4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <linux/posix_acl.h>
36 #include <asm/uaccess.h>
41 /* [Feb-1997 T. Schoebel-Theuer]
42 * Fundamental changes in the pathname lookup mechanisms (namei)
43 * were necessary because of omirr. The reason is that omirr needs
44 * to know the _real_ pathname, not the user-supplied one, in case
45 * of symlinks (and also when transname replacements occur).
47 * The new code replaces the old recursive symlink resolution with
48 * an iterative one (in case of non-nested symlink chains). It does
49 * this with calls to <fs>_follow_link().
50 * As a side effect, dir_namei(), _namei() and follow_link() are now
51 * replaced with a single function lookup_dentry() that can handle all
52 * the special cases of the former code.
54 * With the new dcache, the pathname is stored at each inode, at least as
55 * long as the refcount of the inode is positive. As a side effect, the
56 * size of the dcache depends on the inode cache and thus is dynamic.
58 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
59 * resolution to correspond with current state of the code.
61 * Note that the symlink resolution is not *completely* iterative.
62 * There is still a significant amount of tail- and mid- recursion in
63 * the algorithm. Also, note that <fs>_readlink() is not used in
64 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
65 * may return different results than <fs>_follow_link(). Many virtual
66 * filesystems (including /proc) exhibit this behavior.
69 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
70 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
71 * and the name already exists in form of a symlink, try to create the new
72 * name indicated by the symlink. The old code always complained that the
73 * name already exists, due to not following the symlink even if its target
74 * is nonexistent. The new semantics affects also mknod() and link() when
75 * the name is a symlink pointing to a non-existent name.
77 * I don't know which semantics is the right one, since I have no access
78 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
79 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
80 * "old" one. Personally, I think the new semantics is much more logical.
81 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
82 * file does succeed in both HP-UX and SunOs, but not in Solaris
83 * and in the old Linux semantics.
86 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
87 * semantics. See the comments in "open_namei" and "do_link" below.
89 * [10-Sep-98 Alan Modra] Another symlink change.
92 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
93 * inside the path - always follow.
94 * in the last component in creation/removal/renaming - never follow.
95 * if LOOKUP_FOLLOW passed - follow.
96 * if the pathname has trailing slashes - follow.
97 * otherwise - don't follow.
98 * (applied in that order).
100 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
101 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
102 * During the 2.4 we need to fix the userland stuff depending on it -
103 * hopefully we will be able to get rid of that wart in 2.5. So far only
104 * XEmacs seems to be relying on it...
107 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
108 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
109 * any extra contention...
112 /* In order to reduce some races, while at the same time doing additional
113 * checking and hopefully speeding things up, we copy filenames to the
114 * kernel data space before using them..
116 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
117 * PATH_MAX includes the nul terminator --RR.
119 static int do_getname(const char __user *filename, char *page)
122 unsigned long len = PATH_MAX;
124 if (!segment_eq(get_fs(), KERNEL_DS)) {
125 if ((unsigned long) filename >= TASK_SIZE)
127 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
128 len = TASK_SIZE - (unsigned long) filename;
131 retval = strncpy_from_user(page, filename, len);
135 return -ENAMETOOLONG;
141 static char *getname_flags(const char __user *filename, int flags, int *empty)
143 char *result = __getname();
147 return ERR_PTR(-ENOMEM);
149 retval = do_getname(filename, result);
151 if (retval == -ENOENT && empty)
153 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
155 return ERR_PTR(retval);
158 audit_getname(result);
162 char *getname(const char __user * filename)
164 return getname_flags(filename, 0, NULL);
167 #ifdef CONFIG_AUDITSYSCALL
168 void putname(const char *name)
170 if (unlikely(!audit_dummy_context()))
175 EXPORT_SYMBOL(putname);
178 static int check_acl(struct inode *inode, int mask)
180 #ifdef CONFIG_FS_POSIX_ACL
181 struct posix_acl *acl;
183 if (mask & MAY_NOT_BLOCK) {
184 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
187 /* no ->get_acl() calls in RCU mode... */
188 if (acl == ACL_NOT_CACHED)
190 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
193 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
196 * A filesystem can force a ACL callback by just never filling the
197 * ACL cache. But normally you'd fill the cache either at inode
198 * instantiation time, or on the first ->get_acl call.
200 * If the filesystem doesn't have a get_acl() function at all, we'll
201 * just create the negative cache entry.
203 if (acl == ACL_NOT_CACHED) {
204 if (inode->i_op->get_acl) {
205 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
209 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
215 int error = posix_acl_permission(inode, acl, mask);
216 posix_acl_release(acl);
225 * This does the basic permission checking
227 static int acl_permission_check(struct inode *inode, int mask)
229 unsigned int mode = inode->i_mode;
231 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
234 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
235 int error = check_acl(inode, mask);
236 if (error != -EAGAIN)
240 if (in_group_p(inode->i_gid))
245 * If the DACs are ok we don't need any capability check.
247 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
253 * generic_permission - check for access rights on a Posix-like filesystem
254 * @inode: inode to check access rights for
255 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
257 * Used to check for read/write/execute permissions on a file.
258 * We use "fsuid" for this, letting us set arbitrary permissions
259 * for filesystem access without changing the "normal" uids which
260 * are used for other things.
262 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
263 * request cannot be satisfied (eg. requires blocking or too much complexity).
264 * It would then be called again in ref-walk mode.
266 int generic_permission(struct inode *inode, int mask)
271 * Do the basic permission checks.
273 ret = acl_permission_check(inode, mask);
277 if (S_ISDIR(inode->i_mode)) {
278 /* DACs are overridable for directories */
279 if (inode_capable(inode, CAP_DAC_OVERRIDE))
281 if (!(mask & MAY_WRITE))
282 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
287 * Read/write DACs are always overridable.
288 * Executable DACs are overridable when there is
289 * at least one exec bit set.
291 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
292 if (inode_capable(inode, CAP_DAC_OVERRIDE))
296 * Searching includes executable on directories, else just read.
298 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
299 if (mask == MAY_READ)
300 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
307 * We _really_ want to just do "generic_permission()" without
308 * even looking at the inode->i_op values. So we keep a cache
309 * flag in inode->i_opflags, that says "this has not special
310 * permission function, use the fast case".
312 static inline int do_inode_permission(struct inode *inode, int mask)
314 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
315 if (likely(inode->i_op->permission))
316 return inode->i_op->permission(inode, mask);
318 /* This gets set once for the inode lifetime */
319 spin_lock(&inode->i_lock);
320 inode->i_opflags |= IOP_FASTPERM;
321 spin_unlock(&inode->i_lock);
323 return generic_permission(inode, mask);
327 * inode_permission - check for access rights to a given inode
328 * @inode: inode to check permission on
329 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
331 * Used to check for read/write/execute permissions on an inode.
332 * We use "fsuid" for this, letting us set arbitrary permissions
333 * for filesystem access without changing the "normal" uids which
334 * are used for other things.
336 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
338 int inode_permission(struct inode *inode, int mask)
342 if (unlikely(mask & MAY_WRITE)) {
343 umode_t mode = inode->i_mode;
346 * Nobody gets write access to a read-only fs.
348 if (IS_RDONLY(inode) &&
349 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
353 * Nobody gets write access to an immutable file.
355 if (IS_IMMUTABLE(inode))
359 retval = do_inode_permission(inode, mask);
363 retval = devcgroup_inode_permission(inode, mask);
367 return security_inode_permission(inode, mask);
371 * path_get - get a reference to a path
372 * @path: path to get the reference to
374 * Given a path increment the reference count to the dentry and the vfsmount.
376 void path_get(struct path *path)
381 EXPORT_SYMBOL(path_get);
384 * path_put - put a reference to a path
385 * @path: path to put the reference to
387 * Given a path decrement the reference count to the dentry and the vfsmount.
389 void path_put(struct path *path)
394 EXPORT_SYMBOL(path_put);
397 * Path walking has 2 modes, rcu-walk and ref-walk (see
398 * Documentation/filesystems/path-lookup.txt). In situations when we can't
399 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
400 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
401 * mode. Refcounts are grabbed at the last known good point before rcu-walk
402 * got stuck, so ref-walk may continue from there. If this is not successful
403 * (eg. a seqcount has changed), then failure is returned and it's up to caller
404 * to restart the path walk from the beginning in ref-walk mode.
