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(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(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_REVALIDATE
)))
606 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
609 /* Note: we do not d_invalidate() */
610 status
= d_revalidate(dentry
, nd
->flags
);
621 static __always_inline
void set_root(struct nameidata
*nd
)
624 get_fs_root(current
->fs
, &nd
->root
);
627 static int link_path_walk(const char *, struct nameidata
*);
629 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
632 struct fs_struct
*fs
= current
->fs
;
636 seq
= read_seqcount_begin(&fs
->seq
);
638 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
639 } while (read_seqcount_retry(&fs
->seq
, seq
));
643 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
655 nd
->flags
|= LOOKUP_JUMPED
;
657 nd
->inode
= nd
->path
.dentry
->d_inode
;
659 ret
= link_path_walk(link
, nd
);
663 return PTR_ERR(link
);
666 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
669 if (path
->mnt
!= nd
->path
.mnt
)
673 static inline void path_to_nameidata(const struct path
*path
,
674 struct nameidata
*nd
)
676 if (!(nd
->flags
& LOOKUP_RCU
)) {
677 dput(nd
->path
.dentry
);
678 if (nd
->path
.mnt
!= path
->mnt
)
679 mntput(nd
->path
.mnt
);
681 nd
->path
.mnt
= path
->mnt
;
682 nd
->path
.dentry
= path
->dentry
;
686 * Helper to directly jump to a known parsed path from ->follow_link,
687 * caller must have taken a reference to path beforehand.
689 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
694 nd
->inode
= nd
->path
.dentry
->d_inode
;
695 nd
->flags
|= LOOKUP_JUMPED
;
697 BUG_ON(nd
->inode
->i_op
->follow_link
);
700 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
702 struct inode
*inode
= link
->dentry
->d_inode
;
703 if (inode
->i_op
->put_link
)
704 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
708 int sysctl_protected_symlinks __read_mostly
= 0;
709 int sysctl_protected_hardlinks __read_mostly
= 0;
712 * may_follow_link - Check symlink following for unsafe situations
713 * @link: The path of the symlink
714 * @nd: nameidata pathwalk data
716 * In the case of the sysctl_protected_symlinks sysctl being enabled,
717 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
718 * in a sticky world-writable directory. This is to protect privileged
719 * processes from failing races against path names that may change out
720 * from under them by way of other users creating malicious symlinks.
721 * It will permit symlinks to be followed only when outside a sticky
722 * world-writable directory, or when the uid of the symlink and follower
723 * match, or when the directory owner matches the symlink's owner.
725 * Returns 0 if following the symlink is allowed, -ve on error.
727 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
729 const struct inode
*inode
;
730 const struct inode
*parent
;
732 if (!sysctl_protected_symlinks
)
735 /* Allowed if owner and follower match. */
736 inode
= link
->dentry
->d_inode
;
737 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
740 /* Allowed if parent directory not sticky and world-writable. */
741 parent
= nd
->path
.dentry
->d_inode
;
742 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
745 /* Allowed if parent directory and link owner match. */
746 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
749 audit_log_link_denied("follow_link", link
);
750 path_put_conditional(link
, nd
);
756 * safe_hardlink_source - Check for safe hardlink conditions
757 * @inode: the source inode to hardlink from
759 * Return false if at least one of the following conditions:
760 * - inode is not a regular file
762 * - inode is setgid and group-exec
763 * - access failure for read and write
765 * Otherwise returns true.
767 static bool safe_hardlink_source(struct inode
*inode
)
769 umode_t mode
= inode
->i_mode
;
771 /* Special files should not get pinned to the filesystem. */
775 /* Setuid files should not get pinned to the filesystem. */
779 /* Executable setgid files should not get pinned to the filesystem. */
780 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
783 /* Hardlinking to unreadable or unwritable sources is dangerous. */
784 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
791 * may_linkat - Check permissions for creating a hardlink
792 * @link: the source to hardlink from
794 * Block hardlink when all of:
795 * - sysctl_protected_hardlinks enabled
796 * - fsuid does not match inode
797 * - hardlink source is unsafe (see safe_hardlink_source() above)
800 * Returns 0 if successful, -ve on error.
802 static int may_linkat(struct path
*link
)
804 const struct cred
*cred
;
807 if (!sysctl_protected_hardlinks
)
810 cred
= current_cred();
811 inode
= link
->dentry
->d_inode
;
813 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
814 * otherwise, it must be a safe source.
816 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
820 audit_log_link_denied("linkat", link
);
824 static __always_inline
int
825 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
827 struct dentry
*dentry
= link
->dentry
;
831 BUG_ON(nd
->flags
& LOOKUP_RCU
);
833 if (link
->mnt
== nd
->path
.mnt
)
837 if (unlikely(current
->total_link_count
>= 40))
838 goto out_put_nd_path
;
841 current
->total_link_count
++;
844 nd_set_link(nd
, NULL
);
846 error
= security_inode_follow_link(link
->dentry
, nd
);
848 goto out_put_nd_path
;
850 nd
->last_type
= LAST_BIND
;
851 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
854 goto out_put_nd_path
;
859 error
= __vfs_follow_link(nd
, s
);
861 put_link(nd
, link
, *p
);
873 static int follow_up_rcu(struct path
*path
)
875 struct mount
*mnt
= real_mount(path
->mnt
);
876 struct mount
*parent
;
877 struct dentry
*mountpoint
;
879 parent
= mnt
->mnt_parent
;
880 if (&parent
->mnt
== path
->mnt
)
882 mountpoint
= mnt
->mnt_mountpoint
;
883 path
->dentry
= mountpoint
;
884 path
->mnt
= &parent
->mnt
;
889 * follow_up - Find the mountpoint of path's vfsmount
891 * Given a path, find the mountpoint of its source file system.
892 * Replace @path with the path of the mountpoint in the parent mount.
895 * Return 1 if we went up a level and 0 if we were already at the
898 int follow_up(struct path
*path
)
900 struct mount
*mnt
= real_mount(path
->mnt
);
901 struct mount
*parent
;
902 struct dentry
*mountpoint
;
904 br_read_lock(&vfsmount_lock
);
905 parent
= mnt
->mnt_parent
;
907 br_read_unlock(&vfsmount_lock
);
910 mntget(&parent
->mnt
);
911 mountpoint
= dget(mnt
->mnt_mountpoint
);
912 br_read_unlock(&vfsmount_lock
);
914 path
->dentry
= mountpoint
;
916 path
->mnt
= &parent
->mnt
;
921 * Perform an automount
922 * - return -EISDIR to tell follow_managed() to stop and return the path we
925 static int follow_automount(struct path
*path
, unsigned flags
,
928 struct vfsmount
*mnt
;
931 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
934 /* We don't want to mount if someone's just doing a stat -
935 * unless they're stat'ing a directory and appended a '/' to
938 * We do, however, want to mount if someone wants to open or
939 * create a file of any type under the mountpoint, wants to
940 * traverse through the mountpoint or wants to open the
941 * mounted directory. Also, autofs may mark negative dentries
942 * as being automount points. These will need the attentions
943 * of the daemon to instantiate them before they can be used.
945 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
946 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
947 path
->dentry
->d_inode
)
950 current
->total_link_count
++;
951 if (current
->total_link_count
>= 40)
954 mnt
= path
->dentry
->d_op
->d_automount(path
);
957 * The filesystem is allowed to return -EISDIR here to indicate
958 * it doesn't want to automount. For instance, autofs would do
959 * this so that its userspace daemon can mount on this dentry.
961 * However, we can only permit this if it's a terminal point in
962 * the path being looked up; if it wasn't then the remainder of
963 * the path is inaccessible and we should say so.
965 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
970 if (!mnt
) /* mount collision */
974 /* lock_mount() may release path->mnt on error */
978 err
= finish_automount(mnt
, path
);
982 /* Someone else made a mount here whilst we were busy */
987 path
->dentry
= dget(mnt
->mnt_root
);
996 * Handle a dentry that is managed in some way.
997 * - Flagged for transit management (autofs)
998 * - Flagged as mountpoint
999 * - Flagged as automount point
1001 * This may only be called in refwalk mode.
