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 (dentry && d_need_lookup(dentry)) {
1178 dentry = d_inode_lookup(parent, dentry, nd);
1179 if (IS_ERR(dentry)) {
1180 mutex_unlock(&dir->i_mutex);
1181 return PTR_ERR(dentry);
1183 } else if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1184 status = d_revalidate(dentry, nd);
1185 if (unlikely(status <= 0)) {
1187 mutex_unlock(&dir->i_mutex);
1191 if (!d_invalidate(dentry)) {
1193 dentry = d_alloc_and_lookup(parent, name, nd);
1194 if (IS_ERR(dentry)) {
1195 mutex_unlock(&dir->i_mutex);
1196 return PTR_ERR(dentry);
1200 } else if (!dentry) {
1201 dentry = d_alloc_and_lookup(parent, name, nd);
1202 if (IS_ERR(dentry)) {
1203 mutex_unlock(&dir->i_mutex);
1204 return PTR_ERR(dentry);
1207 mutex_unlock(&dir->i_mutex);
1210 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1211 status = d_revalidate(dentry, nd);
1212 if (unlikely(status <= 0)) {
1217 if (!d_invalidate(dentry)) {
1225 path->dentry = dentry;
1226 err = follow_managed(path, nd->flags);
1227 if (unlikely(err < 0)) {
1228 path_put_conditional(path, nd);
1232 nd->flags |= LOOKUP_JUMPED;
1233 *inode = path->dentry->d_inode;
1237 static inline int may_lookup(struct nameidata *nd)
1239 if (nd->flags & LOOKUP_RCU) {
1240 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1243 if (unlazy_walk(nd, NULL))
1246 return inode_permission(nd->inode, MAY_EXEC);
1249 static inline int handle_dots(struct nameidata *nd, int type)
1251 if (type == LAST_DOTDOT) {
1252 if (nd->flags & LOOKUP_RCU) {
1253 if (follow_dotdot_rcu(nd))
1261 static void terminate_walk(struct nameidata *nd)
1263 if (!(nd->flags & LOOKUP_RCU)) {
1264 path_put(&nd->path);
1266 nd->flags &= ~LOOKUP_RCU;
1267 if (!(nd->flags & LOOKUP_ROOT))
1268 nd->root.mnt = NULL;
1270 br_read_unlock(vfsmount_lock);
1275 * Do we need to follow links? We _really_ want to be able
1276 * to do this check without having to look at inode->i_op,
1277 * so we keep a cache of "no, this doesn't need follow_link"
1278 * for the common case.
1280 static inline int should_follow_link(struct inode *inode, int follow)
1282 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1283 if (likely(inode->i_op->follow_link))
1286 /* This gets set once for the inode lifetime */
1287 spin_lock(&inode->i_lock);
1288 inode->i_opflags |= IOP_NOFOLLOW;
1289 spin_unlock(&inode->i_lock);
1294 static inline int walk_component(struct nameidata *nd, struct path *path,
1295 struct qstr *name, int type, int follow)
1297 struct inode *inode;
1300 * "." and ".." are special - ".." especially so because it has
1301 * to be able to know about the current root directory and
1302 * parent relationships.
1304 if (unlikely(type != LAST_NORM))
1305 return handle_dots(nd, type);
1306 err = do_lookup(nd, name, path, &inode);
1307 if (unlikely(err)) {
1312 path_to_nameidata(path, nd);
1316 if (should_follow_link(inode, follow)) {
1317 if (nd->flags & LOOKUP_RCU) {
1318 if (unlikely(unlazy_walk(nd, path->dentry))) {
1323 BUG_ON(inode != path->dentry->d_inode);
1326 path_to_nameidata(path, nd);
1332 * This limits recursive symlink follows to 8, while
1333 * limiting consecutive symlinks to 40.
1335 * Without that kind of total limit, nasty chains of consecutive
1336 * symlinks can cause almost arbitrarily long lookups.
1338 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1342 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1343 path_put_conditional(path, nd);
1344 path_put(&nd->path);
1347 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1350 current->link_count++;
1353 struct path link = *path;
1356 res = follow_link(&link, nd, &cookie);
1358 res = walk_component(nd, path, &nd->last,
1359 nd->last_type, LOOKUP_FOLLOW);
1360 put_link(nd, &link, cookie);
1363 current->link_count--;
1369 * We really don't want to look at inode->i_op->lookup
1370 * when we don't have to. So we keep a cache bit in
1371 * the inode ->i_opflags field that says "yes, we can
1372 * do lookup on this inode".
1374 static inline int can_lookup(struct inode *inode)
1376 if (likely(inode->i_opflags & IOP_LOOKUP))
1378 if (likely(!inode->i_op->lookup))
1381 /* We do this once for the lifetime of the inode */
1382 spin_lock(&inode->i_lock);
1383 inode->i_opflags |= IOP_LOOKUP;
1384 spin_unlock(&inode->i_lock);
1389 * We can do the critical dentry name comparison and hashing
1390 * operations one word at a time, but we are limited to:
1392 * - Architectures with fast unaligned word accesses. We could
1393 * do a "get_unaligned()" if this helps and is sufficiently
1396 * - Little-endian machines (so that we can generate the mask
1397 * of low bytes efficiently). Again, we *could* do a byte
1398 * swapping load on big-endian architectures if that is not
1399 * expensive enough to make the optimization worthless.
1401 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1402 * do not trap on the (extremely unlikely) case of a page
1403 * crossing operation.