408 * unlazy_walk - try to switch to ref-walk mode.
409 * @nd: nameidata pathwalk data
410 * @dentry: child of nd->path.dentry or NULL
411 * Returns: 0 on success, -ECHILD on failure
413 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
414 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
415 * @nd or NULL. Must be called from rcu-walk context.
417 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
419 struct fs_struct *fs = current->fs;
420 struct dentry *parent = nd->path.dentry;
423 BUG_ON(!(nd->flags & LOOKUP_RCU));
424 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
426 spin_lock(&fs->lock);
427 if (nd->root.mnt != fs->root.mnt ||
428 nd->root.dentry != fs->root.dentry)
431 spin_lock(&parent->d_lock);
433 if (!__d_rcu_to_refcount(parent, nd->seq))
435 BUG_ON(nd->inode != parent->d_inode);
437 if (dentry->d_parent != parent)
439 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
440 if (!__d_rcu_to_refcount(dentry, nd->seq))
443 * If the sequence check on the child dentry passed, then
444 * the child has not been removed from its parent. This
445 * means the parent dentry must be valid and able to take
446 * a reference at this point.
448 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
449 BUG_ON(!parent->d_count);
451 spin_unlock(&dentry->d_lock);
453 spin_unlock(&parent->d_lock);
456 spin_unlock(&fs->lock);
458 mntget(nd->path.mnt);
461 br_read_unlock(vfsmount_lock);
462 nd->flags &= ~LOOKUP_RCU;
466 spin_unlock(&dentry->d_lock);
468 spin_unlock(&parent->d_lock);
471 spin_unlock(&fs->lock);
476 * release_open_intent - free up open intent resources
477 * @nd: pointer to nameidata
479 void release_open_intent(struct nameidata *nd)
481 struct file *file = nd->intent.open.file;
483 if (file && !IS_ERR(file)) {
484 if (file->f_path.dentry == NULL)
491 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
493 return dentry->d_op->d_revalidate(dentry, nd);
497 * complete_walk - successful completion of path walk
498 * @nd: pointer nameidata
500 * If we had been in RCU mode, drop out of it and legitimize nd->path.
501 * Revalidate the final result, unless we'd already done that during
502 * the path walk or the filesystem doesn't ask for it. Return 0 on
503 * success, -error on failure. In case of failure caller does not
504 * need to drop nd->path.
506 static int complete_walk(struct nameidata *nd)
508 struct dentry *dentry = nd->path.dentry;
511 if (nd->flags & LOOKUP_RCU) {
512 nd->flags &= ~LOOKUP_RCU;
513 if (!(nd->flags & LOOKUP_ROOT))
515 spin_lock(&dentry->d_lock);
516 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
517 spin_unlock(&dentry->d_lock);
519 br_read_unlock(vfsmount_lock);
522 BUG_ON(nd->inode != dentry->d_inode);
523 spin_unlock(&dentry->d_lock);
524 mntget(nd->path.mnt);
526 br_read_unlock(vfsmount_lock);
529 if (likely(!(nd->flags & LOOKUP_JUMPED)))
532 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
535 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
538 /* Note: we do not d_invalidate() */
539 status = d_revalidate(dentry, nd);
550 static __always_inline void set_root(struct nameidata *nd)
553 get_fs_root(current->fs, &nd->root);
556 static int link_path_walk(const char *, struct nameidata *);
558 static __always_inline void set_root_rcu(struct nameidata *nd)
561 struct fs_struct *fs = current->fs;
565 seq = read_seqcount_begin(&fs->seq);
567 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
568 } while (read_seqcount_retry(&fs->seq, seq));
572 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
584 nd->flags |= LOOKUP_JUMPED;
586 nd->inode = nd->path.dentry->d_inode;
588 ret = link_path_walk(link, nd);
592 return PTR_ERR(link);
595 static void path_put_conditional(struct path *path, struct nameidata *nd)
598 if (path->mnt != nd->path.mnt)
602 static inline void path_to_nameidata(const struct path *path,
603 struct nameidata *nd)
605 if (!(nd->flags & LOOKUP_RCU)) {
606 dput(nd->path.dentry);
607 if (nd->path.mnt != path->mnt)
608 mntput(nd->path.mnt);
610 nd->path.mnt = path->mnt;
611 nd->path.dentry = path->dentry;
614 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
616 struct inode *inode = link->dentry->d_inode;
617 if (!IS_ERR(cookie) && inode->i_op->put_link)
618 inode->i_op->put_link(link->dentry, nd, cookie);
622 static __always_inline int
623 follow_link(struct path *link, struct nameidata *nd, void **p)
626 struct dentry *dentry = link->dentry;
628 BUG_ON(nd->flags & LOOKUP_RCU);
630 if (link->mnt == nd->path.mnt)
633 if (unlikely(current->total_link_count >= 40)) {
634 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
639 current->total_link_count++;
642 nd_set_link(nd, NULL);
644 error = security_inode_follow_link(link->dentry, nd);
646 *p = ERR_PTR(error); /* no ->put_link(), please */
651 nd->last_type = LAST_BIND;
652 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
655 char *s = nd_get_link(nd);
658 error = __vfs_follow_link(nd, s);
659 else if (nd->last_type == LAST_BIND) {
660 nd->flags |= LOOKUP_JUMPED;
661 nd->inode = nd->path.dentry->d_inode;
662 if (nd->inode->i_op->follow_link) {
663 /* stepped on a _really_ weird one */
672 static int follow_up_rcu(struct path *path)
674 struct mount *mnt = real_mount(path->mnt);
675 struct mount *parent;
676 struct dentry *mountpoint;
678 parent = mnt->mnt_parent;
679 if (&parent->mnt == path->mnt)
681 mountpoint = mnt->mnt_mountpoint;
682 path->dentry = mountpoint;
683 path->mnt = &parent->mnt;
687 int follow_up(struct path *path)
689 struct mount *mnt = real_mount(path->mnt);
690 struct mount *parent;
691 struct dentry *mountpoint;
693 br_read_lock(vfsmount_lock);
694 parent = mnt->mnt_parent;
695 if (&parent->mnt == path->mnt) {
696 br_read_unlock(vfsmount_lock);
699 mntget(&parent->mnt);
700 mountpoint = dget(mnt->mnt_mountpoint);
701 br_read_unlock(vfsmount_lock);
703 path->dentry = mountpoint;
705 path->mnt = &parent->mnt;
710 * Perform an automount
711 * - return -EISDIR to tell follow_managed() to stop and return the path we
714 static int follow_automount(struct path *path, unsigned flags,
717 struct vfsmount *mnt;
720 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
723 /* We don't want to mount if someone's just doing a stat -
724 * unless they're stat'ing a directory and appended a '/' to
727 * We do, however, want to mount if someone wants to open or
728 * create a file of any type under the mountpoint, wants to
729 * traverse through the mountpoint or wants to open the
730 * mounted directory. Also, autofs may mark negative dentries
731 * as being automount points. These will need the attentions
732 * of the daemon to instantiate them before they can be used.
734 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
735 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
736 path->dentry->d_inode)
739 current->total_link_count++;
740 if (current->total_link_count >= 40)
743 mnt = path->dentry->d_op->d_automount(path);
746 * The filesystem is allowed to return -EISDIR here to indicate
747 * it doesn't want to automount. For instance, autofs would do
748 * this so that its userspace daemon can mount on this dentry.
750 * However, we can only permit this if it's a terminal point in
751 * the path being looked up; if it wasn't then the remainder of
752 * the path is inaccessible and we should say so.
754 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
759 if (!mnt) /* mount collision */
763 /* lock_mount() may release path->mnt on error */
767 err = finish_automount(mnt, path);
771 /* Someone else made a mount here whilst we were busy */
776 path->dentry = dget(mnt->mnt_root);
785 * Handle a dentry that is managed in some way.
786 * - Flagged for transit management (autofs)
787 * - Flagged as mountpoint
788 * - Flagged as automount point
790 * This may only be called in refwalk mode.
