4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <asm/uaccess.h>
42 /* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
90 * [10-Sep-98 Alan Modra] Another symlink change.
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 static char *getname_flags(const char __user
*filename
, int flags
, int *empty
)
122 char *result
= __getname(), *err
;
125 if (unlikely(!result
))
126 return ERR_PTR(-ENOMEM
);
128 len
= strncpy_from_user(result
, filename
, PATH_MAX
);
130 if (unlikely(len
< 0))
133 /* The empty path is special. */
134 if (unlikely(!len
)) {
137 err
= ERR_PTR(-ENOENT
);
138 if (!(flags
& LOOKUP_EMPTY
))
142 err
= ERR_PTR(-ENAMETOOLONG
);
143 if (likely(len
< PATH_MAX
)) {
144 audit_getname(result
);
153 char *getname(const char __user
* filename
)
155 return getname_flags(filename
, 0, NULL
);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name
)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname
);
169 static int check_acl(struct inode
*inode
, int mask
)
171 #ifdef CONFIG_FS_POSIX_ACL
172 struct posix_acl
*acl
;
174 if (mask
& MAY_NOT_BLOCK
) {
175 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
178 /* no ->get_acl() calls in RCU mode... */
179 if (acl
== ACL_NOT_CACHED
)
181 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
184 acl
= get_cached_acl(inode
, ACL_TYPE_ACCESS
);
187 * A filesystem can force a ACL callback by just never filling the
188 * ACL cache. But normally you'd fill the cache either at inode
189 * instantiation time, or on the first ->get_acl call.
191 * If the filesystem doesn't have a get_acl() function at all, we'll
192 * just create the negative cache entry.
194 if (acl
== ACL_NOT_CACHED
) {
195 if (inode
->i_op
->get_acl
) {
196 acl
= inode
->i_op
->get_acl(inode
, ACL_TYPE_ACCESS
);
200 set_cached_acl(inode
, ACL_TYPE_ACCESS
, NULL
);
206 int error
= posix_acl_permission(inode
, acl
, mask
);
207 posix_acl_release(acl
);
216 * This does the basic permission checking
218 static int acl_permission_check(struct inode
*inode
, int mask
)
220 unsigned int mode
= inode
->i_mode
;
222 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
225 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
226 int error
= check_acl(inode
, mask
);
227 if (error
!= -EAGAIN
)
231 if (in_group_p(inode
->i_gid
))
236 * If the DACs are ok we don't need any capability check.
238 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
244 * generic_permission - check for access rights on a Posix-like filesystem
245 * @inode: inode to check access rights for
246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
248 * Used to check for read/write/execute permissions on a file.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
251 * are used for other things.
253 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
254 * request cannot be satisfied (eg. requires blocking or too much complexity).
255 * It would then be called again in ref-walk mode.
257 int generic_permission(struct inode
*inode
, int mask
)
262 * Do the basic permission checks.
264 ret
= acl_permission_check(inode
, mask
);
268 if (S_ISDIR(inode
->i_mode
)) {
269 /* DACs are overridable for directories */
270 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
272 if (!(mask
& MAY_WRITE
))
273 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
278 * Read/write DACs are always overridable.
279 * Executable DACs are overridable when there is
280 * at least one exec bit set.
282 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
283 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
287 * Searching includes executable on directories, else just read.
289 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
290 if (mask
== MAY_READ
)
291 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
298 * We _really_ want to just do "generic_permission()" without
299 * even looking at the inode->i_op values. So we keep a cache
300 * flag in inode->i_opflags, that says "this has not special
301 * permission function, use the fast case".
303 static inline int do_inode_permission(struct inode
*inode
, int mask
)
305 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
306 if (likely(inode
->i_op
->permission
))
307 return inode
->i_op
->permission(inode
, mask
);
309 /* This gets set once for the inode lifetime */
310 spin_lock(&inode
->i_lock
);
311 inode
->i_opflags
|= IOP_FASTPERM
;
312 spin_unlock(&inode
->i_lock
);
314 return generic_permission(inode
, mask
);
318 * __inode_permission - Check for access rights to a given inode
319 * @inode: Inode to check permission on
320 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
322 * Check for read/write/execute permissions on an inode.
324 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
326 * This does not check for a read-only file system. You probably want
327 * inode_permission().
329 int __inode_permission(struct inode
*inode
, int mask
)
333 if (unlikely(mask
& MAY_WRITE
)) {
335 * Nobody gets write access to an immutable file.
337 if (IS_IMMUTABLE(inode
))
341 retval
= do_inode_permission(inode
, mask
);
345 retval
= devcgroup_inode_permission(inode
, mask
);
349 return security_inode_permission(inode
, mask
);
353 * sb_permission - Check superblock-level permissions
354 * @sb: Superblock of inode to check permission on
355 * @inode: Inode to check permission on
356 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
358 * Separate out file-system wide checks from inode-specific permission checks.
360 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
362 if (unlikely(mask
& MAY_WRITE
)) {
363 umode_t mode
= inode
->i_mode
;
365 /* Nobody gets write access to a read-only fs. */
366 if ((sb
->s_flags
& MS_RDONLY
) &&
367 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
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. We use fs[ug]id for
379 * this, letting us set arbitrary permissions for filesystem access without
380 * changing the "normal" UIDs which are used for other things.
382 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
384 int inode_permission(struct inode
*inode
, int mask
)
388 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
391 return __inode_permission(inode
, mask
);
395 * path_get - get a reference to a path
396 * @path: path to get the reference to
398 * Given a path increment the reference count to the dentry and the vfsmount.
400 void path_get(struct path
*path
)
405 EXPORT_SYMBOL(path_get
);
408 * path_put - put a reference to a path
409 * @path: path to put the reference to
411 * Given a path decrement the reference count to the dentry and the vfsmount.
413 void path_put(struct path
*path
)
418 EXPORT_SYMBOL(path_put
);
421 * Path walking has 2 modes, rcu-walk and ref-walk (see
422 * Documentation/filesystems/path-lookup.txt). In situations when we can't
423 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
424 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
425 * mode. Refcounts are grabbed at the last known good point before rcu-walk
426 * got stuck, so ref-walk may continue from there. If this is not successful
427 * (eg. a seqcount has changed), then failure is returned and it's up to caller
428 * to restart the path walk from the beginning in ref-walk mode.
431 static inline void lock_rcu_walk(void)
433 br_read_lock(&vfsmount_lock
);
437 static inline void unlock_rcu_walk(void)
440 br_read_unlock(&vfsmount_lock
);
444 * unlazy_walk - try to switch to ref-walk mode.
445 * @nd: nameidata pathwalk data
446 * @dentry: child of nd->path.dentry or NULL
447 * Returns: 0 on success, -ECHILD on failure
449 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
450 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
451 * @nd or NULL. Must be called from rcu-walk context.
453 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
455 struct fs_struct
*fs
= current
->fs
;
456 struct dentry
*parent
= nd
->path
.dentry
;
459 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
460 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
462 spin_lock(&fs
->lock
);
463 if (nd
->root
.mnt
!= fs
->root
.mnt
||
464 nd
->root
.dentry
!= fs
->root
.dentry
)
467 spin_lock(&parent
->d_lock
);
469 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
471 BUG_ON(nd
->inode
!= parent
->d_inode
);
473 if (dentry
->d_parent
!= parent
)
475 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
476 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
479 * If the sequence check on the child dentry passed, then
480 * the child has not been removed from its parent. This
481 * means the parent dentry must be valid and able to take
482 * a reference at this point.
484 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
485 BUG_ON(!parent
->d_count
);
487 spin_unlock(&dentry
->d_lock
);
489 spin_unlock(&parent
->d_lock
);
492 spin_unlock(&fs
->lock
);
494 mntget(nd
->path
.mnt
);
497 nd
->flags
&= ~LOOKUP_RCU
;
501 spin_unlock(&dentry
->d_lock
);
503 spin_unlock(&parent
->d_lock
);
506 spin_unlock(&fs
->lock
);
510 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
512 return dentry
->d_op
->d_revalidate(dentry
, flags
);
516 * complete_walk - successful completion of path walk
517 * @nd: pointer nameidata
519 * If we had been in RCU mode, drop out of it and legitimize nd->path.
520 * Revalidate the final result, unless we'd already done that during
521 * the path walk or the filesystem doesn't ask for it. Return 0 on
522 * success, -error on failure. In case of failure caller does not
523 * need to drop nd->path.
525 static int complete_walk(struct nameidata
*nd
)
527 struct dentry
*dentry
= nd
->path
.dentry
;
530 if (nd
->flags
& LOOKUP_RCU
) {
531 nd
->flags
&= ~LOOKUP_RCU
;
532 if (!(nd
->flags
& LOOKUP_ROOT
))
534 spin_lock(&dentry
->d_lock
);
535 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
536 spin_unlock(&dentry
->d_lock
);
540 BUG_ON(nd
->inode
!= dentry
->d_inode
);
541 spin_unlock(&dentry
->d_lock
);
542 mntget(nd
->path
.mnt
);
546 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
549 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
552 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
555 /* Note: we do not d_invalidate() */
556 status
= d_revalidate(dentry
, nd
->flags
);
567 static __always_inline
void set_root(struct nameidata
*nd
)
570 get_fs_root(current
->fs
, &nd
->root
);
573 static int link_path_walk(const char *, struct nameidata
*);
575 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
578 struct fs_struct
*fs
= current
->fs
;
582 seq
= read_seqcount_begin(&fs
->seq
);
584 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
585 } while (read_seqcount_retry(&fs
->seq
, seq
));
589 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
601 nd
->flags
|= LOOKUP_JUMPED
;
603 nd
->inode
= nd
->path
.dentry
->d_inode
;
605 ret
= link_path_walk(link
, nd
);
609 return PTR_ERR(link
);
612 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
615 if (path
->mnt
!= nd
->path
.mnt
)
619 static inline void path_to_nameidata(const struct path
*path
,
620 struct nameidata
*nd
)
622 if (!(nd
->flags
& LOOKUP_RCU
)) {
623 dput(nd
->path
.dentry
);
624 if (nd
->path
.mnt
!= path
->mnt
)
625 mntput(nd
->path
.mnt
);
627 nd
->path
.mnt
= path
->mnt
;
628 nd
->path
.dentry
= path
->dentry
;
632 * Helper to directly jump to a known parsed path from ->follow_link,
633 * caller must have taken a reference to path beforehand.
