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/module.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>
40 /* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
68 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existent name.
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
85 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
88 * [10-Sep-98 Alan Modra] Another symlink change.
91 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
108 * any extra contention...
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user *filename, char *page)
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
130 retval = strncpy_from_user(page, filename, len);
134 return -ENAMETOOLONG;
140 static char *getname_flags(const char __user *filename, int flags, int *empty)
144 result = ERR_PTR(-ENOMEM);
147 int retval = do_getname(filename, tmp);
151 if (retval == -ENOENT && empty)
153 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
155 result = ERR_PTR(retval);
159 audit_getname(result);
163 char *getname(const char __user * filename)
165 return getname_flags(filename, 0, 0);
168 #ifdef CONFIG_AUDITSYSCALL
169 void putname(const char *name)
171 if (unlikely(!audit_dummy_context()))
176 EXPORT_SYMBOL(putname);
179 static int check_acl(struct inode *inode, int mask)
181 #ifdef CONFIG_FS_POSIX_ACL
182 struct posix_acl *acl;
184 if (mask & MAY_NOT_BLOCK) {
185 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
188 /* no ->get_acl() calls in RCU mode... */
189 if (acl == ACL_NOT_CACHED)
191 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
194 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
197 * A filesystem can force a ACL callback by just never filling the
198 * ACL cache. But normally you'd fill the cache either at inode
199 * instantiation time, or on the first ->get_acl call.
201 * If the filesystem doesn't have a get_acl() function at all, we'll
202 * just create the negative cache entry.
204 if (acl == ACL_NOT_CACHED) {
205 if (inode->i_op->get_acl) {
206 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
210 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
216 int error = posix_acl_permission(inode, acl, mask);
217 posix_acl_release(acl);
226 * This does the basic permission checking
228 static int acl_permission_check(struct inode *inode, int mask)
230 unsigned int mode = inode->i_mode;
232 if (current_user_ns() != inode_userns(inode))
235 if (likely(current_fsuid() == inode->i_uid))
238 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
239 int error = check_acl(inode, mask);
240 if (error != -EAGAIN)
244 if (in_group_p(inode->i_gid))
250 * If the DACs are ok we don't need any capability check.
252 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
258 * generic_permission - check for access rights on a Posix-like filesystem
259 * @inode: inode to check access rights for
260 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
262 * Used to check for read/write/execute permissions on a file.
263 * We use "fsuid" for this, letting us set arbitrary permissions
264 * for filesystem access without changing the "normal" uids which
265 * are used for other things.
267 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
268 * request cannot be satisfied (eg. requires blocking or too much complexity).
269 * It would then be called again in ref-walk mode.
271 int generic_permission(struct inode *inode, int mask)
276 * Do the basic permission checks.
278 ret = acl_permission_check(inode, mask);
282 if (S_ISDIR(inode->i_mode)) {
283 /* DACs are overridable for directories */
284 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
286 if (!(mask & MAY_WRITE))
287 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
292 * Read/write DACs are always overridable.
293 * Executable DACs are overridable when there is
294 * at least one exec bit set.
296 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
297 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
301 * Searching includes executable on directories, else just read.
303 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
304 if (mask == MAY_READ)
305 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
312 * We _really_ want to just do "generic_permission()" without
313 * even looking at the inode->i_op values. So we keep a cache
314 * flag in inode->i_opflags, that says "this has not special
315 * permission function, use the fast case".
317 static inline int do_inode_permission(struct inode *inode, int mask)
319 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
320 if (likely(inode->i_op->permission))
321 return inode->i_op->permission(inode, mask);
323 /* This gets set once for the inode lifetime */
324 spin_lock(&inode->i_lock);
325 inode->i_opflags |= IOP_FASTPERM;
326 spin_unlock(&inode->i_lock);
328 return generic_permission(inode, mask);
332 * inode_permission - check for access rights to a given inode
333 * @inode: inode to check permission on
334 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
336 * Used to check for read/write/execute permissions on an inode.
337 * We use "fsuid" for this, letting us set arbitrary permissions
338 * for filesystem access without changing the "normal" uids which
339 * are used for other things.
341 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
343 int inode_permission(struct inode *inode, int mask)
347 if (unlikely(mask & MAY_WRITE)) {
348 umode_t mode = inode->i_mode;
351 * Nobody gets write access to a read-only fs.
353 if (IS_RDONLY(inode) &&
354 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
358 * Nobody gets write access to an immutable file.
360 if (IS_IMMUTABLE(inode))
364 retval = do_inode_permission(inode, mask);
368 retval = devcgroup_inode_permission(inode, mask);
372 return security_inode_permission(inode, mask);
376 * path_get - get a reference to a path
377 * @path: path to get the reference to
379 * Given a path increment the reference count to the dentry and the vfsmount.
381 void path_get(struct path *path)
386 EXPORT_SYMBOL(path_get);
389 * path_put - put a reference to a path
390 * @path: path to put the reference to
392 * Given a path decrement the reference count to the dentry and the vfsmount.
394 void path_put(struct path *path)
399 EXPORT_SYMBOL(path_put);
402 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
403 * @path: nameidate to verify
405 * Rename can sometimes move a file or directory outside of a bind
406 * mount, path_connected allows those cases to be detected.
408 static bool path_connected(const struct path *path)
410 struct vfsmount *mnt = path->mnt;
412 /* Only bind mounts can have disconnected paths */
413 if (mnt->mnt_root == mnt->mnt_sb->s_root)
416 return is_subdir(path->dentry, mnt->mnt_root);
420 * Path walking has 2 modes, rcu-walk and ref-walk (see
421 * Documentation/filesystems/path-lookup.txt). In situations when we can't
422 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
423 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
424 * mode. Refcounts are grabbed at the last known good point before rcu-walk
425 * got stuck, so ref-walk may continue from there. If this is not successful
426 * (eg. a seqcount has changed), then failure is returned and it's up to caller
427 * to restart the path walk from the beginning in ref-walk mode.
431 * unlazy_walk - try to switch to ref-walk mode.
432 * @nd: nameidata pathwalk data
433 * @dentry: child of nd->path.dentry or NULL
434 * Returns: 0 on success, -ECHILD on failure
436 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
437 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
438 * @nd or NULL. Must be called from rcu-walk context.
440 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
442 struct fs_struct *fs = current->fs;
443 struct dentry *parent = nd->path.dentry;
446 BUG_ON(!(nd->flags & LOOKUP_RCU));
447 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
449 spin_lock(&fs->lock);
450 if (nd->root.mnt != fs->root.mnt ||
451 nd->root.dentry != fs->root.dentry)
454 spin_lock(&parent->d_lock);
456 if (!__d_rcu_to_refcount(parent, nd->seq))
458 BUG_ON(nd->inode != parent->d_inode);
460 if (dentry->d_parent != parent)
462 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
463 if (!__d_rcu_to_refcount(dentry, nd->seq))
466 * If the sequence check on the child dentry passed, then
467 * the child has not been removed from its parent. This
468 * means the parent dentry must be valid and able to take
469 * a reference at this point.
471 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
472 BUG_ON(!parent->d_count);
474 spin_unlock(&dentry->d_lock);
476 spin_unlock(&parent->d_lock);
479 spin_unlock(&fs->lock);
481 mntget(nd->path.mnt);
484 br_read_unlock(vfsmount_lock);
485 nd->flags &= ~LOOKUP_RCU;
489 spin_unlock(&dentry->d_lock);
491 spin_unlock(&parent->d_lock);
494 spin_unlock(&fs->lock);
499 * release_open_intent - free up open intent resources
500 * @nd: pointer to nameidata
502 void release_open_intent(struct nameidata *nd)
504 struct file *file = nd->intent.open.file;
506 if (file && !IS_ERR(file)) {
507 if (file->f_path.dentry == NULL)
514 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
516 return dentry->d_op->d_revalidate(dentry, nd);
520 * complete_walk - successful completion of path walk
521 * @nd: pointer nameidata
523 * If we had been in RCU mode, drop out of it and legitimize nd->path.
524 * Revalidate the final result, unless we'd already done that during
525 * the path walk or the filesystem doesn't ask for it. Return 0 on
526 * success, -error on failure. In case of failure caller does not
527 * need to drop nd->path.
