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/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <linux/posix_acl.h>
36 #include <asm/uaccess.h>
41 /* [Feb-1997 T. Schoebel-Theuer]
42 * Fundamental changes in the pathname lookup mechanisms (namei)
43 * were necessary because of omirr. The reason is that omirr needs
44 * to know the _real_ pathname, not the user-supplied one, in case
45 * of symlinks (and also when transname replacements occur).
47 * The new code replaces the old recursive symlink resolution with
48 * an iterative one (in case of non-nested symlink chains). It does
49 * this with calls to <fs>_follow_link().
50 * As a side effect, dir_namei(), _namei() and follow_link() are now
51 * replaced with a single function lookup_dentry() that can handle all
52 * the special cases of the former code.
54 * With the new dcache, the pathname is stored at each inode, at least as
55 * long as the refcount of the inode is positive. As a side effect, the
56 * size of the dcache depends on the inode cache and thus is dynamic.
58 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
59 * resolution to correspond with current state of the code.
61 * Note that the symlink resolution is not *completely* iterative.
62 * There is still a significant amount of tail- and mid- recursion in
63 * the algorithm. Also, note that <fs>_readlink() is not used in
64 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
65 * may return different results than <fs>_follow_link(). Many virtual
66 * filesystems (including /proc) exhibit this behavior.
69 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
70 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
71 * and the name already exists in form of a symlink, try to create the new
72 * name indicated by the symlink. The old code always complained that the
73 * name already exists, due to not following the symlink even if its target
74 * is nonexistent. The new semantics affects also mknod() and link() when
75 * the name is a symlink pointing to a non-existent name.
77 * I don't know which semantics is the right one, since I have no access
78 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
79 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
80 * "old" one. Personally, I think the new semantics is much more logical.
81 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
82 * file does succeed in both HP-UX and SunOs, but not in Solaris
83 * and in the old Linux semantics.
86 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
87 * semantics. See the comments in "open_namei" and "do_link" below.
89 * [10-Sep-98 Alan Modra] Another symlink change.
92 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
93 * inside the path - always follow.
94 * in the last component in creation/removal/renaming - never follow.
95 * if LOOKUP_FOLLOW passed - follow.
96 * if the pathname has trailing slashes - follow.
97 * otherwise - don't follow.
98 * (applied in that order).
100 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
101 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
102 * During the 2.4 we need to fix the userland stuff depending on it -
103 * hopefully we will be able to get rid of that wart in 2.5. So far only
104 * XEmacs seems to be relying on it...
107 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
108 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
109 * any extra contention...
112 /* In order to reduce some races, while at the same time doing additional
113 * checking and hopefully speeding things up, we copy filenames to the
114 * kernel data space before using them..
116 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
117 * PATH_MAX includes the nul terminator --RR.
119 static int do_getname(const char __user *filename, char *page)
122 unsigned long len = PATH_MAX;
124 if (!segment_eq(get_fs(), KERNEL_DS)) {
125 if ((unsigned long) filename >= TASK_SIZE)
127 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
128 len = TASK_SIZE - (unsigned long) filename;
131 retval = strncpy_from_user(page, filename, len);
135 return -ENAMETOOLONG;
141 static char *getname_flags(const char __user *filename, int flags, int *empty)
143 char *result = __getname();
147 return ERR_PTR(-ENOMEM);
149 retval = do_getname(filename, result);
151 if (retval == -ENOENT && empty)
153 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
155 return ERR_PTR(retval);
158 audit_getname(result);
162 char *getname(const char __user * filename)
164 return getname_flags(filename, 0, NULL);
167 #ifdef CONFIG_AUDITSYSCALL
168 void putname(const char *name)
170 if (unlikely(!audit_dummy_context()))
175 EXPORT_SYMBOL(putname);
178 static int check_acl(struct inode *inode, int mask)
180 #ifdef CONFIG_FS_POSIX_ACL
181 struct posix_acl *acl;
183 if (mask & MAY_NOT_BLOCK) {
184 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
187 /* no ->get_acl() calls in RCU mode... */
188 if (acl == ACL_NOT_CACHED)
190 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
193 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
196 * A filesystem can force a ACL callback by just never filling the
197 * ACL cache. But normally you'd fill the cache either at inode
198 * instantiation time, or on the first ->get_acl call.
200 * If the filesystem doesn't have a get_acl() function at all, we'll
201 * just create the negative cache entry.
203 if (acl == ACL_NOT_CACHED) {
204 if (inode->i_op->get_acl) {
205 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
209 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
215 int error = posix_acl_permission(inode, acl, mask);
216 posix_acl_release(acl);
225 * This does the basic permission checking
227 static int acl_permission_check(struct inode *inode, int mask)
229 unsigned int mode = inode->i_mode;
231 if (current_user_ns() != inode_userns(inode))
234 if (likely(current_fsuid() == inode->i_uid))
237 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
238 int error = check_acl(inode, mask);
239 if (error != -EAGAIN)
243 if (in_group_p(inode->i_gid))
249 * If the DACs are ok we don't need any capability check.
251 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
257 * generic_permission - check for access rights on a Posix-like filesystem
258 * @inode: inode to check access rights for
259 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
261 * Used to check for read/write/execute permissions on a file.
262 * We use "fsuid" for this, letting us set arbitrary permissions
263 * for filesystem access without changing the "normal" uids which
264 * are used for other things.
266 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
267 * request cannot be satisfied (eg. requires blocking or too much complexity).
268 * It would then be called again in ref-walk mode.
270 int generic_permission(struct inode *inode, int mask)
275 * Do the basic permission checks.
277 ret = acl_permission_check(inode, mask);
281 if (S_ISDIR(inode->i_mode)) {
282 /* DACs are overridable for directories */
283 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
285 if (!(mask & MAY_WRITE))
286 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
291 * Read/write DACs are always overridable.
292 * Executable DACs are overridable when there is
293 * at least one exec bit set.
295 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
296 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
300 * Searching includes executable on directories, else just read.
302 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
303 if (mask == MAY_READ)
304 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
311 * We _really_ want to just do "generic_permission()" without
312 * even looking at the inode->i_op values. So we keep a cache
313 * flag in inode->i_opflags, that says "this has not special
314 * permission function, use the fast case".
316 static inline int do_inode_permission(struct inode *inode, int mask)
318 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
319 if (likely(inode->i_op->permission))
320 return inode->i_op->permission(inode, mask);
322 /* This gets set once for the inode lifetime */
323 spin_lock(&inode->i_lock);
324 inode->i_opflags |= IOP_FASTPERM;
325 spin_unlock(&inode->i_lock);
327 return generic_permission(inode, mask);
331 * inode_permission - check for access rights to a given inode
332 * @inode: inode to check permission on
333 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
335 * Used to check for read/write/execute permissions on an inode.
336 * We use "fsuid" for this, letting us set arbitrary permissions
337 * for filesystem access without changing the "normal" uids which
338 * are used for other things.
340 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
342 int inode_permission(struct inode *inode, int mask)
346 if (unlikely(mask & MAY_WRITE)) {
347 umode_t mode = inode->i_mode;
350 * Nobody gets write access to a read-only fs.
352 if (IS_RDONLY(inode) &&
353 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
357 * Nobody gets write access to an immutable file.
359 if (IS_IMMUTABLE(inode))
363 retval = do_inode_permission(inode, mask);
367 retval = devcgroup_inode_permission(inode, mask);
371 return security_inode_permission(inode, mask);
375 * path_get - get a reference to a path
376 * @path: path to get the reference to
378 * Given a path increment the reference count to the dentry and the vfsmount.
380 void path_get(struct path *path)
385 EXPORT_SYMBOL(path_get);
388 * path_put - put a reference to a path
389 * @path: path to put the reference to
391 * Given a path decrement the reference count to the dentry and the vfsmount.
393 void path_put(struct path *path)
398 EXPORT_SYMBOL(path_put);
401 * Path walking has 2 modes, rcu-walk and ref-walk (see
402 * Documentation/filesystems/path-lookup.txt). In situations when we can't
403 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
404 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
405 * mode. Refcounts are grabbed at the last known good point before rcu-walk
406 * got stuck, so ref-walk may continue from there. If this is not successful
407 * (eg. a seqcount has changed), then failure is returned and it's up to caller
408 * to restart the path walk from the beginning in ref-walk mode.
412 * unlazy_walk - try to switch to ref-walk mode.
413 * @nd: nameidata pathwalk data
414 * @dentry: child of nd->path.dentry or NULL
415 * Returns: 0 on success, -ECHILD on failure
417 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
418 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
419 * @nd or NULL. Must be called from rcu-walk context.
421 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
423 struct fs_struct *fs = current->fs;
424 struct dentry *parent = nd->path.dentry;
427 BUG_ON(!(nd->flags & LOOKUP_RCU));
428 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
430 spin_lock(&fs->lock);
431 if (nd->root.mnt != fs->root.mnt ||
432 nd->root.dentry != fs->root.dentry)
435 spin_lock(&parent->d_lock);
437 if (!__d_rcu_to_refcount(parent, nd->seq))
439 BUG_ON(nd->inode != parent->d_inode);
441 if (dentry->d_parent != parent)
443 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
444 if (!__d_rcu_to_refcount(dentry, nd->seq))
447 * If the sequence check on the child dentry passed, then
448 * the child has not been removed from its parent. This
449 * means the parent dentry must be valid and able to take
450 * a reference at this point.