1003 * Serialization is taken care of in namespace.c
1005 static int follow_managed(struct path
*path
, unsigned flags
)
1007 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1009 bool need_mntput
= false;
1012 /* Given that we're not holding a lock here, we retain the value in a
1013 * local variable for each dentry as we look at it so that we don't see
1014 * the components of that value change under us */
1015 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1016 managed
&= DCACHE_MANAGED_DENTRY
,
1017 unlikely(managed
!= 0)) {
1018 /* Allow the filesystem to manage the transit without i_mutex
1020 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1021 BUG_ON(!path
->dentry
->d_op
);
1022 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1023 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1028 /* Transit to a mounted filesystem. */
1029 if (managed
& DCACHE_MOUNTED
) {
1030 struct vfsmount
*mounted
= lookup_mnt(path
);
1035 path
->mnt
= mounted
;
1036 path
->dentry
= dget(mounted
->mnt_root
);
1041 /* Something is mounted on this dentry in another
1042 * namespace and/or whatever was mounted there in this
1043 * namespace got unmounted before we managed to get the
1047 /* Handle an automount point */
1048 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1049 ret
= follow_automount(path
, flags
, &need_mntput
);
1055 /* We didn't change the current path point */
1059 if (need_mntput
&& path
->mnt
== mnt
)
1063 return ret
< 0 ? ret
: need_mntput
;
1066 int follow_down_one(struct path
*path
)
1068 struct vfsmount
*mounted
;
1070 mounted
= lookup_mnt(path
);
1074 path
->mnt
= mounted
;
1075 path
->dentry
= dget(mounted
->mnt_root
);
1081 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
1083 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
1084 dentry
->d_op
->d_manage(dentry
, true) < 0);
1088 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1089 * we meet a managed dentry that would need blocking.
1091 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1092 struct inode
**inode
)
1095 struct mount
*mounted
;
1097 * Don't forget we might have a non-mountpoint managed dentry
1098 * that wants to block transit.
1100 if (unlikely(managed_dentry_might_block(path
->dentry
)))
1103 if (!d_mountpoint(path
->dentry
))
1106 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1109 path
->mnt
= &mounted
->mnt
;
1110 path
->dentry
= mounted
->mnt
.mnt_root
;
1111 nd
->flags
|= LOOKUP_JUMPED
;
1112 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1114 * Update the inode too. We don't need to re-check the
1115 * dentry sequence number here after this d_inode read,
1116 * because a mount-point is always pinned.
1118 *inode
= path
->dentry
->d_inode
;
1123 static void follow_mount_rcu(struct nameidata
*nd
)
1125 while (d_mountpoint(nd
->path
.dentry
)) {
1126 struct mount
*mounted
;
1127 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
1130 nd
->path
.mnt
= &mounted
->mnt
;
1131 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1132 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1136 static int follow_dotdot_rcu(struct nameidata
*nd
)
1141 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1142 nd
->path
.mnt
== nd
->root
.mnt
) {
1145 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1146 struct dentry
*old
= nd
->path
.dentry
;
1147 struct dentry
*parent
= old
->d_parent
;
1150 seq
= read_seqcount_begin(&parent
->d_seq
);
1151 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1153 nd
->path
.dentry
= parent
;
1157 if (!follow_up_rcu(&nd
->path
))
1159 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1161 follow_mount_rcu(nd
);
1162 nd
->inode
= nd
->path
.dentry
->d_inode
;
1166 nd
->flags
&= ~LOOKUP_RCU
;
1167 if (!(nd
->flags
& LOOKUP_ROOT
))
1168 nd
->root
.mnt
= NULL
;
1174 * Follow down to the covering mount currently visible to userspace. At each
1175 * point, the filesystem owning that dentry may be queried as to whether the
1176 * caller is permitted to proceed or not.
1178 int follow_down(struct path
*path
)
1183 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1184 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1185 /* Allow the filesystem to manage the transit without i_mutex
1188 * We indicate to the filesystem if someone is trying to mount
1189 * something here. This gives autofs the chance to deny anyone
1190 * other than its daemon the right to mount on its
1193 * The filesystem may sleep at this point.
1195 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1196 BUG_ON(!path
->dentry
->d_op
);
1197 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1198 ret
= path
->dentry
->d_op
->d_manage(
1199 path
->dentry
, false);
1201 return ret
== -EISDIR
? 0 : ret
;
1204 /* Transit to a mounted filesystem. */
1205 if (managed
& DCACHE_MOUNTED
) {
1206 struct vfsmount
*mounted
= lookup_mnt(path
);
1211 path
->mnt
= mounted
;
1212 path
->dentry
= dget(mounted
->mnt_root
);
1216 /* Don't handle automount points here */
1223 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1225 static void follow_mount(struct path
*path
)
1227 while (d_mountpoint(path
->dentry
)) {
1228 struct vfsmount
*mounted
= lookup_mnt(path
);
1233 path
->mnt
= mounted
;
1234 path
->dentry
= dget(mounted
->mnt_root
);
1238 static void follow_dotdot(struct nameidata
*nd
)
1243 struct dentry
*old
= nd
->path
.dentry
;
1245 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1246 nd
->path
.mnt
== nd
->root
.mnt
) {
1249 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1250 /* rare case of legitimate dget_parent()... */
1251 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1255 if (!follow_up(&nd
->path
))
1258 follow_mount(&nd
->path
);
1259 nd
->inode
= nd
->path
.dentry
->d_inode
;
1263 * This looks up the name in dcache, possibly revalidates the old dentry and
1264 * allocates a new one if not found or not valid. In the need_lookup argument
1265 * returns whether i_op->lookup is necessary.
1267 * dir->d_inode->i_mutex must be held
1269 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1270 unsigned int flags
, bool *need_lookup
)
1272 struct dentry
*dentry
;
1275 *need_lookup
= false;
1276 dentry
= d_lookup(dir
, name
);
1278 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1279 error
= d_revalidate(dentry
, flags
);
1280 if (unlikely(error
<= 0)) {
1283 return ERR_PTR(error
);
1284 } else if (!d_invalidate(dentry
)) {
1293 dentry
= d_alloc(dir
, name
);
1294 if (unlikely(!dentry
))
1295 return ERR_PTR(-ENOMEM
);
1297 *need_lookup
= true;
1303 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1304 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1306 * dir->d_inode->i_mutex must be held
1308 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1313 /* Don't create child dentry for a dead directory. */
1314 if (unlikely(IS_DEADDIR(dir
))) {
1316 return ERR_PTR(-ENOENT
);
1319 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1320 if (unlikely(old
)) {
1327 static struct dentry
*__lookup_hash(struct qstr
*name
,
1328 struct dentry
*base
, unsigned int flags
)
1331 struct dentry
*dentry
;
1333 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1337 return lookup_real(base
->d_inode
, dentry
, flags
);
1341 * It's more convoluted than I'd like it to be, but... it's still fairly
1342 * small and for now I'd prefer to have fast path as straight as possible.
1343 * It _is_ time-critical.
1345 static int lookup_fast(struct nameidata
*nd
, struct qstr
*name
,
1346 struct path
*path
, struct inode
**inode
)
1348 struct vfsmount
*mnt
= nd
->path
.mnt
;
1349 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1355 * Rename seqlock is not required here because in the off chance
1356 * of a false negative due to a concurrent rename, we're going to
1357 * do the non-racy lookup, below.
1359 if (nd
->flags
& LOOKUP_RCU
) {
1361 dentry
= __d_lookup_rcu(parent
, name
, &seq
, nd
->inode
);
1366 * This sequence count validates that the inode matches
1367 * the dentry name information from lookup.
1369 *inode
= dentry
->d_inode
;
1370 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1374 * This sequence count validates that the parent had no
1375 * changes while we did the lookup of the dentry above.
1377 * The memory barrier in read_seqcount_begin of child is
1378 * enough, we can use __read_seqcount_retry here.
