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;
411 struct super_block *sb = mnt->mnt_sb;
413 /* Bind mounts and multi-root filesystems can have disconnected paths */
414 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
417 return is_subdir(path->dentry, mnt->mnt_root);
421 * Path walking has 2 modes, rcu-walk and ref-walk (see
422 * Documentation/filesystems/path-lookup.txt). In situations when we can't
423 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
424 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
425 * mode. Refcounts are grabbed at the last known good point before rcu-walk
426 * got stuck, so ref-walk may continue from there. If this is not successful
427 * (eg. a seqcount has changed), then failure is returned and it's up to caller
428 * to restart the path walk from the beginning in ref-walk mode.
432 * unlazy_walk - try to switch to ref-walk mode.
433 * @nd: nameidata pathwalk data
434 * @dentry: child of nd->path.dentry or NULL
435 * Returns: 0 on success, -ECHILD on failure
437 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
438 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
439 * @nd or NULL. Must be called from rcu-walk context.
441 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
443 struct fs_struct *fs = current->fs;
444 struct dentry *parent = nd->path.dentry;
447 BUG_ON(!(nd->flags & LOOKUP_RCU));
448 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
450 spin_lock(&fs->lock);
451 if (nd->root.mnt != fs->root.mnt ||
452 nd->root.dentry != fs->root.dentry)
455 spin_lock(&parent->d_lock);
457 if (!__d_rcu_to_refcount(parent, nd->seq))
459 BUG_ON(nd->inode != parent->d_inode);
461 if (dentry->d_parent != parent)
463 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
464 if (!__d_rcu_to_refcount(dentry, nd->seq))
467 * If the sequence check on the child dentry passed, then
468 * the child has not been removed from its parent. This
469 * means the parent dentry must be valid and able to take
470 * a reference at this point.
472 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
473 BUG_ON(!parent->d_count);
475 spin_unlock(&dentry->d_lock);
477 spin_unlock(&parent->d_lock);
480 spin_unlock(&fs->lock);
482 mntget(nd->path.mnt);
485 br_read_unlock(vfsmount_lock);
486 nd->flags &= ~LOOKUP_RCU;
490 spin_unlock(&dentry->d_lock);
492 spin_unlock(&parent->d_lock);
495 spin_unlock(&fs->lock);
500 * release_open_intent - free up open intent resources
501 * @nd: pointer to nameidata
503 void release_open_intent(struct nameidata *nd)
505 struct file *file = nd->intent.open.file;
507 if (file && !IS_ERR(file)) {
508 if (file->f_path.dentry == NULL)
515 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
517 return dentry->d_op->d_revalidate(dentry, nd);
521 * complete_walk - successful completion of path walk
522 * @nd: pointer nameidata
524 * If we had been in RCU mode, drop out of it and legitimize nd->path.
525 * Revalidate the final result, unless we'd already done that during
526 * the path walk or the filesystem doesn't ask for it. Return 0 on
527 * success, -error on failure. In case of failure caller does not
528 * need to drop nd->path.
530 static int complete_walk(struct nameidata *nd)
532 struct dentry *dentry = nd->path.dentry;
535 if (nd->flags & LOOKUP_RCU) {
536 nd->flags &= ~LOOKUP_RCU;
537 if (!(nd->flags & LOOKUP_ROOT))
539 spin_lock(&dentry->d_lock);
540 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
541 spin_unlock(&dentry->d_lock);
543 br_read_unlock(vfsmount_lock);
546 BUG_ON(nd->inode != dentry->d_inode);
547 spin_unlock(&dentry->d_lock);
548 mntget(nd->path.mnt);
550 br_read_unlock(vfsmount_lock);
553 if (likely(!(nd->flags & LOOKUP_JUMPED)))
556 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
559 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
562 /* Note: we do not d_invalidate() */
563 status = d_revalidate(dentry, nd);
574 static __always_inline void set_root(struct nameidata *nd)
576 get_fs_root(current->fs, &nd->root);
579 static int link_path_walk(const char *, struct nameidata *);
581 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
583 struct fs_struct *fs = current->fs;
587 seq = read_seqcount_begin(&fs->seq);
589 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
590 } while (read_seqcount_retry(&fs->seq, seq));
594 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
607 nd->flags |= LOOKUP_JUMPED;
609 nd->inode = nd->path.dentry->d_inode;
611 ret = link_path_walk(link, nd);
615 return PTR_ERR(link);
618 static void path_put_conditional(struct path *path, struct nameidata *nd)
621 if (path->mnt != nd->path.mnt)
625 static inline void path_to_nameidata(const struct path *path,
626 struct nameidata *nd)
628 if (!(nd->flags & LOOKUP_RCU)) {
629 dput(nd->path.dentry);
630 if (nd->path.mnt != path->mnt)
631 mntput(nd->path.mnt);
633 nd->path.mnt = path->mnt;
634 nd->path.dentry = path->dentry;
637 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
639 struct inode *inode = link->dentry->d_inode;
640 if (!IS_ERR(cookie) && inode->i_op->put_link)
641 inode->i_op->put_link(link->dentry, nd, cookie);
645 static __always_inline int
646 follow_link(struct path *link, struct nameidata *nd, void **p)
649 struct dentry *dentry = link->dentry;
651 BUG_ON(nd->flags & LOOKUP_RCU);
653 if (link->mnt == nd->path.mnt)
656 if (unlikely(current->total_link_count >= 40)) {
657 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
662 current->total_link_count++;
664 touch_atime(link->mnt, dentry);
665 nd_set_link(nd, NULL);
667 error = security_inode_follow_link(link->dentry, nd);
669 *p = ERR_PTR(error); /* no ->put_link(), please */
674 nd->last_type = LAST_BIND;
675 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
678 char *s = nd_get_link(nd);
681 error = __vfs_follow_link(nd, s);
682 else if (nd->last_type == LAST_BIND) {
683 nd->flags |= LOOKUP_JUMPED;
684 nd->inode = nd->path.dentry->d_inode;
685 if (nd->inode->i_op->follow_link) {
686 /* stepped on a _really_ weird one */
695 static int follow_up_rcu(struct path *path)
697 struct vfsmount *parent;
698 struct dentry *mountpoint;
700 parent = path->mnt->mnt_parent;
701 if (parent == path->mnt)
703 mountpoint = path->mnt->mnt_mountpoint;
704 path->dentry = mountpoint;
709 int follow_up(struct path *path)
711 struct vfsmount *parent;
712 struct dentry *mountpoint;
714 br_read_lock(vfsmount_lock);
715 parent = path->mnt->mnt_parent;
716 if (parent == path->mnt) {
717 br_read_unlock(vfsmount_lock);
721 mountpoint = dget(path->mnt->mnt_mountpoint);
722 br_read_unlock(vfsmount_lock);
724 path->dentry = mountpoint;
731 * Perform an automount
732 * - return -EISDIR to tell follow_managed() to stop and return the path we
735 static int follow_automount(struct path *path, unsigned flags,
738 struct vfsmount *mnt;
741 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
744 /* We don't want to mount if someone's just doing a stat -
745 * unless they're stat'ing a directory and appended a '/' to
748 * We do, however, want to mount if someone wants to open or
749 * create a file of any type under the mountpoint, wants to
750 * traverse through the mountpoint or wants to open the
751 * mounted directory. Also, autofs may mark negative dentries
752 * as being automount points. These will need the attentions
753 * of the daemon to instantiate them before they can be used.