452 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
453 BUG_ON(!parent->d_count);
455 spin_unlock(&dentry->d_lock);
457 spin_unlock(&parent->d_lock);
460 spin_unlock(&fs->lock);
462 mntget(nd->path.mnt);
465 br_read_unlock(vfsmount_lock);
466 nd->flags &= ~LOOKUP_RCU;
470 spin_unlock(&dentry->d_lock);
472 spin_unlock(&parent->d_lock);
475 spin_unlock(&fs->lock);
480 * release_open_intent - free up open intent resources
481 * @nd: pointer to nameidata
483 void release_open_intent(struct nameidata *nd)
485 struct file *file = nd->intent.open.file;
487 if (file && !IS_ERR(file)) {
488 if (file->f_path.dentry == NULL)
495 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
497 return dentry->d_op->d_revalidate(dentry, nd);
501 * complete_walk - successful completion of path walk
502 * @nd: pointer nameidata
504 * If we had been in RCU mode, drop out of it and legitimize nd->path.
505 * Revalidate the final result, unless we'd already done that during
506 * the path walk or the filesystem doesn't ask for it. Return 0 on
507 * success, -error on failure. In case of failure caller does not
508 * need to drop nd->path.
510 static int complete_walk(struct nameidata *nd)
512 struct dentry *dentry = nd->path.dentry;
515 if (nd->flags & LOOKUP_RCU) {
516 nd->flags &= ~LOOKUP_RCU;
517 if (!(nd->flags & LOOKUP_ROOT))
519 spin_lock(&dentry->d_lock);
520 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
521 spin_unlock(&dentry->d_lock);
523 br_read_unlock(vfsmount_lock);
526 BUG_ON(nd->inode != dentry->d_inode);
527 spin_unlock(&dentry->d_lock);
528 mntget(nd->path.mnt);
530 br_read_unlock(vfsmount_lock);
533 if (likely(!(nd->flags & LOOKUP_JUMPED)))
536 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
539 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
542 /* Note: we do not d_invalidate() */
543 status = d_revalidate(dentry, nd);
554 static __always_inline void set_root(struct nameidata *nd)
557 get_fs_root(current->fs, &nd->root);
560 static int link_path_walk(const char *, struct nameidata *);
562 static __always_inline void set_root_rcu(struct nameidata *nd)
565 struct fs_struct *fs = current->fs;
569 seq = read_seqcount_begin(&fs->seq);
571 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
572 } while (read_seqcount_retry(&fs->seq, seq));
576 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
588 nd->flags |= LOOKUP_JUMPED;
590 nd->inode = nd->path.dentry->d_inode;
592 ret = link_path_walk(link, nd);
596 return PTR_ERR(link);
599 static void path_put_conditional(struct path *path, struct nameidata *nd)
602 if (path->mnt != nd->path.mnt)
606 static inline void path_to_nameidata(const struct path *path,
607 struct nameidata *nd)
609 if (!(nd->flags & LOOKUP_RCU)) {
610 dput(nd->path.dentry);
611 if (nd->path.mnt != path->mnt)
612 mntput(nd->path.mnt);
614 nd->path.mnt = path->mnt;
615 nd->path.dentry = path->dentry;
618 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
620 struct inode *inode = link->dentry->d_inode;
621 if (!IS_ERR(cookie) && inode->i_op->put_link)
622 inode->i_op->put_link(link->dentry, nd, cookie);
626 static __always_inline int
627 follow_link(struct path *link, struct nameidata *nd, void **p)
630 struct dentry *dentry = link->dentry;
632 BUG_ON(nd->flags & LOOKUP_RCU);
634 if (link->mnt == nd->path.mnt)
637 if (unlikely(current->total_link_count >= 40)) {
638 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
643 current->total_link_count++;
646 nd_set_link(nd, NULL);
648 error = security_inode_follow_link(link->dentry, nd);
650 *p = ERR_PTR(error); /* no ->put_link(), please */
655 nd->last_type = LAST_BIND;
656 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
659 char *s = nd_get_link(nd);
662 error = __vfs_follow_link(nd, s);
663 else if (nd->last_type == LAST_BIND) {
664 nd->flags |= LOOKUP_JUMPED;
665 nd->inode = nd->path.dentry->d_inode;
666 if (nd->inode->i_op->follow_link) {
667 /* stepped on a _really_ weird one */
676 static int follow_up_rcu(struct path *path)
678 struct mount *mnt = real_mount(path->mnt);
679 struct mount *parent;
680 struct dentry *mountpoint;
682 parent = mnt->mnt_parent;
683 if (&parent->mnt == path->mnt)
685 mountpoint = mnt->mnt_mountpoint;
686 path->dentry = mountpoint;
687 path->mnt = &parent->mnt;
691 int follow_up(struct path *path)
693 struct mount *mnt = real_mount(path->mnt);
694 struct mount *parent;
695 struct dentry *mountpoint;
697 br_read_lock(vfsmount_lock);
698 parent = mnt->mnt_parent;
699 if (&parent->mnt == path->mnt) {
700 br_read_unlock(vfsmount_lock);
703 mntget(&parent->mnt);
704 mountpoint = dget(mnt->mnt_mountpoint);
705 br_read_unlock(vfsmount_lock);
707 path->dentry = mountpoint;
709 path->mnt = &parent->mnt;
714 * Perform an automount
715 * - return -EISDIR to tell follow_managed() to stop and return the path we
718 static int follow_automount(struct path *path, unsigned flags,
721 struct vfsmount *mnt;
724 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
727 /* We don't want to mount if someone's just doing a stat -
728 * unless they're stat'ing a directory and appended a '/' to
731 * We do, however, want to mount if someone wants to open or
732 * create a file of any type under the mountpoint, wants to
733 * traverse through the mountpoint or wants to open the
734 * mounted directory. Also, autofs may mark negative dentries
735 * as being automount points. These will need the attentions
736 * of the daemon to instantiate them before they can be used.
738 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
739 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
740 path->dentry->d_inode)
743 current->total_link_count++;
744 if (current->total_link_count >= 40)
747 mnt = path->dentry->d_op->d_automount(path);
750 * The filesystem is allowed to return -EISDIR here to indicate
751 * it doesn't want to automount. For instance, autofs would do
752 * this so that its userspace daemon can mount on this dentry.
754 * However, we can only permit this if it's a terminal point in
755 * the path being looked up; if it wasn't then the remainder of
756 * the path is inaccessible and we should say so.
758 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
763 if (!mnt) /* mount collision */
767 /* lock_mount() may release path->mnt on error */
771 err = finish_automount(mnt, path);
775 /* Someone else made a mount here whilst we were busy */
780 path->dentry = dget(mnt->mnt_root);
789 * Handle a dentry that is managed in some way.
790 * - Flagged for transit management (autofs)
791 * - Flagged as mountpoint
792 * - Flagged as automount point
794 * This may only be called in refwalk mode.
796 * Serialization is taken care of in namespace.c
798 static int follow_managed(struct path *path, unsigned flags)
800 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
802 bool need_mntput = false;
805 /* Given that we're not holding a lock here, we retain the value in a
806 * local variable for each dentry as we look at it so that we don't see
807 * the components of that value change under us */
808 while (managed = ACCESS_ONCE(path->dentry->d_flags),
809 managed &= DCACHE_MANAGED_DENTRY,
810 unlikely(managed != 0)) {
811 /* Allow the filesystem to manage the transit without i_mutex
813 if (managed & DCACHE_MANAGE_TRANSIT) {
814 BUG_ON(!path->dentry->d_op);
815 BUG_ON(!path->dentry->d_op->d_manage);
816 ret = path->dentry->d_op->d_manage(path->dentry, false);
821 /* Transit to a mounted filesystem. */
822 if (managed & DCACHE_MOUNTED) {
823 struct vfsmount *mounted = lookup_mnt(path);
829 path->dentry = dget(mounted->mnt_root);
834 /* Something is mounted on this dentry in another
835 * namespace and/or whatever was mounted there in this
836 * namespace got unmounted before we managed to get the
840 /* Handle an automount point */
841 if (managed & DCACHE_NEED_AUTOMOUNT) {
842 ret = follow_automount(path, flags, &need_mntput);
848 /* We didn't change the current path point */
852 if (need_mntput && path->mnt == mnt)
856 return ret < 0 ? ret : need_mntput;
859 int follow_down_one(struct path *path)
861 struct vfsmount *mounted;
863 mounted = lookup_mnt(path);
868 path->dentry = dget(mounted->mnt_root);
874 static inline bool managed_dentry_might_block(struct dentry *dentry)
876 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
877 dentry->d_op->d_manage(dentry, true) < 0);
881 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
882 * we meet a managed dentry that would need blocking.