529 static int complete_walk(struct nameidata *nd)
531 struct dentry *dentry = nd->path.dentry;
534 if (nd->flags & LOOKUP_RCU) {
535 nd->flags &= ~LOOKUP_RCU;
536 if (!(nd->flags & LOOKUP_ROOT))
538 spin_lock(&dentry->d_lock);
539 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
540 spin_unlock(&dentry->d_lock);
542 br_read_unlock(vfsmount_lock);
545 BUG_ON(nd->inode != dentry->d_inode);
546 spin_unlock(&dentry->d_lock);
547 mntget(nd->path.mnt);
549 br_read_unlock(vfsmount_lock);
552 if (likely(!(nd->flags & LOOKUP_JUMPED)))
555 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
558 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
561 /* Note: we do not d_invalidate() */
562 status = d_revalidate(dentry, nd);
573 static __always_inline void set_root(struct nameidata *nd)
575 get_fs_root(current->fs, &nd->root);
578 static int link_path_walk(const char *, struct nameidata *);
580 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
582 struct fs_struct *fs = current->fs;
586 seq = read_seqcount_begin(&fs->seq);
588 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
589 } while (read_seqcount_retry(&fs->seq, seq));
593 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
606 nd->flags |= LOOKUP_JUMPED;
608 nd->inode = nd->path.dentry->d_inode;
610 ret = link_path_walk(link, nd);
614 return PTR_ERR(link);
617 static void path_put_conditional(struct path *path, struct nameidata *nd)
620 if (path->mnt != nd->path.mnt)
624 static inline void path_to_nameidata(const struct path *path,
625 struct nameidata *nd)
627 if (!(nd->flags & LOOKUP_RCU)) {
628 dput(nd->path.dentry);
629 if (nd->path.mnt != path->mnt)
630 mntput(nd->path.mnt);
632 nd->path.mnt = path->mnt;
633 nd->path.dentry = path->dentry;
636 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
638 struct inode *inode = link->dentry->d_inode;
639 if (!IS_ERR(cookie) && inode->i_op->put_link)
640 inode->i_op->put_link(link->dentry, nd, cookie);
644 static __always_inline int
645 follow_link(struct path *link, struct nameidata *nd, void **p)
648 struct dentry *dentry = link->dentry;
650 BUG_ON(nd->flags & LOOKUP_RCU);
652 if (link->mnt == nd->path.mnt)
655 if (unlikely(current->total_link_count >= 40)) {
656 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
661 current->total_link_count++;
663 touch_atime(link->mnt, dentry);
664 nd_set_link(nd, NULL);
666 error = security_inode_follow_link(link->dentry, nd);
668 *p = ERR_PTR(error); /* no ->put_link(), please */
673 nd->last_type = LAST_BIND;
674 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
677 char *s = nd_get_link(nd);
680 error = __vfs_follow_link(nd, s);
681 else if (nd->last_type == LAST_BIND) {
682 nd->flags |= LOOKUP_JUMPED;
683 nd->inode = nd->path.dentry->d_inode;
684 if (nd->inode->i_op->follow_link) {
685 /* stepped on a _really_ weird one */
694 static int follow_up_rcu(struct path *path)
696 struct vfsmount *parent;
697 struct dentry *mountpoint;
699 parent = path->mnt->mnt_parent;
700 if (parent == path->mnt)
702 mountpoint = path->mnt->mnt_mountpoint;
703 path->dentry = mountpoint;
708 int follow_up(struct path *path)
710 struct vfsmount *parent;
711 struct dentry *mountpoint;
713 br_read_lock(vfsmount_lock);
714 parent = path->mnt->mnt_parent;
715 if (parent == path->mnt) {
716 br_read_unlock(vfsmount_lock);
720 mountpoint = dget(path->mnt->mnt_mountpoint);
721 br_read_unlock(vfsmount_lock);
723 path->dentry = mountpoint;
730 * Perform an automount
731 * - return -EISDIR to tell follow_managed() to stop and return the path we
734 static int follow_automount(struct path *path, unsigned flags,
737 struct vfsmount *mnt;
740 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
743 /* We don't want to mount if someone's just doing a stat -
744 * unless they're stat'ing a directory and appended a '/' to
747 * We do, however, want to mount if someone wants to open or
748 * create a file of any type under the mountpoint, wants to
749 * traverse through the mountpoint or wants to open the
750 * mounted directory. Also, autofs may mark negative dentries
751 * as being automount points. These will need the attentions
752 * of the daemon to instantiate them before they can be used.
754 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
755 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
756 path->dentry->d_inode)
759 current->total_link_count++;
760 if (current->total_link_count >= 40)
763 mnt = path->dentry->d_op->d_automount(path);
766 * The filesystem is allowed to return -EISDIR here to indicate
767 * it doesn't want to automount. For instance, autofs would do
768 * this so that its userspace daemon can mount on this dentry.
770 * However, we can only permit this if it's a terminal point in
771 * the path being looked up; if it wasn't then the remainder of
772 * the path is inaccessible and we should say so.
774 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
779 if (!mnt) /* mount collision */
783 /* lock_mount() may release path->mnt on error */
787 err = finish_automount(mnt, path);
791 /* Someone else made a mount here whilst we were busy */
796 path->dentry = dget(mnt->mnt_root);
805 * Handle a dentry that is managed in some way.
806 * - Flagged for transit management (autofs)
807 * - Flagged as mountpoint
808 * - Flagged as automount point
810 * This may only be called in refwalk mode.
812 * Serialization is taken care of in namespace.c
814 static int follow_managed(struct path *path, unsigned flags)
816 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
818 bool need_mntput = false;
821 /* Given that we're not holding a lock here, we retain the value in a
822 * local variable for each dentry as we look at it so that we don't see
823 * the components of that value change under us */
824 while (managed = ACCESS_ONCE(path->dentry->d_flags),
825 managed &= DCACHE_MANAGED_DENTRY,
826 unlikely(managed != 0)) {
827 /* Allow the filesystem to manage the transit without i_mutex
829 if (managed & DCACHE_MANAGE_TRANSIT) {
830 BUG_ON(!path->dentry->d_op);
831 BUG_ON(!path->dentry->d_op->d_manage);
832 ret = path->dentry->d_op->d_manage(path->dentry, false);
837 /* Transit to a mounted filesystem. */
838 if (managed & DCACHE_MOUNTED) {
839 struct vfsmount *mounted = lookup_mnt(path);
845 path->dentry = dget(mounted->mnt_root);
850 /* Something is mounted on this dentry in another
851 * namespace and/or whatever was mounted there in this
852 * namespace got unmounted before we managed to get the
856 /* Handle an automount point */
857 if (managed & DCACHE_NEED_AUTOMOUNT) {
858 ret = follow_automount(path, flags, &need_mntput);
864 /* We didn't change the current path point */
868 if (need_mntput && path->mnt == mnt)
872 return ret < 0 ? ret : need_mntput;
875 int follow_down_one(struct path *path)
877 struct vfsmount *mounted;
879 mounted = lookup_mnt(path);
884 path->dentry = dget(mounted->mnt_root);
890 static inline bool managed_dentry_might_block(struct dentry *dentry)
892 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
893 dentry->d_op->d_manage(dentry, true) < 0);
897 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
898 * we meet a managed dentry that would need blocking.
900 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
901 struct inode **inode)
904 struct vfsmount *mounted;
906 * Don't forget we might have a non-mountpoint managed dentry
907 * that wants to block transit.
909 if (unlikely(managed_dentry_might_block(path->dentry)))
912 if (!d_mountpoint(path->dentry))
915 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
919 path->dentry = mounted->mnt_root;
920 nd->flags |= LOOKUP_JUMPED;
921 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
923 * Update the inode too. We don't need to re-check the
924 * dentry sequence number here after this d_inode read,
925 * because a mount-point is always pinned.