635 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
640 nd
->inode
= nd
->path
.dentry
->d_inode
;
641 nd
->flags
|= LOOKUP_JUMPED
;
643 BUG_ON(nd
->inode
->i_op
->follow_link
);
646 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
648 struct inode
*inode
= link
->dentry
->d_inode
;
649 if (inode
->i_op
->put_link
)
650 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
654 int sysctl_protected_symlinks __read_mostly
= 1;
655 int sysctl_protected_hardlinks __read_mostly
= 1;
658 * may_follow_link - Check symlink following for unsafe situations
659 * @link: The path of the symlink
660 * @nd: nameidata pathwalk data
662 * In the case of the sysctl_protected_symlinks sysctl being enabled,
663 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
664 * in a sticky world-writable directory. This is to protect privileged
665 * processes from failing races against path names that may change out
666 * from under them by way of other users creating malicious symlinks.
667 * It will permit symlinks to be followed only when outside a sticky
668 * world-writable directory, or when the uid of the symlink and follower
669 * match, or when the directory owner matches the symlink's owner.
671 * Returns 0 if following the symlink is allowed, -ve on error.
673 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
675 const struct inode
*inode
;
676 const struct inode
*parent
;
678 if (!sysctl_protected_symlinks
)
681 /* Allowed if owner and follower match. */
682 inode
= link
->dentry
->d_inode
;
683 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
686 /* Allowed if parent directory not sticky and world-writable. */
687 parent
= nd
->path
.dentry
->d_inode
;
688 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
691 /* Allowed if parent directory and link owner match. */
692 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
695 path_put_conditional(link
, nd
);
697 audit_log_link_denied("follow_link", link
);
702 * safe_hardlink_source - Check for safe hardlink conditions
703 * @inode: the source inode to hardlink from
705 * Return false if at least one of the following conditions:
706 * - inode is not a regular file
708 * - inode is setgid and group-exec
709 * - access failure for read and write
711 * Otherwise returns true.
713 static bool safe_hardlink_source(struct inode
*inode
)
715 umode_t mode
= inode
->i_mode
;
717 /* Special files should not get pinned to the filesystem. */
721 /* Setuid files should not get pinned to the filesystem. */
725 /* Executable setgid files should not get pinned to the filesystem. */
726 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
729 /* Hardlinking to unreadable or unwritable sources is dangerous. */
730 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
737 * may_linkat - Check permissions for creating a hardlink
738 * @link: the source to hardlink from
740 * Block hardlink when all of:
741 * - sysctl_protected_hardlinks enabled
742 * - fsuid does not match inode
743 * - hardlink source is unsafe (see safe_hardlink_source() above)
746 * Returns 0 if successful, -ve on error.
748 static int may_linkat(struct path
*link
)
750 const struct cred
*cred
;
753 if (!sysctl_protected_hardlinks
)
756 cred
= current_cred();
757 inode
= link
->dentry
->d_inode
;
759 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
760 * otherwise, it must be a safe source.
762 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
766 audit_log_link_denied("linkat", link
);
770 static __always_inline
int
771 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
773 struct dentry
*dentry
= link
->dentry
;
777 BUG_ON(nd
->flags
& LOOKUP_RCU
);
779 if (link
->mnt
== nd
->path
.mnt
)
783 if (unlikely(current
->total_link_count
>= 40))
784 goto out_put_nd_path
;
787 current
->total_link_count
++;
790 nd_set_link(nd
, NULL
);
792 error
= security_inode_follow_link(link
->dentry
, nd
);
794 goto out_put_nd_path
;
796 nd
->last_type
= LAST_BIND
;
797 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
800 goto out_put_nd_path
;
805 error
= __vfs_follow_link(nd
, s
);
807 put_link(nd
, link
, *p
);
818 static int follow_up_rcu(struct path
*path
)
820 struct mount
*mnt
= real_mount(path
->mnt
);
821 struct mount
*parent
;
822 struct dentry
*mountpoint
;
824 parent
= mnt
->mnt_parent
;
825 if (&parent
->mnt
== path
->mnt
)
827 mountpoint
= mnt
->mnt_mountpoint
;
828 path
->dentry
= mountpoint
;
829 path
->mnt
= &parent
->mnt
;
834 * follow_up - Find the mountpoint of path's vfsmount
836 * Given a path, find the mountpoint of its source file system.
837 * Replace @path with the path of the mountpoint in the parent mount.
840 * Return 1 if we went up a level and 0 if we were already at the
843 int follow_up(struct path
*path
)
845 struct mount
*mnt
= real_mount(path
->mnt
);
846 struct mount
*parent
;
847 struct dentry
*mountpoint
;
849 br_read_lock(&vfsmount_lock
);
850 parent
= mnt
->mnt_parent
;
852 br_read_unlock(&vfsmount_lock
);
855 mntget(&parent
->mnt
);
856 mountpoint
= dget(mnt
->mnt_mountpoint
);
857 br_read_unlock(&vfsmount_lock
);
859 path
->dentry
= mountpoint
;
861 path
->mnt
= &parent
->mnt
;
866 * Perform an automount
867 * - return -EISDIR to tell follow_managed() to stop and return the path we
870 static int follow_automount(struct path
*path
, unsigned flags
,
873 struct vfsmount
*mnt
;
876 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
879 /* We don't want to mount if someone's just doing a stat -
880 * unless they're stat'ing a directory and appended a '/' to
883 * We do, however, want to mount if someone wants to open or
884 * create a file of any type under the mountpoint, wants to
885 * traverse through the mountpoint or wants to open the
886 * mounted directory. Also, autofs may mark negative dentries
887 * as being automount points. These will need the attentions
888 * of the daemon to instantiate them before they can be used.
890 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
891 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
892 path
->dentry
->d_inode
)
895 current
->total_link_count
++;
896 if (current
->total_link_count
>= 40)
899 mnt
= path
->dentry
->d_op
->d_automount(path
);
902 * The filesystem is allowed to return -EISDIR here to indicate
903 * it doesn't want to automount. For instance, autofs would do
904 * this so that its userspace daemon can mount on this dentry.
906 * However, we can only permit this if it's a terminal point in
907 * the path being looked up; if it wasn't then the remainder of
908 * the path is inaccessible and we should say so.
910 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
915 if (!mnt
) /* mount collision */
919 /* lock_mount() may release path->mnt on error */
923 err
= finish_automount(mnt
, path
);
927 /* Someone else made a mount here whilst we were busy */
932 path
->dentry
= dget(mnt
->mnt_root
);
941 * Handle a dentry that is managed in some way.
942 * - Flagged for transit management (autofs)
943 * - Flagged as mountpoint
944 * - Flagged as automount point
946 * This may only be called in refwalk mode.
948 * Serialization is taken care of in namespace.c
950 static int follow_managed(struct path
*path
, unsigned flags
)
952 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
954 bool need_mntput
= false;
957 /* Given that we're not holding a lock here, we retain the value in a
958 * local variable for each dentry as we look at it so that we don't see
959 * the components of that value change under us */
960 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
961 managed
&= DCACHE_MANAGED_DENTRY
,
962 unlikely(managed
!= 0)) {
963 /* Allow the filesystem to manage the transit without i_mutex
965 if (managed
& DCACHE_MANAGE_TRANSIT
) {
966 BUG_ON(!path
->dentry
->d_op
);
967 BUG_ON(!path
->dentry
->d_op
->d_manage
);
968 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
973 /* Transit to a mounted filesystem. */
974 if (managed
& DCACHE_MOUNTED
) {
975 struct vfsmount
*mounted
= lookup_mnt(path
);
981 path
->dentry
= dget(mounted
->mnt_root
);
986 /* Something is mounted on this dentry in another
987 * namespace and/or whatever was mounted there in this
988 * namespace got unmounted before we managed to get the
992 /* Handle an automount point */
993 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
994 ret
= follow_automount(path
, flags
, &need_mntput
);
1000 /* We didn't change the current path point */
1004 if (need_mntput
&& path
->mnt
== mnt
)
1008 return ret
< 0 ? ret
: need_mntput
;
1011 int follow_down_one(struct path
*path
)
1013 struct vfsmount
*mounted
;
1015 mounted
= lookup_mnt(path
);
1019 path
->mnt
= mounted
;
1020 path
->dentry
= dget(mounted
->mnt_root
);
1026 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
1028 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
1029 dentry
->d_op
->d_manage(dentry
, true) < 0);
1033 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1034 * we meet a managed dentry that would need blocking.
1036 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1037 struct inode
**inode
)
1040 struct mount
*mounted
;
1042 * Don't forget we might have a non-mountpoint managed dentry
1043 * that wants to block transit.
