4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
121 void final_putname(struct filename
*name
)
123 if (name
->separate
) {
124 __putname(name
->name
);
131 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
133 static struct filename
*
134 getname_flags(const char __user
*filename
, int flags
, int *empty
)
136 struct filename
*result
, *err
;
141 result
= audit_reusename(filename
);
145 result
= __getname();
146 if (unlikely(!result
))
147 return ERR_PTR(-ENOMEM
);
150 * First, try to embed the struct filename inside the names_cache
153 kname
= (char *)result
+ sizeof(*result
);
154 result
->name
= kname
;
155 result
->separate
= false;
156 max
= EMBEDDED_NAME_MAX
;
159 len
= strncpy_from_user(kname
, filename
, max
);
160 if (unlikely(len
< 0)) {
166 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
167 * separate struct filename so we can dedicate the entire
168 * names_cache allocation for the pathname, and re-do the copy from
171 if (len
== EMBEDDED_NAME_MAX
&& max
== EMBEDDED_NAME_MAX
) {
172 kname
= (char *)result
;
174 result
= kzalloc(sizeof(*result
), GFP_KERNEL
);
176 err
= ERR_PTR(-ENOMEM
);
177 result
= (struct filename
*)kname
;
180 result
->name
= kname
;
181 result
->separate
= true;
186 /* The empty path is special. */
187 if (unlikely(!len
)) {
190 err
= ERR_PTR(-ENOENT
);
191 if (!(flags
& LOOKUP_EMPTY
))
195 err
= ERR_PTR(-ENAMETOOLONG
);
196 if (unlikely(len
>= PATH_MAX
))
199 result
->uptr
= filename
;
200 audit_getname(result
);
204 final_putname(result
);
209 getname(const char __user
* filename
)
211 return getname_flags(filename
, 0, NULL
);
213 EXPORT_SYMBOL(getname
);
215 #ifdef CONFIG_AUDITSYSCALL
216 void putname(struct filename
*name
)
218 if (unlikely(!audit_dummy_context()))
219 return audit_putname(name
);
224 static int check_acl(struct inode
*inode
, int mask
)
226 #ifdef CONFIG_FS_POSIX_ACL
227 struct posix_acl
*acl
;
229 if (mask
& MAY_NOT_BLOCK
) {
230 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
233 /* no ->get_acl() calls in RCU mode... */
234 if (acl
== ACL_NOT_CACHED
)
236 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
239 acl
= get_cached_acl(inode
, ACL_TYPE_ACCESS
);
242 * A filesystem can force a ACL callback by just never filling the
243 * ACL cache. But normally you'd fill the cache either at inode
244 * instantiation time, or on the first ->get_acl call.
246 * If the filesystem doesn't have a get_acl() function at all, we'll
247 * just create the negative cache entry.
249 if (acl
== ACL_NOT_CACHED
) {
250 if (inode
->i_op
->get_acl
) {
251 acl
= inode
->i_op
->get_acl(inode
, ACL_TYPE_ACCESS
);
255 set_cached_acl(inode
, ACL_TYPE_ACCESS
, NULL
);
261 int error
= posix_acl_permission(inode
, acl
, mask
);
262 posix_acl_release(acl
);
271 * This does the basic permission checking
273 static int acl_permission_check(struct inode
*inode
, int mask
)
275 unsigned int mode
= inode
->i_mode
;
277 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
280 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
281 int error
= check_acl(inode
, mask
);
282 if (error
!= -EAGAIN
)
286 if (in_group_p(inode
->i_gid
))
291 * If the DACs are ok we don't need any capability check.
293 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
299 * generic_permission - check for access rights on a Posix-like filesystem
300 * @inode: inode to check access rights for
301 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
303 * Used to check for read/write/execute permissions on a file.
304 * We use "fsuid" for this, letting us set arbitrary permissions
305 * for filesystem access without changing the "normal" uids which
306 * are used for other things.
308 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
309 * request cannot be satisfied (eg. requires blocking or too much complexity).
310 * It would then be called again in ref-walk mode.
312 int generic_permission(struct inode
*inode
, int mask
)
317 * Do the basic permission checks.
319 ret
= acl_permission_check(inode
, mask
);
323 if (S_ISDIR(inode
->i_mode
)) {
324 /* DACs are overridable for directories */
325 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
327 if (!(mask
& MAY_WRITE
))
328 if (capable_wrt_inode_uidgid(inode
,
329 CAP_DAC_READ_SEARCH
))
334 * Read/write DACs are always overridable.
335 * Executable DACs are overridable when there is
336 * at least one exec bit set.
338 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
339 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
343 * Searching includes executable on directories, else just read.
345 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
346 if (mask
== MAY_READ
)
347 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_READ_SEARCH
))
354 * We _really_ want to just do "generic_permission()" without
355 * even looking at the inode->i_op values. So we keep a cache
356 * flag in inode->i_opflags, that says "this has not special
357 * permission function, use the fast case".
359 static inline int do_inode_permission(struct inode
*inode
, int mask
)
361 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
362 if (likely(inode
->i_op
->permission
))
363 return inode
->i_op
->permission(inode
, mask
);
365 /* This gets set once for the inode lifetime */
366 spin_lock(&inode
->i_lock
);
367 inode
->i_opflags
|= IOP_FASTPERM
;
368 spin_unlock(&inode
->i_lock
);
370 return generic_permission(inode
, mask
);
374 * __inode_permission - Check for access rights to a given inode
375 * @inode: Inode to check permission on
376 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
378 * Check for read/write/execute permissions on an inode.
380 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
382 * This does not check for a read-only file system. You probably want
383 * inode_permission().
385 int __inode_permission(struct inode
*inode
, int mask
)
389 if (unlikely(mask
& MAY_WRITE
)) {
391 * Nobody gets write access to an immutable file.
393 if (IS_IMMUTABLE(inode
))
397 retval
= do_inode_permission(inode
, mask
);
401 retval
= devcgroup_inode_permission(inode
, mask
);
405 return security_inode_permission(inode
, mask
);
409 * sb_permission - Check superblock-level permissions
410 * @sb: Superblock of inode to check permission on
411 * @inode: Inode to check permission on
412 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
414 * Separate out file-system wide checks from inode-specific permission checks.
416 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
418 if (unlikely(mask
& MAY_WRITE
)) {
419 umode_t mode
= inode
->i_mode
;
421 /* Nobody gets write access to a read-only fs. */
422 if ((sb
->s_flags
& MS_RDONLY
) &&
423 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
430 * inode_permission - Check for access rights to a given inode
431 * @inode: Inode to check permission on
432 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
434 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
435 * this, letting us set arbitrary permissions for filesystem access without
436 * changing the "normal" UIDs which are used for other things.
438 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
440 int inode_permission(struct inode
*inode
, int mask
)
444 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
447 return __inode_permission(inode
, mask
);
451 * path_get - get a reference to a path
452 * @path: path to get the reference to
454 * Given a path increment the reference count to the dentry and the vfsmount.
456 void path_get(const struct path
*path
)
461 EXPORT_SYMBOL(path_get
);
464 * path_put - put a reference to a path
465 * @path: path to put the reference to
467 * Given a path decrement the reference count to the dentry and the vfsmount.
469 void path_put(const struct path
*path
)
474 EXPORT_SYMBOL(path_put
);
477 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
478 * @path: nameidate to verify
480 * Rename can sometimes move a file or directory outside of a bind
481 * mount, path_connected allows those cases to be detected.
483 static bool path_connected(const struct path
*path
)
485 struct vfsmount
*mnt
= path
->mnt
;
487 /* Only bind mounts can have disconnected paths */
488 if (mnt
->mnt_root
== mnt
->mnt_sb
->s_root
)
491 return is_subdir(path
->dentry
, mnt
->mnt_root
);
495 * Path walking has 2 modes, rcu-walk and ref-walk (see
496 * Documentation/filesystems/path-lookup.txt). In situations when we can't
497 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
498 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
499 * mode. Refcounts are grabbed at the last known good point before rcu-walk
500 * got stuck, so ref-walk may continue from there. If this is not successful
501 * (eg. a seqcount has changed), then failure is returned and it's up to caller
502 * to restart the path walk from the beginning in ref-walk mode.
505 static inline void lock_rcu_walk(void)
507 br_read_lock(&vfsmount_lock
);
511 static inline void unlock_rcu_walk(void)
514 br_read_unlock(&vfsmount_lock
);
518 * unlazy_walk - try to switch to ref-walk mode.
519 * @nd: nameidata pathwalk data
520 * @dentry: child of nd->path.dentry or NULL
521 * Returns: 0 on success, -ECHILD on failure
523 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
524 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
525 * @nd or NULL. Must be called from rcu-walk context.
527 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
529 struct fs_struct
*fs
= current
->fs
;
530 struct dentry
*parent
= nd
->path
.dentry
;
533 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
534 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
536 spin_lock(&fs
->lock
);
537 if (nd
->root
.mnt
!= fs
->root
.mnt
||
538 nd
->root
.dentry
!= fs
->root
.dentry
)
541 spin_lock(&parent
->d_lock
);
543 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
545 BUG_ON(nd
->inode
!= parent
->d_inode
);
547 if (dentry
->d_parent
!= parent
)
549 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
550 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
553 * If the sequence check on the child dentry passed, then
554 * the child has not been removed from its parent. This
555 * means the parent dentry must be valid and able to take
556 * a reference at this point.
558 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
559 BUG_ON(!parent
->d_count
);
561 spin_unlock(&dentry
->d_lock
);
563 spin_unlock(&parent
->d_lock
);
566 spin_unlock(&fs
->lock
);
568 mntget(nd
->path
.mnt
);
571 nd
->flags
&= ~LOOKUP_RCU
;
575 spin_unlock(&dentry
->d_lock
);
577 spin_unlock(&parent
->d_lock
);
580 spin_unlock(&fs
->lock
);
584 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
586 return dentry
->d_op
->d_revalidate(dentry
, flags
);
590 * complete_walk - successful completion of path walk
591 * @nd: pointer nameidata
593 * If we had been in RCU mode, drop out of it and legitimize nd->path.
