4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <asm/uaccess.h>
42 /* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
90 * [10-Sep-98 Alan Modra] Another symlink change.
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 static char *getname_flags(const char __user *filename, int flags, int *empty)
122 char *result = __getname(), *err;
125 if (unlikely(!result))
126 return ERR_PTR(-ENOMEM);
128 len = strncpy_from_user(result, filename, PATH_MAX);
130 if (unlikely(len < 0))
133 /* The empty path is special. */
134 if (unlikely(!len)) {
137 err = ERR_PTR(-ENOENT);
138 if (!(flags & LOOKUP_EMPTY))
142 err = ERR_PTR(-ENAMETOOLONG);
143 if (likely(len < PATH_MAX)) {
144 audit_getname(result);
153 char *getname(const char __user * filename)
155 return getname_flags(filename, 0, NULL);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname);
169 static int check_acl(struct inode *inode, int mask)
171 #ifdef CONFIG_FS_POSIX_ACL
172 struct posix_acl *acl;
174 if (mask & MAY_NOT_BLOCK) {
175 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
178 /* no ->get_acl() calls in RCU mode... */
179 if (acl == ACL_NOT_CACHED)
181 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
184 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
187 * A filesystem can force a ACL callback by just never filling the
188 * ACL cache. But normally you'd fill the cache either at inode
189 * instantiation time, or on the first ->get_acl call.
191 * If the filesystem doesn't have a get_acl() function at all, we'll
192 * just create the negative cache entry.
194 if (acl == ACL_NOT_CACHED) {
195 if (inode->i_op->get_acl) {
196 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
200 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
206 int error = posix_acl_permission(inode, acl, mask);
207 posix_acl_release(acl);
216 * This does the basic permission checking
218 static int acl_permission_check(struct inode *inode, int mask)
220 unsigned int mode = inode->i_mode;
222 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
225 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
226 int error = check_acl(inode, mask);
227 if (error != -EAGAIN)
231 if (in_group_p(inode->i_gid))
236 * If the DACs are ok we don't need any capability check.
238 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
244 * generic_permission - check for access rights on a Posix-like filesystem
245 * @inode: inode to check access rights for
246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
248 * Used to check for read/write/execute permissions on a file.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
251 * are used for other things.
253 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
254 * request cannot be satisfied (eg. requires blocking or too much complexity).
255 * It would then be called again in ref-walk mode.
257 int generic_permission(struct inode *inode, int mask)
262 * Do the basic permission checks.
264 ret = acl_permission_check(inode, mask);
268 if (S_ISDIR(inode->i_mode)) {
269 /* DACs are overridable for directories */
270 if (inode_capable(inode, CAP_DAC_OVERRIDE))
272 if (!(mask & MAY_WRITE))
273 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
278 * Read/write DACs are always overridable.
279 * Executable DACs are overridable when there is
280 * at least one exec bit set.
282 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
283 if (inode_capable(inode, CAP_DAC_OVERRIDE))
287 * Searching includes executable on directories, else just read.
289 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
290 if (mask == MAY_READ)
291 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
298 * We _really_ want to just do "generic_permission()" without
299 * even looking at the inode->i_op values. So we keep a cache
300 * flag in inode->i_opflags, that says "this has not special
301 * permission function, use the fast case".
303 static inline int do_inode_permission(struct inode *inode, int mask)
305 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
306 if (likely(inode->i_op->permission))
307 return inode->i_op->permission(inode, mask);
309 /* This gets set once for the inode lifetime */
310 spin_lock(&inode->i_lock);
311 inode->i_opflags |= IOP_FASTPERM;
312 spin_unlock(&inode->i_lock);
314 return generic_permission(inode, mask);
318 * inode_permission - check for access rights to a given inode
319 * @inode: inode to check permission on
320 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
322 * Used to check for read/write/execute permissions on an inode.
323 * We use "fsuid" for this, letting us set arbitrary permissions
324 * for filesystem access without changing the "normal" uids which
325 * are used for other things.
327 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
329 int inode_permission(struct inode *inode, int mask)
333 if (unlikely(mask & MAY_WRITE)) {
334 umode_t mode = inode->i_mode;
337 * Nobody gets write access to a read-only fs.
339 if (IS_RDONLY(inode) &&
340 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
344 * Nobody gets write access to an immutable file.
346 if (IS_IMMUTABLE(inode))
350 retval = do_inode_permission(inode, mask);
354 retval = devcgroup_inode_permission(inode, mask);
358 return security_inode_permission(inode, mask);
362 * path_get - get a reference to a path
363 * @path: path to get the reference to
365 * Given a path increment the reference count to the dentry and the vfsmount.
367 void path_get(struct path *path)
372 EXPORT_SYMBOL(path_get);
375 * path_put - put a reference to a path
376 * @path: path to put the reference to
378 * Given a path decrement the reference count to the dentry and the vfsmount.
380 void path_put(struct path *path)
385 EXPORT_SYMBOL(path_put);
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). In situations when we can't
390 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
391 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
392 * mode. Refcounts are grabbed at the last known good point before rcu-walk
393 * got stuck, so ref-walk may continue from there. If this is not successful
394 * (eg. a seqcount has changed), then failure is returned and it's up to caller
395 * to restart the path walk from the beginning in ref-walk mode.
399 * unlazy_walk - try to switch to ref-walk mode.
400 * @nd: nameidata pathwalk data
401 * @dentry: child of nd->path.dentry or NULL
402 * Returns: 0 on success, -ECHILD on failure
404 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
405 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
406 * @nd or NULL. Must be called from rcu-walk context.
408 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
410 struct fs_struct *fs = current->fs;
411 struct dentry *parent = nd->path.dentry;
414 BUG_ON(!(nd->flags & LOOKUP_RCU));
415 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
417 spin_lock(&fs->lock);
418 if (nd->root.mnt != fs->root.mnt ||
419 nd->root.dentry != fs->root.dentry)
422 spin_lock(&parent->d_lock);
424 if (!__d_rcu_to_refcount(parent, nd->seq))
426 BUG_ON(nd->inode != parent->d_inode);
428 if (dentry->d_parent != parent)
430 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
431 if (!__d_rcu_to_refcount(dentry, nd->seq))
434 * If the sequence check on the child dentry passed, then
435 * the child has not been removed from its parent. This
436 * means the parent dentry must be valid and able to take
437 * a reference at this point.
439 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
440 BUG_ON(!parent->d_count);
442 spin_unlock(&dentry->d_lock);
444 spin_unlock(&parent->d_lock);
447 spin_unlock(&fs->lock);
449 mntget(nd->path.mnt);
452 br_read_unlock(&vfsmount_lock);
453 nd->flags &= ~LOOKUP_RCU;
457 spin_unlock(&dentry->d_lock);
459 spin_unlock(&parent->d_lock);
462 spin_unlock(&fs->lock);
466 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
468 return dentry->d_op->d_revalidate(dentry, flags);
472 * complete_walk - successful completion of path walk
473 * @nd: pointer nameidata
475 * If we had been in RCU mode, drop out of it and legitimize nd->path.
476 * Revalidate the final result, unless we'd already done that during
477 * the path walk or the filesystem doesn't ask for it. Return 0 on
478 * success, -error on failure. In case of failure caller does not
479 * need to drop nd->path.