927 *inode = path->dentry->d_inode;
932 static int follow_dotdot_rcu(struct nameidata *nd)
934 struct inode *inode = nd->inode;
939 if (nd->path.dentry == nd->root.dentry &&
940 nd->path.mnt == nd->root.mnt) {
943 if (nd->path.dentry != nd->path.mnt->mnt_root) {
944 struct dentry *old = nd->path.dentry;
945 struct dentry *parent = old->d_parent;
948 inode = parent->d_inode;
949 seq = read_seqcount_begin(&parent->d_seq);
950 if (read_seqcount_retry(&old->d_seq, nd->seq))
952 nd->path.dentry = parent;
954 if (unlikely(!path_connected(&nd->path)))
958 if (!follow_up_rcu(&nd->path))
960 inode = nd->path.dentry->d_inode;
961 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
963 while (d_mountpoint(nd->path.dentry)) {
964 struct vfsmount *mounted;
965 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
968 nd->path.mnt = mounted;
969 nd->path.dentry = mounted->mnt_root;
970 inode = nd->path.dentry->d_inode;
971 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
977 nd->flags &= ~LOOKUP_RCU;
978 if (!(nd->flags & LOOKUP_ROOT))
981 br_read_unlock(vfsmount_lock);
986 * Follow down to the covering mount currently visible to userspace. At each
987 * point, the filesystem owning that dentry may be queried as to whether the
988 * caller is permitted to proceed or not.
990 int follow_down(struct path *path)
995 while (managed = ACCESS_ONCE(path->dentry->d_flags),
996 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
997 /* Allow the filesystem to manage the transit without i_mutex
1000 * We indicate to the filesystem if someone is trying to mount
1001 * something here. This gives autofs the chance to deny anyone
1002 * other than its daemon the right to mount on its
1005 * The filesystem may sleep at this point.
1007 if (managed & DCACHE_MANAGE_TRANSIT) {
1008 BUG_ON(!path->dentry->d_op);
1009 BUG_ON(!path->dentry->d_op->d_manage);
1010 ret = path->dentry->d_op->d_manage(
1011 path->dentry, false);
1013 return ret == -EISDIR ? 0 : ret;
1016 /* Transit to a mounted filesystem. */
1017 if (managed & DCACHE_MOUNTED) {
1018 struct vfsmount *mounted = lookup_mnt(path);
1023 path->mnt = mounted;
1024 path->dentry = dget(mounted->mnt_root);
1028 /* Don't handle automount points here */
1035 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1037 static void follow_mount(struct path *path)
1039 while (d_mountpoint(path->dentry)) {
1040 struct vfsmount *mounted = lookup_mnt(path);
1045 path->mnt = mounted;
1046 path->dentry = dget(mounted->mnt_root);
1050 static int follow_dotdot(struct nameidata *nd)
1056 struct dentry *old = nd->path.dentry;
1058 if (nd->path.dentry == nd->root.dentry &&
1059 nd->path.mnt == nd->root.mnt) {
1062 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1063 /* rare case of legitimate dget_parent()... */
1064 nd->path.dentry = dget_parent(nd->path.dentry);
1066 if (unlikely(!path_connected(&nd->path))) {
1067 path_put(&nd->path);
1072 if (!follow_up(&nd->path))
1075 follow_mount(&nd->path);
1076 nd->inode = nd->path.dentry->d_inode;
1081 * Allocate a dentry with name and parent, and perform a parent
1082 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1083 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1084 * have verified that no child exists while under i_mutex.
1086 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1087 struct qstr *name, struct nameidata *nd)
1089 struct inode *inode = parent->d_inode;
1090 struct dentry *dentry;
1093 /* Don't create child dentry for a dead directory. */
1094 if (unlikely(IS_DEADDIR(inode)))
1095 return ERR_PTR(-ENOENT);
1097 dentry = d_alloc(parent, name);
1098 if (unlikely(!dentry))
1099 return ERR_PTR(-ENOMEM);
1101 old = inode->i_op->lookup(inode, dentry, nd);
1102 if (unlikely(old)) {
1110 * We already have a dentry, but require a lookup to be performed on the parent
1111 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1112 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1113 * child exists while under i_mutex.
1115 static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1116 struct nameidata *nd)
1118 struct inode *inode = parent->d_inode;
1121 /* Don't create child dentry for a dead directory. */
1122 if (unlikely(IS_DEADDIR(inode))) {
1124 return ERR_PTR(-ENOENT);
1127 old = inode->i_op->lookup(inode, dentry, nd);
1128 if (unlikely(old)) {
1136 * It's more convoluted than I'd like it to be, but... it's still fairly
1137 * small and for now I'd prefer to have fast path as straight as possible.
1138 * It _is_ time-critical.
1140 static int do_lookup(struct nameidata *nd, struct qstr *name,
1141 struct path *path, struct inode **inode)
1143 struct vfsmount *mnt = nd->path.mnt;
1144 struct dentry *dentry, *parent = nd->path.dentry;
1150 * Rename seqlock is not required here because in the off chance
1151 * of a false negative due to a concurrent rename, we're going to
1152 * do the non-racy lookup, below.
1154 if (nd->flags & LOOKUP_RCU) {
1157 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1161 /* Memory barrier in read_seqcount_begin of child is enough */
1162 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1166 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1167 status = d_revalidate(dentry, nd);
1168 if (unlikely(status <= 0)) {
1169 if (status != -ECHILD)
1174 if (unlikely(d_need_lookup(dentry)))
1177 path->dentry = dentry;
1178 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1180 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1184 if (unlazy_walk(nd, dentry))
1187 dentry = __d_lookup(parent, name);
1190 if (dentry && unlikely(d_need_lookup(dentry))) {
1195 if (unlikely(!dentry)) {
1196 struct inode *dir = parent->d_inode;
1197 BUG_ON(nd->inode != dir);
1199 mutex_lock(&dir->i_mutex);
1200 dentry = d_lookup(parent, name);
1201 if (likely(!dentry)) {
1202 dentry = d_alloc_and_lookup(parent, name, nd);
1203 if (IS_ERR(dentry)) {
1204 mutex_unlock(&dir->i_mutex);
1205 return PTR_ERR(dentry);
1210 } else if (unlikely(d_need_lookup(dentry))) {
1211 dentry = d_inode_lookup(parent, dentry, nd);
1212 if (IS_ERR(dentry)) {
1213 mutex_unlock(&dir->i_mutex);
1214 return PTR_ERR(dentry);
1220 mutex_unlock(&dir->i_mutex);
1222 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1223 status = d_revalidate(dentry, nd);
1224 if (unlikely(status <= 0)) {
1229 if (!d_invalidate(dentry)) {
1238 path->dentry = dentry;
1239 err = follow_managed(path, nd->flags);
1240 if (unlikely(err < 0)) {
1241 path_put_conditional(path, nd);
1245 nd->flags |= LOOKUP_JUMPED;
1246 *inode = path->dentry->d_inode;
1250 static inline int may_lookup(struct nameidata *nd)
1252 if (nd->flags & LOOKUP_RCU) {
1253 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1256 if (unlazy_walk(nd, NULL))
1259 return inode_permission(nd->inode, MAY_EXEC);
1262 static inline int handle_dots(struct nameidata *nd, int type)
1264 if (type == LAST_DOTDOT) {
1265 if (nd->flags & LOOKUP_RCU) {
1266 if (follow_dotdot_rcu(nd))
1269 return follow_dotdot(nd);
1274 static void terminate_walk(struct nameidata *nd)
1276 if (!(nd->flags & LOOKUP_RCU)) {
1277 path_put(&nd->path);
1279 nd->flags &= ~LOOKUP_RCU;
1280 if (!(nd->flags & LOOKUP_ROOT))
1281 nd->root.mnt = NULL;
1283 br_read_unlock(vfsmount_lock);
1288 * Do we need to follow links? We _really_ want to be able
1289 * to do this check without having to look at inode->i_op,
1290 * so we keep a cache of "no, this doesn't need follow_link"
1291 * for the common case.
1293 static inline int should_follow_link(struct inode *inode, int follow)
1295 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1296 if (likely(inode->i_op->follow_link))
1299 /* This gets set once for the inode lifetime */
1300 spin_lock(&inode->i_lock);
1301 inode->i_opflags |= IOP_NOFOLLOW;
1302 spin_unlock(&inode->i_lock);
1307 static inline int walk_component(struct nameidata *nd, struct path *path,
1308 struct qstr *name, int type, int follow)
1310 struct inode *inode;
1313 * "." and ".." are special - ".." especially so because it has
1314 * to be able to know about the current root directory and
1315 * parent relationships.