1045 if (unlikely(managed_dentry_might_block(path
->dentry
)))
1048 if (!d_mountpoint(path
->dentry
))
1051 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1054 path
->mnt
= &mounted
->mnt
;
1055 path
->dentry
= mounted
->mnt
.mnt_root
;
1056 nd
->flags
|= LOOKUP_JUMPED
;
1057 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1059 * Update the inode too. We don't need to re-check the
1060 * dentry sequence number here after this d_inode read,
1061 * because a mount-point is always pinned.
1063 *inode
= path
->dentry
->d_inode
;
1068 static void follow_mount_rcu(struct nameidata
*nd
)
1070 while (d_mountpoint(nd
->path
.dentry
)) {
1071 struct mount
*mounted
;
1072 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
1075 nd
->path
.mnt
= &mounted
->mnt
;
1076 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1077 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1081 static int follow_dotdot_rcu(struct nameidata
*nd
)
1086 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1087 nd
->path
.mnt
== nd
->root
.mnt
) {
1090 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1091 struct dentry
*old
= nd
->path
.dentry
;
1092 struct dentry
*parent
= old
->d_parent
;
1095 seq
= read_seqcount_begin(&parent
->d_seq
);
1096 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1098 nd
->path
.dentry
= parent
;
1102 if (!follow_up_rcu(&nd
->path
))
1104 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1106 follow_mount_rcu(nd
);
1107 nd
->inode
= nd
->path
.dentry
->d_inode
;
1111 nd
->flags
&= ~LOOKUP_RCU
;
1112 if (!(nd
->flags
& LOOKUP_ROOT
))
1113 nd
->root
.mnt
= NULL
;
1119 * Follow down to the covering mount currently visible to userspace. At each
1120 * point, the filesystem owning that dentry may be queried as to whether the
1121 * caller is permitted to proceed or not.
1123 int follow_down(struct path
*path
)
1128 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1129 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1130 /* Allow the filesystem to manage the transit without i_mutex
1133 * We indicate to the filesystem if someone is trying to mount
1134 * something here. This gives autofs the chance to deny anyone
1135 * other than its daemon the right to mount on its
1138 * The filesystem may sleep at this point.
1140 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1141 BUG_ON(!path
->dentry
->d_op
);
1142 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1143 ret
= path
->dentry
->d_op
->d_manage(
1144 path
->dentry
, false);
1146 return ret
== -EISDIR
? 0 : ret
;
1149 /* Transit to a mounted filesystem. */
1150 if (managed
& DCACHE_MOUNTED
) {
1151 struct vfsmount
*mounted
= lookup_mnt(path
);
1156 path
->mnt
= mounted
;
1157 path
->dentry
= dget(mounted
->mnt_root
);
1161 /* Don't handle automount points here */
1168 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1170 static void follow_mount(struct path
*path
)
1172 while (d_mountpoint(path
->dentry
)) {
1173 struct vfsmount
*mounted
= lookup_mnt(path
);
1178 path
->mnt
= mounted
;
1179 path
->dentry
= dget(mounted
->mnt_root
);
1183 static void follow_dotdot(struct nameidata
*nd
)
1188 struct dentry
*old
= nd
->path
.dentry
;
1190 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1191 nd
->path
.mnt
== nd
->root
.mnt
) {
1194 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1195 /* rare case of legitimate dget_parent()... */
1196 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1200 if (!follow_up(&nd
->path
))
1203 follow_mount(&nd
->path
);
1204 nd
->inode
= nd
->path
.dentry
->d_inode
;
1208 * This looks up the name in dcache, possibly revalidates the old dentry and
1209 * allocates a new one if not found or not valid. In the need_lookup argument
1210 * returns whether i_op->lookup is necessary.
1212 * dir->d_inode->i_mutex must be held
1214 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1215 unsigned int flags
, bool *need_lookup
)
1217 struct dentry
*dentry
;
1220 *need_lookup
= false;
1221 dentry
= d_lookup(dir
, name
);
1223 if (d_need_lookup(dentry
)) {
1224 *need_lookup
= true;
1225 } else if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1226 error
= d_revalidate(dentry
, flags
);
1227 if (unlikely(error
<= 0)) {
1230 return ERR_PTR(error
);
1231 } else if (!d_invalidate(dentry
)) {
1240 dentry
= d_alloc(dir
, name
);
1241 if (unlikely(!dentry
))
1242 return ERR_PTR(-ENOMEM
);
1244 *need_lookup
= true;
1250 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1251 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1253 * dir->d_inode->i_mutex must be held
1255 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1260 /* Don't create child dentry for a dead directory. */
1261 if (unlikely(IS_DEADDIR(dir
))) {
1263 return ERR_PTR(-ENOENT
);
1266 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1267 if (unlikely(old
)) {
1274 static struct dentry
*__lookup_hash(struct qstr
*name
,
1275 struct dentry
*base
, unsigned int flags
)
1278 struct dentry
*dentry
;
1280 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1284 return lookup_real(base
->d_inode
, dentry
, flags
);
1288 * It's more convoluted than I'd like it to be, but... it's still fairly
1289 * small and for now I'd prefer to have fast path as straight as possible.
1290 * It _is_ time-critical.
1292 static int lookup_fast(struct nameidata
*nd
, struct qstr
*name
,
1293 struct path
*path
, struct inode
**inode
)
1295 struct vfsmount
*mnt
= nd
->path
.mnt
;
1296 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1302 * Rename seqlock is not required here because in the off chance
1303 * of a false negative due to a concurrent rename, we're going to
1304 * do the non-racy lookup, below.
1306 if (nd
->flags
& LOOKUP_RCU
) {
1308 dentry
= __d_lookup_rcu(parent
, name
, &seq
, nd
->inode
);
1313 * This sequence count validates that the inode matches
1314 * the dentry name information from lookup.
1316 *inode
= dentry
->d_inode
;
1317 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1321 * This sequence count validates that the parent had no
1322 * changes while we did the lookup of the dentry above.
1324 * The memory barrier in read_seqcount_begin of child is
1325 * enough, we can use __read_seqcount_retry here.
1327 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1331 if (unlikely(d_need_lookup(dentry
)))
1333 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1334 status
= d_revalidate(dentry
, nd
->flags
);
1335 if (unlikely(status
<= 0)) {
1336 if (status
!= -ECHILD
)
1342 path
->dentry
= dentry
;
1343 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1345 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1349 if (unlazy_walk(nd
, dentry
))
1352 dentry
= __d_lookup(parent
, name
);
1355 if (unlikely(!dentry
))
1358 if (unlikely(d_need_lookup(dentry
))) {
1363 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1364 status
= d_revalidate(dentry
, nd
->flags
);
1365 if (unlikely(status
<= 0)) {
1370 if (!d_invalidate(dentry
)) {
1377 path
->dentry
= dentry
;
1378 err
= follow_managed(path
, nd
->flags
);
1379 if (unlikely(err
< 0)) {
1380 path_put_conditional(path
, nd
);
1384 nd
->flags
|= LOOKUP_JUMPED
;
1385 *inode
= path
->dentry
->d_inode
;
1392 /* Fast lookup failed, do it the slow way */
1393 static int lookup_slow(struct nameidata
*nd
, struct qstr
*name
,
1396 struct dentry
*dentry
, *parent
;
1399 parent
= nd
->path
.dentry
;
1400 BUG_ON(nd
->inode
!= parent
->d_inode
);
1402 mutex_lock(&parent
->d_inode
->i_mutex
);
1403 dentry
= __lookup_hash(name
, parent
, nd
->flags
);
1404 mutex_unlock(&parent
->d_inode
->i_mutex
);
1406 return PTR_ERR(dentry
);
1407 path
->mnt
= nd
->path
.mnt
;
1408 path
->dentry
= dentry
;
1409 err
= follow_managed(path
, nd
->flags
);
1410 if (unlikely(err
< 0)) {
1411 path_put_conditional(path
, nd
);
1415 nd
->flags
|= LOOKUP_JUMPED
;
1419 static inline int may_lookup(struct nameidata
*nd
)
1421 if (nd
->flags
& LOOKUP_RCU
) {
1422 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1425 if (unlazy_walk(nd
, NULL
))
1428 return inode_permission(nd
->inode
, MAY_EXEC
);
1431 static inline int handle_dots(struct nameidata
*nd
, int type
)
1433 if (type
== LAST_DOTDOT
) {
1434 if (nd
->flags
& LOOKUP_RCU
) {
1435 if (follow_dotdot_rcu(nd
))
1443 static void terminate_walk(struct nameidata
*nd
)
1445 if (!(nd
->flags
& LOOKUP_RCU
)) {
1446 path_put(&nd
->path
);
1448 nd
->flags
&= ~LOOKUP_RCU
;
1449 if (!(nd
->flags
& LOOKUP_ROOT
))
1450 nd
->root
.mnt
= NULL
;
1456 * Do we need to follow links? We _really_ want to be able
1457 * to do this check without having to look at inode->i_op,
1458 * so we keep a cache of "no, this doesn't need follow_link"
1459 * for the common case.
1461 static inline int should_follow_link(struct inode
*inode
, int follow
)
1463 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1464 if (likely(inode
->i_op
->follow_link
))
1467 /* This gets set once for the inode lifetime */
1468 spin_lock(&inode
->i_lock
);
1469 inode
->i_opflags
|= IOP_NOFOLLOW
;
1470 spin_unlock(&inode
->i_lock
);
1475 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1476 struct qstr
*name
, int type
, int follow
)
1478 struct inode
*inode
;
1481 * "." and ".." are special - ".." especially so because it has
1482 * to be able to know about the current root directory and
1483 * parent relationships.
