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 * Used to check for read/write/execute permissions on an inode.
323 * We use "fsuid" for this, letting us set arbitrary permissions
324 * for filesystem access without changing the "normal" uids which
325 * are used for other things.
327 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
329 int inode_permission(struct inode *inode, int mask)
333 if (unlikely(mask & MAY_WRITE)) {
334 umode_t mode = inode->i_mode;
337 * Nobody gets write access to a read-only fs.
339 if (IS_RDONLY(inode) &&
340 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
344 * Nobody gets write access to an immutable file.
346 if (IS_IMMUTABLE(inode))
350 retval = do_inode_permission(inode, mask);
354 retval = devcgroup_inode_permission(inode, mask);
358 return security_inode_permission(inode, mask);
362 * path_get - get a reference to a path
363 * @path: path to get the reference to
365 * Given a path increment the reference count to the dentry and the vfsmount.
367 void path_get(struct path *path)
372 EXPORT_SYMBOL(path_get);
375 * path_put - put a reference to a path
376 * @path: path to put the reference to
378 * Given a path decrement the reference count to the dentry and the vfsmount.
380 void path_put(struct path *path)
385 EXPORT_SYMBOL(path_put);
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). In situations when we can't
390 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
391 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
392 * mode. Refcounts are grabbed at the last known good point before rcu-walk
393 * got stuck, so ref-walk may continue from there. If this is not successful
394 * (eg. a seqcount has changed), then failure is returned and it's up to caller
395 * to restart the path walk from the beginning in ref-walk mode.
399 * unlazy_walk - try to switch to ref-walk mode.
400 * @nd: nameidata pathwalk data
401 * @dentry: child of nd->path.dentry or NULL
402 * Returns: 0 on success, -ECHILD on failure
404 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
405 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
406 * @nd or NULL. Must be called from rcu-walk context.
408 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
410 struct fs_struct *fs = current->fs;
411 struct dentry *parent = nd->path.dentry;
414 BUG_ON(!(nd->flags & LOOKUP_RCU));
415 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
417 spin_lock(&fs->lock);
418 if (nd->root.mnt != fs->root.mnt ||
419 nd->root.dentry != fs->root.dentry)
422 spin_lock(&parent->d_lock);
424 if (!__d_rcu_to_refcount(parent, nd->seq))
426 BUG_ON(nd->inode != parent->d_inode);
428 if (dentry->d_parent != parent)
430 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
431 if (!__d_rcu_to_refcount(dentry, nd->seq))
434 * If the sequence check on the child dentry passed, then
435 * the child has not been removed from its parent. This
436 * means the parent dentry must be valid and able to take
437 * a reference at this point.
439 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
440 BUG_ON(!parent->d_count);
442 spin_unlock(&dentry->d_lock);
444 spin_unlock(&parent->d_lock);
447 spin_unlock(&fs->lock);
449 mntget(nd->path.mnt);
452 br_read_unlock(&vfsmount_lock);
453 nd->flags &= ~LOOKUP_RCU;
457 spin_unlock(&dentry->d_lock);
459 spin_unlock(&parent->d_lock);
462 spin_unlock(&fs->lock);
466 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
468 return dentry->d_op->d_revalidate(dentry, flags);
472 * complete_walk - successful completion of path walk
473 * @nd: pointer nameidata
475 * If we had been in RCU mode, drop out of it and legitimize nd->path.
476 * Revalidate the final result, unless we'd already done that during
477 * the path walk or the filesystem doesn't ask for it. Return 0 on
478 * success, -error on failure. In case of failure caller does not
479 * need to drop nd->path.
481 static int complete_walk(struct nameidata *nd)
483 struct dentry *dentry = nd->path.dentry;
486 if (nd->flags & LOOKUP_RCU) {
487 nd->flags &= ~LOOKUP_RCU;
488 if (!(nd->flags & LOOKUP_ROOT))
490 spin_lock(&dentry->d_lock);
491 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
492 spin_unlock(&dentry->d_lock);
494 br_read_unlock(&vfsmount_lock);
497 BUG_ON(nd->inode != dentry->d_inode);
498 spin_unlock(&dentry->d_lock);
499 mntget(nd->path.mnt);
501 br_read_unlock(&vfsmount_lock);
504 if (likely(!(nd->flags & LOOKUP_JUMPED)))
507 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
510 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
513 /* Note: we do not d_invalidate() */
514 status = d_revalidate(dentry, nd->flags);
525 static __always_inline void set_root(struct nameidata *nd)
528 get_fs_root(current->fs, &nd->root);
531 static int link_path_walk(const char *, struct nameidata *);
533 static __always_inline void set_root_rcu(struct nameidata *nd)
536 struct fs_struct *fs = current->fs;
540 seq = read_seqcount_begin(&fs->seq);
542 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
543 } while (read_seqcount_retry(&fs->seq, seq));
547 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
559 nd->flags |= LOOKUP_JUMPED;
561 nd->inode = nd->path.dentry->d_inode;
563 ret = link_path_walk(link, nd);
567 return PTR_ERR(link);
570 static void path_put_conditional(struct path *path, struct nameidata *nd)
573 if (path->mnt != nd->path.mnt)
577 static inline void path_to_nameidata(const struct path *path,
578 struct nameidata *nd)
580 if (!(nd->flags & LOOKUP_RCU)) {
581 dput(nd->path.dentry);
582 if (nd->path.mnt != path->mnt)
583 mntput(nd->path.mnt);
585 nd->path.mnt = path->mnt;
586 nd->path.dentry = path->dentry;
590 * Helper to directly jump to a known parsed path from ->follow_link,
591 * caller must have taken a reference to path beforehand.
593 void nd_jump_link(struct nameidata *nd, struct path *path)
598 nd->inode = nd->path.dentry->d_inode;
599 nd->flags |= LOOKUP_JUMPED;
601 BUG_ON(nd->inode->i_op->follow_link);
604 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
606 struct inode *inode = link->dentry->d_inode;
607 if (inode->i_op->put_link)
608 inode->i_op->put_link(link->dentry, nd, cookie);
612 static __always_inline int
613 follow_link(struct path *link, struct nameidata *nd, void **p)
615 struct dentry *dentry = link->dentry;
619 BUG_ON(nd->flags & LOOKUP_RCU);
621 if (link->mnt == nd->path.mnt)
625 if (unlikely(current->total_link_count >= 40))
626 goto out_put_nd_path;
629 current->total_link_count++;
632 nd_set_link(nd, NULL);
634 error = security_inode_follow_link(link->dentry, nd);
636 goto out_put_nd_path;
638 nd->last_type = LAST_BIND;
639 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
642 goto out_put_nd_path;
647 error = __vfs_follow_link(nd, s);
649 put_link(nd, link, *p);
660 static int follow_up_rcu(struct path *path)
662 struct mount *mnt = real_mount(path->mnt);
663 struct mount *parent;
664 struct dentry *mountpoint;
666 parent = mnt->mnt_parent;
667 if (&parent->mnt == path->mnt)
669 mountpoint = mnt->mnt_mountpoint;
670 path->dentry = mountpoint;
671 path->mnt = &parent->mnt;
675 int follow_up(struct path *path)
677 struct mount *mnt = real_mount(path->mnt);
678 struct mount *parent;
679 struct dentry *mountpoint;
681 br_read_lock(&vfsmount_lock);
682 parent = mnt->mnt_parent;
683 if (&parent->mnt == path->mnt) {
684 br_read_unlock(&vfsmount_lock);
687 mntget(&parent->mnt);
688 mountpoint = dget(mnt->mnt_mountpoint);
689 br_read_unlock(&vfsmount_lock);
691 path->dentry = mountpoint;
693 path->mnt = &parent->mnt;
698 * Perform an automount
699 * - return -EISDIR to tell follow_managed() to stop and return the path we
702 static int follow_automount(struct path *path, unsigned flags,
705 struct vfsmount *mnt;
708 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
711 /* We don't want to mount if someone's just doing a stat -
712 * unless they're stat'ing a directory and appended a '/' to
715 * We do, however, want to mount if someone wants to open or
716 * create a file of any type under the mountpoint, wants to
717 * traverse through the mountpoint or wants to open the
718 * mounted directory. Also, autofs may mark negative dentries
719 * as being automount points. These will need the attentions
720 * of the daemon to instantiate them before they can be used.
722 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
723 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
724 path->dentry->d_inode)
727 current->total_link_count++;
728 if (current->total_link_count >= 40)
731 mnt = path->dentry->d_op->d_automount(path);
734 * The filesystem is allowed to return -EISDIR here to indicate
735 * it doesn't want to automount. For instance, autofs would do
736 * this so that its userspace daemon can mount on this dentry.