1317 if (unlikely(type != LAST_NORM))
1318 return handle_dots(nd, type);
1319 err = do_lookup(nd, name, path, &inode);
1320 if (unlikely(err)) {
1325 path_to_nameidata(path, nd);
1329 if (should_follow_link(inode, follow)) {
1330 if (nd->flags & LOOKUP_RCU) {
1331 if (unlikely(unlazy_walk(nd, path->dentry))) {
1336 BUG_ON(inode != path->dentry->d_inode);
1339 path_to_nameidata(path, nd);
1345 * This limits recursive symlink follows to 8, while
1346 * limiting consecutive symlinks to 40.
1348 * Without that kind of total limit, nasty chains of consecutive
1349 * symlinks can cause almost arbitrarily long lookups.
1351 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1355 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1356 path_put_conditional(path, nd);
1357 path_put(&nd->path);
1360 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1363 current->link_count++;
1366 struct path link = *path;
1369 res = follow_link(&link, nd, &cookie);
1371 res = walk_component(nd, path, &nd->last,
1372 nd->last_type, LOOKUP_FOLLOW);
1373 put_link(nd, &link, cookie);
1376 current->link_count--;
1382 * We really don't want to look at inode->i_op->lookup
1383 * when we don't have to. So we keep a cache bit in
1384 * the inode ->i_opflags field that says "yes, we can
1385 * do lookup on this inode".
1387 static inline int can_lookup(struct inode *inode)
1389 if (likely(inode->i_opflags & IOP_LOOKUP))
1391 if (likely(!inode->i_op->lookup))
1394 /* We do this once for the lifetime of the inode */
1395 spin_lock(&inode->i_lock);
1396 inode->i_opflags |= IOP_LOOKUP;
1397 spin_unlock(&inode->i_lock);
1403 * This is the basic name resolution function, turning a pathname into
1404 * the final dentry. We expect 'base' to be positive and a directory.
1406 * Returns 0 and nd will have valid dentry and mnt on success.
1407 * Returns error and drops reference to input namei data on failure.
1409 static int link_path_walk(const char *name, struct nameidata *nd)
1419 /* At this point we know we have a real path component. */
1426 err = may_lookup(nd);
1431 c = *(const unsigned char *)name;
1433 hash = init_name_hash();
1436 hash = partial_name_hash(c, hash);
1437 c = *(const unsigned char *)name;
1438 } while (c && (c != '/'));
1439 this.len = name - (const char *) this.name;
1440 this.hash = end_name_hash(hash);
1443 if (this.name[0] == '.') switch (this.len) {
1445 if (this.name[1] == '.') {
1447 nd->flags |= LOOKUP_JUMPED;
1453 if (likely(type == LAST_NORM)) {
1454 struct dentry *parent = nd->path.dentry;
1455 nd->flags &= ~LOOKUP_JUMPED;
1456 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1457 err = parent->d_op->d_hash(parent, nd->inode,
1464 /* remove trailing slashes? */
1466 goto last_component;
1467 while (*++name == '/');
1469 goto last_component;
1471 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1476 err = nested_symlink(&next, nd);
1480 if (can_lookup(nd->inode))
1484 /* here ends the main loop */
1488 nd->last_type = type;
1495 static int path_init(int dfd, const char *name, unsigned int flags,
1496 struct nameidata *nd, struct file **fp)
1502 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1503 nd->flags = flags | LOOKUP_JUMPED;
1505 if (flags & LOOKUP_ROOT) {
1506 struct inode *inode = nd->root.dentry->d_inode;
1508 if (!inode->i_op->lookup)
1510 retval = inode_permission(inode, MAY_EXEC);
1514 nd->path = nd->root;
1516 if (flags & LOOKUP_RCU) {
1517 br_read_lock(vfsmount_lock);
1519 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1521 path_get(&nd->path);
1526 nd->root.mnt = NULL;
1529 if (flags & LOOKUP_RCU) {
1530 br_read_lock(vfsmount_lock);
1532 nd->seq = set_root_rcu(nd);
1535 path_get(&nd->root);
1537 nd->path = nd->root;
1538 } else if (dfd == AT_FDCWD) {
1539 if (flags & LOOKUP_RCU) {
1540 struct fs_struct *fs = current->fs;
1543 br_read_lock(vfsmount_lock);
1547 seq = read_seqcount_begin(&fs->seq);
1549 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1550 } while (read_seqcount_retry(&fs->seq, seq));
1552 get_fs_pwd(current->fs, &nd->path);
1555 struct dentry *dentry;
1557 file = fget_raw_light(dfd, &fput_needed);
1562 dentry = file->f_path.dentry;
1566 if (!S_ISDIR(dentry->d_inode->i_mode))
1569 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1574 nd->path = file->f_path;
1575 if (flags & LOOKUP_RCU) {
1578 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1579 br_read_lock(vfsmount_lock);
1582 path_get(&file->f_path);
1583 fput_light(file, fput_needed);
1587 nd->inode = nd->path.dentry->d_inode;
1588 if (!(flags & LOOKUP_RCU))
1590 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1592 if (!(nd->flags & LOOKUP_ROOT))
1593 nd->root.mnt = NULL;
1595 br_read_unlock(vfsmount_lock);
1599 fput_light(file, fput_needed);
1604 static inline int lookup_last(struct nameidata *nd, struct path *path)
1606 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1607 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1609 nd->flags &= ~LOOKUP_PARENT;
1610 return walk_component(nd, path, &nd->last, nd->last_type,
1611 nd->flags & LOOKUP_FOLLOW);
1614 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1615 static int path_lookupat(int dfd, const char *name,
1616 unsigned int flags, struct nameidata *nd)
1618 struct file *base = NULL;
1623 * Path walking is largely split up into 2 different synchronisation
1624 * schemes, rcu-walk and ref-walk (explained in
1625 * Documentation/filesystems/path-lookup.txt). These share much of the
1626 * path walk code, but some things particularly setup, cleanup, and
1627 * following mounts are sufficiently divergent that functions are
1628 * duplicated. Typically there is a function foo(), and its RCU
1629 * analogue, foo_rcu().
1631 * -ECHILD is the error number of choice (just to avoid clashes) that
1632 * is returned if some aspect of an rcu-walk fails. Such an error must
1633 * be handled by restarting a traditional ref-walk (which will always
1634 * be able to complete).
1636 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1641 current->total_link_count = 0;
1642 err = link_path_walk(name, nd);
1644 if (!err && !(flags & LOOKUP_PARENT)) {
1645 err = lookup_last(nd, &path);
1648 struct path link = path;
1649 nd->flags |= LOOKUP_PARENT;
1650 err = follow_link(&link, nd, &cookie);
1652 err = lookup_last(nd, &path);
1653 put_link(nd, &link, cookie);
1658 err = complete_walk(nd);
1660 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1661 if (!nd->inode->i_op->lookup) {
1662 path_put(&nd->path);
1670 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1671 path_put(&nd->root);
1672 nd->root.mnt = NULL;
1677 static int do_path_lookup(int dfd, const char *name,
1678 unsigned int flags, struct nameidata *nd)
1680 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1681 if (unlikely(retval == -ECHILD))
1682 retval = path_lookupat(dfd, name, flags, nd);
1683 if (unlikely(retval == -ESTALE))
1684 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1686 if (likely(!retval)) {
1687 if (unlikely(!audit_dummy_context())) {
1688 if (nd->path.dentry && nd->inode)
1689 audit_inode(name, nd->path.dentry);
1695 int kern_path_parent(const char *name, struct nameidata *nd)
1697 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1700 int kern_path(const char *name, unsigned int flags, struct path *path)
1702 struct nameidata nd;
1703 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1710 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1711 * @dentry: pointer to dentry of the base directory
1712 * @mnt: pointer to vfs mount of the base directory
1713 * @name: pointer to file name
1714 * @flags: lookup flags
1715 * @path: pointer to struct path to fill
1717 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1718 const char *name, unsigned int flags,
1721 struct nameidata nd;
1723 nd.root.dentry = dentry;
1725 BUG_ON(flags & LOOKUP_PARENT);
1726 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1727 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1733 static struct dentry *__lookup_hash(struct qstr *name,
1734 struct dentry *base, struct nameidata *nd)
1736 struct inode *inode = base->d_inode;
1737 struct dentry *dentry;
1740 err = inode_permission(inode, MAY_EXEC);
1742 return ERR_PTR(err);
1745 * Don't bother with __d_lookup: callers are for creat as
1746 * well as unlink, so a lot of the time it would cost
1749 dentry = d_lookup(base, name);
1751 if (dentry && d_need_lookup(dentry)) {
1753 * __lookup_hash is called with the parent dir's i_mutex already
1754 * held, so we are good to go here.