594 * Revalidate the final result, unless we'd already done that during
595 * the path walk or the filesystem doesn't ask for it. Return 0 on
596 * success, -error on failure. In case of failure caller does not
597 * need to drop nd->path.
599 static int complete_walk(struct nameidata
*nd
)
601 struct dentry
*dentry
= nd
->path
.dentry
;
604 if (nd
->flags
& LOOKUP_RCU
) {
605 nd
->flags
&= ~LOOKUP_RCU
;
606 if (!(nd
->flags
& LOOKUP_ROOT
))
608 spin_lock(&dentry
->d_lock
);
609 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
610 spin_unlock(&dentry
->d_lock
);
614 BUG_ON(nd
->inode
!= dentry
->d_inode
);
615 spin_unlock(&dentry
->d_lock
);
616 mntget(nd
->path
.mnt
);
620 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
623 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
626 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
637 static __always_inline
void set_root(struct nameidata
*nd
)
640 get_fs_root(current
->fs
, &nd
->root
);
643 static int link_path_walk(const char *, struct nameidata
*);
645 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
648 struct fs_struct
*fs
= current
->fs
;
652 seq
= read_seqcount_begin(&fs
->seq
);
654 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
655 } while (read_seqcount_retry(&fs
->seq
, seq
));
659 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
671 nd
->flags
|= LOOKUP_JUMPED
;
673 nd
->inode
= nd
->path
.dentry
->d_inode
;
675 ret
= link_path_walk(link
, nd
);
679 return PTR_ERR(link
);
682 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
685 if (path
->mnt
!= nd
->path
.mnt
)
689 static inline void path_to_nameidata(const struct path
*path
,
690 struct nameidata
*nd
)
692 if (!(nd
->flags
& LOOKUP_RCU
)) {
693 dput(nd
->path
.dentry
);
694 if (nd
->path
.mnt
!= path
->mnt
)
695 mntput(nd
->path
.mnt
);
697 nd
->path
.mnt
= path
->mnt
;
698 nd
->path
.dentry
= path
->dentry
;
702 * Helper to directly jump to a known parsed path from ->follow_link,
703 * caller must have taken a reference to path beforehand.
705 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
710 nd
->inode
= nd
->path
.dentry
->d_inode
;
711 nd
->flags
|= LOOKUP_JUMPED
;
714 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
716 struct inode
*inode
= link
->dentry
->d_inode
;
717 if (inode
->i_op
->put_link
)
718 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
722 int sysctl_protected_symlinks __read_mostly
= 0;
723 int sysctl_protected_hardlinks __read_mostly
= 0;
726 * may_follow_link - Check symlink following for unsafe situations
727 * @link: The path of the symlink
728 * @nd: nameidata pathwalk data
730 * In the case of the sysctl_protected_symlinks sysctl being enabled,
731 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
732 * in a sticky world-writable directory. This is to protect privileged
733 * processes from failing races against path names that may change out
734 * from under them by way of other users creating malicious symlinks.
735 * It will permit symlinks to be followed only when outside a sticky
736 * world-writable directory, or when the uid of the symlink and follower
737 * match, or when the directory owner matches the symlink's owner.
739 * Returns 0 if following the symlink is allowed, -ve on error.
741 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
743 const struct inode
*inode
;
744 const struct inode
*parent
;
746 if (!sysctl_protected_symlinks
)
749 /* Allowed if owner and follower match. */
750 inode
= link
->dentry
->d_inode
;
751 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
754 /* Allowed if parent directory not sticky and world-writable. */
755 parent
= nd
->path
.dentry
->d_inode
;
756 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
759 /* Allowed if parent directory and link owner match. */
760 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
763 audit_log_link_denied("follow_link", link
);
764 path_put_conditional(link
, nd
);
770 * safe_hardlink_source - Check for safe hardlink conditions
771 * @inode: the source inode to hardlink from
773 * Return false if at least one of the following conditions:
774 * - inode is not a regular file
776 * - inode is setgid and group-exec
777 * - access failure for read and write
779 * Otherwise returns true.
781 static bool safe_hardlink_source(struct inode
*inode
)
783 umode_t mode
= inode
->i_mode
;
785 /* Special files should not get pinned to the filesystem. */
789 /* Setuid files should not get pinned to the filesystem. */
793 /* Executable setgid files should not get pinned to the filesystem. */
794 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
797 /* Hardlinking to unreadable or unwritable sources is dangerous. */
798 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
805 * may_linkat - Check permissions for creating a hardlink
806 * @link: the source to hardlink from
808 * Block hardlink when all of:
809 * - sysctl_protected_hardlinks enabled
810 * - fsuid does not match inode
811 * - hardlink source is unsafe (see safe_hardlink_source() above)
814 * Returns 0 if successful, -ve on error.
816 static int may_linkat(struct path
*link
)
818 const struct cred
*cred
;
821 if (!sysctl_protected_hardlinks
)
824 cred
= current_cred();
825 inode
= link
->dentry
->d_inode
;
827 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
828 * otherwise, it must be a safe source.
830 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
834 audit_log_link_denied("linkat", link
);
838 static __always_inline
int
839 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
841 struct dentry
*dentry
= link
->dentry
;
845 BUG_ON(nd
->flags
& LOOKUP_RCU
);
847 if (link
->mnt
== nd
->path
.mnt
)
851 if (unlikely(current
->total_link_count
>= 40))
852 goto out_put_nd_path
;
855 current
->total_link_count
++;
858 nd_set_link(nd
, NULL
);
860 error
= security_inode_follow_link(link
->dentry
, nd
);
862 goto out_put_nd_path
;
864 nd
->last_type
= LAST_BIND
;
865 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
868 goto out_put_nd_path
;
873 error
= __vfs_follow_link(nd
, s
);
875 put_link(nd
, link
, *p
);
887 static int follow_up_rcu(struct path
*path
)
889 struct mount
*mnt
= real_mount(path
->mnt
);
890 struct mount
*parent
;
891 struct dentry
*mountpoint
;
893 parent
= mnt
->mnt_parent
;
894 if (&parent
->mnt
== path
->mnt
)
896 mountpoint
= mnt
->mnt_mountpoint
;
897 path
->dentry
= mountpoint
;
898 path
->mnt
= &parent
->mnt
;
903 * follow_up - Find the mountpoint of path's vfsmount
905 * Given a path, find the mountpoint of its source file system.
906 * Replace @path with the path of the mountpoint in the parent mount.
909 * Return 1 if we went up a level and 0 if we were already at the
912 int follow_up(struct path
*path
)
914 struct mount
*mnt
= real_mount(path
->mnt
);
915 struct mount
*parent
;
916 struct dentry
*mountpoint
;
918 br_read_lock(&vfsmount_lock
);
919 parent
= mnt
->mnt_parent
;
921 br_read_unlock(&vfsmount_lock
);
924 mntget(&parent
->mnt
);
925 mountpoint
= dget(mnt
->mnt_mountpoint
);
926 br_read_unlock(&vfsmount_lock
);
928 path
->dentry
= mountpoint
;
930 path
->mnt
= &parent
->mnt
;
935 * Perform an automount
936 * - return -EISDIR to tell follow_managed() to stop and return the path we
939 static int follow_automount(struct path
*path
, unsigned flags
,
942 struct vfsmount
*mnt
;
945 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
948 /* We don't want to mount if someone's just doing a stat -
949 * unless they're stat'ing a directory and appended a '/' to
952 * We do, however, want to mount if someone wants to open or
953 * create a file of any type under the mountpoint, wants to
954 * traverse through the mountpoint or wants to open the
955 * mounted directory. Also, autofs may mark negative dentries
956 * as being automount points. These will need the attentions
957 * of the daemon to instantiate them before they can be used.
959 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
960 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
961 path
->dentry
->d_inode
)
964 current
->total_link_count
++;
965 if (current
->total_link_count
>= 40)
968 mnt
= path
->dentry
->d_op
->d_automount(path
);
971 * The filesystem is allowed to return -EISDIR here to indicate
972 * it doesn't want to automount. For instance, autofs would do
973 * this so that its userspace daemon can mount on this dentry.
975 * However, we can only permit this if it's a terminal point in
976 * the path being looked up; if it wasn't then the remainder of
977 * the path is inaccessible and we should say so.
979 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
984 if (!mnt
) /* mount collision */
988 /* lock_mount() may release path->mnt on error */
992 err
= finish_automount(mnt
, path
);
996 /* Someone else made a mount here whilst we were busy */
1001 path
->dentry
= dget(mnt
->mnt_root
);
1010 * Handle a dentry that is managed in some way.
1011 * - Flagged for transit management (autofs)
1012 * - Flagged as mountpoint
1013 * - Flagged as automount point
1015 * This may only be called in refwalk mode.
1017 * Serialization is taken care of in namespace.c
1019 static int follow_managed(struct path
*path
, unsigned flags
)
1021 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1023 bool need_mntput
= false;
1026 /* Given that we're not holding a lock here, we retain the value in a
1027 * local variable for each dentry as we look at it so that we don't see
1028 * the components of that value change under us */
1029 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1030 managed
&= DCACHE_MANAGED_DENTRY
,
1031 unlikely(managed
!= 0)) {
1032 /* Allow the filesystem to manage the transit without i_mutex
1034 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1035 BUG_ON(!path
->dentry
->d_op
);
1036 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1037 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1042 /* Transit to a mounted filesystem. */
1043 if (managed
& DCACHE_MOUNTED
) {
1044 struct vfsmount
*mounted
= lookup_mnt(path
);
1049 path
->mnt
= mounted
;
1050 path
->dentry
= dget(mounted
->mnt_root
);
1055 /* Something is mounted on this dentry in another
1056 * namespace and/or whatever was mounted there in this
1057 * namespace got unmounted before we managed to get the
1061 /* Handle an automount point */
1062 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1063 ret
= follow_automount(path
, flags
, &need_mntput
);
1069 /* We didn't change the current path point */
1073 if (need_mntput
&& path
->mnt
== mnt
)
1077 return ret
< 0 ? ret
: need_mntput
;
1080 int follow_down_one(struct path
*path
)
1082 struct vfsmount
*mounted
;
1084 mounted
= lookup_mnt(path
);
1088 path
->mnt
= mounted
;
1089 path
->dentry
= dget(mounted
->mnt_root
);
1095 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
1097 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
1098 dentry
->d_op
->d_manage(dentry
, true) < 0);
1102 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1103 * we meet a managed dentry that would need blocking.
