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);
467 * release_open_intent - free up open intent resources
468 * @nd: pointer to nameidata
470 void release_open_intent(struct nameidata *nd)
472 struct file *file = nd->intent.open.file;
474 if (file && !IS_ERR(file)) {
475 if (file->f_path.dentry == NULL)
482 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
484 return dentry->d_op->d_revalidate(dentry, nd);
488 * complete_walk - successful completion of path walk
489 * @nd: pointer nameidata
491 * If we had been in RCU mode, drop out of it and legitimize nd->path.
492 * Revalidate the final result, unless we'd already done that during
493 * the path walk or the filesystem doesn't ask for it. Return 0 on
494 * success, -error on failure. In case of failure caller does not
495 * need to drop nd->path.
497 static int complete_walk(struct nameidata *nd)
499 struct dentry *dentry = nd->path.dentry;
502 if (nd->flags & LOOKUP_RCU) {
503 nd->flags &= ~LOOKUP_RCU;
504 if (!(nd->flags & LOOKUP_ROOT))
506 spin_lock(&dentry->d_lock);
507 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
508 spin_unlock(&dentry->d_lock);
510 br_read_unlock(&vfsmount_lock);
513 BUG_ON(nd->inode != dentry->d_inode);
514 spin_unlock(&dentry->d_lock);
515 mntget(nd->path.mnt);
517 br_read_unlock(&vfsmount_lock);
520 if (likely(!(nd->flags & LOOKUP_JUMPED)))
523 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
526 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
529 /* Note: we do not d_invalidate() */
530 status = d_revalidate(dentry, nd);
541 static __always_inline void set_root(struct nameidata *nd)
544 get_fs_root(current->fs, &nd->root);
547 static int link_path_walk(const char *, struct nameidata *);
549 static __always_inline void set_root_rcu(struct nameidata *nd)
552 struct fs_struct *fs = current->fs;
556 seq = read_seqcount_begin(&fs->seq);
558 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
559 } while (read_seqcount_retry(&fs->seq, seq));
563 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
575 nd->flags |= LOOKUP_JUMPED;
577 nd->inode = nd->path.dentry->d_inode;
579 ret = link_path_walk(link, nd);
583 return PTR_ERR(link);
586 static void path_put_conditional(struct path *path, struct nameidata *nd)
589 if (path->mnt != nd->path.mnt)
593 static inline void path_to_nameidata(const struct path *path,
594 struct nameidata *nd)
596 if (!(nd->flags & LOOKUP_RCU)) {
597 dput(nd->path.dentry);
598 if (nd->path.mnt != path->mnt)
599 mntput(nd->path.mnt);
601 nd->path.mnt = path->mnt;
602 nd->path.dentry = path->dentry;
605 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
607 struct inode *inode = link->dentry->d_inode;
608 if (inode->i_op->put_link)
609 inode->i_op->put_link(link->dentry, nd, cookie);
613 static __always_inline int
614 follow_link(struct path *link, struct nameidata *nd, void **p)
616 struct dentry *dentry = link->dentry;
620 BUG_ON(nd->flags & LOOKUP_RCU);
622 if (link->mnt == nd->path.mnt)
626 if (unlikely(current->total_link_count >= 40))
627 goto out_put_nd_path;
630 current->total_link_count++;
633 nd_set_link(nd, NULL);
635 error = security_inode_follow_link(link->dentry, nd);
637 goto out_put_nd_path;
639 nd->last_type = LAST_BIND;
640 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
648 error = __vfs_follow_link(nd, s);
649 } else if (nd->last_type == LAST_BIND) {
650 nd->flags |= LOOKUP_JUMPED;
651 nd->inode = nd->path.dentry->d_inode;
652 if (nd->inode->i_op->follow_link) {
653 /* stepped on a _really_ weird one */
659 put_link(nd, link, *p);
670 static int follow_up_rcu(struct path *path)
672 struct mount *mnt = real_mount(path->mnt);
673 struct mount *parent;
674 struct dentry *mountpoint;
676 parent = mnt->mnt_parent;
677 if (&parent->mnt == path->mnt)
679 mountpoint = mnt->mnt_mountpoint;
680 path->dentry = mountpoint;
681 path->mnt = &parent->mnt;
685 int follow_up(struct path *path)
687 struct mount *mnt = real_mount(path->mnt);
688 struct mount *parent;
689 struct dentry *mountpoint;
691 br_read_lock(&vfsmount_lock);
692 parent = mnt->mnt_parent;
693 if (&parent->mnt == path->mnt) {
694 br_read_unlock(&vfsmount_lock);
697 mntget(&parent->mnt);
698 mountpoint = dget(mnt->mnt_mountpoint);
699 br_read_unlock(&vfsmount_lock);
701 path->dentry = mountpoint;
703 path->mnt = &parent->mnt;
708 * Perform an automount
709 * - return -EISDIR to tell follow_managed() to stop and return the path we
712 static int follow_automount(struct path *path, unsigned flags,
715 struct vfsmount *mnt;
718 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
721 /* We don't want to mount if someone's just doing a stat -
722 * unless they're stat'ing a directory and appended a '/' to
725 * We do, however, want to mount if someone wants to open or
726 * create a file of any type under the mountpoint, wants to
727 * traverse through the mountpoint or wants to open the
728 * mounted directory. Also, autofs may mark negative dentries
729 * as being automount points. These will need the attentions
730 * of the daemon to instantiate them before they can be used.
732 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
733 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
734 path->dentry->d_inode)
737 current->total_link_count++;
738 if (current->total_link_count >= 40)
741 mnt = path->dentry->d_op->d_automount(path);
744 * The filesystem is allowed to return -EISDIR here to indicate
745 * it doesn't want to automount. For instance, autofs would do
746 * this so that its userspace daemon can mount on this dentry.
748 * However, we can only permit this if it's a terminal point in
749 * the path being looked up; if it wasn't then the remainder of
750 * the path is inaccessible and we should say so.
752 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
757 if (!mnt) /* mount collision */
761 /* lock_mount() may release path->mnt on error */
765 err = finish_automount(mnt, path);
769 /* Someone else made a mount here whilst we were busy */
774 path->dentry = dget(mnt->mnt_root);
783 * Handle a dentry that is managed in some way.
784 * - Flagged for transit management (autofs)
785 * - Flagged as mountpoint
786 * - Flagged as automount point
788 * This may only be called in refwalk mode.
790 * Serialization is taken care of in namespace.c
792 static int follow_managed(struct path *path, unsigned flags)
794 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
796 bool need_mntput = false;
799 /* Given that we're not holding a lock here, we retain the value in a
800 * local variable for each dentry as we look at it so that we don't see
801 * the components of that value change under us */
802 while (managed = ACCESS_ONCE(path->dentry->d_flags),
803 managed &= DCACHE_MANAGED_DENTRY,
804 unlikely(managed != 0)) {
805 /* Allow the filesystem to manage the transit without i_mutex
807 if (managed & DCACHE_MANAGE_TRANSIT) {
808 BUG_ON(!path->dentry->d_op);
809 BUG_ON(!path->dentry->d_op->d_manage);
810 ret = path->dentry->d_op->d_manage(path->dentry, false);
815 /* Transit to a mounted filesystem. */
816 if (managed & DCACHE_MOUNTED) {
817 struct vfsmount *mounted = lookup_mnt(path);
823 path->dentry = dget(mounted->mnt_root);
828 /* Something is mounted on this dentry in another
829 * namespace and/or whatever was mounted there in this
830 * namespace got unmounted before we managed to get the
834 /* Handle an automount point */
835 if (managed & DCACHE_NEED_AUTOMOUNT) {
836 ret = follow_automount(path, flags, &need_mntput);
842 /* We didn't change the current path point */
846 if (need_mntput && path->mnt == mnt)
850 return ret < 0 ? ret : need_mntput;
853 int follow_down_one(struct path *path)
855 struct vfsmount *mounted;
857 mounted = lookup_mnt(path);
862 path->dentry = dget(mounted->mnt_root);
868 static inline bool managed_dentry_might_block(struct dentry *dentry)
870 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
871 dentry->d_op->d_manage(dentry, true) < 0);
875 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
876 * we meet a managed dentry that would need blocking.
