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 void final_putname(struct filename
*name
)
122 if (name
->separate
) {
123 __putname(name
->name
);
130 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
132 static struct filename
*
133 getname_flags(const char __user
*filename
, int flags
, int *empty
)
135 struct filename
*result
, *err
;
140 result
= audit_reusename(filename
);
144 result
= __getname();
145 if (unlikely(!result
))
146 return ERR_PTR(-ENOMEM
);
149 * First, try to embed the struct filename inside the names_cache
152 kname
= (char *)result
+ sizeof(*result
);
153 result
->name
= kname
;
154 result
->separate
= false;
155 max
= EMBEDDED_NAME_MAX
;
158 len
= strncpy_from_user(kname
, filename
, max
);
159 if (unlikely(len
< 0)) {
165 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
166 * separate struct filename so we can dedicate the entire
167 * names_cache allocation for the pathname, and re-do the copy from
170 if (len
== EMBEDDED_NAME_MAX
&& max
== EMBEDDED_NAME_MAX
) {
171 kname
= (char *)result
;
173 result
= kzalloc(sizeof(*result
), GFP_KERNEL
);
175 err
= ERR_PTR(-ENOMEM
);
176 result
= (struct filename
*)kname
;
179 result
->name
= kname
;
180 result
->separate
= true;
185 /* The empty path is special. */
186 if (unlikely(!len
)) {
189 err
= ERR_PTR(-ENOENT
);
190 if (!(flags
& LOOKUP_EMPTY
))
194 err
= ERR_PTR(-ENAMETOOLONG
);
195 if (unlikely(len
>= PATH_MAX
))
198 result
->uptr
= filename
;
199 audit_getname(result
);
203 final_putname(result
);
208 getname(const char __user
* filename
)
210 return getname_flags(filename
, 0, NULL
);
212 EXPORT_SYMBOL(getname
);
214 #ifdef CONFIG_AUDITSYSCALL
215 void putname(struct filename
*name
)
217 if (unlikely(!audit_dummy_context()))
218 return audit_putname(name
);
223 static int check_acl(struct inode
*inode
, int mask
)
225 #ifdef CONFIG_FS_POSIX_ACL
226 struct posix_acl
*acl
;
228 if (mask
& MAY_NOT_BLOCK
) {
229 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
232 /* no ->get_acl() calls in RCU mode... */
233 if (acl
== ACL_NOT_CACHED
)
235 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
238 acl
= get_cached_acl(inode
, ACL_TYPE_ACCESS
);
241 * A filesystem can force a ACL callback by just never filling the
242 * ACL cache. But normally you'd fill the cache either at inode
243 * instantiation time, or on the first ->get_acl call.
245 * If the filesystem doesn't have a get_acl() function at all, we'll
246 * just create the negative cache entry.
248 if (acl
== ACL_NOT_CACHED
) {
249 if (inode
->i_op
->get_acl
) {
250 acl
= inode
->i_op
->get_acl(inode
, ACL_TYPE_ACCESS
);
254 set_cached_acl(inode
, ACL_TYPE_ACCESS
, NULL
);
260 int error
= posix_acl_permission(inode
, acl
, mask
);
261 posix_acl_release(acl
);
270 * This does the basic permission checking
272 static int acl_permission_check(struct inode
*inode
, int mask
)
274 unsigned int mode
= inode
->i_mode
;
276 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
279 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
280 int error
= check_acl(inode
, mask
);
281 if (error
!= -EAGAIN
)
285 if (in_group_p(inode
->i_gid
))
290 * If the DACs are ok we don't need any capability check.
292 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
298 * generic_permission - check for access rights on a Posix-like filesystem
299 * @inode: inode to check access rights for
300 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
302 * Used to check for read/write/execute permissions on a file.
303 * We use "fsuid" for this, letting us set arbitrary permissions
304 * for filesystem access without changing the "normal" uids which
305 * are used for other things.
307 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
308 * request cannot be satisfied (eg. requires blocking or too much complexity).
309 * It would then be called again in ref-walk mode.
311 int generic_permission(struct inode
*inode
, int mask
)
316 * Do the basic permission checks.
318 ret
= acl_permission_check(inode
, mask
);
322 if (S_ISDIR(inode
->i_mode
)) {
323 /* DACs are overridable for directories */
324 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
326 if (!(mask
& MAY_WRITE
))
327 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
332 * Read/write DACs are always overridable.
333 * Executable DACs are overridable when there is
334 * at least one exec bit set.
336 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
337 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
341 * Searching includes executable on directories, else just read.
343 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
344 if (mask
== MAY_READ
)
345 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
352 * We _really_ want to just do "generic_permission()" without
353 * even looking at the inode->i_op values. So we keep a cache
354 * flag in inode->i_opflags, that says "this has not special
355 * permission function, use the fast case".
357 static inline int do_inode_permission(struct inode
*inode
, int mask
)
359 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
360 if (likely(inode
->i_op
->permission
))
361 return inode
->i_op
->permission(inode
, mask
);
363 /* This gets set once for the inode lifetime */
364 spin_lock(&inode
->i_lock
);
365 inode
->i_opflags
|= IOP_FASTPERM
;
366 spin_unlock(&inode
->i_lock
);
368 return generic_permission(inode
, mask
);
372 * __inode_permission - Check for access rights to a given inode
373 * @inode: Inode to check permission on
374 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
376 * Check for read/write/execute permissions on an inode.
378 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
380 * This does not check for a read-only file system. You probably want
381 * inode_permission().
383 int __inode_permission(struct inode
*inode
, int mask
)
387 if (unlikely(mask
& MAY_WRITE
)) {
389 * Nobody gets write access to an immutable file.
391 if (IS_IMMUTABLE(inode
))
395 retval
= do_inode_permission(inode
, mask
);
399 retval
= devcgroup_inode_permission(inode
, mask
);
403 return security_inode_permission(inode
, mask
);
407 * sb_permission - Check superblock-level permissions
408 * @sb: Superblock of inode to check permission on
409 * @inode: Inode to check permission on
410 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
412 * Separate out file-system wide checks from inode-specific permission checks.
414 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
416 if (unlikely(mask
& MAY_WRITE
)) {
417 umode_t mode
= inode
->i_mode
;
419 /* Nobody gets write access to a read-only fs. */
420 if ((sb
->s_flags
& MS_RDONLY
) &&
421 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
428 * inode_permission - Check for access rights to a given inode
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
433 * this, letting us set arbitrary permissions for filesystem access without
434 * changing the "normal" UIDs which are used for other things.
436 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
438 int inode_permission(struct inode
*inode
, int mask
)
442 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
445 return __inode_permission(inode
, mask
);
449 * path_get - get a reference to a path
450 * @path: path to get the reference to
452 * Given a path increment the reference count to the dentry and the vfsmount.
454 void path_get(const struct path
*path
)
459 EXPORT_SYMBOL(path_get
);
462 * path_put - put a reference to a path
463 * @path: path to put the reference to
465 * Given a path decrement the reference count to the dentry and the vfsmount.
467 void path_put(const struct path
*path
)
472 EXPORT_SYMBOL(path_put
);
475 * Path walking has 2 modes, rcu-walk and ref-walk (see
476 * Documentation/filesystems/path-lookup.txt). In situations when we can't
477 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
478 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
479 * mode. Refcounts are grabbed at the last known good point before rcu-walk
480 * got stuck, so ref-walk may continue from there. If this is not successful
481 * (eg. a seqcount has changed), then failure is returned and it's up to caller
482 * to restart the path walk from the beginning in ref-walk mode.
485 static inline void lock_rcu_walk(void)
487 br_read_lock(&vfsmount_lock
);
491 static inline void unlock_rcu_walk(void)
494 br_read_unlock(&vfsmount_lock
);
498 * unlazy_walk - try to switch to ref-walk mode.
499 * @nd: nameidata pathwalk data
500 * @dentry: child of nd->path.dentry or NULL
501 * Returns: 0 on success, -ECHILD on failure
503 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
504 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
505 * @nd or NULL. Must be called from rcu-walk context.
507 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
509 struct fs_struct
*fs
= current
->fs
;
510 struct dentry
*parent
= nd
->path
.dentry
;
513 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
514 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
516 spin_lock(&fs
->lock
);
517 if (nd
->root
.mnt
!= fs
->root
.mnt
||
518 nd
->root
.dentry
!= fs
->root
.dentry
)
521 spin_lock(&parent
->d_lock
);
523 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
525 BUG_ON(nd
->inode
!= parent
->d_inode
);
527 if (dentry
->d_parent
!= parent
)
529 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
530 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
533 * If the sequence check on the child dentry passed, then
534 * the child has not been removed from its parent. This
535 * means the parent dentry must be valid and able to take
536 * a reference at this point.
538 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
539 BUG_ON(!parent
->d_count
);
541 spin_unlock(&dentry
->d_lock
);
543 spin_unlock(&parent
->d_lock
);
546 spin_unlock(&fs
->lock
);
548 mntget(nd
->path
.mnt
);
551 nd
->flags
&= ~LOOKUP_RCU
;
555 spin_unlock(&dentry
->d_lock
);
557 spin_unlock(&parent
->d_lock
);
560 spin_unlock(&fs
->lock
);
564 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
566 return dentry
->d_op
->d_revalidate(dentry
, flags
);
570 * complete_walk - successful completion of path walk
571 * @nd: pointer nameidata
573 * If we had been in RCU mode, drop out of it and legitimize nd->path.
574 * Revalidate the final result, unless we'd already done that during
575 * the path walk or the filesystem doesn't ask for it. Return 0 on
576 * success, -error on failure. In case of failure caller does not
577 * need to drop nd->path.