1105 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1106 struct inode
**inode
)
1109 struct mount
*mounted
;
1111 * Don't forget we might have a non-mountpoint managed dentry
1112 * that wants to block transit.
1114 if (unlikely(managed_dentry_might_block(path
->dentry
)))
1117 if (!d_mountpoint(path
->dentry
))
1120 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1123 path
->mnt
= &mounted
->mnt
;
1124 path
->dentry
= mounted
->mnt
.mnt_root
;
1125 nd
->flags
|= LOOKUP_JUMPED
;
1126 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1128 * Update the inode too. We don't need to re-check the
1129 * dentry sequence number here after this d_inode read,
1130 * because a mount-point is always pinned.
1132 *inode
= path
->dentry
->d_inode
;
1137 static void follow_mount_rcu(struct nameidata
*nd
)
1139 while (d_mountpoint(nd
->path
.dentry
)) {
1140 struct mount
*mounted
;
1141 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
1144 nd
->path
.mnt
= &mounted
->mnt
;
1145 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1146 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1150 static int follow_dotdot_rcu(struct nameidata
*nd
)
1155 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1156 nd
->path
.mnt
== nd
->root
.mnt
) {
1159 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1160 struct dentry
*old
= nd
->path
.dentry
;
1161 struct dentry
*parent
= old
->d_parent
;
1164 seq
= read_seqcount_begin(&parent
->d_seq
);
1165 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1167 nd
->path
.dentry
= parent
;
1169 if (unlikely(!path_connected(&nd
->path
)))
1173 if (!follow_up_rcu(&nd
->path
))
1175 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1177 follow_mount_rcu(nd
);
1178 nd
->inode
= nd
->path
.dentry
->d_inode
;
1182 nd
->flags
&= ~LOOKUP_RCU
;
1183 if (!(nd
->flags
& LOOKUP_ROOT
))
1184 nd
->root
.mnt
= NULL
;
1190 * Follow down to the covering mount currently visible to userspace. At each
1191 * point, the filesystem owning that dentry may be queried as to whether the
1192 * caller is permitted to proceed or not.
1194 int follow_down(struct path
*path
)
1199 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1200 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1201 /* Allow the filesystem to manage the transit without i_mutex
1204 * We indicate to the filesystem if someone is trying to mount
1205 * something here. This gives autofs the chance to deny anyone
1206 * other than its daemon the right to mount on its
1209 * The filesystem may sleep at this point.
1211 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1212 BUG_ON(!path
->dentry
->d_op
);
1213 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1214 ret
= path
->dentry
->d_op
->d_manage(
1215 path
->dentry
, false);
1217 return ret
== -EISDIR
? 0 : ret
;
1220 /* Transit to a mounted filesystem. */
1221 if (managed
& DCACHE_MOUNTED
) {
1222 struct vfsmount
*mounted
= lookup_mnt(path
);
1227 path
->mnt
= mounted
;
1228 path
->dentry
= dget(mounted
->mnt_root
);
1232 /* Don't handle automount points here */
1239 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1241 static void follow_mount(struct path
*path
)
1243 while (d_mountpoint(path
->dentry
)) {
1244 struct vfsmount
*mounted
= lookup_mnt(path
);
1249 path
->mnt
= mounted
;
1250 path
->dentry
= dget(mounted
->mnt_root
);
1254 static int follow_dotdot(struct nameidata
*nd
)
1259 struct dentry
*old
= nd
->path
.dentry
;
1261 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1262 nd
->path
.mnt
== nd
->root
.mnt
) {
1265 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1266 /* rare case of legitimate dget_parent()... */
1267 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1269 if (unlikely(!path_connected(&nd
->path
))) {
1270 path_put(&nd
->path
);
1275 if (!follow_up(&nd
->path
))
1278 follow_mount(&nd
->path
);
1279 nd
->inode
= nd
->path
.dentry
->d_inode
;
1284 * This looks up the name in dcache, possibly revalidates the old dentry and
1285 * allocates a new one if not found or not valid. In the need_lookup argument
1286 * returns whether i_op->lookup is necessary.
1288 * dir->d_inode->i_mutex must be held
1290 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1291 unsigned int flags
, bool *need_lookup
)
1293 struct dentry
*dentry
;
1296 *need_lookup
= false;
1297 dentry
= d_lookup(dir
, name
);
1299 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1300 error
= d_revalidate(dentry
, flags
);
1301 if (unlikely(error
<= 0)) {
1304 return ERR_PTR(error
);
1305 } else if (!d_invalidate(dentry
)) {
1314 dentry
= d_alloc(dir
, name
);
1315 if (unlikely(!dentry
))
1316 return ERR_PTR(-ENOMEM
);
1318 *need_lookup
= true;
1324 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1325 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1327 * dir->d_inode->i_mutex must be held
1329 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1334 /* Don't create child dentry for a dead directory. */
1335 if (unlikely(IS_DEADDIR(dir
))) {
1337 return ERR_PTR(-ENOENT
);
1340 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1341 if (unlikely(old
)) {
1348 static struct dentry
*__lookup_hash(struct qstr
*name
,
1349 struct dentry
*base
, unsigned int flags
)
1352 struct dentry
*dentry
;
1354 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1358 return lookup_real(base
->d_inode
, dentry
, flags
);
1362 * It's more convoluted than I'd like it to be, but... it's still fairly
1363 * small and for now I'd prefer to have fast path as straight as possible.
1364 * It _is_ time-critical.
1366 static int lookup_fast(struct nameidata
*nd
,
1367 struct path
*path
, struct inode
**inode
)
1369 struct vfsmount
*mnt
= nd
->path
.mnt
;
1370 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1376 * Rename seqlock is not required here because in the off chance
1377 * of a false negative due to a concurrent rename, we're going to
1378 * do the non-racy lookup, below.
1380 if (nd
->flags
& LOOKUP_RCU
) {
1382 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
, nd
->inode
);
1387 * This sequence count validates that the inode matches
1388 * the dentry name information from lookup.
1390 *inode
= dentry
->d_inode
;
1391 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1395 * This sequence count validates that the parent had no
1396 * changes while we did the lookup of the dentry above.
1398 * The memory barrier in read_seqcount_begin of child is
1399 * enough, we can use __read_seqcount_retry here.
1401 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1405 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1406 status
= d_revalidate(dentry
, nd
->flags
);
1407 if (unlikely(status
<= 0)) {
1408 if (status
!= -ECHILD
)
1414 path
->dentry
= dentry
;
1415 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1417 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1421 if (unlazy_walk(nd
, dentry
))
1424 dentry
= __d_lookup(parent
, &nd
->last
);
1427 if (unlikely(!dentry
))
1430 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1431 status
= d_revalidate(dentry
, nd
->flags
);
1432 if (unlikely(status
<= 0)) {
1437 if (!d_invalidate(dentry
)) {
1444 path
->dentry
= dentry
;
1445 err
= follow_managed(path
, nd
->flags
);
1446 if (unlikely(err
< 0)) {
1447 path_put_conditional(path
, nd
);
1451 nd
->flags
|= LOOKUP_JUMPED
;
1452 *inode
= path
->dentry
->d_inode
;
1459 /* Fast lookup failed, do it the slow way */
1460 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1462 struct dentry
*dentry
, *parent
;
1465 parent
= nd
->path
.dentry
;
1466 BUG_ON(nd
->inode
!= parent
->d_inode
);
1468 mutex_lock(&parent
->d_inode
->i_mutex
);
1469 dentry
= __lookup_hash(&nd
->last
, parent
, nd
->flags
);
1470 mutex_unlock(&parent
->d_inode
->i_mutex
);
1472 return PTR_ERR(dentry
);
1473 path
->mnt
= nd
->path
.mnt
;
1474 path
->dentry
= dentry
;
1475 err
= follow_managed(path
, nd
->flags
);
1476 if (unlikely(err
< 0)) {
1477 path_put_conditional(path
, nd
);
1481 nd
->flags
|= LOOKUP_JUMPED
;
1485 static inline int may_lookup(struct nameidata
*nd
)
1487 if (nd
->flags
& LOOKUP_RCU
) {
1488 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1491 if (unlazy_walk(nd
, NULL
))
1494 return inode_permission(nd
->inode
, MAY_EXEC
);
1497 static inline int handle_dots(struct nameidata
*nd
, int type
)
1499 if (type
== LAST_DOTDOT
) {
1500 if (nd
->flags
& LOOKUP_RCU
) {
1501 if (follow_dotdot_rcu(nd
))
1504 return follow_dotdot(nd
);
1509 static void terminate_walk(struct nameidata
*nd
)
1511 if (!(nd
->flags
& LOOKUP_RCU
)) {
1512 path_put(&nd
->path
);
1514 nd
->flags
&= ~LOOKUP_RCU
;
1515 if (!(nd
->flags
& LOOKUP_ROOT
))
1516 nd
->root
.mnt
= NULL
;
1522 * Do we need to follow links? We _really_ want to be able
1523 * to do this check without having to look at inode->i_op,
1524 * so we keep a cache of "no, this doesn't need follow_link"
1525 * for the common case.
1527 static inline int should_follow_link(struct inode
*inode
, int follow
)
1529 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1530 if (likely(inode
->i_op
->follow_link
))
1533 /* This gets set once for the inode lifetime */
1534 spin_lock(&inode
->i_lock
);
1535 inode
->i_opflags
|= IOP_NOFOLLOW
;
1536 spin_unlock(&inode
->i_lock
);
1541 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1544 struct inode
*inode
;
1547 * "." and ".." are special - ".." especially so because it has
1548 * to be able to know about the current root directory and
1549 * parent relationships.