792 * Serialization is taken care of in namespace.c
794 static int follow_managed(struct path *path, unsigned flags)
796 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
798 bool need_mntput = false;
801 /* Given that we're not holding a lock here, we retain the value in a
802 * local variable for each dentry as we look at it so that we don't see
803 * the components of that value change under us */
804 while (managed = ACCESS_ONCE(path->dentry->d_flags),
805 managed &= DCACHE_MANAGED_DENTRY,
806 unlikely(managed != 0)) {
807 /* Allow the filesystem to manage the transit without i_mutex
809 if (managed & DCACHE_MANAGE_TRANSIT) {
810 BUG_ON(!path->dentry->d_op);
811 BUG_ON(!path->dentry->d_op->d_manage);
812 ret = path->dentry->d_op->d_manage(path->dentry, false);
817 /* Transit to a mounted filesystem. */
818 if (managed & DCACHE_MOUNTED) {
819 struct vfsmount *mounted = lookup_mnt(path);
825 path->dentry = dget(mounted->mnt_root);
830 /* Something is mounted on this dentry in another
831 * namespace and/or whatever was mounted there in this
832 * namespace got unmounted before we managed to get the
836 /* Handle an automount point */
837 if (managed & DCACHE_NEED_AUTOMOUNT) {
838 ret = follow_automount(path, flags, &need_mntput);
844 /* We didn't change the current path point */
848 if (need_mntput && path->mnt == mnt)
852 return ret < 0 ? ret : need_mntput;
855 int follow_down_one(struct path *path)
857 struct vfsmount *mounted;
859 mounted = lookup_mnt(path);
864 path->dentry = dget(mounted->mnt_root);
870 static inline bool managed_dentry_might_block(struct dentry *dentry)
872 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
873 dentry->d_op->d_manage(dentry, true) < 0);
877 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
878 * we meet a managed dentry that would need blocking.
880 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
881 struct inode **inode)
884 struct mount *mounted;
886 * Don't forget we might have a non-mountpoint managed dentry
887 * that wants to block transit.
889 if (unlikely(managed_dentry_might_block(path->dentry)))
892 if (!d_mountpoint(path->dentry))
895 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
898 path->mnt = &mounted->mnt;
899 path->dentry = mounted->mnt.mnt_root;
900 nd->flags |= LOOKUP_JUMPED;
901 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
903 * Update the inode too. We don't need to re-check the
904 * dentry sequence number here after this d_inode read,
905 * because a mount-point is always pinned.
907 *inode = path->dentry->d_inode;
912 static void follow_mount_rcu(struct nameidata *nd)
914 while (d_mountpoint(nd->path.dentry)) {
915 struct mount *mounted;
916 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
919 nd->path.mnt = &mounted->mnt;
920 nd->path.dentry = mounted->mnt.mnt_root;
921 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
925 static int follow_dotdot_rcu(struct nameidata *nd)
930 if (nd->path.dentry == nd->root.dentry &&
931 nd->path.mnt == nd->root.mnt) {
934 if (nd->path.dentry != nd->path.mnt->mnt_root) {
935 struct dentry *old = nd->path.dentry;
936 struct dentry *parent = old->d_parent;
939 seq = read_seqcount_begin(&parent->d_seq);
940 if (read_seqcount_retry(&old->d_seq, nd->seq))
942 nd->path.dentry = parent;
946 if (!follow_up_rcu(&nd->path))
948 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
950 follow_mount_rcu(nd);
951 nd->inode = nd->path.dentry->d_inode;
955 nd->flags &= ~LOOKUP_RCU;
956 if (!(nd->flags & LOOKUP_ROOT))
959 br_read_unlock(vfsmount_lock);
964 * Follow down to the covering mount currently visible to userspace. At each
965 * point, the filesystem owning that dentry may be queried as to whether the
966 * caller is permitted to proceed or not.
968 int follow_down(struct path *path)
973 while (managed = ACCESS_ONCE(path->dentry->d_flags),
974 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
975 /* Allow the filesystem to manage the transit without i_mutex
978 * We indicate to the filesystem if someone is trying to mount
979 * something here. This gives autofs the chance to deny anyone
980 * other than its daemon the right to mount on its
983 * The filesystem may sleep at this point.
985 if (managed & DCACHE_MANAGE_TRANSIT) {
986 BUG_ON(!path->dentry->d_op);
987 BUG_ON(!path->dentry->d_op->d_manage);
988 ret = path->dentry->d_op->d_manage(
989 path->dentry, false);
991 return ret == -EISDIR ? 0 : ret;
994 /* Transit to a mounted filesystem. */
995 if (managed & DCACHE_MOUNTED) {
996 struct vfsmount *mounted = lookup_mnt(path);
1001 path->mnt = mounted;
1002 path->dentry = dget(mounted->mnt_root);
1006 /* Don't handle automount points here */
1013 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1015 static void follow_mount(struct path *path)
1017 while (d_mountpoint(path->dentry)) {
1018 struct vfsmount *mounted = lookup_mnt(path);
1023 path->mnt = mounted;
1024 path->dentry = dget(mounted->mnt_root);
1028 static void follow_dotdot(struct nameidata *nd)
1033 struct dentry *old = nd->path.dentry;
1035 if (nd->path.dentry == nd->root.dentry &&
1036 nd->path.mnt == nd->root.mnt) {
1039 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1040 /* rare case of legitimate dget_parent()... */
1041 nd->path.dentry = dget_parent(nd->path.dentry);
1045 if (!follow_up(&nd->path))
1048 follow_mount(&nd->path);
1049 nd->inode = nd->path.dentry->d_inode;
1053 * This looks up the name in dcache, possibly revalidates the old dentry and
1054 * allocates a new one if not found or not valid. In the need_lookup argument
1055 * returns whether i_op->lookup is necessary.
1057 * dir->d_inode->i_mutex must be held
1059 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1060 struct nameidata *nd, bool *need_lookup)
1062 struct dentry *dentry;
1065 *need_lookup = false;
1066 dentry = d_lookup(dir, name);
1068 if (d_need_lookup(dentry)) {
1069 *need_lookup = true;
1070 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1071 error = d_revalidate(dentry, nd);
1072 if (unlikely(error <= 0)) {
1075 return ERR_PTR(error);
1076 } else if (!d_invalidate(dentry)) {
1085 dentry = d_alloc(dir, name);
1086 if (unlikely(!dentry))
1087 return ERR_PTR(-ENOMEM);
1089 *need_lookup = true;
1095 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1096 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1098 * dir->d_inode->i_mutex must be held
1100 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1101 struct nameidata *nd)
1105 /* Don't create child dentry for a dead directory. */
1106 if (unlikely(IS_DEADDIR(dir))) {
1108 return ERR_PTR(-ENOENT);
1111 old = dir->i_op->lookup(dir, dentry, nd);
1112 if (unlikely(old)) {
1119 static struct dentry *__lookup_hash(struct qstr *name,
1120 struct dentry *base, struct nameidata *nd)
1123 struct dentry *dentry;
1125 dentry = lookup_dcache(name, base, nd, &need_lookup);
1129 return lookup_real(base->d_inode, dentry, nd);
1133 * It's more convoluted than I'd like it to be, but... it's still fairly
1134 * small and for now I'd prefer to have fast path as straight as possible.
1135 * It _is_ time-critical.
1137 static int do_lookup(struct nameidata *nd, struct qstr *name,
1138 struct path *path, struct inode **inode)
1140 struct vfsmount *mnt = nd->path.mnt;
1141 struct dentry *dentry, *parent = nd->path.dentry;
1147 * Rename seqlock is not required here because in the off chance
1148 * of a false negative due to a concurrent rename, we're going to
1149 * do the non-racy lookup, below.