481 static int complete_walk(struct nameidata *nd)
483 struct dentry *dentry = nd->path.dentry;
486 if (nd->flags & LOOKUP_RCU) {
487 nd->flags &= ~LOOKUP_RCU;
488 if (!(nd->flags & LOOKUP_ROOT))
490 spin_lock(&dentry->d_lock);
491 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
492 spin_unlock(&dentry->d_lock);
494 br_read_unlock(&vfsmount_lock);
497 BUG_ON(nd->inode != dentry->d_inode);
498 spin_unlock(&dentry->d_lock);
499 mntget(nd->path.mnt);
501 br_read_unlock(&vfsmount_lock);
504 if (likely(!(nd->flags & LOOKUP_JUMPED)))
507 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
510 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
513 /* Note: we do not d_invalidate() */
514 status = d_revalidate(dentry, nd->flags);
525 static __always_inline void set_root(struct nameidata *nd)
528 get_fs_root(current->fs, &nd->root);
531 static int link_path_walk(const char *, struct nameidata *);
533 static __always_inline void set_root_rcu(struct nameidata *nd)
536 struct fs_struct *fs = current->fs;
540 seq = read_seqcount_begin(&fs->seq);
542 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
543 } while (read_seqcount_retry(&fs->seq, seq));
547 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
559 nd->flags |= LOOKUP_JUMPED;
561 nd->inode = nd->path.dentry->d_inode;
563 ret = link_path_walk(link, nd);
567 return PTR_ERR(link);
570 static void path_put_conditional(struct path *path, struct nameidata *nd)
573 if (path->mnt != nd->path.mnt)
577 static inline void path_to_nameidata(const struct path *path,
578 struct nameidata *nd)
580 if (!(nd->flags & LOOKUP_RCU)) {
581 dput(nd->path.dentry);
582 if (nd->path.mnt != path->mnt)
583 mntput(nd->path.mnt);
585 nd->path.mnt = path->mnt;
586 nd->path.dentry = path->dentry;
589 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
591 struct inode *inode = link->dentry->d_inode;
592 if (inode->i_op->put_link)
593 inode->i_op->put_link(link->dentry, nd, cookie);
597 static __always_inline int
598 follow_link(struct path *link, struct nameidata *nd, void **p)
600 struct dentry *dentry = link->dentry;
604 BUG_ON(nd->flags & LOOKUP_RCU);
606 if (link->mnt == nd->path.mnt)
610 if (unlikely(current->total_link_count >= 40))
611 goto out_put_nd_path;
614 current->total_link_count++;
617 nd_set_link(nd, NULL);
619 error = security_inode_follow_link(link->dentry, nd);
621 goto out_put_nd_path;
623 nd->last_type = LAST_BIND;
624 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
632 error = __vfs_follow_link(nd, s);
633 } else if (nd->last_type == LAST_BIND) {
634 nd->flags |= LOOKUP_JUMPED;
635 nd->inode = nd->path.dentry->d_inode;
636 if (nd->inode->i_op->follow_link) {
637 /* stepped on a _really_ weird one */
643 put_link(nd, link, *p);
654 static int follow_up_rcu(struct path *path)
656 struct mount *mnt = real_mount(path->mnt);
657 struct mount *parent;
658 struct dentry *mountpoint;
660 parent = mnt->mnt_parent;
661 if (&parent->mnt == path->mnt)
663 mountpoint = mnt->mnt_mountpoint;
664 path->dentry = mountpoint;
665 path->mnt = &parent->mnt;
669 int follow_up(struct path *path)
671 struct mount *mnt = real_mount(path->mnt);
672 struct mount *parent;
673 struct dentry *mountpoint;
675 br_read_lock(&vfsmount_lock);
676 parent = mnt->mnt_parent;
677 if (&parent->mnt == path->mnt) {
678 br_read_unlock(&vfsmount_lock);
681 mntget(&parent->mnt);
682 mountpoint = dget(mnt->mnt_mountpoint);
683 br_read_unlock(&vfsmount_lock);
685 path->dentry = mountpoint;
687 path->mnt = &parent->mnt;
692 * Perform an automount
693 * - return -EISDIR to tell follow_managed() to stop and return the path we
696 static int follow_automount(struct path *path, unsigned flags,
699 struct vfsmount *mnt;
702 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
705 /* We don't want to mount if someone's just doing a stat -
706 * unless they're stat'ing a directory and appended a '/' to
709 * We do, however, want to mount if someone wants to open or
710 * create a file of any type under the mountpoint, wants to
711 * traverse through the mountpoint or wants to open the
712 * mounted directory. Also, autofs may mark negative dentries
713 * as being automount points. These will need the attentions
714 * of the daemon to instantiate them before they can be used.
716 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
717 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
718 path->dentry->d_inode)
721 current->total_link_count++;
722 if (current->total_link_count >= 40)
725 mnt = path->dentry->d_op->d_automount(path);
728 * The filesystem is allowed to return -EISDIR here to indicate
729 * it doesn't want to automount. For instance, autofs would do
730 * this so that its userspace daemon can mount on this dentry.
732 * However, we can only permit this if it's a terminal point in
733 * the path being looked up; if it wasn't then the remainder of
734 * the path is inaccessible and we should say so.
736 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
741 if (!mnt) /* mount collision */
745 /* lock_mount() may release path->mnt on error */
749 err = finish_automount(mnt, path);
753 /* Someone else made a mount here whilst we were busy */
758 path->dentry = dget(mnt->mnt_root);
767 * Handle a dentry that is managed in some way.
768 * - Flagged for transit management (autofs)
769 * - Flagged as mountpoint
770 * - Flagged as automount point
772 * This may only be called in refwalk mode.
774 * Serialization is taken care of in namespace.c
776 static int follow_managed(struct path *path, unsigned flags)
778 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
780 bool need_mntput = false;
783 /* Given that we're not holding a lock here, we retain the value in a
784 * local variable for each dentry as we look at it so that we don't see
785 * the components of that value change under us */
786 while (managed = ACCESS_ONCE(path->dentry->d_flags),
787 managed &= DCACHE_MANAGED_DENTRY,
788 unlikely(managed != 0)) {
789 /* Allow the filesystem to manage the transit without i_mutex
791 if (managed & DCACHE_MANAGE_TRANSIT) {
792 BUG_ON(!path->dentry->d_op);
793 BUG_ON(!path->dentry->d_op->d_manage);
794 ret = path->dentry->d_op->d_manage(path->dentry, false);
799 /* Transit to a mounted filesystem. */
800 if (managed & DCACHE_MOUNTED) {
801 struct vfsmount *mounted = lookup_mnt(path);
807 path->dentry = dget(mounted->mnt_root);
812 /* Something is mounted on this dentry in another
813 * namespace and/or whatever was mounted there in this
814 * namespace got unmounted before we managed to get the
818 /* Handle an automount point */
819 if (managed & DCACHE_NEED_AUTOMOUNT) {
820 ret = follow_automount(path, flags, &need_mntput);
826 /* We didn't change the current path point */
830 if (need_mntput && path->mnt == mnt)
834 return ret < 0 ? ret : need_mntput;
837 int follow_down_one(struct path *path)
839 struct vfsmount *mounted;
841 mounted = lookup_mnt(path);
846 path->dentry = dget(mounted->mnt_root);
852 static inline bool managed_dentry_might_block(struct dentry *dentry)
854 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
855 dentry->d_op->d_manage(dentry, true) < 0);
859 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
860 * we meet a managed dentry that would need blocking.
862 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
863 struct inode **inode)
866 struct mount *mounted;
868 * Don't forget we might have a non-mountpoint managed dentry
869 * that wants to block transit.
871 if (unlikely(managed_dentry_might_block(path->dentry)))
874 if (!d_mountpoint(path->dentry))
877 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
880 path->mnt = &mounted->mnt;
881 path->dentry = mounted->mnt.mnt_root;
882 nd->flags |= LOOKUP_JUMPED;
883 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
885 * Update the inode too. We don't need to re-check the
886 * dentry sequence number here after this d_inode read,
887 * because a mount-point is always pinned.
889 *inode = path->dentry->d_inode;
894 static void follow_mount_rcu(struct nameidata *nd)
896 while (d_mountpoint(nd->path.dentry)) {
897 struct mount *mounted;
898 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
901 nd->path.mnt = &mounted->mnt;
902 nd->path.dentry = mounted->mnt.mnt_root;
903 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
907 static int follow_dotdot_rcu(struct nameidata *nd)
912 if (nd->path.dentry == nd->root.dentry &&
913 nd->path.mnt == nd->root.mnt) {
916 if (nd->path.dentry != nd->path.mnt->mnt_root) {
917 struct dentry *old = nd->path.dentry;
918 struct dentry *parent = old->d_parent;
921 seq = read_seqcount_begin(&parent->d_seq);
922 if (read_seqcount_retry(&old->d_seq, nd->seq))
924 nd->path.dentry = parent;
928 if (!follow_up_rcu(&nd->path))
930 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
932 follow_mount_rcu(nd);
933 nd->inode = nd->path.dentry->d_inode;
937 nd->flags &= ~LOOKUP_RCU;
938 if (!(nd->flags & LOOKUP_ROOT))
941 br_read_unlock(&vfsmount_lock);
946 * Follow down to the covering mount currently visible to userspace. At each
947 * point, the filesystem owning that dentry may be queried as to whether the
948 * caller is permitted to proceed or not.
950 int follow_down(struct path *path)
955 while (managed = ACCESS_ONCE(path->dentry->d_flags),
956 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
957 /* Allow the filesystem to manage the transit without i_mutex
960 * We indicate to the filesystem if someone is trying to mount
961 * something here. This gives autofs the chance to deny anyone
962 * other than its daemon the right to mount on its
965 * The filesystem may sleep at this point.
967 if (managed & DCACHE_MANAGE_TRANSIT) {
968 BUG_ON(!path->dentry->d_op);
969 BUG_ON(!path->dentry->d_op->d_manage);
970 ret = path->dentry->d_op->d_manage(
971 path->dentry, false);
973 return ret == -EISDIR ? 0 : ret;
976 /* Transit to a mounted filesystem. */
977 if (managed & DCACHE_MOUNTED) {
978 struct vfsmount *mounted = lookup_mnt(path);
984 path->dentry = dget(mounted->mnt_root);
988 /* Don't handle automount points here */
995 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
997 static void follow_mount(struct path *path)
999 while (d_mountpoint(path->dentry)) {
1000 struct vfsmount *mounted = lookup_mnt(path);
1005 path->mnt = mounted;
1006 path->dentry = dget(mounted->mnt_root);
1010 static void follow_dotdot(struct nameidata *nd)
1015 struct dentry *old = nd->path.dentry;
1017 if (nd->path.dentry == nd->root.dentry &&
1018 nd->path.mnt == nd->root.mnt) {
1021 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1022 /* rare case of legitimate dget_parent()... */
1023 nd->path.dentry = dget_parent(nd->path.dentry);
1027 if (!follow_up(&nd->path))
1030 follow_mount(&nd->path);
1031 nd->inode = nd->path.dentry->d_inode;
1035 * This looks up the name in dcache, possibly revalidates the old dentry and
1036 * allocates a new one if not found or not valid. In the need_lookup argument
1037 * returns whether i_op->lookup is necessary.