755 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
756 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
757 path->dentry->d_inode)
760 current->total_link_count++;
761 if (current->total_link_count >= 40)
764 mnt = path->dentry->d_op->d_automount(path);
767 * The filesystem is allowed to return -EISDIR here to indicate
768 * it doesn't want to automount. For instance, autofs would do
769 * this so that its userspace daemon can mount on this dentry.
771 * However, we can only permit this if it's a terminal point in
772 * the path being looked up; if it wasn't then the remainder of
773 * the path is inaccessible and we should say so.
775 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
780 if (!mnt) /* mount collision */
784 /* lock_mount() may release path->mnt on error */
788 err = finish_automount(mnt, path);
792 /* Someone else made a mount here whilst we were busy */
797 path->dentry = dget(mnt->mnt_root);
806 * Handle a dentry that is managed in some way.
807 * - Flagged for transit management (autofs)
808 * - Flagged as mountpoint
809 * - Flagged as automount point
811 * This may only be called in refwalk mode.
813 * Serialization is taken care of in namespace.c
815 static int follow_managed(struct path *path, unsigned flags)
817 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
819 bool need_mntput = false;
822 /* Given that we're not holding a lock here, we retain the value in a
823 * local variable for each dentry as we look at it so that we don't see
824 * the components of that value change under us */
825 while (managed = ACCESS_ONCE(path->dentry->d_flags),
826 managed &= DCACHE_MANAGED_DENTRY,
827 unlikely(managed != 0)) {
828 /* Allow the filesystem to manage the transit without i_mutex
830 if (managed & DCACHE_MANAGE_TRANSIT) {
831 BUG_ON(!path->dentry->d_op);
832 BUG_ON(!path->dentry->d_op->d_manage);
833 ret = path->dentry->d_op->d_manage(path->dentry, false);
838 /* Transit to a mounted filesystem. */
839 if (managed & DCACHE_MOUNTED) {
840 struct vfsmount *mounted = lookup_mnt(path);
846 path->dentry = dget(mounted->mnt_root);
851 /* Something is mounted on this dentry in another
852 * namespace and/or whatever was mounted there in this
853 * namespace got unmounted before we managed to get the
857 /* Handle an automount point */
858 if (managed & DCACHE_NEED_AUTOMOUNT) {
859 ret = follow_automount(path, flags, &need_mntput);
865 /* We didn't change the current path point */
869 if (need_mntput && path->mnt == mnt)
873 return ret < 0 ? ret : need_mntput;
876 int follow_down_one(struct path *path)
878 struct vfsmount *mounted;
880 mounted = lookup_mnt(path);
885 path->dentry = dget(mounted->mnt_root);
891 static inline bool managed_dentry_might_block(struct dentry *dentry)
893 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
894 dentry->d_op->d_manage(dentry, true) < 0);
898 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
899 * we meet a managed dentry that would need blocking.
901 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
902 struct inode **inode)
905 struct vfsmount *mounted;
907 * Don't forget we might have a non-mountpoint managed dentry
908 * that wants to block transit.
910 if (unlikely(managed_dentry_might_block(path->dentry)))
913 if (!d_mountpoint(path->dentry))
916 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
920 path->dentry = mounted->mnt_root;
921 nd->flags |= LOOKUP_JUMPED;
922 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
924 * Update the inode too. We don't need to re-check the
925 * dentry sequence number here after this d_inode read,
926 * because a mount-point is always pinned.
928 *inode = path->dentry->d_inode;
933 static int follow_dotdot_rcu(struct nameidata *nd)
935 struct inode *inode = nd->inode;
940 if (nd->path.dentry == nd->root.dentry &&
941 nd->path.mnt == nd->root.mnt) {
944 if (nd->path.dentry != nd->path.mnt->mnt_root) {
945 struct dentry *old = nd->path.dentry;
946 struct dentry *parent = old->d_parent;
949 inode = parent->d_inode;
950 seq = read_seqcount_begin(&parent->d_seq);
951 if (read_seqcount_retry(&old->d_seq, nd->seq))
953 nd->path.dentry = parent;
955 if (unlikely(!path_connected(&nd->path)))
959 if (!follow_up_rcu(&nd->path))
961 inode = nd->path.dentry->d_inode;
962 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
964 while (d_mountpoint(nd->path.dentry)) {
965 struct vfsmount *mounted;
966 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
969 nd->path.mnt = mounted;
970 nd->path.dentry = mounted->mnt_root;
971 inode = nd->path.dentry->d_inode;
972 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
978 nd->flags &= ~LOOKUP_RCU;
979 if (!(nd->flags & LOOKUP_ROOT))
982 br_read_unlock(vfsmount_lock);
987 * Follow down to the covering mount currently visible to userspace. At each
988 * point, the filesystem owning that dentry may be queried as to whether the
989 * caller is permitted to proceed or not.
991 int follow_down(struct path *path)
996 while (managed = ACCESS_ONCE(path->dentry->d_flags),
997 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
998 /* Allow the filesystem to manage the transit without i_mutex
1001 * We indicate to the filesystem if someone is trying to mount
1002 * something here. This gives autofs the chance to deny anyone
1003 * other than its daemon the right to mount on its
1006 * The filesystem may sleep at this point.
1008 if (managed & DCACHE_MANAGE_TRANSIT) {
1009 BUG_ON(!path->dentry->d_op);
1010 BUG_ON(!path->dentry->d_op->d_manage);
1011 ret = path->dentry->d_op->d_manage(
1012 path->dentry, false);
1014 return ret == -EISDIR ? 0 : ret;
1017 /* Transit to a mounted filesystem. */
1018 if (managed & DCACHE_MOUNTED) {
1019 struct vfsmount *mounted = lookup_mnt(path);
1024 path->mnt = mounted;
1025 path->dentry = dget(mounted->mnt_root);
1029 /* Don't handle automount points here */
1036 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1038 static void follow_mount(struct path *path)
1040 while (d_mountpoint(path->dentry)) {
1041 struct vfsmount *mounted = lookup_mnt(path);
1046 path->mnt = mounted;
1047 path->dentry = dget(mounted->mnt_root);
1051 static int follow_dotdot(struct nameidata *nd)
1057 struct dentry *old = nd->path.dentry;
1059 if (nd->path.dentry == nd->root.dentry &&
1060 nd->path.mnt == nd->root.mnt) {
1063 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1064 /* rare case of legitimate dget_parent()... */
1065 nd->path.dentry = dget_parent(nd->path.dentry);
1067 if (unlikely(!path_connected(&nd->path))) {
1068 path_put(&nd->path);
1073 if (!follow_up(&nd->path))
1076 follow_mount(&nd->path);
1077 nd->inode = nd->path.dentry->d_inode;
1082 * Allocate a dentry with name and parent, and perform a parent
1083 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1084 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1085 * have verified that no child exists while under i_mutex.