1151 if (nd->flags & LOOKUP_RCU) {
1154 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1158 /* Memory barrier in read_seqcount_begin of child is enough */
1159 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1163 if (unlikely(d_need_lookup(dentry)))
1165 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1166 status = d_revalidate(dentry, nd);
1167 if (unlikely(status <= 0)) {
1168 if (status != -ECHILD)
1174 path->dentry = dentry;
1175 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1177 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1181 if (unlazy_walk(nd, dentry))
1184 dentry = __d_lookup(parent, name);
1187 if (unlikely(!dentry))
1190 if (unlikely(d_need_lookup(dentry))) {
1195 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1196 status = d_revalidate(dentry, nd);
1197 if (unlikely(status <= 0)) {
1202 if (!d_invalidate(dentry)) {
1209 path->dentry = dentry;
1210 err = follow_managed(path, nd->flags);
1211 if (unlikely(err < 0)) {
1212 path_put_conditional(path, nd);
1216 nd->flags |= LOOKUP_JUMPED;
1217 *inode = path->dentry->d_inode;
1221 BUG_ON(nd->inode != parent->d_inode);
1223 mutex_lock(&parent->d_inode->i_mutex);
1224 dentry = __lookup_hash(name, parent, nd);
1225 mutex_unlock(&parent->d_inode->i_mutex);
1227 return PTR_ERR(dentry);
1231 static inline int may_lookup(struct nameidata *nd)
1233 if (nd->flags & LOOKUP_RCU) {
1234 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1237 if (unlazy_walk(nd, NULL))
1240 return inode_permission(nd->inode, MAY_EXEC);
1243 static inline int handle_dots(struct nameidata *nd, int type)
1245 if (type == LAST_DOTDOT) {
1246 if (nd->flags & LOOKUP_RCU) {
1247 if (follow_dotdot_rcu(nd))
1255 static void terminate_walk(struct nameidata *nd)
1257 if (!(nd->flags & LOOKUP_RCU)) {
1258 path_put(&nd->path);
1260 nd->flags &= ~LOOKUP_RCU;
1261 if (!(nd->flags & LOOKUP_ROOT))
1262 nd->root.mnt = NULL;
1264 br_read_unlock(vfsmount_lock);
1269 * Do we need to follow links? We _really_ want to be able
1270 * to do this check without having to look at inode->i_op,
1271 * so we keep a cache of "no, this doesn't need follow_link"
1272 * for the common case.
1274 static inline int should_follow_link(struct inode *inode, int follow)
1276 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1277 if (likely(inode->i_op->follow_link))
1280 /* This gets set once for the inode lifetime */
1281 spin_lock(&inode->i_lock);
1282 inode->i_opflags |= IOP_NOFOLLOW;
1283 spin_unlock(&inode->i_lock);
1288 static inline int walk_component(struct nameidata *nd, struct path *path,
1289 struct qstr *name, int type, int follow)
1291 struct inode *inode;
1294 * "." and ".." are special - ".." especially so because it has
1295 * to be able to know about the current root directory and
1296 * parent relationships.
1298 if (unlikely(type != LAST_NORM))
1299 return handle_dots(nd, type);
1300 err = do_lookup(nd, name, path, &inode);
1301 if (unlikely(err)) {
1306 path_to_nameidata(path, nd);
1310 if (should_follow_link(inode, follow)) {
1311 if (nd->flags & LOOKUP_RCU) {
1312 if (unlikely(unlazy_walk(nd, path->dentry))) {
1317 BUG_ON(inode != path->dentry->d_inode);
1320 path_to_nameidata(path, nd);
1326 * This limits recursive symlink follows to 8, while
1327 * limiting consecutive symlinks to 40.
1329 * Without that kind of total limit, nasty chains of consecutive
1330 * symlinks can cause almost arbitrarily long lookups.
1332 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1336 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1337 path_put_conditional(path, nd);
1338 path_put(&nd->path);
1341 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1344 current->link_count++;
1347 struct path link = *path;
1350 res = follow_link(&link, nd, &cookie);
1352 res = walk_component(nd, path, &nd->last,
1353 nd->last_type, LOOKUP_FOLLOW);
1354 put_link(nd, &link, cookie);
1357 current->link_count--;
1363 * We really don't want to look at inode->i_op->lookup
1364 * when we don't have to. So we keep a cache bit in
1365 * the inode ->i_opflags field that says "yes, we can
1366 * do lookup on this inode".
1368 static inline int can_lookup(struct inode *inode)
1370 if (likely(inode->i_opflags & IOP_LOOKUP))
1372 if (likely(!inode->i_op->lookup))
1375 /* We do this once for the lifetime of the inode */
1376 spin_lock(&inode->i_lock);
1377 inode->i_opflags |= IOP_LOOKUP;
1378 spin_unlock(&inode->i_lock);
1383 * We can do the critical dentry name comparison and hashing
1384 * operations one word at a time, but we are limited to:
1386 * - Architectures with fast unaligned word accesses. We could
1387 * do a "get_unaligned()" if this helps and is sufficiently
1390 * - Little-endian machines (so that we can generate the mask
1391 * of low bytes efficiently). Again, we *could* do a byte
1392 * swapping load on big-endian architectures if that is not
1393 * expensive enough to make the optimization worthless.
1395 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1396 * do not trap on the (extremely unlikely) case of a page
1397 * crossing operation.
1399 * - Furthermore, we need an efficient 64-bit compile for the
1400 * 64-bit case in order to generate the "number of bytes in
1401 * the final mask". Again, that could be replaced with a
1402 * efficient population count instruction or similar.
1404 #ifdef CONFIG_DCACHE_WORD_ACCESS
1409 * Jan Achrenius on G+: microoptimized version of
1410 * the simpler "(mask & ONEBYTES) * ONEBYTES >> 56"
1411 * that works for the bytemasks without having to
1414 static inline long count_masked_bytes(unsigned long mask)
1416 return mask*0x0001020304050608ul >> 56;
1419 static inline unsigned int fold_hash(unsigned long hash)
1421 hash += hash >> (8*sizeof(int));
1425 #else /* 32-bit case */
1427 /* Carl Chatfield / Jan Achrenius G+ version for 32-bit */
1428 static inline long count_masked_bytes(long mask)
1430 /* (000000 0000ff 00ffff ffffff) -> ( 1 1 2 3 ) */
1431 long a = (0x0ff0001+mask) >> 23;
1432 /* Fix the 1 for 00 case */
1436 #define fold_hash(x) (x)
1440 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1442 unsigned long a, mask;
1443 unsigned long hash = 0;
1446 a = *(unsigned long *)name;
1447 if (len < sizeof(unsigned long))
1451 name += sizeof(unsigned long);
1452 len -= sizeof(unsigned long);
1456 mask = ~(~0ul << len*8);
1459 return fold_hash(hash);
1461 EXPORT_SYMBOL(full_name_hash);
1463 #define REPEAT_BYTE(x) ((~0ul / 0xff) * (x))
1464 #define ONEBYTES REPEAT_BYTE(0x01)
1465 #define SLASHBYTES REPEAT_BYTE('/')
1466 #define HIGHBITS REPEAT_BYTE(0x80)
1468 /* Return the high bit set in the first byte that is a zero */
1469 static inline unsigned long has_zero(unsigned long a)
1471 return ((a - ONEBYTES) & ~a) & HIGHBITS;
1475 * Calculate the length and hash of the path component, and
1476 * return the length of the component;
1478 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1480 unsigned long a, mask, hash, len;
1483 len = -sizeof(unsigned long);
1485 hash = (hash + a) * 9;
1486 len += sizeof(unsigned long);
1487 a = *(unsigned long *)(name+len);
1488 /* Do we have any NUL or '/' bytes in this word? */
1489 mask = has_zero(a) | has_zero(a ^ SLASHBYTES);
1492 /* The mask *below* the first high bit set */
1493 mask = (mask - 1) & ~mask;
1496 *hashp = fold_hash(hash);
1498 return len + count_masked_bytes(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);
1773 err = lookup_last(nd, &path);
1774 put_link(nd, &link, cookie);
1779 err = complete_walk(nd);
1781 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1782 if (!nd->inode->i_op->lookup) {
1783 path_put(&nd->path);
1791 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1792 path_put(&nd->root);
1793 nd->root.mnt = NULL;
1798 static int do_path_lookup(int dfd, const char *name,
1799 unsigned int flags, struct nameidata *nd)
1801 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1802 if (unlikely(retval == -ECHILD))
1803 retval = path_lookupat(dfd, name, flags, nd);
1804 if (unlikely(retval == -ESTALE))
1805 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1807 if (likely(!retval)) {
1808 if (unlikely(!audit_dummy_context())) {
1809 if (nd->path.dentry && nd->inode)
1810 audit_inode(name, nd->path.dentry);
1816 int kern_path_parent(const char *name, struct nameidata *nd)
1818 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1821 int kern_path(const char *name, unsigned int flags, struct path *path)
1823 struct nameidata nd;
1824 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1831 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1832 * @dentry: pointer to dentry of the base directory
1833 * @mnt: pointer to vfs mount of the base directory
1834 * @name: pointer to file name
1835 * @flags: lookup flags
1836 * @path: pointer to struct path to fill
1838 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1839 const char *name, unsigned int flags,
1842 struct nameidata nd;
1844 nd.root.dentry = dentry;
1846 BUG_ON(flags & LOOKUP_PARENT);
1847 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1848 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1855 * Restricted form of lookup. Doesn't follow links, single-component only,
1856 * needs parent already locked. Doesn't follow mounts.