1039 * dir->d_inode->i_mutex must be held
1041 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1042 unsigned int flags, bool *need_lookup)
1044 struct dentry *dentry;
1047 *need_lookup = false;
1048 dentry = d_lookup(dir, name);
1050 if (d_need_lookup(dentry)) {
1051 *need_lookup = true;
1052 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1053 error = d_revalidate(dentry, flags);
1054 if (unlikely(error <= 0)) {
1057 return ERR_PTR(error);
1058 } else if (!d_invalidate(dentry)) {
1067 dentry = d_alloc(dir, name);
1068 if (unlikely(!dentry))
1069 return ERR_PTR(-ENOMEM);
1071 *need_lookup = true;
1077 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1078 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1080 * dir->d_inode->i_mutex must be held
1082 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1087 /* Don't create child dentry for a dead directory. */
1088 if (unlikely(IS_DEADDIR(dir))) {
1090 return ERR_PTR(-ENOENT);
1093 old = dir->i_op->lookup(dir, dentry, flags);
1094 if (unlikely(old)) {
1101 static struct dentry *__lookup_hash(struct qstr *name,
1102 struct dentry *base, unsigned int flags)
1105 struct dentry *dentry;
1107 dentry = lookup_dcache(name, base, flags, &need_lookup);
1111 return lookup_real(base->d_inode, dentry, flags);
1115 * It's more convoluted than I'd like it to be, but... it's still fairly
1116 * small and for now I'd prefer to have fast path as straight as possible.
1117 * It _is_ time-critical.
1119 static int lookup_fast(struct nameidata *nd, struct qstr *name,
1120 struct path *path, struct inode **inode)
1122 struct vfsmount *mnt = nd->path.mnt;
1123 struct dentry *dentry, *parent = nd->path.dentry;
1129 * Rename seqlock is not required here because in the off chance
1130 * of a false negative due to a concurrent rename, we're going to
1131 * do the non-racy lookup, below.
1133 if (nd->flags & LOOKUP_RCU) {
1135 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
1140 * This sequence count validates that the inode matches
1141 * the dentry name information from lookup.
1143 *inode = dentry->d_inode;
1144 if (read_seqcount_retry(&dentry->d_seq, seq))
1148 * This sequence count validates that the parent had no
1149 * changes while we did the lookup of the dentry above.
1151 * The memory barrier in read_seqcount_begin of child is
1152 * enough, we can use __read_seqcount_retry here.
1154 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1158 if (unlikely(d_need_lookup(dentry)))
1160 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1161 status = d_revalidate(dentry, nd->flags);
1162 if (unlikely(status <= 0)) {
1163 if (status != -ECHILD)
1169 path->dentry = dentry;
1170 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1172 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1176 if (unlazy_walk(nd, dentry))
1179 dentry = __d_lookup(parent, name);
1182 if (unlikely(!dentry))
1185 if (unlikely(d_need_lookup(dentry))) {
1190 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1191 status = d_revalidate(dentry, nd->flags);
1192 if (unlikely(status <= 0)) {
1197 if (!d_invalidate(dentry)) {
1204 path->dentry = dentry;
1205 err = follow_managed(path, nd->flags);
1206 if (unlikely(err < 0)) {
1207 path_put_conditional(path, nd);
1211 nd->flags |= LOOKUP_JUMPED;
1212 *inode = path->dentry->d_inode;
1219 /* Fast lookup failed, do it the slow way */
1220 static int lookup_slow(struct nameidata *nd, struct qstr *name,
1223 struct dentry *dentry, *parent;
1226 parent = nd->path.dentry;
1227 BUG_ON(nd->inode != parent->d_inode);
1229 mutex_lock(&parent->d_inode->i_mutex);
1230 dentry = __lookup_hash(name, parent, nd->flags);
1231 mutex_unlock(&parent->d_inode->i_mutex);
1233 return PTR_ERR(dentry);
1234 path->mnt = nd->path.mnt;
1235 path->dentry = dentry;
1236 err = follow_managed(path, nd->flags);
1237 if (unlikely(err < 0)) {
1238 path_put_conditional(path, nd);
1242 nd->flags |= LOOKUP_JUMPED;
1246 static inline int may_lookup(struct nameidata *nd)
1248 if (nd->flags & LOOKUP_RCU) {
1249 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1252 if (unlazy_walk(nd, NULL))
1255 return inode_permission(nd->inode, MAY_EXEC);
1258 static inline int handle_dots(struct nameidata *nd, int type)
1260 if (type == LAST_DOTDOT) {
1261 if (nd->flags & LOOKUP_RCU) {
1262 if (follow_dotdot_rcu(nd))
1270 static void terminate_walk(struct nameidata *nd)
1272 if (!(nd->flags & LOOKUP_RCU)) {
1273 path_put(&nd->path);
1275 nd->flags &= ~LOOKUP_RCU;
1276 if (!(nd->flags & LOOKUP_ROOT))
1277 nd->root.mnt = NULL;
1279 br_read_unlock(&vfsmount_lock);
1284 * Do we need to follow links? We _really_ want to be able
1285 * to do this check without having to look at inode->i_op,
1286 * so we keep a cache of "no, this doesn't need follow_link"
1287 * for the common case.
1289 static inline int should_follow_link(struct inode *inode, int follow)
1291 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1292 if (likely(inode->i_op->follow_link))
1295 /* This gets set once for the inode lifetime */
1296 spin_lock(&inode->i_lock);
1297 inode->i_opflags |= IOP_NOFOLLOW;
1298 spin_unlock(&inode->i_lock);
1303 static inline int walk_component(struct nameidata *nd, struct path *path,
1304 struct qstr *name, int type, int follow)
1306 struct inode *inode;
1309 * "." and ".." are special - ".." especially so because it has
1310 * to be able to know about the current root directory and
1311 * parent relationships.
1313 if (unlikely(type != LAST_NORM))
1314 return handle_dots(nd, type);
1315 err = lookup_fast(nd, name, path, &inode);
1316 if (unlikely(err)) {
1320 err = lookup_slow(nd, name, path);
1324 inode = path->dentry->d_inode;
1330 if (should_follow_link(inode, follow)) {
1331 if (nd->flags & LOOKUP_RCU) {
1332 if (unlikely(unlazy_walk(nd, path->dentry))) {
1337 BUG_ON(inode != path->dentry->d_inode);
1340 path_to_nameidata(path, nd);
1345 path_to_nameidata(path, nd);
1352 * This limits recursive symlink follows to 8, while
1353 * limiting consecutive symlinks to 40.
1355 * Without that kind of total limit, nasty chains of consecutive
1356 * symlinks can cause almost arbitrarily long lookups.
1358 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1362 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1363 path_put_conditional(path, nd);
1364 path_put(&nd->path);
1367 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1370 current->link_count++;
1373 struct path link = *path;
1376 res = follow_link(&link, nd, &cookie);
1379 res = walk_component(nd, path, &nd->last,
1380 nd->last_type, LOOKUP_FOLLOW);
1381 put_link(nd, &link, cookie);
1384 current->link_count--;
1390 * We really don't want to look at inode->i_op->lookup
1391 * when we don't have to. So we keep a cache bit in
1392 * the inode ->i_opflags field that says "yes, we can
1393 * do lookup on this inode".
1395 static inline int can_lookup(struct inode *inode)
1397 if (likely(inode->i_opflags & IOP_LOOKUP))
1399 if (likely(!inode->i_op->lookup))
1402 /* We do this once for the lifetime of the inode */
1403 spin_lock(&inode->i_lock);
1404 inode->i_opflags |= IOP_LOOKUP;
1405 spin_unlock(&inode->i_lock);
1410 * We can do the critical dentry name comparison and hashing
1411 * operations one word at a time, but we are limited to:
1413 * - Architectures with fast unaligned word accesses. We could
1414 * do a "get_unaligned()" if this helps and is sufficiently
1417 * - Little-endian machines (so that we can generate the mask
1418 * of low bytes efficiently). Again, we *could* do a byte
1419 * swapping load on big-endian architectures if that is not
1420 * expensive enough to make the optimization worthless.
1422 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1423 * do not trap on the (extremely unlikely) case of a page
1424 * crossing operation.
1426 * - Furthermore, we need an efficient 64-bit compile for the
1427 * 64-bit case in order to generate the "number of bytes in
1428 * the final mask". Again, that could be replaced with a
1429 * efficient population count instruction or similar.