1087 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1088 struct qstr *name, struct nameidata *nd)
1090 struct inode *inode = parent->d_inode;
1091 struct dentry *dentry;
1094 /* Don't create child dentry for a dead directory. */
1095 if (unlikely(IS_DEADDIR(inode)))
1096 return ERR_PTR(-ENOENT);
1098 dentry = d_alloc(parent, name);
1099 if (unlikely(!dentry))
1100 return ERR_PTR(-ENOMEM);
1102 old = inode->i_op->lookup(inode, dentry, nd);
1103 if (unlikely(old)) {
1111 * We already have a dentry, but require a lookup to be performed on the parent
1112 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1113 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1114 * child exists while under i_mutex.
1116 static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1117 struct nameidata *nd)
1119 struct inode *inode = parent->d_inode;
1122 /* Don't create child dentry for a dead directory. */
1123 if (unlikely(IS_DEADDIR(inode))) {
1125 return ERR_PTR(-ENOENT);
1128 old = inode->i_op->lookup(inode, dentry, nd);
1129 if (unlikely(old)) {
1137 * It's more convoluted than I'd like it to be, but... it's still fairly
1138 * small and for now I'd prefer to have fast path as straight as possible.
1139 * It _is_ time-critical.
1141 static int do_lookup(struct nameidata *nd, struct qstr *name,
1142 struct path *path, struct inode **inode)
1144 struct vfsmount *mnt = nd->path.mnt;
1145 struct dentry *dentry, *parent = nd->path.dentry;
1151 * Rename seqlock is not required here because in the off chance
1152 * of a false negative due to a concurrent rename, we're going to
1153 * do the non-racy lookup, below.
1155 if (nd->flags & LOOKUP_RCU) {
1158 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1162 /* Memory barrier in read_seqcount_begin of child is enough */
1163 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1167 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1168 status = d_revalidate(dentry, nd);
1169 if (unlikely(status <= 0)) {
1170 if (status != -ECHILD)
1175 if (unlikely(d_need_lookup(dentry)))
1178 path->dentry = dentry;
1179 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1181 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1185 if (unlazy_walk(nd, dentry))
1188 dentry = __d_lookup(parent, name);
1191 if (dentry && unlikely(d_need_lookup(dentry))) {
1196 if (unlikely(!dentry)) {
1197 struct inode *dir = parent->d_inode;
1198 BUG_ON(nd->inode != dir);
1200 mutex_lock(&dir->i_mutex);
1201 dentry = d_lookup(parent, name);
1202 if (likely(!dentry)) {
1203 dentry = d_alloc_and_lookup(parent, name, nd);
1204 if (IS_ERR(dentry)) {
1205 mutex_unlock(&dir->i_mutex);
1206 return PTR_ERR(dentry);
1211 } else if (unlikely(d_need_lookup(dentry))) {
1212 dentry = d_inode_lookup(parent, dentry, nd);
1213 if (IS_ERR(dentry)) {
1214 mutex_unlock(&dir->i_mutex);
1215 return PTR_ERR(dentry);
1221 mutex_unlock(&dir->i_mutex);
1223 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1224 status = d_revalidate(dentry, nd);
1225 if (unlikely(status <= 0)) {
1230 if (!d_invalidate(dentry)) {
1239 path->dentry = dentry;
1240 err = follow_managed(path, nd->flags);
1241 if (unlikely(err < 0)) {
1242 path_put_conditional(path, nd);
1246 nd->flags |= LOOKUP_JUMPED;
1247 *inode = path->dentry->d_inode;
1251 static inline int may_lookup(struct nameidata *nd)
1253 if (nd->flags & LOOKUP_RCU) {
1254 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1257 if (unlazy_walk(nd, NULL))
1260 return inode_permission(nd->inode, MAY_EXEC);
1263 static inline int handle_dots(struct nameidata *nd, int type)
1265 if (type == LAST_DOTDOT) {
1266 if (nd->flags & LOOKUP_RCU) {
1267 if (follow_dotdot_rcu(nd))
1270 return follow_dotdot(nd);
1275 static void terminate_walk(struct nameidata *nd)
1277 if (!(nd->flags & LOOKUP_RCU)) {
1278 path_put(&nd->path);
1280 nd->flags &= ~LOOKUP_RCU;
1281 if (!(nd->flags & LOOKUP_ROOT))
1282 nd->root.mnt = NULL;
1284 br_read_unlock(vfsmount_lock);
1289 * Do we need to follow links? We _really_ want to be able
1290 * to do this check without having to look at inode->i_op,
1291 * so we keep a cache of "no, this doesn't need follow_link"
1292 * for the common case.
1294 static inline int should_follow_link(struct inode *inode, int follow)
1296 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1297 if (likely(inode->i_op->follow_link))
1300 /* This gets set once for the inode lifetime */
1301 spin_lock(&inode->i_lock);
1302 inode->i_opflags |= IOP_NOFOLLOW;
1303 spin_unlock(&inode->i_lock);
1308 static inline int walk_component(struct nameidata *nd, struct path *path,
1309 struct qstr *name, int type, int follow)
1311 struct inode *inode;
1314 * "." and ".." are special - ".." especially so because it has
1315 * to be able to know about the current root directory and
1316 * parent relationships.
1318 if (unlikely(type != LAST_NORM))
1319 return handle_dots(nd, type);
1320 err = do_lookup(nd, name, path, &inode);
1321 if (unlikely(err)) {
1326 path_to_nameidata(path, nd);
1330 if (should_follow_link(inode, follow)) {
1331 if (nd->flags & LOOKUP_RCU) {
1332 if (unlikely(unlazy_walk(nd, path->dentry))) {
1337 BUG_ON(inode != path->dentry->d_inode);
1340 path_to_nameidata(path, nd);
1346 * This limits recursive symlink follows to 8, while
1347 * limiting consecutive symlinks to 40.
1349 * Without that kind of total limit, nasty chains of consecutive
1350 * symlinks can cause almost arbitrarily long lookups.