1405 * - Furthermore, we need an efficient 64-bit compile for the
1406 * 64-bit case in order to generate the "number of bytes in
1407 * the final mask". Again, that could be replaced with a
1408 * efficient population count instruction or similar.
1410 #ifdef CONFIG_DCACHE_WORD_ACCESS
1415 * Jan Achrenius on G+: microoptimized version of
1416 * the simpler "(mask & ONEBYTES) * ONEBYTES >> 56"
1417 * that works for the bytemasks without having to
1420 static inline long count_masked_bytes(unsigned long mask)
1422 return mask*0x0001020304050608ul >> 56;
1425 static inline unsigned int fold_hash(unsigned long hash)
1427 hash += hash >> (8*sizeof(int));
1431 #else /* 32-bit case */
1433 /* Carl Chatfield / Jan Achrenius G+ version for 32-bit */
1434 static inline long count_masked_bytes(long mask)
1436 /* (000000 0000ff 00ffff ffffff) -> ( 1 1 2 3 ) */
1437 long a = (0x0ff0001+mask) >> 23;
1438 /* Fix the 1 for 00 case */
1442 #define fold_hash(x) (x)
1446 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1448 unsigned long a, mask;
1449 unsigned long hash = 0;
1452 a = *(unsigned long *)name;
1453 if (len < sizeof(unsigned long))
1457 name += sizeof(unsigned long);
1458 len -= sizeof(unsigned long);
1462 mask = ~(~0ul << len*8);
1465 return fold_hash(hash);
1467 EXPORT_SYMBOL(full_name_hash);
1469 #define REPEAT_BYTE(x) ((~0ul / 0xff) * (x))
1470 #define ONEBYTES REPEAT_BYTE(0x01)
1471 #define SLASHBYTES REPEAT_BYTE('/')
1472 #define HIGHBITS REPEAT_BYTE(0x80)
1474 /* Return the high bit set in the first byte that is a zero */
1475 static inline unsigned long has_zero(unsigned long a)
1477 return ((a - ONEBYTES) & ~a) & HIGHBITS;
1481 * Calculate the length and hash of the path component, and
1482 * return the length of the component;
1484 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1486 unsigned long a, mask, hash, len;
1489 len = -sizeof(unsigned long);
1491 hash = (hash + a) * 9;
1492 len += sizeof(unsigned long);
1493 a = *(unsigned long *)(name+len);
1494 /* Do we have any NUL or '/' bytes in this word? */
1495 mask = has_zero(a) | has_zero(a ^ SLASHBYTES);
1498 /* The mask *below* the first high bit set */
1499 mask = (mask - 1) & ~mask;
1502 *hashp = fold_hash(hash);
1504 return len + count_masked_bytes(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);
1779 err = lookup_last(nd, &path);
1780 put_link(nd, &link, cookie);
1785 err = complete_walk(nd);
1787 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1788 if (!nd->inode->i_op->lookup) {
1789 path_put(&nd->path);
1797 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1798 path_put(&nd->root);
1799 nd->root.mnt = NULL;
1804 static int do_path_lookup(int dfd, const char *name,
1805 unsigned int flags, struct nameidata *nd)
1807 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1808 if (unlikely(retval == -ECHILD))
1809 retval = path_lookupat(dfd, name, flags, nd);
1810 if (unlikely(retval == -ESTALE))
1811 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1813 if (likely(!retval)) {
1814 if (unlikely(!audit_dummy_context())) {
1815 if (nd->path.dentry && nd->inode)
1816 audit_inode(name, nd->path.dentry);
1822 int kern_path_parent(const char *name, struct nameidata *nd)
1824 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1827 int kern_path(const char *name, unsigned int flags, struct path *path)
1829 struct nameidata nd;
1830 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1837 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1838 * @dentry: pointer to dentry of the base directory
1839 * @mnt: pointer to vfs mount of the base directory
1840 * @name: pointer to file name
1841 * @flags: lookup flags
1842 * @path: pointer to struct path to fill
1844 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1845 const char *name, unsigned int flags,
1848 struct nameidata nd;
1850 nd.root.dentry = dentry;
1852 BUG_ON(flags & LOOKUP_PARENT);
1853 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1854 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1860 static struct dentry *__lookup_hash(struct qstr *name,
1861 struct dentry *base, struct nameidata *nd)
1863 struct dentry *dentry;
1866 * Don't bother with __d_lookup: callers are for creat as
1867 * well as unlink, so a lot of the time it would cost
1870 dentry = d_lookup(base, name);
1872 if (dentry && d_need_lookup(dentry)) {
1874 * __lookup_hash is called with the parent dir's i_mutex already
1875 * held, so we are good to go here.
1877 return d_inode_lookup(base, dentry, nd);
1880 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1881 int status = d_revalidate(dentry, nd);
1882 if (unlikely(status <= 0)) {
1884 * The dentry failed validation.
1885 * If d_revalidate returned 0 attempt to invalidate
1886 * the dentry otherwise d_revalidate is asking us
1887 * to return a fail status.
1891 return ERR_PTR(status);
1892 } else if (!d_invalidate(dentry)) {
1900 dentry = d_alloc_and_lookup(base, name, nd);
1906 * Restricted form of lookup. Doesn't follow links, single-component only,
1907 * needs parent already locked. Doesn't follow mounts.