579 static int complete_walk(struct nameidata
*nd
)
581 struct dentry
*dentry
= nd
->path
.dentry
;
584 if (nd
->flags
& LOOKUP_RCU
) {
585 nd
->flags
&= ~LOOKUP_RCU
;
586 if (!(nd
->flags
& LOOKUP_ROOT
))
588 spin_lock(&dentry
->d_lock
);
589 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
590 spin_unlock(&dentry
->d_lock
);
594 BUG_ON(nd
->inode
!= dentry
->d_inode
);
595 spin_unlock(&dentry
->d_lock
);
596 mntget(nd
->path
.mnt
);
600 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
603 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
606 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
617 static __always_inline
void set_root(struct nameidata
*nd
)
620 get_fs_root(current
->fs
, &nd
->root
);
623 static int link_path_walk(const char *, struct nameidata
*);
625 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
628 struct fs_struct
*fs
= current
->fs
;
632 seq
= read_seqcount_begin(&fs
->seq
);
634 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
635 } while (read_seqcount_retry(&fs
->seq
, seq
));
639 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
651 nd
->flags
|= LOOKUP_JUMPED
;
653 nd
->inode
= nd
->path
.dentry
->d_inode
;
655 ret
= link_path_walk(link
, nd
);
659 return PTR_ERR(link
);
662 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
665 if (path
->mnt
!= nd
->path
.mnt
)
669 static inline void path_to_nameidata(const struct path
*path
,
670 struct nameidata
*nd
)
672 if (!(nd
->flags
& LOOKUP_RCU
)) {
673 dput(nd
->path
.dentry
);
674 if (nd
->path
.mnt
!= path
->mnt
)
675 mntput(nd
->path
.mnt
);
677 nd
->path
.mnt
= path
->mnt
;
678 nd
->path
.dentry
= path
->dentry
;
682 * Helper to directly jump to a known parsed path from ->follow_link,
683 * caller must have taken a reference to path beforehand.
685 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
690 nd
->inode
= nd
->path
.dentry
->d_inode
;
691 nd
->flags
|= LOOKUP_JUMPED
;
694 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
696 struct inode
*inode
= link
->dentry
->d_inode
;
697 if (inode
->i_op
->put_link
)
698 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
702 int sysctl_protected_symlinks __read_mostly
= 0;
703 int sysctl_protected_hardlinks __read_mostly
= 0;
706 * may_follow_link - Check symlink following for unsafe situations
707 * @link: The path of the symlink
708 * @nd: nameidata pathwalk data
710 * In the case of the sysctl_protected_symlinks sysctl being enabled,
711 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
712 * in a sticky world-writable directory. This is to protect privileged
713 * processes from failing races against path names that may change out
714 * from under them by way of other users creating malicious symlinks.
715 * It will permit symlinks to be followed only when outside a sticky
716 * world-writable directory, or when the uid of the symlink and follower
717 * match, or when the directory owner matches the symlink's owner.
719 * Returns 0 if following the symlink is allowed, -ve on error.
721 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
723 const struct inode
*inode
;
724 const struct inode
*parent
;
726 if (!sysctl_protected_symlinks
)
729 /* Allowed if owner and follower match. */
730 inode
= link
->dentry
->d_inode
;
731 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
734 /* Allowed if parent directory not sticky and world-writable. */
735 parent
= nd
->path
.dentry
->d_inode
;
736 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
739 /* Allowed if parent directory and link owner match. */
740 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
743 audit_log_link_denied("follow_link", link
);
744 path_put_conditional(link
, nd
);
750 * safe_hardlink_source - Check for safe hardlink conditions
751 * @inode: the source inode to hardlink from
753 * Return false if at least one of the following conditions:
754 * - inode is not a regular file
756 * - inode is setgid and group-exec
757 * - access failure for read and write
759 * Otherwise returns true.
761 static bool safe_hardlink_source(struct inode
*inode
)
763 umode_t mode
= inode
->i_mode
;
765 /* Special files should not get pinned to the filesystem. */
769 /* Setuid files should not get pinned to the filesystem. */
773 /* Executable setgid files should not get pinned to the filesystem. */
774 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
777 /* Hardlinking to unreadable or unwritable sources is dangerous. */
778 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
785 * may_linkat - Check permissions for creating a hardlink
786 * @link: the source to hardlink from
788 * Block hardlink when all of:
789 * - sysctl_protected_hardlinks enabled
790 * - fsuid does not match inode
791 * - hardlink source is unsafe (see safe_hardlink_source() above)
794 * Returns 0 if successful, -ve on error.
796 static int may_linkat(struct path
*link
)
798 const struct cred
*cred
;
801 if (!sysctl_protected_hardlinks
)
804 cred
= current_cred();
805 inode
= link
->dentry
->d_inode
;
807 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
808 * otherwise, it must be a safe source.
810 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
814 audit_log_link_denied("linkat", link
);
818 static __always_inline
int
819 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
821 struct dentry
*dentry
= link
->dentry
;
825 BUG_ON(nd
->flags
& LOOKUP_RCU
);
827 if (link
->mnt
== nd
->path
.mnt
)
831 if (unlikely(current
->total_link_count
>= 40))
832 goto out_put_nd_path
;
835 current
->total_link_count
++;
838 nd_set_link(nd
, NULL
);
840 error
= security_inode_follow_link(link
->dentry
, nd
);
842 goto out_put_nd_path
;
844 nd
->last_type
= LAST_BIND
;
845 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
848 goto out_put_nd_path
;
853 error
= __vfs_follow_link(nd
, s
);
855 put_link(nd
, link
, *p
);
867 static int follow_up_rcu(struct path
*path
)
869 struct mount
*mnt
= real_mount(path
->mnt
);
870 struct mount
*parent
;
871 struct dentry
*mountpoint
;
873 parent
= mnt
->mnt_parent
;
874 if (&parent
->mnt
== path
->mnt
)
876 mountpoint
= mnt
->mnt_mountpoint
;
877 path
->dentry
= mountpoint
;
878 path
->mnt
= &parent
->mnt
;
883 * follow_up - Find the mountpoint of path's vfsmount
885 * Given a path, find the mountpoint of its source file system.
886 * Replace @path with the path of the mountpoint in the parent mount.
889 * Return 1 if we went up a level and 0 if we were already at the
892 int follow_up(struct path
*path
)
894 struct mount
*mnt
= real_mount(path
->mnt
);
895 struct mount
*parent
;
896 struct dentry
*mountpoint
;
898 br_read_lock(&vfsmount_lock
);
899 parent
= mnt
->mnt_parent
;
901 br_read_unlock(&vfsmount_lock
);
904 mntget(&parent
->mnt
);
905 mountpoint
= dget(mnt
->mnt_mountpoint
);
906 br_read_unlock(&vfsmount_lock
);
908 path
->dentry
= mountpoint
;
910 path
->mnt
= &parent
->mnt
;
915 * Perform an automount
916 * - return -EISDIR to tell follow_managed() to stop and return the path we
919 static int follow_automount(struct path
*path
, unsigned flags
,
922 struct vfsmount
*mnt
;
925 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
928 /* We don't want to mount if someone's just doing a stat -
929 * unless they're stat'ing a directory and appended a '/' to
932 * We do, however, want to mount if someone wants to open or
933 * create a file of any type under the mountpoint, wants to
934 * traverse through the mountpoint or wants to open the
935 * mounted directory. Also, autofs may mark negative dentries
936 * as being automount points. These will need the attentions
937 * of the daemon to instantiate them before they can be used.
939 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
940 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
941 path
->dentry
->d_inode
)
944 current
->total_link_count
++;
945 if (current
->total_link_count
>= 40)
948 mnt
= path
->dentry
->d_op
->d_automount(path
);
951 * The filesystem is allowed to return -EISDIR here to indicate
952 * it doesn't want to automount. For instance, autofs would do
953 * this so that its userspace daemon can mount on this dentry.
955 * However, we can only permit this if it's a terminal point in
956 * the path being looked up; if it wasn't then the remainder of
957 * the path is inaccessible and we should say so.
959 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
964 if (!mnt
) /* mount collision */
968 /* lock_mount() may release path->mnt on error */
972 err
= finish_automount(mnt
, path
);
976 /* Someone else made a mount here whilst we were busy */
981 path
->dentry
= dget(mnt
->mnt_root
);
990 * Handle a dentry that is managed in some way.
991 * - Flagged for transit management (autofs)
992 * - Flagged as mountpoint
993 * - Flagged as automount point
995 * This may only be called in refwalk mode.
997 * Serialization is taken care of in namespace.c
999 static int follow_managed(struct path
*path
, unsigned flags
)
1001 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1003 bool need_mntput
= false;
1006 /* Given that we're not holding a lock here, we retain the value in a
1007 * local variable for each dentry as we look at it so that we don't see
1008 * the components of that value change under us */
1009 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1010 managed
&= DCACHE_MANAGED_DENTRY
,
1011 unlikely(managed
!= 0)) {
1012 /* Allow the filesystem to manage the transit without i_mutex
1014 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1015 BUG_ON(!path
->dentry
->d_op
);
1016 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1017 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1022 /* Transit to a mounted filesystem. */
1023 if (managed
& DCACHE_MOUNTED
) {
1024 struct vfsmount
*mounted
= lookup_mnt(path
);
1029 path
->mnt
= mounted
;
1030 path
->dentry
= dget(mounted
->mnt_root
);
1035 /* Something is mounted on this dentry in another
1036 * namespace and/or whatever was mounted there in this
1037 * namespace got unmounted before we managed to get the
1041 /* Handle an automount point */
1042 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1043 ret
= follow_automount(path
, flags
, &need_mntput
);
1049 /* We didn't change the current path point */
1053 if (need_mntput
&& path
->mnt
== mnt
)
1057 return ret
< 0 ? ret
: need_mntput
;
1060 int follow_down_one(struct path
*path
)
1062 struct vfsmount
*mounted
;
1064 mounted
= lookup_mnt(path
);
1068 path
->mnt
= mounted
;
1069 path
->dentry
= dget(mounted
->mnt_root
);
1075 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
1077 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
1078 dentry
->d_op
->d_manage(dentry
, true) < 0);
1082 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1083 * we meet a managed dentry that would need blocking.