1551 if (unlikely(nd
->last_type
!= LAST_NORM
))
1552 return handle_dots(nd
, nd
->last_type
);
1553 err
= lookup_fast(nd
, path
, &inode
);
1554 if (unlikely(err
)) {
1558 err
= lookup_slow(nd
, path
);
1562 inode
= path
->dentry
->d_inode
;
1568 if (should_follow_link(inode
, follow
)) {
1569 if (nd
->flags
& LOOKUP_RCU
) {
1570 if (unlikely(nd
->path
.mnt
!= path
->mnt
||
1571 unlazy_walk(nd
, path
->dentry
))) {
1576 BUG_ON(inode
!= path
->dentry
->d_inode
);
1579 path_to_nameidata(path
, nd
);
1584 path_to_nameidata(path
, nd
);
1591 * This limits recursive symlink follows to 8, while
1592 * limiting consecutive symlinks to 40.
1594 * Without that kind of total limit, nasty chains of consecutive
1595 * symlinks can cause almost arbitrarily long lookups.
1597 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1601 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1602 path_put_conditional(path
, nd
);
1603 path_put(&nd
->path
);
1606 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1609 current
->link_count
++;
1612 struct path link
= *path
;
1615 res
= follow_link(&link
, nd
, &cookie
);
1618 res
= walk_component(nd
, path
, LOOKUP_FOLLOW
);
1619 put_link(nd
, &link
, cookie
);
1622 current
->link_count
--;
1628 * We really don't want to look at inode->i_op->lookup
1629 * when we don't have to. So we keep a cache bit in
1630 * the inode ->i_opflags field that says "yes, we can
1631 * do lookup on this inode".
1633 static inline int can_lookup(struct inode
*inode
)
1635 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1637 if (likely(!inode
->i_op
->lookup
))
1640 /* We do this once for the lifetime of the inode */
1641 spin_lock(&inode
->i_lock
);
1642 inode
->i_opflags
|= IOP_LOOKUP
;
1643 spin_unlock(&inode
->i_lock
);
1648 * We can do the critical dentry name comparison and hashing
1649 * operations one word at a time, but we are limited to:
1651 * - Architectures with fast unaligned word accesses. We could
1652 * do a "get_unaligned()" if this helps and is sufficiently
1655 * - Little-endian machines (so that we can generate the mask
1656 * of low bytes efficiently). Again, we *could* do a byte
1657 * swapping load on big-endian architectures if that is not
1658 * expensive enough to make the optimization worthless.
1660 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1661 * do not trap on the (extremely unlikely) case of a page
1662 * crossing operation.
1664 * - Furthermore, we need an efficient 64-bit compile for the
1665 * 64-bit case in order to generate the "number of bytes in
1666 * the final mask". Again, that could be replaced with a
1667 * efficient population count instruction or similar.
1669 #ifdef CONFIG_DCACHE_WORD_ACCESS
1671 #include <asm/word-at-a-time.h>
1675 static inline unsigned int fold_hash(unsigned long hash
)
1677 return hash_64(hash
, 32);
1680 #else /* 32-bit case */
1682 #define fold_hash(x) (x)
1686 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1688 unsigned long a
, mask
;
1689 unsigned long hash
= 0;
1692 a
= load_unaligned_zeropad(name
);
1693 if (len
< sizeof(unsigned long))
1697 name
+= sizeof(unsigned long);
1698 len
-= sizeof(unsigned long);
1702 mask
= ~(~0ul << len
*8);
1705 return fold_hash(hash
);
1707 EXPORT_SYMBOL(full_name_hash
);
1710 * Calculate the length and hash of the path component, and
1711 * return the length of the component;
1713 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1715 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1716 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1719 len
= -sizeof(unsigned long);
1721 hash
= (hash
+ a
) * 9;
1722 len
+= sizeof(unsigned long);
1723 a
= load_unaligned_zeropad(name
+len
);
1724 b
= a
^ REPEAT_BYTE('/');
1725 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1727 adata
= prep_zero_mask(a
, adata
, &constants
);
1728 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1730 mask
= create_zero_mask(adata
| bdata
);
1732 hash
+= a
& zero_bytemask(mask
);
1733 *hashp
= fold_hash(hash
);
1735 return len
+ find_zero(mask
);
1740 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1742 unsigned long hash
= init_name_hash();
1744 hash
= partial_name_hash(*name
++, hash
);
1745 return end_name_hash(hash
);
1747 EXPORT_SYMBOL(full_name_hash
);
1750 * We know there's a real path component here of at least
1753 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1755 unsigned long hash
= init_name_hash();
1756 unsigned long len
= 0, c
;
1758 c
= (unsigned char)*name
;
1761 hash
= partial_name_hash(c
, hash
);
1762 c
= (unsigned char)name
[len
];
1763 } while (c
&& c
!= '/');
1764 *hashp
= end_name_hash(hash
);
1772 * This is the basic name resolution function, turning a pathname into
1773 * the final dentry. We expect 'base' to be positive and a directory.
1775 * Returns 0 and nd will have valid dentry and mnt on success.
1776 * Returns error and drops reference to input namei data on failure.
1778 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1788 /* At this point we know we have a real path component. */
1794 err
= may_lookup(nd
);
1798 len
= hash_name(name
, &this.hash
);
1803 if (name
[0] == '.') switch (len
) {
1805 if (name
[1] == '.') {
1807 nd
->flags
|= LOOKUP_JUMPED
;
1813 if (likely(type
== LAST_NORM
)) {
1814 struct dentry
*parent
= nd
->path
.dentry
;
1815 nd
->flags
&= ~LOOKUP_JUMPED
;
1816 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1817 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1825 nd
->last_type
= type
;
1830 * If it wasn't NUL, we know it was '/'. Skip that
1831 * slash, and continue until no more slashes.
1835 } while (unlikely(name
[len
] == '/'));
1841 err
= walk_component(nd
, &next
, LOOKUP_FOLLOW
);
1846 err
= nested_symlink(&next
, nd
);
1850 if (!can_lookup(nd
->inode
)) {
1859 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1860 struct nameidata
*nd
, struct file
**fp
)
1864 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1865 nd
->flags
= flags
| LOOKUP_JUMPED
;
1867 if (flags
& LOOKUP_ROOT
) {
1868 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1870 if (!can_lookup(inode
))
1872 retval
= inode_permission(inode
, MAY_EXEC
);
1876 nd
->path
= nd
->root
;
1878 if (flags
& LOOKUP_RCU
) {
1880 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1882 path_get(&nd
->path
);
1887 nd
->root
.mnt
= NULL
;
1890 if (flags
& LOOKUP_RCU
) {
1895 path_get(&nd
->root
);
1897 nd
->path
= nd
->root
;
1898 } else if (dfd
== AT_FDCWD
) {
1899 if (flags
& LOOKUP_RCU
) {
1900 struct fs_struct
*fs
= current
->fs
;
1906 seq
= read_seqcount_begin(&fs
->seq
);
1908 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1909 } while (read_seqcount_retry(&fs
->seq
, seq
));
1911 get_fs_pwd(current
->fs
, &nd
->path
);
1914 /* Caller must check execute permissions on the starting path component */
1915 struct fd f
= fdget_raw(dfd
);
1916 struct dentry
*dentry
;
1921 dentry
= f
.file
->f_path
.dentry
;
1924 if (!can_lookup(dentry
->d_inode
)) {
1930 nd
->path
= f
.file
->f_path
;
1931 if (flags
& LOOKUP_RCU
) {
1934 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1937 path_get(&nd
->path
);
1942 nd
->inode
= nd
->path
.dentry
->d_inode
;
1946 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1948 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1949 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1951 nd
->flags
&= ~LOOKUP_PARENT
;
1952 return walk_component(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
);
1955 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1956 static int path_lookupat(int dfd
, const char *name
,
1957 unsigned int flags
, struct nameidata
*nd
)
1959 struct file
*base
= NULL
;
1964 * Path walking is largely split up into 2 different synchronisation
1965 * schemes, rcu-walk and ref-walk (explained in
1966 * Documentation/filesystems/path-lookup.txt). These share much of the
1967 * path walk code, but some things particularly setup, cleanup, and
1968 * following mounts are sufficiently divergent that functions are
1969 * duplicated. Typically there is a function foo(), and its RCU
1970 * analogue, foo_rcu().
1972 * -ECHILD is the error number of choice (just to avoid clashes) that
1973 * is returned if some aspect of an rcu-walk fails. Such an error must
1974 * be handled by restarting a traditional ref-walk (which will always
1975 * be able to complete).
1977 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1982 current
->total_link_count
= 0;
1983 err
= link_path_walk(name
, nd
);
1985 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1986 err
= lookup_last(nd
, &path
);
1989 struct path link
= path
;
1990 err
= may_follow_link(&link
, nd
);
1993 nd
->flags
|= LOOKUP_PARENT
;
1994 err
= follow_link(&link
, nd
, &cookie
);
1997 err
= lookup_last(nd
, &path
);
1998 put_link(nd
, &link
, cookie
);
2003 err
= complete_walk(nd
);
2005 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
2006 if (!can_lookup(nd
->inode
)) {
2007 path_put(&nd
->path
);
2013 struct super_block
*sb
= nd
->inode
->i_sb
;
2014 if (sb
->s_flags
& MS_RDONLY
) {
2015 if (d_is_su(nd
->path
.dentry
) && !su_visible()) {
2016 path_put(&nd
->path
);
2025 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
2026 path_put(&nd
->root
);
2027 nd
->root
.mnt
= NULL
;
2032 static int filename_lookup(int dfd
, struct filename
*name
,
2033 unsigned int flags
, struct nameidata
*nd
)
2035 int retval
= path_lookupat(dfd
, name
->name
, flags
| LOOKUP_RCU
, nd
);
2036 if (unlikely(retval
== -ECHILD
))
2037 retval
= path_lookupat(dfd
, name
->name
, flags
, nd
);
2038 if (unlikely(retval
== -ESTALE
))
2039 retval
= path_lookupat(dfd
, name
->name
,
2040 flags
| LOOKUP_REVAL
, nd
);
2042 if (likely(!retval
))
2043 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2047 static int do_path_lookup(int dfd
, const char *name
,
2048 unsigned int flags
, struct nameidata
*nd
)
2050 struct filename filename
= { .name
= name
};
2052 return filename_lookup(dfd
, &filename
, flags
, nd
);
2055 /* does lookup, returns the object with parent locked */
2056 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2058 struct nameidata nd
;
2060 int err
= do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, &nd
);
2062 return ERR_PTR(err
);
2063 if (nd
.last_type
!= LAST_NORM
) {
2065 return ERR_PTR(-EINVAL
);
2067 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2068 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2070 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2078 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2080 struct nameidata nd
;
2081 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
2088 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2089 * @dentry: pointer to dentry of the base directory
2090 * @mnt: pointer to vfs mount of the base directory
2091 * @name: pointer to file name
2092 * @flags: lookup flags
2093 * @path: pointer to struct path to fill
2095 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2096 const char *name
, unsigned int flags
,
2099 struct nameidata nd
;
2101 nd
.root
.dentry
= dentry
;
2103 BUG_ON(flags
& LOOKUP_PARENT
);
2104 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2105 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
2112 * Restricted form of lookup. Doesn't follow links, single-component only,
2113 * needs parent already locked. Doesn't follow mounts.