1352 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1356 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1357 path_put_conditional(path, nd);
1358 path_put(&nd->path);
1361 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1364 current->link_count++;
1367 struct path link = *path;
1370 res = follow_link(&link, nd, &cookie);
1372 res = walk_component(nd, path, &nd->last,
1373 nd->last_type, LOOKUP_FOLLOW);
1374 put_link(nd, &link, cookie);
1377 current->link_count--;
1383 * We really don't want to look at inode->i_op->lookup
1384 * when we don't have to. So we keep a cache bit in
1385 * the inode ->i_opflags field that says "yes, we can
1386 * do lookup on this inode".
1388 static inline int can_lookup(struct inode *inode)
1390 if (likely(inode->i_opflags & IOP_LOOKUP))
1392 if (likely(!inode->i_op->lookup))
1395 /* We do this once for the lifetime of the inode */
1396 spin_lock(&inode->i_lock);
1397 inode->i_opflags |= IOP_LOOKUP;
1398 spin_unlock(&inode->i_lock);
1404 * This is the basic name resolution function, turning a pathname into
1405 * the final dentry. We expect 'base' to be positive and a directory.
1407 * Returns 0 and nd will have valid dentry and mnt on success.
1408 * Returns error and drops reference to input namei data on failure.
1410 static int link_path_walk(const char *name, struct nameidata *nd)
1420 /* At this point we know we have a real path component. */
1427 err = may_lookup(nd);
1432 c = *(const unsigned char *)name;
1434 hash = init_name_hash();
1437 hash = partial_name_hash(c, hash);
1438 c = *(const unsigned char *)name;
1439 } while (c && (c != '/'));
1440 this.len = name - (const char *) this.name;
1441 this.hash = end_name_hash(hash);
1444 if (this.name[0] == '.') switch (this.len) {
1446 if (this.name[1] == '.') {
1448 nd->flags |= LOOKUP_JUMPED;
1454 if (likely(type == LAST_NORM)) {
1455 struct dentry *parent = nd->path.dentry;
1456 nd->flags &= ~LOOKUP_JUMPED;
1457 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1458 err = parent->d_op->d_hash(parent, nd->inode,
1465 /* remove trailing slashes? */
1467 goto last_component;
1468 while (*++name == '/');
1470 goto last_component;
1472 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1477 err = nested_symlink(&next, nd);
1481 if (can_lookup(nd->inode))
1485 /* here ends the main loop */
1489 nd->last_type = type;
1496 static int path_init(int dfd, const char *name, unsigned int flags,
1497 struct nameidata *nd, struct file **fp)
1503 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1504 nd->flags = flags | LOOKUP_JUMPED;
1506 if (flags & LOOKUP_ROOT) {
1507 struct inode *inode = nd->root.dentry->d_inode;
1509 if (!inode->i_op->lookup)
1511 retval = inode_permission(inode, MAY_EXEC);
1515 nd->path = nd->root;
1517 if (flags & LOOKUP_RCU) {
1518 br_read_lock(vfsmount_lock);
1520 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1522 path_get(&nd->path);
1527 nd->root.mnt = NULL;
1530 if (flags & LOOKUP_RCU) {
1531 br_read_lock(vfsmount_lock);
1533 nd->seq = set_root_rcu(nd);
1536 path_get(&nd->root);
1538 nd->path = nd->root;
1539 } else if (dfd == AT_FDCWD) {
1540 if (flags & LOOKUP_RCU) {
1541 struct fs_struct *fs = current->fs;
1544 br_read_lock(vfsmount_lock);
1548 seq = read_seqcount_begin(&fs->seq);
1550 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1551 } while (read_seqcount_retry(&fs->seq, seq));
1553 get_fs_pwd(current->fs, &nd->path);
1556 struct dentry *dentry;
1558 file = fget_raw_light(dfd, &fput_needed);
1563 dentry = file->f_path.dentry;
1567 if (!S_ISDIR(dentry->d_inode->i_mode))
1570 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1575 nd->path = file->f_path;
1576 if (flags & LOOKUP_RCU) {
1579 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1580 br_read_lock(vfsmount_lock);
1583 path_get(&file->f_path);
1584 fput_light(file, fput_needed);
1588 nd->inode = nd->path.dentry->d_inode;
1589 if (!(flags & LOOKUP_RCU))
1591 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1593 if (!(nd->flags & LOOKUP_ROOT))
1594 nd->root.mnt = NULL;
1596 br_read_unlock(vfsmount_lock);
1600 fput_light(file, fput_needed);
1605 static inline int lookup_last(struct nameidata *nd, struct path *path)
1607 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1608 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1610 nd->flags &= ~LOOKUP_PARENT;
1611 return walk_component(nd, path, &nd->last, nd->last_type,
1612 nd->flags & LOOKUP_FOLLOW);
1615 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1616 static int path_lookupat(int dfd, const char *name,
1617 unsigned int flags, struct nameidata *nd)
1619 struct file *base = NULL;
1624 * Path walking is largely split up into 2 different synchronisation
1625 * schemes, rcu-walk and ref-walk (explained in
1626 * Documentation/filesystems/path-lookup.txt). These share much of the
1627 * path walk code, but some things particularly setup, cleanup, and
1628 * following mounts are sufficiently divergent that functions are
1629 * duplicated. Typically there is a function foo(), and its RCU
1630 * analogue, foo_rcu().
1632 * -ECHILD is the error number of choice (just to avoid clashes) that
1633 * is returned if some aspect of an rcu-walk fails. Such an error must
1634 * be handled by restarting a traditional ref-walk (which will always
1635 * be able to complete).
1637 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1642 current->total_link_count = 0;
1643 err = link_path_walk(name, nd);
1645 if (!err && !(flags & LOOKUP_PARENT)) {
1646 err = lookup_last(nd, &path);
1649 struct path link = path;
1650 nd->flags |= LOOKUP_PARENT;
1651 err = follow_link(&link, nd, &cookie);
1653 err = lookup_last(nd, &path);
1654 put_link(nd, &link, cookie);
1659 err = complete_walk(nd);
1661 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1662 if (!nd->inode->i_op->lookup) {
1663 path_put(&nd->path);
1671 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1672 path_put(&nd->root);
1673 nd->root.mnt = NULL;
1678 static int do_path_lookup(int dfd, const char *name,
1679 unsigned int flags, struct nameidata *nd)
1681 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1682 if (unlikely(retval == -ECHILD))
1683 retval = path_lookupat(dfd, name, flags, nd);
1684 if (unlikely(retval == -ESTALE))
1685 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1687 if (likely(!retval)) {
1688 if (unlikely(!audit_dummy_context())) {
1689 if (nd->path.dentry && nd->inode)
1690 audit_inode(name, nd->path.dentry);
1696 int kern_path_parent(const char *name, struct nameidata *nd)
1698 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1701 int kern_path(const char *name, unsigned int flags, struct path *path)
1703 struct nameidata nd;
1704 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1711 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1712 * @dentry: pointer to dentry of the base directory
1713 * @mnt: pointer to vfs mount of the base directory
1714 * @name: pointer to file name
1715 * @flags: lookup flags
1716 * @path: pointer to struct path to fill
1718 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1719 const char *name, unsigned int flags,
1722 struct nameidata nd;
1724 nd.root.dentry = dentry;
1726 BUG_ON(flags & LOOKUP_PARENT);
1727 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1728 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1734 static struct dentry *__lookup_hash(struct qstr *name,
1735 struct dentry *base, struct nameidata *nd)
1737 struct inode *inode = base->d_inode;
1738 struct dentry *dentry;
1741 err = inode_permission(inode, MAY_EXEC);
1743 return ERR_PTR(err);
1746 * Don't bother with __d_lookup: callers are for creat as
1747 * well as unlink, so a lot of the time it would cost
1750 dentry = d_lookup(base, name);
1752 if (dentry && d_need_lookup(dentry)) {
1754 * __lookup_hash is called with the parent dir's i_mutex already
1755 * held, so we are good to go here.