1380 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1384 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1385 status
= d_revalidate(dentry
, nd
->flags
);
1386 if (unlikely(status
<= 0)) {
1387 if (status
!= -ECHILD
)
1393 path
->dentry
= dentry
;
1394 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1396 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1400 if (unlazy_walk(nd
, dentry
))
1403 dentry
= __d_lookup(parent
, name
);
1406 if (unlikely(!dentry
))
1409 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1410 status
= d_revalidate(dentry
, nd
->flags
);
1411 if (unlikely(status
<= 0)) {
1416 if (!d_invalidate(dentry
)) {
1423 path
->dentry
= dentry
;
1424 err
= follow_managed(path
, nd
->flags
);
1425 if (unlikely(err
< 0)) {
1426 path_put_conditional(path
, nd
);
1430 nd
->flags
|= LOOKUP_JUMPED
;
1431 *inode
= path
->dentry
->d_inode
;
1438 /* Fast lookup failed, do it the slow way */
1439 static int lookup_slow(struct nameidata
*nd
, struct qstr
*name
,
1442 struct dentry
*dentry
, *parent
;
1445 parent
= nd
->path
.dentry
;
1446 BUG_ON(nd
->inode
!= parent
->d_inode
);
1448 mutex_lock(&parent
->d_inode
->i_mutex
);
1449 dentry
= __lookup_hash(name
, parent
, nd
->flags
);
1450 mutex_unlock(&parent
->d_inode
->i_mutex
);
1452 return PTR_ERR(dentry
);
1453 path
->mnt
= nd
->path
.mnt
;
1454 path
->dentry
= dentry
;
1455 err
= follow_managed(path
, nd
->flags
);
1456 if (unlikely(err
< 0)) {
1457 path_put_conditional(path
, nd
);
1461 nd
->flags
|= LOOKUP_JUMPED
;
1465 static inline int may_lookup(struct nameidata
*nd
)
1467 if (nd
->flags
& LOOKUP_RCU
) {
1468 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1471 if (unlazy_walk(nd
, NULL
))
1474 return inode_permission(nd
->inode
, MAY_EXEC
);
1477 static inline int handle_dots(struct nameidata
*nd
, int type
)
1479 if (type
== LAST_DOTDOT
) {
1480 if (nd
->flags
& LOOKUP_RCU
) {
1481 if (follow_dotdot_rcu(nd
))
1489 static void terminate_walk(struct nameidata
*nd
)
1491 if (!(nd
->flags
& LOOKUP_RCU
)) {
1492 path_put(&nd
->path
);
1494 nd
->flags
&= ~LOOKUP_RCU
;
1495 if (!(nd
->flags
& LOOKUP_ROOT
))
1496 nd
->root
.mnt
= NULL
;
1502 * Do we need to follow links? We _really_ want to be able
1503 * to do this check without having to look at inode->i_op,
1504 * so we keep a cache of "no, this doesn't need follow_link"
1505 * for the common case.
1507 static inline int should_follow_link(struct inode
*inode
, int follow
)
1509 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1510 if (likely(inode
->i_op
->follow_link
))
1513 /* This gets set once for the inode lifetime */
1514 spin_lock(&inode
->i_lock
);
1515 inode
->i_opflags
|= IOP_NOFOLLOW
;
1516 spin_unlock(&inode
->i_lock
);
1521 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1522 struct qstr
*name
, int type
, int follow
)
1524 struct inode
*inode
;
1527 * "." and ".." are special - ".." especially so because it has
1528 * to be able to know about the current root directory and
1529 * parent relationships.
1531 if (unlikely(type
!= LAST_NORM
))
1532 return handle_dots(nd
, type
);
1533 err
= lookup_fast(nd
, name
, path
, &inode
);
1534 if (unlikely(err
)) {
1538 err
= lookup_slow(nd
, name
, path
);
1542 inode
= path
->dentry
->d_inode
;
1548 if (should_follow_link(inode
, follow
)) {
1549 if (nd
->flags
& LOOKUP_RCU
) {
1550 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1555 BUG_ON(inode
!= path
->dentry
->d_inode
);
1558 path_to_nameidata(path
, nd
);
1563 path_to_nameidata(path
, nd
);
1570 * This limits recursive symlink follows to 8, while
1571 * limiting consecutive symlinks to 40.
1573 * Without that kind of total limit, nasty chains of consecutive
1574 * symlinks can cause almost arbitrarily long lookups.
1576 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1580 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1581 path_put_conditional(path
, nd
);
1582 path_put(&nd
->path
);
1585 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1588 current
->link_count
++;
1591 struct path link
= *path
;
1594 res
= follow_link(&link
, nd
, &cookie
);
1597 res
= walk_component(nd
, path
, &nd
->last
,
1598 nd
->last_type
, LOOKUP_FOLLOW
);
1599 put_link(nd
, &link
, cookie
);
1602 current
->link_count
--;
1608 * We really don't want to look at inode->i_op->lookup
1609 * when we don't have to. So we keep a cache bit in
1610 * the inode ->i_opflags field that says "yes, we can
1611 * do lookup on this inode".
1613 static inline int can_lookup(struct inode
*inode
)
1615 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1617 if (likely(!inode
->i_op
->lookup
))
1620 /* We do this once for the lifetime of the inode */
1621 spin_lock(&inode
->i_lock
);
1622 inode
->i_opflags
|= IOP_LOOKUP
;
1623 spin_unlock(&inode
->i_lock
);
1628 * We can do the critical dentry name comparison and hashing
1629 * operations one word at a time, but we are limited to:
1631 * - Architectures with fast unaligned word accesses. We could
1632 * do a "get_unaligned()" if this helps and is sufficiently
1635 * - Little-endian machines (so that we can generate the mask
1636 * of low bytes efficiently). Again, we *could* do a byte
1637 * swapping load on big-endian architectures if that is not
1638 * expensive enough to make the optimization worthless.
1640 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1641 * do not trap on the (extremely unlikely) case of a page
1642 * crossing operation.
1644 * - Furthermore, we need an efficient 64-bit compile for the
1645 * 64-bit case in order to generate the "number of bytes in
1646 * the final mask". Again, that could be replaced with a
1647 * efficient population count instruction or similar.
1649 #ifdef CONFIG_DCACHE_WORD_ACCESS
1651 #include <asm/word-at-a-time.h>
1655 static inline unsigned int fold_hash(unsigned long hash
)
1657 hash
+= hash
>> (8*sizeof(int));
1661 #else /* 32-bit case */
1663 #define fold_hash(x) (x)
1667 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1669 unsigned long a
, mask
;
1670 unsigned long hash
= 0;
1673 a
= load_unaligned_zeropad(name
);
1674 if (len
< sizeof(unsigned long))
1678 name
+= sizeof(unsigned long);
1679 len
-= sizeof(unsigned long);
1683 mask
= ~(~0ul << len
*8);
1686 return fold_hash(hash
);
1688 EXPORT_SYMBOL(full_name_hash
);
1691 * Calculate the length and hash of the path component, and
1692 * return the length of the component;
1694 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1696 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1697 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1700 len
= -sizeof(unsigned long);
1702 hash
= (hash
+ a
) * 9;
1703 len
+= sizeof(unsigned long);
1704 a
= load_unaligned_zeropad(name
+len
);
1705 b
= a
^ REPEAT_BYTE('/');
1706 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1708 adata
= prep_zero_mask(a
, adata
, &constants
);
1709 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1711 mask
= create_zero_mask(adata
| bdata
);
1713 hash
+= a
& zero_bytemask(mask
);
1714 *hashp
= fold_hash(hash
);
1716 return len
+ find_zero(mask
);
1721 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1723 unsigned long hash
= init_name_hash();
1725 hash
= partial_name_hash(*name
++, hash
);
1726 return end_name_hash(hash
);
1728 EXPORT_SYMBOL(full_name_hash
);
1731 * We know there's a real path component here of at least
1734 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1736 unsigned long hash
= init_name_hash();
1737 unsigned long len
= 0, c
;
1739 c
= (unsigned char)*name
;
1742 hash
= partial_name_hash(c
, hash
);
1743 c
= (unsigned char)name
[len
];
1744 } while (c
&& c
!= '/');
1745 *hashp
= end_name_hash(hash
);
1753 * This is the basic name resolution function, turning a pathname into
1754 * the final dentry. We expect 'base' to be positive and a directory.
1756 * Returns 0 and nd will have valid dentry and mnt on success.
1757 * Returns error and drops reference to input namei data on failure.
1759 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1769 /* At this point we know we have a real path component. */
1775 err
= may_lookup(nd
);
1779 len
= hash_name(name
, &this.hash
);
1784 if (name
[0] == '.') switch (len
) {
1786 if (name
[1] == '.') {
1788 nd
->flags
|= LOOKUP_JUMPED
;
1794 if (likely(type
== LAST_NORM
)) {
1795 struct dentry
*parent
= nd
->path
.dentry
;
1796 nd
->flags
&= ~LOOKUP_JUMPED
;
1797 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1798 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1806 nd
->last_type
= type
;
1811 * If it wasn't NUL, we know it was '/'. Skip that
1812 * slash, and continue until no more slashes.