1910 static struct dentry *lookup_hash(struct nameidata *nd)
1912 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1916 * lookup_one_len - filesystem helper to lookup single pathname component
1917 * @name: pathname component to lookup
1918 * @base: base directory to lookup from
1919 * @len: maximum length @len should be interpreted to
1921 * Note that this routine is purely a helper for filesystem usage and should
1922 * not be called by generic code. Also note that by using this function the
1923 * nameidata argument is passed to the filesystem methods and a filesystem
1924 * using this helper needs to be prepared for that.
1926 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1932 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1936 this.hash = full_name_hash(name, len);
1938 return ERR_PTR(-EACCES);
1941 c = *(const unsigned char *)name++;
1942 if (c == '/' || c == '\0')
1943 return ERR_PTR(-EACCES);
1946 * See if the low-level filesystem might want
1947 * to use its own hash..
1949 if (base->d_flags & DCACHE_OP_HASH) {
1950 int err = base->d_op->d_hash(base, base->d_inode, &this);
1952 return ERR_PTR(err);
1955 err = inode_permission(base->d_inode, MAY_EXEC);
1957 return ERR_PTR(err);
1959 return __lookup_hash(&this, base, NULL);
1962 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1963 struct path *path, int *empty)
1965 struct nameidata nd;
1966 char *tmp = getname_flags(name, flags, empty);
1967 int err = PTR_ERR(tmp);
1970 BUG_ON(flags & LOOKUP_PARENT);
1972 err = do_path_lookup(dfd, tmp, flags, &nd);
1980 int user_path_at(int dfd, const char __user *name, unsigned flags,
1983 return user_path_at_empty(dfd, name, flags, path, NULL);
1986 static int user_path_parent(int dfd, const char __user *path,
1987 struct nameidata *nd, char **name)
1989 char *s = getname(path);
1995 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
2005 * It's inline, so penalty for filesystems that don't use sticky bit is
2008 static inline int check_sticky(struct inode *dir, struct inode *inode)
2010 uid_t fsuid = current_fsuid();
2012 if (!(dir->i_mode & S_ISVTX))
2014 if (current_user_ns() != inode_userns(inode))
2016 if (inode->i_uid == fsuid)
2018 if (dir->i_uid == fsuid)
2022 return !ns_capable(inode_userns(inode), CAP_FOWNER);
2026 * Check whether we can remove a link victim from directory dir, check
2027 * whether the type of victim is right.
2028 * 1. We can't do it if dir is read-only (done in permission())
2029 * 2. We should have write and exec permissions on dir
2030 * 3. We can't remove anything from append-only dir
2031 * 4. We can't do anything with immutable dir (done in permission())
2032 * 5. If the sticky bit on dir is set we should either
2033 * a. be owner of dir, or
2034 * b. be owner of victim, or
2035 * c. have CAP_FOWNER capability
2036 * 6. If the victim is append-only or immutable we can't do antyhing with
2037 * links pointing to it.
2038 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2039 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2040 * 9. We can't remove a root or mountpoint.
2041 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2042 * nfs_async_unlink().
2044 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2048 if (!victim->d_inode)
2051 BUG_ON(victim->d_parent->d_inode != dir);
2052 audit_inode_child(victim, dir);
2054 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2059 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2060 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2063 if (!S_ISDIR(victim->d_inode->i_mode))
2065 if (IS_ROOT(victim))
2067 } else if (S_ISDIR(victim->d_inode->i_mode))
2069 if (IS_DEADDIR(dir))
2071 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2076 /* Check whether we can create an object with dentry child in directory
2078 * 1. We can't do it if child already exists (open has special treatment for
2079 * this case, but since we are inlined it's OK)
2080 * 2. We can't do it if dir is read-only (done in permission())
2081 * 3. We should have write and exec permissions on dir
2082 * 4. We can't do it if dir is immutable (done in permission())
2084 static inline int may_create(struct inode *dir, struct dentry *child)
2088 if (IS_DEADDIR(dir))
2090 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2094 * p1 and p2 should be directories on the same fs.
2096 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2101 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2105 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2107 p = d_ancestor(p2, p1);
2109 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2110 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2114 p = d_ancestor(p1, p2);
2116 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2117 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
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 void unlock_rename(struct dentry *p1, struct dentry *p2)
2128 mutex_unlock(&p1->d_inode->i_mutex);
2130 mutex_unlock(&p2->d_inode->i_mutex);
2131 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2135 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2136 struct nameidata *nd)
2138 int error = may_create(dir, dentry);
2143 if (!dir->i_op->create)
2144 return -EACCES; /* shouldn't it be ENOSYS? */
2147 error = security_inode_create(dir, dentry, mode);
2150 error = dir->i_op->create(dir, dentry, mode, nd);
2152 fsnotify_create(dir, dentry);
2156 static int may_open(struct path *path, int acc_mode, int flag)
2158 struct dentry *dentry = path->dentry;
2159 struct inode *inode = dentry->d_inode;
2169 switch (inode->i_mode & S_IFMT) {
2173 if (acc_mode & MAY_WRITE)
2178 if (path->mnt->mnt_flags & MNT_NODEV)
2187 error = inode_permission(inode, acc_mode);
2192 * An append-only file must be opened in append mode for writing.
2194 if (IS_APPEND(inode)) {
2195 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2201 /* O_NOATIME can only be set by the owner or superuser */
2202 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2208 static int handle_truncate(struct file *filp)
2210 struct path *path = &filp->f_path;
2211 struct inode *inode = path->dentry->d_inode;
2212 int error = get_write_access(inode);
2216 * Refuse to truncate files with mandatory locks held on them.