738 * However, we can only permit this if it's a terminal point in
739 * the path being looked up; if it wasn't then the remainder of
740 * the path is inaccessible and we should say so.
742 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
747 if (!mnt) /* mount collision */
751 /* lock_mount() may release path->mnt on error */
755 err = finish_automount(mnt, path);
759 /* Someone else made a mount here whilst we were busy */
764 path->dentry = dget(mnt->mnt_root);
773 * Handle a dentry that is managed in some way.
774 * - Flagged for transit management (autofs)
775 * - Flagged as mountpoint
776 * - Flagged as automount point
778 * This may only be called in refwalk mode.
780 * Serialization is taken care of in namespace.c
782 static int follow_managed(struct path *path, unsigned flags)
784 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
786 bool need_mntput = false;
789 /* Given that we're not holding a lock here, we retain the value in a
790 * local variable for each dentry as we look at it so that we don't see
791 * the components of that value change under us */
792 while (managed = ACCESS_ONCE(path->dentry->d_flags),
793 managed &= DCACHE_MANAGED_DENTRY,
794 unlikely(managed != 0)) {
795 /* Allow the filesystem to manage the transit without i_mutex
797 if (managed & DCACHE_MANAGE_TRANSIT) {
798 BUG_ON(!path->dentry->d_op);
799 BUG_ON(!path->dentry->d_op->d_manage);
800 ret = path->dentry->d_op->d_manage(path->dentry, false);
805 /* Transit to a mounted filesystem. */
806 if (managed & DCACHE_MOUNTED) {
807 struct vfsmount *mounted = lookup_mnt(path);
813 path->dentry = dget(mounted->mnt_root);
818 /* Something is mounted on this dentry in another
819 * namespace and/or whatever was mounted there in this
820 * namespace got unmounted before we managed to get the
824 /* Handle an automount point */
825 if (managed & DCACHE_NEED_AUTOMOUNT) {
826 ret = follow_automount(path, flags, &need_mntput);
832 /* We didn't change the current path point */
836 if (need_mntput && path->mnt == mnt)
840 return ret < 0 ? ret : need_mntput;
843 int follow_down_one(struct path *path)
845 struct vfsmount *mounted;
847 mounted = lookup_mnt(path);
852 path->dentry = dget(mounted->mnt_root);
858 static inline bool managed_dentry_might_block(struct dentry *dentry)
860 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
861 dentry->d_op->d_manage(dentry, true) < 0);
865 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
866 * we meet a managed dentry that would need blocking.
868 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
869 struct inode **inode)
872 struct mount *mounted;
874 * Don't forget we might have a non-mountpoint managed dentry
875 * that wants to block transit.
877 if (unlikely(managed_dentry_might_block(path->dentry)))
880 if (!d_mountpoint(path->dentry))
883 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
886 path->mnt = &mounted->mnt;
887 path->dentry = mounted->mnt.mnt_root;
888 nd->flags |= LOOKUP_JUMPED;
889 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
891 * Update the inode too. We don't need to re-check the
892 * dentry sequence number here after this d_inode read,
893 * because a mount-point is always pinned.
895 *inode = path->dentry->d_inode;
900 static void follow_mount_rcu(struct nameidata *nd)
902 while (d_mountpoint(nd->path.dentry)) {
903 struct mount *mounted;
904 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
907 nd->path.mnt = &mounted->mnt;
908 nd->path.dentry = mounted->mnt.mnt_root;
909 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
913 static int follow_dotdot_rcu(struct nameidata *nd)
918 if (nd->path.dentry == nd->root.dentry &&
919 nd->path.mnt == nd->root.mnt) {
922 if (nd->path.dentry != nd->path.mnt->mnt_root) {
923 struct dentry *old = nd->path.dentry;
924 struct dentry *parent = old->d_parent;
927 seq = read_seqcount_begin(&parent->d_seq);
928 if (read_seqcount_retry(&old->d_seq, nd->seq))
930 nd->path.dentry = parent;
934 if (!follow_up_rcu(&nd->path))
936 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
938 follow_mount_rcu(nd);
939 nd->inode = nd->path.dentry->d_inode;
943 nd->flags &= ~LOOKUP_RCU;
944 if (!(nd->flags & LOOKUP_ROOT))
947 br_read_unlock(&vfsmount_lock);
952 * Follow down to the covering mount currently visible to userspace. At each
953 * point, the filesystem owning that dentry may be queried as to whether the
954 * caller is permitted to proceed or not.
956 int follow_down(struct path *path)
961 while (managed = ACCESS_ONCE(path->dentry->d_flags),
962 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
963 /* Allow the filesystem to manage the transit without i_mutex
966 * We indicate to the filesystem if someone is trying to mount
967 * something here. This gives autofs the chance to deny anyone
968 * other than its daemon the right to mount on its
971 * The filesystem may sleep at this point.
973 if (managed & DCACHE_MANAGE_TRANSIT) {
974 BUG_ON(!path->dentry->d_op);
975 BUG_ON(!path->dentry->d_op->d_manage);
976 ret = path->dentry->d_op->d_manage(
977 path->dentry, false);
979 return ret == -EISDIR ? 0 : ret;
982 /* Transit to a mounted filesystem. */
983 if (managed & DCACHE_MOUNTED) {
984 struct vfsmount *mounted = lookup_mnt(path);
990 path->dentry = dget(mounted->mnt_root);
994 /* Don't handle automount points here */
1001 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1003 static void follow_mount(struct path *path)
1005 while (d_mountpoint(path->dentry)) {
1006 struct vfsmount *mounted = lookup_mnt(path);
1011 path->mnt = mounted;
1012 path->dentry = dget(mounted->mnt_root);
1016 static void follow_dotdot(struct nameidata *nd)
1021 struct dentry *old = nd->path.dentry;
1023 if (nd->path.dentry == nd->root.dentry &&
1024 nd->path.mnt == nd->root.mnt) {
1027 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1028 /* rare case of legitimate dget_parent()... */
1029 nd->path.dentry = dget_parent(nd->path.dentry);
1033 if (!follow_up(&nd->path))
1036 follow_mount(&nd->path);
1037 nd->inode = nd->path.dentry->d_inode;
1041 * This looks up the name in dcache, possibly revalidates the old dentry and
1042 * allocates a new one if not found or not valid. In the need_lookup argument
1043 * returns whether i_op->lookup is necessary.
1045 * dir->d_inode->i_mutex must be held
1047 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1048 unsigned int flags, bool *need_lookup)
1050 struct dentry *dentry;
1053 *need_lookup = false;
1054 dentry = d_lookup(dir, name);
1056 if (d_need_lookup(dentry)) {
1057 *need_lookup = true;
1058 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1059 error = d_revalidate(dentry, flags);
1060 if (unlikely(error <= 0)) {
1063 return ERR_PTR(error);
1064 } else if (!d_invalidate(dentry)) {
1073 dentry = d_alloc(dir, name);
1074 if (unlikely(!dentry))
1075 return ERR_PTR(-ENOMEM);
1077 *need_lookup = true;
1083 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1084 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1086 * dir->d_inode->i_mutex must be held
1088 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1093 /* Don't create child dentry for a dead directory. */
1094 if (unlikely(IS_DEADDIR(dir))) {
1096 return ERR_PTR(-ENOENT);
1099 old = dir->i_op->lookup(dir, dentry, flags);
1100 if (unlikely(old)) {
1107 static struct dentry *__lookup_hash(struct qstr *name,
1108 struct dentry *base, unsigned int flags)
1111 struct dentry *dentry;
1113 dentry = lookup_dcache(name, base, flags, &need_lookup);
1117 return lookup_real(base->d_inode, dentry, flags);
1121 * It's more convoluted than I'd like it to be, but... it's still fairly
1122 * small and for now I'd prefer to have fast path as straight as possible.
1123 * It _is_ time-critical.
1125 static int lookup_fast(struct nameidata *nd, struct qstr *name,
1126 struct path *path, struct inode **inode)
1128 struct vfsmount *mnt = nd->path.mnt;
1129 struct dentry *dentry, *parent = nd->path.dentry;
1135 * Rename seqlock is not required here because in the off chance
1136 * of a false negative due to a concurrent rename, we're going to
1137 * do the non-racy lookup, below.
1139 if (nd->flags & LOOKUP_RCU) {
1141 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
1146 * This sequence count validates that the inode matches
1147 * the dentry name information from lookup.
1149 *inode = dentry->d_inode;
1150 if (read_seqcount_retry(&dentry->d_seq, seq))
1154 * This sequence count validates that the parent had no
1155 * changes while we did the lookup of the dentry above.
1157 * The memory barrier in read_seqcount_begin of child is
1158 * enough, we can use __read_seqcount_retry here.