1085 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1086 struct inode
**inode
)
1089 struct mount
*mounted
;
1091 * Don't forget we might have a non-mountpoint managed dentry
1092 * that wants to block transit.
1094 if (unlikely(managed_dentry_might_block(path
->dentry
)))
1097 if (!d_mountpoint(path
->dentry
))
1100 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1103 path
->mnt
= &mounted
->mnt
;
1104 path
->dentry
= mounted
->mnt
.mnt_root
;
1105 nd
->flags
|= LOOKUP_JUMPED
;
1106 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1108 * Update the inode too. We don't need to re-check the
1109 * dentry sequence number here after this d_inode read,
1110 * because a mount-point is always pinned.
1112 *inode
= path
->dentry
->d_inode
;
1117 static void follow_mount_rcu(struct nameidata
*nd
)
1119 while (d_mountpoint(nd
->path
.dentry
)) {
1120 struct mount
*mounted
;
1121 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
1124 nd
->path
.mnt
= &mounted
->mnt
;
1125 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1126 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1130 static int follow_dotdot_rcu(struct nameidata
*nd
)
1135 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1136 nd
->path
.mnt
== nd
->root
.mnt
) {
1139 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1140 struct dentry
*old
= nd
->path
.dentry
;
1141 struct dentry
*parent
= old
->d_parent
;
1144 seq
= read_seqcount_begin(&parent
->d_seq
);
1145 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1147 nd
->path
.dentry
= parent
;
1151 if (!follow_up_rcu(&nd
->path
))
1153 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1155 follow_mount_rcu(nd
);
1156 nd
->inode
= nd
->path
.dentry
->d_inode
;
1160 nd
->flags
&= ~LOOKUP_RCU
;
1161 if (!(nd
->flags
& LOOKUP_ROOT
))
1162 nd
->root
.mnt
= NULL
;
1168 * Follow down to the covering mount currently visible to userspace. At each
1169 * point, the filesystem owning that dentry may be queried as to whether the
1170 * caller is permitted to proceed or not.
1172 int follow_down(struct path
*path
)
1177 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1178 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1179 /* Allow the filesystem to manage the transit without i_mutex
1182 * We indicate to the filesystem if someone is trying to mount
1183 * something here. This gives autofs the chance to deny anyone
1184 * other than its daemon the right to mount on its
1187 * The filesystem may sleep at this point.
1189 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1190 BUG_ON(!path
->dentry
->d_op
);
1191 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1192 ret
= path
->dentry
->d_op
->d_manage(
1193 path
->dentry
, false);
1195 return ret
== -EISDIR
? 0 : ret
;
1198 /* Transit to a mounted filesystem. */
1199 if (managed
& DCACHE_MOUNTED
) {
1200 struct vfsmount
*mounted
= lookup_mnt(path
);
1205 path
->mnt
= mounted
;
1206 path
->dentry
= dget(mounted
->mnt_root
);
1210 /* Don't handle automount points here */
1217 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1219 static void follow_mount(struct path
*path
)
1221 while (d_mountpoint(path
->dentry
)) {
1222 struct vfsmount
*mounted
= lookup_mnt(path
);
1227 path
->mnt
= mounted
;
1228 path
->dentry
= dget(mounted
->mnt_root
);
1232 static void follow_dotdot(struct nameidata
*nd
)
1237 struct dentry
*old
= nd
->path
.dentry
;
1239 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1240 nd
->path
.mnt
== nd
->root
.mnt
) {
1243 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1244 /* rare case of legitimate dget_parent()... */
1245 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1249 if (!follow_up(&nd
->path
))
1252 follow_mount(&nd
->path
);
1253 nd
->inode
= nd
->path
.dentry
->d_inode
;
1257 * This looks up the name in dcache, possibly revalidates the old dentry and
1258 * allocates a new one if not found or not valid. In the need_lookup argument
1259 * returns whether i_op->lookup is necessary.
1261 * dir->d_inode->i_mutex must be held
1263 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1264 unsigned int flags
, bool *need_lookup
)
1266 struct dentry
*dentry
;
1269 *need_lookup
= false;
1270 dentry
= d_lookup(dir
, name
);
1272 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1273 error
= d_revalidate(dentry
, flags
);
1274 if (unlikely(error
<= 0)) {
1277 return ERR_PTR(error
);
1278 } else if (!d_invalidate(dentry
)) {
1287 dentry
= d_alloc(dir
, name
);
1288 if (unlikely(!dentry
))
1289 return ERR_PTR(-ENOMEM
);
1291 *need_lookup
= true;
1297 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1298 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1300 * dir->d_inode->i_mutex must be held
1302 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1307 /* Don't create child dentry for a dead directory. */
1308 if (unlikely(IS_DEADDIR(dir
))) {
1310 return ERR_PTR(-ENOENT
);
1313 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1314 if (unlikely(old
)) {
1321 static struct dentry
*__lookup_hash(struct qstr
*name
,
1322 struct dentry
*base
, unsigned int flags
)
1325 struct dentry
*dentry
;
1327 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1331 return lookup_real(base
->d_inode
, dentry
, flags
);
1335 * It's more convoluted than I'd like it to be, but... it's still fairly
1336 * small and for now I'd prefer to have fast path as straight as possible.
1337 * It _is_ time-critical.
1339 static int lookup_fast(struct nameidata
*nd
,
1340 struct path
*path
, struct inode
**inode
)
1342 struct vfsmount
*mnt
= nd
->path
.mnt
;
1343 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1349 * Rename seqlock is not required here because in the off chance
1350 * of a false negative due to a concurrent rename, we're going to
1351 * do the non-racy lookup, below.
1353 if (nd
->flags
& LOOKUP_RCU
) {
1355 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
, nd
->inode
);
1360 * This sequence count validates that the inode matches
1361 * the dentry name information from lookup.
1363 *inode
= dentry
->d_inode
;
1364 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1368 * This sequence count validates that the parent had no
1369 * changes while we did the lookup of the dentry above.
1371 * The memory barrier in read_seqcount_begin of child is
1372 * enough, we can use __read_seqcount_retry here.
1374 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1378 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1379 status
= d_revalidate(dentry
, nd
->flags
);
1380 if (unlikely(status
<= 0)) {
1381 if (status
!= -ECHILD
)
1387 path
->dentry
= dentry
;
1388 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1390 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1394 if (unlazy_walk(nd
, dentry
))
1397 dentry
= __d_lookup(parent
, &nd
->last
);
1400 if (unlikely(!dentry
))
1403 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1404 status
= d_revalidate(dentry
, nd
->flags
);
1405 if (unlikely(status
<= 0)) {
1410 if (!d_invalidate(dentry
)) {
1417 path
->dentry
= dentry
;
1418 err
= follow_managed(path
, nd
->flags
);
1419 if (unlikely(err
< 0)) {
1420 path_put_conditional(path
, nd
);
1424 nd
->flags
|= LOOKUP_JUMPED
;
1425 *inode
= path
->dentry
->d_inode
;
1432 /* Fast lookup failed, do it the slow way */
1433 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1435 struct dentry
*dentry
, *parent
;
1438 parent
= nd
->path
.dentry
;
1439 BUG_ON(nd
->inode
!= parent
->d_inode
);
1441 mutex_lock(&parent
->d_inode
->i_mutex
);
1442 dentry
= __lookup_hash(&nd
->last
, parent
, nd
->flags
);
1443 mutex_unlock(&parent
->d_inode
->i_mutex
);
1445 return PTR_ERR(dentry
);
1446 path
->mnt
= nd
->path
.mnt
;
1447 path
->dentry
= dentry
;
1448 err
= follow_managed(path
, nd
->flags
);
1449 if (unlikely(err
< 0)) {
1450 path_put_conditional(path
, nd
);
1454 nd
->flags
|= LOOKUP_JUMPED
;
1458 static inline int may_lookup(struct nameidata
*nd
)
1460 if (nd
->flags
& LOOKUP_RCU
) {
1461 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1464 if (unlazy_walk(nd
, NULL
))
1467 return inode_permission(nd
->inode
, MAY_EXEC
);
1470 static inline int handle_dots(struct nameidata
*nd
, int type
)
1472 if (type
== LAST_DOTDOT
) {
1473 if (nd
->flags
& LOOKUP_RCU
) {
1474 if (follow_dotdot_rcu(nd
))
1482 static void terminate_walk(struct nameidata
*nd
)
1484 if (!(nd
->flags
& LOOKUP_RCU
)) {
1485 path_put(&nd
->path
);
1487 nd
->flags
&= ~LOOKUP_RCU
;
1488 if (!(nd
->flags
& LOOKUP_ROOT
))
1489 nd
->root
.mnt
= NULL
;
1495 * Do we need to follow links? We _really_ want to be able
1496 * to do this check without having to look at inode->i_op,
1497 * so we keep a cache of "no, this doesn't need follow_link"
1498 * for the common case.
1500 static inline int should_follow_link(struct inode
*inode
, int follow
)
1502 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1503 if (likely(inode
->i_op
->follow_link
))
1506 /* This gets set once for the inode lifetime */
1507 spin_lock(&inode
->i_lock
);
1508 inode
->i_opflags
|= IOP_NOFOLLOW
;
1509 spin_unlock(&inode
->i_lock
);
1514 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1517 struct inode
*inode
;
1520 * "." and ".." are special - ".." especially so because it has
1521 * to be able to know about the current root directory and
1522 * parent relationships.