1816 } while (unlikely(name
[len
] == '/'));
1822 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1827 err
= nested_symlink(&next
, nd
);
1831 if (!can_lookup(nd
->inode
)) {
1840 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1841 struct nameidata
*nd
, struct file
**fp
)
1845 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1846 nd
->flags
= flags
| LOOKUP_JUMPED
;
1848 if (flags
& LOOKUP_ROOT
) {
1849 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1851 if (!can_lookup(inode
))
1853 retval
= inode_permission(inode
, MAY_EXEC
);
1857 nd
->path
= nd
->root
;
1859 if (flags
& LOOKUP_RCU
) {
1861 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1863 path_get(&nd
->path
);
1868 nd
->root
.mnt
= NULL
;
1871 if (flags
& LOOKUP_RCU
) {
1876 path_get(&nd
->root
);
1878 nd
->path
= nd
->root
;
1879 } else if (dfd
== AT_FDCWD
) {
1880 if (flags
& LOOKUP_RCU
) {
1881 struct fs_struct
*fs
= current
->fs
;
1887 seq
= read_seqcount_begin(&fs
->seq
);
1889 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1890 } while (read_seqcount_retry(&fs
->seq
, seq
));
1892 get_fs_pwd(current
->fs
, &nd
->path
);
1895 /* Caller must check execute permissions on the starting path component */
1896 struct fd f
= fdget_raw(dfd
);
1897 struct dentry
*dentry
;
1902 dentry
= f
.file
->f_path
.dentry
;
1905 if (!can_lookup(dentry
->d_inode
)) {
1911 nd
->path
= f
.file
->f_path
;
1912 if (flags
& LOOKUP_RCU
) {
1915 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1918 path_get(&nd
->path
);
1923 nd
->inode
= nd
->path
.dentry
->d_inode
;
1927 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1929 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1930 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1932 nd
->flags
&= ~LOOKUP_PARENT
;
1933 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1934 nd
->flags
& LOOKUP_FOLLOW
);
1937 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1938 static int path_lookupat(int dfd
, const char *name
,
1939 unsigned int flags
, struct nameidata
*nd
)
1941 struct file
*base
= NULL
;
1946 * Path walking is largely split up into 2 different synchronisation
1947 * schemes, rcu-walk and ref-walk (explained in
1948 * Documentation/filesystems/path-lookup.txt). These share much of the
1949 * path walk code, but some things particularly setup, cleanup, and
1950 * following mounts are sufficiently divergent that functions are
1951 * duplicated. Typically there is a function foo(), and its RCU
1952 * analogue, foo_rcu().
1954 * -ECHILD is the error number of choice (just to avoid clashes) that
1955 * is returned if some aspect of an rcu-walk fails. Such an error must
1956 * be handled by restarting a traditional ref-walk (which will always
1957 * be able to complete).
1959 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1964 current
->total_link_count
= 0;
1965 err
= link_path_walk(name
, nd
);
1967 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1968 err
= lookup_last(nd
, &path
);
1971 struct path link
= path
;
1972 err
= may_follow_link(&link
, nd
);
1975 nd
->flags
|= LOOKUP_PARENT
;
1976 err
= follow_link(&link
, nd
, &cookie
);
1979 err
= lookup_last(nd
, &path
);
1980 put_link(nd
, &link
, cookie
);
1985 err
= complete_walk(nd
);
1987 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1988 if (!nd
->inode
->i_op
->lookup
) {
1989 path_put(&nd
->path
);
1997 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1998 path_put(&nd
->root
);
1999 nd
->root
.mnt
= NULL
;
2004 static int filename_lookup(int dfd
, struct filename
*name
,
2005 unsigned int flags
, struct nameidata
*nd
)
2007 int retval
= path_lookupat(dfd
, name
->name
, flags
| LOOKUP_RCU
, nd
);
2008 if (unlikely(retval
== -ECHILD
))
2009 retval
= path_lookupat(dfd
, name
->name
, flags
, nd
);
2010 if (unlikely(retval
== -ESTALE
))
2011 retval
= path_lookupat(dfd
, name
->name
,
2012 flags
| LOOKUP_REVAL
, nd
);
2014 if (likely(!retval
))
2015 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2019 static int do_path_lookup(int dfd
, const char *name
,
2020 unsigned int flags
, struct nameidata
*nd
)
2022 struct filename filename
= { .name
= name
};
2024 return filename_lookup(dfd
, &filename
, flags
, nd
);
2027 /* does lookup, returns the object with parent locked */
2028 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2030 struct nameidata nd
;
2032 int err
= do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, &nd
);
2034 return ERR_PTR(err
);
2035 if (nd
.last_type
!= LAST_NORM
) {
2037 return ERR_PTR(-EINVAL
);
2039 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2040 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2042 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2050 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2052 struct nameidata nd
;
2053 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
2060 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2061 * @dentry: pointer to dentry of the base directory
2062 * @mnt: pointer to vfs mount of the base directory
2063 * @name: pointer to file name
2064 * @flags: lookup flags
2065 * @path: pointer to struct path to fill
2067 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2068 const char *name
, unsigned int flags
,
2071 struct nameidata nd
;
2073 nd
.root
.dentry
= dentry
;
2075 BUG_ON(flags
& LOOKUP_PARENT
);
2076 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2077 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
2084 * Restricted form of lookup. Doesn't follow links, single-component only,
2085 * needs parent already locked. Doesn't follow mounts.
2088 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2090 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2094 * lookup_one_len - filesystem helper to lookup single pathname component
2095 * @name: pathname component to lookup
2096 * @base: base directory to lookup from
2097 * @len: maximum length @len should be interpreted to
2099 * Note that this routine is purely a helper for filesystem usage and should
2100 * not be called by generic code. Also note that by using this function the
2101 * nameidata argument is passed to the filesystem methods and a filesystem
2102 * using this helper needs to be prepared for that.
2104 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2110 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2114 this.hash
= full_name_hash(name
, len
);
2116 return ERR_PTR(-EACCES
);
2118 if (unlikely(name
[0] == '.')) {
2119 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2120 return ERR_PTR(-EACCES
);
2124 c
= *(const unsigned char *)name
++;
2125 if (c
== '/' || c
== '\0')
2126 return ERR_PTR(-EACCES
);
2129 * See if the low-level filesystem might want
2130 * to use its own hash..
2132 if (base
->d_flags
& DCACHE_OP_HASH
) {
2133 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
2135 return ERR_PTR(err
);
2138 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2140 return ERR_PTR(err
);
2142 return __lookup_hash(&this, base
, 0);
2145 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2146 struct path
*path
, int *empty
)
2148 struct nameidata nd
;
2149 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2150 int err
= PTR_ERR(tmp
);
2153 BUG_ON(flags
& LOOKUP_PARENT
);
2155 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2163 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2166 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2170 * NB: most callers don't do anything directly with the reference to the
2171 * to struct filename, but the nd->last pointer points into the name string
2172 * allocated by getname. So we must hold the reference to it until all
2173 * path-walking is complete.
2175 static struct filename
*
2176 user_path_parent(int dfd
, const char __user
*path
, struct nameidata
*nd
,
2179 struct filename
*s
= getname(path
);
2182 /* only LOOKUP_REVAL is allowed in extra flags */
2183 flags
&= LOOKUP_REVAL
;
2188 error
= filename_lookup(dfd
, s
, flags
| LOOKUP_PARENT
, nd
);
2191 return ERR_PTR(error
);
2198 * It's inline, so penalty for filesystems that don't use sticky bit is
2201 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
2203 kuid_t fsuid
= current_fsuid();
2205 if (!(dir
->i_mode
& S_ISVTX
))
2207 if (uid_eq(inode
->i_uid
, fsuid
))
2209 if (uid_eq(dir
->i_uid
, fsuid
))
2211 return !inode_capable(inode
, CAP_FOWNER
);
2215 * Check whether we can remove a link victim from directory dir, check
2216 * whether the type of victim is right.
2217 * 1. We can't do it if dir is read-only (done in permission())
2218 * 2. We should have write and exec permissions on dir
2219 * 3. We can't remove anything from append-only dir
2220 * 4. We can't do anything with immutable dir (done in permission())
2221 * 5. If the sticky bit on dir is set we should either
2222 * a. be owner of dir, or
2223 * b. be owner of victim, or
2224 * c. have CAP_FOWNER capability
2225 * 6. If the victim is append-only or immutable we can't do antyhing with
2226 * links pointing to it.
2227 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2228 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2229 * 9. We can't remove a root or mountpoint.
2230 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2231 * nfs_async_unlink().
2233 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
2237 if (!victim
->d_inode
)
2240 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2241 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2243 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2248 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
2249 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
2252 if (!S_ISDIR(victim
->d_inode
->i_mode
))
2254 if (IS_ROOT(victim
))
2256 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
2258 if (IS_DEADDIR(dir
))
2260 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2265 /* Check whether we can create an object with dentry child in directory
2267 * 1. We can't do it if child already exists (open has special treatment for
2268 * this case, but since we are inlined it's OK)
2269 * 2. We can't do it if dir is read-only (done in permission())
2270 * 3. We should have write and exec permissions on dir
2271 * 4. We can't do it if dir is immutable (done in permission())
2273 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2277 if (IS_DEADDIR(dir
))
2279 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2283 * p1 and p2 should be directories on the same fs.