884 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
885 struct inode **inode)
888 struct mount *mounted;
890 * Don't forget we might have a non-mountpoint managed dentry
891 * that wants to block transit.
893 if (unlikely(managed_dentry_might_block(path->dentry)))
896 if (!d_mountpoint(path->dentry))
899 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
902 path->mnt = &mounted->mnt;
903 path->dentry = mounted->mnt.mnt_root;
904 nd->flags |= LOOKUP_JUMPED;
905 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
907 * Update the inode too. We don't need to re-check the
908 * dentry sequence number here after this d_inode read,
909 * because a mount-point is always pinned.
911 *inode = path->dentry->d_inode;
916 static void follow_mount_rcu(struct nameidata *nd)
918 while (d_mountpoint(nd->path.dentry)) {
919 struct mount *mounted;
920 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
923 nd->path.mnt = &mounted->mnt;
924 nd->path.dentry = mounted->mnt.mnt_root;
925 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
929 static int follow_dotdot_rcu(struct nameidata *nd)
934 if (nd->path.dentry == nd->root.dentry &&
935 nd->path.mnt == nd->root.mnt) {
938 if (nd->path.dentry != nd->path.mnt->mnt_root) {
939 struct dentry *old = nd->path.dentry;
940 struct dentry *parent = old->d_parent;
943 seq = read_seqcount_begin(&parent->d_seq);
944 if (read_seqcount_retry(&old->d_seq, nd->seq))
946 nd->path.dentry = parent;
950 if (!follow_up_rcu(&nd->path))
952 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
954 follow_mount_rcu(nd);
955 nd->inode = nd->path.dentry->d_inode;
959 nd->flags &= ~LOOKUP_RCU;
960 if (!(nd->flags & LOOKUP_ROOT))
963 br_read_unlock(vfsmount_lock);
968 * Follow down to the covering mount currently visible to userspace. At each
969 * point, the filesystem owning that dentry may be queried as to whether the
970 * caller is permitted to proceed or not.
972 int follow_down(struct path *path)
977 while (managed = ACCESS_ONCE(path->dentry->d_flags),
978 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
979 /* Allow the filesystem to manage the transit without i_mutex
982 * We indicate to the filesystem if someone is trying to mount
983 * something here. This gives autofs the chance to deny anyone
984 * other than its daemon the right to mount on its
987 * The filesystem may sleep at this point.
989 if (managed & DCACHE_MANAGE_TRANSIT) {
990 BUG_ON(!path->dentry->d_op);
991 BUG_ON(!path->dentry->d_op->d_manage);
992 ret = path->dentry->d_op->d_manage(
993 path->dentry, false);
995 return ret == -EISDIR ? 0 : ret;
998 /* Transit to a mounted filesystem. */
999 if (managed & DCACHE_MOUNTED) {
1000 struct vfsmount *mounted = lookup_mnt(path);
1005 path->mnt = mounted;
1006 path->dentry = dget(mounted->mnt_root);
1010 /* Don't handle automount points here */
1017 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1019 static void follow_mount(struct path *path)
1021 while (d_mountpoint(path->dentry)) {
1022 struct vfsmount *mounted = lookup_mnt(path);
1027 path->mnt = mounted;
1028 path->dentry = dget(mounted->mnt_root);
1032 static void follow_dotdot(struct nameidata *nd)
1037 struct dentry *old = nd->path.dentry;
1039 if (nd->path.dentry == nd->root.dentry &&
1040 nd->path.mnt == nd->root.mnt) {
1043 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1044 /* rare case of legitimate dget_parent()... */
1045 nd->path.dentry = dget_parent(nd->path.dentry);
1049 if (!follow_up(&nd->path))
1052 follow_mount(&nd->path);
1053 nd->inode = nd->path.dentry->d_inode;
1057 * Allocate a dentry with name and parent, and perform a parent
1058 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1059 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1060 * have verified that no child exists while under i_mutex.
1062 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1063 struct qstr *name, struct nameidata *nd)
1065 struct inode *inode = parent->d_inode;
1066 struct dentry *dentry;
1069 /* Don't create child dentry for a dead directory. */
1070 if (unlikely(IS_DEADDIR(inode)))
1071 return ERR_PTR(-ENOENT);
1073 dentry = d_alloc(parent, name);
1074 if (unlikely(!dentry))
1075 return ERR_PTR(-ENOMEM);
1077 old = inode->i_op->lookup(inode, dentry, nd);
1078 if (unlikely(old)) {
1086 * We already have a dentry, but require a lookup to be performed on the parent
1087 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1088 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1089 * child exists while under i_mutex.
1091 static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1092 struct nameidata *nd)
1094 struct inode *inode = parent->d_inode;
1097 /* Don't create child dentry for a dead directory. */
1098 if (unlikely(IS_DEADDIR(inode))) {
1100 return ERR_PTR(-ENOENT);
1103 old = inode->i_op->lookup(inode, dentry, nd);
1104 if (unlikely(old)) {
1112 * It's more convoluted than I'd like it to be, but... it's still fairly
1113 * small and for now I'd prefer to have fast path as straight as possible.
1114 * It _is_ time-critical.
1116 static int do_lookup(struct nameidata *nd, struct qstr *name,
1117 struct path *path, struct inode **inode)
1119 struct vfsmount *mnt = nd->path.mnt;
1120 struct dentry *dentry, *parent = nd->path.dentry;
1126 * Rename seqlock is not required here because in the off chance
1127 * of a false negative due to a concurrent rename, we're going to
1128 * do the non-racy lookup, below.
1130 if (nd->flags & LOOKUP_RCU) {
1133 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1137 /* Memory barrier in read_seqcount_begin of child is enough */
1138 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1142 if (unlikely(d_need_lookup(dentry)))
1144 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1145 status = d_revalidate(dentry, nd);
1146 if (unlikely(status <= 0)) {
1147 if (status != -ECHILD)
1153 path->dentry = dentry;
1154 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1156 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1160 if (unlazy_walk(nd, dentry))
1163 dentry = __d_lookup(parent, name);
1166 if (dentry && unlikely(d_need_lookup(dentry))) {
1171 if (unlikely(!dentry)) {
1172 struct inode *dir = parent->d_inode;
1173 BUG_ON(nd->inode != dir);
1175 mutex_lock(&dir->i_mutex);
1176 dentry = d_lookup(parent, name);
1177 if (likely(!dentry)) {
1178 dentry = d_alloc_and_lookup(parent, name, nd);
1179 if (IS_ERR(dentry)) {
1180 mutex_unlock(&dir->i_mutex);
1181 return PTR_ERR(dentry);
1185 } else if (unlikely(d_need_lookup(dentry))) {
1186 dentry = d_inode_lookup(parent, dentry, nd);
1187 if (IS_ERR(dentry)) {
1188 mutex_unlock(&dir->i_mutex);
1189 return PTR_ERR(dentry);
1193 } else if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1194 status = d_revalidate(dentry, nd);
1195 if (unlikely(status <= 0)) {
1197 mutex_unlock(&dir->i_mutex);
1201 if (!d_invalidate(dentry)) {
1203 dentry = d_alloc_and_lookup(parent, name, nd);
1204 if (IS_ERR(dentry)) {
1205 mutex_unlock(&dir->i_mutex);
1206 return PTR_ERR(dentry);
1212 mutex_unlock(&dir->i_mutex);
1215 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1216 status = d_revalidate(dentry, nd);
1217 if (unlikely(status <= 0)) {
1222 if (!d_invalidate(dentry)) {
1230 path->dentry = dentry;
1231 err = follow_managed(path, nd->flags);
1232 if (unlikely(err < 0)) {
1233 path_put_conditional(path, nd);
1237 nd->flags |= LOOKUP_JUMPED;
1238 *inode = path->dentry->d_inode;
1242 static inline int may_lookup(struct nameidata *nd)
1244 if (nd->flags & LOOKUP_RCU) {
1245 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1248 if (unlazy_walk(nd, NULL))
1251 return inode_permission(nd->inode, MAY_EXEC);
1254 static inline int handle_dots(struct nameidata *nd, int type)
1256 if (type == LAST_DOTDOT) {
1257 if (nd->flags & LOOKUP_RCU) {
1258 if (follow_dotdot_rcu(nd))
1266 static void terminate_walk(struct nameidata *nd)
1268 if (!(nd->flags & LOOKUP_RCU)) {
1269 path_put(&nd->path);
1271 nd->flags &= ~LOOKUP_RCU;
1272 if (!(nd->flags & LOOKUP_ROOT))
1273 nd->root.mnt = NULL;
1275 br_read_unlock(vfsmount_lock);
1280 * Do we need to follow links? We _really_ want to be able
1281 * to do this check without having to look at inode->i_op,
1282 * so we keep a cache of "no, this doesn't need follow_link"
1283 * for the common case.