2218 error = locks_verify_locked(inode);
2220 error = security_path_truncate(path);
2222 error = do_truncate(path->dentry, 0,
2223 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2226 put_write_access(inode);
2230 static inline int open_to_namei_flags(int flag)
2232 if ((flag & O_ACCMODE) == 3)
2238 * Handle the last step of open()
2240 static struct file *do_last(struct nameidata *nd, struct path *path,
2241 const struct open_flags *op, const char *pathname)
2243 struct dentry *dir = nd->path.dentry;
2244 struct dentry *dentry;
2245 int open_flag = op->open_flag;
2246 int will_truncate = open_flag & O_TRUNC;
2248 int acc_mode = op->acc_mode;
2252 nd->flags &= ~LOOKUP_PARENT;
2253 nd->flags |= op->intent;
2255 switch (nd->last_type) {
2258 error = handle_dots(nd, nd->last_type);
2260 return ERR_PTR(error);
2263 error = complete_walk(nd);
2265 return ERR_PTR(error);
2266 audit_inode(pathname, nd->path.dentry);
2267 if (open_flag & O_CREAT) {
2273 error = complete_walk(nd);
2275 return ERR_PTR(error);
2276 audit_inode(pathname, dir);
2280 if (!(open_flag & O_CREAT)) {
2282 if (nd->last.name[nd->last.len])
2283 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2284 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2286 /* we _can_ be in RCU mode here */
2287 error = walk_component(nd, path, &nd->last, LAST_NORM,
2290 return ERR_PTR(error);
2291 if (error) /* symlink */
2294 error = complete_walk(nd);
2296 return ERR_PTR(error);
2299 if (nd->flags & LOOKUP_DIRECTORY) {
2300 if (!nd->inode->i_op->lookup)
2303 audit_inode(pathname, nd->path.dentry);
2307 /* create side of things */
2309 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2310 * cleared when we got to the last component we are about to look up
2312 error = complete_walk(nd);
2314 return ERR_PTR(error);
2316 audit_inode(pathname, dir);
2318 /* trailing slashes? */
2319 if (nd->last.name[nd->last.len])
2322 mutex_lock(&dir->d_inode->i_mutex);
2324 dentry = lookup_hash(nd);
2325 error = PTR_ERR(dentry);
2326 if (IS_ERR(dentry)) {
2327 mutex_unlock(&dir->d_inode->i_mutex);
2331 path->dentry = dentry;
2332 path->mnt = nd->path.mnt;
2334 /* Negative dentry, just create the file */
2335 if (!dentry->d_inode) {
2336 umode_t mode = op->mode;
2337 if (!IS_POSIXACL(dir->d_inode))
2338 mode &= ~current_umask();
2340 * This write is needed to ensure that a
2341 * rw->ro transition does not occur between
2342 * the time when the file is created and when
2343 * a permanent write count is taken through
2344 * the 'struct file' in nameidata_to_filp().
2346 error = mnt_want_write(nd->path.mnt);
2348 goto exit_mutex_unlock;
2350 /* Don't check for write permission, don't truncate */
2351 open_flag &= ~O_TRUNC;
2353 acc_mode = MAY_OPEN;
2354 error = security_path_mknod(&nd->path, dentry, mode, 0);
2356 goto exit_mutex_unlock;
2357 error = vfs_create(dir->d_inode, dentry, mode, nd);
2359 goto exit_mutex_unlock;
2360 mutex_unlock(&dir->d_inode->i_mutex);
2361 dput(nd->path.dentry);
2362 nd->path.dentry = dentry;
2367 * It already exists.
2369 mutex_unlock(&dir->d_inode->i_mutex);
2370 audit_inode(pathname, path->dentry);
2373 if (open_flag & O_EXCL)
2376 error = follow_managed(path, nd->flags);
2381 nd->flags |= LOOKUP_JUMPED;
2384 if (!path->dentry->d_inode)
2387 if (path->dentry->d_inode->i_op->follow_link)
2390 path_to_nameidata(path, nd);
2391 nd->inode = path->dentry->d_inode;
2392 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2393 error = complete_walk(nd);
2395 return ERR_PTR(error);
2397 if (S_ISDIR(nd->inode->i_mode))
2400 if (!S_ISREG(nd->inode->i_mode))
2403 if (will_truncate) {
2404 error = mnt_want_write(nd->path.mnt);
2410 error = may_open(&nd->path, acc_mode, open_flag);
2413 filp = nameidata_to_filp(nd);
2414 if (!IS_ERR(filp)) {
2415 error = ima_file_check(filp, op->acc_mode);
2418 filp = ERR_PTR(error);
2421 if (!IS_ERR(filp)) {
2422 if (will_truncate) {
2423 error = handle_truncate(filp);
2426 filp = ERR_PTR(error);
2432 mnt_drop_write(nd->path.mnt);
2433 path_put(&nd->path);
2437 mutex_unlock(&dir->d_inode->i_mutex);
2439 path_put_conditional(path, nd);
2441 filp = ERR_PTR(error);
2445 static struct file *path_openat(int dfd, const char *pathname,
2446 struct nameidata *nd, const struct open_flags *op, int flags)
2448 struct file *base = NULL;
2453 filp = get_empty_filp();
2455 return ERR_PTR(-ENFILE);
2457 filp->f_flags = op->open_flag;
2458 nd->intent.open.file = filp;
2459 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2460 nd->intent.open.create_mode = op->mode;
2462 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2463 if (unlikely(error))
2466 current->total_link_count = 0;
2467 error = link_path_walk(pathname, nd);
2468 if (unlikely(error))
2471 filp = do_last(nd, &path, op, pathname);
2472 while (unlikely(!filp)) { /* trailing symlink */
2473 struct path link = path;
2475 if (!(nd->flags & LOOKUP_FOLLOW)) {
2476 path_put_conditional(&path, nd);
2477 path_put(&nd->path);
2478 filp = ERR_PTR(-ELOOP);
2481 nd->flags |= LOOKUP_PARENT;
2482 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2483 error = follow_link(&link, nd, &cookie);
2484 if (unlikely(error))
2485 filp = ERR_PTR(error);