1160 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1164 if (unlikely(d_need_lookup(dentry)))
1166 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1167 status = d_revalidate(dentry, nd->flags);
1168 if (unlikely(status <= 0)) {
1169 if (status != -ECHILD)
1175 path->dentry = dentry;
1176 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1178 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1182 if (unlazy_walk(nd, dentry))
1185 dentry = __d_lookup(parent, name);
1188 if (unlikely(!dentry))
1191 if (unlikely(d_need_lookup(dentry))) {
1196 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1197 status = d_revalidate(dentry, nd->flags);
1198 if (unlikely(status <= 0)) {
1203 if (!d_invalidate(dentry)) {
1210 path->dentry = dentry;
1211 err = follow_managed(path, nd->flags);
1212 if (unlikely(err < 0)) {
1213 path_put_conditional(path, nd);
1217 nd->flags |= LOOKUP_JUMPED;
1218 *inode = path->dentry->d_inode;
1225 /* Fast lookup failed, do it the slow way */
1226 static int lookup_slow(struct nameidata *nd, struct qstr *name,
1229 struct dentry *dentry, *parent;
1232 parent = nd->path.dentry;
1233 BUG_ON(nd->inode != parent->d_inode);
1235 mutex_lock(&parent->d_inode->i_mutex);
1236 dentry = __lookup_hash(name, parent, nd->flags);
1237 mutex_unlock(&parent->d_inode->i_mutex);
1239 return PTR_ERR(dentry);
1240 path->mnt = nd->path.mnt;
1241 path->dentry = dentry;
1242 err = follow_managed(path, nd->flags);
1243 if (unlikely(err < 0)) {
1244 path_put_conditional(path, nd);
1248 nd->flags |= LOOKUP_JUMPED;
1252 static inline int may_lookup(struct nameidata *nd)
1254 if (nd->flags & LOOKUP_RCU) {
1255 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1258 if (unlazy_walk(nd, NULL))
1261 return inode_permission(nd->inode, MAY_EXEC);
1264 static inline int handle_dots(struct nameidata *nd, int type)
1266 if (type == LAST_DOTDOT) {
1267 if (nd->flags & LOOKUP_RCU) {
1268 if (follow_dotdot_rcu(nd))
1276 static void terminate_walk(struct nameidata *nd)
1278 if (!(nd->flags & LOOKUP_RCU)) {
1279 path_put(&nd->path);
1281 nd->flags &= ~LOOKUP_RCU;
1282 if (!(nd->flags & LOOKUP_ROOT))
1283 nd->root.mnt = NULL;
1285 br_read_unlock(&vfsmount_lock);
1290 * Do we need to follow links? We _really_ want to be able
1291 * to do this check without having to look at inode->i_op,
1292 * so we keep a cache of "no, this doesn't need follow_link"
1293 * for the common case.
1295 static inline int should_follow_link(struct inode *inode, int follow)
1297 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1298 if (likely(inode->i_op->follow_link))
1301 /* This gets set once for the inode lifetime */
1302 spin_lock(&inode->i_lock);
1303 inode->i_opflags |= IOP_NOFOLLOW;
1304 spin_unlock(&inode->i_lock);
1309 static inline int walk_component(struct nameidata *nd, struct path *path,
1310 struct qstr *name, int type, int follow)
1312 struct inode *inode;
1315 * "." and ".." are special - ".." especially so because it has
1316 * to be able to know about the current root directory and
1317 * parent relationships.
1319 if (unlikely(type != LAST_NORM))
1320 return handle_dots(nd, type);
1321 err = lookup_fast(nd, name, path, &inode);
1322 if (unlikely(err)) {
1326 err = lookup_slow(nd, name, path);
1330 inode = path->dentry->d_inode;
1336 if (should_follow_link(inode, follow)) {
1337 if (nd->flags & LOOKUP_RCU) {
1338 if (unlikely(unlazy_walk(nd, path->dentry))) {
1343 BUG_ON(inode != path->dentry->d_inode);
1346 path_to_nameidata(path, nd);
1351 path_to_nameidata(path, nd);
1358 * This limits recursive symlink follows to 8, while
1359 * limiting consecutive symlinks to 40.
1361 * Without that kind of total limit, nasty chains of consecutive
1362 * symlinks can cause almost arbitrarily long lookups.
1364 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1368 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1369 path_put_conditional(path, nd);
1370 path_put(&nd->path);
1373 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1376 current->link_count++;
1379 struct path link = *path;
1382 res = follow_link(&link, nd, &cookie);
1385 res = walk_component(nd, path, &nd->last,
1386 nd->last_type, LOOKUP_FOLLOW);
1387 put_link(nd, &link, cookie);
1390 current->link_count--;
1396 * We really don't want to look at inode->i_op->lookup
1397 * when we don't have to. So we keep a cache bit in
1398 * the inode ->i_opflags field that says "yes, we can
1399 * do lookup on this inode".
1401 static inline int can_lookup(struct inode *inode)
1403 if (likely(inode->i_opflags & IOP_LOOKUP))
1405 if (likely(!inode->i_op->lookup))
1408 /* We do this once for the lifetime of the inode */
1409 spin_lock(&inode->i_lock);
1410 inode->i_opflags |= IOP_LOOKUP;
1411 spin_unlock(&inode->i_lock);
1416 * We can do the critical dentry name comparison and hashing
1417 * operations one word at a time, but we are limited to:
1419 * - Architectures with fast unaligned word accesses. We could
1420 * do a "get_unaligned()" if this helps and is sufficiently
1423 * - Little-endian machines (so that we can generate the mask
1424 * of low bytes efficiently). Again, we *could* do a byte
1425 * swapping load on big-endian architectures if that is not
1426 * expensive enough to make the optimization worthless.
1428 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1429 * do not trap on the (extremely unlikely) case of a page
1430 * crossing operation.
1432 * - Furthermore, we need an efficient 64-bit compile for the
1433 * 64-bit case in order to generate the "number of bytes in
1434 * the final mask". Again, that could be replaced with a
1435 * efficient population count instruction or similar.
1437 #ifdef CONFIG_DCACHE_WORD_ACCESS
1439 #include <asm/word-at-a-time.h>
1443 static inline unsigned int fold_hash(unsigned long hash)
1445 hash += hash >> (8*sizeof(int));
1449 #else /* 32-bit case */
1451 #define fold_hash(x) (x)
1455 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1457 unsigned long a, mask;
1458 unsigned long hash = 0;
1461 a = load_unaligned_zeropad(name);
1462 if (len < sizeof(unsigned long))
1466 name += sizeof(unsigned long);
1467 len -= sizeof(unsigned long);
1471 mask = ~(~0ul << len*8);
1474 return fold_hash(hash);
1476 EXPORT_SYMBOL(full_name_hash);
1479 * Calculate the length and hash of the path component, and
1480 * return the length of the component;
1482 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1484 unsigned long a, b, adata, bdata, mask, hash, len;
1485 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1488 len = -sizeof(unsigned long);
1490 hash = (hash + a) * 9;
1491 len += sizeof(unsigned long);
1492 a = load_unaligned_zeropad(name+len);
1493 b = a ^ REPEAT_BYTE('/');
1494 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1496 adata = prep_zero_mask(a, adata, &constants);
1497 bdata = prep_zero_mask(b, bdata, &constants);
1499 mask = create_zero_mask(adata | bdata);
1501 hash += a & zero_bytemask(mask);
1502 *hashp = fold_hash(hash);
1504 return len + find_zero(mask);
1509 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1511 unsigned long hash = init_name_hash();
1513 hash = partial_name_hash(*name++, hash);
1514 return end_name_hash(hash);
1516 EXPORT_SYMBOL(full_name_hash);
1519 * We know there's a real path component here of at least
1522 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1524 unsigned long hash = init_name_hash();
1525 unsigned long len = 0, c;
1527 c = (unsigned char)*name;
1530 hash = partial_name_hash(c, hash);
1531 c = (unsigned char)name[len];
1532 } while (c && c != '/');
1533 *hashp = end_name_hash(hash);
1541 * This is the basic name resolution function, turning a pathname into
1542 * the final dentry. We expect 'base' to be positive and a directory.
1544 * Returns 0 and nd will have valid dentry and mnt on success.
1545 * Returns error and drops reference to input namei data on failure.
1547 static int link_path_walk(const char *name, struct nameidata *nd)
1557 /* At this point we know we have a real path component. */
1563 err = may_lookup(nd);
1567 len = hash_name(name, &this.hash);
1572 if (name[0] == '.') switch (len) {
1574 if (name[1] == '.') {
1576 nd->flags |= LOOKUP_JUMPED;
1582 if (likely(type == LAST_NORM)) {
1583 struct dentry *parent = nd->path.dentry;
1584 nd->flags &= ~LOOKUP_JUMPED;
1585 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1586 err = parent->d_op->d_hash(parent, nd->inode,
1594 goto last_component;
1596 * If it wasn't NUL, we know it was '/'. Skip that
1597 * slash, and continue until no more slashes.