1524 if (unlikely(nd
->last_type
!= LAST_NORM
))
1525 return handle_dots(nd
, nd
->last_type
);
1526 err
= lookup_fast(nd
, path
, &inode
);
1527 if (unlikely(err
)) {
1531 err
= lookup_slow(nd
, path
);
1535 inode
= path
->dentry
->d_inode
;
1541 if (should_follow_link(inode
, follow
)) {
1542 if (nd
->flags
& LOOKUP_RCU
) {
1543 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1548 BUG_ON(inode
!= path
->dentry
->d_inode
);
1551 path_to_nameidata(path
, nd
);
1556 path_to_nameidata(path
, nd
);
1563 * This limits recursive symlink follows to 8, while
1564 * limiting consecutive symlinks to 40.
1566 * Without that kind of total limit, nasty chains of consecutive
1567 * symlinks can cause almost arbitrarily long lookups.
1569 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1573 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1574 path_put_conditional(path
, nd
);
1575 path_put(&nd
->path
);
1578 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1581 current
->link_count
++;
1584 struct path link
= *path
;
1587 res
= follow_link(&link
, nd
, &cookie
);
1590 res
= walk_component(nd
, path
, LOOKUP_FOLLOW
);
1591 put_link(nd
, &link
, cookie
);
1594 current
->link_count
--;
1600 * We really don't want to look at inode->i_op->lookup
1601 * when we don't have to. So we keep a cache bit in
1602 * the inode ->i_opflags field that says "yes, we can
1603 * do lookup on this inode".
1605 static inline int can_lookup(struct inode
*inode
)
1607 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1609 if (likely(!inode
->i_op
->lookup
))
1612 /* We do this once for the lifetime of the inode */
1613 spin_lock(&inode
->i_lock
);
1614 inode
->i_opflags
|= IOP_LOOKUP
;
1615 spin_unlock(&inode
->i_lock
);
1620 * We can do the critical dentry name comparison and hashing
1621 * operations one word at a time, but we are limited to:
1623 * - Architectures with fast unaligned word accesses. We could
1624 * do a "get_unaligned()" if this helps and is sufficiently
1627 * - Little-endian machines (so that we can generate the mask
1628 * of low bytes efficiently). Again, we *could* do a byte
1629 * swapping load on big-endian architectures if that is not
1630 * expensive enough to make the optimization worthless.
1632 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1633 * do not trap on the (extremely unlikely) case of a page
1634 * crossing operation.
1636 * - Furthermore, we need an efficient 64-bit compile for the
1637 * 64-bit case in order to generate the "number of bytes in
1638 * the final mask". Again, that could be replaced with a
1639 * efficient population count instruction or similar.
1641 #ifdef CONFIG_DCACHE_WORD_ACCESS
1643 #include <asm/word-at-a-time.h>
1647 static inline unsigned int fold_hash(unsigned long hash
)
1649 hash
+= hash
>> (8*sizeof(int));
1653 #else /* 32-bit case */
1655 #define fold_hash(x) (x)
1659 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1661 unsigned long a
, mask
;
1662 unsigned long hash
= 0;
1665 a
= load_unaligned_zeropad(name
);
1666 if (len
< sizeof(unsigned long))
1670 name
+= sizeof(unsigned long);
1671 len
-= sizeof(unsigned long);
1675 mask
= ~(~0ul << len
*8);
1678 return fold_hash(hash
);
1680 EXPORT_SYMBOL(full_name_hash
);
1683 * Calculate the length and hash of the path component, and
1684 * return the length of the component;
1686 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1688 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1689 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1692 len
= -sizeof(unsigned long);
1694 hash
= (hash
+ a
) * 9;
1695 len
+= sizeof(unsigned long);
1696 a
= load_unaligned_zeropad(name
+len
);
1697 b
= a
^ REPEAT_BYTE('/');
1698 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1700 adata
= prep_zero_mask(a
, adata
, &constants
);
1701 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1703 mask
= create_zero_mask(adata
| bdata
);
1705 hash
+= a
& zero_bytemask(mask
);
1706 *hashp
= fold_hash(hash
);
1708 return len
+ find_zero(mask
);
1713 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1715 unsigned long hash
= init_name_hash();
1717 hash
= partial_name_hash(*name
++, hash
);
1718 return end_name_hash(hash
);
1720 EXPORT_SYMBOL(full_name_hash
);
1723 * We know there's a real path component here of at least
1726 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1728 unsigned long hash
= init_name_hash();
1729 unsigned long len
= 0, c
;
1731 c
= (unsigned char)*name
;
1734 hash
= partial_name_hash(c
, hash
);
1735 c
= (unsigned char)name
[len
];
1736 } while (c
&& c
!= '/');
1737 *hashp
= end_name_hash(hash
);
1745 * This is the basic name resolution function, turning a pathname into
1746 * the final dentry. We expect 'base' to be positive and a directory.
1748 * Returns 0 and nd will have valid dentry and mnt on success.
1749 * Returns error and drops reference to input namei data on failure.
1751 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1761 /* At this point we know we have a real path component. */
1767 err
= may_lookup(nd
);
1771 len
= hash_name(name
, &this.hash
);
1776 if (name
[0] == '.') switch (len
) {
1778 if (name
[1] == '.') {
1780 nd
->flags
|= LOOKUP_JUMPED
;
1786 if (likely(type
== LAST_NORM
)) {
1787 struct dentry
*parent
= nd
->path
.dentry
;
1788 nd
->flags
&= ~LOOKUP_JUMPED
;
1789 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1790 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1798 nd
->last_type
= type
;
1803 * If it wasn't NUL, we know it was '/'. Skip that
1804 * slash, and continue until no more slashes.
1808 } while (unlikely(name
[len
] == '/'));
1814 err
= walk_component(nd
, &next
, LOOKUP_FOLLOW
);
1819 err
= nested_symlink(&next
, nd
);
1823 if (!can_lookup(nd
->inode
)) {
1832 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1833 struct nameidata
*nd
, struct file
**fp
)
1837 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1838 nd
->flags
= flags
| LOOKUP_JUMPED
;
1840 if (flags
& LOOKUP_ROOT
) {
1841 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1843 if (!can_lookup(inode
))
1845 retval
= inode_permission(inode
, MAY_EXEC
);
1849 nd
->path
= nd
->root
;
1851 if (flags
& LOOKUP_RCU
) {
1853 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1855 path_get(&nd
->path
);
1860 nd
->root
.mnt
= NULL
;
1863 if (flags
& LOOKUP_RCU
) {
1868 path_get(&nd
->root
);
1870 nd
->path
= nd
->root
;
1871 } else if (dfd
== AT_FDCWD
) {
1872 if (flags
& LOOKUP_RCU
) {
1873 struct fs_struct
*fs
= current
->fs
;
1879 seq
= read_seqcount_begin(&fs
->seq
);
1881 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1882 } while (read_seqcount_retry(&fs
->seq
, seq
));
1884 get_fs_pwd(current
->fs
, &nd
->path
);
1887 /* Caller must check execute permissions on the starting path component */
1888 struct fd f
= fdget_raw(dfd
);
1889 struct dentry
*dentry
;
1894 dentry
= f
.file
->f_path
.dentry
;
1897 if (!can_lookup(dentry
->d_inode
)) {
1903 nd
->path
= f
.file
->f_path
;
1904 if (flags
& LOOKUP_RCU
) {
1907 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1910 path_get(&nd
->path
);
1915 nd
->inode
= nd
->path
.dentry
->d_inode
;
1919 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1921 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1922 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1924 nd
->flags
&= ~LOOKUP_PARENT
;
1925 return walk_component(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
);
1928 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1929 static int path_lookupat(int dfd
, const char *name
,
1930 unsigned int flags
, struct nameidata
*nd
)
1932 struct file
*base
= NULL
;
1937 * Path walking is largely split up into 2 different synchronisation
1938 * schemes, rcu-walk and ref-walk (explained in
1939 * Documentation/filesystems/path-lookup.txt). These share much of the
1940 * path walk code, but some things particularly setup, cleanup, and
1941 * following mounts are sufficiently divergent that functions are
1942 * duplicated. Typically there is a function foo(), and its RCU
1943 * analogue, foo_rcu().
1945 * -ECHILD is the error number of choice (just to avoid clashes) that
1946 * is returned if some aspect of an rcu-walk fails. Such an error must
1947 * be handled by restarting a traditional ref-walk (which will always
1948 * be able to complete).
1950 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1955 current
->total_link_count
= 0;
1956 err
= link_path_walk(name
, nd
);
1958 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1959 err
= lookup_last(nd
, &path
);
1962 struct path link
= path
;
1963 err
= may_follow_link(&link
, nd
);
1966 nd
->flags
|= LOOKUP_PARENT
;
1967 err
= follow_link(&link
, nd
, &cookie
);
1970 err
= lookup_last(nd
, &path
);
1971 put_link(nd
, &link
, cookie
);
1976 err
= complete_walk(nd
);
1978 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1979 if (!can_lookup(nd
->inode
)) {
1980 path_put(&nd
->path
);
1988 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1989 path_put(&nd
->root
);
1990 nd
->root
.mnt
= NULL
;
1995 static int filename_lookup(int dfd
, struct filename
*name
,
1996 unsigned int flags
, struct nameidata
*nd
)
1998 int retval
= path_lookupat(dfd
, name
->name
, flags
| LOOKUP_RCU
, nd
);
1999 if (unlikely(retval
== -ECHILD
))
2000 retval
= path_lookupat(dfd
, name
->name
, flags
, nd
);
2001 if (unlikely(retval
== -ESTALE
))
2002 retval
= path_lookupat(dfd
, name
->name
,
2003 flags
| LOOKUP_REVAL
, nd
);
2005 if (likely(!retval
))
2006 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2010 static int do_path_lookup(int dfd
, const char *name
,
2011 unsigned int flags
, struct nameidata
*nd
)
2013 struct filename filename
= { .name
= name
};
2015 return filename_lookup(dfd
, &filename
, flags
, nd
);
2018 /* does lookup, returns the object with parent locked */
2019 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2021 struct nameidata nd
;
2023 int err
= do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, &nd
);
2025 return ERR_PTR(err
);
2026 if (nd
.last_type
!= LAST_NORM
) {
2028 return ERR_PTR(-EINVAL
);
2030 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2031 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2033 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2041 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2043 struct nameidata nd
;
2044 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
2051 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2052 * @dentry: pointer to dentry of the base directory
2053 * @mnt: pointer to vfs mount of the base directory
2054 * @name: pointer to file name
2055 * @flags: lookup flags
2056 * @path: pointer to struct path to fill
2058 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2059 const char *name
, unsigned int flags
,
2062 struct nameidata nd
;
2064 nd
.root
.dentry
= dentry
;
2066 BUG_ON(flags
& LOOKUP_PARENT
);
2067 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2068 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
2075 * Restricted form of lookup. Doesn't follow links, single-component only,
2076 * needs parent already locked. Doesn't follow mounts.
