_mapping_page(mapping
, 0, NULL
);
3886 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3890 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3892 struct page
*page
= NULL
;
3893 char *s
= page_getlink(dentry
, &page
);
3894 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3897 page_cache_release(page
);
3902 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3904 struct page
*page
= NULL
;
3905 nd_set_link(nd
, page_getlink(dentry
, &page
));
3909 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3911 struct page
*page
= cookie
;
3915 page_cache_release(page
);
3920 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3922 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3924 struct address_space
*mapping
= inode
->i_mapping
;
3929 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3931 flags
|= AOP_FLAG_NOFS
;
3934 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3935 flags
, &page
, &fsdata
);
3939 kaddr
= kmap_atomic(page
);
3940 memcpy(kaddr
, symname
, len
-1);
3941 kunmap_atomic(kaddr
);
3943 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3950 mark_inode_dirty(inode
);
3956 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3958 return __page_symlink(inode
, symname
, len
,
3959 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3962 const struct inode_operations page_symlink_inode_operations
= {
3963 .readlink
= generic_readlink
,
3964 .follow_link
= page_follow_link_light
,
3965 .put_link
= page_put_link
,
3968 EXPORT_SYMBOL(user_path_at
);
3969 EXPORT_SYMBOL(follow_down_one
);
3970 EXPORT_SYMBOL(follow_down
);
3971 EXPORT_SYMBOL(follow_up
);
3972 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3973 EXPORT_SYMBOL(getname
);
3974 EXPORT_SYMBOL(lock_rename
);
3975 EXPORT_SYMBOL(lookup_one_len
);
3976 EXPORT_SYMBOL(page_follow_link_light
);
3977 EXPORT_SYMBOL(page_put_link
);
3978 EXPORT_SYMBOL(page_readlink
);
3979 EXPORT_SYMBOL(__page_symlink
);
3980 EXPORT_SYMBOL(page_symlink
);
3981 EXPORT_SYMBOL(page_symlink_inode_operations
);
3982 EXPORT_SYMBOL(kern_path
);
3983 EXPORT_SYMBOL(vfs_path_lookup
);
3984 EXPORT_SYMBOL(inode_permission
);
3985 EXPORT_SYMBOL(unlock_rename
);
3986 EXPORT_SYMBOL(vfs_create
);
3987 EXPORT_SYMBOL(vfs_follow_link
);
3988 EXPORT_SYMBOL(vfs_link
);
3989 EXPORT_SYMBOL(vfs_mkdir
);
3990 EXPORT_SYMBOL(vfs_mknod
);
3991 EXPORT_SYMBOL(generic_permission
);
3992 EXPORT_SYMBOL(vfs_readlink
);
3993 EXPORT_SYMBOL(vfs_rename
);
3994 EXPORT_SYMBOL(vfs_rmdir
);
3995 EXPORT_SYMBOL(vfs_symlink
);
3996 EXPORT_SYMBOL(vfs_unlink
);
3997 EXPORT_SYMBOL(dentry_unhash
);
3998 EXPORT_SYMBOL(generic_readlink
);