878 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
879 struct inode **inode)
882 struct mount *mounted;
884 * Don't forget we might have a non-mountpoint managed dentry
885 * that wants to block transit.
887 if (unlikely(managed_dentry_might_block(path->dentry)))
890 if (!d_mountpoint(path->dentry))
893 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
896 path->mnt = &mounted->mnt;
897 path->dentry = mounted->mnt.mnt_root;
898 nd->flags |= LOOKUP_JUMPED;
899 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
901 * Update the inode too. We don't need to re-check the
902 * dentry sequence number here after this d_inode read,
903 * because a mount-point is always pinned.
905 *inode = path->dentry->d_inode;
910 static void follow_mount_rcu(struct nameidata *nd)
912 while (d_mountpoint(nd->path.dentry)) {
913 struct mount *mounted;
914 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
917 nd->path.mnt = &mounted->mnt;
918 nd->path.dentry = mounted->mnt.mnt_root;
919 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
923 static int follow_dotdot_rcu(struct nameidata *nd)
928 if (nd->path.dentry == nd->root.dentry &&
929 nd->path.mnt == nd->root.mnt) {
932 if (nd->path.dentry != nd->path.mnt->mnt_root) {
933 struct dentry *old = nd->path.dentry;
934 struct dentry *parent = old->d_parent;
937 seq = read_seqcount_begin(&parent->d_seq);
938 if (read_seqcount_retry(&old->d_seq, nd->seq))
940 nd->path.dentry = parent;
944 if (!follow_up_rcu(&nd->path))
946 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
948 follow_mount_rcu(nd);
949 nd->inode = nd->path.dentry->d_inode;
953 nd->flags &= ~LOOKUP_RCU;
954 if (!(nd->flags & LOOKUP_ROOT))
957 br_read_unlock(&vfsmount_lock);
962 * Follow down to the covering mount currently visible to userspace. At each
963 * point, the filesystem owning that dentry may be queried as to whether the
964 * caller is permitted to proceed or not.
966 int follow_down(struct path *path)
971 while (managed = ACCESS_ONCE(path->dentry->d_flags),
972 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
973 /* Allow the filesystem to manage the transit without i_mutex
976 * We indicate to the filesystem if someone is trying to mount
977 * something here. This gives autofs the chance to deny anyone
978 * other than its daemon the right to mount on its
981 * The filesystem may sleep at this point.
983 if (managed & DCACHE_MANAGE_TRANSIT) {
984 BUG_ON(!path->dentry->d_op);
985 BUG_ON(!path->dentry->d_op->d_manage);
986 ret = path->dentry->d_op->d_manage(
987 path->dentry, false);
989 return ret == -EISDIR ? 0 : ret;
992 /* Transit to a mounted filesystem. */
993 if (managed & DCACHE_MOUNTED) {
994 struct vfsmount *mounted = lookup_mnt(path);
1000 path->dentry = dget(mounted->mnt_root);
1004 /* Don't handle automount points here */
1011 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1013 static void follow_mount(struct path *path)
1015 while (d_mountpoint(path->dentry)) {
1016 struct vfsmount *mounted = lookup_mnt(path);
1021 path->mnt = mounted;
1022 path->dentry = dget(mounted->mnt_root);
1026 static void follow_dotdot(struct nameidata *nd)
1031 struct dentry *old = nd->path.dentry;
1033 if (nd->path.dentry == nd->root.dentry &&
1034 nd->path.mnt == nd->root.mnt) {
1037 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1038 /* rare case of legitimate dget_parent()... */
1039 nd->path.dentry = dget_parent(nd->path.dentry);
1043 if (!follow_up(&nd->path))
1046 follow_mount(&nd->path);
1047 nd->inode = nd->path.dentry->d_inode;
1051 * This looks up the name in dcache, possibly revalidates the old dentry and
1052 * allocates a new one if not found or not valid. In the need_lookup argument
1053 * returns whether i_op->lookup is necessary.
1055 * dir->d_inode->i_mutex must be held
1057 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1058 struct nameidata *nd, bool *need_lookup)
1060 struct dentry *dentry;
1063 *need_lookup = false;
1064 dentry = d_lookup(dir, name);
1066 if (d_need_lookup(dentry)) {
1067 *need_lookup = true;
1068 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1069 error = d_revalidate(dentry, nd);
1070 if (unlikely(error <= 0)) {
1073 return ERR_PTR(error);
1074 } else if (!d_invalidate(dentry)) {
1083 dentry = d_alloc(dir, name);
1084 if (unlikely(!dentry))
1085 return ERR_PTR(-ENOMEM);
1087 *need_lookup = true;
1093 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1094 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1096 * dir->d_inode->i_mutex must be held
1098 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1099 struct nameidata *nd)
1103 /* Don't create child dentry for a dead directory. */
1104 if (unlikely(IS_DEADDIR(dir))) {
1106 return ERR_PTR(-ENOENT);
1109 old = dir->i_op->lookup(dir, dentry, nd);
1110 if (unlikely(old)) {
1117 static struct dentry *__lookup_hash(struct qstr *name,
1118 struct dentry *base, struct nameidata *nd)
1121 struct dentry *dentry;
1123 dentry = lookup_dcache(name, base, nd, &need_lookup);
1127 return lookup_real(base->d_inode, dentry, nd);
1131 * It's more convoluted than I'd like it to be, but... it's still fairly
1132 * small and for now I'd prefer to have fast path as straight as possible.
1133 * It _is_ time-critical.
1135 static int lookup_fast(struct nameidata *nd, struct qstr *name,
1136 struct path *path, struct inode **inode)
1138 struct vfsmount *mnt = nd->path.mnt;
1139 struct dentry *dentry, *parent = nd->path.dentry;
1145 * Rename seqlock is not required here because in the off chance
1146 * of a false negative due to a concurrent rename, we're going to
1147 * do the non-racy lookup, below.
1149 if (nd->flags & LOOKUP_RCU) {
1151 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
1156 * This sequence count validates that the inode matches
1157 * the dentry name information from lookup.
1159 *inode = dentry->d_inode;
1160 if (read_seqcount_retry(&dentry->d_seq, seq))
1164 * This sequence count validates that the parent had no
1165 * changes while we did the lookup of the dentry above.
1167 * The memory barrier in read_seqcount_begin of child is
1168 * enough, we can use __read_seqcount_retry here.
1170 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1174 if (unlikely(d_need_lookup(dentry)))
1176 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1177 status = d_revalidate(dentry, nd);
1178 if (unlikely(status <= 0)) {
1179 if (status != -ECHILD)
1185 path->dentry = dentry;
1186 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1188 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1192 if (unlazy_walk(nd, dentry))
1195 dentry = __d_lookup(parent, name);
1198 if (unlikely(!dentry))
1201 if (unlikely(d_need_lookup(dentry))) {
1206 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1207 status = d_revalidate(dentry, nd);
1208 if (unlikely(status <= 0)) {
1213 if (!d_invalidate(dentry)) {
1220 path->dentry = dentry;
1221 err = follow_managed(path, nd->flags);
1222 if (unlikely(err < 0)) {
1223 path_put_conditional(path, nd);
1227 nd->flags |= LOOKUP_JUMPED;
1228 *inode = path->dentry->d_inode;
1235 /* Fast lookup failed, do it the slow way */
1236 static int lookup_slow(struct nameidata *nd, struct qstr *name,
1239 struct dentry *dentry, *parent;
1242 parent = nd->path.dentry;
1243 BUG_ON(nd->inode != parent->d_inode);
1245 mutex_lock(&parent->d_inode->i_mutex);
1246 dentry = __lookup_hash(name, parent, nd);
1247 mutex_unlock(&parent->d_inode->i_mutex);
1249 return PTR_ERR(dentry);
1250 path->mnt = nd->path.mnt;
1251 path->dentry = dentry;
1252 err = follow_managed(path, nd->flags);
1253 if (unlikely(err < 0)) {
1254 path_put_conditional(path, nd);
1258 nd->flags |= LOOKUP_JUMPED;
1262 static inline int may_lookup(struct nameidata *nd)
1264 if (nd->flags & LOOKUP_RCU) {
1265 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1268 if (unlazy_walk(nd, NULL))
1271 return inode_permission(nd->inode, MAY_EXEC);
1274 static inline int handle_dots(struct nameidata *nd, int type)
1276 if (type == LAST_DOTDOT) {
1277 if (nd->flags & LOOKUP_RCU) {
1278 if (follow_dotdot_rcu(nd))
1286 static void terminate_walk(struct nameidata *nd)
1288 if (!(nd->flags & LOOKUP_RCU)) {
1289 path_put(&nd->path);
1291 nd->flags &= ~LOOKUP_RCU;
1292 if (!(nd->flags & LOOKUP_ROOT))
1293 nd->root.mnt = NULL;
1295 br_read_unlock(&vfsmount_lock);
1300 * Do we need to follow links? We _really_ want to be able
1301 * to do this check without having to look at inode->i_op,
1302 * so we keep a cache of "no, this doesn't need follow_link"
1303 * for the common case.