1601 } while (unlikely(name[len] == '/'));
1603 goto last_component;
1606 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1611 err = nested_symlink(&next, nd);
1615 if (can_lookup(nd->inode))
1619 /* here ends the main loop */
1623 nd->last_type = type;
1630 static int path_init(int dfd, const char *name, unsigned int flags,
1631 struct nameidata *nd, struct file **fp)
1637 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1638 nd->flags = flags | LOOKUP_JUMPED;
1640 if (flags & LOOKUP_ROOT) {
1641 struct inode *inode = nd->root.dentry->d_inode;
1643 if (!inode->i_op->lookup)
1645 retval = inode_permission(inode, MAY_EXEC);
1649 nd->path = nd->root;
1651 if (flags & LOOKUP_RCU) {
1652 br_read_lock(&vfsmount_lock);
1654 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1656 path_get(&nd->path);
1661 nd->root.mnt = NULL;
1664 if (flags & LOOKUP_RCU) {
1665 br_read_lock(&vfsmount_lock);
1670 path_get(&nd->root);
1672 nd->path = nd->root;
1673 } else if (dfd == AT_FDCWD) {
1674 if (flags & LOOKUP_RCU) {
1675 struct fs_struct *fs = current->fs;
1678 br_read_lock(&vfsmount_lock);
1682 seq = read_seqcount_begin(&fs->seq);
1684 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1685 } while (read_seqcount_retry(&fs->seq, seq));
1687 get_fs_pwd(current->fs, &nd->path);
1690 struct dentry *dentry;
1692 file = fget_raw_light(dfd, &fput_needed);
1697 dentry = file->f_path.dentry;
1701 if (!S_ISDIR(dentry->d_inode->i_mode))
1704 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1709 nd->path = file->f_path;
1710 if (flags & LOOKUP_RCU) {
1713 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1714 br_read_lock(&vfsmount_lock);
1717 path_get(&file->f_path);
1718 fput_light(file, fput_needed);
1722 nd->inode = nd->path.dentry->d_inode;
1726 fput_light(file, fput_needed);
1731 static inline int lookup_last(struct nameidata *nd, struct path *path)
1733 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1734 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1736 nd->flags &= ~LOOKUP_PARENT;
1737 return walk_component(nd, path, &nd->last, nd->last_type,
1738 nd->flags & LOOKUP_FOLLOW);
1741 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1742 static int path_lookupat(int dfd, const char *name,
1743 unsigned int flags, struct nameidata *nd)
1745 struct file *base = NULL;
1750 * Path walking is largely split up into 2 different synchronisation
1751 * schemes, rcu-walk and ref-walk (explained in
1752 * Documentation/filesystems/path-lookup.txt). These share much of the
1753 * path walk code, but some things particularly setup, cleanup, and
1754 * following mounts are sufficiently divergent that functions are
1755 * duplicated. Typically there is a function foo(), and its RCU
1756 * analogue, foo_rcu().
1758 * -ECHILD is the error number of choice (just to avoid clashes) that
1759 * is returned if some aspect of an rcu-walk fails. Such an error must
1760 * be handled by restarting a traditional ref-walk (which will always
1761 * be able to complete).
1763 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1768 current->total_link_count = 0;
1769 err = link_path_walk(name, nd);
1771 if (!err && !(flags & LOOKUP_PARENT)) {
1772 err = lookup_last(nd, &path);
1775 struct path link = path;
1776 nd->flags |= LOOKUP_PARENT;
1777 err = follow_link(&link, nd, &cookie);
1780 err = lookup_last(nd, &path);
1781 put_link(nd, &link, cookie);
1786 err = complete_walk(nd);
1788 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1789 if (!nd->inode->i_op->lookup) {
1790 path_put(&nd->path);
1798 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1799 path_put(&nd->root);
1800 nd->root.mnt = NULL;
1805 static int do_path_lookup(int dfd, const char *name,
1806 unsigned int flags, struct nameidata *nd)
1808 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1809 if (unlikely(retval == -ECHILD))
1810 retval = path_lookupat(dfd, name, flags, nd);
1811 if (unlikely(retval == -ESTALE))
1812 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1814 if (likely(!retval)) {
1815 if (unlikely(!audit_dummy_context())) {
1816 if (nd->path.dentry && nd->inode)
1817 audit_inode(name, nd->path.dentry);
1823 /* does lookup, returns the object with parent locked */
1824 struct dentry *kern_path_locked(const char *name, struct path *path)
1826 struct nameidata nd;
1828 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
1830 return ERR_PTR(err);
1831 if (nd.last_type != LAST_NORM) {
1833 return ERR_PTR(-EINVAL);
1835 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1836 d = lookup_one_len(nd.last.name, nd.path.dentry, nd.last.len);
1838 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1846 int kern_path(const char *name, unsigned int flags, struct path *path)
1848 struct nameidata nd;
1849 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1856 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1857 * @dentry: pointer to dentry of the base directory
1858 * @mnt: pointer to vfs mount of the base directory
1859 * @name: pointer to file name
1860 * @flags: lookup flags
1861 * @path: pointer to struct path to fill
1863 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1864 const char *name, unsigned int flags,
1867 struct nameidata nd;
1869 nd.root.dentry = dentry;
1871 BUG_ON(flags & LOOKUP_PARENT);
1872 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1873 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1880 * Restricted form of lookup. Doesn't follow links, single-component only,
1881 * needs parent already locked. Doesn't follow mounts.
1884 static struct dentry *lookup_hash(struct nameidata *nd)
1886 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
1890 * lookup_one_len - filesystem helper to lookup single pathname component
1891 * @name: pathname component to lookup
1892 * @base: base directory to lookup from
1893 * @len: maximum length @len should be interpreted to
1895 * Note that this routine is purely a helper for filesystem usage and should
1896 * not be called by generic code. Also note that by using this function the
1897 * nameidata argument is passed to the filesystem methods and a filesystem
1898 * using this helper needs to be prepared for that.
1900 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1906 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1910 this.hash = full_name_hash(name, len);
1912 return ERR_PTR(-EACCES);
1915 c = *(const unsigned char *)name++;
1916 if (c == '/' || c == '\0')
1917 return ERR_PTR(-EACCES);
1920 * See if the low-level filesystem might want
1921 * to use its own hash..
1923 if (base->d_flags & DCACHE_OP_HASH) {
1924 int err = base->d_op->d_hash(base, base->d_inode, &this);
1926 return ERR_PTR(err);
1929 err = inode_permission(base->d_inode, MAY_EXEC);
1931 return ERR_PTR(err);
1933 return __lookup_hash(&this, base, 0);
1936 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1937 struct path *path, int *empty)
1939 struct nameidata nd;
1940 char *tmp = getname_flags(name, flags, empty);
1941 int err = PTR_ERR(tmp);
1944 BUG_ON(flags & LOOKUP_PARENT);
1946 err = do_path_lookup(dfd, tmp, flags, &nd);
1954 int user_path_at(int dfd, const char __user *name, unsigned flags,
1957 return user_path_at_empty(dfd, name, flags, path, NULL);
1960 static int user_path_parent(int dfd, const char __user *path,
1961 struct nameidata *nd, char **name)
1963 char *s = getname(path);
1969 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1979 * It's inline, so penalty for filesystems that don't use sticky bit is
1982 static inline int check_sticky(struct inode *dir, struct inode *inode)
1984 kuid_t fsuid = current_fsuid();
1986 if (!(dir->i_mode & S_ISVTX))
1988 if (uid_eq(inode->i_uid, fsuid))
1990 if (uid_eq(dir->i_uid, fsuid))
1992 return !inode_capable(inode, CAP_FOWNER);
1996 * Check whether we can remove a link victim from directory dir, check
1997 * whether the type of victim is right.
1998 * 1. We can't do it if dir is read-only (done in permission())
1999 * 2. We should have write and exec permissions on dir
2000 * 3. We can't remove anything from append-only dir
2001 * 4. We can't do anything with immutable dir (done in permission())
2002 * 5. If the sticky bit on dir is set we should either
2003 * a. be owner of dir, or
2004 * b. be owner of victim, or
2005 * c. have CAP_FOWNER capability
2006 * 6. If the victim is append-only or immutable we can't do antyhing with
2007 * links pointing to it.
2008 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2009 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2010 * 9. We can't remove a root or mountpoint.
2011 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2012 * nfs_async_unlink().