2116 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2118 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2122 * lookup_one_len - filesystem helper to lookup single pathname component
2123 * @name: pathname component to lookup
2124 * @base: base directory to lookup from
2125 * @len: maximum length @len should be interpreted to
2127 * Note that this routine is purely a helper for filesystem usage and should
2128 * not be called by generic code. Also note that by using this function the
2129 * nameidata argument is passed to the filesystem methods and a filesystem
2130 * using this helper needs to be prepared for that.
2132 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2138 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2142 this.hash
= full_name_hash(name
, len
);
2144 return ERR_PTR(-EACCES
);
2146 if (unlikely(name
[0] == '.')) {
2147 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2148 return ERR_PTR(-EACCES
);
2152 c
= *(const unsigned char *)name
++;
2153 if (c
== '/' || c
== '\0')
2154 return ERR_PTR(-EACCES
);
2157 * See if the low-level filesystem might want
2158 * to use its own hash..
2160 if (base
->d_flags
& DCACHE_OP_HASH
) {
2161 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
2163 return ERR_PTR(err
);
2166 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2168 return ERR_PTR(err
);
2170 return __lookup_hash(&this, base
, 0);
2173 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2174 struct path
*path
, int *empty
)
2176 struct nameidata nd
;
2177 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2178 int err
= PTR_ERR(tmp
);
2181 BUG_ON(flags
& LOOKUP_PARENT
);
2183 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2191 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2194 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2198 * NB: most callers don't do anything directly with the reference to the
2199 * to struct filename, but the nd->last pointer points into the name string
2200 * allocated by getname. So we must hold the reference to it until all
2201 * path-walking is complete.
2203 static struct filename
*
2204 user_path_parent(int dfd
, const char __user
*path
, struct nameidata
*nd
,
2207 struct filename
*s
= getname(path
);
2210 /* only LOOKUP_REVAL is allowed in extra flags */
2211 flags
&= LOOKUP_REVAL
;
2216 error
= filename_lookup(dfd
, s
, flags
| LOOKUP_PARENT
, nd
);
2219 return ERR_PTR(error
);
2226 * It's inline, so penalty for filesystems that don't use sticky bit is
2229 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
2231 kuid_t fsuid
= current_fsuid();
2233 if (!(dir
->i_mode
& S_ISVTX
))
2235 if (uid_eq(inode
->i_uid
, fsuid
))
2237 if (uid_eq(dir
->i_uid
, fsuid
))
2239 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2243 * Check whether we can remove a link victim from directory dir, check
2244 * whether the type of victim is right.
2245 * 1. We can't do it if dir is read-only (done in permission())
2246 * 2. We should have write and exec permissions on dir
2247 * 3. We can't remove anything from append-only dir
2248 * 4. We can't do anything with immutable dir (done in permission())
2249 * 5. If the sticky bit on dir is set we should either
2250 * a. be owner of dir, or
2251 * b. be owner of victim, or
2252 * c. have CAP_FOWNER capability
2253 * 6. If the victim is append-only or immutable we can't do antyhing with
2254 * links pointing to it.
2255 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2256 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2257 * 9. We can't remove a root or mountpoint.
2258 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2259 * nfs_async_unlink().
2261 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
2265 if (!victim
->d_inode
)
2268 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2269 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2271 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2276 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
2277 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
2280 if (!S_ISDIR(victim
->d_inode
->i_mode
))
2282 if (IS_ROOT(victim
))
2284 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
2286 if (IS_DEADDIR(dir
))
2288 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2293 /* Check whether we can create an object with dentry child in directory
2295 * 1. We can't do it if child already exists (open has special treatment for
2296 * this case, but since we are inlined it's OK)
2297 * 2. We can't do it if dir is read-only (done in permission())
2298 * 3. We should have write and exec permissions on dir
2299 * 4. We can't do it if dir is immutable (done in permission())
2301 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2303 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2306 if (IS_DEADDIR(dir
))
2308 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2312 * p1 and p2 should be directories on the same fs.
2314 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2319 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2323 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2325 p
= d_ancestor(p2
, p1
);
2327 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2328 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2332 p
= d_ancestor(p1
, p2
);
2334 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2335 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2339 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2340 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2344 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2346 mutex_unlock(&p1
->d_inode
->i_mutex
);
2348 mutex_unlock(&p2
->d_inode
->i_mutex
);
2349 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2353 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2356 int error
= may_create(dir
, dentry
);
2360 if (!dir
->i_op
->create
)
2361 return -EACCES
; /* shouldn't it be ENOSYS? */
2364 error
= security_inode_create(dir
, dentry
, mode
);
2367 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2369 fsnotify_create(dir
, dentry
);
2373 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2375 struct dentry
*dentry
= path
->dentry
;
2376 struct inode
*inode
= dentry
->d_inode
;
2386 switch (inode
->i_mode
& S_IFMT
) {
2390 if (acc_mode
& MAY_WRITE
)
2395 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2404 error
= inode_permission(inode
, acc_mode
);
2409 * An append-only file must be opened in append mode for writing.
2411 if (IS_APPEND(inode
)) {
2412 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2418 /* O_NOATIME can only be set by the owner or superuser */
2419 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2425 static int handle_truncate(struct file
*filp
)
2427 struct path
*path
= &filp
->f_path
;
2428 struct inode
*inode
= path
->dentry
->d_inode
;
2429 int error
= get_write_access(inode
);
2433 * Refuse to truncate files with mandatory locks held on them.
2435 error
= locks_verify_locked(inode
);
2437 error
= security_path_truncate(path
);
2439 error
= do_truncate(path
->dentry
, 0,
2440 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2443 put_write_access(inode
);
2447 static inline int open_to_namei_flags(int flag
)
2449 if ((flag
& O_ACCMODE
) == 3)
2454 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2456 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2460 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2464 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2468 * Attempt to atomically look up, create and open a file from a negative
2471 * Returns 0 if successful. The file will have been created and attached to
2472 * @file by the filesystem calling finish_open().
2474 * Returns 1 if the file was looked up only or didn't need creating. The
2475 * caller will need to perform the open themselves. @path will have been
2476 * updated to point to the new dentry. This may be negative.
2478 * Returns an error code otherwise.
2480 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2481 struct path
*path
, struct file
*file
,
2482 const struct open_flags
*op
,
2483 bool got_write
, bool need_lookup
,
2486 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2487 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2491 int create_error
= 0;
2492 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2494 BUG_ON(dentry
->d_inode
);
2496 /* Don't create child dentry for a dead directory. */
2497 if (unlikely(IS_DEADDIR(dir
))) {
2503 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2504 mode
&= ~current_umask();
2506 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
)) {
2507 open_flag
&= ~O_TRUNC
;
2508 *opened
|= FILE_CREATED
;
2512 * Checking write permission is tricky, bacuse we don't know if we are
2513 * going to actually need it: O_CREAT opens should work as long as the
2514 * file exists. But checking existence breaks atomicity. The trick is
2515 * to check access and if not granted clear O_CREAT from the flags.
2517 * Another problem is returing the "right" error value (e.g. for an
2518 * O_EXCL open we want to return EEXIST not EROFS).
2520 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2521 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2522 if (!(open_flag
& O_CREAT
)) {
2524 * No O_CREATE -> atomicity not a requirement -> fall
2525 * back to lookup + open
2528 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2529 /* Fall back and fail with the right error */
2530 create_error
= -EROFS
;
2533 /* No side effects, safe to clear O_CREAT */
2534 create_error
= -EROFS
;
2535 open_flag
&= ~O_CREAT
;
2539 if (open_flag
& O_CREAT
) {
2540 error
= may_o_create(&nd
->path
, dentry
, mode
);
2542 create_error
= error
;
2543 if (open_flag
& O_EXCL
)
2545 open_flag
&= ~O_CREAT
;
2549 if (nd
->flags
& LOOKUP_DIRECTORY
)
2550 open_flag
|= O_DIRECTORY
;
2552 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2553 file
->f_path
.mnt
= nd
->path
.mnt
;
2554 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2557 if (create_error
&& error
== -ENOENT
)
2558 error
= create_error
;
2562 acc_mode
= op
->acc_mode
;
2563 if (*opened
& FILE_CREATED
) {
2564 fsnotify_create(dir
, dentry
);
2565 acc_mode
= MAY_OPEN
;
2568 if (error
) { /* returned 1, that is */
2569 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2573 if (file
->f_path
.dentry
) {
2575 dentry
= file
->f_path
.dentry
;
2577 if (create_error
&& dentry
->d_inode
== NULL
) {
2578 error
= create_error
;
2585 * We didn't have the inode before the open, so check open permission
2588 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2598 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2600 return PTR_ERR(dentry
);
2603 int open_flag
= op
->open_flag
;
2605 error
= create_error
;
2606 if ((open_flag
& O_EXCL
)) {
2607 if (!dentry
->d_inode
)
2609 } else if (!dentry
->d_inode
) {
2611 } else if ((open_flag
& O_TRUNC
) &&
2612 S_ISREG(dentry
->d_inode
->i_mode
)) {
2615 /* will fail later, go on to get the right error */
2619 path
->dentry
= dentry
;
2620 path
->mnt
= nd
->path
.mnt
;
2625 * Look up and maybe create and open the last component.