1485 if (unlikely(type
!= LAST_NORM
))
1486 return handle_dots(nd
, type
);
1487 err
= lookup_fast(nd
, name
, path
, &inode
);
1488 if (unlikely(err
)) {
1492 err
= lookup_slow(nd
, name
, path
);
1496 inode
= path
->dentry
->d_inode
;
1502 if (should_follow_link(inode
, follow
)) {
1503 if (nd
->flags
& LOOKUP_RCU
) {
1504 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1509 BUG_ON(inode
!= path
->dentry
->d_inode
);
1512 path_to_nameidata(path
, nd
);
1517 path_to_nameidata(path
, nd
);
1524 * This limits recursive symlink follows to 8, while
1525 * limiting consecutive symlinks to 40.
1527 * Without that kind of total limit, nasty chains of consecutive
1528 * symlinks can cause almost arbitrarily long lookups.
1530 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1534 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1535 path_put_conditional(path
, nd
);
1536 path_put(&nd
->path
);
1539 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1542 current
->link_count
++;
1545 struct path link
= *path
;
1548 res
= follow_link(&link
, nd
, &cookie
);
1551 res
= walk_component(nd
, path
, &nd
->last
,
1552 nd
->last_type
, LOOKUP_FOLLOW
);
1553 put_link(nd
, &link
, cookie
);
1556 current
->link_count
--;
1562 * We really don't want to look at inode->i_op->lookup
1563 * when we don't have to. So we keep a cache bit in
1564 * the inode ->i_opflags field that says "yes, we can
1565 * do lookup on this inode".
1567 static inline int can_lookup(struct inode
*inode
)
1569 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1571 if (likely(!inode
->i_op
->lookup
))
1574 /* We do this once for the lifetime of the inode */
1575 spin_lock(&inode
->i_lock
);
1576 inode
->i_opflags
|= IOP_LOOKUP
;
1577 spin_unlock(&inode
->i_lock
);
1582 * We can do the critical dentry name comparison and hashing
1583 * operations one word at a time, but we are limited to:
1585 * - Architectures with fast unaligned word accesses. We could
1586 * do a "get_unaligned()" if this helps and is sufficiently
1589 * - Little-endian machines (so that we can generate the mask
1590 * of low bytes efficiently). Again, we *could* do a byte
1591 * swapping load on big-endian architectures if that is not
1592 * expensive enough to make the optimization worthless.
1594 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1595 * do not trap on the (extremely unlikely) case of a page
1596 * crossing operation.
1598 * - Furthermore, we need an efficient 64-bit compile for the
1599 * 64-bit case in order to generate the "number of bytes in
1600 * the final mask". Again, that could be replaced with a
1601 * efficient population count instruction or similar.
1603 #ifdef CONFIG_DCACHE_WORD_ACCESS
1605 #include <asm/word-at-a-time.h>
1609 static inline unsigned int fold_hash(unsigned long hash
)
1611 hash
+= hash
>> (8*sizeof(int));
1615 #else /* 32-bit case */
1617 #define fold_hash(x) (x)
1621 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1623 unsigned long a
, mask
;
1624 unsigned long hash
= 0;
1627 a
= load_unaligned_zeropad(name
);
1628 if (len
< sizeof(unsigned long))
1632 name
+= sizeof(unsigned long);
1633 len
-= sizeof(unsigned long);
1637 mask
= ~(~0ul << len
*8);
1640 return fold_hash(hash
);
1642 EXPORT_SYMBOL(full_name_hash
);
1645 * Calculate the length and hash of the path component, and
1646 * return the length of the component;
1648 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1650 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1651 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1654 len
= -sizeof(unsigned long);
1656 hash
= (hash
+ a
) * 9;
1657 len
+= sizeof(unsigned long);
1658 a
= load_unaligned_zeropad(name
+len
);
1659 b
= a
^ REPEAT_BYTE('/');
1660 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1662 adata
= prep_zero_mask(a
, adata
, &constants
);
1663 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1665 mask
= create_zero_mask(adata
| bdata
);
1667 hash
+= a
& zero_bytemask(mask
);
1668 *hashp
= fold_hash(hash
);
1670 return len
+ find_zero(mask
);
1675 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1677 unsigned long hash
= init_name_hash();
1679 hash
= partial_name_hash(*name
++, hash
);
1680 return end_name_hash(hash
);
1682 EXPORT_SYMBOL(full_name_hash
);
1685 * We know there's a real path component here of at least
1688 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1690 unsigned long hash
= init_name_hash();
1691 unsigned long len
= 0, c
;
1693 c
= (unsigned char)*name
;
1696 hash
= partial_name_hash(c
, hash
);
1697 c
= (unsigned char)name
[len
];
1698 } while (c
&& c
!= '/');
1699 *hashp
= end_name_hash(hash
);
1707 * This is the basic name resolution function, turning a pathname into
1708 * the final dentry. We expect 'base' to be positive and a directory.
1710 * Returns 0 and nd will have valid dentry and mnt on success.
1711 * Returns error and drops reference to input namei data on failure.
1713 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1723 /* At this point we know we have a real path component. */
1729 err
= may_lookup(nd
);
1733 len
= hash_name(name
, &this.hash
);
1738 if (name
[0] == '.') switch (len
) {
1740 if (name
[1] == '.') {
1742 nd
->flags
|= LOOKUP_JUMPED
;
1748 if (likely(type
== LAST_NORM
)) {
1749 struct dentry
*parent
= nd
->path
.dentry
;
1750 nd
->flags
&= ~LOOKUP_JUMPED
;
1751 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1752 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1760 goto last_component
;
1762 * If it wasn't NUL, we know it was '/'. Skip that
1763 * slash, and continue until no more slashes.
1767 } while (unlikely(name
[len
] == '/'));
1769 goto last_component
;
1772 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1777 err
= nested_symlink(&next
, nd
);
1781 if (can_lookup(nd
->inode
))
1785 /* here ends the main loop */
1789 nd
->last_type
= type
;
1796 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1797 struct nameidata
*nd
, struct file
**fp
)
1801 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1802 nd
->flags
= flags
| LOOKUP_JUMPED
;
1804 if (flags
& LOOKUP_ROOT
) {
1805 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1807 if (!inode
->i_op
->lookup
)
1809 retval
= inode_permission(inode
, MAY_EXEC
);
1813 nd
->path
= nd
->root
;
1815 if (flags
& LOOKUP_RCU
) {
1817 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1819 path_get(&nd
->path
);
1824 nd
->root
.mnt
= NULL
;
1827 if (flags
& LOOKUP_RCU
) {
1832 path_get(&nd
->root
);
1834 nd
->path
= nd
->root
;
1835 } else if (dfd
== AT_FDCWD
) {
1836 if (flags
& LOOKUP_RCU
) {
1837 struct fs_struct
*fs
= current
->fs
;
1843 seq
= read_seqcount_begin(&fs
->seq
);
1845 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1846 } while (read_seqcount_retry(&fs
->seq
, seq
));
1848 get_fs_pwd(current
->fs
, &nd
->path
);
1851 struct fd f
= fdget_raw(dfd
);
1852 struct dentry
*dentry
;
1857 dentry
= f
.file
->f_path
.dentry
;
1860 if (!S_ISDIR(dentry
->d_inode
->i_mode
)) {
1865 retval
= inode_permission(dentry
->d_inode
, MAY_EXEC
);
1872 nd
->path
= f
.file
->f_path
;
1873 if (flags
& LOOKUP_RCU
) {
1876 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1879 path_get(&nd
->path
);
1884 nd
->inode
= nd
->path
.dentry
->d_inode
;
1888 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1890 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1891 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1893 nd
->flags
&= ~LOOKUP_PARENT
;
1894 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1895 nd
->flags
& LOOKUP_FOLLOW
);
1898 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1899 static int path_lookupat(int dfd
, const char *name
,
1900 unsigned int flags
, struct nameidata
*nd
)
1902 struct file
*base
= NULL
;
1907 * Path walking is largely split up into 2 different synchronisation
1908 * schemes, rcu-walk and ref-walk (explained in
1909 * Documentation/filesystems/path-lookup.txt). These share much of the
1910 * path walk code, but some things particularly setup, cleanup, and
1911 * following mounts are sufficiently divergent that functions are
1912 * duplicated. Typically there is a function foo(), and its RCU
1913 * analogue, foo_rcu().
1915 * -ECHILD is the error number of choice (just to avoid clashes) that
1916 * is returned if some aspect of an rcu-walk fails. Such an error must
1917 * be handled by restarting a traditional ref-walk (which will always
1918 * be able to complete).