1756 dentry = d_inode_lookup(base, dentry, nd);
1761 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1762 int status = d_revalidate(dentry, nd);
1763 if (unlikely(status <= 0)) {
1765 * The dentry failed validation.
1766 * If d_revalidate returned 0 attempt to invalidate
1767 * the dentry otherwise d_revalidate is asking us
1768 * to return a fail status.
1772 return ERR_PTR(status);
1773 } else if (!d_invalidate(dentry)) {
1781 dentry = d_alloc_and_lookup(base, name, nd);
1787 * Restricted form of lookup. Doesn't follow links, single-component only,
1788 * needs parent already locked. Doesn't follow mounts.
1791 static struct dentry *lookup_hash(struct nameidata *nd)
1793 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1797 * lookup_one_len - filesystem helper to lookup single pathname component
1798 * @name: pathname component to lookup
1799 * @base: base directory to lookup from
1800 * @len: maximum length @len should be interpreted to
1802 * Note that this routine is purely a helper for filesystem usage and should
1803 * not be called by generic code. Also note that by using this function the
1804 * nameidata argument is passed to the filesystem methods and a filesystem
1805 * using this helper needs to be prepared for that.
1807 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1813 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1818 return ERR_PTR(-EACCES);
1820 hash = init_name_hash();
1822 c = *(const unsigned char *)name++;
1823 if (c == '/' || c == '\0')
1824 return ERR_PTR(-EACCES);
1825 hash = partial_name_hash(c, hash);
1827 this.hash = end_name_hash(hash);
1829 * See if the low-level filesystem might want
1830 * to use its own hash..
1832 if (base->d_flags & DCACHE_OP_HASH) {
1833 int err = base->d_op->d_hash(base, base->d_inode, &this);
1835 return ERR_PTR(err);
1838 return __lookup_hash(&this, base, NULL);
1841 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1842 struct path *path, int *empty)
1844 struct nameidata nd;
1845 char *tmp = getname_flags(name, flags, empty);
1846 int err = PTR_ERR(tmp);
1849 BUG_ON(flags & LOOKUP_PARENT);
1851 err = do_path_lookup(dfd, tmp, flags, &nd);
1859 int user_path_at(int dfd, const char __user *name, unsigned flags,
1862 return user_path_at_empty(dfd, name, flags, path, 0);
1865 static int user_path_parent(int dfd, const char __user *path,
1866 struct nameidata *nd, char **name)
1868 char *s = getname(path);
1874 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1884 * It's inline, so penalty for filesystems that don't use sticky bit is
1887 static inline int check_sticky(struct inode *dir, struct inode *inode)
1889 uid_t fsuid = current_fsuid();
1891 if (!(dir->i_mode & S_ISVTX))
1893 if (current_user_ns() != inode_userns(inode))
1895 if (inode->i_uid == fsuid)
1897 if (dir->i_uid == fsuid)
1901 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1905 * Check whether we can remove a link victim from directory dir, check
1906 * whether the type of victim is right.
1907 * 1. We can't do it if dir is read-only (done in permission())
1908 * 2. We should have write and exec permissions on dir
1909 * 3. We can't remove anything from append-only dir
1910 * 4. We can't do anything with immutable dir (done in permission())
1911 * 5. If the sticky bit on dir is set we should either
1912 * a. be owner of dir, or
1913 * b. be owner of victim, or
1914 * c. have CAP_FOWNER capability
1915 * 6. If the victim is append-only or immutable we can't do antyhing with
1916 * links pointing to it.
1917 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1918 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1919 * 9. We can't remove a root or mountpoint.
1920 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1921 * nfs_async_unlink().
1923 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1927 if (!victim->d_inode)
1930 BUG_ON(victim->d_parent->d_inode != dir);
1931 audit_inode_child(victim, dir);
1933 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1938 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1939 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1942 if (!S_ISDIR(victim->d_inode->i_mode))
1944 if (IS_ROOT(victim))
1946 } else if (S_ISDIR(victim->d_inode->i_mode))
1948 if (IS_DEADDIR(dir))
1950 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1955 /* Check whether we can create an object with dentry child in directory
1957 * 1. We can't do it if child already exists (open has special treatment for
1958 * this case, but since we are inlined it's OK)
1959 * 2. We can't do it if dir is read-only (done in permission())
1960 * 3. We should have write and exec permissions on dir
1961 * 4. We can't do it if dir is immutable (done in permission())
1963 static inline int may_create(struct inode *dir, struct dentry *child)
1967 if (IS_DEADDIR(dir))
1969 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1973 * p1 and p2 should be directories on the same fs.
1975 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1980 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1984 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1986 p = d_ancestor(p2, p1);
1988 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1989 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1993 p = d_ancestor(p1, p2);
1995 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1996 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2000 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2001 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2005 void unlock_rename(struct dentry *p1, struct dentry *p2)
2007 mutex_unlock(&p1->d_inode->i_mutex);
2009 mutex_unlock(&p2->d_inode->i_mutex);
2010 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2014 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
2015 struct nameidata *nd)
2017 int error = may_create(dir, dentry);
2022 if (!dir->i_op->create)
2023 return -EACCES; /* shouldn't it be ENOSYS? */
2026 error = security_inode_create(dir, dentry, mode);
2029 error = dir->i_op->create(dir, dentry, mode, nd);
2031 fsnotify_create(dir, dentry);
2035 static int may_open(struct path *path, int acc_mode, int flag)
2037 struct dentry *dentry = path->dentry;
2038 struct inode *inode = dentry->d_inode;
2048 switch (inode->i_mode & S_IFMT) {
2052 if (acc_mode & MAY_WRITE)
2057 if (path->mnt->mnt_flags & MNT_NODEV)
2066 error = inode_permission(inode, acc_mode);
2071 * An append-only file must be opened in append mode for writing.
2073 if (IS_APPEND(inode)) {
2074 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2080 /* O_NOATIME can only be set by the owner or superuser */
2081 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2087 static int handle_truncate(struct file *filp)
2089 struct path *path = &filp->f_path;
2090 struct inode *inode = path->dentry->d_inode;
2091 int error = get_write_access(inode);
2095 * Refuse to truncate files with mandatory locks held on them.
2097 error = locks_verify_locked(inode);
2099 error = security_path_truncate(path);
2101 error = do_truncate(path->dentry, 0,
2102 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2105 put_write_access(inode);
2109 static inline int open_to_namei_flags(int flag)
2111 if ((flag & O_ACCMODE) == 3)
2117 * Handle the last step of open()
2119 static struct file *do_last(struct nameidata *nd, struct path *path,
2120 const struct open_flags *op, const char *pathname)
2122 struct dentry *dir = nd->path.dentry;
2123 struct dentry *dentry;
2124 int open_flag = op->open_flag;
2125 int will_truncate = open_flag & O_TRUNC;
2127 int acc_mode = op->acc_mode;
2131 nd->flags &= ~LOOKUP_PARENT;
2132 nd->flags |= op->intent;
2134 switch (nd->last_type) {
2137 error = handle_dots(nd, nd->last_type);
2139 return ERR_PTR(error);
2142 error = complete_walk(nd);
2144 return ERR_PTR(error);
2145 audit_inode(pathname, nd->path.dentry);
2146 if (open_flag & O_CREAT) {
2152 error = complete_walk(nd);
2154 return ERR_PTR(error);
2155 audit_inode(pathname, dir);
2159 if (!(open_flag & O_CREAT)) {
2161 if (nd->last.name[nd->last.len])
2162 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2163 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2165 /* we _can_ be in RCU mode here */
2166 error = walk_component(nd, path, &nd->last, LAST_NORM,
2169 return ERR_PTR(error);
2170 if (error) /* symlink */
2173 error = complete_walk(nd);
2175 return ERR_PTR(error);
2178 if (nd->flags & LOOKUP_DIRECTORY) {
2179 if (!nd->inode->i_op->lookup)
2182 audit_inode(pathname, nd->path.dentry);
2186 /* create side of things */
2188 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2189 * cleared when we got to the last component we are about to look up
2191 error = complete_walk(nd);
2193 return ERR_PTR(error);
2195 audit_inode(pathname, dir);
2197 /* trailing slashes? */
2198 if (nd->last.name[nd->last.len])
2201 mutex_lock(&dir->d_inode->i_mutex);
2203 dentry = lookup_hash(nd);
2204 error = PTR_ERR(dentry);
2205 if (IS_ERR(dentry)) {
2206 mutex_unlock(&dir->d_inode->i_mutex);
2210 path->dentry = dentry;
2211 path->mnt = nd->path.mnt;
2213 /* Negative dentry, just create the file */
2214 if (!dentry->d_inode) {
2215 int mode = op->mode;
2216 if (!IS_POSIXACL(dir->d_inode))
2217 mode &= ~current_umask();
2219 * This write is needed to ensure that a
2220 * rw->ro transition does not occur between
2221 * the time when the file is created and when
2222 * a permanent write count is taken through
2223 * the 'struct file' in nameidata_to_filp().