2079 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2081 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2085 * lookup_one_len - filesystem helper to lookup single pathname component
2086 * @name: pathname component to lookup
2087 * @base: base directory to lookup from
2088 * @len: maximum length @len should be interpreted to
2090 * Note that this routine is purely a helper for filesystem usage and should
2091 * not be called by generic code. Also note that by using this function the
2092 * nameidata argument is passed to the filesystem methods and a filesystem
2093 * using this helper needs to be prepared for that.
2095 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2101 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2105 this.hash
= full_name_hash(name
, len
);
2107 return ERR_PTR(-EACCES
);
2109 if (unlikely(name
[0] == '.')) {
2110 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2111 return ERR_PTR(-EACCES
);
2115 c
= *(const unsigned char *)name
++;
2116 if (c
== '/' || c
== '\0')
2117 return ERR_PTR(-EACCES
);
2120 * See if the low-level filesystem might want
2121 * to use its own hash..
2123 if (base
->d_flags
& DCACHE_OP_HASH
) {
2124 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
2126 return ERR_PTR(err
);
2129 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2131 return ERR_PTR(err
);
2133 return __lookup_hash(&this, base
, 0);
2136 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2137 struct path
*path
, int *empty
)
2139 struct nameidata nd
;
2140 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2141 int err
= PTR_ERR(tmp
);
2144 BUG_ON(flags
& LOOKUP_PARENT
);
2146 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2154 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2157 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2161 * NB: most callers don't do anything directly with the reference to the
2162 * to struct filename, but the nd->last pointer points into the name string
2163 * allocated by getname. So we must hold the reference to it until all
2164 * path-walking is complete.
2166 static struct filename
*
2167 user_path_parent(int dfd
, const char __user
*path
, struct nameidata
*nd
,
2170 struct filename
*s
= getname(path
);
2173 /* only LOOKUP_REVAL is allowed in extra flags */
2174 flags
&= LOOKUP_REVAL
;
2179 error
= filename_lookup(dfd
, s
, flags
| LOOKUP_PARENT
, nd
);
2182 return ERR_PTR(error
);
2189 * It's inline, so penalty for filesystems that don't use sticky bit is
2192 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
2194 kuid_t fsuid
= current_fsuid();
2196 if (!(dir
->i_mode
& S_ISVTX
))
2198 if (uid_eq(inode
->i_uid
, fsuid
))
2200 if (uid_eq(dir
->i_uid
, fsuid
))
2202 return !inode_capable(inode
, CAP_FOWNER
);
2206 * Check whether we can remove a link victim from directory dir, check
2207 * whether the type of victim is right.
2208 * 1. We can't do it if dir is read-only (done in permission())
2209 * 2. We should have write and exec permissions on dir
2210 * 3. We can't remove anything from append-only dir
2211 * 4. We can't do anything with immutable dir (done in permission())
2212 * 5. If the sticky bit on dir is set we should either
2213 * a. be owner of dir, or
2214 * b. be owner of victim, or
2215 * c. have CAP_FOWNER capability
2216 * 6. If the victim is append-only or immutable we can't do antyhing with
2217 * links pointing to it.
2218 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2219 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2220 * 9. We can't remove a root or mountpoint.
2221 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2222 * nfs_async_unlink().
2224 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
2228 if (!victim
->d_inode
)
2231 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2232 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2234 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2239 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
2240 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
2243 if (!S_ISDIR(victim
->d_inode
->i_mode
))
2245 if (IS_ROOT(victim
))
2247 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
2249 if (IS_DEADDIR(dir
))
2251 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2256 /* Check whether we can create an object with dentry child in directory
2258 * 1. We can't do it if child already exists (open has special treatment for
2259 * this case, but since we are inlined it's OK)
2260 * 2. We can't do it if dir is read-only (done in permission())
2261 * 3. We should have write and exec permissions on dir
2262 * 4. We can't do it if dir is immutable (done in permission())
2264 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2266 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2269 if (IS_DEADDIR(dir
))
2271 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2275 * p1 and p2 should be directories on the same fs.
2277 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2282 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2286 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2288 p
= d_ancestor(p2
, p1
);
2290 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2291 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2295 p
= d_ancestor(p1
, p2
);
2297 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2298 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2302 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2303 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2307 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2309 mutex_unlock(&p1
->d_inode
->i_mutex
);
2311 mutex_unlock(&p2
->d_inode
->i_mutex
);
2312 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2316 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2319 int error
= may_create(dir
, dentry
);
2323 if (!dir
->i_op
->create
)
2324 return -EACCES
; /* shouldn't it be ENOSYS? */
2327 error
= security_inode_create(dir
, dentry
, mode
);
2330 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2332 fsnotify_create(dir
, dentry
);
2336 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2338 struct dentry
*dentry
= path
->dentry
;
2339 struct inode
*inode
= dentry
->d_inode
;
2349 switch (inode
->i_mode
& S_IFMT
) {
2353 if (acc_mode
& MAY_WRITE
)
2358 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2367 error
= inode_permission(inode
, acc_mode
);
2372 * An append-only file must be opened in append mode for writing.
2374 if (IS_APPEND(inode
)) {
2375 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2381 /* O_NOATIME can only be set by the owner or superuser */
2382 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2388 static int handle_truncate(struct file
*filp
)
2390 struct path
*path
= &filp
->f_path
;
2391 struct inode
*inode
= path
->dentry
->d_inode
;
2392 int error
= get_write_access(inode
);
2396 * Refuse to truncate files with mandatory locks held on them.
2398 error
= locks_verify_locked(inode
);
2400 error
= security_path_truncate(path
);
2402 error
= do_truncate(path
->dentry
, 0,
2403 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2406 put_write_access(inode
);
2410 static inline int open_to_namei_flags(int flag
)
2412 if ((flag
& O_ACCMODE
) == 3)
2417 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2419 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2423 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2427 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2431 * Attempt to atomically look up, create and open a file from a negative
2434 * Returns 0 if successful. The file will have been created and attached to
2435 * @file by the filesystem calling finish_open().
2437 * Returns 1 if the file was looked up only or didn't need creating. The
2438 * caller will need to perform the open themselves. @path will have been
2439 * updated to point to the new dentry. This may be negative.
2441 * Returns an error code otherwise.
2443 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2444 struct path
*path
, struct file
*file
,
2445 const struct open_flags
*op
,
2446 bool got_write
, bool need_lookup
,
2449 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2450 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2454 int create_error
= 0;
2455 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2457 BUG_ON(dentry
->d_inode
);
2459 /* Don't create child dentry for a dead directory. */
2460 if (unlikely(IS_DEADDIR(dir
))) {
2466 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2467 mode
&= ~current_umask();
2469 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
)) {
2470 open_flag
&= ~O_TRUNC
;
2471 *opened
|= FILE_CREATED
;
2475 * Checking write permission is tricky, bacuse we don't know if we are
2476 * going to actually need it: O_CREAT opens should work as long as the
2477 * file exists. But checking existence breaks atomicity. The trick is
2478 * to check access and if not granted clear O_CREAT from the flags.
2480 * Another problem is returing the "right" error value (e.g. for an
2481 * O_EXCL open we want to return EEXIST not EROFS).
2483 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2484 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2485 if (!(open_flag
& O_CREAT
)) {
2487 * No O_CREATE -> atomicity not a requirement -> fall
2488 * back to lookup + open
2491 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2492 /* Fall back and fail with the right error */
2493 create_error
= -EROFS
;
2496 /* No side effects, safe to clear O_CREAT */
2497 create_error
= -EROFS
;
2498 open_flag
&= ~O_CREAT
;
2502 if (open_flag
& O_CREAT
) {
2503 error
= may_o_create(&nd
->path
, dentry
, mode
);
2505 create_error
= error
;
2506 if (open_flag
& O_EXCL
)
2508 open_flag
&= ~O_CREAT
;
2512 if (nd
->flags
& LOOKUP_DIRECTORY
)
2513 open_flag
|= O_DIRECTORY
;
2515 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2516 file
->f_path
.mnt
= nd
->path
.mnt
;
2517 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2520 if (create_error
&& error
== -ENOENT
)
2521 error
= create_error
;
2525 acc_mode
= op
->acc_mode
;
2526 if (*opened
& FILE_CREATED
) {
2527 fsnotify_create(dir
, dentry
);
2528 acc_mode
= MAY_OPEN
;
2531 if (error
) { /* returned 1, that is */
2532 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2536 if (file
->f_path
.dentry
) {
2538 dentry
= file
->f_path
.dentry
;
2540 if (create_error
&& dentry
->d_inode
== NULL
) {
2541 error
= create_error
;
2548 * We didn't have the inode before the open, so check open permission
2551 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2561 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2563 return PTR_ERR(dentry
);
2566 int open_flag
= op
->open_flag
;
2568 error
= create_error
;
2569 if ((open_flag
& O_EXCL
)) {
2570 if (!dentry
->d_inode
)
2572 } else if (!dentry
->d_inode
) {
2574 } else if ((open_flag
& O_TRUNC
) &&
2575 S_ISREG(dentry
->d_inode
->i_mode
)) {
2578 /* will fail later, go on to get the right error */
2582 path
->dentry
= dentry
;
2583 path
->mnt
= nd
->path
.mnt
;
2588 * Look up and maybe create and open the last component.