1920 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1925 current
->total_link_count
= 0;
1926 err
= link_path_walk(name
, nd
);
1928 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1929 err
= lookup_last(nd
, &path
);
1932 struct path link
= path
;
1933 err
= may_follow_link(&link
, nd
);
1936 nd
->flags
|= LOOKUP_PARENT
;
1937 err
= follow_link(&link
, nd
, &cookie
);
1940 err
= lookup_last(nd
, &path
);
1941 put_link(nd
, &link
, cookie
);
1946 err
= complete_walk(nd
);
1948 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1949 if (!nd
->inode
->i_op
->lookup
) {
1950 path_put(&nd
->path
);
1958 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1959 path_put(&nd
->root
);
1960 nd
->root
.mnt
= NULL
;
1965 static int do_path_lookup(int dfd
, const char *name
,
1966 unsigned int flags
, struct nameidata
*nd
)
1968 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1969 if (unlikely(retval
== -ECHILD
))
1970 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1971 if (unlikely(retval
== -ESTALE
))
1972 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1974 if (likely(!retval
)) {
1975 if (unlikely(!audit_dummy_context())) {
1976 if (nd
->path
.dentry
&& nd
->inode
)
1977 audit_inode(name
, nd
->path
.dentry
);
1983 /* does lookup, returns the object with parent locked */
1984 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
1986 struct nameidata nd
;
1988 int err
= do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, &nd
);
1990 return ERR_PTR(err
);
1991 if (nd
.last_type
!= LAST_NORM
) {
1993 return ERR_PTR(-EINVAL
);
1995 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1996 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
1998 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2006 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2008 struct nameidata nd
;
2009 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
2016 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2017 * @dentry: pointer to dentry of the base directory
2018 * @mnt: pointer to vfs mount of the base directory
2019 * @name: pointer to file name
2020 * @flags: lookup flags
2021 * @path: pointer to struct path to fill
2023 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2024 const char *name
, unsigned int flags
,
2027 struct nameidata nd
;
2029 nd
.root
.dentry
= dentry
;
2031 BUG_ON(flags
& LOOKUP_PARENT
);
2032 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2033 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
2040 * Restricted form of lookup. Doesn't follow links, single-component only,
2041 * needs parent already locked. Doesn't follow mounts.
2044 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2046 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2050 * lookup_one_len - filesystem helper to lookup single pathname component
2051 * @name: pathname component to lookup
2052 * @base: base directory to lookup from
2053 * @len: maximum length @len should be interpreted to
2055 * Note that this routine is purely a helper for filesystem usage and should
2056 * not be called by generic code. Also note that by using this function the
2057 * nameidata argument is passed to the filesystem methods and a filesystem
2058 * using this helper needs to be prepared for that.
2060 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2066 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2070 this.hash
= full_name_hash(name
, len
);
2072 return ERR_PTR(-EACCES
);
2075 c
= *(const unsigned char *)name
++;
2076 if (c
== '/' || c
== '\0')
2077 return ERR_PTR(-EACCES
);
2080 * See if the low-level filesystem might want
2081 * to use its own hash..
2083 if (base
->d_flags
& DCACHE_OP_HASH
) {
2084 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
2086 return ERR_PTR(err
);
2089 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2091 return ERR_PTR(err
);
2093 return __lookup_hash(&this, base
, 0);
2096 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2097 struct path
*path
, int *empty
)
2099 struct nameidata nd
;
2100 char *tmp
= getname_flags(name
, flags
, empty
);
2101 int err
= PTR_ERR(tmp
);
2104 BUG_ON(flags
& LOOKUP_PARENT
);
2106 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
2114 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2117 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2120 static int user_path_parent(int dfd
, const char __user
*path
,
2121 struct nameidata
*nd
, char **name
)
2123 char *s
= getname(path
);
2129 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
2139 * It's inline, so penalty for filesystems that don't use sticky bit is
2142 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
2144 kuid_t fsuid
= current_fsuid();
2146 if (!(dir
->i_mode
& S_ISVTX
))
2148 if (uid_eq(inode
->i_uid
, fsuid
))
2150 if (uid_eq(dir
->i_uid
, fsuid
))
2152 return !inode_capable(inode
, CAP_FOWNER
);
2156 * Check whether we can remove a link victim from directory dir, check
2157 * whether the type of victim is right.
2158 * 1. We can't do it if dir is read-only (done in permission())
2159 * 2. We should have write and exec permissions on dir
2160 * 3. We can't remove anything from append-only dir
2161 * 4. We can't do anything with immutable dir (done in permission())
2162 * 5. If the sticky bit on dir is set we should either
2163 * a. be owner of dir, or
2164 * b. be owner of victim, or
2165 * c. have CAP_FOWNER capability
2166 * 6. If the victim is append-only or immutable we can't do antyhing with
2167 * links pointing to it.
2168 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2169 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2170 * 9. We can't remove a root or mountpoint.
2171 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2172 * nfs_async_unlink().
2174 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
2178 if (!victim
->d_inode
)
2181 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2182 audit_inode_child(victim
, dir
);
2184 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2189 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
2190 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
2193 if (!S_ISDIR(victim
->d_inode
->i_mode
))
2195 if (IS_ROOT(victim
))
2197 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
2199 if (IS_DEADDIR(dir
))
2201 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2206 /* Check whether we can create an object with dentry child in directory
2208 * 1. We can't do it if child already exists (open has special treatment for
2209 * this case, but since we are inlined it's OK)
2210 * 2. We can't do it if dir is read-only (done in permission())
2211 * 3. We should have write and exec permissions on dir
2212 * 4. We can't do it if dir is immutable (done in permission())
2214 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2218 if (IS_DEADDIR(dir
))
2220 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2224 * p1 and p2 should be directories on the same fs.
2226 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2231 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2235 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2237 p
= d_ancestor(p2
, p1
);
2239 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2240 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2244 p
= d_ancestor(p1
, p2
);
2246 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2247 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2251 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2252 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2256 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2258 mutex_unlock(&p1
->d_inode
->i_mutex
);
2260 mutex_unlock(&p2
->d_inode
->i_mutex
);
2261 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2265 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2268 int error
= may_create(dir
, dentry
);
2272 if (!dir
->i_op
->create
)
2273 return -EACCES
; /* shouldn't it be ENOSYS? */
2276 error
= security_inode_create(dir
, dentry
, mode
);
2279 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2281 fsnotify_create(dir
, dentry
);
2285 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2287 struct dentry
*dentry
= path
->dentry
;
2288 struct inode
*inode
= dentry
->d_inode
;
2298 switch (inode
->i_mode
& S_IFMT
) {
2302 if (acc_mode
& MAY_WRITE
)
2307 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2316 error
= inode_permission(inode
, acc_mode
);
2321 * An append-only file must be opened in append mode for writing.
2323 if (IS_APPEND(inode
)) {
2324 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2330 /* O_NOATIME can only be set by the owner or superuser */
2331 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2337 static int handle_truncate(struct file
*filp
)
2339 struct path
*path
= &filp
->f_path
;
2340 struct inode
*inode
= path
->dentry
->d_inode
;
2341 int error
= get_write_access(inode
);
2345 * Refuse to truncate files with mandatory locks held on them.
2347 error
= locks_verify_locked(inode
);
2349 error
= security_path_truncate(path
);
2351 error
= do_truncate(path
->dentry
, 0,
2352 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2355 put_write_access(inode
);
2359 static inline int open_to_namei_flags(int flag
)
2361 if ((flag
& O_ACCMODE
) == 3)
2366 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2368 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2372 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2376 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2380 * Attempt to atomically look up, create and open a file from a negative
2383 * Returns 0 if successful. The file will have been created and attached to
2384 * @file by the filesystem calling finish_open().
2386 * Returns 1 if the file was looked up only or didn't need creating. The
2387 * caller will need to perform the open themselves. @path will have been
2388 * updated to point to the new dentry. This may be negative.
2390 * Returns an error code otherwise.
2392 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2393 struct path
*path
, struct file
*file
,
2394 const struct open_flags
*op
,
2395 bool got_write
, bool need_lookup
,
2398 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2399 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2403 int create_error
= 0;
2404 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2406 BUG_ON(dentry
->d_inode
);
2408 /* Don't create child dentry for a dead directory. */
2409 if (unlikely(IS_DEADDIR(dir
))) {
2415 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2416 mode
&= ~current_umask();
2418 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
)) {
2419 open_flag
&= ~O_TRUNC
;
2420 *opened
|= FILE_CREATED
;
2424 * Checking write permission is tricky, bacuse we don't know if we are
2425 * going to actually need it: O_CREAT opens should work as long as the
2426 * file exists. But checking existence breaks atomicity. The trick is
2427 * to check access and if not granted clear O_CREAT from the flags.
2429 * Another problem is returing the "right" error value (e.g. for an
2430 * O_EXCL open we want to return EEXIST not EROFS).
2432 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2433 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2434 if (!(open_flag
& O_CREAT
)) {
2436 * No O_CREATE -> atomicity not a requirement -> fall
2437 * back to lookup + open
2440 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2441 /* Fall back and fail with the right error */
2442 create_error
= -EROFS
;
2445 /* No side effects, safe to clear O_CREAT */
2446 create_error
= -EROFS
;
2447 open_flag
&= ~O_CREAT
;
2451 if (open_flag
& O_CREAT
) {
2452 error
= may_o_create(&nd
->path
, dentry
, mode
);
2454 create_error
= error
;
2455 if (open_flag
& O_EXCL
)
2457 open_flag
&= ~O_CREAT
;
2461 if (nd
->flags
& LOOKUP_DIRECTORY
)
2462 open_flag
|= O_DIRECTORY
;
2464 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2465 file
->f_path
.mnt
= nd
->path
.mnt
;
2466 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2469 if (create_error
&& error
== -ENOENT
)
2470 error
= create_error
;
2474 acc_mode
= op
->acc_mode
;
2475 if (*opened
& FILE_CREATED
) {
2476 fsnotify_create(dir
, dentry
);
2477 acc_mode
= MAY_OPEN
;
2480 if (error
) { /* returned 1, that is */
2481 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2485 if (file
->f_path
.dentry
) {
2487 dentry
= file
->f_path
.dentry
;
2489 if (create_error
&& dentry
->d_inode
== NULL
) {
2490 error
= create_error
;
2497 * We didn't have the inode before the open, so check open permission
2500 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2510 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2512 return PTR_ERR(dentry
);
2515 int open_flag
= op
->open_flag
;
2517 error
= create_error
;
2518 if ((open_flag
& O_EXCL
)) {
2519 if (!dentry
->d_inode
)
2521 } else if (!dentry
->d_inode
) {
2523 } else if ((open_flag
& O_TRUNC
) &&
2524 S_ISREG(dentry
->d_inode
->i_mode
)) {
2527 /* will fail later, go on to get the right error */
2531 path
->dentry
= dentry
;
2532 path
->mnt
= nd
->path
.mnt
;
2537 * Look up and maybe create and open the last component.