1285 static inline int should_follow_link(struct inode *inode, int follow)
1287 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1288 if (likely(inode->i_op->follow_link))
1291 /* This gets set once for the inode lifetime */
1292 spin_lock(&inode->i_lock);
1293 inode->i_opflags |= IOP_NOFOLLOW;
1294 spin_unlock(&inode->i_lock);
1299 static inline int walk_component(struct nameidata *nd, struct path *path,
1300 struct qstr *name, int type, int follow)
1302 struct inode *inode;
1305 * "." and ".." are special - ".." especially so because it has
1306 * to be able to know about the current root directory and
1307 * parent relationships.
1309 if (unlikely(type != LAST_NORM))
1310 return handle_dots(nd, type);
1311 err = do_lookup(nd, name, path, &inode);
1312 if (unlikely(err)) {
1317 path_to_nameidata(path, nd);
1321 if (should_follow_link(inode, follow)) {
1322 if (nd->flags & LOOKUP_RCU) {
1323 if (unlikely(unlazy_walk(nd, path->dentry))) {
1328 BUG_ON(inode != path->dentry->d_inode);
1331 path_to_nameidata(path, nd);
1337 * This limits recursive symlink follows to 8, while
1338 * limiting consecutive symlinks to 40.
1340 * Without that kind of total limit, nasty chains of consecutive
1341 * symlinks can cause almost arbitrarily long lookups.
1343 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1347 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1348 path_put_conditional(path, nd);
1349 path_put(&nd->path);
1352 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1355 current->link_count++;
1358 struct path link = *path;
1361 res = follow_link(&link, nd, &cookie);
1363 res = walk_component(nd, path, &nd->last,
1364 nd->last_type, LOOKUP_FOLLOW);
1365 put_link(nd, &link, cookie);
1368 current->link_count--;
1374 * We really don't want to look at inode->i_op->lookup
1375 * when we don't have to. So we keep a cache bit in
1376 * the inode ->i_opflags field that says "yes, we can
1377 * do lookup on this inode".
1379 static inline int can_lookup(struct inode *inode)
1381 if (likely(inode->i_opflags & IOP_LOOKUP))
1383 if (likely(!inode->i_op->lookup))
1386 /* We do this once for the lifetime of the inode */
1387 spin_lock(&inode->i_lock);
1388 inode->i_opflags |= IOP_LOOKUP;
1389 spin_unlock(&inode->i_lock);
1394 * We can do the critical dentry name comparison and hashing
1395 * operations one word at a time, but we are limited to:
1397 * - Architectures with fast unaligned word accesses. We could
1398 * do a "get_unaligned()" if this helps and is sufficiently
1401 * - Little-endian machines (so that we can generate the mask
1402 * of low bytes efficiently). Again, we *could* do a byte
1403 * swapping load on big-endian architectures if that is not
1404 * expensive enough to make the optimization worthless.
1406 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1407 * do not trap on the (extremely unlikely) case of a page
1408 * crossing operation.
1410 * - Furthermore, we need an efficient 64-bit compile for the
1411 * 64-bit case in order to generate the "number of bytes in
1412 * the final mask". Again, that could be replaced with a
1413 * efficient population count instruction or similar.
1415 #ifdef CONFIG_DCACHE_WORD_ACCESS
1420 * Jan Achrenius on G+: microoptimized version of
1421 * the simpler "(mask & ONEBYTES) * ONEBYTES >> 56"
1422 * that works for the bytemasks without having to
1425 static inline long count_masked_bytes(unsigned long mask)
1427 return mask*0x0001020304050608ul >> 56;
1430 static inline unsigned int fold_hash(unsigned long hash)
1432 hash += hash >> (8*sizeof(int));
1436 #else /* 32-bit case */
1438 /* Carl Chatfield / Jan Achrenius G+ version for 32-bit */
1439 static inline long count_masked_bytes(long mask)
1441 /* (000000 0000ff 00ffff ffffff) -> ( 1 1 2 3 ) */
1442 long a = (0x0ff0001+mask) >> 23;
1443 /* Fix the 1 for 00 case */
1447 #define fold_hash(x) (x)
1451 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1453 unsigned long a, mask;
1454 unsigned long hash = 0;
1457 a = *(unsigned long *)name;
1458 if (len < sizeof(unsigned long))
1462 name += sizeof(unsigned long);
1463 len -= sizeof(unsigned long);
1467 mask = ~(~0ul << len*8);
1470 return fold_hash(hash);
1472 EXPORT_SYMBOL(full_name_hash);
1474 #define REPEAT_BYTE(x) ((~0ul / 0xff) * (x))
1475 #define ONEBYTES REPEAT_BYTE(0x01)
1476 #define SLASHBYTES REPEAT_BYTE('/')
1477 #define HIGHBITS REPEAT_BYTE(0x80)
1479 /* Return the high bit set in the first byte that is a zero */
1480 static inline unsigned long has_zero(unsigned long a)
1482 return ((a - ONEBYTES) & ~a) & HIGHBITS;
1486 * Calculate the length and hash of the path component, and
1487 * return the length of the component;
1489 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1491 unsigned long a, mask, hash, len;
1494 len = -sizeof(unsigned long);
1496 hash = (hash + a) * 9;
1497 len += sizeof(unsigned long);
1498 a = *(unsigned long *)(name+len);
1499 /* Do we have any NUL or '/' bytes in this word? */
1500 mask = has_zero(a) | has_zero(a ^ SLASHBYTES);
1503 /* The mask *below* the first high bit set */
1504 mask = (mask - 1) & ~mask;
1507 *hashp = fold_hash(hash);
1509 return len + count_masked_bytes(mask);
1514 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1516 unsigned long hash = init_name_hash();
1518 hash = partial_name_hash(*name++, hash);
1519 return end_name_hash(hash);
1521 EXPORT_SYMBOL(full_name_hash);
1524 * We know there's a real path component here of at least
1527 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1529 unsigned long hash = init_name_hash();
1530 unsigned long len = 0, c;
1532 c = (unsigned char)*name;
1535 hash = partial_name_hash(c, hash);
1536 c = (unsigned char)name[len];
1537 } while (c && c != '/');
1538 *hashp = end_name_hash(hash);
1546 * This is the basic name resolution function, turning a pathname into
1547 * the final dentry. We expect 'base' to be positive and a directory.
1549 * Returns 0 and nd will have valid dentry and mnt on success.
1550 * Returns error and drops reference to input namei data on failure.
1552 static int link_path_walk(const char *name, struct nameidata *nd)
1562 /* At this point we know we have a real path component. */
1568 err = may_lookup(nd);
1572 len = hash_name(name, &this.hash);
1577 if (name[0] == '.') switch (len) {
1579 if (name[1] == '.') {
1581 nd->flags |= LOOKUP_JUMPED;
1587 if (likely(type == LAST_NORM)) {
1588 struct dentry *parent = nd->path.dentry;
1589 nd->flags &= ~LOOKUP_JUMPED;
1590 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1591 err = parent->d_op->d_hash(parent, nd->inode,
1599 goto last_component;
1601 * If it wasn't NUL, we know it was '/'. Skip that
1602 * slash, and continue until no more slashes.
1606 } while (unlikely(name[len] == '/'));
1608 goto last_component;
1611 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1616 err = nested_symlink(&next, nd);
1620 if (can_lookup(nd->inode))
1624 /* here ends the main loop */
1628 nd->last_type = type;
1635 static int path_init(int dfd, const char *name, unsigned int flags,
1636 struct nameidata *nd, struct file **fp)
1642 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1643 nd->flags = flags | LOOKUP_JUMPED;
1645 if (flags & LOOKUP_ROOT) {
1646 struct inode *inode = nd->root.dentry->d_inode;
1648 if (!inode->i_op->lookup)
1650 retval = inode_permission(inode, MAY_EXEC);
1654 nd->path = nd->root;
1656 if (flags & LOOKUP_RCU) {
1657 br_read_lock(vfsmount_lock);
1659 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1661 path_get(&nd->path);
1666 nd->root.mnt = NULL;
1669 if (flags & LOOKUP_RCU) {
1670 br_read_lock(vfsmount_lock);
1675 path_get(&nd->root);
1677 nd->path = nd->root;
1678 } else if (dfd == AT_FDCWD) {
1679 if (flags & LOOKUP_RCU) {
1680 struct fs_struct *fs = current->fs;
1683 br_read_lock(vfsmount_lock);
1687 seq = read_seqcount_begin(&fs->seq);
1689 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1690 } while (read_seqcount_retry(&fs->seq, seq));
1692 get_fs_pwd(current->fs, &nd->path);
1695 struct dentry *dentry;
1697 file = fget_raw_light(dfd, &fput_needed);
1702 dentry = file->f_path.dentry;
1706 if (!S_ISDIR(dentry->d_inode->i_mode))
1709 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1714 nd->path = file->f_path;
1715 if (flags & LOOKUP_RCU) {
1718 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1719 br_read_lock(vfsmount_lock);
1722 path_get(&file->f_path);
1723 fput_light(file, fput_needed);
1727 nd->inode = nd->path.dentry->d_inode;
1731 fput_light(file, fput_needed);
1736 static inline int lookup_last(struct nameidata *nd, struct path *path)
1738 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1739 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1741 nd->flags &= ~LOOKUP_PARENT;
1742 return walk_component(nd, path, &nd->last, nd->last_type,
1743 nd->flags & LOOKUP_FOLLOW);
1746 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1747 static int path_lookupat(int dfd, const char *name,
1748 unsigned int flags, struct nameidata *nd)
1750 struct file *base = NULL;
1755 * Path walking is largely split up into 2 different synchronisation
1756 * schemes, rcu-walk and ref-walk (explained in
1757 * Documentation/filesystems/path-lookup.txt). These share much of the
1758 * path walk code, but some things particularly setup, cleanup, and
1759 * following mounts are sufficiently divergent that functions are
1760 * duplicated. Typically there is a function foo(), and its RCU
1761 * analogue, foo_rcu().