1431 #ifdef CONFIG_DCACHE_WORD_ACCESS
1433 #include <asm/word-at-a-time.h>
1437 static inline unsigned int fold_hash(unsigned long hash)
1439 hash += hash >> (8*sizeof(int));
1443 #else /* 32-bit case */
1445 #define fold_hash(x) (x)
1449 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1451 unsigned long a, mask;
1452 unsigned long hash = 0;
1455 a = load_unaligned_zeropad(name);
1456 if (len < sizeof(unsigned long))
1460 name += sizeof(unsigned long);
1461 len -= sizeof(unsigned long);
1465 mask = ~(~0ul << len*8);
1468 return fold_hash(hash);
1470 EXPORT_SYMBOL(full_name_hash);
1473 * Calculate the length and hash of the path component, and
1474 * return the length of the component;
1476 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1478 unsigned long a, b, adata, bdata, mask, hash, len;
1479 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1482 len = -sizeof(unsigned long);
1484 hash = (hash + a) * 9;
1485 len += sizeof(unsigned long);
1486 a = load_unaligned_zeropad(name+len);
1487 b = a ^ REPEAT_BYTE('/');
1488 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1490 adata = prep_zero_mask(a, adata, &constants);
1491 bdata = prep_zero_mask(b, bdata, &constants);
1493 mask = create_zero_mask(adata | bdata);
1495 hash += a & zero_bytemask(mask);
1496 *hashp = fold_hash(hash);
1498 return len + find_zero(mask);
1503 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1505 unsigned long hash = init_name_hash();
1507 hash = partial_name_hash(*name++, hash);
1508 return end_name_hash(hash);
1510 EXPORT_SYMBOL(full_name_hash);
1513 * We know there's a real path component here of at least
1516 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1518 unsigned long hash = init_name_hash();
1519 unsigned long len = 0, c;
1521 c = (unsigned char)*name;
1524 hash = partial_name_hash(c, hash);
1525 c = (unsigned char)name[len];
1526 } while (c && c != '/');
1527 *hashp = end_name_hash(hash);
1535 * This is the basic name resolution function, turning a pathname into
1536 * the final dentry. We expect 'base' to be positive and a directory.
1538 * Returns 0 and nd will have valid dentry and mnt on success.
1539 * Returns error and drops reference to input namei data on failure.
1541 static int link_path_walk(const char *name, struct nameidata *nd)
1551 /* At this point we know we have a real path component. */
1557 err = may_lookup(nd);
1561 len = hash_name(name, &this.hash);
1566 if (name[0] == '.') switch (len) {
1568 if (name[1] == '.') {
1570 nd->flags |= LOOKUP_JUMPED;
1576 if (likely(type == LAST_NORM)) {
1577 struct dentry *parent = nd->path.dentry;
1578 nd->flags &= ~LOOKUP_JUMPED;
1579 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1580 err = parent->d_op->d_hash(parent, nd->inode,
1588 goto last_component;
1590 * If it wasn't NUL, we know it was '/'. Skip that
1591 * slash, and continue until no more slashes.
1595 } while (unlikely(name[len] == '/'));
1597 goto last_component;
1600 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1605 err = nested_symlink(&next, nd);
1609 if (can_lookup(nd->inode))
1613 /* here ends the main loop */
1617 nd->last_type = type;
1624 static int path_init(int dfd, const char *name, unsigned int flags,
1625 struct nameidata *nd, struct file **fp)
1631 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1632 nd->flags = flags | LOOKUP_JUMPED;
1634 if (flags & LOOKUP_ROOT) {
1635 struct inode *inode = nd->root.dentry->d_inode;
1637 if (!inode->i_op->lookup)
1639 retval = inode_permission(inode, MAY_EXEC);
1643 nd->path = nd->root;
1645 if (flags & LOOKUP_RCU) {
1646 br_read_lock(&vfsmount_lock);
1648 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1650 path_get(&nd->path);
1655 nd->root.mnt = NULL;
1658 if (flags & LOOKUP_RCU) {
1659 br_read_lock(&vfsmount_lock);
1664 path_get(&nd->root);
1666 nd->path = nd->root;
1667 } else if (dfd == AT_FDCWD) {
1668 if (flags & LOOKUP_RCU) {
1669 struct fs_struct *fs = current->fs;
1672 br_read_lock(&vfsmount_lock);
1676 seq = read_seqcount_begin(&fs->seq);
1678 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1679 } while (read_seqcount_retry(&fs->seq, seq));
1681 get_fs_pwd(current->fs, &nd->path);
1684 struct dentry *dentry;
1686 file = fget_raw_light(dfd, &fput_needed);
1691 dentry = file->f_path.dentry;
1695 if (!S_ISDIR(dentry->d_inode->i_mode))
1698 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1703 nd->path = file->f_path;
1704 if (flags & LOOKUP_RCU) {
1707 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1708 br_read_lock(&vfsmount_lock);
1711 path_get(&file->f_path);
1712 fput_light(file, fput_needed);
1716 nd->inode = nd->path.dentry->d_inode;
1720 fput_light(file, fput_needed);
1725 static inline int lookup_last(struct nameidata *nd, struct path *path)
1727 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1728 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1730 nd->flags &= ~LOOKUP_PARENT;
1731 return walk_component(nd, path, &nd->last, nd->last_type,
1732 nd->flags & LOOKUP_FOLLOW);
1735 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1736 static int path_lookupat(int dfd, const char *name,
1737 unsigned int flags, struct nameidata *nd)
1739 struct file *base = NULL;
1744 * Path walking is largely split up into 2 different synchronisation
1745 * schemes, rcu-walk and ref-walk (explained in
1746 * Documentation/filesystems/path-lookup.txt). These share much of the
1747 * path walk code, but some things particularly setup, cleanup, and
1748 * following mounts are sufficiently divergent that functions are
1749 * duplicated. Typically there is a function foo(), and its RCU
1750 * analogue, foo_rcu().
1752 * -ECHILD is the error number of choice (just to avoid clashes) that
1753 * is returned if some aspect of an rcu-walk fails. Such an error must
1754 * be handled by restarting a traditional ref-walk (which will always
1755 * be able to complete).
1757 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1762 current->total_link_count = 0;
1763 err = link_path_walk(name, nd);
1765 if (!err && !(flags & LOOKUP_PARENT)) {
1766 err = lookup_last(nd, &path);
1769 struct path link = path;
1770 nd->flags |= LOOKUP_PARENT;
1771 err = follow_link(&link, nd, &cookie);
1774 err = lookup_last(nd, &path);
1775 put_link(nd, &link, cookie);
1780 err = complete_walk(nd);
1782 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1783 if (!nd->inode->i_op->lookup) {
1784 path_put(&nd->path);
1792 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1793 path_put(&nd->root);
1794 nd->root.mnt = NULL;
1799 static int do_path_lookup(int dfd, const char *name,
1800 unsigned int flags, struct nameidata *nd)
1802 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1803 if (unlikely(retval == -ECHILD))
1804 retval = path_lookupat(dfd, name, flags, nd);
1805 if (unlikely(retval == -ESTALE))
1806 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1808 if (likely(!retval)) {
1809 if (unlikely(!audit_dummy_context())) {
1810 if (nd->path.dentry && nd->inode)
1811 audit_inode(name, nd->path.dentry);
1817 /* does lookup, returns the object with parent locked */
1818 struct dentry *kern_path_locked(const char *name, struct path *path)
1820 struct nameidata nd;
1822 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
1824 return ERR_PTR(err);
1825 if (nd.last_type != LAST_NORM) {
1827 return ERR_PTR(-EINVAL);
1829 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1830 d = lookup_one_len(nd.last.name, nd.path.dentry, nd.last.len);
1832 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1840 int kern_path(const char *name, unsigned int flags, struct path *path)
1842 struct nameidata nd;
1843 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1850 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1851 * @dentry: pointer to dentry of the base directory
1852 * @mnt: pointer to vfs mount of the base directory
1853 * @name: pointer to file name
1854 * @flags: lookup flags
1855 * @path: pointer to struct path to fill
1857 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1858 const char *name, unsigned int flags,
1861 struct nameidata nd;
1863 nd.root.dentry = dentry;
1865 BUG_ON(flags & LOOKUP_PARENT);
1866 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1867 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1874 * Restricted form of lookup. Doesn't follow links, single-component only,
1875 * needs parent already locked. Doesn't follow mounts.
1878 static struct dentry *lookup_hash(struct nameidata *nd)
1880 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
1884 * lookup_one_len - filesystem helper to lookup single pathname component
1885 * @name: pathname component to lookup
1886 * @base: base directory to lookup from
1887 * @len: maximum length @len should be interpreted to
1889 * Note that this routine is purely a helper for filesystem usage and should
1890 * not be called by generic code. Also note that by using this function the
1891 * nameidata argument is passed to the filesystem methods and a filesystem
1892 * using this helper needs to be prepared for that.
1894 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1900 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1904 this.hash = full_name_hash(name, len);
1906 return ERR_PTR(-EACCES);
1909 c = *(const unsigned char *)name++;
1910 if (c == '/' || c == '\0')
1911 return ERR_PTR(-EACCES);
1914 * See if the low-level filesystem might want
1915 * to use its own hash..