1305 static inline int should_follow_link(struct inode *inode, int follow)
1307 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1308 if (likely(inode->i_op->follow_link))
1311 /* This gets set once for the inode lifetime */
1312 spin_lock(&inode->i_lock);
1313 inode->i_opflags |= IOP_NOFOLLOW;
1314 spin_unlock(&inode->i_lock);
1319 static inline int walk_component(struct nameidata *nd, struct path *path,
1320 struct qstr *name, int type, int follow)
1322 struct inode *inode;
1325 * "." and ".." are special - ".." especially so because it has
1326 * to be able to know about the current root directory and
1327 * parent relationships.
1329 if (unlikely(type != LAST_NORM))
1330 return handle_dots(nd, type);
1331 err = lookup_fast(nd, name, path, &inode);
1332 if (unlikely(err)) {
1336 err = lookup_slow(nd, name, path);
1340 inode = path->dentry->d_inode;
1346 if (should_follow_link(inode, follow)) {
1347 if (nd->flags & LOOKUP_RCU) {
1348 if (unlikely(unlazy_walk(nd, path->dentry))) {
1353 BUG_ON(inode != path->dentry->d_inode);
1356 path_to_nameidata(path, nd);
1361 path_to_nameidata(path, nd);
1368 * This limits recursive symlink follows to 8, while
1369 * limiting consecutive symlinks to 40.
1371 * Without that kind of total limit, nasty chains of consecutive
1372 * symlinks can cause almost arbitrarily long lookups.
1374 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1378 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1379 path_put_conditional(path, nd);
1380 path_put(&nd->path);
1383 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1386 current->link_count++;
1389 struct path link = *path;
1392 res = follow_link(&link, nd, &cookie);
1395 res = walk_component(nd, path, &nd->last,
1396 nd->last_type, LOOKUP_FOLLOW);
1397 put_link(nd, &link, cookie);
1400 current->link_count--;
1406 * We really don't want to look at inode->i_op->lookup
1407 * when we don't have to. So we keep a cache bit in
1408 * the inode ->i_opflags field that says "yes, we can
1409 * do lookup on this inode".
1411 static inline int can_lookup(struct inode *inode)
1413 if (likely(inode->i_opflags & IOP_LOOKUP))
1415 if (likely(!inode->i_op->lookup))
1418 /* We do this once for the lifetime of the inode */
1419 spin_lock(&inode->i_lock);
1420 inode->i_opflags |= IOP_LOOKUP;
1421 spin_unlock(&inode->i_lock);
1426 * We can do the critical dentry name comparison and hashing
1427 * operations one word at a time, but we are limited to:
1429 * - Architectures with fast unaligned word accesses. We could
1430 * do a "get_unaligned()" if this helps and is sufficiently
1433 * - Little-endian machines (so that we can generate the mask
1434 * of low bytes efficiently). Again, we *could* do a byte
1435 * swapping load on big-endian architectures if that is not
1436 * expensive enough to make the optimization worthless.
1438 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1439 * do not trap on the (extremely unlikely) case of a page
1440 * crossing operation.
1442 * - Furthermore, we need an efficient 64-bit compile for the
1443 * 64-bit case in order to generate the "number of bytes in
1444 * the final mask". Again, that could be replaced with a
1445 * efficient population count instruction or similar.
1447 #ifdef CONFIG_DCACHE_WORD_ACCESS
1449 #include <asm/word-at-a-time.h>
1453 static inline unsigned int fold_hash(unsigned long hash)
1455 hash += hash >> (8*sizeof(int));
1459 #else /* 32-bit case */
1461 #define fold_hash(x) (x)
1465 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1467 unsigned long a, mask;
1468 unsigned long hash = 0;
1471 a = load_unaligned_zeropad(name);
1472 if (len < sizeof(unsigned long))
1476 name += sizeof(unsigned long);
1477 len -= sizeof(unsigned long);
1481 mask = ~(~0ul << len*8);
1484 return fold_hash(hash);
1486 EXPORT_SYMBOL(full_name_hash);
1489 * Calculate the length and hash of the path component, and
1490 * return the length of the component;
1492 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1494 unsigned long a, b, adata, bdata, mask, hash, len;
1495 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1498 len = -sizeof(unsigned long);
1500 hash = (hash + a) * 9;
1501 len += sizeof(unsigned long);
1502 a = load_unaligned_zeropad(name+len);
1503 b = a ^ REPEAT_BYTE('/');
1504 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1506 adata = prep_zero_mask(a, adata, &constants);
1507 bdata = prep_zero_mask(b, bdata, &constants);
1509 mask = create_zero_mask(adata | bdata);
1511 hash += a & zero_bytemask(mask);
1512 *hashp = fold_hash(hash);
1514 return len + find_zero(mask);
1519 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1521 unsigned long hash = init_name_hash();
1523 hash = partial_name_hash(*name++, hash);
1524 return end_name_hash(hash);
1526 EXPORT_SYMBOL(full_name_hash);
1529 * We know there's a real path component here of at least
1532 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1534 unsigned long hash = init_name_hash();
1535 unsigned long len = 0, c;
1537 c = (unsigned char)*name;
1540 hash = partial_name_hash(c, hash);
1541 c = (unsigned char)name[len];
1542 } while (c && c != '/');
1543 *hashp = end_name_hash(hash);
1551 * This is the basic name resolution function, turning a pathname into
1552 * the final dentry. We expect 'base' to be positive and a directory.
1554 * Returns 0 and nd will have valid dentry and mnt on success.
1555 * Returns error and drops reference to input namei data on failure.
1557 static int link_path_walk(const char *name, struct nameidata *nd)
1567 /* At this point we know we have a real path component. */
1573 err = may_lookup(nd);
1577 len = hash_name(name, &this.hash);
1582 if (name[0] == '.') switch (len) {
1584 if (name[1] == '.') {
1586 nd->flags |= LOOKUP_JUMPED;
1592 if (likely(type == LAST_NORM)) {
1593 struct dentry *parent = nd->path.dentry;
1594 nd->flags &= ~LOOKUP_JUMPED;
1595 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1596 err = parent->d_op->d_hash(parent, nd->inode,
1604 goto last_component;
1606 * If it wasn't NUL, we know it was '/'. Skip that
1607 * slash, and continue until no more slashes.