2627 * Must be called with i_mutex held on parent.
2629 * Returns 0 if the file was successfully atomically created (if necessary) and
2630 * opened. In this case the file will be returned attached to @file.
2632 * Returns 1 if the file was not completely opened at this time, though lookups
2633 * and creations will have been performed and the dentry returned in @path will
2634 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2635 * specified then a negative dentry may be returned.
2637 * An error code is returned otherwise.
2639 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2640 * cleared otherwise prior to returning.
2642 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2644 const struct open_flags
*op
,
2645 bool got_write
, int *opened
)
2647 struct dentry
*dir
= nd
->path
.dentry
;
2648 struct inode
*dir_inode
= dir
->d_inode
;
2649 struct dentry
*dentry
;
2653 *opened
&= ~FILE_CREATED
;
2654 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2656 return PTR_ERR(dentry
);
2658 /* Cached positive dentry: will open in f_op->open */
2659 if (!need_lookup
&& dentry
->d_inode
)
2662 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2663 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2664 need_lookup
, opened
);
2668 BUG_ON(dentry
->d_inode
);
2670 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2672 return PTR_ERR(dentry
);
2675 /* Negative dentry, just create the file */
2676 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2677 umode_t mode
= op
->mode
;
2678 if (!IS_POSIXACL(dir
->d_inode
))
2679 mode
&= ~current_umask();
2681 * This write is needed to ensure that a
2682 * rw->ro transition does not occur between
2683 * the time when the file is created and when
2684 * a permanent write count is taken through
2685 * the 'struct file' in finish_open().
2691 *opened
|= FILE_CREATED
;
2692 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2695 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2696 nd
->flags
& LOOKUP_EXCL
);
2701 path
->dentry
= dentry
;
2702 path
->mnt
= nd
->path
.mnt
;
2711 * Handle the last step of open()
2713 static int do_last(struct nameidata
*nd
, struct path
*path
,
2714 struct file
*file
, const struct open_flags
*op
,
2715 int *opened
, struct filename
*name
)
2717 struct dentry
*dir
= nd
->path
.dentry
;
2718 int open_flag
= op
->open_flag
;
2719 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2720 bool got_write
= false;
2721 int acc_mode
= op
->acc_mode
;
2722 struct inode
*inode
;
2723 bool symlink_ok
= false;
2724 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2725 bool retried
= false;
2728 nd
->flags
&= ~LOOKUP_PARENT
;
2729 nd
->flags
|= op
->intent
;
2731 switch (nd
->last_type
) {
2734 error
= handle_dots(nd
, nd
->last_type
);
2739 error
= complete_walk(nd
);
2742 audit_inode(name
, nd
->path
.dentry
, 0);
2743 if (open_flag
& O_CREAT
) {
2749 error
= complete_walk(nd
);
2752 audit_inode(name
, dir
, 0);
2756 if (!(open_flag
& O_CREAT
)) {
2757 if (nd
->last
.name
[nd
->last
.len
])
2758 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2759 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2761 /* we _can_ be in RCU mode here */
2762 error
= lookup_fast(nd
, path
, &inode
);
2769 BUG_ON(nd
->inode
!= dir
->d_inode
);
2771 /* create side of things */
2773 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2774 * has been cleared when we got to the last component we are
2777 error
= complete_walk(nd
);
2781 audit_inode(name
, dir
, LOOKUP_PARENT
);
2783 /* trailing slashes? */
2784 if (nd
->last
.name
[nd
->last
.len
])
2789 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2790 error
= mnt_want_write(nd
->path
.mnt
);
2794 * do _not_ fail yet - we might not need that or fail with
2795 * a different error; let lookup_open() decide; we'll be
2796 * dropping this one anyway.
2799 mutex_lock(&dir
->d_inode
->i_mutex
);
2800 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2801 mutex_unlock(&dir
->d_inode
->i_mutex
);
2807 if ((*opened
& FILE_CREATED
) ||
2808 !S_ISREG(file_inode(file
)->i_mode
))
2809 will_truncate
= false;
2811 audit_inode(name
, file
->f_path
.dentry
, 0);
2815 if (*opened
& FILE_CREATED
) {
2816 /* Don't check for write permission, don't truncate */
2817 open_flag
&= ~O_TRUNC
;
2818 will_truncate
= false;
2819 acc_mode
= MAY_OPEN
;
2820 path_to_nameidata(path
, nd
);
2821 goto finish_open_created
;
2825 * create/update audit record if it already exists.
2827 if (path
->dentry
->d_inode
)
2828 audit_inode(name
, path
->dentry
, 0);
2831 * If atomic_open() acquired write access it is dropped now due to
2832 * possible mount and symlink following (this might be optimized away if
2836 mnt_drop_write(nd
->path
.mnt
);
2841 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
2844 error
= follow_managed(path
, nd
->flags
);
2849 nd
->flags
|= LOOKUP_JUMPED
;
2851 BUG_ON(nd
->flags
& LOOKUP_RCU
);
2852 inode
= path
->dentry
->d_inode
;
2854 /* we _can_ be in RCU mode here */
2857 path_to_nameidata(path
, nd
);
2861 if (should_follow_link(inode
, !symlink_ok
)) {
2862 if (nd
->flags
& LOOKUP_RCU
) {
2863 if (unlikely(nd
->path
.mnt
!= path
->mnt
||
2864 unlazy_walk(nd
, path
->dentry
))) {
2869 BUG_ON(inode
!= path
->dentry
->d_inode
);
2873 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
2874 path_to_nameidata(path
, nd
);
2876 save_parent
.dentry
= nd
->path
.dentry
;
2877 save_parent
.mnt
= mntget(path
->mnt
);
2878 nd
->path
.dentry
= path
->dentry
;
2882 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2883 error
= complete_walk(nd
);
2885 path_put(&save_parent
);
2889 if ((open_flag
& O_CREAT
) && S_ISDIR(nd
->inode
->i_mode
))
2892 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !can_lookup(nd
->inode
))
2894 audit_inode(name
, nd
->path
.dentry
, 0);
2896 if (!S_ISREG(nd
->inode
->i_mode
))
2897 will_truncate
= false;
2899 if (will_truncate
) {
2900 error
= mnt_want_write(nd
->path
.mnt
);
2905 finish_open_created
:
2906 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2909 file
->f_path
.mnt
= nd
->path
.mnt
;
2910 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
2912 if (error
== -EOPENSTALE
)
2917 error
= open_check_o_direct(file
);
2920 error
= ima_file_check(file
, op
->acc_mode
);
2924 if (will_truncate
) {
2925 error
= handle_truncate(file
);
2930 if (unlikely(error
> 0)) {
2935 mnt_drop_write(nd
->path
.mnt
);
2936 path_put(&save_parent
);
2941 path_put_conditional(path
, nd
);
2948 /* If no saved parent or already retried then can't retry */
2949 if (!save_parent
.dentry
|| retried
)
2952 BUG_ON(save_parent
.dentry
!= dir
);
2953 path_put(&nd
->path
);
2954 nd
->path
= save_parent
;
2955 nd
->inode
= dir
->d_inode
;
2956 save_parent
.mnt
= NULL
;
2957 save_parent
.dentry
= NULL
;
2959 mnt_drop_write(nd
->path
.mnt
);
2966 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
2967 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2969 struct file
*base
= NULL
;
2975 file
= get_empty_filp();
2979 file
->f_flags
= op
->open_flag
;
2981 error
= path_init(dfd
, pathname
->name
, flags
| LOOKUP_PARENT
, nd
, &base
);
2982 if (unlikely(error
))
2985 current
->total_link_count
= 0;
2986 error
= link_path_walk(pathname
->name
, nd
);
2987 if (unlikely(error
))
2990 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2991 while (unlikely(error
> 0)) { /* trailing symlink */
2992 struct path link
= path
;
2994 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2995 path_put_conditional(&path
, nd
);
2996 path_put(&nd
->path
);
3000 error
= may_follow_link(&link
, nd
);
3001 if (unlikely(error
))
3003 nd
->flags
|= LOOKUP_PARENT
;
3004 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3005 error
= follow_link(&link
, nd
, &cookie
);
3006 if (unlikely(error
))
3008 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3009 put_link(nd
, &link
, cookie
);
3012 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
3013 path_put(&nd
->root
);
3016 if (!(opened
& FILE_OPENED
)) {
3020 if (unlikely(error
)) {
3021 if (error
== -EOPENSTALE
) {
3022 if (flags
& LOOKUP_RCU
)
3027 file
= ERR_PTR(error
);
3032 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3033 const struct open_flags
*op
, int flags
)
3035 struct nameidata nd
;
3038 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3039 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3040 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3041 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3042 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3046 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3047 const char *name
, const struct open_flags
*op
, int flags
)
3049 struct nameidata nd
;
3051 struct filename filename
= { .name
= name
};
3054 nd
.root
.dentry
= dentry
;
3056 flags
|= LOOKUP_ROOT
;
3058 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
3059 return ERR_PTR(-ELOOP
);
3061 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3062 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3063 file
= path_openat(-1, &filename
, &nd
, op
, flags
);
3064 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3065 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3069 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3070 struct path
*path
, unsigned int lookup_flags
)
3072 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3073 struct nameidata nd
;
3076 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3079 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3080 * other flags passed in are ignored!
3082 lookup_flags
&= LOOKUP_REVAL
;
3084 error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
|lookup_flags
, &nd
);
3086 return ERR_PTR(error
);
3089 * Yucky last component or no last component at all?