2487 filp = do_last(nd, &path, op, pathname);
2488 put_link(nd, &link, cookie);
2491 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2492 path_put(&nd->root);
2495 release_open_intent(nd);
2499 filp = ERR_PTR(error);
2503 struct file *do_filp_open(int dfd, const char *pathname,
2504 const struct open_flags *op, int flags)
2506 struct nameidata nd;
2509 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2510 if (unlikely(filp == ERR_PTR(-ECHILD)))
2511 filp = path_openat(dfd, pathname, &nd, op, flags);
2512 if (unlikely(filp == ERR_PTR(-ESTALE)))
2513 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2517 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2518 const char *name, const struct open_flags *op, int flags)
2520 struct nameidata nd;
2524 nd.root.dentry = dentry;
2526 flags |= LOOKUP_ROOT;
2528 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2529 return ERR_PTR(-ELOOP);
2531 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2532 if (unlikely(file == ERR_PTR(-ECHILD)))
2533 file = path_openat(-1, name, &nd, op, flags);
2534 if (unlikely(file == ERR_PTR(-ESTALE)))
2535 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2539 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2541 struct dentry *dentry = ERR_PTR(-EEXIST);
2542 struct nameidata nd;
2543 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2545 return ERR_PTR(error);
2548 * Yucky last component or no last component at all?
2549 * (foo/., foo/.., /////)
2551 if (nd.last_type != LAST_NORM)
2553 nd.flags &= ~LOOKUP_PARENT;
2554 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2555 nd.intent.open.flags = O_EXCL;
2558 * Do the final lookup.
2560 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2561 dentry = lookup_hash(&nd);
2565 if (dentry->d_inode)
2568 * Special case - lookup gave negative, but... we had foo/bar/
2569 * From the vfs_mknod() POV we just have a negative dentry -
2570 * all is fine. Let's be bastards - you had / on the end, you've
2571 * been asking for (non-existent) directory. -ENOENT for you.
2573 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2575 dentry = ERR_PTR(-ENOENT);
2582 dentry = ERR_PTR(-EEXIST);
2584 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2589 EXPORT_SYMBOL(kern_path_create);
2591 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2593 char *tmp = getname(pathname);
2596 return ERR_CAST(tmp);
2597 res = kern_path_create(dfd, tmp, path, is_dir);
2601 EXPORT_SYMBOL(user_path_create);
2603 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2605 int error = may_create(dir, dentry);
2610 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2611 !ns_capable(inode_userns(dir), CAP_MKNOD))
2614 if (!dir->i_op->mknod)
2617 error = devcgroup_inode_mknod(mode, dev);
2621 error = security_inode_mknod(dir, dentry, mode, dev);
2625 error = dir->i_op->mknod(dir, dentry, mode, dev);
2627 fsnotify_create(dir, dentry);
2631 static int may_mknod(umode_t mode)
2633 switch (mode & S_IFMT) {
2639 case 0: /* zero mode translates to S_IFREG */
2648 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2651 struct dentry *dentry;
2658 dentry = user_path_create(dfd, filename, &path, 0);
2660 return PTR_ERR(dentry);
2662 if (!IS_POSIXACL(path.dentry->d_inode))
2663 mode &= ~current_umask();
2664 error = may_mknod(mode);
2667 error = mnt_want_write(path.mnt);
2670 error = security_path_mknod(&path, dentry, mode, dev);
2672 goto out_drop_write;
2673 switch (mode & S_IFMT) {
2674 case 0: case S_IFREG:
2675 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2677 case S_IFCHR: case S_IFBLK:
2678 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2679 new_decode_dev(dev));
2681 case S_IFIFO: case S_IFSOCK:
2682 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2686 mnt_drop_write(path.mnt);
2689 mutex_unlock(&path.dentry->d_inode->i_mutex);
2695 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
2697 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2700 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2702 int error = may_create(dir, dentry);
2703 unsigned max_links = dir->i_sb->s_max_links;
2708 if (!dir->i_op->mkdir)
2711 mode &= (S_IRWXUGO|S_ISVTX);
2712 error = security_inode_mkdir(dir, dentry, mode);
2716 if (max_links && dir->i_nlink >= max_links)
2719 error = dir->i_op->mkdir(dir, dentry, mode);
2721 fsnotify_mkdir(dir, dentry);
2725 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
2727 struct dentry *dentry;
2731 dentry = user_path_create(dfd, pathname, &path, 1);
2733 return PTR_ERR(dentry);
2735 if (!IS_POSIXACL(path.dentry->d_inode))
2736 mode &= ~current_umask();
2737 error = mnt_want_write(path.mnt);
2740 error = security_path_mkdir(&path, dentry, mode);
2742 goto out_drop_write;
2743 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2745 mnt_drop_write(path.mnt);
2748 mutex_unlock(&path.dentry->d_inode->i_mutex);
2753 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
2755 return sys_mkdirat(AT_FDCWD, pathname, mode);
2759 * The dentry_unhash() helper will try to drop the dentry early: we
2760 * should have a usage count of 2 if we're the only user of this
2761 * dentry, and if that is true (possibly after pruning the dcache),
2762 * then we drop the dentry now.