2285 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2290 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2294 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2296 p
= d_ancestor(p2
, p1
);
2298 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2299 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2303 p
= d_ancestor(p1
, p2
);
2305 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2306 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2310 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2311 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2315 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2317 mutex_unlock(&p1
->d_inode
->i_mutex
);
2319 mutex_unlock(&p2
->d_inode
->i_mutex
);
2320 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2324 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2327 int error
= may_create(dir
, dentry
);
2331 if (!dir
->i_op
->create
)
2332 return -EACCES
; /* shouldn't it be ENOSYS? */
2335 error
= security_inode_create(dir
, dentry
, mode
);
2338 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2340 fsnotify_create(dir
, dentry
);
2344 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2346 struct dentry
*dentry
= path
->dentry
;
2347 struct inode
*inode
= dentry
->d_inode
;
2357 switch (inode
->i_mode
& S_IFMT
) {
2361 if (acc_mode
& MAY_WRITE
)
2366 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2375 error
= inode_permission(inode
, acc_mode
);
2380 * An append-only file must be opened in append mode for writing.
2382 if (IS_APPEND(inode
)) {
2383 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2389 /* O_NOATIME can only be set by the owner or superuser */
2390 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2396 static int handle_truncate(struct file
*filp
)
2398 struct path
*path
= &filp
->f_path
;
2399 struct inode
*inode
= path
->dentry
->d_inode
;
2400 int error
= get_write_access(inode
);
2404 * Refuse to truncate files with mandatory locks held on them.
2406 error
= locks_verify_locked(inode
);
2408 error
= security_path_truncate(path
);
2410 error
= do_truncate(path
->dentry
, 0,
2411 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2414 put_write_access(inode
);
2418 static inline int open_to_namei_flags(int flag
)
2420 if ((flag
& O_ACCMODE
) == 3)
2425 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2427 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2431 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2435 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2439 * Attempt to atomically look up, create and open a file from a negative
2442 * Returns 0 if successful. The file will have been created and attached to
2443 * @file by the filesystem calling finish_open().
2445 * Returns 1 if the file was looked up only or didn't need creating. The
2446 * caller will need to perform the open themselves. @path will have been
2447 * updated to point to the new dentry. This may be negative.
2449 * Returns an error code otherwise.
2451 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2452 struct path
*path
, struct file
*file
,
2453 const struct open_flags
*op
,
2454 bool got_write
, bool need_lookup
,
2457 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2458 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2462 int create_error
= 0;
2463 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2465 BUG_ON(dentry
->d_inode
);
2467 /* Don't create child dentry for a dead directory. */
2468 if (unlikely(IS_DEADDIR(dir
))) {
2474 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2475 mode
&= ~current_umask();
2477 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
)) {
2478 open_flag
&= ~O_TRUNC
;
2479 *opened
|= FILE_CREATED
;
2483 * Checking write permission is tricky, bacuse we don't know if we are
2484 * going to actually need it: O_CREAT opens should work as long as the
2485 * file exists. But checking existence breaks atomicity. The trick is
2486 * to check access and if not granted clear O_CREAT from the flags.
2488 * Another problem is returing the "right" error value (e.g. for an
2489 * O_EXCL open we want to return EEXIST not EROFS).
2491 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2492 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2493 if (!(open_flag
& O_CREAT
)) {
2495 * No O_CREATE -> atomicity not a requirement -> fall
2496 * back to lookup + open
2499 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2500 /* Fall back and fail with the right error */
2501 create_error
= -EROFS
;
2504 /* No side effects, safe to clear O_CREAT */
2505 create_error
= -EROFS
;
2506 open_flag
&= ~O_CREAT
;
2510 if (open_flag
& O_CREAT
) {
2511 error
= may_o_create(&nd
->path
, dentry
, mode
);
2513 create_error
= error
;
2514 if (open_flag
& O_EXCL
)
2516 open_flag
&= ~O_CREAT
;
2520 if (nd
->flags
& LOOKUP_DIRECTORY
)
2521 open_flag
|= O_DIRECTORY
;
2523 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2524 file
->f_path
.mnt
= nd
->path
.mnt
;
2525 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2528 if (create_error
&& error
== -ENOENT
)
2529 error
= create_error
;
2533 acc_mode
= op
->acc_mode
;
2534 if (*opened
& FILE_CREATED
) {
2535 fsnotify_create(dir
, dentry
);
2536 acc_mode
= MAY_OPEN
;
2539 if (error
) { /* returned 1, that is */
2540 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2544 if (file
->f_path
.dentry
) {
2546 dentry
= file
->f_path
.dentry
;
2548 if (create_error
&& dentry
->d_inode
== NULL
) {
2549 error
= create_error
;
2556 * We didn't have the inode before the open, so check open permission
2559 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2569 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2571 return PTR_ERR(dentry
);
2574 int open_flag
= op
->open_flag
;
2576 error
= create_error
;
2577 if ((open_flag
& O_EXCL
)) {
2578 if (!dentry
->d_inode
)
2580 } else if (!dentry
->d_inode
) {
2582 } else if ((open_flag
& O_TRUNC
) &&
2583 S_ISREG(dentry
->d_inode
->i_mode
)) {
2586 /* will fail later, go on to get the right error */
2590 path
->dentry
= dentry
;
2591 path
->mnt
= nd
->path
.mnt
;
2596 * Look up and maybe create and open the last component.
2598 * Must be called with i_mutex held on parent.
2600 * Returns 0 if the file was successfully atomically created (if necessary) and
2601 * opened. In this case the file will be returned attached to @file.
2603 * Returns 1 if the file was not completely opened at this time, though lookups
2604 * and creations will have been performed and the dentry returned in @path will
2605 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2606 * specified then a negative dentry may be returned.
2608 * An error code is returned otherwise.
2610 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2611 * cleared otherwise prior to returning.
2613 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2615 const struct open_flags
*op
,
2616 bool got_write
, int *opened
)
2618 struct dentry
*dir
= nd
->path
.dentry
;
2619 struct inode
*dir_inode
= dir
->d_inode
;
2620 struct dentry
*dentry
;
2624 *opened
&= ~FILE_CREATED
;
2625 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2627 return PTR_ERR(dentry
);
2629 /* Cached positive dentry: will open in f_op->open */
2630 if (!need_lookup
&& dentry
->d_inode
)
2633 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2634 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2635 need_lookup
, opened
);
2639 BUG_ON(dentry
->d_inode
);
2641 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2643 return PTR_ERR(dentry
);
2646 /* Negative dentry, just create the file */
2647 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2648 umode_t mode
= op
->mode
;
2649 if (!IS_POSIXACL(dir
->d_inode
))
2650 mode
&= ~current_umask();
2652 * This write is needed to ensure that a
2653 * rw->ro transition does not occur between
2654 * the time when the file is created and when
2655 * a permanent write count is taken through
2656 * the 'struct file' in finish_open().
2662 *opened
|= FILE_CREATED
;
2663 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2666 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2667 nd
->flags
& LOOKUP_EXCL
);
2672 path
->dentry
= dentry
;
2673 path
->mnt
= nd
->path
.mnt
;
2682 * Handle the last step of open()
2684 static int do_last(struct nameidata
*nd
, struct path
*path
,
2685 struct file
*file
, const struct open_flags
*op
,
2686 int *opened
, struct filename
*name
)
2688 struct dentry
*dir
= nd
->path
.dentry
;
2689 int open_flag
= op
->open_flag
;
2690 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2691 bool got_write
= false;
2692 int acc_mode
= op
->acc_mode
;
2693 struct inode
*inode
;
2694 bool symlink_ok
= false;
2695 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2696 bool retried
= false;
2699 nd
->flags
&= ~LOOKUP_PARENT
;
2700 nd
->flags
|= op
->intent
;
2702 switch (nd
->last_type
) {
2705 error
= handle_dots(nd
, nd
->last_type
);
2710 error
= complete_walk(nd
);
2713 audit_inode(name
, nd
->path
.dentry
, 0);
2714 if (open_flag
& O_CREAT
) {
2720 error
= complete_walk(nd
);
2723 audit_inode(name
, dir
, 0);
2727 if (!(open_flag
& O_CREAT
)) {
2728 if (nd
->last
.name
[nd
->last
.len
])
2729 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2730 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2732 /* we _can_ be in RCU mode here */
2733 error
= lookup_fast(nd
, &nd
->last
, path
, &inode
);
2740 BUG_ON(nd
->inode
!= dir
->d_inode
);
2742 /* create side of things */
2744 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2745 * has been cleared when we got to the last component we are
2748 error
= complete_walk(nd
);
2752 audit_inode(name
, dir
, 0);
2754 /* trailing slashes? */
2755 if (nd
->last
.name
[nd
->last
.len
])
2760 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2761 error
= mnt_want_write(nd
->path
.mnt
);
2765 * do _not_ fail yet - we might not need that or fail with
2766 * a different error; let lookup_open() decide; we'll be
2767 * dropping this one anyway.