1917 if (base->d_flags & DCACHE_OP_HASH) {
1918 int err = base->d_op->d_hash(base, base->d_inode, &this);
1920 return ERR_PTR(err);
1923 err = inode_permission(base->d_inode, MAY_EXEC);
1925 return ERR_PTR(err);
1927 return __lookup_hash(&this, base, 0);
1930 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1931 struct path *path, int *empty)
1933 struct nameidata nd;
1934 char *tmp = getname_flags(name, flags, empty);
1935 int err = PTR_ERR(tmp);
1938 BUG_ON(flags & LOOKUP_PARENT);
1940 err = do_path_lookup(dfd, tmp, flags, &nd);
1948 int user_path_at(int dfd, const char __user *name, unsigned flags,
1951 return user_path_at_empty(dfd, name, flags, path, NULL);
1954 static int user_path_parent(int dfd, const char __user *path,
1955 struct nameidata *nd, char **name)
1957 char *s = getname(path);
1963 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1973 * It's inline, so penalty for filesystems that don't use sticky bit is
1976 static inline int check_sticky(struct inode *dir, struct inode *inode)
1978 kuid_t fsuid = current_fsuid();
1980 if (!(dir->i_mode & S_ISVTX))
1982 if (uid_eq(inode->i_uid, fsuid))
1984 if (uid_eq(dir->i_uid, fsuid))
1986 return !inode_capable(inode, CAP_FOWNER);
1990 * Check whether we can remove a link victim from directory dir, check
1991 * whether the type of victim is right.
1992 * 1. We can't do it if dir is read-only (done in permission())
1993 * 2. We should have write and exec permissions on dir
1994 * 3. We can't remove anything from append-only dir
1995 * 4. We can't do anything with immutable dir (done in permission())
1996 * 5. If the sticky bit on dir is set we should either
1997 * a. be owner of dir, or
1998 * b. be owner of victim, or
1999 * c. have CAP_FOWNER capability
2000 * 6. If the victim is append-only or immutable we can't do antyhing with
2001 * links pointing to it.
2002 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2003 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2004 * 9. We can't remove a root or mountpoint.
2005 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2006 * nfs_async_unlink().
2008 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2012 if (!victim->d_inode)
2015 BUG_ON(victim->d_parent->d_inode != dir);
2016 audit_inode_child(victim, dir);
2018 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2023 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2024 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2027 if (!S_ISDIR(victim->d_inode->i_mode))
2029 if (IS_ROOT(victim))
2031 } else if (S_ISDIR(victim->d_inode->i_mode))
2033 if (IS_DEADDIR(dir))
2035 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2040 /* Check whether we can create an object with dentry child in directory
2042 * 1. We can't do it if child already exists (open has special treatment for
2043 * this case, but since we are inlined it's OK)
2044 * 2. We can't do it if dir is read-only (done in permission())
2045 * 3. We should have write and exec permissions on dir
2046 * 4. We can't do it if dir is immutable (done in permission())
2048 static inline int may_create(struct inode *dir, struct dentry *child)
2052 if (IS_DEADDIR(dir))
2054 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2058 * p1 and p2 should be directories on the same fs.
2060 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2065 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2069 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2071 p = d_ancestor(p2, p1);
2073 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2074 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2078 p = d_ancestor(p1, p2);
2080 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2081 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2085 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2086 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2090 void unlock_rename(struct dentry *p1, struct dentry *p2)
2092 mutex_unlock(&p1->d_inode->i_mutex);
2094 mutex_unlock(&p2->d_inode->i_mutex);
2095 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2099 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2102 int error = may_create(dir, dentry);
2106 if (!dir->i_op->create)
2107 return -EACCES; /* shouldn't it be ENOSYS? */
2110 error = security_inode_create(dir, dentry, mode);
2113 error = dir->i_op->create(dir, dentry, mode, want_excl);
2115 fsnotify_create(dir, dentry);
2119 static int may_open(struct path *path, int acc_mode, int flag)
2121 struct dentry *dentry = path->dentry;
2122 struct inode *inode = dentry->d_inode;
2132 switch (inode->i_mode & S_IFMT) {
2136 if (acc_mode & MAY_WRITE)
2141 if (path->mnt->mnt_flags & MNT_NODEV)
2150 error = inode_permission(inode, acc_mode);
2155 * An append-only file must be opened in append mode for writing.
2157 if (IS_APPEND(inode)) {
2158 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2164 /* O_NOATIME can only be set by the owner or superuser */
2165 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2171 static int handle_truncate(struct file *filp)
2173 struct path *path = &filp->f_path;
2174 struct inode *inode = path->dentry->d_inode;
2175 int error = get_write_access(inode);
2179 * Refuse to truncate files with mandatory locks held on them.
2181 error = locks_verify_locked(inode);
2183 error = security_path_truncate(path);
2185 error = do_truncate(path->dentry, 0,
2186 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2189 put_write_access(inode);
2193 static inline int open_to_namei_flags(int flag)
2195 if ((flag & O_ACCMODE) == 3)
2200 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2202 int error = security_path_mknod(dir, dentry, mode, 0);
2206 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2210 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2214 * Attempt to atomically look up, create and open a file from a negative
2217 * Returns 0 if successful. The file will have been created and attached to
2218 * @file by the filesystem calling finish_open().
2220 * Returns 1 if the file was looked up only or didn't need creating. The
2221 * caller will need to perform the open themselves. @path will have been
2222 * updated to point to the new dentry. This may be negative.
2224 * Returns an error code otherwise.
2226 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2227 struct path *path, struct file *file,
2228 const struct open_flags *op,
2229 bool *want_write, bool need_lookup,
2232 struct inode *dir = nd->path.dentry->d_inode;
2233 unsigned open_flag = open_to_namei_flags(op->open_flag);
2237 int create_error = 0;
2238 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2240 BUG_ON(dentry->d_inode);
2242 /* Don't create child dentry for a dead directory. */
2243 if (unlikely(IS_DEADDIR(dir))) {
2248 mode = op->mode & S_IALLUGO;
2249 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2250 mode &= ~current_umask();
2252 if (open_flag & O_EXCL) {
2253 open_flag &= ~O_TRUNC;
2254 *opened |= FILE_CREATED;
2258 * Checking write permission is tricky, bacuse we don't know if we are
2259 * going to actually need it: O_CREAT opens should work as long as the
2260 * file exists. But checking existence breaks atomicity. The trick is
2261 * to check access and if not granted clear O_CREAT from the flags.
2263 * Another problem is returing the "right" error value (e.g. for an
2264 * O_EXCL open we want to return EEXIST not EROFS).
2266 if ((open_flag & (O_CREAT | O_TRUNC)) ||
2267 (open_flag & O_ACCMODE) != O_RDONLY) {
2268 error = mnt_want_write(nd->path.mnt);
2271 } else if (!(open_flag & O_CREAT)) {
2273 * No O_CREATE -> atomicity not a requirement -> fall
2274 * back to lookup + open
2277 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2278 /* Fall back and fail with the right error */
2279 create_error = error;
2282 /* No side effects, safe to clear O_CREAT */
2283 create_error = error;
2284 open_flag &= ~O_CREAT;
2288 if (open_flag & O_CREAT) {
2289 error = may_o_create(&nd->path, dentry, op->mode);
2291 create_error = error;
2292 if (open_flag & O_EXCL)
2294 open_flag &= ~O_CREAT;
2298 if (nd->flags & LOOKUP_DIRECTORY)
2299 open_flag |= O_DIRECTORY;
2301 file->f_path.dentry = DENTRY_NOT_SET;
2302 file->f_path.mnt = nd->path.mnt;
2303 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2306 if (create_error && error == -ENOENT)
2307 error = create_error;
2311 acc_mode = op->acc_mode;
2312 if (*opened & FILE_CREATED) {
2313 fsnotify_create(dir, dentry);
2314 acc_mode = MAY_OPEN;
2317 if (error) { /* returned 1, that is */
2318 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2322 if (file->f_path.dentry) {
2324 dentry = file->f_path.dentry;
2330 * We didn't have the inode before the open, so check open permission
2333 error = may_open(&file->f_path, acc_mode, open_flag);
2343 dentry = lookup_real(dir, dentry, nd->flags);
2345 return PTR_ERR(dentry);
2348 int open_flag = op->open_flag;
2350 error = create_error;
2351 if ((open_flag & O_EXCL)) {
2352 if (!dentry->d_inode)
2354 } else if (!dentry->d_inode) {
2356 } else if ((open_flag & O_TRUNC) &&
2357 S_ISREG(dentry->d_inode->i_mode)) {
2360 /* will fail later, go on to get the right error */
2364 path->dentry = dentry;
2365 path->mnt = nd->path.mnt;
2370 * Look up and maybe create and open the last component.
2372 * Must be called with i_mutex held on parent.
2374 * Returns 0 if the file was successfully atomically created (if necessary) and
2375 * opened. In this case the file will be returned attached to @file.
2377 * Returns 1 if the file was not completely opened at this time, though lookups
2378 * and creations will have been performed and the dentry returned in @path will
2379 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2380 * specified then a negative dentry may be returned.
2382 * An error code is returned otherwise.
2384 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2385 * cleared otherwise prior to returning.