1611 } while (unlikely(name[len] == '/'));
1613 goto last_component;
1616 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1621 err = nested_symlink(&next, nd);
1625 if (can_lookup(nd->inode))
1629 /* here ends the main loop */
1633 nd->last_type = type;
1640 static int path_init(int dfd, const char *name, unsigned int flags,
1641 struct nameidata *nd, struct file **fp)
1647 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1648 nd->flags = flags | LOOKUP_JUMPED;
1650 if (flags & LOOKUP_ROOT) {
1651 struct inode *inode = nd->root.dentry->d_inode;
1653 if (!inode->i_op->lookup)
1655 retval = inode_permission(inode, MAY_EXEC);
1659 nd->path = nd->root;
1661 if (flags & LOOKUP_RCU) {
1662 br_read_lock(&vfsmount_lock);
1664 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1666 path_get(&nd->path);
1671 nd->root.mnt = NULL;
1674 if (flags & LOOKUP_RCU) {
1675 br_read_lock(&vfsmount_lock);
1680 path_get(&nd->root);
1682 nd->path = nd->root;
1683 } else if (dfd == AT_FDCWD) {
1684 if (flags & LOOKUP_RCU) {
1685 struct fs_struct *fs = current->fs;
1688 br_read_lock(&vfsmount_lock);
1692 seq = read_seqcount_begin(&fs->seq);
1694 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1695 } while (read_seqcount_retry(&fs->seq, seq));
1697 get_fs_pwd(current->fs, &nd->path);
1700 struct dentry *dentry;
1702 file = fget_raw_light(dfd, &fput_needed);
1707 dentry = file->f_path.dentry;
1711 if (!S_ISDIR(dentry->d_inode->i_mode))
1714 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1719 nd->path = file->f_path;
1720 if (flags & LOOKUP_RCU) {
1723 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1724 br_read_lock(&vfsmount_lock);
1727 path_get(&file->f_path);
1728 fput_light(file, fput_needed);
1732 nd->inode = nd->path.dentry->d_inode;
1736 fput_light(file, fput_needed);
1741 static inline int lookup_last(struct nameidata *nd, struct path *path)
1743 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1744 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1746 nd->flags &= ~LOOKUP_PARENT;
1747 return walk_component(nd, path, &nd->last, nd->last_type,
1748 nd->flags & LOOKUP_FOLLOW);
1751 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1752 static int path_lookupat(int dfd, const char *name,
1753 unsigned int flags, struct nameidata *nd)
1755 struct file *base = NULL;
1760 * Path walking is largely split up into 2 different synchronisation
1761 * schemes, rcu-walk and ref-walk (explained in
1762 * Documentation/filesystems/path-lookup.txt). These share much of the
1763 * path walk code, but some things particularly setup, cleanup, and
1764 * following mounts are sufficiently divergent that functions are
1765 * duplicated. Typically there is a function foo(), and its RCU
1766 * analogue, foo_rcu().
1768 * -ECHILD is the error number of choice (just to avoid clashes) that
1769 * is returned if some aspect of an rcu-walk fails. Such an error must
1770 * be handled by restarting a traditional ref-walk (which will always
1771 * be able to complete).
1773 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1778 current->total_link_count = 0;
1779 err = link_path_walk(name, nd);
1781 if (!err && !(flags & LOOKUP_PARENT)) {
1782 err = lookup_last(nd, &path);
1785 struct path link = path;
1786 nd->flags |= LOOKUP_PARENT;
1787 err = follow_link(&link, nd, &cookie);
1790 err = lookup_last(nd, &path);
1791 put_link(nd, &link, cookie);
1796 err = complete_walk(nd);
1798 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1799 if (!nd->inode->i_op->lookup) {
1800 path_put(&nd->path);
1808 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1809 path_put(&nd->root);
1810 nd->root.mnt = NULL;
1815 static int do_path_lookup(int dfd, const char *name,
1816 unsigned int flags, struct nameidata *nd)
1818 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1819 if (unlikely(retval == -ECHILD))
1820 retval = path_lookupat(dfd, name, flags, nd);
1821 if (unlikely(retval == -ESTALE))
1822 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1824 if (likely(!retval)) {
1825 if (unlikely(!audit_dummy_context())) {
1826 if (nd->path.dentry && nd->inode)
1827 audit_inode(name, nd->path.dentry);
1833 int kern_path_parent(const char *name, struct nameidata *nd)
1835 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1838 int kern_path(const char *name, unsigned int flags, struct path *path)
1840 struct nameidata nd;
1841 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1848 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1849 * @dentry: pointer to dentry of the base directory
1850 * @mnt: pointer to vfs mount of the base directory
1851 * @name: pointer to file name
1852 * @flags: lookup flags
1853 * @path: pointer to struct path to fill
1855 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1856 const char *name, unsigned int flags,
1859 struct nameidata nd;
1861 nd.root.dentry = dentry;
1863 BUG_ON(flags & LOOKUP_PARENT);
1864 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1865 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1872 * Restricted form of lookup. Doesn't follow links, single-component only,
1873 * needs parent already locked. Doesn't follow mounts.
1876 static struct dentry *lookup_hash(struct nameidata *nd)
1878 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1882 * lookup_one_len - filesystem helper to lookup single pathname component
1883 * @name: pathname component to lookup
1884 * @base: base directory to lookup from
1885 * @len: maximum length @len should be interpreted to
1887 * Note that this routine is purely a helper for filesystem usage and should
1888 * not be called by generic code. Also note that by using this function the
1889 * nameidata argument is passed to the filesystem methods and a filesystem
1890 * using this helper needs to be prepared for that.
1892 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1898 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1902 this.hash = full_name_hash(name, len);
1904 return ERR_PTR(-EACCES);
1907 c = *(const unsigned char *)name++;
1908 if (c == '/' || c == '\0')
1909 return ERR_PTR(-EACCES);
1912 * See if the low-level filesystem might want
1913 * to use its own hash..
1915 if (base->d_flags & DCACHE_OP_HASH) {
1916 int err = base->d_op->d_hash(base, base->d_inode, &this);
1918 return ERR_PTR(err);
1921 err = inode_permission(base->d_inode, MAY_EXEC);
1923 return ERR_PTR(err);
1925 return __lookup_hash(&this, base, NULL);
1928 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1929 struct path *path, int *empty)
1931 struct nameidata nd;
1932 char *tmp = getname_flags(name, flags, empty);
1933 int err = PTR_ERR(tmp);
1936 BUG_ON(flags & LOOKUP_PARENT);
1938 err = do_path_lookup(dfd, tmp, flags, &nd);
1946 int user_path_at(int dfd, const char __user *name, unsigned flags,
1949 return user_path_at_empty(dfd, name, flags, path, NULL);
1952 static int user_path_parent(int dfd, const char __user *path,
1953 struct nameidata *nd, char **name)
1955 char *s = getname(path);
1961 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1971 * It's inline, so penalty for filesystems that don't use sticky bit is
1974 static inline int check_sticky(struct inode *dir, struct inode *inode)
1976 kuid_t fsuid = current_fsuid();
1978 if (!(dir->i_mode & S_ISVTX))
1980 if (uid_eq(inode->i_uid, fsuid))
1982 if (uid_eq(dir->i_uid, fsuid))
1984 return !inode_capable(inode, CAP_FOWNER);
1988 * Check whether we can remove a link victim from directory dir, check
1989 * whether the type of victim is right.
1990 * 1. We can't do it if dir is read-only (done in permission())
1991 * 2. We should have write and exec permissions on dir
1992 * 3. We can't remove anything from append-only dir
1993 * 4. We can't do anything with immutable dir (done in permission())
1994 * 5. If the sticky bit on dir is set we should either
1995 * a. be owner of dir, or
1996 * b. be owner of victim, or
1997 * c. have CAP_FOWNER capability
1998 * 6. If the victim is append-only or immutable we can't do antyhing with
1999 * links pointing to it.
2000 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2001 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2002 * 9. We can't remove a root or mountpoint.
2003 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2004 * nfs_async_unlink().
2006 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2010 if (!victim->d_inode)
2013 BUG_ON(victim->d_parent->d_inode != dir);
2014 audit_inode_child(victim, dir);
2016 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2021 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2022 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2025 if (!S_ISDIR(victim->d_inode->i_mode))
2027 if (IS_ROOT(victim))
2029 } else if (S_ISDIR(victim->d_inode->i_mode))
2031 if (IS_DEADDIR(dir))
2033 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2038 /* Check whether we can create an object with dentry child in directory
2040 * 1. We can't do it if child already exists (open has special treatment for
2041 * this case, but since we are inlined it's OK)
2042 * 2. We can't do it if dir is read-only (done in permission())
2043 * 3. We should have write and exec permissions on dir
2044 * 4. We can't do it if dir is immutable (done in permission())
2046 static inline int may_create(struct inode *dir, struct dentry *child)
2050 if (IS_DEADDIR(dir))
2052 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2056 * p1 and p2 should be directories on the same fs.