2539 * Must be called with i_mutex held on parent.
2541 * Returns 0 if the file was successfully atomically created (if necessary) and
2542 * opened. In this case the file will be returned attached to @file.
2544 * Returns 1 if the file was not completely opened at this time, though lookups
2545 * and creations will have been performed and the dentry returned in @path will
2546 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2547 * specified then a negative dentry may be returned.
2549 * An error code is returned otherwise.
2551 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2552 * cleared otherwise prior to returning.
2554 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2556 const struct open_flags
*op
,
2557 bool got_write
, int *opened
)
2559 struct dentry
*dir
= nd
->path
.dentry
;
2560 struct inode
*dir_inode
= dir
->d_inode
;
2561 struct dentry
*dentry
;
2565 *opened
&= ~FILE_CREATED
;
2566 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2568 return PTR_ERR(dentry
);
2570 /* Cached positive dentry: will open in f_op->open */
2571 if (!need_lookup
&& dentry
->d_inode
)
2574 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2575 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2576 need_lookup
, opened
);
2580 BUG_ON(dentry
->d_inode
);
2582 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2584 return PTR_ERR(dentry
);
2587 /* Negative dentry, just create the file */
2588 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2589 umode_t mode
= op
->mode
;
2590 if (!IS_POSIXACL(dir
->d_inode
))
2591 mode
&= ~current_umask();
2593 * This write is needed to ensure that a
2594 * rw->ro transition does not occur between
2595 * the time when the file is created and when
2596 * a permanent write count is taken through
2597 * the 'struct file' in finish_open().
2603 *opened
|= FILE_CREATED
;
2604 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2607 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2608 nd
->flags
& LOOKUP_EXCL
);
2613 path
->dentry
= dentry
;
2614 path
->mnt
= nd
->path
.mnt
;
2623 * Handle the last step of open()
2625 static int do_last(struct nameidata
*nd
, struct path
*path
,
2626 struct file
*file
, const struct open_flags
*op
,
2627 int *opened
, const char *pathname
)
2629 struct dentry
*dir
= nd
->path
.dentry
;
2630 int open_flag
= op
->open_flag
;
2631 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2632 bool got_write
= false;
2633 int acc_mode
= op
->acc_mode
;
2634 struct inode
*inode
;
2635 bool symlink_ok
= false;
2636 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2637 bool retried
= false;
2640 nd
->flags
&= ~LOOKUP_PARENT
;
2641 nd
->flags
|= op
->intent
;
2643 switch (nd
->last_type
) {
2646 error
= handle_dots(nd
, nd
->last_type
);
2651 error
= complete_walk(nd
);
2654 audit_inode(pathname
, nd
->path
.dentry
);
2655 if (open_flag
& O_CREAT
) {
2661 error
= complete_walk(nd
);
2664 audit_inode(pathname
, dir
);
2668 if (!(open_flag
& O_CREAT
)) {
2669 if (nd
->last
.name
[nd
->last
.len
])
2670 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2671 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2673 /* we _can_ be in RCU mode here */
2674 error
= lookup_fast(nd
, &nd
->last
, path
, &inode
);
2681 BUG_ON(nd
->inode
!= dir
->d_inode
);
2683 /* create side of things */
2685 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2686 * has been cleared when we got to the last component we are
2689 error
= complete_walk(nd
);
2693 audit_inode(pathname
, dir
);
2695 /* trailing slashes? */
2696 if (nd
->last
.name
[nd
->last
.len
])
2701 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2702 error
= mnt_want_write(nd
->path
.mnt
);
2706 * do _not_ fail yet - we might not need that or fail with
2707 * a different error; let lookup_open() decide; we'll be
2708 * dropping this one anyway.
2711 mutex_lock(&dir
->d_inode
->i_mutex
);
2712 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2713 mutex_unlock(&dir
->d_inode
->i_mutex
);
2719 if ((*opened
& FILE_CREATED
) ||
2720 !S_ISREG(file
->f_path
.dentry
->d_inode
->i_mode
))
2721 will_truncate
= false;
2723 audit_inode(pathname
, file
->f_path
.dentry
);
2727 if (*opened
& FILE_CREATED
) {
2728 /* Don't check for write permission, don't truncate */
2729 open_flag
&= ~O_TRUNC
;
2730 will_truncate
= false;
2731 acc_mode
= MAY_OPEN
;
2732 path_to_nameidata(path
, nd
);
2733 goto finish_open_created
;
2737 * create/update audit record if it already exists.
2739 if (path
->dentry
->d_inode
)
2740 audit_inode(pathname
, path
->dentry
);
2743 * If atomic_open() acquired write access it is dropped now due to
2744 * possible mount and symlink following (this might be optimized away if
2748 mnt_drop_write(nd
->path
.mnt
);
2753 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
2756 error
= follow_managed(path
, nd
->flags
);
2761 nd
->flags
|= LOOKUP_JUMPED
;
2763 BUG_ON(nd
->flags
& LOOKUP_RCU
);
2764 inode
= path
->dentry
->d_inode
;
2766 /* we _can_ be in RCU mode here */
2769 path_to_nameidata(path
, nd
);
2773 if (should_follow_link(inode
, !symlink_ok
)) {
2774 if (nd
->flags
& LOOKUP_RCU
) {
2775 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
2780 BUG_ON(inode
!= path
->dentry
->d_inode
);
2784 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
2785 path_to_nameidata(path
, nd
);
2787 save_parent
.dentry
= nd
->path
.dentry
;
2788 save_parent
.mnt
= mntget(path
->mnt
);
2789 nd
->path
.dentry
= path
->dentry
;
2793 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2794 error
= complete_walk(nd
);
2796 path_put(&save_parent
);
2800 if ((open_flag
& O_CREAT
) && S_ISDIR(nd
->inode
->i_mode
))
2803 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !nd
->inode
->i_op
->lookup
)
2805 audit_inode(pathname
, nd
->path
.dentry
);
2807 if (!S_ISREG(nd
->inode
->i_mode
))
2808 will_truncate
= false;
2810 if (will_truncate
) {
2811 error
= mnt_want_write(nd
->path
.mnt
);
2816 finish_open_created
:
2817 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2820 file
->f_path
.mnt
= nd
->path
.mnt
;
2821 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
2823 if (error
== -EOPENSTALE
)
2828 error
= open_check_o_direct(file
);
2831 error
= ima_file_check(file
, op
->acc_mode
);
2835 if (will_truncate
) {
2836 error
= handle_truncate(file
);
2842 mnt_drop_write(nd
->path
.mnt
);
2843 path_put(&save_parent
);
2848 path_put_conditional(path
, nd
);
2855 /* If no saved parent or already retried then can't retry */
2856 if (!save_parent
.dentry
|| retried
)
2859 BUG_ON(save_parent
.dentry
!= dir
);
2860 path_put(&nd
->path
);
2861 nd
->path
= save_parent
;
2862 nd
->inode
= dir
->d_inode
;
2863 save_parent
.mnt
= NULL
;
2864 save_parent
.dentry
= NULL
;
2866 mnt_drop_write(nd
->path
.mnt
);
2873 static struct file
*path_openat(int dfd
, const char *pathname
,
2874 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2876 struct file
*base
= NULL
;
2882 file
= get_empty_filp();
2884 return ERR_PTR(-ENFILE
);
2886 file
->f_flags
= op
->open_flag
;
2888 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2889 if (unlikely(error
))
2892 current
->total_link_count
= 0;
2893 error
= link_path_walk(pathname
, nd
);
2894 if (unlikely(error
))
2897 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2898 while (unlikely(error
> 0)) { /* trailing symlink */
2899 struct path link
= path
;
2901 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2902 path_put_conditional(&path
, nd
);
2903 path_put(&nd
->path
);
2907 error
= may_follow_link(&link
, nd
);
2908 if (unlikely(error
))
2910 nd
->flags
|= LOOKUP_PARENT
;
2911 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2912 error
= follow_link(&link
, nd
, &cookie
);
2913 if (unlikely(error
))
2915 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2916 put_link(nd
, &link
, cookie
);
2919 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2920 path_put(&nd
->root
);
2923 if (!(opened
& FILE_OPENED
)) {
2927 if (unlikely(error
)) {
2928 if (error
== -EOPENSTALE
) {
2929 if (flags
& LOOKUP_RCU
)
2934 file
= ERR_PTR(error
);
2939 struct file
*do_filp_open(int dfd
, const char *pathname
,
2940 const struct open_flags
*op
, int flags
)
2942 struct nameidata nd
;
2945 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2946 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2947 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2948 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2949 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2953 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2954 const char *name
, const struct open_flags
*op
, int flags
)
2956 struct nameidata nd
;
2960 nd
.root
.dentry
= dentry
;
2962 flags
|= LOOKUP_ROOT
;
2964 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2965 return ERR_PTR(-ELOOP
);
2967 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2968 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2969 file
= path_openat(-1, name
, &nd
, op
, flags
);
2970 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2971 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2975 struct dentry
*kern_path_create(int dfd
, const char *pathname
, struct path
*path
, int is_dir
)
2977 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2978 struct nameidata nd
;
2980 int error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
, &nd
);
2982 return ERR_PTR(error
);
2985 * Yucky last component or no last component at all?