2387 static int lookup_open(struct nameidata *nd, struct path *path,
2389 const struct open_flags *op,
2390 bool *want_write, int *opened)
2392 struct dentry *dir = nd->path.dentry;
2393 struct inode *dir_inode = dir->d_inode;
2394 struct dentry *dentry;
2398 *opened &= ~FILE_CREATED;
2399 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2401 return PTR_ERR(dentry);
2403 /* Cached positive dentry: will open in f_op->open */
2404 if (!need_lookup && dentry->d_inode)
2407 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2408 return atomic_open(nd, dentry, path, file, op, want_write,
2409 need_lookup, opened);
2413 BUG_ON(dentry->d_inode);
2415 dentry = lookup_real(dir_inode, dentry, nd->flags);
2417 return PTR_ERR(dentry);
2420 /* Negative dentry, just create the file */
2421 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2422 umode_t mode = op->mode;
2423 if (!IS_POSIXACL(dir->d_inode))
2424 mode &= ~current_umask();
2426 * This write is needed to ensure that a
2427 * rw->ro transition does not occur between
2428 * the time when the file is created and when
2429 * a permanent write count is taken through
2430 * the 'struct file' in finish_open().
2432 error = mnt_want_write(nd->path.mnt);
2436 *opened |= FILE_CREATED;
2437 error = security_path_mknod(&nd->path, dentry, mode, 0);
2440 error = vfs_create(dir->d_inode, dentry, mode,
2441 nd->flags & LOOKUP_EXCL);
2446 path->dentry = dentry;
2447 path->mnt = nd->path.mnt;
2456 * Handle the last step of open()
2458 static int do_last(struct nameidata *nd, struct path *path,
2459 struct file *file, const struct open_flags *op,
2460 int *opened, const char *pathname)
2462 struct dentry *dir = nd->path.dentry;
2463 int open_flag = op->open_flag;
2464 bool will_truncate = (open_flag & O_TRUNC) != 0;
2465 bool want_write = false;
2466 int acc_mode = op->acc_mode;
2467 struct inode *inode;
2468 bool symlink_ok = false;
2469 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2470 bool retried = false;
2473 nd->flags &= ~LOOKUP_PARENT;
2474 nd->flags |= op->intent;
2476 switch (nd->last_type) {
2479 error = handle_dots(nd, nd->last_type);
2484 error = complete_walk(nd);
2487 audit_inode(pathname, nd->path.dentry);
2488 if (open_flag & O_CREAT) {
2494 error = complete_walk(nd);
2497 audit_inode(pathname, dir);
2501 if (!(open_flag & O_CREAT)) {
2502 if (nd->last.name[nd->last.len])
2503 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2504 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2506 /* we _can_ be in RCU mode here */
2507 error = lookup_fast(nd, &nd->last, path, &inode);
2514 BUG_ON(nd->inode != dir->d_inode);
2516 /* create side of things */
2518 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2519 * has been cleared when we got to the last component we are
2522 error = complete_walk(nd);
2526 audit_inode(pathname, dir);
2528 /* trailing slashes? */
2529 if (nd->last.name[nd->last.len])
2534 mutex_lock(&dir->d_inode->i_mutex);
2535 error = lookup_open(nd, path, file, op, &want_write, opened);
2536 mutex_unlock(&dir->d_inode->i_mutex);
2542 if ((*opened & FILE_CREATED) ||
2543 !S_ISREG(file->f_path.dentry->d_inode->i_mode))
2544 will_truncate = false;
2546 audit_inode(pathname, file->f_path.dentry);
2550 if (*opened & FILE_CREATED) {
2551 /* Don't check for write permission, don't truncate */
2552 open_flag &= ~O_TRUNC;
2553 will_truncate = false;
2554 acc_mode = MAY_OPEN;
2555 path_to_nameidata(path, nd);
2556 goto finish_open_created;
2560 * It already exists.
2562 audit_inode(pathname, path->dentry);
2565 * If atomic_open() acquired write access it is dropped now due to
2566 * possible mount and symlink following (this might be optimized away if
2570 mnt_drop_write(nd->path.mnt);
2575 if (open_flag & O_EXCL)
2578 error = follow_managed(path, nd->flags);
2583 nd->flags |= LOOKUP_JUMPED;
2585 BUG_ON(nd->flags & LOOKUP_RCU);
2586 inode = path->dentry->d_inode;
2588 /* we _can_ be in RCU mode here */
2591 path_to_nameidata(path, nd);
2595 if (should_follow_link(inode, !symlink_ok)) {
2596 if (nd->flags & LOOKUP_RCU) {
2597 if (unlikely(unlazy_walk(nd, path->dentry))) {
2602 BUG_ON(inode != path->dentry->d_inode);
2606 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2607 path_to_nameidata(path, nd);
2609 save_parent.dentry = nd->path.dentry;
2610 save_parent.mnt = mntget(path->mnt);
2611 nd->path.dentry = path->dentry;
2615 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2616 error = complete_walk(nd);
2618 path_put(&save_parent);
2622 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
2625 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
2627 audit_inode(pathname, nd->path.dentry);
2629 if (!S_ISREG(nd->inode->i_mode))
2630 will_truncate = false;
2632 if (will_truncate) {
2633 error = mnt_want_write(nd->path.mnt);
2638 finish_open_created:
2639 error = may_open(&nd->path, acc_mode, open_flag);
2642 file->f_path.mnt = nd->path.mnt;
2643 error = finish_open(file, nd->path.dentry, NULL, opened);
2645 if (error == -EOPENSTALE)
2650 error = open_check_o_direct(file);
2653 error = ima_file_check(file, op->acc_mode);
2657 if (will_truncate) {
2658 error = handle_truncate(file);
2664 mnt_drop_write(nd->path.mnt);
2665 path_put(&save_parent);
2670 path_put_conditional(path, nd);
2677 /* If no saved parent or already retried then can't retry */
2678 if (!save_parent.dentry || retried)
2681 BUG_ON(save_parent.dentry != dir);
2682 path_put(&nd->path);
2683 nd->path = save_parent;
2684 nd->inode = dir->d_inode;
2685 save_parent.mnt = NULL;
2686 save_parent.dentry = NULL;
2688 mnt_drop_write(nd->path.mnt);
2695 static struct file *path_openat(int dfd, const char *pathname,
2696 struct nameidata *nd, const struct open_flags *op, int flags)
2698 struct file *base = NULL;
2704 file = get_empty_filp();
2706 return ERR_PTR(-ENFILE);
2708 file->f_flags = op->open_flag;
2710 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2711 if (unlikely(error))
2714 current->total_link_count = 0;
2715 error = link_path_walk(pathname, nd);
2716 if (unlikely(error))
2719 error = do_last(nd, &path, file, op, &opened, pathname);
2720 while (unlikely(error > 0)) { /* trailing symlink */
2721 struct path link = path;
2723 if (!(nd->flags & LOOKUP_FOLLOW)) {
2724 path_put_conditional(&path, nd);
2725 path_put(&nd->path);
2729 nd->flags |= LOOKUP_PARENT;
2730 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2731 error = follow_link(&link, nd, &cookie);
2732 if (unlikely(error))
2734 error = do_last(nd, &path, file, op, &opened, pathname);
2735 put_link(nd, &link, cookie);
2738 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2739 path_put(&nd->root);
2742 if (!(opened & FILE_OPENED)) {
2746 if (unlikely(error)) {
2747 if (error == -EOPENSTALE) {
2748 if (flags & LOOKUP_RCU)
2753 file = ERR_PTR(error);
2758 struct file *do_filp_open(int dfd, const char *pathname,
2759 const struct open_flags *op, int flags)
2761 struct nameidata nd;
2764 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2765 if (unlikely(filp == ERR_PTR(-ECHILD)))
2766 filp = path_openat(dfd, pathname, &nd, op, flags);
2767 if (unlikely(filp == ERR_PTR(-ESTALE)))
2768 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2772 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2773 const char *name, const struct open_flags *op, int flags)
2775 struct nameidata nd;
2779 nd.root.dentry = dentry;
2781 flags |= LOOKUP_ROOT;
2783 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2784 return ERR_PTR(-ELOOP);
2786 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2787 if (unlikely(file == ERR_PTR(-ECHILD)))
2788 file = path_openat(-1, name, &nd, op, flags);
2789 if (unlikely(file == ERR_PTR(-ESTALE)))
2790 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2794 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2796 struct dentry *dentry = ERR_PTR(-EEXIST);
2797 struct nameidata nd;
2798 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2800 return ERR_PTR(error);
2803 * Yucky last component or no last component at all?
2804 * (foo/., foo/.., /////)
2806 if (nd.last_type != LAST_NORM)
2808 nd.flags &= ~LOOKUP_PARENT;
2809 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2812 * Do the final lookup.
2814 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2815 dentry = lookup_hash(&nd);
2819 if (dentry->d_inode)
2822 * Special case - lookup gave negative, but... we had foo/bar/
2823 * From the vfs_mknod() POV we just have a negative dentry -
2824 * all is fine. Let's be bastards - you had / on the end, you've
2825 * been asking for (non-existent) directory. -ENOENT for you.