1859 static struct dentry *lookup_hash(struct nameidata *nd)
1861 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1865 * lookup_one_len - filesystem helper to lookup single pathname component
1866 * @name: pathname component to lookup
1867 * @base: base directory to lookup from
1868 * @len: maximum length @len should be interpreted to
1870 * Note that this routine is purely a helper for filesystem usage and should
1871 * not be called by generic code. Also note that by using this function the
1872 * nameidata argument is passed to the filesystem methods and a filesystem
1873 * using this helper needs to be prepared for that.
1875 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1881 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1885 this.hash = full_name_hash(name, len);
1887 return ERR_PTR(-EACCES);
1890 c = *(const unsigned char *)name++;
1891 if (c == '/' || c == '\0')
1892 return ERR_PTR(-EACCES);
1895 * See if the low-level filesystem might want
1896 * to use its own hash..
1898 if (base->d_flags & DCACHE_OP_HASH) {
1899 int err = base->d_op->d_hash(base, base->d_inode, &this);
1901 return ERR_PTR(err);
1904 err = inode_permission(base->d_inode, MAY_EXEC);
1906 return ERR_PTR(err);
1908 return __lookup_hash(&this, base, NULL);
1911 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1912 struct path *path, int *empty)
1914 struct nameidata nd;
1915 char *tmp = getname_flags(name, flags, empty);
1916 int err = PTR_ERR(tmp);
1919 BUG_ON(flags & LOOKUP_PARENT);
1921 err = do_path_lookup(dfd, tmp, flags, &nd);
1929 int user_path_at(int dfd, const char __user *name, unsigned flags,
1932 return user_path_at_empty(dfd, name, flags, path, NULL);
1935 static int user_path_parent(int dfd, const char __user *path,
1936 struct nameidata *nd, char **name)
1938 char *s = getname(path);
1944 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1954 * It's inline, so penalty for filesystems that don't use sticky bit is
1957 static inline int check_sticky(struct inode *dir, struct inode *inode)
1959 kuid_t fsuid = current_fsuid();
1961 if (!(dir->i_mode & S_ISVTX))
1963 if (uid_eq(inode->i_uid, fsuid))
1965 if (uid_eq(dir->i_uid, fsuid))
1967 return !inode_capable(inode, CAP_FOWNER);
1971 * Check whether we can remove a link victim from directory dir, check
1972 * whether the type of victim is right.
1973 * 1. We can't do it if dir is read-only (done in permission())
1974 * 2. We should have write and exec permissions on dir
1975 * 3. We can't remove anything from append-only dir
1976 * 4. We can't do anything with immutable dir (done in permission())
1977 * 5. If the sticky bit on dir is set we should either
1978 * a. be owner of dir, or
1979 * b. be owner of victim, or
1980 * c. have CAP_FOWNER capability
1981 * 6. If the victim is append-only or immutable we can't do antyhing with
1982 * links pointing to it.
1983 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1984 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1985 * 9. We can't remove a root or mountpoint.
1986 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1987 * nfs_async_unlink().
1989 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1993 if (!victim->d_inode)
1996 BUG_ON(victim->d_parent->d_inode != dir);
1997 audit_inode_child(victim, dir);
1999 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2004 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2005 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2008 if (!S_ISDIR(victim->d_inode->i_mode))
2010 if (IS_ROOT(victim))
2012 } else if (S_ISDIR(victim->d_inode->i_mode))
2014 if (IS_DEADDIR(dir))
2016 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2021 /* Check whether we can create an object with dentry child in directory
2023 * 1. We can't do it if child already exists (open has special treatment for
2024 * this case, but since we are inlined it's OK)
2025 * 2. We can't do it if dir is read-only (done in permission())
2026 * 3. We should have write and exec permissions on dir
2027 * 4. We can't do it if dir is immutable (done in permission())
2029 static inline int may_create(struct inode *dir, struct dentry *child)
2033 if (IS_DEADDIR(dir))
2035 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2039 * p1 and p2 should be directories on the same fs.
2041 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2046 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2050 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2052 p = d_ancestor(p2, p1);
2054 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2055 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2059 p = d_ancestor(p1, p2);
2061 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2062 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2066 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2067 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2071 void unlock_rename(struct dentry *p1, struct dentry *p2)
2073 mutex_unlock(&p1->d_inode->i_mutex);
2075 mutex_unlock(&p2->d_inode->i_mutex);
2076 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2080 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2081 struct nameidata *nd)
2083 int error = may_create(dir, dentry);
2088 if (!dir->i_op->create)
2089 return -EACCES; /* shouldn't it be ENOSYS? */
2092 error = security_inode_create(dir, dentry, mode);
2095 error = dir->i_op->create(dir, dentry, mode, nd);
2097 fsnotify_create(dir, dentry);
2101 static int may_open(struct path *path, int acc_mode, int flag)
2103 struct dentry *dentry = path->dentry;
2104 struct inode *inode = dentry->d_inode;
2114 switch (inode->i_mode & S_IFMT) {
2118 if (acc_mode & MAY_WRITE)
2123 if (path->mnt->mnt_flags & MNT_NODEV)
2132 error = inode_permission(inode, acc_mode);
2137 * An append-only file must be opened in append mode for writing.
2139 if (IS_APPEND(inode)) {
2140 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2146 /* O_NOATIME can only be set by the owner or superuser */
2147 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2153 static int handle_truncate(struct file *filp)
2155 struct path *path = &filp->f_path;
2156 struct inode *inode = path->dentry->d_inode;
2157 int error = get_write_access(inode);
2161 * Refuse to truncate files with mandatory locks held on them.
2163 error = locks_verify_locked(inode);
2165 error = security_path_truncate(path);
2167 error = do_truncate(path->dentry, 0,
2168 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2171 put_write_access(inode);
2175 static inline int open_to_namei_flags(int flag)
2177 if ((flag & O_ACCMODE) == 3)
2183 * Handle the last step of open()
2185 static struct file *do_last(struct nameidata *nd, struct path *path,
2186 const struct open_flags *op, const char *pathname)
2188 struct dentry *dir = nd->path.dentry;
2189 struct dentry *dentry;
2190 int open_flag = op->open_flag;
2191 int will_truncate = open_flag & O_TRUNC;
2193 int acc_mode = op->acc_mode;
2197 nd->flags &= ~LOOKUP_PARENT;
2198 nd->flags |= op->intent;
2200 switch (nd->last_type) {
2203 error = handle_dots(nd, nd->last_type);
2205 return ERR_PTR(error);
2208 error = complete_walk(nd);
2210 return ERR_PTR(error);
2211 audit_inode(pathname, nd->path.dentry);
2212 if (open_flag & O_CREAT) {
2218 error = complete_walk(nd);
2220 return ERR_PTR(error);
2221 audit_inode(pathname, dir);
2225 if (!(open_flag & O_CREAT)) {
2227 if (nd->last.name[nd->last.len])
2228 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2229 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2231 /* we _can_ be in RCU mode here */
2232 error = walk_component(nd, path, &nd->last, LAST_NORM,
2235 return ERR_PTR(error);
2236 if (error) /* symlink */
2239 error = complete_walk(nd);
2241 return ERR_PTR(error);
2244 if (nd->flags & LOOKUP_DIRECTORY) {
2245 if (!nd->inode->i_op->lookup)
2248 audit_inode(pathname, nd->path.dentry);
2252 /* create side of things */
2254 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2255 * cleared when we got to the last component we are about to look up
2257 error = complete_walk(nd);
2259 return ERR_PTR(error);
2261 audit_inode(pathname, dir);
2263 /* trailing slashes? */
2264 if (nd->last.name[nd->last.len])
2267 mutex_lock(&dir->d_inode->i_mutex);
2269 dentry = lookup_hash(nd);
2270 error = PTR_ERR(dentry);
2271 if (IS_ERR(dentry)) {
2272 mutex_unlock(&dir->d_inode->i_mutex);
2276 path->dentry = dentry;
2277 path->mnt = nd->path.mnt;
2279 /* Negative dentry, just create the file */
2280 if (!dentry->d_inode) {
2281 umode_t mode = op->mode;
2282 if (!IS_POSIXACL(dir->d_inode))
2283 mode &= ~current_umask();
2285 * This write is needed to ensure that a
2286 * rw->ro transition does not occur between
2287 * the time when the file is created and when
2288 * a permanent write count is taken through
2289 * the 'struct file' in nameidata_to_filp().