1763 * -ECHILD is the error number of choice (just to avoid clashes) that
1764 * is returned if some aspect of an rcu-walk fails. Such an error must
1765 * be handled by restarting a traditional ref-walk (which will always
1766 * be able to complete).
1768 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1773 current->total_link_count = 0;
1774 err = link_path_walk(name, nd);
1776 if (!err && !(flags & LOOKUP_PARENT)) {
1777 err = lookup_last(nd, &path);
1780 struct path link = path;
1781 nd->flags |= LOOKUP_PARENT;
1782 err = follow_link(&link, nd, &cookie);
1784 err = lookup_last(nd, &path);
1785 put_link(nd, &link, cookie);
1790 err = complete_walk(nd);
1792 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1793 if (!nd->inode->i_op->lookup) {
1794 path_put(&nd->path);
1802 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1803 path_put(&nd->root);
1804 nd->root.mnt = NULL;
1809 static int do_path_lookup(int dfd, const char *name,
1810 unsigned int flags, struct nameidata *nd)
1812 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1813 if (unlikely(retval == -ECHILD))
1814 retval = path_lookupat(dfd, name, flags, nd);
1815 if (unlikely(retval == -ESTALE))
1816 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1818 if (likely(!retval)) {
1819 if (unlikely(!audit_dummy_context())) {
1820 if (nd->path.dentry && nd->inode)
1821 audit_inode(name, nd->path.dentry);
1827 int kern_path_parent(const char *name, struct nameidata *nd)
1829 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1832 int kern_path(const char *name, unsigned int flags, struct path *path)
1834 struct nameidata nd;
1835 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1842 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1843 * @dentry: pointer to dentry of the base directory
1844 * @mnt: pointer to vfs mount of the base directory
1845 * @name: pointer to file name
1846 * @flags: lookup flags
1847 * @path: pointer to struct path to fill
1849 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1850 const char *name, unsigned int flags,
1853 struct nameidata nd;
1855 nd.root.dentry = dentry;
1857 BUG_ON(flags & LOOKUP_PARENT);
1858 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1859 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1865 static struct dentry *__lookup_hash(struct qstr *name,
1866 struct dentry *base, struct nameidata *nd)
1868 struct dentry *dentry;
1871 * Don't bother with __d_lookup: callers are for creat as
1872 * well as unlink, so a lot of the time it would cost
1875 dentry = d_lookup(base, name);
1877 if (dentry && d_need_lookup(dentry)) {
1879 * __lookup_hash is called with the parent dir's i_mutex already
1880 * held, so we are good to go here.
1882 return d_inode_lookup(base, dentry, nd);
1885 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1886 int status = d_revalidate(dentry, nd);
1887 if (unlikely(status <= 0)) {
1889 * The dentry failed validation.
1890 * If d_revalidate returned 0 attempt to invalidate
1891 * the dentry otherwise d_revalidate is asking us
1892 * to return a fail status.
1896 return ERR_PTR(status);
1897 } else if (!d_invalidate(dentry)) {
1905 dentry = d_alloc_and_lookup(base, name, nd);
1911 * Restricted form of lookup. Doesn't follow links, single-component only,
1912 * needs parent already locked. Doesn't follow mounts.
1915 static struct dentry *lookup_hash(struct nameidata *nd)
1917 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1921 * lookup_one_len - filesystem helper to lookup single pathname component
1922 * @name: pathname component to lookup
1923 * @base: base directory to lookup from
1924 * @len: maximum length @len should be interpreted to
1926 * Note that this routine is purely a helper for filesystem usage and should
1927 * not be called by generic code. Also note that by using this function the
1928 * nameidata argument is passed to the filesystem methods and a filesystem
1929 * using this helper needs to be prepared for that.
1931 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1937 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1941 this.hash = full_name_hash(name, len);
1943 return ERR_PTR(-EACCES);
1946 c = *(const unsigned char *)name++;
1947 if (c == '/' || c == '\0')
1948 return ERR_PTR(-EACCES);
1951 * See if the low-level filesystem might want
1952 * to use its own hash..
1954 if (base->d_flags & DCACHE_OP_HASH) {
1955 int err = base->d_op->d_hash(base, base->d_inode, &this);
1957 return ERR_PTR(err);
1960 err = inode_permission(base->d_inode, MAY_EXEC);
1962 return ERR_PTR(err);
1964 return __lookup_hash(&this, base, NULL);
1967 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1968 struct path *path, int *empty)
1970 struct nameidata nd;
1971 char *tmp = getname_flags(name, flags, empty);
1972 int err = PTR_ERR(tmp);
1975 BUG_ON(flags & LOOKUP_PARENT);
1977 err = do_path_lookup(dfd, tmp, flags, &nd);
1985 int user_path_at(int dfd, const char __user *name, unsigned flags,
1988 return user_path_at_empty(dfd, name, flags, path, NULL);
1991 static int user_path_parent(int dfd, const char __user *path,
1992 struct nameidata *nd, char **name)
1994 char *s = getname(path);
2000 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
2010 * It's inline, so penalty for filesystems that don't use sticky bit is
2013 static inline int check_sticky(struct inode *dir, struct inode *inode)
2015 uid_t fsuid = current_fsuid();
2017 if (!(dir->i_mode & S_ISVTX))
2019 if (current_user_ns() != inode_userns(inode))
2021 if (inode->i_uid == fsuid)
2023 if (dir->i_uid == fsuid)
2027 return !ns_capable(inode_userns(inode), CAP_FOWNER);
2031 * Check whether we can remove a link victim from directory dir, check
2032 * whether the type of victim is right.
2033 * 1. We can't do it if dir is read-only (done in permission())
2034 * 2. We should have write and exec permissions on dir
2035 * 3. We can't remove anything from append-only dir
2036 * 4. We can't do anything with immutable dir (done in permission())
2037 * 5. If the sticky bit on dir is set we should either
2038 * a. be owner of dir, or
2039 * b. be owner of victim, or
2040 * c. have CAP_FOWNER capability
2041 * 6. If the victim is append-only or immutable we can't do antyhing with
2042 * links pointing to it.
2043 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2044 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2045 * 9. We can't remove a root or mountpoint.
2046 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2047 * nfs_async_unlink().
2049 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2053 if (!victim->d_inode)
2056 BUG_ON(victim->d_parent->d_inode != dir);
2057 audit_inode_child(victim, dir);
2059 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2064 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2065 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2068 if (!S_ISDIR(victim->d_inode->i_mode))
2070 if (IS_ROOT(victim))
2072 } else if (S_ISDIR(victim->d_inode->i_mode))
2074 if (IS_DEADDIR(dir))
2076 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2081 /* Check whether we can create an object with dentry child in directory
2083 * 1. We can't do it if child already exists (open has special treatment for
2084 * this case, but since we are inlined it's OK)
2085 * 2. We can't do it if dir is read-only (done in permission())
2086 * 3. We should have write and exec permissions on dir
2087 * 4. We can't do it if dir is immutable (done in permission())
2089 static inline int may_create(struct inode *dir, struct dentry *child)
2093 if (IS_DEADDIR(dir))
2095 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2099 * p1 and p2 should be directories on the same fs.
2101 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2106 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2110 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2112 p = d_ancestor(p2, p1);
2114 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2115 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2119 p = d_ancestor(p1, p2);
2121 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2122 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2126 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2127 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2131 void unlock_rename(struct dentry *p1, struct dentry *p2)
2133 mutex_unlock(&p1->d_inode->i_mutex);
2135 mutex_unlock(&p2->d_inode->i_mutex);
2136 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2140 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2141 struct nameidata *nd)
2143 int error = may_create(dir, dentry);
2148 if (!dir->i_op->create)
2149 return -EACCES; /* shouldn't it be ENOSYS? */
2152 error = security_inode_create(dir, dentry, mode);
2155 error = dir->i_op->create(dir, dentry, mode, nd);
2157 fsnotify_create(dir, dentry);
2161 static int may_open(struct path *path, int acc_mode, int flag)
2163 struct dentry *dentry = path->dentry;
2164 struct inode *inode = dentry->d_inode;
2174 switch (inode->i_mode & S_IFMT) {
2178 if (acc_mode & MAY_WRITE)
2183 if (path->mnt->mnt_flags & MNT_NODEV)
2192 error = inode_permission(inode, acc_mode);
2197 * An append-only file must be opened in append mode for writing.