2225 error = mnt_want_write(nd->path.mnt);
2227 goto exit_mutex_unlock;
2229 /* Don't check for write permission, don't truncate */
2230 open_flag &= ~O_TRUNC;
2232 acc_mode = MAY_OPEN;
2233 error = security_path_mknod(&nd->path, dentry, mode, 0);
2235 goto exit_mutex_unlock;
2236 error = vfs_create(dir->d_inode, dentry, mode, nd);
2238 goto exit_mutex_unlock;
2239 mutex_unlock(&dir->d_inode->i_mutex);
2240 dput(nd->path.dentry);
2241 nd->path.dentry = dentry;
2246 * It already exists.
2248 mutex_unlock(&dir->d_inode->i_mutex);
2249 audit_inode(pathname, path->dentry);
2252 if (open_flag & O_EXCL)
2255 error = follow_managed(path, nd->flags);
2260 nd->flags |= LOOKUP_JUMPED;
2263 if (!path->dentry->d_inode)
2266 if (path->dentry->d_inode->i_op->follow_link)
2269 path_to_nameidata(path, nd);
2270 nd->inode = path->dentry->d_inode;
2271 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2272 error = complete_walk(nd);
2274 return ERR_PTR(error);
2276 if (S_ISDIR(nd->inode->i_mode))
2279 if (!S_ISREG(nd->inode->i_mode))
2282 if (will_truncate) {
2283 error = mnt_want_write(nd->path.mnt);
2289 error = may_open(&nd->path, acc_mode, open_flag);
2292 filp = nameidata_to_filp(nd);
2293 if (!IS_ERR(filp)) {
2294 error = ima_file_check(filp, op->acc_mode);
2297 filp = ERR_PTR(error);
2300 if (!IS_ERR(filp)) {
2301 if (will_truncate) {
2302 error = handle_truncate(filp);
2305 filp = ERR_PTR(error);
2311 mnt_drop_write(nd->path.mnt);
2312 path_put(&nd->path);
2316 mutex_unlock(&dir->d_inode->i_mutex);
2318 path_put_conditional(path, nd);
2320 filp = ERR_PTR(error);
2324 static struct file *path_openat(int dfd, const char *pathname,
2325 struct nameidata *nd, const struct open_flags *op, int flags)
2327 struct file *base = NULL;
2332 filp = get_empty_filp();
2334 return ERR_PTR(-ENFILE);
2336 filp->f_flags = op->open_flag;
2337 nd->intent.open.file = filp;
2338 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2339 nd->intent.open.create_mode = op->mode;
2341 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2342 if (unlikely(error))
2345 current->total_link_count = 0;
2346 error = link_path_walk(pathname, nd);
2347 if (unlikely(error))
2350 filp = do_last(nd, &path, op, pathname);
2351 while (unlikely(!filp)) { /* trailing symlink */
2352 struct path link = path;
2354 if (!(nd->flags & LOOKUP_FOLLOW)) {
2355 path_put_conditional(&path, nd);
2356 path_put(&nd->path);
2357 filp = ERR_PTR(-ELOOP);
2360 nd->flags |= LOOKUP_PARENT;
2361 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2362 error = follow_link(&link, nd, &cookie);
2363 if (unlikely(error))
2364 filp = ERR_PTR(error);
2366 filp = do_last(nd, &path, op, pathname);
2367 put_link(nd, &link, cookie);
2370 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2371 path_put(&nd->root);
2374 release_open_intent(nd);
2378 filp = ERR_PTR(error);
2382 struct file *do_filp_open(int dfd, const char *pathname,
2383 const struct open_flags *op, int flags)
2385 struct nameidata nd;
2388 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2389 if (unlikely(filp == ERR_PTR(-ECHILD)))
2390 filp = path_openat(dfd, pathname, &nd, op, flags);
2391 if (unlikely(filp == ERR_PTR(-ESTALE)))
2392 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2396 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2397 const char *name, const struct open_flags *op, int flags)
2399 struct nameidata nd;
2403 nd.root.dentry = dentry;
2405 flags |= LOOKUP_ROOT;
2407 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2408 return ERR_PTR(-ELOOP);
2410 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2411 if (unlikely(file == ERR_PTR(-ECHILD)))
2412 file = path_openat(-1, name, &nd, op, flags);
2413 if (unlikely(file == ERR_PTR(-ESTALE)))
2414 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2418 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2420 struct dentry *dentry = ERR_PTR(-EEXIST);
2421 struct nameidata nd;
2422 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2424 return ERR_PTR(error);
2427 * Yucky last component or no last component at all?
2428 * (foo/., foo/.., /////)
2430 if (nd.last_type != LAST_NORM)
2432 nd.flags &= ~LOOKUP_PARENT;
2433 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2434 nd.intent.open.flags = O_EXCL;
2437 * Do the final lookup.
2439 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2440 dentry = lookup_hash(&nd);
2444 if (dentry->d_inode)
2447 * Special case - lookup gave negative, but... we had foo/bar/
2448 * From the vfs_mknod() POV we just have a negative dentry -
2449 * all is fine. Let's be bastards - you had / on the end, you've
2450 * been asking for (non-existent) directory. -ENOENT for you.
2452 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2454 dentry = ERR_PTR(-ENOENT);
2461 dentry = ERR_PTR(-EEXIST);
2463 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2468 EXPORT_SYMBOL(kern_path_create);
2470 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2472 char *tmp = getname(pathname);
2475 return ERR_CAST(tmp);
2476 res = kern_path_create(dfd, tmp, path, is_dir);
2480 EXPORT_SYMBOL(user_path_create);
2482 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2484 int error = may_create(dir, dentry);
2489 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2490 !ns_capable(inode_userns(dir), CAP_MKNOD))
2493 if (!dir->i_op->mknod)
2496 error = devcgroup_inode_mknod(mode, dev);
2500 error = security_inode_mknod(dir, dentry, mode, dev);
2504 error = dir->i_op->mknod(dir, dentry, mode, dev);
2506 fsnotify_create(dir, dentry);
2510 static int may_mknod(mode_t mode)
2512 switch (mode & S_IFMT) {
2518 case 0: /* zero mode translates to S_IFREG */
2527 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2530 struct dentry *dentry;
2537 dentry = user_path_create(dfd, filename, &path, 0);
2539 return PTR_ERR(dentry);
2541 if (!IS_POSIXACL(path.dentry->d_inode))
2542 mode &= ~current_umask();
2543 error = may_mknod(mode);
2546 error = mnt_want_write(path.mnt);
2549 error = security_path_mknod(&path, dentry, mode, dev);
2551 goto out_drop_write;
2552 switch (mode & S_IFMT) {
2553 case 0: case S_IFREG:
2554 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2556 case S_IFCHR: case S_IFBLK:
2557 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2558 new_decode_dev(dev));
2560 case S_IFIFO: case S_IFSOCK:
2561 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2565 mnt_drop_write(path.mnt);
2568 mutex_unlock(&path.dentry->d_inode->i_mutex);
2574 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2576 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2579 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2581 int error = may_create(dir, dentry);
2586 if (!dir->i_op->mkdir)
2589 mode &= (S_IRWXUGO|S_ISVTX);
2590 error = security_inode_mkdir(dir, dentry, mode);
2594 error = dir->i_op->mkdir(dir, dentry, mode);
2596 fsnotify_mkdir(dir, dentry);
2600 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2602 struct dentry *dentry;
2606 dentry = user_path_create(dfd, pathname, &path, 1);
2608 return PTR_ERR(dentry);
2610 if (!IS_POSIXACL(path.dentry->d_inode))
2611 mode &= ~current_umask();
2612 error = mnt_want_write(path.mnt);
2615 error = security_path_mkdir(&path, dentry, mode);
2617 goto out_drop_write;
2618 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2620 mnt_drop_write(path.mnt);
2623 mutex_unlock(&path.dentry->d_inode->i_mutex);
2628 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2630 return sys_mkdirat(AT_FDCWD, pathname, mode);
2634 * The dentry_unhash() helper will try to drop the dentry early: we
2635 * should have a usage count of 2 if we're the only user of this
2636 * dentry, and if that is true (possibly after pruning the dcache),
2637 * then we drop the dentry now.