2770 mutex_lock(&dir
->d_inode
->i_mutex
);
2771 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2772 mutex_unlock(&dir
->d_inode
->i_mutex
);
2778 if ((*opened
& FILE_CREATED
) ||
2779 !S_ISREG(file_inode(file
)->i_mode
))
2780 will_truncate
= false;
2782 audit_inode(name
, file
->f_path
.dentry
, 0);
2786 if (*opened
& FILE_CREATED
) {
2787 /* Don't check for write permission, don't truncate */
2788 open_flag
&= ~O_TRUNC
;
2789 will_truncate
= false;
2790 acc_mode
= MAY_OPEN
;
2791 path_to_nameidata(path
, nd
);
2792 goto finish_open_created
;
2796 * create/update audit record if it already exists.
2798 if (path
->dentry
->d_inode
)
2799 audit_inode(name
, path
->dentry
, 0);
2802 * If atomic_open() acquired write access it is dropped now due to
2803 * possible mount and symlink following (this might be optimized away if
2807 mnt_drop_write(nd
->path
.mnt
);
2812 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
2815 error
= follow_managed(path
, nd
->flags
);
2820 nd
->flags
|= LOOKUP_JUMPED
;
2822 BUG_ON(nd
->flags
& LOOKUP_RCU
);
2823 inode
= path
->dentry
->d_inode
;
2825 /* we _can_ be in RCU mode here */
2828 path_to_nameidata(path
, nd
);
2832 if (should_follow_link(inode
, !symlink_ok
)) {
2833 if (nd
->flags
& LOOKUP_RCU
) {
2834 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
2839 BUG_ON(inode
!= path
->dentry
->d_inode
);
2843 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
2844 path_to_nameidata(path
, nd
);
2846 save_parent
.dentry
= nd
->path
.dentry
;
2847 save_parent
.mnt
= mntget(path
->mnt
);
2848 nd
->path
.dentry
= path
->dentry
;
2852 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2853 error
= complete_walk(nd
);
2855 path_put(&save_parent
);
2859 if ((open_flag
& O_CREAT
) && S_ISDIR(nd
->inode
->i_mode
))
2862 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !nd
->inode
->i_op
->lookup
)
2864 audit_inode(name
, nd
->path
.dentry
, 0);
2866 if (!S_ISREG(nd
->inode
->i_mode
))
2867 will_truncate
= false;
2869 if (will_truncate
) {
2870 error
= mnt_want_write(nd
->path
.mnt
);
2875 finish_open_created
:
2876 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2879 file
->f_path
.mnt
= nd
->path
.mnt
;
2880 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
2882 if (error
== -EOPENSTALE
)
2887 error
= open_check_o_direct(file
);
2890 error
= ima_file_check(file
, op
->acc_mode
);
2894 if (will_truncate
) {
2895 error
= handle_truncate(file
);
2901 mnt_drop_write(nd
->path
.mnt
);
2902 path_put(&save_parent
);
2907 path_put_conditional(path
, nd
);
2914 /* If no saved parent or already retried then can't retry */
2915 if (!save_parent
.dentry
|| retried
)
2918 BUG_ON(save_parent
.dentry
!= dir
);
2919 path_put(&nd
->path
);
2920 nd
->path
= save_parent
;
2921 nd
->inode
= dir
->d_inode
;
2922 save_parent
.mnt
= NULL
;
2923 save_parent
.dentry
= NULL
;
2925 mnt_drop_write(nd
->path
.mnt
);
2932 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
2933 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2935 struct file
*base
= NULL
;
2941 file
= get_empty_filp();
2945 file
->f_flags
= op
->open_flag
;
2947 error
= path_init(dfd
, pathname
->name
, flags
| LOOKUP_PARENT
, nd
, &base
);
2948 if (unlikely(error
))
2951 current
->total_link_count
= 0;
2952 error
= link_path_walk(pathname
->name
, nd
);
2953 if (unlikely(error
))
2956 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2957 while (unlikely(error
> 0)) { /* trailing symlink */
2958 struct path link
= path
;
2960 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2961 path_put_conditional(&path
, nd
);
2962 path_put(&nd
->path
);
2966 error
= may_follow_link(&link
, nd
);
2967 if (unlikely(error
))
2969 nd
->flags
|= LOOKUP_PARENT
;
2970 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2971 error
= follow_link(&link
, nd
, &cookie
);
2972 if (unlikely(error
))
2974 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2975 put_link(nd
, &link
, cookie
);
2978 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2979 path_put(&nd
->root
);
2982 if (!(opened
& FILE_OPENED
)) {
2986 if (unlikely(error
)) {
2987 if (error
== -EOPENSTALE
) {
2988 if (flags
& LOOKUP_RCU
)
2993 file
= ERR_PTR(error
);
2998 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
2999 const struct open_flags
*op
, int flags
)
3001 struct nameidata nd
;
3004 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3005 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3006 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3007 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3008 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3012 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3013 const char *name
, const struct open_flags
*op
, int flags
)
3015 struct nameidata nd
;
3017 struct filename filename
= { .name
= name
};
3020 nd
.root
.dentry
= dentry
;
3022 flags
|= LOOKUP_ROOT
;
3024 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
3025 return ERR_PTR(-ELOOP
);
3027 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3028 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3029 file
= path_openat(-1, &filename
, &nd
, op
, flags
);
3030 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3031 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3035 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3036 struct path
*path
, unsigned int lookup_flags
)
3038 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3039 struct nameidata nd
;
3042 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3045 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3046 * other flags passed in are ignored!
3048 lookup_flags
&= LOOKUP_REVAL
;
3050 error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
|lookup_flags
, &nd
);
3052 return ERR_PTR(error
);
3055 * Yucky last component or no last component at all?
3056 * (foo/., foo/.., /////)
3058 if (nd
.last_type
!= LAST_NORM
)
3060 nd
.flags
&= ~LOOKUP_PARENT
;
3061 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3063 /* don't fail immediately if it's r/o, at least try to report other errors */
3064 err2
= mnt_want_write(nd
.path
.mnt
);
3066 * Do the final lookup.
3068 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3069 dentry
= lookup_hash(&nd
);
3074 if (dentry
->d_inode
)
3077 * Special case - lookup gave negative, but... we had foo/bar/
3078 * From the vfs_mknod() POV we just have a negative dentry -
3079 * all is fine. Let's be bastards - you had / on the end, you've
3080 * been asking for (non-existent) directory. -ENOENT for you.
3082 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3086 if (unlikely(err2
)) {
3094 dentry
= ERR_PTR(error
);
3096 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3098 mnt_drop_write(nd
.path
.mnt
);
3103 EXPORT_SYMBOL(kern_path_create
);
3105 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3108 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3109 mnt_drop_write(path
->mnt
);
3112 EXPORT_SYMBOL(done_path_create
);
3114 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3115 struct path
*path
, unsigned int lookup_flags
)
3117 struct filename
*tmp
= getname(pathname
);
3120 return ERR_CAST(tmp
);
3121 res
= kern_path_create(dfd
, tmp
->name
, path
, lookup_flags
);
3125 EXPORT_SYMBOL(user_path_create
);
3127 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3129 int error
= may_create(dir
, dentry
);
3134 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3137 if (!dir
->i_op
->mknod
)
3140 error
= devcgroup_inode_mknod(mode
, dev
);
3144 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3148 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3150 fsnotify_create(dir
, dentry
);
3154 static int may_mknod(umode_t mode
)
3156 switch (mode
& S_IFMT
) {
3162 case 0: /* zero mode translates to S_IFREG */
3171 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3174 struct dentry
*dentry
;
3177 unsigned int lookup_flags
= 0;
3179 error
= may_mknod(mode
);
3183 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3185 return PTR_ERR(dentry
);
3187 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3188 mode
&= ~current_umask();
3189 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3192 switch (mode
& S_IFMT
) {
3193 case 0: case S_IFREG
:
3194 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3196 case S_IFCHR
: case S_IFBLK
:
3197 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3198 new_decode_dev(dev
));
3200 case S_IFIFO
: case S_IFSOCK
:
3201 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3205 done_path_create(&path
, dentry
);
3206 if (retry_estale(error
, lookup_flags
)) {
3207 lookup_flags
|= LOOKUP_REVAL
;
3213 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3215 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3218 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3220 int error
= may_create(dir
, dentry
);
3221 unsigned max_links
= dir
->i_sb
->s_max_links
;
3226 if (!dir
->i_op
->mkdir
)
3229 mode
&= (S_IRWXUGO
|S_ISVTX
);
3230 error
= security_inode_mkdir(dir
, dentry
, mode
);
3234 if (max_links
&& dir
->i_nlink
>= max_links
)
3237 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3239 fsnotify_mkdir(dir
, dentry
);
3243 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3245 struct dentry
*dentry
;
3248 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3251 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3253 return PTR_ERR(dentry
);
3255 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3256 mode
&= ~current_umask();
3257 error
= security_path_mkdir(&path
, dentry
, mode
);
3259 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3260 done_path_create(&path
, dentry
);
3261 if (retry_estale(error
, lookup_flags
)) {
3262 lookup_flags
|= LOOKUP_REVAL
;
3268 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3270 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3274 * The dentry_unhash() helper will try to drop the dentry early: we
3275 * should have a usage count of 1 if we're the only user of this
3276 * dentry, and if that is true (possibly after pruning the dcache),
3277 * then we drop the dentry now.