2199 if (IS_APPEND(inode)) {
2200 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2206 /* O_NOATIME can only be set by the owner or superuser */
2207 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2213 static int handle_truncate(struct file *filp)
2215 struct path *path = &filp->f_path;
2216 struct inode *inode = path->dentry->d_inode;
2217 int error = get_write_access(inode);
2221 * Refuse to truncate files with mandatory locks held on them.
2223 error = locks_verify_locked(inode);
2225 error = security_path_truncate(path);
2227 error = do_truncate(path->dentry, 0,
2228 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2231 put_write_access(inode);
2235 static inline int open_to_namei_flags(int flag)
2237 if ((flag & O_ACCMODE) == 3)
2243 * Handle the last step of open()
2245 static struct file *do_last(struct nameidata *nd, struct path *path,
2246 const struct open_flags *op, const char *pathname)
2248 struct dentry *dir = nd->path.dentry;
2249 struct dentry *dentry;
2250 int open_flag = op->open_flag;
2251 int will_truncate = open_flag & O_TRUNC;
2253 int acc_mode = op->acc_mode;
2257 nd->flags &= ~LOOKUP_PARENT;
2258 nd->flags |= op->intent;
2260 switch (nd->last_type) {
2263 error = handle_dots(nd, nd->last_type);
2265 return ERR_PTR(error);
2268 error = complete_walk(nd);
2270 return ERR_PTR(error);
2271 audit_inode(pathname, nd->path.dentry);
2272 if (open_flag & O_CREAT) {
2278 error = complete_walk(nd);
2280 return ERR_PTR(error);
2281 audit_inode(pathname, dir);
2285 if (!(open_flag & O_CREAT)) {
2287 if (nd->last.name[nd->last.len])
2288 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2289 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2291 /* we _can_ be in RCU mode here */
2292 error = walk_component(nd, path, &nd->last, LAST_NORM,
2295 return ERR_PTR(error);
2296 if (error) /* symlink */
2299 error = complete_walk(nd);
2301 return ERR_PTR(error);
2304 if (nd->flags & LOOKUP_DIRECTORY) {
2305 if (!nd->inode->i_op->lookup)
2308 audit_inode(pathname, nd->path.dentry);
2312 /* create side of things */
2314 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2315 * cleared when we got to the last component we are about to look up
2317 error = complete_walk(nd);
2319 return ERR_PTR(error);
2321 audit_inode(pathname, dir);
2323 /* trailing slashes? */
2324 if (nd->last.name[nd->last.len])
2327 mutex_lock(&dir->d_inode->i_mutex);
2329 dentry = lookup_hash(nd);
2330 error = PTR_ERR(dentry);
2331 if (IS_ERR(dentry)) {
2332 mutex_unlock(&dir->d_inode->i_mutex);
2336 path->dentry = dentry;
2337 path->mnt = nd->path.mnt;
2339 /* Negative dentry, just create the file */
2340 if (!dentry->d_inode) {
2341 umode_t mode = op->mode;
2342 if (!IS_POSIXACL(dir->d_inode))
2343 mode &= ~current_umask();
2345 * This write is needed to ensure that a
2346 * rw->ro transition does not occur between
2347 * the time when the file is created and when
2348 * a permanent write count is taken through
2349 * the 'struct file' in nameidata_to_filp().
2351 error = mnt_want_write(nd->path.mnt);
2353 goto exit_mutex_unlock;
2355 /* Don't check for write permission, don't truncate */
2356 open_flag &= ~O_TRUNC;
2358 acc_mode = MAY_OPEN;
2359 error = security_path_mknod(&nd->path, dentry, mode, 0);
2361 goto exit_mutex_unlock;
2362 error = vfs_create(dir->d_inode, dentry, mode, nd);
2364 goto exit_mutex_unlock;
2365 mutex_unlock(&dir->d_inode->i_mutex);
2366 dput(nd->path.dentry);
2367 nd->path.dentry = dentry;
2372 * It already exists.
2374 mutex_unlock(&dir->d_inode->i_mutex);
2375 audit_inode(pathname, path->dentry);
2378 if (open_flag & O_EXCL)
2381 error = follow_managed(path, nd->flags);
2386 nd->flags |= LOOKUP_JUMPED;
2389 if (!path->dentry->d_inode)
2392 if (path->dentry->d_inode->i_op->follow_link)
2395 path_to_nameidata(path, nd);
2396 nd->inode = path->dentry->d_inode;
2397 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2398 error = complete_walk(nd);
2400 return ERR_PTR(error);
2402 if (S_ISDIR(nd->inode->i_mode))
2405 if (!S_ISREG(nd->inode->i_mode))
2408 if (will_truncate) {
2409 error = mnt_want_write(nd->path.mnt);
2415 error = may_open(&nd->path, acc_mode, open_flag);
2418 filp = nameidata_to_filp(nd);
2419 if (!IS_ERR(filp)) {
2420 error = ima_file_check(filp, op->acc_mode);
2423 filp = ERR_PTR(error);
2426 if (!IS_ERR(filp)) {
2427 if (will_truncate) {
2428 error = handle_truncate(filp);
2431 filp = ERR_PTR(error);
2437 mnt_drop_write(nd->path.mnt);
2438 path_put(&nd->path);
2442 mutex_unlock(&dir->d_inode->i_mutex);
2444 path_put_conditional(path, nd);
2446 filp = ERR_PTR(error);
2450 static struct file *path_openat(int dfd, const char *pathname,
2451 struct nameidata *nd, const struct open_flags *op, int flags)
2453 struct file *base = NULL;
2458 filp = get_empty_filp();
2460 return ERR_PTR(-ENFILE);
2462 filp->f_flags = op->open_flag;
2463 nd->intent.open.file = filp;
2464 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2465 nd->intent.open.create_mode = op->mode;
2467 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2468 if (unlikely(error))
2471 current->total_link_count = 0;
2472 error = link_path_walk(pathname, nd);
2473 if (unlikely(error))
2476 filp = do_last(nd, &path, op, pathname);
2477 while (unlikely(!filp)) { /* trailing symlink */
2478 struct path link = path;
2480 if (!(nd->flags & LOOKUP_FOLLOW)) {
2481 path_put_conditional(&path, nd);
2482 path_put(&nd->path);
2483 filp = ERR_PTR(-ELOOP);
2486 nd->flags |= LOOKUP_PARENT;
2487 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2488 error = follow_link(&link, nd, &cookie);
2489 if (unlikely(error))
2490 filp = ERR_PTR(error);
2492 filp = do_last(nd, &path, op, pathname);
2493 put_link(nd, &link, cookie);
2496 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2497 path_put(&nd->root);
2500 release_open_intent(nd);
2504 filp = ERR_PTR(error);
2508 struct file *do_filp_open(int dfd, const char *pathname,
2509 const struct open_flags *op, int flags)
2511 struct nameidata nd;
2514 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2515 if (unlikely(filp == ERR_PTR(-ECHILD)))
2516 filp = path_openat(dfd, pathname, &nd, op, flags);
2517 if (unlikely(filp == ERR_PTR(-ESTALE)))
2518 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2522 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2523 const char *name, const struct open_flags *op, int flags)
2525 struct nameidata nd;
2529 nd.root.dentry = dentry;
2531 flags |= LOOKUP_ROOT;
2533 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2534 return ERR_PTR(-ELOOP);
2536 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2537 if (unlikely(file == ERR_PTR(-ECHILD)))
2538 file = path_openat(-1, name, &nd, op, flags);
2539 if (unlikely(file == ERR_PTR(-ESTALE)))
2540 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2544 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2546 struct dentry *dentry = ERR_PTR(-EEXIST);
2547 struct nameidata nd;
2548 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2550 return ERR_PTR(error);
2553 * Yucky last component or no last component at all?
2554 * (foo/., foo/.., /////)
2556 if (nd.last_type != LAST_NORM)
2558 nd.flags &= ~LOOKUP_PARENT;
2559 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2560 nd.intent.open.flags = O_EXCL;
2563 * Do the final lookup.
2565 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2566 dentry = lookup_hash(&nd);
2570 if (dentry->d_inode)
2573 * Special case - lookup gave negative, but... we had foo/bar/
2574 * From the vfs_mknod() POV we just have a negative dentry -
2575 * all is fine. Let's be bastards - you had / on the end, you've
2576 * been asking for (non-existent) directory. -ENOENT for you.