2014 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2018 if (!victim->d_inode)
2021 BUG_ON(victim->d_parent->d_inode != dir);
2022 audit_inode_child(victim, dir);
2024 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2029 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2030 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2033 if (!S_ISDIR(victim->d_inode->i_mode))
2035 if (IS_ROOT(victim))
2037 } else if (S_ISDIR(victim->d_inode->i_mode))
2039 if (IS_DEADDIR(dir))
2041 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2046 /* Check whether we can create an object with dentry child in directory
2048 * 1. We can't do it if child already exists (open has special treatment for
2049 * this case, but since we are inlined it's OK)
2050 * 2. We can't do it if dir is read-only (done in permission())
2051 * 3. We should have write and exec permissions on dir
2052 * 4. We can't do it if dir is immutable (done in permission())
2054 static inline int may_create(struct inode *dir, struct dentry *child)
2058 if (IS_DEADDIR(dir))
2060 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2064 * p1 and p2 should be directories on the same fs.
2066 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2071 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2075 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2077 p = d_ancestor(p2, p1);
2079 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2080 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2084 p = d_ancestor(p1, p2);
2086 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2087 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2091 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2092 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2096 void unlock_rename(struct dentry *p1, struct dentry *p2)
2098 mutex_unlock(&p1->d_inode->i_mutex);
2100 mutex_unlock(&p2->d_inode->i_mutex);
2101 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2105 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2108 int error = may_create(dir, dentry);
2112 if (!dir->i_op->create)
2113 return -EACCES; /* shouldn't it be ENOSYS? */
2116 error = security_inode_create(dir, dentry, mode);
2119 error = dir->i_op->create(dir, dentry, mode, want_excl);
2121 fsnotify_create(dir, dentry);
2125 static int may_open(struct path *path, int acc_mode, int flag)
2127 struct dentry *dentry = path->dentry;
2128 struct inode *inode = dentry->d_inode;
2138 switch (inode->i_mode & S_IFMT) {
2142 if (acc_mode & MAY_WRITE)
2147 if (path->mnt->mnt_flags & MNT_NODEV)
2156 error = inode_permission(inode, acc_mode);
2161 * An append-only file must be opened in append mode for writing.
2163 if (IS_APPEND(inode)) {
2164 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2170 /* O_NOATIME can only be set by the owner or superuser */
2171 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2177 static int handle_truncate(struct file *filp)
2179 struct path *path = &filp->f_path;
2180 struct inode *inode = path->dentry->d_inode;
2181 int error = get_write_access(inode);
2185 * Refuse to truncate files with mandatory locks held on them.
2187 error = locks_verify_locked(inode);
2189 error = security_path_truncate(path);
2191 error = do_truncate(path->dentry, 0,
2192 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2195 put_write_access(inode);
2199 static inline int open_to_namei_flags(int flag)
2201 if ((flag & O_ACCMODE) == 3)
2206 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2208 int error = security_path_mknod(dir, dentry, mode, 0);
2212 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2216 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2220 * Attempt to atomically look up, create and open a file from a negative
2223 * Returns 0 if successful. The file will have been created and attached to
2224 * @file by the filesystem calling finish_open().
2226 * Returns 1 if the file was looked up only or didn't need creating. The
2227 * caller will need to perform the open themselves. @path will have been
2228 * updated to point to the new dentry. This may be negative.
2230 * Returns an error code otherwise.
2232 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2233 struct path *path, struct file *file,
2234 const struct open_flags *op,
2235 bool *want_write, bool need_lookup,
2238 struct inode *dir = nd->path.dentry->d_inode;
2239 unsigned open_flag = open_to_namei_flags(op->open_flag);
2243 int create_error = 0;
2244 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2246 BUG_ON(dentry->d_inode);
2248 /* Don't create child dentry for a dead directory. */
2249 if (unlikely(IS_DEADDIR(dir))) {
2254 mode = op->mode & S_IALLUGO;
2255 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2256 mode &= ~current_umask();
2258 if (open_flag & O_EXCL) {
2259 open_flag &= ~O_TRUNC;
2260 *opened |= FILE_CREATED;
2264 * Checking write permission is tricky, bacuse we don't know if we are
2265 * going to actually need it: O_CREAT opens should work as long as the
2266 * file exists. But checking existence breaks atomicity. The trick is
2267 * to check access and if not granted clear O_CREAT from the flags.
2269 * Another problem is returing the "right" error value (e.g. for an
2270 * O_EXCL open we want to return EEXIST not EROFS).
2272 if ((open_flag & (O_CREAT | O_TRUNC)) ||
2273 (open_flag & O_ACCMODE) != O_RDONLY) {
2274 error = mnt_want_write(nd->path.mnt);
2277 } else if (!(open_flag & O_CREAT)) {
2279 * No O_CREATE -> atomicity not a requirement -> fall
2280 * back to lookup + open
2283 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2284 /* Fall back and fail with the right error */
2285 create_error = error;
2288 /* No side effects, safe to clear O_CREAT */
2289 create_error = error;
2290 open_flag &= ~O_CREAT;
2294 if (open_flag & O_CREAT) {
2295 error = may_o_create(&nd->path, dentry, op->mode);
2297 create_error = error;
2298 if (open_flag & O_EXCL)
2300 open_flag &= ~O_CREAT;
2304 if (nd->flags & LOOKUP_DIRECTORY)
2305 open_flag |= O_DIRECTORY;
2307 file->f_path.dentry = DENTRY_NOT_SET;
2308 file->f_path.mnt = nd->path.mnt;
2309 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2312 if (create_error && error == -ENOENT)
2313 error = create_error;
2317 acc_mode = op->acc_mode;
2318 if (*opened & FILE_CREATED) {
2319 fsnotify_create(dir, dentry);
2320 acc_mode = MAY_OPEN;
2323 if (error) { /* returned 1, that is */
2324 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2328 if (file->f_path.dentry) {
2330 dentry = file->f_path.dentry;
2336 * We didn't have the inode before the open, so check open permission
2339 error = may_open(&file->f_path, acc_mode, open_flag);
2349 dentry = lookup_real(dir, dentry, nd->flags);
2351 return PTR_ERR(dentry);
2354 int open_flag = op->open_flag;
2356 error = create_error;
2357 if ((open_flag & O_EXCL)) {
2358 if (!dentry->d_inode)
2360 } else if (!dentry->d_inode) {
2362 } else if ((open_flag & O_TRUNC) &&
2363 S_ISREG(dentry->d_inode->i_mode)) {
2366 /* will fail later, go on to get the right error */
2370 path->dentry = dentry;
2371 path->mnt = nd->path.mnt;
2376 * Look up and maybe create and open the last component.
2378 * Must be called with i_mutex held on parent.
2380 * Returns 0 if the file was successfully atomically created (if necessary) and
2381 * opened. In this case the file will be returned attached to @file.
2383 * Returns 1 if the file was not completely opened at this time, though lookups
2384 * and creations will have been performed and the dentry returned in @path will
2385 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2386 * specified then a negative dentry may be returned.
2388 * An error code is returned otherwise.
2390 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2391 * cleared otherwise prior to returning.
2393 static int lookup_open(struct nameidata *nd, struct path *path,
2395 const struct open_flags *op,
2396 bool *want_write, int *opened)
2398 struct dentry *dir = nd->path.dentry;
2399 struct inode *dir_inode = dir->d_inode;
2400 struct dentry *dentry;
2404 *opened &= ~FILE_CREATED;
2405 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2407 return PTR_ERR(dentry);
2409 /* Cached positive dentry: will open in f_op->open */
2410 if (!need_lookup && dentry->d_inode)
2413 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2414 return atomic_open(nd, dentry, path, file, op, want_write,
2415 need_lookup, opened);
2419 BUG_ON(dentry->d_inode);
2421 dentry = lookup_real(dir_inode, dentry, nd->flags);
2423 return PTR_ERR(dentry);
2426 /* Negative dentry, just create the file */
2427 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2428 umode_t mode = op->mode;
2429 if (!IS_POSIXACL(dir->d_inode))
2430 mode &= ~current_umask();
2432 * This write is needed to ensure that a
2433 * rw->ro transition does not occur between
2434 * the time when the file is created and when
2435 * a permanent write count is taken through
2436 * the 'struct file' in finish_open().