2590 * Must be called with i_mutex held on parent.
2592 * Returns 0 if the file was successfully atomically created (if necessary) and
2593 * opened. In this case the file will be returned attached to @file.
2595 * Returns 1 if the file was not completely opened at this time, though lookups
2596 * and creations will have been performed and the dentry returned in @path will
2597 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2598 * specified then a negative dentry may be returned.
2600 * An error code is returned otherwise.
2602 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2603 * cleared otherwise prior to returning.
2605 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2607 const struct open_flags
*op
,
2608 bool got_write
, int *opened
)
2610 struct dentry
*dir
= nd
->path
.dentry
;
2611 struct inode
*dir_inode
= dir
->d_inode
;
2612 struct dentry
*dentry
;
2616 *opened
&= ~FILE_CREATED
;
2617 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2619 return PTR_ERR(dentry
);
2621 /* Cached positive dentry: will open in f_op->open */
2622 if (!need_lookup
&& dentry
->d_inode
)
2625 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2626 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2627 need_lookup
, opened
);
2631 BUG_ON(dentry
->d_inode
);
2633 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2635 return PTR_ERR(dentry
);
2638 /* Negative dentry, just create the file */
2639 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2640 umode_t mode
= op
->mode
;
2641 if (!IS_POSIXACL(dir
->d_inode
))
2642 mode
&= ~current_umask();
2644 * This write is needed to ensure that a
2645 * rw->ro transition does not occur between
2646 * the time when the file is created and when
2647 * a permanent write count is taken through
2648 * the 'struct file' in finish_open().
2654 *opened
|= FILE_CREATED
;
2655 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2658 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2659 nd
->flags
& LOOKUP_EXCL
);
2664 path
->dentry
= dentry
;
2665 path
->mnt
= nd
->path
.mnt
;
2674 * Handle the last step of open()
2676 static int do_last(struct nameidata
*nd
, struct path
*path
,
2677 struct file
*file
, const struct open_flags
*op
,
2678 int *opened
, struct filename
*name
)
2680 struct dentry
*dir
= nd
->path
.dentry
;
2681 int open_flag
= op
->open_flag
;
2682 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2683 bool got_write
= false;
2684 int acc_mode
= op
->acc_mode
;
2685 struct inode
*inode
;
2686 bool symlink_ok
= false;
2687 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2688 bool retried
= false;
2691 nd
->flags
&= ~LOOKUP_PARENT
;
2692 nd
->flags
|= op
->intent
;
2694 switch (nd
->last_type
) {
2697 error
= handle_dots(nd
, nd
->last_type
);
2702 error
= complete_walk(nd
);
2705 audit_inode(name
, nd
->path
.dentry
, 0);
2706 if (open_flag
& O_CREAT
) {
2712 error
= complete_walk(nd
);
2715 audit_inode(name
, dir
, 0);
2719 if (!(open_flag
& O_CREAT
)) {
2720 if (nd
->last
.name
[nd
->last
.len
])
2721 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2722 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2724 /* we _can_ be in RCU mode here */
2725 error
= lookup_fast(nd
, path
, &inode
);
2732 BUG_ON(nd
->inode
!= dir
->d_inode
);
2734 /* create side of things */
2736 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2737 * has been cleared when we got to the last component we are
2740 error
= complete_walk(nd
);
2744 audit_inode(name
, dir
, LOOKUP_PARENT
);
2746 /* trailing slashes? */
2747 if (nd
->last
.name
[nd
->last
.len
])
2752 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2753 error
= mnt_want_write(nd
->path
.mnt
);
2757 * do _not_ fail yet - we might not need that or fail with
2758 * a different error; let lookup_open() decide; we'll be
2759 * dropping this one anyway.
2762 mutex_lock(&dir
->d_inode
->i_mutex
);
2763 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2764 mutex_unlock(&dir
->d_inode
->i_mutex
);
2770 if ((*opened
& FILE_CREATED
) ||
2771 !S_ISREG(file_inode(file
)->i_mode
))
2772 will_truncate
= false;
2774 audit_inode(name
, file
->f_path
.dentry
, 0);
2778 if (*opened
& FILE_CREATED
) {
2779 /* Don't check for write permission, don't truncate */
2780 open_flag
&= ~O_TRUNC
;
2781 will_truncate
= false;
2782 acc_mode
= MAY_OPEN
;
2783 path_to_nameidata(path
, nd
);
2784 goto finish_open_created
;
2788 * create/update audit record if it already exists.
2790 if (path
->dentry
->d_inode
)
2791 audit_inode(name
, path
->dentry
, 0);
2794 * If atomic_open() acquired write access it is dropped now due to
2795 * possible mount and symlink following (this might be optimized away if
2799 mnt_drop_write(nd
->path
.mnt
);
2804 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
2807 error
= follow_managed(path
, nd
->flags
);
2812 nd
->flags
|= LOOKUP_JUMPED
;
2814 BUG_ON(nd
->flags
& LOOKUP_RCU
);
2815 inode
= path
->dentry
->d_inode
;
2817 /* we _can_ be in RCU mode here */
2820 path_to_nameidata(path
, nd
);
2824 if (should_follow_link(inode
, !symlink_ok
)) {
2825 if (nd
->flags
& LOOKUP_RCU
) {
2826 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
2831 BUG_ON(inode
!= path
->dentry
->d_inode
);
2835 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
2836 path_to_nameidata(path
, nd
);
2838 save_parent
.dentry
= nd
->path
.dentry
;
2839 save_parent
.mnt
= mntget(path
->mnt
);
2840 nd
->path
.dentry
= path
->dentry
;
2844 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2845 error
= complete_walk(nd
);
2847 path_put(&save_parent
);
2851 if ((open_flag
& O_CREAT
) && S_ISDIR(nd
->inode
->i_mode
))
2854 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !can_lookup(nd
->inode
))
2856 audit_inode(name
, nd
->path
.dentry
, 0);
2858 if (!S_ISREG(nd
->inode
->i_mode
))
2859 will_truncate
= false;
2861 if (will_truncate
) {
2862 error
= mnt_want_write(nd
->path
.mnt
);
2867 finish_open_created
:
2868 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2871 file
->f_path
.mnt
= nd
->path
.mnt
;
2872 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
2874 if (error
== -EOPENSTALE
)
2879 error
= open_check_o_direct(file
);
2882 error
= ima_file_check(file
, op
->acc_mode
);
2886 if (will_truncate
) {
2887 error
= handle_truncate(file
);
2893 mnt_drop_write(nd
->path
.mnt
);
2894 path_put(&save_parent
);
2899 path_put_conditional(path
, nd
);
2906 /* If no saved parent or already retried then can't retry */
2907 if (!save_parent
.dentry
|| retried
)
2910 BUG_ON(save_parent
.dentry
!= dir
);
2911 path_put(&nd
->path
);
2912 nd
->path
= save_parent
;
2913 nd
->inode
= dir
->d_inode
;
2914 save_parent
.mnt
= NULL
;
2915 save_parent
.dentry
= NULL
;
2917 mnt_drop_write(nd
->path
.mnt
);
2924 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
2925 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2927 struct file
*base
= NULL
;
2933 file
= get_empty_filp();
2937 file
->f_flags
= op
->open_flag
;
2939 error
= path_init(dfd
, pathname
->name
, flags
| LOOKUP_PARENT
, nd
, &base
);
2940 if (unlikely(error
))
2943 current
->total_link_count
= 0;
2944 error
= link_path_walk(pathname
->name
, nd
);
2945 if (unlikely(error
))
2948 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2949 while (unlikely(error
> 0)) { /* trailing symlink */
2950 struct path link
= path
;
2952 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2953 path_put_conditional(&path
, nd
);
2954 path_put(&nd
->path
);
2958 error
= may_follow_link(&link
, nd
);
2959 if (unlikely(error
))
2961 nd
->flags
|= LOOKUP_PARENT
;
2962 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2963 error
= follow_link(&link
, nd
, &cookie
);
2964 if (unlikely(error
))
2966 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2967 put_link(nd
, &link
, cookie
);
2970 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2971 path_put(&nd
->root
);
2974 if (!(opened
& FILE_OPENED
)) {
2978 if (unlikely(error
)) {
2979 if (error
== -EOPENSTALE
) {
2980 if (flags
& LOOKUP_RCU
)
2985 file
= ERR_PTR(error
);
2990 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
2991 const struct open_flags
*op
, int flags
)
2993 struct nameidata nd
;
2996 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2997 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2998 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2999 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3000 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3004 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3005 const char *name
, const struct open_flags
*op
, int flags
)
3007 struct nameidata nd
;
3009 struct filename filename
= { .name
= name
};
3012 nd
.root
.dentry
= dentry
;
3014 flags
|= LOOKUP_ROOT
;
3016 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
3017 return ERR_PTR(-ELOOP
);
3019 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3020 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3021 file
= path_openat(-1, &filename
, &nd
, op
, flags
);
3022 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3023 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3027 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3028 struct path
*path
, unsigned int lookup_flags
)
3030 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3031 struct nameidata nd
;
3034 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3037 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3038 * other flags passed in are ignored!
3040 lookup_flags
&= LOOKUP_REVAL
;
3042 error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
|lookup_flags
, &nd
);
3044 return ERR_PTR(error
);
3047 * Yucky last component or no last component at all?
3048 * (foo/., foo/.., /////)
3050 if (nd
.last_type
!= LAST_NORM
)
3052 nd
.flags
&= ~LOOKUP_PARENT
;
3053 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3055 /* don't fail immediately if it's r/o, at least try to report other errors */
3056 err2
= mnt_want_write(nd
.path
.mnt
);
3058 * Do the final lookup.