2058 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2063 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2067 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2069 p = d_ancestor(p2, p1);
2071 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2072 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2076 p = d_ancestor(p1, p2);
2078 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2079 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2083 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2084 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2088 void unlock_rename(struct dentry *p1, struct dentry *p2)
2090 mutex_unlock(&p1->d_inode->i_mutex);
2092 mutex_unlock(&p2->d_inode->i_mutex);
2093 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2097 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2098 struct nameidata *nd)
2100 int error = may_create(dir, dentry);
2105 if (!dir->i_op->create)
2106 return -EACCES; /* shouldn't it be ENOSYS? */
2109 error = security_inode_create(dir, dentry, mode);
2112 error = dir->i_op->create(dir, dentry, mode, nd);
2114 fsnotify_create(dir, dentry);
2118 static int may_open(struct path *path, int acc_mode, int flag)
2120 struct dentry *dentry = path->dentry;
2121 struct inode *inode = dentry->d_inode;
2131 switch (inode->i_mode & S_IFMT) {
2135 if (acc_mode & MAY_WRITE)
2140 if (path->mnt->mnt_flags & MNT_NODEV)
2149 error = inode_permission(inode, acc_mode);
2154 * An append-only file must be opened in append mode for writing.
2156 if (IS_APPEND(inode)) {
2157 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2163 /* O_NOATIME can only be set by the owner or superuser */
2164 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2170 static int handle_truncate(struct file *filp)
2172 struct path *path = &filp->f_path;
2173 struct inode *inode = path->dentry->d_inode;
2174 int error = get_write_access(inode);
2178 * Refuse to truncate files with mandatory locks held on them.
2180 error = locks_verify_locked(inode);
2182 error = security_path_truncate(path);
2184 error = do_truncate(path->dentry, 0,
2185 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2188 put_write_access(inode);
2192 static inline int open_to_namei_flags(int flag)
2194 if ((flag & O_ACCMODE) == 3)
2199 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2201 int error = security_path_mknod(dir, dentry, mode, 0);
2205 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2209 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2212 static struct file *atomic_open(struct nameidata *nd, struct dentry *dentry,
2213 struct path *path, const struct open_flags *op,
2214 int *want_write, bool need_lookup,
2217 struct inode *dir = nd->path.dentry->d_inode;
2218 unsigned open_flag = open_to_namei_flags(op->open_flag);
2224 int create_error = 0;
2225 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2227 BUG_ON(dentry->d_inode);
2229 /* Don't create child dentry for a dead directory. */
2230 if (unlikely(IS_DEADDIR(dir))) {
2231 filp = ERR_PTR(-ENOENT);
2235 mode = op->mode & S_IALLUGO;
2236 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2237 mode &= ~current_umask();
2239 if (open_flag & O_EXCL) {
2240 open_flag &= ~O_TRUNC;
2245 * Checking write permission is tricky, bacuse we don't know if we are
2246 * going to actually need it: O_CREAT opens should work as long as the
2247 * file exists. But checking existence breaks atomicity. The trick is
2248 * to check access and if not granted clear O_CREAT from the flags.
2250 * Another problem is returing the "right" error value (e.g. for an
2251 * O_EXCL open we want to return EEXIST not EROFS).
2253 if ((open_flag & (O_CREAT | O_TRUNC)) ||
2254 (open_flag & O_ACCMODE) != O_RDONLY) {
2255 error = mnt_want_write(nd->path.mnt);
2258 } else if (!(open_flag & O_CREAT)) {
2260 * No O_CREATE -> atomicity not a requirement -> fall
2261 * back to lookup + open
2264 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2265 /* Fall back and fail with the right error */
2266 create_error = error;
2269 /* No side effects, safe to clear O_CREAT */
2270 create_error = error;
2271 open_flag &= ~O_CREAT;
2275 if (open_flag & O_CREAT) {
2276 error = may_o_create(&nd->path, dentry, op->mode);
2278 create_error = error;
2279 if (open_flag & O_EXCL)
2281 open_flag &= ~O_CREAT;
2285 if (nd->flags & LOOKUP_DIRECTORY)
2286 open_flag |= O_DIRECTORY;
2288 od.dentry = DENTRY_NOT_SET;
2289 od.mnt = nd->path.mnt;
2290 od.filp = &nd->intent.open.file;
2291 filp = dir->i_op->atomic_open(dir, dentry, &od, open_flag, mode,
2294 if (WARN_ON(od.dentry != DENTRY_NOT_SET))
2297 if (create_error && PTR_ERR(filp) == -ENOENT)
2298 filp = ERR_PTR(create_error);
2302 acc_mode = op->acc_mode;
2304 fsnotify_create(dir, dentry);
2305 acc_mode = MAY_OPEN;
2309 if (WARN_ON(od.dentry == DENTRY_NOT_SET)) {
2310 filp = ERR_PTR(-EIO);
2321 * We didn't have the inode before the open, so check open permission
2324 error = may_open(&filp->f_path, acc_mode, open_flag);
2328 error = open_check_o_direct(filp);
2338 filp = ERR_PTR(error);
2343 dentry = lookup_real(dir, dentry, nd);
2345 return ERR_CAST(dentry);
2348 int open_flag = op->open_flag;
2350 filp = ERR_PTR(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 * Lookup, maybe create and open the last component
2372 * Must be called with i_mutex held on parent.
2374 * Returns open file or NULL on success, error otherwise. NULL means no open
2375 * was performed, only lookup.
2377 static struct file *lookup_open(struct nameidata *nd, struct path *path,
2378 const struct open_flags *op,
2379 int *want_write, bool *created)
2381 struct dentry *dir = nd->path.dentry;
2382 struct inode *dir_inode = dir->d_inode;
2383 struct dentry *dentry;
2388 dentry = lookup_dcache(&nd->last, dir, nd, &need_lookup);
2390 return ERR_CAST(dentry);
2392 /* Cached positive dentry: will open in f_op->open */
2393 if (!need_lookup && dentry->d_inode)
2396 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2397 return atomic_open(nd, dentry, path, op, want_write,
2398 need_lookup, created);
2402 BUG_ON(dentry->d_inode);
2404 dentry = lookup_real(dir_inode, dentry, nd);
2406 return ERR_CAST(dentry);
2409 /* Negative dentry, just create the file */
2410 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2411 umode_t mode = op->mode;
2412 if (!IS_POSIXACL(dir->d_inode))
2413 mode &= ~current_umask();
2415 * This write is needed to ensure that a
2416 * rw->ro transition does not occur between
2417 * the time when the file is created and when
2418 * a permanent write count is taken through
2419 * the 'struct file' in nameidata_to_filp().
2421 error = mnt_want_write(nd->path.mnt);
2426 error = security_path_mknod(&nd->path, dentry, mode, 0);
2429 error = vfs_create(dir->d_inode, dentry, mode, nd);
2434 path->dentry = dentry;
2435 path->mnt = nd->path.mnt;
2440 return ERR_PTR(error);
2444 * Handle the last step of open()
2446 static struct file *do_last(struct nameidata *nd, struct path *path,
2447 const struct open_flags *op, const char *pathname)
2449 struct dentry *dir = nd->path.dentry;
2450 int open_flag = op->open_flag;
2451 int will_truncate = open_flag & O_TRUNC;
2453 int acc_mode = op->acc_mode;
2455 struct inode *inode;
2458 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2459 bool retried = false;
2462 nd->flags &= ~LOOKUP_PARENT;
2463 nd->flags |= op->intent;
2465 switch (nd->last_type) {
2468 error = handle_dots(nd, nd->last_type);
2470 return ERR_PTR(error);
2473 error = complete_walk(nd);
2475 return ERR_PTR(error);
2476 audit_inode(pathname, nd->path.dentry);
2477 if (open_flag & O_CREAT) {
2483 error = complete_walk(nd);
2485 return ERR_PTR(error);
2486 audit_inode(pathname, dir);
2490 if (!(open_flag & O_CREAT)) {
2491 if (nd->last.name[nd->last.len])
2492 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2493 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2495 /* we _can_ be in RCU mode here */
2496 error = lookup_fast(nd, &nd->last, path, &inode);
2503 BUG_ON(nd->inode != dir->d_inode);
2505 /* create side of things */
2507 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2508 * has been cleared when we got to the last component we are
2511 error = complete_walk(nd);
2513 return ERR_PTR(error);
2515 audit_inode(pathname, dir);
2517 /* trailing slashes? */
2518 if (nd->last.name[nd->last.len])
2523 mutex_lock(&dir->d_inode->i_mutex);
2524 filp = lookup_open(nd, path, op, &want_write, &created);
2525 mutex_unlock(&dir->d_inode->i_mutex);
2531 if (created || !S_ISREG(filp->f_path.dentry->d_inode->i_mode))
2534 audit_inode(pathname, filp->f_path.dentry);
2539 /* Don't check for write permission, don't truncate */
2540 open_flag &= ~O_TRUNC;
2542 acc_mode = MAY_OPEN;
2543 path_to_nameidata(path, nd);
2548 * It already exists.