3279 * A low-level filesystem can, if it choses, legally
3282 * if (!d_unhashed(dentry))
3285 * if it cannot handle the case of removing a directory
3286 * that is still in use by something else..
3288 void dentry_unhash(struct dentry
*dentry
)
3290 shrink_dcache_parent(dentry
);
3291 spin_lock(&dentry
->d_lock
);
3292 if (dentry
->d_count
== 1)
3294 spin_unlock(&dentry
->d_lock
);
3297 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3299 int error
= may_delete(dir
, dentry
, 1);
3304 if (!dir
->i_op
->rmdir
)
3308 mutex_lock(&dentry
->d_inode
->i_mutex
);
3311 if (d_mountpoint(dentry
))
3314 error
= security_inode_rmdir(dir
, dentry
);
3318 shrink_dcache_parent(dentry
);
3319 error
= dir
->i_op
->rmdir(dir
, dentry
);
3323 dentry
->d_inode
->i_flags
|= S_DEAD
;
3327 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3334 static long do_rmdir(int dfd
, const char __user
*pathname
)
3337 struct filename
*name
;
3338 struct dentry
*dentry
;
3339 struct nameidata nd
;
3340 unsigned int lookup_flags
= 0;
3342 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3344 return PTR_ERR(name
);
3346 switch(nd
.last_type
) {
3358 nd
.flags
&= ~LOOKUP_PARENT
;
3359 error
= mnt_want_write(nd
.path
.mnt
);
3363 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3364 dentry
= lookup_hash(&nd
);
3365 error
= PTR_ERR(dentry
);
3368 if (!dentry
->d_inode
) {
3372 error
= security_path_rmdir(&nd
.path
, dentry
);
3375 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3379 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3380 mnt_drop_write(nd
.path
.mnt
);
3384 if (retry_estale(error
, lookup_flags
)) {
3385 lookup_flags
|= LOOKUP_REVAL
;
3391 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3393 return do_rmdir(AT_FDCWD
, pathname
);
3396 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3398 int error
= may_delete(dir
, dentry
, 0);
3403 if (!dir
->i_op
->unlink
)
3406 mutex_lock(&dentry
->d_inode
->i_mutex
);
3407 if (d_mountpoint(dentry
))
3410 error
= security_inode_unlink(dir
, dentry
);
3412 error
= dir
->i_op
->unlink(dir
, dentry
);
3417 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3419 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3420 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3421 fsnotify_link_count(dentry
->d_inode
);
3429 * Make sure that the actual truncation of the file will occur outside its
3430 * directory's i_mutex. Truncate can take a long time if there is a lot of
3431 * writeout happening, and we don't want to prevent access to the directory
3432 * while waiting on the I/O.
3434 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3437 struct filename
*name
;
3438 struct dentry
*dentry
;
3439 struct nameidata nd
;
3440 struct inode
*inode
= NULL
;
3441 unsigned int lookup_flags
= 0;
3443 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3445 return PTR_ERR(name
);
3448 if (nd
.last_type
!= LAST_NORM
)
3451 nd
.flags
&= ~LOOKUP_PARENT
;
3452 error
= mnt_want_write(nd
.path
.mnt
);
3456 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3457 dentry
= lookup_hash(&nd
);
3458 error
= PTR_ERR(dentry
);
3459 if (!IS_ERR(dentry
)) {
3460 /* Why not before? Because we want correct error value */
3461 if (nd
.last
.name
[nd
.last
.len
])
3463 inode
= dentry
->d_inode
;
3467 error
= security_path_unlink(&nd
.path
, dentry
);
3470 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
3474 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3476 iput(inode
); /* truncate the inode here */
3477 mnt_drop_write(nd
.path
.mnt
);
3481 if (retry_estale(error
, lookup_flags
)) {
3482 lookup_flags
|= LOOKUP_REVAL
;
3489 error
= !dentry
->d_inode
? -ENOENT
:
3490 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
3494 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3496 if ((flag
& ~AT_REMOVEDIR
) != 0)
3499 if (flag
& AT_REMOVEDIR
)
3500 return do_rmdir(dfd
, pathname
);
3502 return do_unlinkat(dfd
, pathname
);
3505 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3507 return do_unlinkat(AT_FDCWD
, pathname
);
3510 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3512 int error
= may_create(dir
, dentry
);
3517 if (!dir
->i_op
->symlink
)
3520 error
= security_inode_symlink(dir
, dentry
, oldname
);
3524 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3526 fsnotify_create(dir
, dentry
);
3530 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3531 int, newdfd
, const char __user
*, newname
)
3534 struct filename
*from
;
3535 struct dentry
*dentry
;
3537 unsigned int lookup_flags
= 0;
3539 from
= getname(oldname
);
3541 return PTR_ERR(from
);
3543 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3544 error
= PTR_ERR(dentry
);
3548 error
= security_path_symlink(&path
, dentry
, from
->name
);
3550 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3551 done_path_create(&path
, dentry
);
3552 if (retry_estale(error
, lookup_flags
)) {
3553 lookup_flags
|= LOOKUP_REVAL
;
3561 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3563 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3566 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3568 struct inode
*inode
= old_dentry
->d_inode
;
3569 unsigned max_links
= dir
->i_sb
->s_max_links
;
3575 error
= may_create(dir
, new_dentry
);
3579 if (dir
->i_sb
!= inode
->i_sb
)
3583 * A link to an append-only or immutable file cannot be created.
3585 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3587 if (!dir
->i_op
->link
)
3589 if (S_ISDIR(inode
->i_mode
))
3592 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3596 mutex_lock(&inode
->i_mutex
);
3597 /* Make sure we don't allow creating hardlink to an unlinked file */
3598 if (inode
->i_nlink
== 0)
3600 else if (max_links
&& inode
->i_nlink
>= max_links
)
3603 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3604 mutex_unlock(&inode
->i_mutex
);
3606 fsnotify_link(dir
, inode
, new_dentry
);
3611 * Hardlinks are often used in delicate situations. We avoid
3612 * security-related surprises by not following symlinks on the
3615 * We don't follow them on the oldname either to be compatible
3616 * with linux 2.0, and to avoid hard-linking to directories
3617 * and other special files. --ADM
3619 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3620 int, newdfd
, const char __user
*, newname
, int, flags
)
3622 struct dentry
*new_dentry
;
3623 struct path old_path
, new_path
;
3627 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3630 * To use null names we require CAP_DAC_READ_SEARCH
3631 * This ensures that not everyone will be able to create
3632 * handlink using the passed filedescriptor.
3634 if (flags
& AT_EMPTY_PATH
) {
3635 if (!capable(CAP_DAC_READ_SEARCH
))
3640 if (flags
& AT_SYMLINK_FOLLOW
)
3641 how
|= LOOKUP_FOLLOW
;
3643 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3647 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
3648 (how
& LOOKUP_REVAL
));
3649 error
= PTR_ERR(new_dentry
);
3650 if (IS_ERR(new_dentry
))
3654 if (old_path
.mnt
!= new_path
.mnt
)
3656 error
= may_linkat(&old_path
);
3657 if (unlikely(error
))
3659 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3662 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3664 done_path_create(&new_path
, new_dentry
);
3665 if (retry_estale(error
, how
)) {
3666 how
|= LOOKUP_REVAL
;
3670 path_put(&old_path
);
3675 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3677 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3681 * The worst of all namespace operations - renaming directory. "Perverted"
3682 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3684 * a) we can get into loop creation. Check is done in is_subdir().
3685 * b) race potential - two innocent renames can create a loop together.
3686 * That's where 4.4 screws up. Current fix: serialization on
3687 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3689 * c) we have to lock _three_ objects - parents and victim (if it exists).