1757 dentry = d_inode_lookup(base, dentry, nd);
1762 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1763 int status = d_revalidate(dentry, nd);
1764 if (unlikely(status <= 0)) {
1766 * The dentry failed validation.
1767 * If d_revalidate returned 0 attempt to invalidate
1768 * the dentry otherwise d_revalidate is asking us
1769 * to return a fail status.
1773 return ERR_PTR(status);
1774 } else if (!d_invalidate(dentry)) {
1782 dentry = d_alloc_and_lookup(base, name, nd);
1788 * Restricted form of lookup. Doesn't follow links, single-component only,
1789 * needs parent already locked. Doesn't follow mounts.
1792 struct dentry *lookup_hash(struct nameidata *nd)
1794 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1796 EXPORT_SYMBOL(lookup_hash);
1799 * lookup_one_len - filesystem helper to lookup single pathname component
1800 * @name: pathname component to lookup
1801 * @base: base directory to lookup from
1802 * @len: maximum length @len should be interpreted to
1804 * Note that this routine is purely a helper for filesystem usage and should
1805 * not be called by generic code. Also note that by using this function the
1806 * nameidata argument is passed to the filesystem methods and a filesystem
1807 * using this helper needs to be prepared for that.
1809 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1815 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1820 return ERR_PTR(-EACCES);
1822 hash = init_name_hash();
1824 c = *(const unsigned char *)name++;
1825 if (c == '/' || c == '\0')
1826 return ERR_PTR(-EACCES);
1827 hash = partial_name_hash(c, hash);
1829 this.hash = end_name_hash(hash);
1831 * See if the low-level filesystem might want
1832 * to use its own hash..
1834 if (base->d_flags & DCACHE_OP_HASH) {
1835 int err = base->d_op->d_hash(base, base->d_inode, &this);
1837 return ERR_PTR(err);
1840 return __lookup_hash(&this, base, NULL);
1843 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1844 struct path *path, int *empty)
1846 struct nameidata nd;
1847 char *tmp = getname_flags(name, flags, empty);
1848 int err = PTR_ERR(tmp);
1851 BUG_ON(flags & LOOKUP_PARENT);
1853 err = do_path_lookup(dfd, tmp, flags, &nd);
1861 int user_path_at(int dfd, const char __user *name, unsigned flags,
1864 return user_path_at_empty(dfd, name, flags, path, 0);
1867 static int user_path_parent(int dfd, const char __user *path,
1868 struct nameidata *nd, char **name)
1870 char *s = getname(path);
1876 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1886 * It's inline, so penalty for filesystems that don't use sticky bit is
1889 static inline int check_sticky(struct inode *dir, struct inode *inode)
1891 uid_t fsuid = current_fsuid();
1893 if (!(dir->i_mode & S_ISVTX))
1895 if (current_user_ns() != inode_userns(inode))
1897 if (inode->i_uid == fsuid)
1899 if (dir->i_uid == fsuid)
1903 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1907 * Check whether we can remove a link victim from directory dir, check
1908 * whether the type of victim is right.
1909 * 1. We can't do it if dir is read-only (done in permission())
1910 * 2. We should have write and exec permissions on dir
1911 * 3. We can't remove anything from append-only dir
1912 * 4. We can't do anything with immutable dir (done in permission())
1913 * 5. If the sticky bit on dir is set we should either
1914 * a. be owner of dir, or
1915 * b. be owner of victim, or
1916 * c. have CAP_FOWNER capability
1917 * 6. If the victim is append-only or immutable we can't do antyhing with
1918 * links pointing to it.
1919 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1920 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1921 * 9. We can't remove a root or mountpoint.
1922 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1923 * nfs_async_unlink().
1925 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1929 if (!victim->d_inode)
1932 BUG_ON(victim->d_parent->d_inode != dir);
1933 audit_inode_child(victim, dir);
1935 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1940 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1941 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1944 if (!S_ISDIR(victim->d_inode->i_mode))
1946 if (IS_ROOT(victim))
1948 } else if (S_ISDIR(victim->d_inode->i_mode))
1950 if (IS_DEADDIR(dir))
1952 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1957 /* Check whether we can create an object with dentry child in directory
1959 * 1. We can't do it if child already exists (open has special treatment for
1960 * this case, but since we are inlined it's OK)
1961 * 2. We can't do it if dir is read-only (done in permission())
1962 * 3. We should have write and exec permissions on dir
1963 * 4. We can't do it if dir is immutable (done in permission())
1965 static inline int may_create(struct inode *dir, struct dentry *child)
1969 if (IS_DEADDIR(dir))
1971 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1975 * p1 and p2 should be directories on the same fs.
1977 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1982 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1986 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1988 p = d_ancestor(p2, p1);
1990 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1991 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1995 p = d_ancestor(p1, p2);
1997 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1998 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2002 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2003 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2007 void unlock_rename(struct dentry *p1, struct dentry *p2)
2009 mutex_unlock(&p1->d_inode->i_mutex);
2011 mutex_unlock(&p2->d_inode->i_mutex);
2012 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2016 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
2017 struct nameidata *nd)
2019 int error = may_create(dir, dentry);
2024 if (!dir->i_op->create)
2025 return -EACCES; /* shouldn't it be ENOSYS? */
2028 error = security_inode_create(dir, dentry, mode);
2031 error = dir->i_op->create(dir, dentry, mode, nd);
2033 fsnotify_create(dir, dentry);
2037 static int may_open(struct path *path, int acc_mode, int flag)
2039 struct dentry *dentry = path->dentry;
2040 struct inode *inode = dentry->d_inode;
2050 switch (inode->i_mode & S_IFMT) {
2054 if (acc_mode & MAY_WRITE)
2059 if (path->mnt->mnt_flags & MNT_NODEV)
2068 error = inode_permission(inode, acc_mode);
2073 * An append-only file must be opened in append mode for writing.