2550 audit_inode(pathname, path->dentry);
2553 * If atomic_open() acquired write access it is dropped now due to
2554 * possible mount and symlink following (this might be optimized away if
2558 mnt_drop_write(nd->path.mnt);
2563 if (open_flag & O_EXCL)
2566 error = follow_managed(path, nd->flags);
2571 nd->flags |= LOOKUP_JUMPED;
2573 BUG_ON(nd->flags & LOOKUP_RCU);
2574 inode = path->dentry->d_inode;
2576 /* we _can_ be in RCU mode here */
2579 path_to_nameidata(path, nd);
2583 if (should_follow_link(inode, !symlink_ok)) {
2584 if (nd->flags & LOOKUP_RCU) {
2585 if (unlikely(unlazy_walk(nd, path->dentry))) {
2590 BUG_ON(inode != path->dentry->d_inode);
2594 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2595 path_to_nameidata(path, nd);
2597 save_parent.dentry = nd->path.dentry;
2598 save_parent.mnt = mntget(path->mnt);
2599 nd->path.dentry = path->dentry;
2603 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2604 error = complete_walk(nd);
2606 path_put(&save_parent);
2607 return ERR_PTR(error);
2610 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
2613 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
2615 audit_inode(pathname, nd->path.dentry);
2617 if (!S_ISREG(nd->inode->i_mode))
2620 if (will_truncate) {
2621 error = mnt_want_write(nd->path.mnt);
2627 error = may_open(&nd->path, acc_mode, open_flag);
2630 filp = nameidata_to_filp(nd);
2631 if (filp == ERR_PTR(-EOPENSTALE) && save_parent.dentry && !retried) {
2632 BUG_ON(save_parent.dentry != dir);
2633 path_put(&nd->path);
2634 nd->path = save_parent;
2635 nd->inode = dir->d_inode;
2636 save_parent.mnt = NULL;
2637 save_parent.dentry = NULL;
2639 mnt_drop_write(nd->path.mnt);
2646 if (!IS_ERR(filp)) {
2647 error = ima_file_check(filp, op->acc_mode);
2650 filp = ERR_PTR(error);
2653 if (!IS_ERR(filp)) {
2654 if (will_truncate) {
2655 error = handle_truncate(filp);
2658 filp = ERR_PTR(error);
2664 mnt_drop_write(nd->path.mnt);
2665 path_put(&save_parent);
2670 path_put_conditional(path, nd);
2672 filp = ERR_PTR(error);
2676 static struct file *path_openat(int dfd, const char *pathname,
2677 struct nameidata *nd, const struct open_flags *op, int flags)
2679 struct file *base = NULL;
2684 filp = get_empty_filp();
2686 return ERR_PTR(-ENFILE);
2688 filp->f_flags = op->open_flag;
2689 nd->intent.open.file = filp;
2690 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2691 nd->intent.open.create_mode = op->mode;
2693 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2694 if (unlikely(error))
2697 current->total_link_count = 0;
2698 error = link_path_walk(pathname, nd);
2699 if (unlikely(error))
2702 filp = do_last(nd, &path, op, pathname);
2703 while (unlikely(!filp)) { /* trailing symlink */
2704 struct path link = path;
2706 if (!(nd->flags & LOOKUP_FOLLOW)) {
2707 path_put_conditional(&path, nd);
2708 path_put(&nd->path);
2709 filp = ERR_PTR(-ELOOP);
2712 nd->flags |= LOOKUP_PARENT;
2713 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2714 error = follow_link(&link, nd, &cookie);
2715 if (unlikely(error))
2717 filp = do_last(nd, &path, op, pathname);
2718 put_link(nd, &link, cookie);
2721 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2722 path_put(&nd->root);
2725 release_open_intent(nd);
2726 if (filp == ERR_PTR(-EOPENSTALE)) {
2727 if (flags & LOOKUP_RCU)
2728 filp = ERR_PTR(-ECHILD);
2730 filp = ERR_PTR(-ESTALE);
2735 filp = ERR_PTR(error);
2739 struct file *do_filp_open(int dfd, const char *pathname,
2740 const struct open_flags *op, int flags)
2742 struct nameidata nd;
2745 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2746 if (unlikely(filp == ERR_PTR(-ECHILD)))
2747 filp = path_openat(dfd, pathname, &nd, op, flags);
2748 if (unlikely(filp == ERR_PTR(-ESTALE)))
2749 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2753 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2754 const char *name, const struct open_flags *op, int flags)
2756 struct nameidata nd;
2760 nd.root.dentry = dentry;
2762 flags |= LOOKUP_ROOT;
2764 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2765 return ERR_PTR(-ELOOP);
2767 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2768 if (unlikely(file == ERR_PTR(-ECHILD)))
2769 file = path_openat(-1, name, &nd, op, flags);
2770 if (unlikely(file == ERR_PTR(-ESTALE)))
2771 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2775 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2777 struct dentry *dentry = ERR_PTR(-EEXIST);
2778 struct nameidata nd;
2779 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2781 return ERR_PTR(error);
2784 * Yucky last component or no last component at all?
2785 * (foo/., foo/.., /////)
2787 if (nd.last_type != LAST_NORM)
2789 nd.flags &= ~LOOKUP_PARENT;
2790 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2791 nd.intent.open.flags = O_EXCL;
2794 * Do the final lookup.
2796 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2797 dentry = lookup_hash(&nd);
2801 if (dentry->d_inode)
2804 * Special case - lookup gave negative, but... we had foo/bar/
2805 * From the vfs_mknod() POV we just have a negative dentry -
2806 * all is fine. Let's be bastards - you had / on the end, you've
2807 * been asking for (non-existent) directory. -ENOENT for you.
2809 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2811 dentry = ERR_PTR(-ENOENT);
2818 dentry = ERR_PTR(-EEXIST);
2820 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2825 EXPORT_SYMBOL(kern_path_create);
2827 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2829 char *tmp = getname(pathname);
2832 return ERR_CAST(tmp);
2833 res = kern_path_create(dfd, tmp, path, is_dir);
2837 EXPORT_SYMBOL(user_path_create);
2839 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2841 int error = may_create(dir, dentry);
2846 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2849 if (!dir->i_op->mknod)
2852 error = devcgroup_inode_mknod(mode, dev);
2856 error = security_inode_mknod(dir, dentry, mode, dev);
2860 error = dir->i_op->mknod(dir, dentry, mode, dev);
2862 fsnotify_create(dir, dentry);
2866 static int may_mknod(umode_t mode)
2868 switch (mode & S_IFMT) {
2874 case 0: /* zero mode translates to S_IFREG */
2883 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2886 struct dentry *dentry;
2893 dentry = user_path_create(dfd, filename, &path, 0);
2895 return PTR_ERR(dentry);
2897 if (!IS_POSIXACL(path.dentry->d_inode))
2898 mode &= ~current_umask();
2899 error = may_mknod(mode);
2902 error = mnt_want_write(path.mnt);
2905 error = security_path_mknod(&path, dentry, mode, dev);
2907 goto out_drop_write;
2908 switch (mode & S_IFMT) {
2909 case 0: case S_IFREG:
2910 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2912 case S_IFCHR: case S_IFBLK:
2913 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2914 new_decode_dev(dev));
2916 case S_IFIFO: case S_IFSOCK:
2917 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2921 mnt_drop_write(path.mnt);
2924 mutex_unlock(&path.dentry->d_inode->i_mutex);
2930 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
2932 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2935 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2937 int error = may_create(dir, dentry);
2938 unsigned max_links = dir->i_sb->s_max_links;
2943 if (!dir->i_op->mkdir)
2946 mode &= (S_IRWXUGO|S_ISVTX);
2947 error = security_inode_mkdir(dir, dentry, mode);
2951 if (max_links && dir->i_nlink >= max_links)
2954 error = dir->i_op->mkdir(dir, dentry, mode);
2956 fsnotify_mkdir(dir, dentry);
2960 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
2962 struct dentry *dentry;
2966 dentry = user_path_create(dfd, pathname, &path, 1);
2968 return PTR_ERR(dentry);
2970 if (!IS_POSIXACL(path.dentry->d_inode))
2971 mode &= ~current_umask();
2972 error = mnt_want_write(path.mnt);
2975 error = security_path_mkdir(&path, dentry, mode);
2977 goto out_drop_write;
2978 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2980 mnt_drop_write(path.mnt);
2983 mutex_unlock(&path.dentry->d_inode->i_mutex);
2988 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
2990 return sys_mkdirat(AT_FDCWD, pathname, mode);
2994 * The dentry_unhash() helper will try to drop the dentry early: we
2995 * should have a usage count of 1 if we're the only user of this
2996 * dentry, and if that is true (possibly after pruning the dcache),
2997 * then we drop the dentry now.