2764 * A low-level filesystem can, if it choses, legally
2767 * if (!d_unhashed(dentry))
2770 * if it cannot handle the case of removing a directory
2771 * that is still in use by something else..
2773 void dentry_unhash(struct dentry *dentry)
2775 shrink_dcache_parent(dentry);
2776 spin_lock(&dentry->d_lock);
2777 if (dentry->d_count == 1)
2779 spin_unlock(&dentry->d_lock);
2782 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2784 int error = may_delete(dir, dentry, 1);
2789 if (!dir->i_op->rmdir)
2793 mutex_lock(&dentry->d_inode->i_mutex);
2796 if (d_mountpoint(dentry))
2799 error = security_inode_rmdir(dir, dentry);
2803 shrink_dcache_parent(dentry);
2804 error = dir->i_op->rmdir(dir, dentry);
2808 dentry->d_inode->i_flags |= S_DEAD;
2812 mutex_unlock(&dentry->d_inode->i_mutex);
2819 static long do_rmdir(int dfd, const char __user *pathname)
2823 struct dentry *dentry;
2824 struct nameidata nd;
2826 error = user_path_parent(dfd, pathname, &nd, &name);
2830 switch(nd.last_type) {
2842 nd.flags &= ~LOOKUP_PARENT;
2844 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2845 dentry = lookup_hash(&nd);
2846 error = PTR_ERR(dentry);
2849 if (!dentry->d_inode) {
2853 error = mnt_want_write(nd.path.mnt);
2856 error = security_path_rmdir(&nd.path, dentry);
2859 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2861 mnt_drop_write(nd.path.mnt);
2865 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2872 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2874 return do_rmdir(AT_FDCWD, pathname);
2877 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2879 int error = may_delete(dir, dentry, 0);
2884 if (!dir->i_op->unlink)
2887 mutex_lock(&dentry->d_inode->i_mutex);
2888 if (d_mountpoint(dentry))
2891 error = security_inode_unlink(dir, dentry);
2893 error = dir->i_op->unlink(dir, dentry);
2898 mutex_unlock(&dentry->d_inode->i_mutex);
2900 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2901 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2902 fsnotify_link_count(dentry->d_inode);
2910 * Make sure that the actual truncation of the file will occur outside its
2911 * directory's i_mutex. Truncate can take a long time if there is a lot of
2912 * writeout happening, and we don't want to prevent access to the directory
2913 * while waiting on the I/O.
2915 static long do_unlinkat(int dfd, const char __user *pathname)
2919 struct dentry *dentry;
2920 struct nameidata nd;
2921 struct inode *inode = NULL;
2923 error = user_path_parent(dfd, pathname, &nd, &name);
2928 if (nd.last_type != LAST_NORM)
2931 nd.flags &= ~LOOKUP_PARENT;
2933 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2934 dentry = lookup_hash(&nd);
2935 error = PTR_ERR(dentry);
2936 if (!IS_ERR(dentry)) {
2937 /* Why not before? Because we want correct error value */
2938 if (nd.last.name[nd.last.len])
2940 inode = dentry->d_inode;
2944 error = mnt_want_write(nd.path.mnt);
2947 error = security_path_unlink(&nd.path, dentry);
2950 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2952 mnt_drop_write(nd.path.mnt);
2956 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2958 iput(inode); /* truncate the inode here */
2965 error = !dentry->d_inode ? -ENOENT :
2966 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2970 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2972 if ((flag & ~AT_REMOVEDIR) != 0)
2975 if (flag & AT_REMOVEDIR)
2976 return do_rmdir(dfd, pathname);
2978 return do_unlinkat(dfd, pathname);
2981 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2983 return do_unlinkat(AT_FDCWD, pathname);
2986 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2988 int error = may_create(dir, dentry);
2993 if (!dir->i_op->symlink)
2996 error = security_inode_symlink(dir, dentry, oldname);
3000 error = dir->i_op->symlink(dir, dentry, oldname);
3002 fsnotify_create(dir, dentry);
3006 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3007 int, newdfd, const char __user *, newname)
3011 struct dentry *dentry;
3014 from = getname(oldname);
3016 return PTR_ERR(from);
3018 dentry = user_path_create(newdfd, newname, &path, 0);
3019 error = PTR_ERR(dentry);
3023 error = mnt_want_write(path.mnt);
3026 error = security_path_symlink(&path, dentry, from);
3028 goto out_drop_write;
3029 error = vfs_symlink(path.dentry->d_inode, dentry, from);
3031 mnt_drop_write(path.mnt);
3034 mutex_unlock(&path.dentry->d_inode->i_mutex);
3041 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3043 return sys_symlinkat(oldname, AT_FDCWD, newname);
3046 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3048 struct inode *inode = old_dentry->d_inode;
3049 unsigned max_links = dir->i_sb->s_max_links;
3055 error = may_create(dir, new_dentry);
3059 if (dir->i_sb != inode->i_sb)
3063 * A link to an append-only or immutable file cannot be created.