2639 * A low-level filesystem can, if it choses, legally
2642 * if (!d_unhashed(dentry))
2645 * if it cannot handle the case of removing a directory
2646 * that is still in use by something else..
2648 void dentry_unhash(struct dentry *dentry)
2650 shrink_dcache_parent(dentry);
2651 spin_lock(&dentry->d_lock);
2652 if (dentry->d_count == 1)
2654 spin_unlock(&dentry->d_lock);
2657 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2659 int error = may_delete(dir, dentry, 1);
2664 if (!dir->i_op->rmdir)
2668 mutex_lock(&dentry->d_inode->i_mutex);
2671 if (d_mountpoint(dentry))
2674 error = security_inode_rmdir(dir, dentry);
2678 shrink_dcache_parent(dentry);
2679 error = dir->i_op->rmdir(dir, dentry);
2683 dentry->d_inode->i_flags |= S_DEAD;
2687 mutex_unlock(&dentry->d_inode->i_mutex);
2694 static long do_rmdir(int dfd, const char __user *pathname)
2698 struct dentry *dentry;
2699 struct nameidata nd;
2701 error = user_path_parent(dfd, pathname, &nd, &name);
2705 switch(nd.last_type) {
2717 nd.flags &= ~LOOKUP_PARENT;
2719 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2720 dentry = lookup_hash(&nd);
2721 error = PTR_ERR(dentry);
2724 if (!dentry->d_inode) {
2728 error = mnt_want_write(nd.path.mnt);
2731 error = security_path_rmdir(&nd.path, dentry);
2734 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2736 mnt_drop_write(nd.path.mnt);
2740 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2747 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2749 return do_rmdir(AT_FDCWD, pathname);
2752 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2754 int error = may_delete(dir, dentry, 0);
2759 if (!dir->i_op->unlink)
2762 mutex_lock(&dentry->d_inode->i_mutex);
2763 if (d_mountpoint(dentry))
2766 error = security_inode_unlink(dir, dentry);
2768 error = dir->i_op->unlink(dir, dentry);
2773 mutex_unlock(&dentry->d_inode->i_mutex);
2775 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2776 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2777 fsnotify_link_count(dentry->d_inode);
2785 * Make sure that the actual truncation of the file will occur outside its
2786 * directory's i_mutex. Truncate can take a long time if there is a lot of
2787 * writeout happening, and we don't want to prevent access to the directory
2788 * while waiting on the I/O.
2790 static long do_unlinkat(int dfd, const char __user *pathname)
2794 struct dentry *dentry;
2795 struct nameidata nd;
2796 struct inode *inode = NULL;
2798 error = user_path_parent(dfd, pathname, &nd, &name);
2803 if (nd.last_type != LAST_NORM)
2806 nd.flags &= ~LOOKUP_PARENT;
2808 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2809 dentry = lookup_hash(&nd);
2810 error = PTR_ERR(dentry);
2811 if (!IS_ERR(dentry)) {
2812 /* Why not before? Because we want correct error value */
2813 if (nd.last.name[nd.last.len])
2815 inode = dentry->d_inode;
2819 error = mnt_want_write(nd.path.mnt);
2822 error = security_path_unlink(&nd.path, dentry);
2825 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2827 mnt_drop_write(nd.path.mnt);
2831 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2833 iput(inode); /* truncate the inode here */
2840 error = !dentry->d_inode ? -ENOENT :
2841 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2845 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2847 if ((flag & ~AT_REMOVEDIR) != 0)
2850 if (flag & AT_REMOVEDIR)
2851 return do_rmdir(dfd, pathname);
2853 return do_unlinkat(dfd, pathname);
2856 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2858 return do_unlinkat(AT_FDCWD, pathname);
2861 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2863 int error = may_create(dir, dentry);
2868 if (!dir->i_op->symlink)
2871 error = security_inode_symlink(dir, dentry, oldname);
2875 error = dir->i_op->symlink(dir, dentry, oldname);
2877 fsnotify_create(dir, dentry);
2881 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2882 int, newdfd, const char __user *, newname)
2886 struct dentry *dentry;
2889 from = getname(oldname);
2891 return PTR_ERR(from);
2893 dentry = user_path_create(newdfd, newname, &path, 0);
2894 error = PTR_ERR(dentry);
2898 error = mnt_want_write(path.mnt);
2901 error = security_path_symlink(&path, dentry, from);
2903 goto out_drop_write;
2904 error = vfs_symlink(path.dentry->d_inode, dentry, from);
2906 mnt_drop_write(path.mnt);
2909 mutex_unlock(&path.dentry->d_inode->i_mutex);
2916 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2918 return sys_symlinkat(oldname, AT_FDCWD, newname);
2921 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2923 struct inode *inode = old_dentry->d_inode;
2929 error = may_create(dir, new_dentry);
2933 if (dir->i_sb != inode->i_sb)
2937 * A link to an append-only or immutable file cannot be created.
2939 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2941 if (!dir->i_op->link)
2943 if (S_ISDIR(inode->i_mode))
2946 error = security_inode_link(old_dentry, dir, new_dentry);
2950 mutex_lock(&inode->i_mutex);
2951 /* Make sure we don't allow creating hardlink to an unlinked file */
2952 if (inode->i_nlink == 0)
2955 error = dir->i_op->link(old_dentry, dir, new_dentry);
2956 mutex_unlock(&inode->i_mutex);
2958 fsnotify_link(dir, inode, new_dentry);
2963 * Hardlinks are often used in delicate situations. We avoid
2964 * security-related surprises by not following symlinks on the
2967 * We don't follow them on the oldname either to be compatible
2968 * with linux 2.0, and to avoid hard-linking to directories
2969 * and other special files. --ADM
2971 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2972 int, newdfd, const char __user *, newname, int, flags)
2974 struct dentry *new_dentry;
2975 struct path old_path, new_path;
2979 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
2982 * To use null names we require CAP_DAC_READ_SEARCH
2983 * This ensures that not everyone will be able to create
2984 * handlink using the passed filedescriptor.
2986 if (flags & AT_EMPTY_PATH) {
2987 if (!capable(CAP_DAC_READ_SEARCH))
2992 if (flags & AT_SYMLINK_FOLLOW)
2993 how |= LOOKUP_FOLLOW;
2995 error = user_path_at(olddfd, oldname, how, &old_path);
2999 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3000 error = PTR_ERR(new_dentry);
3001 if (IS_ERR(new_dentry))
3005 if (old_path.mnt != new_path.mnt)
3007 error = mnt_want_write(new_path.mnt);
3010 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3012 goto out_drop_write;
3013 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3015 mnt_drop_write(new_path.mnt);
3018 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3019 path_put(&new_path);
3021 path_put(&old_path);
3026 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3028 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3032 * The worst of all namespace operations - renaming directory. "Perverted"
3033 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3035 * a) we can get into loop creation. Check is done in is_subdir().
3036 * b) race potential - two innocent renames can create a loop together.