2986 * (foo/., foo/.., /////)
2988 if (nd
.last_type
!= LAST_NORM
)
2990 nd
.flags
&= ~LOOKUP_PARENT
;
2991 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2993 /* don't fail immediately if it's r/o, at least try to report other errors */
2994 err2
= mnt_want_write(nd
.path
.mnt
);
2996 * Do the final lookup.
2998 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2999 dentry
= lookup_hash(&nd
);
3004 if (dentry
->d_inode
)
3007 * Special case - lookup gave negative, but... we had foo/bar/
3008 * From the vfs_mknod() POV we just have a negative dentry -
3009 * all is fine. Let's be bastards - you had / on the end, you've
3010 * been asking for (non-existent) directory. -ENOENT for you.
3012 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3016 if (unlikely(err2
)) {
3024 dentry
= ERR_PTR(error
);
3026 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3028 mnt_drop_write(nd
.path
.mnt
);
3033 EXPORT_SYMBOL(kern_path_create
);
3035 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3038 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3039 mnt_drop_write(path
->mnt
);
3042 EXPORT_SYMBOL(done_path_create
);
3044 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
, struct path
*path
, int is_dir
)
3046 char *tmp
= getname(pathname
);
3049 return ERR_CAST(tmp
);
3050 res
= kern_path_create(dfd
, tmp
, path
, is_dir
);
3054 EXPORT_SYMBOL(user_path_create
);
3056 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3058 int error
= may_create(dir
, dentry
);
3063 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3066 if (!dir
->i_op
->mknod
)
3069 error
= devcgroup_inode_mknod(mode
, dev
);
3073 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3077 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3079 fsnotify_create(dir
, dentry
);
3083 static int may_mknod(umode_t mode
)
3085 switch (mode
& S_IFMT
) {
3091 case 0: /* zero mode translates to S_IFREG */
3100 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3103 struct dentry
*dentry
;
3107 error
= may_mknod(mode
);
3111 dentry
= user_path_create(dfd
, filename
, &path
, 0);
3113 return PTR_ERR(dentry
);
3115 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3116 mode
&= ~current_umask();
3117 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3120 switch (mode
& S_IFMT
) {
3121 case 0: case S_IFREG
:
3122 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3124 case S_IFCHR
: case S_IFBLK
:
3125 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3126 new_decode_dev(dev
));
3128 case S_IFIFO
: case S_IFSOCK
:
3129 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3133 done_path_create(&path
, dentry
);
3137 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3139 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3142 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3144 int error
= may_create(dir
, dentry
);
3145 unsigned max_links
= dir
->i_sb
->s_max_links
;
3150 if (!dir
->i_op
->mkdir
)
3153 mode
&= (S_IRWXUGO
|S_ISVTX
);
3154 error
= security_inode_mkdir(dir
, dentry
, mode
);
3158 if (max_links
&& dir
->i_nlink
>= max_links
)
3161 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3163 fsnotify_mkdir(dir
, dentry
);
3167 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3169 struct dentry
*dentry
;
3173 dentry
= user_path_create(dfd
, pathname
, &path
, 1);
3175 return PTR_ERR(dentry
);
3177 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3178 mode
&= ~current_umask();
3179 error
= security_path_mkdir(&path
, dentry
, mode
);
3181 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3182 done_path_create(&path
, dentry
);
3186 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3188 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3192 * The dentry_unhash() helper will try to drop the dentry early: we
3193 * should have a usage count of 1 if we're the only user of this
3194 * dentry, and if that is true (possibly after pruning the dcache),
3195 * then we drop the dentry now.
3197 * A low-level filesystem can, if it choses, legally
3200 * if (!d_unhashed(dentry))
3203 * if it cannot handle the case of removing a directory
3204 * that is still in use by something else..
3206 void dentry_unhash(struct dentry
*dentry
)
3208 shrink_dcache_parent(dentry
);
3209 spin_lock(&dentry
->d_lock
);
3210 if (dentry
->d_count
== 1)
3212 spin_unlock(&dentry
->d_lock
);
3215 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3217 int error
= may_delete(dir
, dentry
, 1);
3222 if (!dir
->i_op
->rmdir
)
3226 mutex_lock(&dentry
->d_inode
->i_mutex
);
3229 if (d_mountpoint(dentry
))
3232 error
= security_inode_rmdir(dir
, dentry
);
3236 shrink_dcache_parent(dentry
);
3237 error
= dir
->i_op
->rmdir(dir
, dentry
);
3241 dentry
->d_inode
->i_flags
|= S_DEAD
;
3245 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3252 static long do_rmdir(int dfd
, const char __user
*pathname
)
3256 struct dentry
*dentry
;
3257 struct nameidata nd
;
3259 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
3263 switch(nd
.last_type
) {
3275 nd
.flags
&= ~LOOKUP_PARENT
;
3276 error
= mnt_want_write(nd
.path
.mnt
);
3280 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3281 dentry
= lookup_hash(&nd
);
3282 error
= PTR_ERR(dentry
);
3285 if (!dentry
->d_inode
) {
3289 error
= security_path_rmdir(&nd
.path
, dentry
);
3292 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3296 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3297 mnt_drop_write(nd
.path
.mnt
);
3304 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3306 return do_rmdir(AT_FDCWD
, pathname
);
3309 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3311 int error
= may_delete(dir
, dentry
, 0);
3316 if (!dir
->i_op
->unlink
)
3319 mutex_lock(&dentry
->d_inode
->i_mutex
);
3320 if (d_mountpoint(dentry
))
3323 error
= security_inode_unlink(dir
, dentry
);
3325 error
= dir
->i_op
->unlink(dir
, dentry
);
3330 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3332 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3333 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3334 fsnotify_link_count(dentry
->d_inode
);
3342 * Make sure that the actual truncation of the file will occur outside its
3343 * directory's i_mutex. Truncate can take a long time if there is a lot of
3344 * writeout happening, and we don't want to prevent access to the directory
3345 * while waiting on the I/O.
3347 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3351 struct dentry
*dentry
;
3352 struct nameidata nd
;
3353 struct inode
*inode
= NULL
;
3355 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
3360 if (nd
.last_type
!= LAST_NORM
)
3363 nd
.flags
&= ~LOOKUP_PARENT
;
3364 error
= mnt_want_write(nd
.path
.mnt
);
3368 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3369 dentry
= lookup_hash(&nd
);
3370 error
= PTR_ERR(dentry
);
3371 if (!IS_ERR(dentry
)) {
3372 /* Why not before? Because we want correct error value */
3373 if (nd
.last
.name
[nd
.last
.len
])
3375 inode
= dentry
->d_inode
;
3379 error
= security_path_unlink(&nd
.path
, dentry
);
3382 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
3386 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3388 iput(inode
); /* truncate the inode here */
3389 mnt_drop_write(nd
.path
.mnt
);
3396 error
= !dentry
->d_inode
? -ENOENT
:
3397 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
3401 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3403 if ((flag
& ~AT_REMOVEDIR
) != 0)
3406 if (flag
& AT_REMOVEDIR
)
3407 return do_rmdir(dfd
, pathname
);
3409 return do_unlinkat(dfd
, pathname
);
3412 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3414 return do_unlinkat(AT_FDCWD
, pathname
);
3417 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3419 int error
= may_create(dir
, dentry
);
3424 if (!dir
->i_op
->symlink
)
3427 error
= security_inode_symlink(dir
, dentry
, oldname
);
3431 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3433 fsnotify_create(dir
, dentry
);
3437 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3438 int, newdfd
, const char __user
*, newname
)
3442 struct dentry
*dentry
;
3445 from
= getname(oldname
);
3447 return PTR_ERR(from
);
3449 dentry
= user_path_create(newdfd
, newname
, &path
, 0);
3450 error
= PTR_ERR(dentry
);
3454 error
= security_path_symlink(&path
, dentry
, from
);
3456 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
);
3457 done_path_create(&path
, dentry
);
3463 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3465 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3468 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3470 struct inode
*inode
= old_dentry
->d_inode
;
3471 unsigned max_links
= dir
->i_sb
->s_max_links
;
3477 error
= may_create(dir
, new_dentry
);
3481 if (dir
->i_sb
!= inode
->i_sb
)
3485 * A link to an append-only or immutable file cannot be created.
3487 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3489 if (!dir
->i_op
->link
)
3491 if (S_ISDIR(inode
->i_mode
))
3494 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3498 mutex_lock(&inode
->i_mutex
);
3499 /* Make sure we don't allow creating hardlink to an unlinked file */
3500 if (inode
->i_nlink
== 0)
3502 else if (max_links
&& inode
->i_nlink
>= max_links
)
3505 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3506 mutex_unlock(&inode
->i_mutex
);
3508 fsnotify_link(dir
, inode
, new_dentry
);
3513 * Hardlinks are often used in delicate situations. We avoid
3514 * security-related surprises by not following symlinks on the
3517 * We don't follow them on the oldname either to be compatible
3518 * with linux 2.0, and to avoid hard-linking to directories
3519 * and other special files. --ADM
3521 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3522 int, newdfd
, const char __user
*, newname
, int, flags
)
3524 struct dentry
*new_dentry
;
3525 struct path old_path
, new_path
;
3529 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3532 * To use null names we require CAP_DAC_READ_SEARCH
3533 * This ensures that not everyone will be able to create
3534 * handlink using the passed filedescriptor.