2438 error = mnt_want_write(nd->path.mnt);
2442 *opened |= FILE_CREATED;
2443 error = security_path_mknod(&nd->path, dentry, mode, 0);
2446 error = vfs_create(dir->d_inode, dentry, mode,
2447 nd->flags & LOOKUP_EXCL);
2452 path->dentry = dentry;
2453 path->mnt = nd->path.mnt;
2462 * Handle the last step of open()
2464 static int do_last(struct nameidata *nd, struct path *path,
2465 struct file *file, const struct open_flags *op,
2466 int *opened, const char *pathname)
2468 struct dentry *dir = nd->path.dentry;
2469 int open_flag = op->open_flag;
2470 bool will_truncate = (open_flag & O_TRUNC) != 0;
2471 bool want_write = false;
2472 int acc_mode = op->acc_mode;
2473 struct inode *inode;
2474 bool symlink_ok = false;
2475 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2476 bool retried = false;
2479 nd->flags &= ~LOOKUP_PARENT;
2480 nd->flags |= op->intent;
2482 switch (nd->last_type) {
2485 error = handle_dots(nd, nd->last_type);
2490 error = complete_walk(nd);
2493 audit_inode(pathname, nd->path.dentry);
2494 if (open_flag & O_CREAT) {
2500 error = complete_walk(nd);
2503 audit_inode(pathname, dir);
2507 if (!(open_flag & O_CREAT)) {
2508 if (nd->last.name[nd->last.len])
2509 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2510 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2512 /* we _can_ be in RCU mode here */
2513 error = lookup_fast(nd, &nd->last, path, &inode);
2520 BUG_ON(nd->inode != dir->d_inode);
2522 /* create side of things */
2524 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2525 * has been cleared when we got to the last component we are
2528 error = complete_walk(nd);
2532 audit_inode(pathname, dir);
2534 /* trailing slashes? */
2535 if (nd->last.name[nd->last.len])
2540 mutex_lock(&dir->d_inode->i_mutex);
2541 error = lookup_open(nd, path, file, op, &want_write, opened);
2542 mutex_unlock(&dir->d_inode->i_mutex);
2548 if ((*opened & FILE_CREATED) ||
2549 !S_ISREG(file->f_path.dentry->d_inode->i_mode))
2550 will_truncate = false;
2552 audit_inode(pathname, file->f_path.dentry);
2556 if (*opened & FILE_CREATED) {
2557 /* Don't check for write permission, don't truncate */
2558 open_flag &= ~O_TRUNC;
2559 will_truncate = false;
2560 acc_mode = MAY_OPEN;
2561 path_to_nameidata(path, nd);
2562 goto finish_open_created;
2566 * It already exists.
2568 audit_inode(pathname, path->dentry);
2571 * If atomic_open() acquired write access it is dropped now due to
2572 * possible mount and symlink following (this might be optimized away if
2576 mnt_drop_write(nd->path.mnt);
2581 if (open_flag & O_EXCL)
2584 error = follow_managed(path, nd->flags);
2589 nd->flags |= LOOKUP_JUMPED;
2591 BUG_ON(nd->flags & LOOKUP_RCU);
2592 inode = path->dentry->d_inode;
2594 /* we _can_ be in RCU mode here */
2597 path_to_nameidata(path, nd);
2601 if (should_follow_link(inode, !symlink_ok)) {
2602 if (nd->flags & LOOKUP_RCU) {
2603 if (unlikely(unlazy_walk(nd, path->dentry))) {
2608 BUG_ON(inode != path->dentry->d_inode);
2612 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2613 path_to_nameidata(path, nd);
2615 save_parent.dentry = nd->path.dentry;
2616 save_parent.mnt = mntget(path->mnt);
2617 nd->path.dentry = path->dentry;
2621 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2622 error = complete_walk(nd);
2624 path_put(&save_parent);
2628 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
2631 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
2633 audit_inode(pathname, nd->path.dentry);
2635 if (!S_ISREG(nd->inode->i_mode))
2636 will_truncate = false;
2638 if (will_truncate) {
2639 error = mnt_want_write(nd->path.mnt);
2644 finish_open_created:
2645 error = may_open(&nd->path, acc_mode, open_flag);
2648 file->f_path.mnt = nd->path.mnt;
2649 error = finish_open(file, nd->path.dentry, NULL, opened);
2651 if (error == -EOPENSTALE)
2656 error = open_check_o_direct(file);
2659 error = ima_file_check(file, op->acc_mode);
2663 if (will_truncate) {
2664 error = handle_truncate(file);
2670 mnt_drop_write(nd->path.mnt);
2671 path_put(&save_parent);
2676 path_put_conditional(path, nd);
2683 /* If no saved parent or already retried then can't retry */
2684 if (!save_parent.dentry || retried)
2687 BUG_ON(save_parent.dentry != dir);
2688 path_put(&nd->path);
2689 nd->path = save_parent;
2690 nd->inode = dir->d_inode;
2691 save_parent.mnt = NULL;
2692 save_parent.dentry = NULL;
2694 mnt_drop_write(nd->path.mnt);
2701 static struct file *path_openat(int dfd, const char *pathname,
2702 struct nameidata *nd, const struct open_flags *op, int flags)
2704 struct file *base = NULL;
2710 file = get_empty_filp();
2712 return ERR_PTR(-ENFILE);
2714 file->f_flags = op->open_flag;
2716 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2717 if (unlikely(error))
2720 current->total_link_count = 0;
2721 error = link_path_walk(pathname, nd);
2722 if (unlikely(error))
2725 error = do_last(nd, &path, file, op, &opened, pathname);
2726 while (unlikely(error > 0)) { /* trailing symlink */
2727 struct path link = path;
2729 if (!(nd->flags & LOOKUP_FOLLOW)) {
2730 path_put_conditional(&path, nd);
2731 path_put(&nd->path);
2735 nd->flags |= LOOKUP_PARENT;
2736 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2737 error = follow_link(&link, nd, &cookie);
2738 if (unlikely(error))
2740 error = do_last(nd, &path, file, op, &opened, pathname);
2741 put_link(nd, &link, cookie);
2744 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2745 path_put(&nd->root);
2748 if (!(opened & FILE_OPENED)) {
2752 if (unlikely(error)) {
2753 if (error == -EOPENSTALE) {
2754 if (flags & LOOKUP_RCU)
2759 file = ERR_PTR(error);
2764 struct file *do_filp_open(int dfd, const char *pathname,
2765 const struct open_flags *op, int flags)
2767 struct nameidata nd;
2770 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2771 if (unlikely(filp == ERR_PTR(-ECHILD)))
2772 filp = path_openat(dfd, pathname, &nd, op, flags);
2773 if (unlikely(filp == ERR_PTR(-ESTALE)))
2774 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2778 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2779 const char *name, const struct open_flags *op, int flags)
2781 struct nameidata nd;
2785 nd.root.dentry = dentry;
2787 flags |= LOOKUP_ROOT;
2789 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2790 return ERR_PTR(-ELOOP);
2792 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2793 if (unlikely(file == ERR_PTR(-ECHILD)))
2794 file = path_openat(-1, name, &nd, op, flags);
2795 if (unlikely(file == ERR_PTR(-ESTALE)))
2796 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2800 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2802 struct dentry *dentry = ERR_PTR(-EEXIST);
2803 struct nameidata nd;
2804 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2806 return ERR_PTR(error);
2809 * Yucky last component or no last component at all?
2810 * (foo/., foo/.., /////)
2812 if (nd.last_type != LAST_NORM)
2814 nd.flags &= ~LOOKUP_PARENT;
2815 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2818 * Do the final lookup.
2820 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2821 dentry = lookup_hash(&nd);
2825 if (dentry->d_inode)
2828 * Special case - lookup gave negative, but... we had foo/bar/
2829 * From the vfs_mknod() POV we just have a negative dentry -
2830 * all is fine. Let's be bastards - you had / on the end, you've
2831 * been asking for (non-existent) directory. -ENOENT for you.