2999 * A low-level filesystem can, if it choses, legally
3002 * if (!d_unhashed(dentry))
3005 * if it cannot handle the case of removing a directory
3006 * that is still in use by something else..
3008 void dentry_unhash(struct dentry *dentry)
3010 shrink_dcache_parent(dentry);
3011 spin_lock(&dentry->d_lock);
3012 if (dentry->d_count == 1)
3014 spin_unlock(&dentry->d_lock);
3017 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3019 int error = may_delete(dir, dentry, 1);
3024 if (!dir->i_op->rmdir)
3028 mutex_lock(&dentry->d_inode->i_mutex);
3031 if (d_mountpoint(dentry))
3034 error = security_inode_rmdir(dir, dentry);
3038 shrink_dcache_parent(dentry);
3039 error = dir->i_op->rmdir(dir, dentry);
3043 dentry->d_inode->i_flags |= S_DEAD;
3047 mutex_unlock(&dentry->d_inode->i_mutex);
3054 static long do_rmdir(int dfd, const char __user *pathname)
3058 struct dentry *dentry;
3059 struct nameidata nd;
3061 error = user_path_parent(dfd, pathname, &nd, &name);
3065 switch(nd.last_type) {
3077 nd.flags &= ~LOOKUP_PARENT;
3079 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3080 dentry = lookup_hash(&nd);
3081 error = PTR_ERR(dentry);
3084 if (!dentry->d_inode) {
3088 error = mnt_want_write(nd.path.mnt);
3091 error = security_path_rmdir(&nd.path, dentry);
3094 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3096 mnt_drop_write(nd.path.mnt);
3100 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3107 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3109 return do_rmdir(AT_FDCWD, pathname);
3112 int vfs_unlink(struct inode *dir, struct dentry *dentry)
3114 int error = may_delete(dir, dentry, 0);
3119 if (!dir->i_op->unlink)
3122 mutex_lock(&dentry->d_inode->i_mutex);
3123 if (d_mountpoint(dentry))
3126 error = security_inode_unlink(dir, dentry);
3128 error = dir->i_op->unlink(dir, dentry);
3133 mutex_unlock(&dentry->d_inode->i_mutex);
3135 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3136 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3137 fsnotify_link_count(dentry->d_inode);
3145 * Make sure that the actual truncation of the file will occur outside its
3146 * directory's i_mutex. Truncate can take a long time if there is a lot of
3147 * writeout happening, and we don't want to prevent access to the directory
3148 * while waiting on the I/O.
3150 static long do_unlinkat(int dfd, const char __user *pathname)
3154 struct dentry *dentry;
3155 struct nameidata nd;
3156 struct inode *inode = NULL;
3158 error = user_path_parent(dfd, pathname, &nd, &name);
3163 if (nd.last_type != LAST_NORM)
3166 nd.flags &= ~LOOKUP_PARENT;
3168 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3169 dentry = lookup_hash(&nd);
3170 error = PTR_ERR(dentry);
3171 if (!IS_ERR(dentry)) {
3172 /* Why not before? Because we want correct error value */
3173 if (nd.last.name[nd.last.len])
3175 inode = dentry->d_inode;
3179 error = mnt_want_write(nd.path.mnt);
3182 error = security_path_unlink(&nd.path, dentry);
3185 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
3187 mnt_drop_write(nd.path.mnt);
3191 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3193 iput(inode); /* truncate the inode here */
3200 error = !dentry->d_inode ? -ENOENT :
3201 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3205 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3207 if ((flag & ~AT_REMOVEDIR) != 0)
3210 if (flag & AT_REMOVEDIR)
3211 return do_rmdir(dfd, pathname);
3213 return do_unlinkat(dfd, pathname);
3216 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3218 return do_unlinkat(AT_FDCWD, pathname);
3221 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3223 int error = may_create(dir, dentry);
3228 if (!dir->i_op->symlink)
3231 error = security_inode_symlink(dir, dentry, oldname);
3235 error = dir->i_op->symlink(dir, dentry, oldname);
3237 fsnotify_create(dir, dentry);
3241 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3242 int, newdfd, const char __user *, newname)
3246 struct dentry *dentry;
3249 from = getname(oldname);
3251 return PTR_ERR(from);
3253 dentry = user_path_create(newdfd, newname, &path, 0);
3254 error = PTR_ERR(dentry);
3258 error = mnt_want_write(path.mnt);
3261 error = security_path_symlink(&path, dentry, from);
3263 goto out_drop_write;
3264 error = vfs_symlink(path.dentry->d_inode, dentry, from);
3266 mnt_drop_write(path.mnt);
3269 mutex_unlock(&path.dentry->d_inode->i_mutex);
3276 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3278 return sys_symlinkat(oldname, AT_FDCWD, newname);
3281 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3283 struct inode *inode = old_dentry->d_inode;
3284 unsigned max_links = dir->i_sb->s_max_links;
3290 error = may_create(dir, new_dentry);
3294 if (dir->i_sb != inode->i_sb)
3298 * A link to an append-only or immutable file cannot be created.
3300 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3302 if (!dir->i_op->link)
3304 if (S_ISDIR(inode->i_mode))
3307 error = security_inode_link(old_dentry, dir, new_dentry);
3311 mutex_lock(&inode->i_mutex);
3312 /* Make sure we don't allow creating hardlink to an unlinked file */
3313 if (inode->i_nlink == 0)
3315 else if (max_links && inode->i_nlink >= max_links)
3318 error = dir->i_op->link(old_dentry, dir, new_dentry);
3319 mutex_unlock(&inode->i_mutex);
3321 fsnotify_link(dir, inode, new_dentry);
3326 * Hardlinks are often used in delicate situations. We avoid
3327 * security-related surprises by not following symlinks on the
3330 * We don't follow them on the oldname either to be compatible
3331 * with linux 2.0, and to avoid hard-linking to directories
3332 * and other special files. --ADM
3334 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3335 int, newdfd, const char __user *, newname, int, flags)
3337 struct dentry *new_dentry;
3338 struct path old_path, new_path;
3342 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3345 * To use null names we require CAP_DAC_READ_SEARCH
3346 * This ensures that not everyone will be able to create
3347 * handlink using the passed filedescriptor.
3349 if (flags & AT_EMPTY_PATH) {
3350 if (!capable(CAP_DAC_READ_SEARCH))
3355 if (flags & AT_SYMLINK_FOLLOW)
3356 how |= LOOKUP_FOLLOW;
3358 error = user_path_at(olddfd, oldname, how, &old_path);
3362 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3363 error = PTR_ERR(new_dentry);
3364 if (IS_ERR(new_dentry))
3368 if (old_path.mnt != new_path.mnt)
3370 error = mnt_want_write(new_path.mnt);
3373 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3375 goto out_drop_write;
3376 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3378 mnt_drop_write(new_path.mnt);
3381 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3382 path_put(&new_path);
3384 path_put(&old_path);
3389 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3391 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3395 * The worst of all namespace operations - renaming directory. "Perverted"
3396 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3398 * a) we can get into loop creation. Check is done in is_subdir().
3399 * b) race potential - two innocent renames can create a loop together.
3400 * That's where 4.4 screws up. Current fix: serialization on
3401 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3403 * c) we have to lock _three_ objects - parents and victim (if it exists).