2291 error = mnt_want_write(nd->path.mnt);
2293 goto exit_mutex_unlock;
2295 /* Don't check for write permission, don't truncate */
2296 open_flag &= ~O_TRUNC;
2298 acc_mode = MAY_OPEN;
2299 error = security_path_mknod(&nd->path, dentry, mode, 0);
2301 goto exit_mutex_unlock;
2302 error = vfs_create(dir->d_inode, dentry, mode, nd);
2304 goto exit_mutex_unlock;
2305 mutex_unlock(&dir->d_inode->i_mutex);
2306 dput(nd->path.dentry);
2307 nd->path.dentry = dentry;
2312 * It already exists.
2314 mutex_unlock(&dir->d_inode->i_mutex);
2315 audit_inode(pathname, path->dentry);
2318 if (open_flag & O_EXCL)
2321 error = follow_managed(path, nd->flags);
2326 nd->flags |= LOOKUP_JUMPED;
2329 if (!path->dentry->d_inode)
2332 if (path->dentry->d_inode->i_op->follow_link)
2335 path_to_nameidata(path, nd);
2336 nd->inode = path->dentry->d_inode;
2337 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2338 error = complete_walk(nd);
2340 return ERR_PTR(error);
2342 if (S_ISDIR(nd->inode->i_mode))
2345 if (!S_ISREG(nd->inode->i_mode))
2348 if (will_truncate) {
2349 error = mnt_want_write(nd->path.mnt);
2355 error = may_open(&nd->path, acc_mode, open_flag);
2358 filp = nameidata_to_filp(nd);
2359 if (!IS_ERR(filp)) {
2360 error = ima_file_check(filp, op->acc_mode);
2363 filp = ERR_PTR(error);
2366 if (!IS_ERR(filp)) {
2367 if (will_truncate) {
2368 error = handle_truncate(filp);
2371 filp = ERR_PTR(error);
2377 mnt_drop_write(nd->path.mnt);
2378 path_put(&nd->path);
2382 mutex_unlock(&dir->d_inode->i_mutex);
2384 path_put_conditional(path, nd);
2386 filp = ERR_PTR(error);
2390 static struct file *path_openat(int dfd, const char *pathname,
2391 struct nameidata *nd, const struct open_flags *op, int flags)
2393 struct file *base = NULL;
2398 filp = get_empty_filp();
2400 return ERR_PTR(-ENFILE);
2402 filp->f_flags = op->open_flag;
2403 nd->intent.open.file = filp;
2404 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2405 nd->intent.open.create_mode = op->mode;
2407 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2408 if (unlikely(error))
2411 current->total_link_count = 0;
2412 error = link_path_walk(pathname, nd);
2413 if (unlikely(error))
2416 filp = do_last(nd, &path, op, pathname);
2417 while (unlikely(!filp)) { /* trailing symlink */
2418 struct path link = path;
2420 if (!(nd->flags & LOOKUP_FOLLOW)) {
2421 path_put_conditional(&path, nd);
2422 path_put(&nd->path);
2423 filp = ERR_PTR(-ELOOP);
2426 nd->flags |= LOOKUP_PARENT;
2427 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2428 error = follow_link(&link, nd, &cookie);
2429 if (unlikely(error))
2430 filp = ERR_PTR(error);
2432 filp = do_last(nd, &path, op, pathname);
2433 put_link(nd, &link, cookie);
2436 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2437 path_put(&nd->root);
2440 release_open_intent(nd);
2444 filp = ERR_PTR(error);
2448 struct file *do_filp_open(int dfd, const char *pathname,
2449 const struct open_flags *op, int flags)
2451 struct nameidata nd;
2454 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2455 if (unlikely(filp == ERR_PTR(-ECHILD)))
2456 filp = path_openat(dfd, pathname, &nd, op, flags);
2457 if (unlikely(filp == ERR_PTR(-ESTALE)))
2458 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2462 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2463 const char *name, const struct open_flags *op, int flags)
2465 struct nameidata nd;
2469 nd.root.dentry = dentry;
2471 flags |= LOOKUP_ROOT;
2473 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2474 return ERR_PTR(-ELOOP);
2476 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2477 if (unlikely(file == ERR_PTR(-ECHILD)))
2478 file = path_openat(-1, name, &nd, op, flags);
2479 if (unlikely(file == ERR_PTR(-ESTALE)))
2480 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2484 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2486 struct dentry *dentry = ERR_PTR(-EEXIST);
2487 struct nameidata nd;
2488 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2490 return ERR_PTR(error);
2493 * Yucky last component or no last component at all?
2494 * (foo/., foo/.., /////)
2496 if (nd.last_type != LAST_NORM)
2498 nd.flags &= ~LOOKUP_PARENT;
2499 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2500 nd.intent.open.flags = O_EXCL;
2503 * Do the final lookup.
2505 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2506 dentry = lookup_hash(&nd);
2510 if (dentry->d_inode)
2513 * Special case - lookup gave negative, but... we had foo/bar/
2514 * From the vfs_mknod() POV we just have a negative dentry -
2515 * all is fine. Let's be bastards - you had / on the end, you've
2516 * been asking for (non-existent) directory. -ENOENT for you.
2518 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2520 dentry = ERR_PTR(-ENOENT);
2527 dentry = ERR_PTR(-EEXIST);
2529 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2534 EXPORT_SYMBOL(kern_path_create);
2536 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2538 char *tmp = getname(pathname);
2541 return ERR_CAST(tmp);
2542 res = kern_path_create(dfd, tmp, path, is_dir);
2546 EXPORT_SYMBOL(user_path_create);
2548 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2550 int error = may_create(dir, dentry);
2555 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2558 if (!dir->i_op->mknod)
2561 error = devcgroup_inode_mknod(mode, dev);
2565 error = security_inode_mknod(dir, dentry, mode, dev);
2569 error = dir->i_op->mknod(dir, dentry, mode, dev);
2571 fsnotify_create(dir, dentry);
2575 static int may_mknod(umode_t mode)
2577 switch (mode & S_IFMT) {
2583 case 0: /* zero mode translates to S_IFREG */
2592 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2595 struct dentry *dentry;
2602 dentry = user_path_create(dfd, filename, &path, 0);
2604 return PTR_ERR(dentry);
2606 if (!IS_POSIXACL(path.dentry->d_inode))
2607 mode &= ~current_umask();
2608 error = may_mknod(mode);
2611 error = mnt_want_write(path.mnt);
2614 error = security_path_mknod(&path, dentry, mode, dev);
2616 goto out_drop_write;
2617 switch (mode & S_IFMT) {
2618 case 0: case S_IFREG:
2619 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2621 case S_IFCHR: case S_IFBLK:
2622 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2623 new_decode_dev(dev));
2625 case S_IFIFO: case S_IFSOCK:
2626 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2630 mnt_drop_write(path.mnt);
2633 mutex_unlock(&path.dentry->d_inode->i_mutex);
2639 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
2641 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2644 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2646 int error = may_create(dir, dentry);
2647 unsigned max_links = dir->i_sb->s_max_links;
2652 if (!dir->i_op->mkdir)
2655 mode &= (S_IRWXUGO|S_ISVTX);
2656 error = security_inode_mkdir(dir, dentry, mode);
2660 if (max_links && dir->i_nlink >= max_links)
2663 error = dir->i_op->mkdir(dir, dentry, mode);
2665 fsnotify_mkdir(dir, dentry);
2669 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
2671 struct dentry *dentry;
2675 dentry = user_path_create(dfd, pathname, &path, 1);
2677 return PTR_ERR(dentry);
2679 if (!IS_POSIXACL(path.dentry->d_inode))
2680 mode &= ~current_umask();
2681 error = mnt_want_write(path.mnt);
2684 error = security_path_mkdir(&path, dentry, mode);
2686 goto out_drop_write;
2687 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2689 mnt_drop_write(path.mnt);
2692 mutex_unlock(&path.dentry->d_inode->i_mutex);
2697 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
2699 return sys_mkdirat(AT_FDCWD, pathname, mode);
2703 * The dentry_unhash() helper will try to drop the dentry early: we
2704 * should have a usage count of 1 if we're the only user of this
2705 * dentry, and if that is true (possibly after pruning the dcache),
2706 * then we drop the dentry now.