2578 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2580 dentry = ERR_PTR(-ENOENT);
2587 dentry = ERR_PTR(-EEXIST);
2589 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2594 EXPORT_SYMBOL(kern_path_create);
2596 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2598 char *tmp = getname(pathname);
2601 return ERR_CAST(tmp);
2602 res = kern_path_create(dfd, tmp, path, is_dir);
2606 EXPORT_SYMBOL(user_path_create);
2608 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2610 int error = may_create(dir, dentry);
2615 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2616 !ns_capable(inode_userns(dir), CAP_MKNOD))
2619 if (!dir->i_op->mknod)
2622 error = devcgroup_inode_mknod(mode, dev);
2626 error = security_inode_mknod(dir, dentry, mode, dev);
2630 error = dir->i_op->mknod(dir, dentry, mode, dev);
2632 fsnotify_create(dir, dentry);
2636 static int may_mknod(umode_t mode)
2638 switch (mode & S_IFMT) {
2644 case 0: /* zero mode translates to S_IFREG */
2653 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2656 struct dentry *dentry;
2663 dentry = user_path_create(dfd, filename, &path, 0);
2665 return PTR_ERR(dentry);
2667 if (!IS_POSIXACL(path.dentry->d_inode))
2668 mode &= ~current_umask();
2669 error = may_mknod(mode);
2672 error = mnt_want_write(path.mnt);
2675 error = security_path_mknod(&path, dentry, mode, dev);
2677 goto out_drop_write;
2678 switch (mode & S_IFMT) {
2679 case 0: case S_IFREG:
2680 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2682 case S_IFCHR: case S_IFBLK:
2683 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2684 new_decode_dev(dev));
2686 case S_IFIFO: case S_IFSOCK:
2687 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2691 mnt_drop_write(path.mnt);
2694 mutex_unlock(&path.dentry->d_inode->i_mutex);
2700 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
2702 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2705 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2707 int error = may_create(dir, dentry);
2708 unsigned max_links = dir->i_sb->s_max_links;
2713 if (!dir->i_op->mkdir)
2716 mode &= (S_IRWXUGO|S_ISVTX);
2717 error = security_inode_mkdir(dir, dentry, mode);
2721 if (max_links && dir->i_nlink >= max_links)
2724 error = dir->i_op->mkdir(dir, dentry, mode);
2726 fsnotify_mkdir(dir, dentry);
2730 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
2732 struct dentry *dentry;
2736 dentry = user_path_create(dfd, pathname, &path, 1);
2738 return PTR_ERR(dentry);
2740 if (!IS_POSIXACL(path.dentry->d_inode))
2741 mode &= ~current_umask();
2742 error = mnt_want_write(path.mnt);
2745 error = security_path_mkdir(&path, dentry, mode);
2747 goto out_drop_write;
2748 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2750 mnt_drop_write(path.mnt);
2753 mutex_unlock(&path.dentry->d_inode->i_mutex);
2758 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
2760 return sys_mkdirat(AT_FDCWD, pathname, mode);
2764 * The dentry_unhash() helper will try to drop the dentry early: we
2765 * should have a usage count of 2 if we're the only user of this
2766 * dentry, and if that is true (possibly after pruning the dcache),
2767 * then we drop the dentry now.
2769 * A low-level filesystem can, if it choses, legally
2772 * if (!d_unhashed(dentry))
2775 * if it cannot handle the case of removing a directory
2776 * that is still in use by something else..
2778 void dentry_unhash(struct dentry *dentry)
2780 shrink_dcache_parent(dentry);
2781 spin_lock(&dentry->d_lock);
2782 if (dentry->d_count == 1)
2784 spin_unlock(&dentry->d_lock);
2787 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2789 int error = may_delete(dir, dentry, 1);
2794 if (!dir->i_op->rmdir)
2798 mutex_lock(&dentry->d_inode->i_mutex);
2801 if (d_mountpoint(dentry))
2804 error = security_inode_rmdir(dir, dentry);
2808 shrink_dcache_parent(dentry);
2809 error = dir->i_op->rmdir(dir, dentry);
2813 dentry->d_inode->i_flags |= S_DEAD;
2817 mutex_unlock(&dentry->d_inode->i_mutex);
2824 static long do_rmdir(int dfd, const char __user *pathname)
2828 struct dentry *dentry;
2829 struct nameidata nd;
2831 error = user_path_parent(dfd, pathname, &nd, &name);
2835 switch(nd.last_type) {
2847 nd.flags &= ~LOOKUP_PARENT;
2849 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2850 dentry = lookup_hash(&nd);
2851 error = PTR_ERR(dentry);
2854 if (!dentry->d_inode) {
2858 error = mnt_want_write(nd.path.mnt);
2861 error = security_path_rmdir(&nd.path, dentry);
2864 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2866 mnt_drop_write(nd.path.mnt);
2870 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2877 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2879 return do_rmdir(AT_FDCWD, pathname);
2882 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2884 int error = may_delete(dir, dentry, 0);
2889 if (!dir->i_op->unlink)
2892 mutex_lock(&dentry->d_inode->i_mutex);
2893 if (d_mountpoint(dentry))
2896 error = security_inode_unlink(dir, dentry);
2898 error = dir->i_op->unlink(dir, dentry);
2903 mutex_unlock(&dentry->d_inode->i_mutex);
2905 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2906 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2907 fsnotify_link_count(dentry->d_inode);
2915 * Make sure that the actual truncation of the file will occur outside its
2916 * directory's i_mutex. Truncate can take a long time if there is a lot of
2917 * writeout happening, and we don't want to prevent access to the directory
2918 * while waiting on the I/O.
2920 static long do_unlinkat(int dfd, const char __user *pathname)
2924 struct dentry *dentry;
2925 struct nameidata nd;
2926 struct inode *inode = NULL;
2928 error = user_path_parent(dfd, pathname, &nd, &name);
2933 if (nd.last_type != LAST_NORM)
2936 nd.flags &= ~LOOKUP_PARENT;
2938 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2939 dentry = lookup_hash(&nd);
2940 error = PTR_ERR(dentry);
2941 if (!IS_ERR(dentry)) {
2942 /* Why not before? Because we want correct error value */
2943 if (nd.last.name[nd.last.len])
2945 inode = dentry->d_inode;
2949 error = mnt_want_write(nd.path.mnt);
2952 error = security_path_unlink(&nd.path, dentry);
2955 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2957 mnt_drop_write(nd.path.mnt);
2961 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2963 iput(inode); /* truncate the inode here */
2970 error = !dentry->d_inode ? -ENOENT :
2971 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2975 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2977 if ((flag & ~AT_REMOVEDIR) != 0)
2980 if (flag & AT_REMOVEDIR)
2981 return do_rmdir(dfd, pathname);
2983 return do_unlinkat(dfd, pathname);
2986 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2988 return do_unlinkat(AT_FDCWD, pathname);
2991 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2993 int error = may_create(dir, dentry);
2998 if (!dir->i_op->symlink)
3001 error = security_inode_symlink(dir, dentry, oldname);
3005 error = dir->i_op->symlink(dir, dentry, oldname);
3007 fsnotify_create(dir, dentry);
3011 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3012 int, newdfd, const char __user *, newname)
3016 struct dentry *dentry;
3019 from = getname(oldname);
3021 return PTR_ERR(from);
3023 dentry = user_path_create(newdfd, newname, &path, 0);
3024 error = PTR_ERR(dentry);
3028 error = mnt_want_write(path.mnt);
3031 error = security_path_symlink(&path, dentry, from);
3033 goto out_drop_write;
3034 error = vfs_symlink(path.dentry->d_inode, dentry, from);
3036 mnt_drop_write(path.mnt);
3039 mutex_unlock(&path.dentry->d_inode->i_mutex);
3046 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3048 return sys_symlinkat(oldname, AT_FDCWD, newname);
3051 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3053 struct inode *inode = old_dentry->d_inode;
3054 unsigned max_links = dir->i_sb->s_max_links;
3060 error = may_create(dir, new_dentry);
3064 if (dir->i_sb != inode->i_sb)
3068 * A link to an append-only or immutable file cannot be created.
3070 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3072 if (!dir->i_op->link)
3074 if (S_ISDIR(inode->i_mode))
3077 error = security_inode_link(old_dentry, dir, new_dentry);
3081 mutex_lock(&inode->i_mutex);
3082 /* Make sure we don't allow creating hardlink to an unlinked file */
3083 if (inode->i_nlink == 0)
3085 else if (max_links && inode->i_nlink >= max_links)
3088 error = dir->i_op->link(old_dentry, dir, new_dentry);
3089 mutex_unlock(&inode->i_mutex);
3091 fsnotify_link(dir, inode, new_dentry);
3096 * Hardlinks are often used in delicate situations. We avoid
3097 * security-related surprises by not following symlinks on the
3100 * We don't follow them on the oldname either to be compatible
3101 * with linux 2.0, and to avoid hard-linking to directories
3102 * and other special files. --ADM
3104 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3105 int, newdfd, const char __user *, newname, int, flags)
3107 struct dentry *new_dentry;
3108 struct path old_path, new_path;
3112 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3115 * To use null names we require CAP_DAC_READ_SEARCH
3116 * This ensures that not everyone will be able to create
3117 * handlink using the passed filedescriptor.