3536 if (flags
& AT_EMPTY_PATH
) {
3537 if (!capable(CAP_DAC_READ_SEARCH
))
3542 if (flags
& AT_SYMLINK_FOLLOW
)
3543 how
|= LOOKUP_FOLLOW
;
3545 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3549 new_dentry
= user_path_create(newdfd
, newname
, &new_path
, 0);
3550 error
= PTR_ERR(new_dentry
);
3551 if (IS_ERR(new_dentry
))
3555 if (old_path
.mnt
!= new_path
.mnt
)
3557 error
= may_linkat(&old_path
);
3558 if (unlikely(error
))
3560 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3563 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3565 done_path_create(&new_path
, new_dentry
);
3567 path_put(&old_path
);
3572 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3574 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3578 * The worst of all namespace operations - renaming directory. "Perverted"
3579 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3581 * a) we can get into loop creation. Check is done in is_subdir().
3582 * b) race potential - two innocent renames can create a loop together.
3583 * That's where 4.4 screws up. Current fix: serialization on
3584 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3586 * c) we have to lock _three_ objects - parents and victim (if it exists).
3587 * And that - after we got ->i_mutex on parents (until then we don't know
3588 * whether the target exists). Solution: try to be smart with locking
3589 * order for inodes. We rely on the fact that tree topology may change
3590 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3591 * move will be locked. Thus we can rank directories by the tree
3592 * (ancestors first) and rank all non-directories after them.
3593 * That works since everybody except rename does "lock parent, lookup,
3594 * lock child" and rename is under ->s_vfs_rename_mutex.
3595 * HOWEVER, it relies on the assumption that any object with ->lookup()
3596 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3597 * we'd better make sure that there's no link(2) for them.
3598 * d) conversion from fhandle to dentry may come in the wrong moment - when
3599 * we are removing the target. Solution: we will have to grab ->i_mutex
3600 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3601 * ->i_mutex on parents, which works but leads to some truly excessive
3604 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3605 struct inode
*new_dir
, struct dentry
*new_dentry
)
3608 struct inode
*target
= new_dentry
->d_inode
;
3609 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3612 * If we are going to change the parent - check write permissions,
3613 * we'll need to flip '..'.
3615 if (new_dir
!= old_dir
) {
3616 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3621 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3627 mutex_lock(&target
->i_mutex
);
3630 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3634 if (max_links
&& !target
&& new_dir
!= old_dir
&&
3635 new_dir
->i_nlink
>= max_links
)
3639 shrink_dcache_parent(new_dentry
);
3640 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3645 target
->i_flags
|= S_DEAD
;
3646 dont_mount(new_dentry
);
3650 mutex_unlock(&target
->i_mutex
);
3653 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3654 d_move(old_dentry
,new_dentry
);
3658 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3659 struct inode
*new_dir
, struct dentry
*new_dentry
)
3661 struct inode
*target
= new_dentry
->d_inode
;
3664 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3670 mutex_lock(&target
->i_mutex
);
3673 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3676 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3681 dont_mount(new_dentry
);
3682 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3683 d_move(old_dentry
, new_dentry
);
3686 mutex_unlock(&target
->i_mutex
);
3691 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3692 struct inode
*new_dir
, struct dentry
*new_dentry
)
3695 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3696 const unsigned char *old_name
;
3698 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3701 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3705 if (!new_dentry
->d_inode
)
3706 error
= may_create(new_dir
, new_dentry
);
3708 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3712 if (!old_dir
->i_op
->rename
)
3715 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3718 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3720 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3722 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3723 new_dentry
->d_inode
, old_dentry
);
3724 fsnotify_oldname_free(old_name
);
3729 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3730 int, newdfd
, const char __user
*, newname
)
3732 struct dentry
*old_dir
, *new_dir
;
3733 struct dentry
*old_dentry
, *new_dentry
;
3734 struct dentry
*trap
;
3735 struct nameidata oldnd
, newnd
;
3740 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3744 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3749 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3752 old_dir
= oldnd
.path
.dentry
;
3754 if (oldnd
.last_type
!= LAST_NORM
)
3757 new_dir
= newnd
.path
.dentry
;
3758 if (newnd
.last_type
!= LAST_NORM
)
3761 error
= mnt_want_write(oldnd
.path
.mnt
);
3765 oldnd
.flags
&= ~LOOKUP_PARENT
;
3766 newnd
.flags
&= ~LOOKUP_PARENT
;
3767 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3769 trap
= lock_rename(new_dir
, old_dir
);
3771 old_dentry
= lookup_hash(&oldnd
);
3772 error
= PTR_ERR(old_dentry
);
3773 if (IS_ERR(old_dentry
))
3775 /* source must exist */
3777 if (!old_dentry
->d_inode
)
3779 /* unless the source is a directory trailing slashes give -ENOTDIR */
3780 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3782 if (oldnd
.last
.name
[oldnd
.last
.len
])
3784 if (newnd
.last
.name
[newnd
.last
.len
])
3787 /* source should not be ancestor of target */
3789 if (old_dentry
== trap
)
3791 new_dentry
= lookup_hash(&newnd
);
3792 error
= PTR_ERR(new_dentry
);
3793 if (IS_ERR(new_dentry
))
3795 /* target should not be an ancestor of source */
3797 if (new_dentry
== trap
)
3800 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3801 &newnd
.path
, new_dentry
);
3804 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3805 new_dir
->d_inode
, new_dentry
);
3811 unlock_rename(new_dir
, old_dir
);
3812 mnt_drop_write(oldnd
.path
.mnt
);
3814 path_put(&newnd
.path
);
3817 path_put(&oldnd
.path
);
3823 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3825 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3828 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3832 len
= PTR_ERR(link
);
3837 if (len
> (unsigned) buflen
)
3839 if (copy_to_user(buffer
, link
, len
))
3846 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3847 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3848 * using) it for any given inode is up to filesystem.
3850 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3852 struct nameidata nd
;
3857 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3859 return PTR_ERR(cookie
);
3861 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3862 if (dentry
->d_inode
->i_op
->put_link
)
3863 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3867 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3869 return __vfs_follow_link(nd
, link
);
3872 /* get the link contents into pagecache */
3873 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3877 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3878 page
= read_mapping_page(mapping
, 0, NULL
);
3883 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3887 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3889 struct page
*page
= NULL
;
3890 char *s
= page_getlink(dentry
, &page
);
3891 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3894 page_cache_release(page
);
3899 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3901 struct page
*page
= NULL
;
3902 nd_set_link(nd
, page_getlink(dentry
, &page
));
3906 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3908 struct page
*page
= cookie
;
3912 page_cache_release(page
);
3917 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3919 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3921 struct address_space
*mapping
= inode
->i_mapping
;
3926 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3928 flags
|= AOP_FLAG_NOFS
;
3931 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3932 flags
, &page
, &fsdata
);
3936 kaddr
= kmap_atomic(page
);
3937 memcpy(kaddr
, symname
, len
-1);
3938 kunmap_atomic(kaddr
);
3940 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3947 mark_inode_dirty(inode
);
3953 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3955 return __page_symlink(inode
, symname
, len
,
3956 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3959 const struct inode_operations page_symlink_inode_operations
= {
3960 .readlink
= generic_readlink
,
3961 .follow_link
= page_follow_link_light
,
3962 .put_link
= page_put_link
,
3965 EXPORT_SYMBOL(user_path_at
);
3966 EXPORT_SYMBOL(follow_down_one
);
3967 EXPORT_SYMBOL(follow_down
);
3968 EXPORT_SYMBOL(follow_up
);
3969 EXPORT_SYMBOL(get_write_access
); /* nfsd */
3970 EXPORT_SYMBOL(getname
);
3971 EXPORT_SYMBOL(lock_rename
);
3972 EXPORT_SYMBOL(lookup_one_len
);
3973 EXPORT_SYMBOL(page_follow_link_light
);
3974 EXPORT_SYMBOL(page_put_link
);
3975 EXPORT_SYMBOL(page_readlink
);
3976 EXPORT_SYMBOL(__page_symlink
);
3977 EXPORT_SYMBOL(page_symlink
);
3978 EXPORT_SYMBOL(page_symlink_inode_operations
);
3979 EXPORT_SYMBOL(kern_path
);
3980 EXPORT_SYMBOL(vfs_path_lookup
);
3981 EXPORT_SYMBOL(inode_permission
);
3982 EXPORT_SYMBOL(unlock_rename
);
3983 EXPORT_SYMBOL(vfs_create
);
3984 EXPORT_SYMBOL(vfs_follow_link
);
3985 EXPORT_SYMBOL(vfs_link
);
3986 EXPORT_SYMBOL(vfs_mkdir
);
3987 EXPORT_SYMBOL(vfs_mknod
);
3988 EXPORT_SYMBOL(generic_permission
);
3989 EXPORT_SYMBOL(vfs_readlink
);
3990 EXPORT_SYMBOL(vfs_rename
);
3991 EXPORT_SYMBOL(vfs_rmdir
);
3992 EXPORT_SYMBOL(vfs_symlink
);
3993 EXPORT_SYMBOL(vfs_unlink
);
3994 EXPORT_SYMBOL(dentry_unhash
);
3995 EXPORT_SYMBOL(generic_readlink
);