2708 * A low-level filesystem can, if it choses, legally
2711 * if (!d_unhashed(dentry))
2714 * if it cannot handle the case of removing a directory
2715 * that is still in use by something else..
2717 void dentry_unhash(struct dentry *dentry)
2719 shrink_dcache_parent(dentry);
2720 spin_lock(&dentry->d_lock);
2721 if (dentry->d_count == 1)
2723 spin_unlock(&dentry->d_lock);
2726 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2728 int error = may_delete(dir, dentry, 1);
2733 if (!dir->i_op->rmdir)
2737 mutex_lock(&dentry->d_inode->i_mutex);
2740 if (d_mountpoint(dentry))
2743 error = security_inode_rmdir(dir, dentry);
2747 shrink_dcache_parent(dentry);
2748 error = dir->i_op->rmdir(dir, dentry);
2752 dentry->d_inode->i_flags |= S_DEAD;
2756 mutex_unlock(&dentry->d_inode->i_mutex);
2763 static long do_rmdir(int dfd, const char __user *pathname)
2767 struct dentry *dentry;
2768 struct nameidata nd;
2770 error = user_path_parent(dfd, pathname, &nd, &name);
2774 switch(nd.last_type) {
2786 nd.flags &= ~LOOKUP_PARENT;
2788 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2789 dentry = lookup_hash(&nd);
2790 error = PTR_ERR(dentry);
2793 if (!dentry->d_inode) {
2797 error = mnt_want_write(nd.path.mnt);
2800 error = security_path_rmdir(&nd.path, dentry);
2803 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2805 mnt_drop_write(nd.path.mnt);
2809 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2816 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2818 return do_rmdir(AT_FDCWD, pathname);
2821 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2823 int error = may_delete(dir, dentry, 0);
2828 if (!dir->i_op->unlink)
2831 mutex_lock(&dentry->d_inode->i_mutex);
2832 if (d_mountpoint(dentry))
2835 error = security_inode_unlink(dir, dentry);
2837 error = dir->i_op->unlink(dir, dentry);
2842 mutex_unlock(&dentry->d_inode->i_mutex);
2844 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2845 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2846 fsnotify_link_count(dentry->d_inode);
2854 * Make sure that the actual truncation of the file will occur outside its
2855 * directory's i_mutex. Truncate can take a long time if there is a lot of
2856 * writeout happening, and we don't want to prevent access to the directory
2857 * while waiting on the I/O.
2859 static long do_unlinkat(int dfd, const char __user *pathname)
2863 struct dentry *dentry;
2864 struct nameidata nd;
2865 struct inode *inode = NULL;
2867 error = user_path_parent(dfd, pathname, &nd, &name);
2872 if (nd.last_type != LAST_NORM)
2875 nd.flags &= ~LOOKUP_PARENT;
2877 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2878 dentry = lookup_hash(&nd);
2879 error = PTR_ERR(dentry);
2880 if (!IS_ERR(dentry)) {
2881 /* Why not before? Because we want correct error value */
2882 if (nd.last.name[nd.last.len])
2884 inode = dentry->d_inode;
2888 error = mnt_want_write(nd.path.mnt);
2891 error = security_path_unlink(&nd.path, dentry);
2894 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2896 mnt_drop_write(nd.path.mnt);
2900 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2902 iput(inode); /* truncate the inode here */
2909 error = !dentry->d_inode ? -ENOENT :
2910 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2914 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2916 if ((flag & ~AT_REMOVEDIR) != 0)
2919 if (flag & AT_REMOVEDIR)
2920 return do_rmdir(dfd, pathname);
2922 return do_unlinkat(dfd, pathname);
2925 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2927 return do_unlinkat(AT_FDCWD, pathname);
2930 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2932 int error = may_create(dir, dentry);
2937 if (!dir->i_op->symlink)
2940 error = security_inode_symlink(dir, dentry, oldname);
2944 error = dir->i_op->symlink(dir, dentry, oldname);
2946 fsnotify_create(dir, dentry);
2950 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2951 int, newdfd, const char __user *, newname)
2955 struct dentry *dentry;
2958 from = getname(oldname);
2960 return PTR_ERR(from);
2962 dentry = user_path_create(newdfd, newname, &path, 0);
2963 error = PTR_ERR(dentry);
2967 error = mnt_want_write(path.mnt);
2970 error = security_path_symlink(&path, dentry, from);
2972 goto out_drop_write;
2973 error = vfs_symlink(path.dentry->d_inode, dentry, from);
2975 mnt_drop_write(path.mnt);
2978 mutex_unlock(&path.dentry->d_inode->i_mutex);
2985 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2987 return sys_symlinkat(oldname, AT_FDCWD, newname);
2990 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2992 struct inode *inode = old_dentry->d_inode;
2993 unsigned max_links = dir->i_sb->s_max_links;
2999 error = may_create(dir, new_dentry);
3003 if (dir->i_sb != inode->i_sb)
3007 * A link to an append-only or immutable file cannot be created.
3009 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3011 if (!dir->i_op->link)
3013 if (S_ISDIR(inode->i_mode))
3016 error = security_inode_link(old_dentry, dir, new_dentry);
3020 mutex_lock(&inode->i_mutex);
3021 /* Make sure we don't allow creating hardlink to an unlinked file */
3022 if (inode->i_nlink == 0)
3024 else if (max_links && inode->i_nlink >= max_links)
3027 error = dir->i_op->link(old_dentry, dir, new_dentry);
3028 mutex_unlock(&inode->i_mutex);
3030 fsnotify_link(dir, inode, new_dentry);
3035 * Hardlinks are often used in delicate situations. We avoid
3036 * security-related surprises by not following symlinks on the
3039 * We don't follow them on the oldname either to be compatible
3040 * with linux 2.0, and to avoid hard-linking to directories
3041 * and other special files. --ADM
3043 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3044 int, newdfd, const char __user *, newname, int, flags)
3046 struct dentry *new_dentry;
3047 struct path old_path, new_path;
3051 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3054 * To use null names we require CAP_DAC_READ_SEARCH
3055 * This ensures that not everyone will be able to create
3056 * handlink using the passed filedescriptor.
3058 if (flags & AT_EMPTY_PATH) {
3059 if (!capable(CAP_DAC_READ_SEARCH))
3064 if (flags & AT_SYMLINK_FOLLOW)
3065 how |= LOOKUP_FOLLOW;
3067 error = user_path_at(olddfd, oldname, how, &old_path);
3071 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3072 error = PTR_ERR(new_dentry);
3073 if (IS_ERR(new_dentry))
3077 if (old_path.mnt != new_path.mnt)
3079 error = mnt_want_write(new_path.mnt);
3082 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3084 goto out_drop_write;
3085 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3087 mnt_drop_write(new_path.mnt);
3090 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3091 path_put(&new_path);
3093 path_put(&old_path);
3098 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3100 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3104 * The worst of all namespace operations - renaming directory. "Perverted"
3105 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3107 * a) we can get into loop creation. Check is done in is_subdir().
3108 * b) race potential - two innocent renames can create a loop together.
3109 * That's where 4.4 screws up. Current fix: serialization on
3110 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3112 * c) we have to lock _three_ objects - parents and victim (if it exists).