3060 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3061 dentry
= lookup_hash(&nd
);
3066 if (dentry
->d_inode
)
3069 * Special case - lookup gave negative, but... we had foo/bar/
3070 * From the vfs_mknod() POV we just have a negative dentry -
3071 * all is fine. Let's be bastards - you had / on the end, you've
3072 * been asking for (non-existent) directory. -ENOENT for you.
3074 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3078 if (unlikely(err2
)) {
3086 dentry
= ERR_PTR(error
);
3088 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3090 mnt_drop_write(nd
.path
.mnt
);
3095 EXPORT_SYMBOL(kern_path_create
);
3097 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3100 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3101 mnt_drop_write(path
->mnt
);
3104 EXPORT_SYMBOL(done_path_create
);
3106 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3107 struct path
*path
, unsigned int lookup_flags
)
3109 struct filename
*tmp
= getname(pathname
);
3112 return ERR_CAST(tmp
);
3113 res
= kern_path_create(dfd
, tmp
->name
, path
, lookup_flags
);
3117 EXPORT_SYMBOL(user_path_create
);
3119 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3121 int error
= may_create(dir
, dentry
);
3126 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3129 if (!dir
->i_op
->mknod
)
3132 error
= devcgroup_inode_mknod(mode
, dev
);
3136 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3140 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3142 fsnotify_create(dir
, dentry
);
3146 static int may_mknod(umode_t mode
)
3148 switch (mode
& S_IFMT
) {
3154 case 0: /* zero mode translates to S_IFREG */
3163 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3166 struct dentry
*dentry
;
3169 unsigned int lookup_flags
= 0;
3171 error
= may_mknod(mode
);
3175 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3177 return PTR_ERR(dentry
);
3179 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3180 mode
&= ~current_umask();
3181 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3184 switch (mode
& S_IFMT
) {
3185 case 0: case S_IFREG
:
3186 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3188 case S_IFCHR
: case S_IFBLK
:
3189 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3190 new_decode_dev(dev
));
3192 case S_IFIFO
: case S_IFSOCK
:
3193 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3197 done_path_create(&path
, dentry
);
3198 if (retry_estale(error
, lookup_flags
)) {
3199 lookup_flags
|= LOOKUP_REVAL
;
3205 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3207 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3210 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3212 int error
= may_create(dir
, dentry
);
3213 unsigned max_links
= dir
->i_sb
->s_max_links
;
3218 if (!dir
->i_op
->mkdir
)
3221 mode
&= (S_IRWXUGO
|S_ISVTX
);
3222 error
= security_inode_mkdir(dir
, dentry
, mode
);
3226 if (max_links
&& dir
->i_nlink
>= max_links
)
3229 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3231 fsnotify_mkdir(dir
, dentry
);
3235 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3237 struct dentry
*dentry
;
3240 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3243 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3245 return PTR_ERR(dentry
);
3247 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3248 mode
&= ~current_umask();
3249 error
= security_path_mkdir(&path
, dentry
, mode
);
3251 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3252 done_path_create(&path
, dentry
);
3253 if (retry_estale(error
, lookup_flags
)) {
3254 lookup_flags
|= LOOKUP_REVAL
;
3260 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3262 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3266 * The dentry_unhash() helper will try to drop the dentry early: we
3267 * should have a usage count of 1 if we're the only user of this
3268 * dentry, and if that is true (possibly after pruning the dcache),
3269 * then we drop the dentry now.
3271 * A low-level filesystem can, if it choses, legally
3274 * if (!d_unhashed(dentry))
3277 * if it cannot handle the case of removing a directory
3278 * that is still in use by something else..
3280 void dentry_unhash(struct dentry
*dentry
)
3282 shrink_dcache_parent(dentry
);
3283 spin_lock(&dentry
->d_lock
);
3284 if (dentry
->d_count
== 1)
3286 spin_unlock(&dentry
->d_lock
);
3289 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3291 int error
= may_delete(dir
, dentry
, 1);
3296 if (!dir
->i_op
->rmdir
)
3300 mutex_lock(&dentry
->d_inode
->i_mutex
);
3303 if (d_mountpoint(dentry
))
3306 error
= security_inode_rmdir(dir
, dentry
);
3310 shrink_dcache_parent(dentry
);
3311 error
= dir
->i_op
->rmdir(dir
, dentry
);
3315 dentry
->d_inode
->i_flags
|= S_DEAD
;
3319 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3326 static long do_rmdir(int dfd
, const char __user
*pathname
)
3329 struct filename
*name
;
3330 struct dentry
*dentry
;
3331 struct nameidata nd
;
3332 unsigned int lookup_flags
= 0;
3334 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3336 return PTR_ERR(name
);
3338 switch(nd
.last_type
) {
3350 nd
.flags
&= ~LOOKUP_PARENT
;
3351 error
= mnt_want_write(nd
.path
.mnt
);
3355 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3356 dentry
= lookup_hash(&nd
);
3357 error
= PTR_ERR(dentry
);
3360 if (!dentry
->d_inode
) {
3364 error
= security_path_rmdir(&nd
.path
, dentry
);
3367 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3371 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3372 mnt_drop_write(nd
.path
.mnt
);
3376 if (retry_estale(error
, lookup_flags
)) {
3377 lookup_flags
|= LOOKUP_REVAL
;
3383 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3385 return do_rmdir(AT_FDCWD
, pathname
);
3388 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3390 int error
= may_delete(dir
, dentry
, 0);
3395 if (!dir
->i_op
->unlink
)
3398 mutex_lock(&dentry
->d_inode
->i_mutex
);
3399 if (d_mountpoint(dentry
))
3402 error
= security_inode_unlink(dir
, dentry
);
3404 error
= dir
->i_op
->unlink(dir
, dentry
);
3409 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3411 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3412 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3413 fsnotify_link_count(dentry
->d_inode
);
3421 * Make sure that the actual truncation of the file will occur outside its
3422 * directory's i_mutex. Truncate can take a long time if there is a lot of
3423 * writeout happening, and we don't want to prevent access to the directory
3424 * while waiting on the I/O.
3426 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3429 struct filename
*name
;
3430 struct dentry
*dentry
;
3431 struct nameidata nd
;
3432 struct inode
*inode
= NULL
;
3433 unsigned int lookup_flags
= 0;
3435 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3437 return PTR_ERR(name
);
3440 if (nd
.last_type
!= LAST_NORM
)
3443 nd
.flags
&= ~LOOKUP_PARENT
;
3444 error
= mnt_want_write(nd
.path
.mnt
);
3448 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3449 dentry
= lookup_hash(&nd
);
3450 error
= PTR_ERR(dentry
);
3451 if (!IS_ERR(dentry
)) {
3452 /* Why not before? Because we want correct error value */
3453 if (nd
.last
.name
[nd
.last
.len
])
3455 inode
= dentry
->d_inode
;
3459 error
= security_path_unlink(&nd
.path
, dentry
);
3462 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
3466 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3468 iput(inode
); /* truncate the inode here */
3469 mnt_drop_write(nd
.path
.mnt
);
3473 if (retry_estale(error
, lookup_flags
)) {
3474 lookup_flags
|= LOOKUP_REVAL
;
3481 error
= !dentry
->d_inode
? -ENOENT
:
3482 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
3486 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3488 if ((flag
& ~AT_REMOVEDIR
) != 0)
3491 if (flag
& AT_REMOVEDIR
)
3492 return do_rmdir(dfd
, pathname
);
3494 return do_unlinkat(dfd
, pathname
);
3497 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3499 return do_unlinkat(AT_FDCWD
, pathname
);
3502 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3504 int error
= may_create(dir
, dentry
);
3509 if (!dir
->i_op
->symlink
)
3512 error
= security_inode_symlink(dir
, dentry
, oldname
);
3516 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3518 fsnotify_create(dir
, dentry
);
3522 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3523 int, newdfd
, const char __user
*, newname
)
3526 struct filename
*from
;
3527 struct dentry
*dentry
;
3529 unsigned int lookup_flags
= 0;
3531 from
= getname(oldname
);
3533 return PTR_ERR(from
);
3535 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3536 error
= PTR_ERR(dentry
);
3540 error
= security_path_symlink(&path
, dentry
, from
->name
);
3542 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3543 done_path_create(&path
, dentry
);
3544 if (retry_estale(error
, lookup_flags
)) {
3545 lookup_flags
|= LOOKUP_REVAL
;
3553 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3555 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3558 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3560 struct inode
*inode
= old_dentry
->d_inode
;
3561 unsigned max_links
= dir
->i_sb
->s_max_links
;
3567 error
= may_create(dir
, new_dentry
);
3571 if (dir
->i_sb
!= inode
->i_sb
)
3575 * A link to an append-only or immutable file cannot be created.
3577 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3579 if (!dir
->i_op
->link
)
3581 if (S_ISDIR(inode
->i_mode
))
3584 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3588 mutex_lock(&inode
->i_mutex
);
3589 /* Make sure we don't allow creating hardlink to an unlinked file */
3590 if (inode
->i_nlink
== 0)
3592 else if (max_links
&& inode
->i_nlink
>= max_links
)
3595 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3596 mutex_unlock(&inode
->i_mutex
);
3598 fsnotify_link(dir
, inode
, new_dentry
);
3603 * Hardlinks are often used in delicate situations. We avoid
3604 * security-related surprises by not following symlinks on the
3607 * We don't follow them on the oldname either to be compatible
3608 * with linux 2.0, and to avoid hard-linking to directories
3609 * and other special files. --ADM
3611 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3612 int, newdfd
, const char __user
*, newname
, int, flags
)
3614 struct dentry
*new_dentry
;
3615 struct path old_path
, new_path
;
3619 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3622 * To use null names we require CAP_DAC_READ_SEARCH
3623 * This ensures that not everyone will be able to create
3624 * handlink using the passed filedescriptor.