2075 if (IS_APPEND(inode)) {
2076 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2082 /* O_NOATIME can only be set by the owner or superuser */
2083 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2089 static int handle_truncate(struct file *filp)
2091 struct path *path = &filp->f_path;
2092 struct inode *inode = path->dentry->d_inode;
2093 int error = get_write_access(inode);
2097 * Refuse to truncate files with mandatory locks held on them.
2099 error = locks_verify_locked(inode);
2101 error = security_path_truncate(path);
2103 error = do_truncate(path->dentry, 0,
2104 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2107 put_write_access(inode);
2111 static inline int open_to_namei_flags(int flag)
2113 if ((flag & O_ACCMODE) == 3)
2119 * Handle the last step of open()
2121 static struct file *do_last(struct nameidata *nd, struct path *path,
2122 const struct open_flags *op, const char *pathname)
2124 struct dentry *dir = nd->path.dentry;
2125 struct dentry *dentry;
2126 int open_flag = op->open_flag;
2127 int will_truncate = open_flag & O_TRUNC;
2129 int acc_mode = op->acc_mode;
2133 nd->flags &= ~LOOKUP_PARENT;
2134 nd->flags |= op->intent;
2136 switch (nd->last_type) {
2139 error = handle_dots(nd, nd->last_type);
2141 return ERR_PTR(error);
2144 error = complete_walk(nd);
2146 return ERR_PTR(error);
2147 audit_inode(pathname, nd->path.dentry);
2148 if (open_flag & O_CREAT) {
2154 error = complete_walk(nd);
2156 return ERR_PTR(error);
2157 audit_inode(pathname, dir);
2161 if (!(open_flag & O_CREAT)) {
2163 if (nd->last.name[nd->last.len])
2164 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2165 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2167 /* we _can_ be in RCU mode here */
2168 error = walk_component(nd, path, &nd->last, LAST_NORM,
2171 return ERR_PTR(error);
2172 if (error) /* symlink */
2175 error = complete_walk(nd);
2177 return ERR_PTR(error);
2180 if (nd->flags & LOOKUP_DIRECTORY) {
2181 if (!nd->inode->i_op->lookup)
2184 audit_inode(pathname, nd->path.dentry);
2188 /* create side of things */
2190 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2191 * cleared when we got to the last component we are about to look up
2193 error = complete_walk(nd);
2195 return ERR_PTR(error);
2197 audit_inode(pathname, dir);
2199 /* trailing slashes? */
2200 if (nd->last.name[nd->last.len])
2203 mutex_lock(&dir->d_inode->i_mutex);
2205 dentry = lookup_hash(nd);
2206 error = PTR_ERR(dentry);
2207 if (IS_ERR(dentry)) {
2208 mutex_unlock(&dir->d_inode->i_mutex);
2212 path->dentry = dentry;
2213 path->mnt = nd->path.mnt;
2215 /* Negative dentry, just create the file */
2216 if (!dentry->d_inode) {
2217 int mode = op->mode;
2218 if (!IS_POSIXACL(dir->d_inode))
2219 mode &= ~current_umask();
2221 * This write is needed to ensure that a
2222 * rw->ro transition does not occur between
2223 * the time when the file is created and when
2224 * a permanent write count is taken through
2225 * the 'struct file' in nameidata_to_filp().
2227 error = mnt_want_write(nd->path.mnt);
2229 goto exit_mutex_unlock;
2231 /* Don't check for write permission, don't truncate */
2232 open_flag &= ~O_TRUNC;
2234 acc_mode = MAY_OPEN;
2235 error = security_path_mknod(&nd->path, dentry, mode, 0);
2237 goto exit_mutex_unlock;
2238 error = vfs_create(dir->d_inode, dentry, mode, nd);
2240 goto exit_mutex_unlock;
2241 mutex_unlock(&dir->d_inode->i_mutex);
2242 dput(nd->path.dentry);
2243 nd->path.dentry = dentry;
2248 * It already exists.
2250 mutex_unlock(&dir->d_inode->i_mutex);
2251 audit_inode(pathname, path->dentry);
2254 if (open_flag & O_EXCL)
2257 error = follow_managed(path, nd->flags);
2262 nd->flags |= LOOKUP_JUMPED;
2265 if (!path->dentry->d_inode)
2268 if (path->dentry->d_inode->i_op->follow_link)
2271 path_to_nameidata(path, nd);
2272 nd->inode = path->dentry->d_inode;
2273 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2274 error = complete_walk(nd);
2276 return ERR_PTR(error);
2278 if (S_ISDIR(nd->inode->i_mode))
2281 if (!S_ISREG(nd->inode->i_mode))
2284 if (will_truncate) {
2285 error = mnt_want_write(nd->path.mnt);
2291 error = may_open(&nd->path, acc_mode, open_flag);
2294 filp = nameidata_to_filp(nd);
2295 if (!IS_ERR(filp)) {
2296 error = ima_file_check(filp, op->acc_mode);
2299 filp = ERR_PTR(error);
2302 if (!IS_ERR(filp)) {
2303 if (will_truncate) {
2304 error = handle_truncate(filp);
2307 filp = ERR_PTR(error);
2313 mnt_drop_write(nd->path.mnt);
2314 path_put(&nd->path);
2318 mutex_unlock(&dir->d_inode->i_mutex);
2320 path_put_conditional(path, nd);
2322 filp = ERR_PTR(error);
2326 static struct file *path_openat(int dfd, const char *pathname,
2327 struct nameidata *nd, const struct open_flags *op, int flags)
2329 struct file *base = NULL;
2334 filp = get_empty_filp();
2336 return ERR_PTR(-ENFILE);
2338 filp->f_flags = op->open_flag;
2339 nd->intent.open.file = filp;
2340 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2341 nd->intent.open.create_mode = op->mode;
2343 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2344 if (unlikely(error))
2347 current->total_link_count = 0;
2348 error = link_path_walk(pathname, nd);
2349 if (unlikely(error))
2352 filp = do_last(nd, &path, op, pathname);
2353 while (unlikely(!filp)) { /* trailing symlink */
2354 struct path link = path;
2356 if (!(nd->flags & LOOKUP_FOLLOW)) {
2357 path_put_conditional(&path, nd);
2358 path_put(&nd->path);
2359 filp = ERR_PTR(-ELOOP);
2362 nd->flags |= LOOKUP_PARENT;
2363 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2364 error = follow_link(&link, nd, &cookie);
2365 if (unlikely(error))
2366 filp = ERR_PTR(error);
2368 filp = do_last(nd, &path, op, pathname);
2369 put_link(nd, &link, cookie);
2372 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2373 path_put(&nd->root);
2376 release_open_intent(nd);
2380 filp = ERR_PTR(error);
2384 struct file *do_filp_open(int dfd, const char *pathname,
2385 const struct open_flags *op, int flags)
2387 struct nameidata nd;
2390 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2391 if (unlikely(filp == ERR_PTR(-ECHILD)))
2392 filp = path_openat(dfd, pathname, &nd, op, flags);
2393 if (unlikely(filp == ERR_PTR(-ESTALE)))
2394 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2398 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2399 const char *name, const struct open_flags *op, int flags)
2401 struct nameidata nd;
2405 nd.root.dentry = dentry;
2407 flags |= LOOKUP_ROOT;
2409 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2410 return ERR_PTR(-ELOOP);
2412 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2413 if (unlikely(file == ERR_PTR(-ECHILD)))
2414 file = path_openat(-1, name, &nd, op, flags);
2415 if (unlikely(file == ERR_PTR(-ESTALE)))
2416 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2420 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2422 struct dentry *dentry = ERR_PTR(-EEXIST);
2423 struct nameidata nd;
2424 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2426 return ERR_PTR(error);
2429 * Yucky last component or no last component at all?