3037 * That's where 4.4 screws up. Current fix: serialization on
3038 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3040 * c) we have to lock _three_ objects - parents and victim (if it exists).
3041 * And that - after we got ->i_mutex on parents (until then we don't know
3042 * whether the target exists). Solution: try to be smart with locking
3043 * order for inodes. We rely on the fact that tree topology may change
3044 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3045 * move will be locked. Thus we can rank directories by the tree
3046 * (ancestors first) and rank all non-directories after them.
3047 * That works since everybody except rename does "lock parent, lookup,
3048 * lock child" and rename is under ->s_vfs_rename_mutex.
3049 * HOWEVER, it relies on the assumption that any object with ->lookup()
3050 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3051 * we'd better make sure that there's no link(2) for them.
3052 * d) conversion from fhandle to dentry may come in the wrong moment - when
3053 * we are removing the target. Solution: we will have to grab ->i_mutex
3054 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3055 * ->i_mutex on parents, which works but leads to some truly excessive
3058 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3059 struct inode *new_dir, struct dentry *new_dentry)
3062 struct inode *target = new_dentry->d_inode;
3065 * If we are going to change the parent - check write permissions,
3066 * we'll need to flip '..'.
3068 if (new_dir != old_dir) {
3069 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3074 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3080 mutex_lock(&target->i_mutex);
3083 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3087 shrink_dcache_parent(new_dentry);
3088 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3093 target->i_flags |= S_DEAD;
3094 dont_mount(new_dentry);
3098 mutex_unlock(&target->i_mutex);
3101 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3102 d_move(old_dentry,new_dentry);
3106 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3107 struct inode *new_dir, struct dentry *new_dentry)
3109 struct inode *target = new_dentry->d_inode;
3112 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3118 mutex_lock(&target->i_mutex);
3121 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3124 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3129 dont_mount(new_dentry);
3130 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3131 d_move(old_dentry, new_dentry);
3134 mutex_unlock(&target->i_mutex);
3139 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3140 struct inode *new_dir, struct dentry *new_dentry)
3143 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3144 const unsigned char *old_name;
3146 if (old_dentry->d_inode == new_dentry->d_inode)
3149 error = may_delete(old_dir, old_dentry, is_dir);
3153 if (!new_dentry->d_inode)
3154 error = may_create(new_dir, new_dentry);
3156 error = may_delete(new_dir, new_dentry, is_dir);
3160 if (!old_dir->i_op->rename)
3163 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3166 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3168 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3170 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3171 new_dentry->d_inode, old_dentry);
3172 fsnotify_oldname_free(old_name);
3177 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3178 int, newdfd, const char __user *, newname)
3180 struct dentry *old_dir, *new_dir;
3181 struct dentry *old_dentry, *new_dentry;
3182 struct dentry *trap;
3183 struct nameidata oldnd, newnd;
3188 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3192 error = user_path_parent(newdfd, newname, &newnd, &to);
3197 if (oldnd.path.mnt != newnd.path.mnt)
3200 old_dir = oldnd.path.dentry;
3202 if (oldnd.last_type != LAST_NORM)
3205 new_dir = newnd.path.dentry;
3206 if (newnd.last_type != LAST_NORM)
3209 oldnd.flags &= ~LOOKUP_PARENT;
3210 newnd.flags &= ~LOOKUP_PARENT;
3211 newnd.flags |= LOOKUP_RENAME_TARGET;
3213 trap = lock_rename(new_dir, old_dir);
3215 old_dentry = lookup_hash(&oldnd);
3216 error = PTR_ERR(old_dentry);
3217 if (IS_ERR(old_dentry))
3219 /* source must exist */
3221 if (!old_dentry->d_inode)
3223 /* unless the source is a directory trailing slashes give -ENOTDIR */
3224 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3226 if (oldnd.last.name[oldnd.last.len])
3228 if (newnd.last.name[newnd.last.len])
3231 /* source should not be ancestor of target */
3233 if (old_dentry == trap)
3235 new_dentry = lookup_hash(&newnd);
3236 error = PTR_ERR(new_dentry);
3237 if (IS_ERR(new_dentry))
3239 /* target should not be an ancestor of source */
3241 if (new_dentry == trap)
3244 error = mnt_want_write(oldnd.path.mnt);
3247 error = security_path_rename(&oldnd.path, old_dentry,
3248 &newnd.path, new_dentry);
3251 error = vfs_rename(old_dir->d_inode, old_dentry,
3252 new_dir->d_inode, new_dentry);
3254 mnt_drop_write(oldnd.path.mnt);
3260 unlock_rename(new_dir, old_dir);
3262 path_put(&newnd.path);
3265 path_put(&oldnd.path);
3271 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3273 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3276 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3280 len = PTR_ERR(link);
3285 if (len > (unsigned) buflen)
3287 if (copy_to_user(buffer, link, len))
3294 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3295 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3296 * using) it for any given inode is up to filesystem.
3298 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3300 struct nameidata nd;
3305 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3307 return PTR_ERR(cookie);
3309 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3310 if (dentry->d_inode->i_op->put_link)
3311 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3315 int vfs_follow_link(struct nameidata *nd, const char *link)
3317 return __vfs_follow_link(nd, link);
3320 /* get the link contents into pagecache */
3321 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3325 struct address_space *mapping = dentry->d_inode->i_mapping;
3326 page = read_mapping_page(mapping, 0, NULL);
3331 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3335 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3337 struct page *page = NULL;
3338 char *s = page_getlink(dentry, &page);
3339 int res = vfs_readlink(dentry,buffer,buflen,s);
3342 page_cache_release(page);
3347 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3349 struct page *page = NULL;
3350 nd_set_link(nd, page_getlink(dentry, &page));
3354 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3356 struct page *page = cookie;
3360 page_cache_release(page);
3365 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3367 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3369 struct address_space *mapping = inode->i_mapping;
3374 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3376 flags |= AOP_FLAG_NOFS;
3379 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3380 flags, &page, &fsdata);
3384 kaddr = kmap_atomic(page, KM_USER0);
3385 memcpy(kaddr, symname, len-1);
3386 kunmap_atomic(kaddr, KM_USER0);
3388 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3395 mark_inode_dirty(inode);
3401 int page_symlink(struct inode *inode, const char *symname, int len)
3403 return __page_symlink(inode, symname, len,
3404 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3407 const struct inode_operations page_symlink_inode_operations = {
3408 .readlink = generic_readlink,
3409 .follow_link = page_follow_link_light,
3410 .put_link = page_put_link,
3413 EXPORT_SYMBOL(user_path_at);
3414 EXPORT_SYMBOL(follow_down_one);
3415 EXPORT_SYMBOL(follow_down);
3416 EXPORT_SYMBOL(follow_up);
3417 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3418 EXPORT_SYMBOL(getname);
3419 EXPORT_SYMBOL(lock_rename);
3420 EXPORT_SYMBOL(lookup_one_len);
3421 EXPORT_SYMBOL(page_follow_link_light);
3422 EXPORT_SYMBOL(page_put_link);
3423 EXPORT_SYMBOL(page_readlink);
3424 EXPORT_SYMBOL(__page_symlink);
3425 EXPORT_SYMBOL(page_symlink);
3426 EXPORT_SYMBOL(page_symlink_inode_operations);
3427 EXPORT_SYMBOL(kern_path);
3428 EXPORT_SYMBOL(vfs_path_lookup);
3429 EXPORT_SYMBOL(inode_permission);
3430 EXPORT_SYMBOL(unlock_rename);
3431 EXPORT_SYMBOL(vfs_create);
3432 EXPORT_SYMBOL(vfs_follow_link);
3433 EXPORT_SYMBOL(vfs_link);
3434 EXPORT_SYMBOL(vfs_mkdir);
3435 EXPORT_SYMBOL(vfs_mknod);
3436 EXPORT_SYMBOL(generic_permission);
3437 EXPORT_SYMBOL(vfs_readlink);
3438 EXPORT_SYMBOL(vfs_rename);
3439 EXPORT_SYMBOL(vfs_rmdir);
3440 EXPORT_SYMBOL(vfs_symlink);
3441 EXPORT_SYMBOL(vfs_unlink);
3442 EXPORT_SYMBOL(dentry_unhash);
3443 EXPORT_SYMBOL(generic_readlink);