3065 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3067 if (!dir->i_op->link)
3069 if (S_ISDIR(inode->i_mode))
3072 error = security_inode_link(old_dentry, dir, new_dentry);
3076 mutex_lock(&inode->i_mutex);
3077 /* Make sure we don't allow creating hardlink to an unlinked file */
3078 if (inode->i_nlink == 0)
3080 else if (max_links && inode->i_nlink >= max_links)
3083 error = dir->i_op->link(old_dentry, dir, new_dentry);
3084 mutex_unlock(&inode->i_mutex);
3086 fsnotify_link(dir, inode, new_dentry);
3091 * Hardlinks are often used in delicate situations. We avoid
3092 * security-related surprises by not following symlinks on the
3095 * We don't follow them on the oldname either to be compatible
3096 * with linux 2.0, and to avoid hard-linking to directories
3097 * and other special files. --ADM
3099 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3100 int, newdfd, const char __user *, newname, int, flags)
3102 struct dentry *new_dentry;
3103 struct path old_path, new_path;
3107 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3110 * To use null names we require CAP_DAC_READ_SEARCH
3111 * This ensures that not everyone will be able to create
3112 * handlink using the passed filedescriptor.
3114 if (flags & AT_EMPTY_PATH) {
3115 if (!capable(CAP_DAC_READ_SEARCH))
3120 if (flags & AT_SYMLINK_FOLLOW)
3121 how |= LOOKUP_FOLLOW;
3123 error = user_path_at(olddfd, oldname, how, &old_path);
3127 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3128 error = PTR_ERR(new_dentry);
3129 if (IS_ERR(new_dentry))
3133 if (old_path.mnt != new_path.mnt)
3135 error = mnt_want_write(new_path.mnt);
3138 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3140 goto out_drop_write;
3141 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3143 mnt_drop_write(new_path.mnt);
3146 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3147 path_put(&new_path);
3149 path_put(&old_path);
3154 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3156 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3160 * The worst of all namespace operations - renaming directory. "Perverted"
3161 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3163 * a) we can get into loop creation. Check is done in is_subdir().
3164 * b) race potential - two innocent renames can create a loop together.
3165 * That's where 4.4 screws up. Current fix: serialization on
3166 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3168 * c) we have to lock _three_ objects - parents and victim (if it exists).
3169 * And that - after we got ->i_mutex on parents (until then we don't know
3170 * whether the target exists). Solution: try to be smart with locking
3171 * order for inodes. We rely on the fact that tree topology may change
3172 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3173 * move will be locked. Thus we can rank directories by the tree
3174 * (ancestors first) and rank all non-directories after them.
3175 * That works since everybody except rename does "lock parent, lookup,
3176 * lock child" and rename is under ->s_vfs_rename_mutex.
3177 * HOWEVER, it relies on the assumption that any object with ->lookup()
3178 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3179 * we'd better make sure that there's no link(2) for them.
3180 * d) conversion from fhandle to dentry may come in the wrong moment - when
3181 * we are removing the target. Solution: we will have to grab ->i_mutex
3182 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3183 * ->i_mutex on parents, which works but leads to some truly excessive
3186 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3187 struct inode *new_dir, struct dentry *new_dentry)
3190 struct inode *target = new_dentry->d_inode;
3191 unsigned max_links = new_dir->i_sb->s_max_links;
3194 * If we are going to change the parent - check write permissions,
3195 * we'll need to flip '..'.
3197 if (new_dir != old_dir) {
3198 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3203 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3209 mutex_lock(&target->i_mutex);
3212 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3216 if (max_links && !target && new_dir != old_dir &&
3217 new_dir->i_nlink >= max_links)
3221 shrink_dcache_parent(new_dentry);
3222 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3227 target->i_flags |= S_DEAD;
3228 dont_mount(new_dentry);
3232 mutex_unlock(&target->i_mutex);
3235 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3236 d_move(old_dentry,new_dentry);
3240 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3241 struct inode *new_dir, struct dentry *new_dentry)
3243 struct inode *target = new_dentry->d_inode;
3246 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3252 mutex_lock(&target->i_mutex);
3255 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3258 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3263 dont_mount(new_dentry);
3264 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3265 d_move(old_dentry, new_dentry);
3268 mutex_unlock(&target->i_mutex);
3273 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3274 struct inode *new_dir, struct dentry *new_dentry)
3277 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3278 const unsigned char *old_name;
3280 if (old_dentry->d_inode == new_dentry->d_inode)
3283 error = may_delete(old_dir, old_dentry, is_dir);
3287 if (!new_dentry->d_inode)
3288 error = may_create(new_dir, new_dentry);
3290 error = may_delete(new_dir, new_dentry, is_dir);
3294 if (!old_dir->i_op->rename)