2827 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2829 dentry = ERR_PTR(-ENOENT);
2836 dentry = ERR_PTR(-EEXIST);
2838 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2843 EXPORT_SYMBOL(kern_path_create);
2845 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2847 char *tmp = getname(pathname);
2850 return ERR_CAST(tmp);
2851 res = kern_path_create(dfd, tmp, path, is_dir);
2855 EXPORT_SYMBOL(user_path_create);
2857 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2859 int error = may_create(dir, dentry);
2864 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2867 if (!dir->i_op->mknod)
2870 error = devcgroup_inode_mknod(mode, dev);
2874 error = security_inode_mknod(dir, dentry, mode, dev);
2878 error = dir->i_op->mknod(dir, dentry, mode, dev);
2880 fsnotify_create(dir, dentry);
2884 static int may_mknod(umode_t mode)
2886 switch (mode & S_IFMT) {
2892 case 0: /* zero mode translates to S_IFREG */
2901 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2904 struct dentry *dentry;
2911 dentry = user_path_create(dfd, filename, &path, 0);
2913 return PTR_ERR(dentry);
2915 if (!IS_POSIXACL(path.dentry->d_inode))
2916 mode &= ~current_umask();
2917 error = may_mknod(mode);
2920 error = mnt_want_write(path.mnt);
2923 error = security_path_mknod(&path, dentry, mode, dev);
2925 goto out_drop_write;
2926 switch (mode & S_IFMT) {
2927 case 0: case S_IFREG:
2928 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
2930 case S_IFCHR: case S_IFBLK:
2931 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2932 new_decode_dev(dev));
2934 case S_IFIFO: case S_IFSOCK:
2935 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2939 mnt_drop_write(path.mnt);
2942 mutex_unlock(&path.dentry->d_inode->i_mutex);
2948 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
2950 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2953 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2955 int error = may_create(dir, dentry);
2956 unsigned max_links = dir->i_sb->s_max_links;
2961 if (!dir->i_op->mkdir)
2964 mode &= (S_IRWXUGO|S_ISVTX);
2965 error = security_inode_mkdir(dir, dentry, mode);
2969 if (max_links && dir->i_nlink >= max_links)
2972 error = dir->i_op->mkdir(dir, dentry, mode);
2974 fsnotify_mkdir(dir, dentry);
2978 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
2980 struct dentry *dentry;
2984 dentry = user_path_create(dfd, pathname, &path, 1);
2986 return PTR_ERR(dentry);
2988 if (!IS_POSIXACL(path.dentry->d_inode))
2989 mode &= ~current_umask();
2990 error = mnt_want_write(path.mnt);
2993 error = security_path_mkdir(&path, dentry, mode);
2995 goto out_drop_write;
2996 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2998 mnt_drop_write(path.mnt);
3001 mutex_unlock(&path.dentry->d_inode->i_mutex);
3006 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3008 return sys_mkdirat(AT_FDCWD, pathname, mode);
3012 * The dentry_unhash() helper will try to drop the dentry early: we
3013 * should have a usage count of 1 if we're the only user of this
3014 * dentry, and if that is true (possibly after pruning the dcache),
3015 * then we drop the dentry now.
3017 * A low-level filesystem can, if it choses, legally
3020 * if (!d_unhashed(dentry))
3023 * if it cannot handle the case of removing a directory
3024 * that is still in use by something else..
3026 void dentry_unhash(struct dentry *dentry)
3028 shrink_dcache_parent(dentry);
3029 spin_lock(&dentry->d_lock);
3030 if (dentry->d_count == 1)
3032 spin_unlock(&dentry->d_lock);
3035 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3037 int error = may_delete(dir, dentry, 1);
3042 if (!dir->i_op->rmdir)
3046 mutex_lock(&dentry->d_inode->i_mutex);
3049 if (d_mountpoint(dentry))
3052 error = security_inode_rmdir(dir, dentry);
3056 shrink_dcache_parent(dentry);
3057 error = dir->i_op->rmdir(dir, dentry);
3061 dentry->d_inode->i_flags |= S_DEAD;
3065 mutex_unlock(&dentry->d_inode->i_mutex);
3072 static long do_rmdir(int dfd, const char __user *pathname)
3076 struct dentry *dentry;
3077 struct nameidata nd;
3079 error = user_path_parent(dfd, pathname, &nd, &name);
3083 switch(nd.last_type) {
3095 nd.flags &= ~LOOKUP_PARENT;
3097 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3098 dentry = lookup_hash(&nd);
3099 error = PTR_ERR(dentry);
3102 if (!dentry->d_inode) {
3106 error = mnt_want_write(nd.path.mnt);
3109 error = security_path_rmdir(&nd.path, dentry);
3112 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3114 mnt_drop_write(nd.path.mnt);
3118 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3125 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3127 return do_rmdir(AT_FDCWD, pathname);
3130 int vfs_unlink(struct inode *dir, struct dentry *dentry)
3132 int error = may_delete(dir, dentry, 0);
3137 if (!dir->i_op->unlink)
3140 mutex_lock(&dentry->d_inode->i_mutex);
3141 if (d_mountpoint(dentry))
3144 error = security_inode_unlink(dir, dentry);
3146 error = dir->i_op->unlink(dir, dentry);
3151 mutex_unlock(&dentry->d_inode->i_mutex);
3153 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3154 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3155 fsnotify_link_count(dentry->d_inode);
3163 * Make sure that the actual truncation of the file will occur outside its
3164 * directory's i_mutex. Truncate can take a long time if there is a lot of
3165 * writeout happening, and we don't want to prevent access to the directory
3166 * while waiting on the I/O.
3168 static long do_unlinkat(int dfd, const char __user *pathname)
3172 struct dentry *dentry;
3173 struct nameidata nd;
3174 struct inode *inode = NULL;
3176 error = user_path_parent(dfd, pathname, &nd, &name);
3181 if (nd.last_type != LAST_NORM)
3184 nd.flags &= ~LOOKUP_PARENT;
3186 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3187 dentry = lookup_hash(&nd);
3188 error = PTR_ERR(dentry);
3189 if (!IS_ERR(dentry)) {
3190 /* Why not before? Because we want correct error value */
3191 if (nd.last.name[nd.last.len])
3193 inode = dentry->d_inode;
3197 error = mnt_want_write(nd.path.mnt);
3200 error = security_path_unlink(&nd.path, dentry);
3203 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
3205 mnt_drop_write(nd.path.mnt);
3209 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3211 iput(inode); /* truncate the inode here */
3218 error = !dentry->d_inode ? -ENOENT :
3219 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3223 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3225 if ((flag & ~AT_REMOVEDIR) != 0)
3228 if (flag & AT_REMOVEDIR)
3229 return do_rmdir(dfd, pathname);
3231 return do_unlinkat(dfd, pathname);
3234 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3236 return do_unlinkat(AT_FDCWD, pathname);
3239 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3241 int error = may_create(dir, dentry);
3246 if (!dir->i_op->symlink)
3249 error = security_inode_symlink(dir, dentry, oldname);
3253 error = dir->i_op->symlink(dir, dentry, oldname);
3255 fsnotify_create(dir, dentry);
3259 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3260 int, newdfd, const char __user *, newname)
3264 struct dentry *dentry;
3267 from = getname(oldname);
3269 return PTR_ERR(from);
3271 dentry = user_path_create(newdfd, newname, &path, 0);
3272 error = PTR_ERR(dentry);
3276 error = mnt_want_write(path.mnt);
3279 error = security_path_symlink(&path, dentry, from);
3281 goto out_drop_write;
3282 error = vfs_symlink(path.dentry->d_inode, dentry, from);
3284 mnt_drop_write(path.mnt);
3287 mutex_unlock(&path.dentry->d_inode->i_mutex);
3294 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3296 return sys_symlinkat(oldname, AT_FDCWD, newname);
3299 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3301 struct inode *inode = old_dentry->d_inode;
3302 unsigned max_links = dir->i_sb->s_max_links;
3308 error = may_create(dir, new_dentry);
3312 if (dir->i_sb != inode->i_sb)
3316 * A link to an append-only or immutable file cannot be created.
3318 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3320 if (!dir->i_op->link)
3322 if (S_ISDIR(inode->i_mode))
3325 error = security_inode_link(old_dentry, dir, new_dentry);
3329 mutex_lock(&inode->i_mutex);
3330 /* Make sure we don't allow creating hardlink to an unlinked file */
3331 if (inode->i_nlink == 0)
3333 else if (max_links && inode->i_nlink >= max_links)
3336 error = dir->i_op->link(old_dentry, dir, new_dentry);
3337 mutex_unlock(&inode->i_mutex);
3339 fsnotify_link(dir, inode, new_dentry);
3344 * Hardlinks are often used in delicate situations. We avoid
3345 * security-related surprises by not following symlinks on the
3348 * We don't follow them on the oldname either to be compatible
3349 * with linux 2.0, and to avoid hard-linking to directories
3350 * and other special files. --ADM
3352 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3353 int, newdfd, const char __user *, newname, int, flags)
3355 struct dentry *new_dentry;
3356 struct path old_path, new_path;
3360 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3363 * To use null names we require CAP_DAC_READ_SEARCH
3364 * This ensures that not everyone will be able to create
3365 * handlink using the passed filedescriptor.