2833 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2835 dentry = ERR_PTR(-ENOENT);
2842 dentry = ERR_PTR(-EEXIST);
2844 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2849 EXPORT_SYMBOL(kern_path_create);
2851 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2853 char *tmp = getname(pathname);
2856 return ERR_CAST(tmp);
2857 res = kern_path_create(dfd, tmp, path, is_dir);
2861 EXPORT_SYMBOL(user_path_create);
2863 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2865 int error = may_create(dir, dentry);
2870 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2873 if (!dir->i_op->mknod)
2876 error = devcgroup_inode_mknod(mode, dev);
2880 error = security_inode_mknod(dir, dentry, mode, dev);
2884 error = dir->i_op->mknod(dir, dentry, mode, dev);
2886 fsnotify_create(dir, dentry);
2890 static int may_mknod(umode_t mode)
2892 switch (mode & S_IFMT) {
2898 case 0: /* zero mode translates to S_IFREG */
2907 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2910 struct dentry *dentry;
2917 dentry = user_path_create(dfd, filename, &path, 0);
2919 return PTR_ERR(dentry);
2921 if (!IS_POSIXACL(path.dentry->d_inode))
2922 mode &= ~current_umask();
2923 error = may_mknod(mode);
2926 error = mnt_want_write(path.mnt);
2929 error = security_path_mknod(&path, dentry, mode, dev);
2931 goto out_drop_write;
2932 switch (mode & S_IFMT) {
2933 case 0: case S_IFREG:
2934 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
2936 case S_IFCHR: case S_IFBLK:
2937 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2938 new_decode_dev(dev));
2940 case S_IFIFO: case S_IFSOCK:
2941 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2945 mnt_drop_write(path.mnt);
2948 mutex_unlock(&path.dentry->d_inode->i_mutex);
2954 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
2956 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2959 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2961 int error = may_create(dir, dentry);
2962 unsigned max_links = dir->i_sb->s_max_links;
2967 if (!dir->i_op->mkdir)
2970 mode &= (S_IRWXUGO|S_ISVTX);
2971 error = security_inode_mkdir(dir, dentry, mode);
2975 if (max_links && dir->i_nlink >= max_links)
2978 error = dir->i_op->mkdir(dir, dentry, mode);
2980 fsnotify_mkdir(dir, dentry);
2984 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
2986 struct dentry *dentry;
2990 dentry = user_path_create(dfd, pathname, &path, 1);
2992 return PTR_ERR(dentry);
2994 if (!IS_POSIXACL(path.dentry->d_inode))
2995 mode &= ~current_umask();
2996 error = mnt_want_write(path.mnt);
2999 error = security_path_mkdir(&path, dentry, mode);
3001 goto out_drop_write;
3002 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3004 mnt_drop_write(path.mnt);
3007 mutex_unlock(&path.dentry->d_inode->i_mutex);
3012 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3014 return sys_mkdirat(AT_FDCWD, pathname, mode);
3018 * The dentry_unhash() helper will try to drop the dentry early: we
3019 * should have a usage count of 1 if we're the only user of this
3020 * dentry, and if that is true (possibly after pruning the dcache),
3021 * then we drop the dentry now.
3023 * A low-level filesystem can, if it choses, legally
3026 * if (!d_unhashed(dentry))
3029 * if it cannot handle the case of removing a directory
3030 * that is still in use by something else..
3032 void dentry_unhash(struct dentry *dentry)
3034 shrink_dcache_parent(dentry);
3035 spin_lock(&dentry->d_lock);
3036 if (dentry->d_count == 1)
3038 spin_unlock(&dentry->d_lock);
3041 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3043 int error = may_delete(dir, dentry, 1);
3048 if (!dir->i_op->rmdir)
3052 mutex_lock(&dentry->d_inode->i_mutex);
3055 if (d_mountpoint(dentry))
3058 error = security_inode_rmdir(dir, dentry);
3062 shrink_dcache_parent(dentry);
3063 error = dir->i_op->rmdir(dir, dentry);
3067 dentry->d_inode->i_flags |= S_DEAD;
3071 mutex_unlock(&dentry->d_inode->i_mutex);
3078 static long do_rmdir(int dfd, const char __user *pathname)
3082 struct dentry *dentry;
3083 struct nameidata nd;
3085 error = user_path_parent(dfd, pathname, &nd, &name);
3089 switch(nd.last_type) {
3101 nd.flags &= ~LOOKUP_PARENT;
3103 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3104 dentry = lookup_hash(&nd);
3105 error = PTR_ERR(dentry);
3108 if (!dentry->d_inode) {
3112 error = mnt_want_write(nd.path.mnt);
3115 error = security_path_rmdir(&nd.path, dentry);
3118 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3120 mnt_drop_write(nd.path.mnt);
3124 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3131 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3133 return do_rmdir(AT_FDCWD, pathname);
3136 int vfs_unlink(struct inode *dir, struct dentry *dentry)
3138 int error = may_delete(dir, dentry, 0);
3143 if (!dir->i_op->unlink)
3146 mutex_lock(&dentry->d_inode->i_mutex);
3147 if (d_mountpoint(dentry))
3150 error = security_inode_unlink(dir, dentry);
3152 error = dir->i_op->unlink(dir, dentry);
3157 mutex_unlock(&dentry->d_inode->i_mutex);
3159 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3160 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3161 fsnotify_link_count(dentry->d_inode);
3169 * Make sure that the actual truncation of the file will occur outside its
3170 * directory's i_mutex. Truncate can take a long time if there is a lot of
3171 * writeout happening, and we don't want to prevent access to the directory
3172 * while waiting on the I/O.
3174 static long do_unlinkat(int dfd, const char __user *pathname)
3178 struct dentry *dentry;
3179 struct nameidata nd;
3180 struct inode *inode = NULL;
3182 error = user_path_parent(dfd, pathname, &nd, &name);
3187 if (nd.last_type != LAST_NORM)
3190 nd.flags &= ~LOOKUP_PARENT;
3192 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3193 dentry = lookup_hash(&nd);
3194 error = PTR_ERR(dentry);
3195 if (!IS_ERR(dentry)) {
3196 /* Why not before? Because we want correct error value */
3197 if (nd.last.name[nd.last.len])
3199 inode = dentry->d_inode;
3203 error = mnt_want_write(nd.path.mnt);
3206 error = security_path_unlink(&nd.path, dentry);
3209 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
3211 mnt_drop_write(nd.path.mnt);
3215 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3217 iput(inode); /* truncate the inode here */
3224 error = !dentry->d_inode ? -ENOENT :
3225 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3229 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3231 if ((flag & ~AT_REMOVEDIR) != 0)
3234 if (flag & AT_REMOVEDIR)
3235 return do_rmdir(dfd, pathname);
3237 return do_unlinkat(dfd, pathname);
3240 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3242 return do_unlinkat(AT_FDCWD, pathname);
3245 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3247 int error = may_create(dir, dentry);
3252 if (!dir->i_op->symlink)
3255 error = security_inode_symlink(dir, dentry, oldname);
3259 error = dir->i_op->symlink(dir, dentry, oldname);
3261 fsnotify_create(dir, dentry);
3265 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3266 int, newdfd, const char __user *, newname)
3270 struct dentry *dentry;
3273 from = getname(oldname);
3275 return PTR_ERR(from);
3277 dentry = user_path_create(newdfd, newname, &path, 0);
3278 error = PTR_ERR(dentry);
3282 error = mnt_want_write(path.mnt);
3285 error = security_path_symlink(&path, dentry, from);
3287 goto out_drop_write;
3288 error = vfs_symlink(path.dentry->d_inode, dentry, from);
3290 mnt_drop_write(path.mnt);
3293 mutex_unlock(&path.dentry->d_inode->i_mutex);
3300 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3302 return sys_symlinkat(oldname, AT_FDCWD, newname);
3305 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3307 struct inode *inode = old_dentry->d_inode;
3308 unsigned max_links = dir->i_sb->s_max_links;
3314 error = may_create(dir, new_dentry);
3318 if (dir->i_sb != inode->i_sb)
3322 * A link to an append-only or immutable file cannot be created.
3324 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3326 if (!dir->i_op->link)
3328 if (S_ISDIR(inode->i_mode))
3331 error = security_inode_link(old_dentry, dir, new_dentry);
3335 mutex_lock(&inode->i_mutex);
3336 /* Make sure we don't allow creating hardlink to an unlinked file */
3337 if (inode->i_nlink == 0)
3339 else if (max_links && inode->i_nlink >= max_links)
3342 error = dir->i_op->link(old_dentry, dir, new_dentry);
3343 mutex_unlock(&inode->i_mutex);
3345 fsnotify_link(dir, inode, new_dentry);
3350 * Hardlinks are often used in delicate situations. We avoid
3351 * security-related surprises by not following symlinks on the
3354 * We don't follow them on the oldname either to be compatible
3355 * with linux 2.0, and to avoid hard-linking to directories
3356 * and other special files. --ADM
3358 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3359 int, newdfd, const char __user *, newname, int, flags)
3361 struct dentry *new_dentry;
3362 struct path old_path, new_path;
3366 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3369 * To use null names we require CAP_DAC_READ_SEARCH
3370 * This ensures that not everyone will be able to create
3371 * handlink using the passed filedescriptor.
3373 if (flags & AT_EMPTY_PATH) {
3374 if (!capable(CAP_DAC_READ_SEARCH))
3379 if (flags & AT_SYMLINK_FOLLOW)
3380 how |= LOOKUP_FOLLOW;
3382 error = user_path_at(olddfd, oldname, how, &old_path);
3386 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3387 error = PTR_ERR(new_dentry);
3388 if (IS_ERR(new_dentry))
3392 if (old_path.mnt != new_path.mnt)
3394 error = mnt_want_write(new_path.mnt);
3397 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3399 goto out_drop_write;
3400 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3402 mnt_drop_write(new_path.mnt);
3405 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3406 path_put(&new_path);
3408 path_put(&old_path);
3413 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3415 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3419 * The worst of all namespace operations - renaming directory. "Perverted"
3420 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3422 * a) we can get into loop creation. Check is done in is_subdir().