3404 * And that - after we got ->i_mutex on parents (until then we don't know
3405 * whether the target exists). Solution: try to be smart with locking
3406 * order for inodes. We rely on the fact that tree topology may change
3407 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3408 * move will be locked. Thus we can rank directories by the tree
3409 * (ancestors first) and rank all non-directories after them.
3410 * That works since everybody except rename does "lock parent, lookup,
3411 * lock child" and rename is under ->s_vfs_rename_mutex.
3412 * HOWEVER, it relies on the assumption that any object with ->lookup()
3413 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3414 * we'd better make sure that there's no link(2) for them.
3415 * d) conversion from fhandle to dentry may come in the wrong moment - when
3416 * we are removing the target. Solution: we will have to grab ->i_mutex
3417 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3418 * ->i_mutex on parents, which works but leads to some truly excessive
3421 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3422 struct inode *new_dir, struct dentry *new_dentry)
3425 struct inode *target = new_dentry->d_inode;
3426 unsigned max_links = new_dir->i_sb->s_max_links;
3429 * If we are going to change the parent - check write permissions,
3430 * we'll need to flip '..'.
3432 if (new_dir != old_dir) {
3433 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3438 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3444 mutex_lock(&target->i_mutex);
3447 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3451 if (max_links && !target && new_dir != old_dir &&
3452 new_dir->i_nlink >= max_links)
3456 shrink_dcache_parent(new_dentry);
3457 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3462 target->i_flags |= S_DEAD;
3463 dont_mount(new_dentry);
3467 mutex_unlock(&target->i_mutex);
3470 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3471 d_move(old_dentry,new_dentry);
3475 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3476 struct inode *new_dir, struct dentry *new_dentry)
3478 struct inode *target = new_dentry->d_inode;
3481 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3487 mutex_lock(&target->i_mutex);
3490 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3493 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3498 dont_mount(new_dentry);
3499 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3500 d_move(old_dentry, new_dentry);
3503 mutex_unlock(&target->i_mutex);
3508 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3509 struct inode *new_dir, struct dentry *new_dentry)
3512 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3513 const unsigned char *old_name;
3515 if (old_dentry->d_inode == new_dentry->d_inode)
3518 error = may_delete(old_dir, old_dentry, is_dir);
3522 if (!new_dentry->d_inode)
3523 error = may_create(new_dir, new_dentry);
3525 error = may_delete(new_dir, new_dentry, is_dir);
3529 if (!old_dir->i_op->rename)
3532 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3535 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3537 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3539 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3540 new_dentry->d_inode, old_dentry);
3541 fsnotify_oldname_free(old_name);
3546 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3547 int, newdfd, const char __user *, newname)
3549 struct dentry *old_dir, *new_dir;
3550 struct dentry *old_dentry, *new_dentry;
3551 struct dentry *trap;
3552 struct nameidata oldnd, newnd;
3557 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3561 error = user_path_parent(newdfd, newname, &newnd, &to);
3566 if (oldnd.path.mnt != newnd.path.mnt)
3569 old_dir = oldnd.path.dentry;
3571 if (oldnd.last_type != LAST_NORM)
3574 new_dir = newnd.path.dentry;
3575 if (newnd.last_type != LAST_NORM)
3578 oldnd.flags &= ~LOOKUP_PARENT;
3579 newnd.flags &= ~LOOKUP_PARENT;
3580 newnd.flags |= LOOKUP_RENAME_TARGET;
3582 trap = lock_rename(new_dir, old_dir);
3584 old_dentry = lookup_hash(&oldnd);
3585 error = PTR_ERR(old_dentry);
3586 if (IS_ERR(old_dentry))
3588 /* source must exist */
3590 if (!old_dentry->d_inode)
3592 /* unless the source is a directory trailing slashes give -ENOTDIR */
3593 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3595 if (oldnd.last.name[oldnd.last.len])
3597 if (newnd.last.name[newnd.last.len])
3600 /* source should not be ancestor of target */
3602 if (old_dentry == trap)
3604 new_dentry = lookup_hash(&newnd);
3605 error = PTR_ERR(new_dentry);
3606 if (IS_ERR(new_dentry))
3608 /* target should not be an ancestor of source */
3610 if (new_dentry == trap)
3613 error = mnt_want_write(oldnd.path.mnt);
3616 error = security_path_rename(&oldnd.path, old_dentry,
3617 &newnd.path, new_dentry);
3620 error = vfs_rename(old_dir->d_inode, old_dentry,
3621 new_dir->d_inode, new_dentry);
3623 mnt_drop_write(oldnd.path.mnt);
3629 unlock_rename(new_dir, old_dir);
3631 path_put(&newnd.path);
3634 path_put(&oldnd.path);
3640 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3642 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3645 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3649 len = PTR_ERR(link);
3654 if (len > (unsigned) buflen)
3656 if (copy_to_user(buffer, link, len))
3663 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3664 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3665 * using) it for any given inode is up to filesystem.
3667 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3669 struct nameidata nd;
3674 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3676 return PTR_ERR(cookie);
3678 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3679 if (dentry->d_inode->i_op->put_link)
3680 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3684 int vfs_follow_link(struct nameidata *nd, const char *link)
3686 return __vfs_follow_link(nd, link);
3689 /* get the link contents into pagecache */
3690 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3694 struct address_space *mapping = dentry->d_inode->i_mapping;
3695 page = read_mapping_page(mapping, 0, NULL);
3700 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3704 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3706 struct page *page = NULL;
3707 char *s = page_getlink(dentry, &page);
3708 int res = vfs_readlink(dentry,buffer,buflen,s);
3711 page_cache_release(page);
3716 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3718 struct page *page = NULL;
3719 nd_set_link(nd, page_getlink(dentry, &page));
3723 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3725 struct page *page = cookie;
3729 page_cache_release(page);
3734 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3736 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3738 struct address_space *mapping = inode->i_mapping;
3743 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3745 flags |= AOP_FLAG_NOFS;
3748 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3749 flags, &page, &fsdata);
3753 kaddr = kmap_atomic(page);
3754 memcpy(kaddr, symname, len-1);
3755 kunmap_atomic(kaddr);
3757 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3764 mark_inode_dirty(inode);
3770 int page_symlink(struct inode *inode, const char *symname, int len)
3772 return __page_symlink(inode, symname, len,
3773 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3776 const struct inode_operations page_symlink_inode_operations = {
3777 .readlink = generic_readlink,
3778 .follow_link = page_follow_link_light,
3779 .put_link = page_put_link,
3782 EXPORT_SYMBOL(user_path_at);
3783 EXPORT_SYMBOL(follow_down_one);
3784 EXPORT_SYMBOL(follow_down);
3785 EXPORT_SYMBOL(follow_up);
3786 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3787 EXPORT_SYMBOL(getname);
3788 EXPORT_SYMBOL(lock_rename);
3789 EXPORT_SYMBOL(lookup_one_len);
3790 EXPORT_SYMBOL(page_follow_link_light);
3791 EXPORT_SYMBOL(page_put_link);
3792 EXPORT_SYMBOL(page_readlink);
3793 EXPORT_SYMBOL(__page_symlink);
3794 EXPORT_SYMBOL(page_symlink);
3795 EXPORT_SYMBOL(page_symlink_inode_operations);
3796 EXPORT_SYMBOL(kern_path);
3797 EXPORT_SYMBOL(vfs_path_lookup);
3798 EXPORT_SYMBOL(inode_permission);
3799 EXPORT_SYMBOL(unlock_rename);
3800 EXPORT_SYMBOL(vfs_create);
3801 EXPORT_SYMBOL(vfs_follow_link);
3802 EXPORT_SYMBOL(vfs_link);
3803 EXPORT_SYMBOL(vfs_mkdir);
3804 EXPORT_SYMBOL(vfs_mknod);
3805 EXPORT_SYMBOL(generic_permission);
3806 EXPORT_SYMBOL(vfs_readlink);
3807 EXPORT_SYMBOL(vfs_rename);
3808 EXPORT_SYMBOL(vfs_rmdir);
3809 EXPORT_SYMBOL(vfs_symlink);
3810 EXPORT_SYMBOL(vfs_unlink);
3811 EXPORT_SYMBOL(dentry_unhash);
3812 EXPORT_SYMBOL(generic_readlink);