3367 if (flags & AT_EMPTY_PATH) {
3368 if (!capable(CAP_DAC_READ_SEARCH))
3373 if (flags & AT_SYMLINK_FOLLOW)
3374 how |= LOOKUP_FOLLOW;
3376 error = user_path_at(olddfd, oldname, how, &old_path);
3380 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3381 error = PTR_ERR(new_dentry);
3382 if (IS_ERR(new_dentry))
3386 if (old_path.mnt != new_path.mnt)
3388 error = mnt_want_write(new_path.mnt);
3391 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3393 goto out_drop_write;
3394 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3396 mnt_drop_write(new_path.mnt);
3399 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3400 path_put(&new_path);
3402 path_put(&old_path);
3407 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3409 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3413 * The worst of all namespace operations - renaming directory. "Perverted"
3414 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3416 * a) we can get into loop creation. Check is done in is_subdir().
3417 * b) race potential - two innocent renames can create a loop together.
3418 * That's where 4.4 screws up. Current fix: serialization on
3419 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3421 * c) we have to lock _three_ objects - parents and victim (if it exists).
3422 * And that - after we got ->i_mutex on parents (until then we don't know
3423 * whether the target exists). Solution: try to be smart with locking
3424 * order for inodes. We rely on the fact that tree topology may change
3425 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3426 * move will be locked. Thus we can rank directories by the tree
3427 * (ancestors first) and rank all non-directories after them.
3428 * That works since everybody except rename does "lock parent, lookup,
3429 * lock child" and rename is under ->s_vfs_rename_mutex.
3430 * HOWEVER, it relies on the assumption that any object with ->lookup()
3431 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3432 * we'd better make sure that there's no link(2) for them.
3433 * d) conversion from fhandle to dentry may come in the wrong moment - when
3434 * we are removing the target. Solution: we will have to grab ->i_mutex
3435 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3436 * ->i_mutex on parents, which works but leads to some truly excessive
3439 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3440 struct inode *new_dir, struct dentry *new_dentry)
3443 struct inode *target = new_dentry->d_inode;
3444 unsigned max_links = new_dir->i_sb->s_max_links;
3447 * If we are going to change the parent - check write permissions,
3448 * we'll need to flip '..'.
3450 if (new_dir != old_dir) {
3451 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3456 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3462 mutex_lock(&target->i_mutex);
3465 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3469 if (max_links && !target && new_dir != old_dir &&
3470 new_dir->i_nlink >= max_links)
3474 shrink_dcache_parent(new_dentry);
3475 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3480 target->i_flags |= S_DEAD;
3481 dont_mount(new_dentry);
3485 mutex_unlock(&target->i_mutex);
3488 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3489 d_move(old_dentry,new_dentry);
3493 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3494 struct inode *new_dir, struct dentry *new_dentry)
3496 struct inode *target = new_dentry->d_inode;
3499 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3505 mutex_lock(&target->i_mutex);
3508 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3511 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3516 dont_mount(new_dentry);
3517 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3518 d_move(old_dentry, new_dentry);
3521 mutex_unlock(&target->i_mutex);
3526 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3527 struct inode *new_dir, struct dentry *new_dentry)
3530 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3531 const unsigned char *old_name;
3533 if (old_dentry->d_inode == new_dentry->d_inode)
3536 error = may_delete(old_dir, old_dentry, is_dir);
3540 if (!new_dentry->d_inode)
3541 error = may_create(new_dir, new_dentry);
3543 error = may_delete(new_dir, new_dentry, is_dir);
3547 if (!old_dir->i_op->rename)
3550 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3553 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3555 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3557 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3558 new_dentry->d_inode, old_dentry);
3559 fsnotify_oldname_free(old_name);
3564 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3565 int, newdfd, const char __user *, newname)
3567 struct dentry *old_dir, *new_dir;
3568 struct dentry *old_dentry, *new_dentry;
3569 struct dentry *trap;
3570 struct nameidata oldnd, newnd;
3575 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3579 error = user_path_parent(newdfd, newname, &newnd, &to);
3584 if (oldnd.path.mnt != newnd.path.mnt)
3587 old_dir = oldnd.path.dentry;
3589 if (oldnd.last_type != LAST_NORM)
3592 new_dir = newnd.path.dentry;
3593 if (newnd.last_type != LAST_NORM)
3596 oldnd.flags &= ~LOOKUP_PARENT;
3597 newnd.flags &= ~LOOKUP_PARENT;
3598 newnd.flags |= LOOKUP_RENAME_TARGET;
3600 trap = lock_rename(new_dir, old_dir);
3602 old_dentry = lookup_hash(&oldnd);
3603 error = PTR_ERR(old_dentry);
3604 if (IS_ERR(old_dentry))
3606 /* source must exist */
3608 if (!old_dentry->d_inode)
3610 /* unless the source is a directory trailing slashes give -ENOTDIR */
3611 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3613 if (oldnd.last.name[oldnd.last.len])
3615 if (newnd.last.name[newnd.last.len])
3618 /* source should not be ancestor of target */
3620 if (old_dentry == trap)
3622 new_dentry = lookup_hash(&newnd);
3623 error = PTR_ERR(new_dentry);
3624 if (IS_ERR(new_dentry))
3626 /* target should not be an ancestor of source */
3628 if (new_dentry == trap)
3631 error = mnt_want_write(oldnd.path.mnt);
3634 error = security_path_rename(&oldnd.path, old_dentry,
3635 &newnd.path, new_dentry);
3638 error = vfs_rename(old_dir->d_inode, old_dentry,
3639 new_dir->d_inode, new_dentry);
3641 mnt_drop_write(oldnd.path.mnt);
3647 unlock_rename(new_dir, old_dir);
3649 path_put(&newnd.path);
3652 path_put(&oldnd.path);
3658 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3660 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3663 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3667 len = PTR_ERR(link);
3672 if (len > (unsigned) buflen)
3674 if (copy_to_user(buffer, link, len))
3681 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3682 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3683 * using) it for any given inode is up to filesystem.
3685 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3687 struct nameidata nd;
3692 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3694 return PTR_ERR(cookie);
3696 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3697 if (dentry->d_inode->i_op->put_link)
3698 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3702 int vfs_follow_link(struct nameidata *nd, const char *link)
3704 return __vfs_follow_link(nd, link);
3707 /* get the link contents into pagecache */
3708 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3712 struct address_space *mapping = dentry->d_inode->i_mapping;
3713 page = read_mapping_page(mapping, 0, NULL);
3718 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3722 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3724 struct page *page = NULL;
3725 char *s = page_getlink(dentry, &page);
3726 int res = vfs_readlink(dentry,buffer,buflen,s);
3729 page_cache_release(page);
3734 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3736 struct page *page = NULL;
3737 nd_set_link(nd, page_getlink(dentry, &page));
3741 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3743 struct page *page = cookie;
3747 page_cache_release(page);
3752 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3754 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3756 struct address_space *mapping = inode->i_mapping;
3761 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3763 flags |= AOP_FLAG_NOFS;
3766 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3767 flags, &page, &fsdata);
3771 kaddr = kmap_atomic(page);
3772 memcpy(kaddr, symname, len-1);
3773 kunmap_atomic(kaddr);
3775 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3782 mark_inode_dirty(inode);
3788 int page_symlink(struct inode *inode, const char *symname, int len)
3790 return __page_symlink(inode, symname, len,
3791 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3794 const struct inode_operations page_symlink_inode_operations = {
3795 .readlink = generic_readlink,
3796 .follow_link = page_follow_link_light,
3797 .put_link = page_put_link,
3800 EXPORT_SYMBOL(user_path_at);
3801 EXPORT_SYMBOL(follow_down_one);
3802 EXPORT_SYMBOL(follow_down);
3803 EXPORT_SYMBOL(follow_up);
3804 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3805 EXPORT_SYMBOL(getname);
3806 EXPORT_SYMBOL(lock_rename);
3807 EXPORT_SYMBOL(lookup_one_len);
3808 EXPORT_SYMBOL(page_follow_link_light);
3809 EXPORT_SYMBOL(page_put_link);
3810 EXPORT_SYMBOL(page_readlink);
3811 EXPORT_SYMBOL(__page_symlink);
3812 EXPORT_SYMBOL(page_symlink);
3813 EXPORT_SYMBOL(page_symlink_inode_operations);
3814 EXPORT_SYMBOL(kern_path);
3815 EXPORT_SYMBOL(vfs_path_lookup);
3816 EXPORT_SYMBOL(inode_permission);
3817 EXPORT_SYMBOL(unlock_rename);
3818 EXPORT_SYMBOL(vfs_create);
3819 EXPORT_SYMBOL(vfs_follow_link);
3820 EXPORT_SYMBOL(vfs_link);
3821 EXPORT_SYMBOL(vfs_mkdir);
3822 EXPORT_SYMBOL(vfs_mknod);
3823 EXPORT_SYMBOL(generic_permission);
3824 EXPORT_SYMBOL(vfs_readlink);
3825 EXPORT_SYMBOL(vfs_rename);
3826 EXPORT_SYMBOL(vfs_rmdir);
3827 EXPORT_SYMBOL(vfs_symlink);
3828 EXPORT_SYMBOL(vfs_unlink);
3829 EXPORT_SYMBOL(dentry_unhash);
3830 EXPORT_SYMBOL(generic_readlink);