3090 * (foo/., foo/.., /////)
3092 if (nd
.last_type
!= LAST_NORM
)
3094 nd
.flags
&= ~LOOKUP_PARENT
;
3095 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3097 /* don't fail immediately if it's r/o, at least try to report other errors */
3098 err2
= mnt_want_write(nd
.path
.mnt
);
3100 * Do the final lookup.
3102 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3103 dentry
= lookup_hash(&nd
);
3108 if (dentry
->d_inode
)
3111 * Special case - lookup gave negative, but... we had foo/bar/
3112 * From the vfs_mknod() POV we just have a negative dentry -
3113 * all is fine. Let's be bastards - you had / on the end, you've
3114 * been asking for (non-existent) directory. -ENOENT for you.
3116 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3120 if (unlikely(err2
)) {
3128 dentry
= ERR_PTR(error
);
3130 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3132 mnt_drop_write(nd
.path
.mnt
);
3137 EXPORT_SYMBOL(kern_path_create
);
3139 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3142 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3143 mnt_drop_write(path
->mnt
);
3146 EXPORT_SYMBOL(done_path_create
);
3148 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3149 struct path
*path
, unsigned int lookup_flags
)
3151 struct filename
*tmp
= getname(pathname
);
3154 return ERR_CAST(tmp
);
3155 res
= kern_path_create(dfd
, tmp
->name
, path
, lookup_flags
);
3159 EXPORT_SYMBOL(user_path_create
);
3161 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3163 int error
= may_create(dir
, dentry
);
3168 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3171 if (!dir
->i_op
->mknod
)
3174 error
= devcgroup_inode_mknod(mode
, dev
);
3178 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3182 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3184 fsnotify_create(dir
, dentry
);
3188 static int may_mknod(umode_t mode
)
3190 switch (mode
& S_IFMT
) {
3196 case 0: /* zero mode translates to S_IFREG */
3205 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3208 struct dentry
*dentry
;
3211 unsigned int lookup_flags
= 0;
3213 error
= may_mknod(mode
);
3217 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3219 return PTR_ERR(dentry
);
3221 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3222 mode
&= ~current_umask();
3223 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3226 switch (mode
& S_IFMT
) {
3227 case 0: case S_IFREG
:
3228 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3230 case S_IFCHR
: case S_IFBLK
:
3231 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3232 new_decode_dev(dev
));
3234 case S_IFIFO
: case S_IFSOCK
:
3235 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3239 done_path_create(&path
, dentry
);
3240 if (retry_estale(error
, lookup_flags
)) {
3241 lookup_flags
|= LOOKUP_REVAL
;
3247 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3249 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3252 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3254 int error
= may_create(dir
, dentry
);
3255 unsigned max_links
= dir
->i_sb
->s_max_links
;
3260 if (!dir
->i_op
->mkdir
)
3263 mode
&= (S_IRWXUGO
|S_ISVTX
);
3264 error
= security_inode_mkdir(dir
, dentry
, mode
);
3268 if (max_links
&& dir
->i_nlink
>= max_links
)
3271 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3273 fsnotify_mkdir(dir
, dentry
);
3277 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3279 struct dentry
*dentry
;
3282 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3285 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3287 return PTR_ERR(dentry
);
3289 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3290 mode
&= ~current_umask();
3291 error
= security_path_mkdir(&path
, dentry
, mode
);
3293 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3294 done_path_create(&path
, dentry
);
3295 if (retry_estale(error
, lookup_flags
)) {
3296 lookup_flags
|= LOOKUP_REVAL
;
3302 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3304 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3308 * The dentry_unhash() helper will try to drop the dentry early: we
3309 * should have a usage count of 1 if we're the only user of this
3310 * dentry, and if that is true (possibly after pruning the dcache),
3311 * then we drop the dentry now.
3313 * A low-level filesystem can, if it choses, legally
3316 * if (!d_unhashed(dentry))
3319 * if it cannot handle the case of removing a directory
3320 * that is still in use by something else..
3322 void dentry_unhash(struct dentry
*dentry
)
3324 shrink_dcache_parent(dentry
);
3325 spin_lock(&dentry
->d_lock
);
3326 if (dentry
->d_count
== 1)
3328 spin_unlock(&dentry
->d_lock
);
3331 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3333 int error
= may_delete(dir
, dentry
, 1);
3338 if (!dir
->i_op
->rmdir
)
3342 mutex_lock(&dentry
->d_inode
->i_mutex
);
3345 if (d_mountpoint(dentry
))
3348 error
= security_inode_rmdir(dir
, dentry
);
3352 shrink_dcache_parent(dentry
);
3353 error
= dir
->i_op
->rmdir(dir
, dentry
);
3357 dentry
->d_inode
->i_flags
|= S_DEAD
;
3361 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3368 static long do_rmdir(int dfd
, const char __user
*pathname
)
3371 struct filename
*name
;
3372 struct dentry
*dentry
;
3373 struct nameidata nd
;
3374 unsigned int lookup_flags
= 0;
3376 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3378 return PTR_ERR(name
);
3380 switch(nd
.last_type
) {
3392 nd
.flags
&= ~LOOKUP_PARENT
;
3393 error
= mnt_want_write(nd
.path
.mnt
);
3397 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3398 dentry
= lookup_hash(&nd
);
3399 error
= PTR_ERR(dentry
);
3402 if (!dentry
->d_inode
) {
3406 error
= security_path_rmdir(&nd
.path
, dentry
);
3409 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3413 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3414 mnt_drop_write(nd
.path
.mnt
);
3418 if (retry_estale(error
, lookup_flags
)) {
3419 lookup_flags
|= LOOKUP_REVAL
;
3425 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3427 return do_rmdir(AT_FDCWD
, pathname
);
3430 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3432 int error
= may_delete(dir
, dentry
, 0);
3437 if (!dir
->i_op
->unlink
)
3440 mutex_lock(&dentry
->d_inode
->i_mutex
);
3441 if (d_mountpoint(dentry
))
3444 error
= security_inode_unlink(dir
, dentry
);
3446 error
= dir
->i_op
->unlink(dir
, dentry
);
3451 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3453 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3454 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3455 fsnotify_link_count(dentry
->d_inode
);
3463 * Make sure that the actual truncation of the file will occur outside its
3464 * directory's i_mutex. Truncate can take a long time if there is a lot of
3465 * writeout happening, and we don't want to prevent access to the directory
3466 * while waiting on the I/O.
3468 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3471 struct filename
*name
;
3472 struct dentry
*dentry
;
3473 struct nameidata nd
;
3474 struct inode
*inode
= NULL
;
3475 unsigned int lookup_flags
= 0;
3477 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3479 return PTR_ERR(name
);
3482 if (nd
.last_type
!= LAST_NORM
)
3485 nd
.flags
&= ~LOOKUP_PARENT
;
3486 error
= mnt_want_write(nd
.path
.mnt
);
3490 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3491 dentry
= lookup_hash(&nd
);
3492 error
= PTR_ERR(dentry
);
3493 if (!IS_ERR(dentry
)) {
3494 /* Why not before? Because we want correct error value */
3495 if (nd
.last
.name
[nd
.last
.len
])
3497 inode
= dentry
->d_inode
;
3501 error
= security_path_unlink(&nd
.path
, dentry
);
3504 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
3508 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3510 iput(inode
); /* truncate the inode here */
3511 mnt_drop_write(nd
.path
.mnt
);
3515 if (retry_estale(error
, lookup_flags
)) {
3516 lookup_flags
|= LOOKUP_REVAL
;
3523 error
= !dentry
->d_inode
? -ENOENT
:
3524 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
3528 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3530 if ((flag
& ~AT_REMOVEDIR
) != 0)
3533 if (flag
& AT_REMOVEDIR
)
3534 return do_rmdir(dfd
, pathname
);
3536 return do_unlinkat(dfd
, pathname
);
3539 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3541 return do_unlinkat(AT_FDCWD
, pathname
);
3544 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3546 int error
= may_create(dir
, dentry
);
3551 if (!dir
->i_op
->symlink
)
3554 error
= security_inode_symlink(dir
, dentry
, oldname
);
3558 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3560 fsnotify_create(dir
, dentry
);
3564 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3565 int, newdfd
, const char __user
*, newname
)
3568 struct filename
*from
;
3569 struct dentry
*dentry
;
3571 unsigned int lookup_flags
= 0;
3573 from
= getname(oldname
);
3575 return PTR_ERR(from
);
3577 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3578 error
= PTR_ERR(dentry
);
3582 error
= security_path_symlink(&path
, dentry
, from
->name
);
3584 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3585 done_path_create(&path
, dentry
);
3586 if (retry_estale(error
, lookup_flags
)) {
3587 lookup_flags
|= LOOKUP_REVAL
;
3595 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3597 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3600 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3602 struct inode
*inode
= old_dentry
->d_inode
;
3603 unsigned max_links
= dir
->i_sb
->s_max_links
;
3609 error
= may_create(dir
, new_dentry
);
3613 if (dir
->i_sb
!= inode
->i_sb
)
3617 * A link to an append-only or immutable file cannot be created.
3619 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3621 if (!dir
->i_op
->link
)
3623 if (S_ISDIR(inode
->i_mode
))
3626 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3630 mutex_lock(&inode
->i_mutex
);
3631 /* Make sure we don't allow creating hardlink to an unlinked file */
3632 if (inode
->i_nlink
== 0)
3634 else if (max_links
&& inode
->i_nlink
>= max_links
)
3637 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3638 mutex_unlock(&inode
->i_mutex
);
3640 fsnotify_link(dir
, inode
, new_dentry
);
3645 * Hardlinks are often used in delicate situations. We avoid
3646 * security-related surprises by not following symlinks on the
3649 * We don't follow them on the oldname either to be compatible
3650 * with linux 2.0, and to avoid hard-linking to directories
3651 * and other special files. --ADM
3653 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3654 int, newdfd
, const char __user
*, newname
, int, flags
)
3656 struct dentry
*new_dentry
;
3657 struct path old_path
, new_path
;
3661 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3664 * To use null names we require CAP_DAC_READ_SEARCH
3665 * This ensures that not everyone will be able to create
3666 * handlink using the passed filedescriptor.