3423 * b) race potential - two innocent renames can create a loop together.
3424 * That's where 4.4 screws up. Current fix: serialization on
3425 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3427 * c) we have to lock _three_ objects - parents and victim (if it exists).
3428 * And that - after we got ->i_mutex on parents (until then we don't know
3429 * whether the target exists). Solution: try to be smart with locking
3430 * order for inodes. We rely on the fact that tree topology may change
3431 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3432 * move will be locked. Thus we can rank directories by the tree
3433 * (ancestors first) and rank all non-directories after them.
3434 * That works since everybody except rename does "lock parent, lookup,
3435 * lock child" and rename is under ->s_vfs_rename_mutex.
3436 * HOWEVER, it relies on the assumption that any object with ->lookup()
3437 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3438 * we'd better make sure that there's no link(2) for them.
3439 * d) conversion from fhandle to dentry may come in the wrong moment - when
3440 * we are removing the target. Solution: we will have to grab ->i_mutex
3441 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3442 * ->i_mutex on parents, which works but leads to some truly excessive
3445 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3446 struct inode *new_dir, struct dentry *new_dentry)
3449 struct inode *target = new_dentry->d_inode;
3450 unsigned max_links = new_dir->i_sb->s_max_links;
3453 * If we are going to change the parent - check write permissions,
3454 * we'll need to flip '..'.
3456 if (new_dir != old_dir) {
3457 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3462 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3468 mutex_lock(&target->i_mutex);
3471 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3475 if (max_links && !target && new_dir != old_dir &&
3476 new_dir->i_nlink >= max_links)
3480 shrink_dcache_parent(new_dentry);
3481 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3486 target->i_flags |= S_DEAD;
3487 dont_mount(new_dentry);
3491 mutex_unlock(&target->i_mutex);
3494 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3495 d_move(old_dentry,new_dentry);
3499 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3500 struct inode *new_dir, struct dentry *new_dentry)
3502 struct inode *target = new_dentry->d_inode;
3505 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3511 mutex_lock(&target->i_mutex);
3514 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3517 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3522 dont_mount(new_dentry);
3523 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3524 d_move(old_dentry, new_dentry);
3527 mutex_unlock(&target->i_mutex);
3532 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3533 struct inode *new_dir, struct dentry *new_dentry)
3536 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3537 const unsigned char *old_name;
3539 if (old_dentry->d_inode == new_dentry->d_inode)
3542 error = may_delete(old_dir, old_dentry, is_dir);
3546 if (!new_dentry->d_inode)
3547 error = may_create(new_dir, new_dentry);
3549 error = may_delete(new_dir, new_dentry, is_dir);
3553 if (!old_dir->i_op->rename)
3556 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3559 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3561 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3563 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3564 new_dentry->d_inode, old_dentry);
3565 fsnotify_oldname_free(old_name);
3570 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3571 int, newdfd, const char __user *, newname)
3573 struct dentry *old_dir, *new_dir;
3574 struct dentry *old_dentry, *new_dentry;
3575 struct dentry *trap;
3576 struct nameidata oldnd, newnd;
3581 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3585 error = user_path_parent(newdfd, newname, &newnd, &to);
3590 if (oldnd.path.mnt != newnd.path.mnt)
3593 old_dir = oldnd.path.dentry;
3595 if (oldnd.last_type != LAST_NORM)
3598 new_dir = newnd.path.dentry;
3599 if (newnd.last_type != LAST_NORM)
3602 oldnd.flags &= ~LOOKUP_PARENT;
3603 newnd.flags &= ~LOOKUP_PARENT;
3604 newnd.flags |= LOOKUP_RENAME_TARGET;
3606 trap = lock_rename(new_dir, old_dir);
3608 old_dentry = lookup_hash(&oldnd);
3609 error = PTR_ERR(old_dentry);
3610 if (IS_ERR(old_dentry))
3612 /* source must exist */
3614 if (!old_dentry->d_inode)
3616 /* unless the source is a directory trailing slashes give -ENOTDIR */
3617 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3619 if (oldnd.last.name[oldnd.last.len])
3621 if (newnd.last.name[newnd.last.len])
3624 /* source should not be ancestor of target */
3626 if (old_dentry == trap)
3628 new_dentry = lookup_hash(&newnd);
3629 error = PTR_ERR(new_dentry);
3630 if (IS_ERR(new_dentry))
3632 /* target should not be an ancestor of source */
3634 if (new_dentry == trap)
3637 error = mnt_want_write(oldnd.path.mnt);
3640 error = security_path_rename(&oldnd.path, old_dentry,
3641 &newnd.path, new_dentry);
3644 error = vfs_rename(old_dir->d_inode, old_dentry,
3645 new_dir->d_inode, new_dentry);
3647 mnt_drop_write(oldnd.path.mnt);
3653 unlock_rename(new_dir, old_dir);
3655 path_put(&newnd.path);
3658 path_put(&oldnd.path);
3664 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3666 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3669 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3673 len = PTR_ERR(link);
3678 if (len > (unsigned) buflen)
3680 if (copy_to_user(buffer, link, len))
3687 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3688 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3689 * using) it for any given inode is up to filesystem.
3691 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3693 struct nameidata nd;
3698 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3700 return PTR_ERR(cookie);
3702 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3703 if (dentry->d_inode->i_op->put_link)
3704 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3708 int vfs_follow_link(struct nameidata *nd, const char *link)
3710 return __vfs_follow_link(nd, link);
3713 /* get the link contents into pagecache */
3714 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3718 struct address_space *mapping = dentry->d_inode->i_mapping;
3719 page = read_mapping_page(mapping, 0, NULL);
3724 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3728 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3730 struct page *page = NULL;
3731 char *s = page_getlink(dentry, &page);
3732 int res = vfs_readlink(dentry,buffer,buflen,s);
3735 page_cache_release(page);
3740 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3742 struct page *page = NULL;
3743 nd_set_link(nd, page_getlink(dentry, &page));
3747 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3749 struct page *page = cookie;
3753 page_cache_release(page);
3758 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3760 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3762 struct address_space *mapping = inode->i_mapping;
3767 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3769 flags |= AOP_FLAG_NOFS;
3772 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3773 flags, &page, &fsdata);
3777 kaddr = kmap_atomic(page);
3778 memcpy(kaddr, symname, len-1);
3779 kunmap_atomic(kaddr);
3781 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3788 mark_inode_dirty(inode);
3794 int page_symlink(struct inode *inode, const char *symname, int len)
3796 return __page_symlink(inode, symname, len,
3797 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3800 const struct inode_operations page_symlink_inode_operations = {
3801 .readlink = generic_readlink,
3802 .follow_link = page_follow_link_light,
3803 .put_link = page_put_link,
3806 EXPORT_SYMBOL(user_path_at);
3807 EXPORT_SYMBOL(follow_down_one);
3808 EXPORT_SYMBOL(follow_down);
3809 EXPORT_SYMBOL(follow_up);
3810 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3811 EXPORT_SYMBOL(getname);
3812 EXPORT_SYMBOL(lock_rename);
3813 EXPORT_SYMBOL(lookup_one_len);
3814 EXPORT_SYMBOL(page_follow_link_light);
3815 EXPORT_SYMBOL(page_put_link);
3816 EXPORT_SYMBOL(page_readlink);
3817 EXPORT_SYMBOL(__page_symlink);
3818 EXPORT_SYMBOL(page_symlink);
3819 EXPORT_SYMBOL(page_symlink_inode_operations);
3820 EXPORT_SYMBOL(kern_path);
3821 EXPORT_SYMBOL(vfs_path_lookup);
3822 EXPORT_SYMBOL(inode_permission);
3823 EXPORT_SYMBOL(unlock_rename);
3824 EXPORT_SYMBOL(vfs_create);
3825 EXPORT_SYMBOL(vfs_follow_link);
3826 EXPORT_SYMBOL(vfs_link);
3827 EXPORT_SYMBOL(vfs_mkdir);
3828 EXPORT_SYMBOL(vfs_mknod);
3829 EXPORT_SYMBOL(generic_permission);
3830 EXPORT_SYMBOL(vfs_readlink);
3831 EXPORT_SYMBOL(vfs_rename);
3832 EXPORT_SYMBOL(vfs_rmdir);
3833 EXPORT_SYMBOL(vfs_symlink);
3834 EXPORT_SYMBOL(vfs_unlink);
3835 EXPORT_SYMBOL(dentry_unhash);
3836 EXPORT_SYMBOL(generic_readlink);