2430 * (foo/., foo/.., /////)
2432 if (nd.last_type != LAST_NORM)
2434 nd.flags &= ~LOOKUP_PARENT;
2435 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2436 nd.intent.open.flags = O_EXCL;
2439 * Do the final lookup.
2441 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2442 dentry = lookup_hash(&nd);
2446 if (dentry->d_inode)
2449 * Special case - lookup gave negative, but... we had foo/bar/
2450 * From the vfs_mknod() POV we just have a negative dentry -
2451 * all is fine. Let's be bastards - you had / on the end, you've
2452 * been asking for (non-existent) directory. -ENOENT for you.
2454 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2456 dentry = ERR_PTR(-ENOENT);
2463 dentry = ERR_PTR(-EEXIST);
2465 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2470 EXPORT_SYMBOL(kern_path_create);
2472 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2474 char *tmp = getname(pathname);
2477 return ERR_CAST(tmp);
2478 res = kern_path_create(dfd, tmp, path, is_dir);
2482 EXPORT_SYMBOL(user_path_create);
2484 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2486 int error = may_create(dir, dentry);
2491 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2492 !ns_capable(inode_userns(dir), CAP_MKNOD))
2495 if (!dir->i_op->mknod)
2498 error = devcgroup_inode_mknod(mode, dev);
2502 error = security_inode_mknod(dir, dentry, mode, dev);
2506 error = dir->i_op->mknod(dir, dentry, mode, dev);
2508 fsnotify_create(dir, dentry);
2512 static int may_mknod(mode_t mode)
2514 switch (mode & S_IFMT) {
2520 case 0: /* zero mode translates to S_IFREG */
2529 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2532 struct dentry *dentry;
2539 dentry = user_path_create(dfd, filename, &path, 0);
2541 return PTR_ERR(dentry);
2543 if (!IS_POSIXACL(path.dentry->d_inode))
2544 mode &= ~current_umask();
2545 error = may_mknod(mode);
2548 error = mnt_want_write(path.mnt);
2551 error = security_path_mknod(&path, dentry, mode, dev);
2553 goto out_drop_write;
2554 switch (mode & S_IFMT) {
2555 case 0: case S_IFREG:
2556 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2558 case S_IFCHR: case S_IFBLK:
2559 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2560 new_decode_dev(dev));
2562 case S_IFIFO: case S_IFSOCK:
2563 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2567 mnt_drop_write(path.mnt);
2570 mutex_unlock(&path.dentry->d_inode->i_mutex);
2576 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2578 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2581 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2583 int error = may_create(dir, dentry);
2588 if (!dir->i_op->mkdir)
2591 mode &= (S_IRWXUGO|S_ISVTX);
2592 error = security_inode_mkdir(dir, dentry, mode);
2596 error = dir->i_op->mkdir(dir, dentry, mode);
2598 fsnotify_mkdir(dir, dentry);
2602 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2604 struct dentry *dentry;
2608 dentry = user_path_create(dfd, pathname, &path, 1);
2610 return PTR_ERR(dentry);
2612 if (!IS_POSIXACL(path.dentry->d_inode))
2613 mode &= ~current_umask();
2614 error = mnt_want_write(path.mnt);
2617 error = security_path_mkdir(&path, dentry, mode);
2619 goto out_drop_write;
2620 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2622 mnt_drop_write(path.mnt);
2625 mutex_unlock(&path.dentry->d_inode->i_mutex);
2630 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2632 return sys_mkdirat(AT_FDCWD, pathname, mode);
2636 * The dentry_unhash() helper will try to drop the dentry early: we
2637 * should have a usage count of 2 if we're the only user of this
2638 * dentry, and if that is true (possibly after pruning the dcache),
2639 * then we drop the dentry now.
2641 * A low-level filesystem can, if it choses, legally
2644 * if (!d_unhashed(dentry))
2647 * if it cannot handle the case of removing a directory
2648 * that is still in use by something else..
2650 void dentry_unhash(struct dentry *dentry)
2652 shrink_dcache_parent(dentry);
2653 spin_lock(&dentry->d_lock);
2654 if (dentry->d_count == 1)
2656 spin_unlock(&dentry->d_lock);
2659 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2661 int error = may_delete(dir, dentry, 1);
2666 if (!dir->i_op->rmdir)
2670 mutex_lock(&dentry->d_inode->i_mutex);
2673 if (d_mountpoint(dentry))
2676 error = security_inode_rmdir(dir, dentry);
2680 shrink_dcache_parent(dentry);
2681 error = dir->i_op->rmdir(dir, dentry);
2685 dentry->d_inode->i_flags |= S_DEAD;
2689 mutex_unlock(&dentry->d_inode->i_mutex);
2696 static long do_rmdir(int dfd, const char __user *pathname)
2700 struct dentry *dentry;
2701 struct nameidata nd;
2703 error = user_path_parent(dfd, pathname, &nd, &name);
2707 switch(nd.last_type) {
2719 nd.flags &= ~LOOKUP_PARENT;
2721 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2722 dentry = lookup_hash(&nd);
2723 error = PTR_ERR(dentry);
2726 if (!dentry->d_inode) {
2730 error = mnt_want_write(nd.path.mnt);
2733 error = security_path_rmdir(&nd.path, dentry);
2736 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2738 mnt_drop_write(nd.path.mnt);
2742 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2749 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2751 return do_rmdir(AT_FDCWD, pathname);
2754 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2756 int error = may_delete(dir, dentry, 0);
2761 if (!dir->i_op->unlink)
2764 mutex_lock(&dentry->d_inode->i_mutex);
2765 if (d_mountpoint(dentry))
2768 error = security_inode_unlink(dir, dentry);
2770 error = dir->i_op->unlink(dir, dentry);
2775 mutex_unlock(&dentry->d_inode->i_mutex);
2777 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2778 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2779 fsnotify_link_count(dentry->d_inode);
2787 * Make sure that the actual truncation of the file will occur outside its
2788 * directory's i_mutex. Truncate can take a long time if there is a lot of
2789 * writeout happening, and we don't want to prevent access to the directory
2790 * while waiting on the I/O.