3668 if (flags
& AT_EMPTY_PATH
) {
3669 if (!capable(CAP_DAC_READ_SEARCH
))
3674 if (flags
& AT_SYMLINK_FOLLOW
)
3675 how
|= LOOKUP_FOLLOW
;
3677 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3681 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
3682 (how
& LOOKUP_REVAL
));
3683 error
= PTR_ERR(new_dentry
);
3684 if (IS_ERR(new_dentry
))
3688 if (old_path
.mnt
!= new_path
.mnt
)
3690 error
= may_linkat(&old_path
);
3691 if (unlikely(error
))
3693 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3696 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3698 done_path_create(&new_path
, new_dentry
);
3699 if (retry_estale(error
, how
)) {
3700 path_put(&old_path
);
3701 how
|= LOOKUP_REVAL
;
3705 path_put(&old_path
);
3710 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3712 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3716 * The worst of all namespace operations - renaming directory. "Perverted"
3717 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3719 * a) we can get into loop creation. Check is done in is_subdir().
3720 * b) race potential - two innocent renames can create a loop together.
3721 * That's where 4.4 screws up. Current fix: serialization on
3722 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3724 * c) we have to lock _three_ objects - parents and victim (if it exists).
3725 * And that - after we got ->i_mutex on parents (until then we don't know
3726 * whether the target exists). Solution: try to be smart with locking
3727 * order for inodes. We rely on the fact that tree topology may change
3728 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3729 * move will be locked. Thus we can rank directories by the tree
3730 * (ancestors first) and rank all non-directories after them.
3731 * That works since everybody except rename does "lock parent, lookup,
3732 * lock child" and rename is under ->s_vfs_rename_mutex.
3733 * HOWEVER, it relies on the assumption that any object with ->lookup()
3734 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3735 * we'd better make sure that there's no link(2) for them.
3736 * d) conversion from fhandle to dentry may come in the wrong moment - when
3737 * we are removing the target. Solution: we will have to grab ->i_mutex
3738 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3739 * ->i_mutex on parents, which works but leads to some truly excessive
3742 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3743 struct inode
*new_dir
, struct dentry
*new_dentry
)
3746 struct inode
*target
= new_dentry
->d_inode
;
3747 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3750 * If we are going to change the parent - check write permissions,
3751 * we'll need to flip '..'.
3753 if (new_dir
!= old_dir
) {
3754 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3759 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3765 mutex_lock(&target
->i_mutex
);
3768 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3772 if (max_links
&& !target
&& new_dir
!= old_dir
&&
3773 new_dir
->i_nlink
>= max_links
)
3777 shrink_dcache_parent(new_dentry
);
3778 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3783 target
->i_flags
|= S_DEAD
;
3784 dont_mount(new_dentry
);
3788 mutex_unlock(&target
->i_mutex
);
3791 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3792 d_move(old_dentry
,new_dentry
);
3796 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3797 struct inode
*new_dir
, struct dentry
*new_dentry
)
3799 struct inode
*target
= new_dentry
->d_inode
;
3802 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3808 mutex_lock(&target
->i_mutex
);
3811 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3814 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3819 dont_mount(new_dentry
);
3820 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3821 d_move(old_dentry
, new_dentry
);
3824 mutex_unlock(&target
->i_mutex
);
3829 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3830 struct inode
*new_dir
, struct dentry
*new_dentry
)
3833 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3834 const unsigned char *old_name
;
3836 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3839 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3843 if (!new_dentry
->d_inode
)
3844 error
= may_create(new_dir
, new_dentry
);
3846 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3850 if (!old_dir
->i_op
->rename
)
3853 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3856 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3858 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3860 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3861 new_dentry
->d_inode
, old_dentry
);
3862 fsnotify_oldname_free(old_name
);
3867 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3868 int, newdfd
, const char __user
*, newname
)
3870 struct dentry
*old_dir
, *new_dir
;
3871 struct dentry
*old_dentry
, *new_dentry
;
3872 struct dentry
*trap
;
3873 struct nameidata oldnd
, newnd
;
3874 struct filename
*from
;
3875 struct filename
*to
;
3876 unsigned int lookup_flags
= 0;
3877 bool should_retry
= false;
3880 from
= user_path_parent(olddfd
, oldname
, &oldnd
, lookup_flags
);
3882 error
= PTR_ERR(from
);
3886 to
= user_path_parent(newdfd
, newname
, &newnd
, lookup_flags
);
3888 error
= PTR_ERR(to
);
3893 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3896 old_dir
= oldnd
.path
.dentry
;
3898 if (oldnd
.last_type
!= LAST_NORM
)
3901 new_dir
= newnd
.path
.dentry
;
3902 if (newnd
.last_type
!= LAST_NORM
)
3905 error
= mnt_want_write(oldnd
.path
.mnt
);
3909 oldnd
.flags
&= ~LOOKUP_PARENT
;
3910 newnd
.flags
&= ~LOOKUP_PARENT
;
3911 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3913 trap
= lock_rename(new_dir
, old_dir
);
3915 old_dentry
= lookup_hash(&oldnd
);
3916 error
= PTR_ERR(old_dentry
);
3917 if (IS_ERR(old_dentry
))
3919 /* source must exist */
3921 if (!old_dentry
->d_inode
)
3923 /* unless the source is a directory trailing slashes give -ENOTDIR */
3924 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3926 if (oldnd
.last
.name
[oldnd
.last
.len
])
3928 if (newnd
.last
.name
[newnd
.last
.len
])
3931 /* source should not be ancestor of target */
3933 if (old_dentry
== trap
)
3935 new_dentry
= lookup_hash(&newnd
);
3936 error
= PTR_ERR(new_dentry
);
3937 if (IS_ERR(new_dentry
))
3939 /* target should not be an ancestor of source */
3941 if (new_dentry
== trap
)
3944 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3945 &newnd
.path
, new_dentry
);
3948 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3949 new_dir
->d_inode
, new_dentry
);
3955 unlock_rename(new_dir
, old_dir
);
3956 mnt_drop_write(oldnd
.path
.mnt
);
3958 if (retry_estale(error
, lookup_flags
))
3959 should_retry
= true;
3960 path_put(&newnd
.path
);
3963 path_put(&oldnd
.path
);
3966 should_retry
= false;
3967 lookup_flags
|= LOOKUP_REVAL
;
3974 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3976 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3979 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3983 len
= PTR_ERR(link
);
3988 if (len
> (unsigned) buflen
)
3990 if (copy_to_user(buffer
, link
, len
))
3997 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3998 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3999 * using) it for any given inode is up to filesystem.
4001 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4003 struct nameidata nd
;
4008 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
4010 return PTR_ERR(cookie
);
4012 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
4013 if (dentry
->d_inode
->i_op
->put_link
)
4014 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
4018 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
4020 return __vfs_follow_link(nd
, link
);
4023 /* get the link contents into pagecache */
4024 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
4028 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
4029 page
= read_mapping_page(mapping
, 0, NULL
);
4034 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
4038 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4040 struct page
*page
= NULL
;
4041 char *s
= page_getlink(dentry
, &page
);
4042 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
4045 page_cache_release(page
);
4050 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
4052 struct page
*page
= NULL
;
4053 nd_set_link(nd
, page_getlink(dentry
, &page
));
4057 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
4059 struct page
*page
= cookie
;
4063 page_cache_release(page
);
4068 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4070 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4072 struct address_space
*mapping
= inode
->i_mapping
;
4077 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4079 flags
|= AOP_FLAG_NOFS
;
4082 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4083 flags
, &page
, &fsdata
);
4087 kaddr
= kmap_atomic(page
);
4088 memcpy(kaddr
, symname
, len
-1);
4089 kunmap_atomic(kaddr
);
4091 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4098 mark_inode_dirty(inode
);
4104 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4106 return __page_symlink(inode
, symname
, len
,
4107 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4110 const struct inode_operations page_symlink_inode_operations
= {
4111 .readlink
= generic_readlink
,
4112 .follow_link
= page_follow_link_light
,
4113 .put_link
= page_put_link
,
4116 EXPORT_SYMBOL(user_path_at
);
4117 EXPORT_SYMBOL(follow_down_one
);
4118 EXPORT_SYMBOL(follow_down
);
4119 EXPORT_SYMBOL(follow_up
);
4120 EXPORT_SYMBOL(get_write_access
); /* nfsd */
4121 EXPORT_SYMBOL(lock_rename
);
4122 EXPORT_SYMBOL(lookup_one_len
);
4123 EXPORT_SYMBOL(page_follow_link_light
);
4124 EXPORT_SYMBOL(page_put_link
);
4125 EXPORT_SYMBOL(page_readlink
);
4126 EXPORT_SYMBOL(__page_symlink
);
4127 EXPORT_SYMBOL(page_symlink
);
4128 EXPORT_SYMBOL(page_symlink_inode_operations
);
4129 EXPORT_SYMBOL(kern_path
);
4130 EXPORT_SYMBOL(vfs_path_lookup
);
4131 EXPORT_SYMBOL(inode_permission
);
4132 EXPORT_SYMBOL(unlock_rename
);
4133 EXPORT_SYMBOL(vfs_create
);
4134 EXPORT_SYMBOL(vfs_follow_link
);
4135 EXPORT_SYMBOL(vfs_link
);
4136 EXPORT_SYMBOL(vfs_mkdir
);
4137 EXPORT_SYMBOL(vfs_mknod
);
4138 EXPORT_SYMBOL(generic_permission
);
4139 EXPORT_SYMBOL(vfs_readlink
);
4140 EXPORT_SYMBOL(vfs_rename
);
4141 EXPORT_SYMBOL(vfs_rmdir
);
4142 EXPORT_SYMBOL(vfs_symlink
);
4143 EXPORT_SYMBOL(vfs_unlink
);
4144 EXPORT_SYMBOL(dentry_unhash
);
4145 EXPORT_SYMBOL(generic_readlink
);