3119 if (flags & AT_EMPTY_PATH) {
3120 if (!capable(CAP_DAC_READ_SEARCH))
3125 if (flags & AT_SYMLINK_FOLLOW)
3126 how |= LOOKUP_FOLLOW;
3128 error = user_path_at(olddfd, oldname, how, &old_path);
3132 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3133 error = PTR_ERR(new_dentry);
3134 if (IS_ERR(new_dentry))
3138 if (old_path.mnt != new_path.mnt)
3140 error = mnt_want_write(new_path.mnt);
3143 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3145 goto out_drop_write;
3146 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3148 mnt_drop_write(new_path.mnt);
3151 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3152 path_put(&new_path);
3154 path_put(&old_path);
3159 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3161 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3165 * The worst of all namespace operations - renaming directory. "Perverted"
3166 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3168 * a) we can get into loop creation. Check is done in is_subdir().
3169 * b) race potential - two innocent renames can create a loop together.
3170 * That's where 4.4 screws up. Current fix: serialization on
3171 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3173 * c) we have to lock _three_ objects - parents and victim (if it exists).
3174 * And that - after we got ->i_mutex on parents (until then we don't know
3175 * whether the target exists). Solution: try to be smart with locking
3176 * order for inodes. We rely on the fact that tree topology may change
3177 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3178 * move will be locked. Thus we can rank directories by the tree
3179 * (ancestors first) and rank all non-directories after them.
3180 * That works since everybody except rename does "lock parent, lookup,
3181 * lock child" and rename is under ->s_vfs_rename_mutex.
3182 * HOWEVER, it relies on the assumption that any object with ->lookup()
3183 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3184 * we'd better make sure that there's no link(2) for them.
3185 * d) conversion from fhandle to dentry may come in the wrong moment - when
3186 * we are removing the target. Solution: we will have to grab ->i_mutex
3187 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3188 * ->i_mutex on parents, which works but leads to some truly excessive
3191 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3192 struct inode *new_dir, struct dentry *new_dentry)
3195 struct inode *target = new_dentry->d_inode;
3196 unsigned max_links = new_dir->i_sb->s_max_links;
3199 * If we are going to change the parent - check write permissions,
3200 * we'll need to flip '..'.
3202 if (new_dir != old_dir) {
3203 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3208 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3214 mutex_lock(&target->i_mutex);
3217 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3221 if (max_links && !target && new_dir != old_dir &&
3222 new_dir->i_nlink >= max_links)
3226 shrink_dcache_parent(new_dentry);
3227 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3232 target->i_flags |= S_DEAD;
3233 dont_mount(new_dentry);
3237 mutex_unlock(&target->i_mutex);
3240 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3241 d_move(old_dentry,new_dentry);
3245 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3246 struct inode *new_dir, struct dentry *new_dentry)
3248 struct inode *target = new_dentry->d_inode;
3251 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3257 mutex_lock(&target->i_mutex);
3260 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3263 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3268 dont_mount(new_dentry);
3269 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3270 d_move(old_dentry, new_dentry);
3273 mutex_unlock(&target->i_mutex);
3278 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3279 struct inode *new_dir, struct dentry *new_dentry)
3282 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3283 const unsigned char *old_name;
3285 if (old_dentry->d_inode == new_dentry->d_inode)
3288 error = may_delete(old_dir, old_dentry, is_dir);
3292 if (!new_dentry->d_inode)
3293 error = may_create(new_dir, new_dentry);
3295 error = may_delete(new_dir, new_dentry, is_dir);
3299 if (!old_dir->i_op->rename)
3302 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3305 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3307 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3309 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3310 new_dentry->d_inode, old_dentry);
3311 fsnotify_oldname_free(old_name);
3316 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3317 int, newdfd, const char __user *, newname)
3319 struct dentry *old_dir, *new_dir;
3320 struct dentry *old_dentry, *new_dentry;
3321 struct dentry *trap;
3322 struct nameidata oldnd, newnd;
3327 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3331 error = user_path_parent(newdfd, newname, &newnd, &to);
3336 if (oldnd.path.mnt != newnd.path.mnt)
3339 old_dir = oldnd.path.dentry;
3341 if (oldnd.last_type != LAST_NORM)
3344 new_dir = newnd.path.dentry;
3345 if (newnd.last_type != LAST_NORM)
3348 oldnd.flags &= ~LOOKUP_PARENT;
3349 newnd.flags &= ~LOOKUP_PARENT;
3350 newnd.flags |= LOOKUP_RENAME_TARGET;
3352 trap = lock_rename(new_dir, old_dir);
3354 old_dentry = lookup_hash(&oldnd);
3355 error = PTR_ERR(old_dentry);
3356 if (IS_ERR(old_dentry))
3358 /* source must exist */
3360 if (!old_dentry->d_inode)
3362 /* unless the source is a directory trailing slashes give -ENOTDIR */
3363 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3365 if (oldnd.last.name[oldnd.last.len])
3367 if (newnd.last.name[newnd.last.len])
3370 /* source should not be ancestor of target */
3372 if (old_dentry == trap)
3374 new_dentry = lookup_hash(&newnd);
3375 error = PTR_ERR(new_dentry);
3376 if (IS_ERR(new_dentry))
3378 /* target should not be an ancestor of source */
3380 if (new_dentry == trap)
3383 error = mnt_want_write(oldnd.path.mnt);
3386 error = security_path_rename(&oldnd.path, old_dentry,
3387 &newnd.path, new_dentry);
3390 error = vfs_rename(old_dir->d_inode, old_dentry,
3391 new_dir->d_inode, new_dentry);
3393 mnt_drop_write(oldnd.path.mnt);
3399 unlock_rename(new_dir, old_dir);
3401 path_put(&newnd.path);
3404 path_put(&oldnd.path);
3410 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3412 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3415 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3419 len = PTR_ERR(link);
3424 if (len > (unsigned) buflen)
3426 if (copy_to_user(buffer, link, len))
3433 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3434 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3435 * using) it for any given inode is up to filesystem.
3437 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3439 struct nameidata nd;
3444 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3446 return PTR_ERR(cookie);
3448 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3449 if (dentry->d_inode->i_op->put_link)
3450 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3454 int vfs_follow_link(struct nameidata *nd, const char *link)
3456 return __vfs_follow_link(nd, link);
3459 /* get the link contents into pagecache */
3460 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3464 struct address_space *mapping = dentry->d_inode->i_mapping;
3465 page = read_mapping_page(mapping, 0, NULL);
3470 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3474 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3476 struct page *page = NULL;
3477 char *s = page_getlink(dentry, &page);
3478 int res = vfs_readlink(dentry,buffer,buflen,s);
3481 page_cache_release(page);
3486 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3488 struct page *page = NULL;
3489 nd_set_link(nd, page_getlink(dentry, &page));
3493 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3495 struct page *page = cookie;
3499 page_cache_release(page);
3504 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3506 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3508 struct address_space *mapping = inode->i_mapping;
3513 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3515 flags |= AOP_FLAG_NOFS;
3518 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3519 flags, &page, &fsdata);
3523 kaddr = kmap_atomic(page);
3524 memcpy(kaddr, symname, len-1);
3525 kunmap_atomic(kaddr);
3527 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3534 mark_inode_dirty(inode);
3540 int page_symlink(struct inode *inode, const char *symname, int len)
3542 return __page_symlink(inode, symname, len,
3543 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3546 const struct inode_operations page_symlink_inode_operations = {
3547 .readlink = generic_readlink,
3548 .follow_link = page_follow_link_light,
3549 .put_link = page_put_link,
3552 EXPORT_SYMBOL(user_path_at);
3553 EXPORT_SYMBOL(follow_down_one);
3554 EXPORT_SYMBOL(follow_down);
3555 EXPORT_SYMBOL(follow_up);
3556 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3557 EXPORT_SYMBOL(getname);
3558 EXPORT_SYMBOL(lock_rename);
3559 EXPORT_SYMBOL(lookup_one_len);
3560 EXPORT_SYMBOL(page_follow_link_light);
3561 EXPORT_SYMBOL(page_put_link);
3562 EXPORT_SYMBOL(page_readlink);
3563 EXPORT_SYMBOL(__page_symlink);
3564 EXPORT_SYMBOL(page_symlink);
3565 EXPORT_SYMBOL(page_symlink_inode_operations);
3566 EXPORT_SYMBOL(kern_path);
3567 EXPORT_SYMBOL(vfs_path_lookup);
3568 EXPORT_SYMBOL(inode_permission);
3569 EXPORT_SYMBOL(unlock_rename);
3570 EXPORT_SYMBOL(vfs_create);
3571 EXPORT_SYMBOL(vfs_follow_link);
3572 EXPORT_SYMBOL(vfs_link);
3573 EXPORT_SYMBOL(vfs_mkdir);
3574 EXPORT_SYMBOL(vfs_mknod);
3575 EXPORT_SYMBOL(generic_permission);
3576 EXPORT_SYMBOL(vfs_readlink);
3577 EXPORT_SYMBOL(vfs_rename);
3578 EXPORT_SYMBOL(vfs_rmdir);
3579 EXPORT_SYMBOL(vfs_symlink);
3580 EXPORT_SYMBOL(vfs_unlink);
3581 EXPORT_SYMBOL(dentry_unhash);
3582 EXPORT_SYMBOL(generic_readlink);