3690 * And that - after we got ->i_mutex on parents (until then we don't know
3691 * whether the target exists). Solution: try to be smart with locking
3692 * order for inodes. We rely on the fact that tree topology may change
3693 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3694 * move will be locked. Thus we can rank directories by the tree
3695 * (ancestors first) and rank all non-directories after them.
3696 * That works since everybody except rename does "lock parent, lookup,
3697 * lock child" and rename is under ->s_vfs_rename_mutex.
3698 * HOWEVER, it relies on the assumption that any object with ->lookup()
3699 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3700 * we'd better make sure that there's no link(2) for them.
3701 * d) conversion from fhandle to dentry may come in the wrong moment - when
3702 * we are removing the target. Solution: we will have to grab ->i_mutex
3703 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3704 * ->i_mutex on parents, which works but leads to some truly excessive
3707 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3708 struct inode
*new_dir
, struct dentry
*new_dentry
)
3711 struct inode
*target
= new_dentry
->d_inode
;
3712 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3715 * If we are going to change the parent - check write permissions,
3716 * we'll need to flip '..'.
3718 if (new_dir
!= old_dir
) {
3719 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3724 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3730 mutex_lock(&target
->i_mutex
);
3733 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3737 if (max_links
&& !target
&& new_dir
!= old_dir
&&
3738 new_dir
->i_nlink
>= max_links
)
3742 shrink_dcache_parent(new_dentry
);
3743 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3748 target
->i_flags
|= S_DEAD
;
3749 dont_mount(new_dentry
);
3753 mutex_unlock(&target
->i_mutex
);
3756 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3757 d_move(old_dentry
,new_dentry
);
3761 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3762 struct inode
*new_dir
, struct dentry
*new_dentry
)
3764 struct inode
*target
= new_dentry
->d_inode
;
3767 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3773 mutex_lock(&target
->i_mutex
);
3776 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3779 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3784 dont_mount(new_dentry
);
3785 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3786 d_move(old_dentry
, new_dentry
);
3789 mutex_unlock(&target
->i_mutex
);
3794 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3795 struct inode
*new_dir
, struct dentry
*new_dentry
)
3798 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3799 const unsigned char *old_name
;
3801 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3804 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3808 if (!new_dentry
->d_inode
)
3809 error
= may_create(new_dir
, new_dentry
);
3811 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3815 if (!old_dir
->i_op
->rename
)
3818 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3821 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3823 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3825 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3826 new_dentry
->d_inode
, old_dentry
);
3827 fsnotify_oldname_free(old_name
);
3832 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3833 int, newdfd
, const char __user
*, newname
)
3835 struct dentry
*old_dir
, *new_dir
;
3836 struct dentry
*old_dentry
, *new_dentry
;
3837 struct dentry
*trap
;
3838 struct nameidata oldnd
, newnd
;
3839 struct filename
*from
;
3840 struct filename
*to
;
3841 unsigned int lookup_flags
= 0;
3842 bool should_retry
= false;
3845 from
= user_path_parent(olddfd
, oldname
, &oldnd
, lookup_flags
);
3847 error
= PTR_ERR(from
);
3851 to
= user_path_parent(newdfd
, newname
, &newnd
, lookup_flags
);
3853 error
= PTR_ERR(to
);
3858 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3861 old_dir
= oldnd
.path
.dentry
;
3863 if (oldnd
.last_type
!= LAST_NORM
)
3866 new_dir
= newnd
.path
.dentry
;
3867 if (newnd
.last_type
!= LAST_NORM
)
3870 error
= mnt_want_write(oldnd
.path
.mnt
);
3874 oldnd
.flags
&= ~LOOKUP_PARENT
;
3875 newnd
.flags
&= ~LOOKUP_PARENT
;
3876 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3878 trap
= lock_rename(new_dir
, old_dir
);
3880 old_dentry
= lookup_hash(&oldnd
);
3881 error
= PTR_ERR(old_dentry
);
3882 if (IS_ERR(old_dentry
))
3884 /* source must exist */
3886 if (!old_dentry
->d_inode
)
3888 /* unless the source is a directory trailing slashes give -ENOTDIR */
3889 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3891 if (oldnd
.last
.name
[oldnd
.last
.len
])
3893 if (newnd
.last
.name
[newnd
.last
.len
])
3896 /* source should not be ancestor of target */
3898 if (old_dentry
== trap
)
3900 new_dentry
= lookup_hash(&newnd
);
3901 error
= PTR_ERR(new_dentry
);
3902 if (IS_ERR(new_dentry
))
3904 /* target should not be an ancestor of source */
3906 if (new_dentry
== trap
)
3909 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3910 &newnd
.path
, new_dentry
);
3913 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3914 new_dir
->d_inode
, new_dentry
);
3920 unlock_rename(new_dir
, old_dir
);
3921 mnt_drop_write(oldnd
.path
.mnt
);
3923 if (retry_estale(error
, lookup_flags
))
3924 should_retry
= true;
3925 path_put(&newnd
.path
);
3928 path_put(&oldnd
.path
);
3931 should_retry
= false;
3932 lookup_flags
|= LOOKUP_REVAL
;
3939 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3941 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3944 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3948 len
= PTR_ERR(link
);
3953 if (len
> (unsigned) buflen
)
3955 if (copy_to_user(buffer
, link
, len
))
3962 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3963 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3964 * using) it for any given inode is up to filesystem.
3966 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3968 struct nameidata nd
;
3973 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3975 return PTR_ERR(cookie
);
3977 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3978 if (dentry
->d_inode
->i_op
->put_link
)
3979 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3983 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3985 return __vfs_follow_link(nd
, link
);
3988 /* get the link contents into pagecache */
3989 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3993 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3994 page
= read_mapping_page(mapping
, 0, NULL
);
3999 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
4003 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4005 struct page
*page
= NULL
;
4006 char *s
= page_getlink(dentry
, &page
);
4007 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
4010 page_cache_release(page
);
4015 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
4017 struct page
*page
= NULL
;
4018 nd_set_link(nd
, page_getlink(dentry
, &page
));
4022 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
4024 struct page
*page
= cookie
;
4028 page_cache_release(page
);
4033 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4035 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4037 struct address_space
*mapping
= inode
->i_mapping
;
4042 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4044 flags
|= AOP_FLAG_NOFS
;
4047 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4048 flags
, &page
, &fsdata
);
4052 kaddr
= kmap_atomic(page
);
4053 memcpy(kaddr
, symname
, len
-1);
4054 kunmap_atomic(kaddr
);
4056 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4063 mark_inode_dirty(inode
);
4069 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4071 return __page_symlink(inode
, symname
, len
,
4072 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4075 const struct inode_operations page_symlink_inode_operations
= {
4076 .readlink
= generic_readlink
,
4077 .follow_link
= page_follow_link_light
,
4078 .put_link
= page_put_link
,
4081 EXPORT_SYMBOL(user_path_at
);
4082 EXPORT_SYMBOL(follow_down_one
);
4083 EXPORT_SYMBOL(follow_down
);
4084 EXPORT_SYMBOL(follow_up
);
4085 EXPORT_SYMBOL(get_write_access
); /* nfsd */
4086 EXPORT_SYMBOL(lock_rename
);
4087 EXPORT_SYMBOL(lookup_one_len
);
4088 EXPORT_SYMBOL(page_follow_link_light
);
4089 EXPORT_SYMBOL(page_put_link
);
4090 EXPORT_SYMBOL(page_readlink
);
4091 EXPORT_SYMBOL(__page_symlink
);
4092 EXPORT_SYMBOL(page_symlink
);
4093 EXPORT_SYMBOL(page_symlink_inode_operations
);
4094 EXPORT_SYMBOL(kern_path
);
4095 EXPORT_SYMBOL(vfs_path_lookup
);
4096 EXPORT_SYMBOL(inode_permission
);
4097 EXPORT_SYMBOL(unlock_rename
);
4098 EXPORT_SYMBOL(vfs_create
);
4099 EXPORT_SYMBOL(vfs_follow_link
);
4100 EXPORT_SYMBOL(vfs_link
);
4101 EXPORT_SYMBOL(vfs_mkdir
);
4102 EXPORT_SYMBOL(vfs_mknod
);
4103 EXPORT_SYMBOL(generic_permission
);
4104 EXPORT_SYMBOL(vfs_readlink
);
4105 EXPORT_SYMBOL(vfs_rename
);
4106 EXPORT_SYMBOL(vfs_rmdir
);
4107 EXPORT_SYMBOL(vfs_symlink
);
4108 EXPORT_SYMBOL(vfs_unlink
);
4109 EXPORT_SYMBOL(dentry_unhash
);
4110 EXPORT_SYMBOL(generic_readlink
);