3113 * And that - after we got ->i_mutex on parents (until then we don't know
3114 * whether the target exists). Solution: try to be smart with locking
3115 * order for inodes. We rely on the fact that tree topology may change
3116 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3117 * move will be locked. Thus we can rank directories by the tree
3118 * (ancestors first) and rank all non-directories after them.
3119 * That works since everybody except rename does "lock parent, lookup,
3120 * lock child" and rename is under ->s_vfs_rename_mutex.
3121 * HOWEVER, it relies on the assumption that any object with ->lookup()
3122 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3123 * we'd better make sure that there's no link(2) for them.
3124 * d) conversion from fhandle to dentry may come in the wrong moment - when
3125 * we are removing the target. Solution: we will have to grab ->i_mutex
3126 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3127 * ->i_mutex on parents, which works but leads to some truly excessive
3130 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3131 struct inode *new_dir, struct dentry *new_dentry)
3134 struct inode *target = new_dentry->d_inode;
3135 unsigned max_links = new_dir->i_sb->s_max_links;
3138 * If we are going to change the parent - check write permissions,
3139 * we'll need to flip '..'.
3141 if (new_dir != old_dir) {
3142 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3147 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3153 mutex_lock(&target->i_mutex);
3156 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3160 if (max_links && !target && new_dir != old_dir &&
3161 new_dir->i_nlink >= max_links)
3165 shrink_dcache_parent(new_dentry);
3166 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3171 target->i_flags |= S_DEAD;
3172 dont_mount(new_dentry);
3176 mutex_unlock(&target->i_mutex);
3179 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3180 d_move(old_dentry,new_dentry);
3184 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3185 struct inode *new_dir, struct dentry *new_dentry)
3187 struct inode *target = new_dentry->d_inode;
3190 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3196 mutex_lock(&target->i_mutex);
3199 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3202 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3207 dont_mount(new_dentry);
3208 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3209 d_move(old_dentry, new_dentry);
3212 mutex_unlock(&target->i_mutex);
3217 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3218 struct inode *new_dir, struct dentry *new_dentry)
3221 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3222 const unsigned char *old_name;
3224 if (old_dentry->d_inode == new_dentry->d_inode)
3227 error = may_delete(old_dir, old_dentry, is_dir);
3231 if (!new_dentry->d_inode)
3232 error = may_create(new_dir, new_dentry);
3234 error = may_delete(new_dir, new_dentry, is_dir);
3238 if (!old_dir->i_op->rename)
3241 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3244 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3246 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3248 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3249 new_dentry->d_inode, old_dentry);
3250 fsnotify_oldname_free(old_name);
3255 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3256 int, newdfd, const char __user *, newname)
3258 struct dentry *old_dir, *new_dir;
3259 struct dentry *old_dentry, *new_dentry;
3260 struct dentry *trap;
3261 struct nameidata oldnd, newnd;
3266 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3270 error = user_path_parent(newdfd, newname, &newnd, &to);
3275 if (oldnd.path.mnt != newnd.path.mnt)
3278 old_dir = oldnd.path.dentry;
3280 if (oldnd.last_type != LAST_NORM)
3283 new_dir = newnd.path.dentry;
3284 if (newnd.last_type != LAST_NORM)
3287 oldnd.flags &= ~LOOKUP_PARENT;
3288 newnd.flags &= ~LOOKUP_PARENT;
3289 newnd.flags |= LOOKUP_RENAME_TARGET;
3291 trap = lock_rename(new_dir, old_dir);
3293 old_dentry = lookup_hash(&oldnd);
3294 error = PTR_ERR(old_dentry);
3295 if (IS_ERR(old_dentry))
3297 /* source must exist */
3299 if (!old_dentry->d_inode)
3301 /* unless the source is a directory trailing slashes give -ENOTDIR */
3302 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3304 if (oldnd.last.name[oldnd.last.len])
3306 if (newnd.last.name[newnd.last.len])
3309 /* source should not be ancestor of target */
3311 if (old_dentry == trap)
3313 new_dentry = lookup_hash(&newnd);
3314 error = PTR_ERR(new_dentry);
3315 if (IS_ERR(new_dentry))
3317 /* target should not be an ancestor of source */
3319 if (new_dentry == trap)
3322 error = mnt_want_write(oldnd.path.mnt);
3325 error = security_path_rename(&oldnd.path, old_dentry,
3326 &newnd.path, new_dentry);
3329 error = vfs_rename(old_dir->d_inode, old_dentry,
3330 new_dir->d_inode, new_dentry);
3332 mnt_drop_write(oldnd.path.mnt);
3338 unlock_rename(new_dir, old_dir);
3340 path_put(&newnd.path);
3343 path_put(&oldnd.path);
3349 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3351 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3354 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3358 len = PTR_ERR(link);
3363 if (len > (unsigned) buflen)
3365 if (copy_to_user(buffer, link, len))
3372 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3373 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3374 * using) it for any given inode is up to filesystem.
3376 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3378 struct nameidata nd;
3383 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3385 return PTR_ERR(cookie);
3387 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3388 if (dentry->d_inode->i_op->put_link)
3389 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3393 int vfs_follow_link(struct nameidata *nd, const char *link)
3395 return __vfs_follow_link(nd, link);
3398 /* get the link contents into pagecache */
3399 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3403 struct address_space *mapping = dentry->d_inode->i_mapping;
3404 page = read_mapping_page(mapping, 0, NULL);
3409 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3413 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3415 struct page *page = NULL;
3416 char *s = page_getlink(dentry, &page);
3417 int res = vfs_readlink(dentry,buffer,buflen,s);
3420 page_cache_release(page);
3425 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3427 struct page *page = NULL;
3428 nd_set_link(nd, page_getlink(dentry, &page));
3432 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3434 struct page *page = cookie;
3438 page_cache_release(page);
3443 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3445 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3447 struct address_space *mapping = inode->i_mapping;
3452 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3454 flags |= AOP_FLAG_NOFS;
3457 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3458 flags, &page, &fsdata);
3462 kaddr = kmap_atomic(page);
3463 memcpy(kaddr, symname, len-1);
3464 kunmap_atomic(kaddr);
3466 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3473 mark_inode_dirty(inode);
3479 int page_symlink(struct inode *inode, const char *symname, int len)
3481 return __page_symlink(inode, symname, len,
3482 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3485 const struct inode_operations page_symlink_inode_operations = {
3486 .readlink = generic_readlink,
3487 .follow_link = page_follow_link_light,
3488 .put_link = page_put_link,
3491 EXPORT_SYMBOL(user_path_at);
3492 EXPORT_SYMBOL(follow_down_one);
3493 EXPORT_SYMBOL(follow_down);
3494 EXPORT_SYMBOL(follow_up);
3495 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3496 EXPORT_SYMBOL(getname);
3497 EXPORT_SYMBOL(lock_rename);
3498 EXPORT_SYMBOL(lookup_one_len);
3499 EXPORT_SYMBOL(page_follow_link_light);
3500 EXPORT_SYMBOL(page_put_link);
3501 EXPORT_SYMBOL(page_readlink);
3502 EXPORT_SYMBOL(__page_symlink);
3503 EXPORT_SYMBOL(page_symlink);
3504 EXPORT_SYMBOL(page_symlink_inode_operations);
3505 EXPORT_SYMBOL(kern_path);
3506 EXPORT_SYMBOL(vfs_path_lookup);
3507 EXPORT_SYMBOL(inode_permission);
3508 EXPORT_SYMBOL(unlock_rename);
3509 EXPORT_SYMBOL(vfs_create);
3510 EXPORT_SYMBOL(vfs_follow_link);
3511 EXPORT_SYMBOL(vfs_link);
3512 EXPORT_SYMBOL(vfs_mkdir);
3513 EXPORT_SYMBOL(vfs_mknod);
3514 EXPORT_SYMBOL(generic_permission);
3515 EXPORT_SYMBOL(vfs_readlink);
3516 EXPORT_SYMBOL(vfs_rename);
3517 EXPORT_SYMBOL(vfs_rmdir);
3518 EXPORT_SYMBOL(vfs_symlink);
3519 EXPORT_SYMBOL(vfs_unlink);
3520 EXPORT_SYMBOL(dentry_unhash);
3521 EXPORT_SYMBOL(generic_readlink);