3297 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3300 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3302 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3304 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3305 new_dentry->d_inode, old_dentry);
3306 fsnotify_oldname_free(old_name);
3311 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3312 int, newdfd, const char __user *, newname)
3314 struct dentry *old_dir, *new_dir;
3315 struct dentry *old_dentry, *new_dentry;
3316 struct dentry *trap;
3317 struct nameidata oldnd, newnd;
3322 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3326 error = user_path_parent(newdfd, newname, &newnd, &to);
3331 if (oldnd.path.mnt != newnd.path.mnt)
3334 old_dir = oldnd.path.dentry;
3336 if (oldnd.last_type != LAST_NORM)
3339 new_dir = newnd.path.dentry;
3340 if (newnd.last_type != LAST_NORM)
3343 oldnd.flags &= ~LOOKUP_PARENT;
3344 newnd.flags &= ~LOOKUP_PARENT;
3345 newnd.flags |= LOOKUP_RENAME_TARGET;
3347 trap = lock_rename(new_dir, old_dir);
3349 old_dentry = lookup_hash(&oldnd);
3350 error = PTR_ERR(old_dentry);
3351 if (IS_ERR(old_dentry))
3353 /* source must exist */
3355 if (!old_dentry->d_inode)
3357 /* unless the source is a directory trailing slashes give -ENOTDIR */
3358 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3360 if (oldnd.last.name[oldnd.last.len])
3362 if (newnd.last.name[newnd.last.len])
3365 /* source should not be ancestor of target */
3367 if (old_dentry == trap)
3369 new_dentry = lookup_hash(&newnd);
3370 error = PTR_ERR(new_dentry);
3371 if (IS_ERR(new_dentry))
3373 /* target should not be an ancestor of source */
3375 if (new_dentry == trap)
3378 error = mnt_want_write(oldnd.path.mnt);
3381 error = security_path_rename(&oldnd.path, old_dentry,
3382 &newnd.path, new_dentry);
3385 error = vfs_rename(old_dir->d_inode, old_dentry,
3386 new_dir->d_inode, new_dentry);
3388 mnt_drop_write(oldnd.path.mnt);
3394 unlock_rename(new_dir, old_dir);
3396 path_put(&newnd.path);
3399 path_put(&oldnd.path);
3405 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3407 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3410 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3414 len = PTR_ERR(link);
3419 if (len > (unsigned) buflen)
3421 if (copy_to_user(buffer, link, len))
3428 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3429 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3430 * using) it for any given inode is up to filesystem.
3432 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3434 struct nameidata nd;
3439 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3441 return PTR_ERR(cookie);
3443 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3444 if (dentry->d_inode->i_op->put_link)
3445 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3449 int vfs_follow_link(struct nameidata *nd, const char *link)
3451 return __vfs_follow_link(nd, link);
3454 /* get the link contents into pagecache */
3455 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3459 struct address_space *mapping = dentry->d_inode->i_mapping;
3460 page = read_mapping_page(mapping, 0, NULL);
3465 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3469 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3471 struct page *page = NULL;
3472 char *s = page_getlink(dentry, &page);
3473 int res = vfs_readlink(dentry,buffer,buflen,s);
3476 page_cache_release(page);
3481 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3483 struct page *page = NULL;
3484 nd_set_link(nd, page_getlink(dentry, &page));
3488 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3490 struct page *page = cookie;
3494 page_cache_release(page);
3499 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3501 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3503 struct address_space *mapping = inode->i_mapping;
3508 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3510 flags |= AOP_FLAG_NOFS;
3513 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3514 flags, &page, &fsdata);
3518 kaddr = kmap_atomic(page);
3519 memcpy(kaddr, symname, len-1);
3520 kunmap_atomic(kaddr);
3522 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3529 mark_inode_dirty(inode);
3535 int page_symlink(struct inode *inode, const char *symname, int len)
3537 return __page_symlink(inode, symname, len,
3538 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3541 const struct inode_operations page_symlink_inode_operations = {
3542 .readlink = generic_readlink,
3543 .follow_link = page_follow_link_light,
3544 .put_link = page_put_link,
3547 EXPORT_SYMBOL(user_path_at);
3548 EXPORT_SYMBOL(follow_down_one);
3549 EXPORT_SYMBOL(follow_down);
3550 EXPORT_SYMBOL(follow_up);
3551 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3552 EXPORT_SYMBOL(getname);
3553 EXPORT_SYMBOL(lock_rename);
3554 EXPORT_SYMBOL(lookup_one_len);
3555 EXPORT_SYMBOL(page_follow_link_light);
3556 EXPORT_SYMBOL(page_put_link);
3557 EXPORT_SYMBOL(page_readlink);
3558 EXPORT_SYMBOL(__page_symlink);
3559 EXPORT_SYMBOL(page_symlink);
3560 EXPORT_SYMBOL(page_symlink_inode_operations);
3561 EXPORT_SYMBOL(kern_path);
3562 EXPORT_SYMBOL(vfs_path_lookup);
3563 EXPORT_SYMBOL(inode_permission);
3564 EXPORT_SYMBOL(unlock_rename);
3565 EXPORT_SYMBOL(vfs_create);
3566 EXPORT_SYMBOL(vfs_follow_link);
3567 EXPORT_SYMBOL(vfs_link);
3568 EXPORT_SYMBOL(vfs_mkdir);
3569 EXPORT_SYMBOL(vfs_mknod);
3570 EXPORT_SYMBOL(generic_permission);
3571 EXPORT_SYMBOL(vfs_readlink);
3572 EXPORT_SYMBOL(vfs_rename);
3573 EXPORT_SYMBOL(vfs_rmdir);
3574 EXPORT_SYMBOL(vfs_symlink);
3575 EXPORT_SYMBOL(vfs_unlink);
3576 EXPORT_SYMBOL(dentry_unhash);
3577 EXPORT_SYMBOL(generic_readlink);