3626 if (flags
& AT_EMPTY_PATH
) {
3627 if (!capable(CAP_DAC_READ_SEARCH
))
3632 if (flags
& AT_SYMLINK_FOLLOW
)
3633 how
|= LOOKUP_FOLLOW
;
3635 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3639 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
3640 (how
& LOOKUP_REVAL
));
3641 error
= PTR_ERR(new_dentry
);
3642 if (IS_ERR(new_dentry
))
3646 if (old_path
.mnt
!= new_path
.mnt
)
3648 error
= may_linkat(&old_path
);
3649 if (unlikely(error
))
3651 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3654 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3656 done_path_create(&new_path
, new_dentry
);
3657 if (retry_estale(error
, how
)) {
3658 path_put(&old_path
);
3659 how
|= LOOKUP_REVAL
;
3663 path_put(&old_path
);
3668 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3670 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3674 * The worst of all namespace operations - renaming directory. "Perverted"
3675 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3677 * a) we can get into loop creation. Check is done in is_subdir().
3678 * b) race potential - two innocent renames can create a loop together.
3679 * That's where 4.4 screws up. Current fix: serialization on
3680 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3682 * c) we have to lock _three_ objects - parents and victim (if it exists).
3683 * And that - after we got ->i_mutex on parents (until then we don't know
3684 * whether the target exists). Solution: try to be smart with locking
3685 * order for inodes. We rely on the fact that tree topology may change
3686 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3687 * move will be locked. Thus we can rank directories by the tree
3688 * (ancestors first) and rank all non-directories after them.
3689 * That works since everybody except rename does "lock parent, lookup,
3690 * lock child" and rename is under ->s_vfs_rename_mutex.
3691 * HOWEVER, it relies on the assumption that any object with ->lookup()
3692 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3693 * we'd better make sure that there's no link(2) for them.
3694 * d) conversion from fhandle to dentry may come in the wrong moment - when
3695 * we are removing the target. Solution: we will have to grab ->i_mutex
3696 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3697 * ->i_mutex on parents, which works but leads to some truly excessive
3700 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3701 struct inode
*new_dir
, struct dentry
*new_dentry
)
3704 struct inode
*target
= new_dentry
->d_inode
;
3705 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3708 * If we are going to change the parent - check write permissions,
3709 * we'll need to flip '..'.
3711 if (new_dir
!= old_dir
) {
3712 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3717 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3723 mutex_lock(&target
->i_mutex
);
3726 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3730 if (max_links
&& !target
&& new_dir
!= old_dir
&&
3731 new_dir
->i_nlink
>= max_links
)
3735 shrink_dcache_parent(new_dentry
);
3736 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3741 target
->i_flags
|= S_DEAD
;
3742 dont_mount(new_dentry
);
3746 mutex_unlock(&target
->i_mutex
);
3749 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3750 d_move(old_dentry
,new_dentry
);
3754 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3755 struct inode
*new_dir
, struct dentry
*new_dentry
)
3757 struct inode
*target
= new_dentry
->d_inode
;
3760 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3766 mutex_lock(&target
->i_mutex
);
3769 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3772 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3777 dont_mount(new_dentry
);
3778 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3779 d_move(old_dentry
, new_dentry
);
3782 mutex_unlock(&target
->i_mutex
);
3787 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3788 struct inode
*new_dir
, struct dentry
*new_dentry
)
3791 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3792 const unsigned char *old_name
;
3794 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3797 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3801 if (!new_dentry
->d_inode
)
3802 error
= may_create(new_dir
, new_dentry
);
3804 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3808 if (!old_dir
->i_op
->rename
)
3811 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3814 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3816 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3818 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3819 new_dentry
->d_inode
, old_dentry
);
3820 fsnotify_oldname_free(old_name
);
3825 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3826 int, newdfd
, const char __user
*, newname
)
3828 struct dentry
*old_dir
, *new_dir
;
3829 struct dentry
*old_dentry
, *new_dentry
;
3830 struct dentry
*trap
;
3831 struct nameidata oldnd
, newnd
;
3832 struct filename
*from
;
3833 struct filename
*to
;
3834 unsigned int lookup_flags
= 0;
3835 bool should_retry
= false;
3838 from
= user_path_parent(olddfd
, oldname
, &oldnd
, lookup_flags
);
3840 error
= PTR_ERR(from
);
3844 to
= user_path_parent(newdfd
, newname
, &newnd
, lookup_flags
);
3846 error
= PTR_ERR(to
);
3851 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3854 old_dir
= oldnd
.path
.dentry
;
3856 if (oldnd
.last_type
!= LAST_NORM
)
3859 new_dir
= newnd
.path
.dentry
;
3860 if (newnd
.last_type
!= LAST_NORM
)
3863 error
= mnt_want_write(oldnd
.path
.mnt
);
3867 oldnd
.flags
&= ~LOOKUP_PARENT
;
3868 newnd
.flags
&= ~LOOKUP_PARENT
;
3869 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3871 trap
= lock_rename(new_dir
, old_dir
);
3873 old_dentry
= lookup_hash(&oldnd
);
3874 error
= PTR_ERR(old_dentry
);
3875 if (IS_ERR(old_dentry
))
3877 /* source must exist */
3879 if (!old_dentry
->d_inode
)
3881 /* unless the source is a directory trailing slashes give -ENOTDIR */
3882 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3884 if (oldnd
.last
.name
[oldnd
.last
.len
])
3886 if (newnd
.last
.name
[newnd
.last
.len
])
3889 /* source should not be ancestor of target */
3891 if (old_dentry
== trap
)
3893 new_dentry
= lookup_hash(&newnd
);
3894 error
= PTR_ERR(new_dentry
);
3895 if (IS_ERR(new_dentry
))
3897 /* target should not be an ancestor of source */
3899 if (new_dentry
== trap
)
3902 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3903 &newnd
.path
, new_dentry
);
3906 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3907 new_dir
->d_inode
, new_dentry
);
3913 unlock_rename(new_dir
, old_dir
);
3914 mnt_drop_write(oldnd
.path
.mnt
);
3916 if (retry_estale(error
, lookup_flags
))
3917 should_retry
= true;
3918 path_put(&newnd
.path
);
3921 path_put(&oldnd
.path
);
3924 should_retry
= false;
3925 lookup_flags
|= LOOKUP_REVAL
;
3932 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3934 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3937 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3941 len
= PTR_ERR(link
);
3946 if (len
> (unsigned) buflen
)
3948 if (copy_to_user(buffer
, link
, len
))
3955 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3956 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3957 * using) it for any given inode is up to filesystem.
3959 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3961 struct nameidata nd
;
3966 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3968 return PTR_ERR(cookie
);
3970 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3971 if (dentry
->d_inode
->i_op
->put_link
)
3972 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3976 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3978 return __vfs_follow_link(nd
, link
);
3981 /* get the link contents into pagecache */
3982 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3986 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3987 page
= read_mapping_page(mapping
, 0, NULL
);
3992 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3996 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3998 struct page
*page
= NULL
;
3999 char *s
= page_getlink(dentry
, &page
);
4000 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
4003 page_cache_release(page
);
4008 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
4010 struct page
*page
= NULL
;
4011 nd_set_link(nd
, page_getlink(dentry
, &page
));
4015 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
4017 struct page
*page
= cookie
;
4021 page_cache_release(page
);
4026 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4028 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4030 struct address_space
*mapping
= inode
->i_mapping
;
4035 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4037 flags
|= AOP_FLAG_NOFS
;
4040 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4041 flags
, &page
, &fsdata
);
4045 kaddr
= kmap_atomic(page
);
4046 memcpy(kaddr
, symname
, len
-1);
4047 kunmap_atomic(kaddr
);
4049 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4056 mark_inode_dirty(inode
);
4062 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4064 return __page_symlink(inode
, symname
, len
,
4065 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4068 const struct inode_operations page_symlink_inode_operations
= {
4069 .readlink
= generic_readlink
,
4070 .follow_link
= page_follow_link_light
,
4071 .put_link
= page_put_link
,
4074 EXPORT_SYMBOL(user_path_at
);
4075 EXPORT_SYMBOL(follow_down_one
);
4076 EXPORT_SYMBOL(follow_down
);
4077 EXPORT_SYMBOL(follow_up
);
4078 EXPORT_SYMBOL(get_write_access
); /* nfsd */
4079 EXPORT_SYMBOL(lock_rename
);
4080 EXPORT_SYMBOL(lookup_one_len
);
4081 EXPORT_SYMBOL(page_follow_link_light
);
4082 EXPORT_SYMBOL(page_put_link
);
4083 EXPORT_SYMBOL(page_readlink
);
4084 EXPORT_SYMBOL(__page_symlink
);
4085 EXPORT_SYMBOL(page_symlink
);
4086 EXPORT_SYMBOL(page_symlink_inode_operations
);
4087 EXPORT_SYMBOL(kern_path
);
4088 EXPORT_SYMBOL(vfs_path_lookup
);
4089 EXPORT_SYMBOL(inode_permission
);
4090 EXPORT_SYMBOL(unlock_rename
);
4091 EXPORT_SYMBOL(vfs_create
);
4092 EXPORT_SYMBOL(vfs_follow_link
);
4093 EXPORT_SYMBOL(vfs_link
);
4094 EXPORT_SYMBOL(vfs_mkdir
);
4095 EXPORT_SYMBOL(vfs_mknod
);
4096 EXPORT_SYMBOL(generic_permission
);
4097 EXPORT_SYMBOL(vfs_readlink
);
4098 EXPORT_SYMBOL(vfs_rename
);
4099 EXPORT_SYMBOL(vfs_rmdir
);
4100 EXPORT_SYMBOL(vfs_symlink
);
4101 EXPORT_SYMBOL(vfs_unlink
);
4102 EXPORT_SYMBOL(dentry_unhash
);
4103 EXPORT_SYMBOL(generic_readlink
);