2792 static long do_unlinkat(int dfd, const char __user *pathname)
2796 struct dentry *dentry;
2797 struct nameidata nd;
2798 struct inode *inode = NULL;
2800 error = user_path_parent(dfd, pathname, &nd, &name);
2805 if (nd.last_type != LAST_NORM)
2808 nd.flags &= ~LOOKUP_PARENT;
2810 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2811 dentry = lookup_hash(&nd);
2812 error = PTR_ERR(dentry);
2813 if (!IS_ERR(dentry)) {
2814 /* Why not before? Because we want correct error value */
2815 if (nd.last.name[nd.last.len])
2817 inode = dentry->d_inode;
2821 error = mnt_want_write(nd.path.mnt);
2824 error = security_path_unlink(&nd.path, dentry);
2827 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2829 mnt_drop_write(nd.path.mnt);
2833 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2835 iput(inode); /* truncate the inode here */
2842 error = !dentry->d_inode ? -ENOENT :
2843 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2847 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2849 if ((flag & ~AT_REMOVEDIR) != 0)
2852 if (flag & AT_REMOVEDIR)
2853 return do_rmdir(dfd, pathname);
2855 return do_unlinkat(dfd, pathname);
2858 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2860 return do_unlinkat(AT_FDCWD, pathname);
2863 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2865 int error = may_create(dir, dentry);
2870 if (!dir->i_op->symlink)
2873 error = security_inode_symlink(dir, dentry, oldname);
2877 error = dir->i_op->symlink(dir, dentry, oldname);
2879 fsnotify_create(dir, dentry);
2883 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2884 int, newdfd, const char __user *, newname)
2888 struct dentry *dentry;
2891 from = getname(oldname);
2893 return PTR_ERR(from);
2895 dentry = user_path_create(newdfd, newname, &path, 0);
2896 error = PTR_ERR(dentry);
2900 error = mnt_want_write(path.mnt);
2903 error = security_path_symlink(&path, dentry, from);
2905 goto out_drop_write;
2906 error = vfs_symlink(path.dentry->d_inode, dentry, from);
2908 mnt_drop_write(path.mnt);
2911 mutex_unlock(&path.dentry->d_inode->i_mutex);
2918 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2920 return sys_symlinkat(oldname, AT_FDCWD, newname);
2923 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2925 struct inode *inode = old_dentry->d_inode;
2931 error = may_create(dir, new_dentry);
2935 if (dir->i_sb != inode->i_sb)
2939 * A link to an append-only or immutable file cannot be created.
2941 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2943 if (!dir->i_op->link)
2945 if (S_ISDIR(inode->i_mode))
2948 error = security_inode_link(old_dentry, dir, new_dentry);
2952 mutex_lock(&inode->i_mutex);
2953 /* Make sure we don't allow creating hardlink to an unlinked file */
2954 if (inode->i_nlink == 0)
2957 error = dir->i_op->link(old_dentry, dir, new_dentry);
2958 mutex_unlock(&inode->i_mutex);
2960 fsnotify_link(dir, inode, new_dentry);
2965 * Hardlinks are often used in delicate situations. We avoid
2966 * security-related surprises by not following symlinks on the
2969 * We don't follow them on the oldname either to be compatible
2970 * with linux 2.0, and to avoid hard-linking to directories
2971 * and other special files. --ADM
2973 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2974 int, newdfd, const char __user *, newname, int, flags)
2976 struct dentry *new_dentry;
2977 struct path old_path, new_path;
2981 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
2984 * To use null names we require CAP_DAC_READ_SEARCH
2985 * This ensures that not everyone will be able to create
2986 * handlink using the passed filedescriptor.
2988 if (flags & AT_EMPTY_PATH) {
2989 if (!capable(CAP_DAC_READ_SEARCH))
2994 if (flags & AT_SYMLINK_FOLLOW)
2995 how |= LOOKUP_FOLLOW;
2997 error = user_path_at(olddfd, oldname, how, &old_path);
3001 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3002 error = PTR_ERR(new_dentry);
3003 if (IS_ERR(new_dentry))
3007 if (old_path.mnt != new_path.mnt)
3009 error = mnt_want_write(new_path.mnt);
3012 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3014 goto out_drop_write;
3015 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3017 mnt_drop_write(new_path.mnt);
3020 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3021 path_put(&new_path);
3023 path_put(&old_path);
3028 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3030 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3034 * The worst of all namespace operations - renaming directory. "Perverted"
3035 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3037 * a) we can get into loop creation. Check is done in is_subdir().
3038 * b) race potential - two innocent renames can create a loop together.
3039 * That's where 4.4 screws up. Current fix: serialization on
3040 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3042 * c) we have to lock _three_ objects - parents and victim (if it exists).
3043 * And that - after we got ->i_mutex on parents (until then we don't know
3044 * whether the target exists). Solution: try to be smart with locking
3045 * order for inodes. We rely on the fact that tree topology may change
3046 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3047 * move will be locked. Thus we can rank directories by the tree
3048 * (ancestors first) and rank all non-directories after them.
3049 * That works since everybody except rename does "lock parent, lookup,
3050 * lock child" and rename is under ->s_vfs_rename_mutex.
3051 * HOWEVER, it relies on the assumption that any object with ->lookup()
3052 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3053 * we'd better make sure that there's no link(2) for them.
3054 * d) conversion from fhandle to dentry may come in the wrong moment - when
3055 * we are removing the target. Solution: we will have to grab ->i_mutex
3056 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3057 * ->i_mutex on parents, which works but leads to some truly excessive
3060 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3061 struct inode *new_dir, struct dentry *new_dentry)
3064 struct inode *target = new_dentry->d_inode;
3067 * If we are going to change the parent - check write permissions,
3068 * we'll need to flip '..'.
3070 if (new_dir != old_dir) {
3071 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3076 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3082 mutex_lock(&target->i_mutex);
3085 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3089 shrink_dcache_parent(new_dentry);
3090 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3095 target->i_flags |= S_DEAD;
3096 dont_mount(new_dentry);
3100 mutex_unlock(&target->i_mutex);
3103 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3104 d_move(old_dentry,new_dentry);
3108 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3109 struct inode *new_dir, struct dentry *new_dentry)
3111 struct inode *target = new_dentry->d_inode;
3114 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3120 mutex_lock(&target->i_mutex);
3123 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3126 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3131 dont_mount(new_dentry);
3132 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3133 d_move(old_dentry, new_dentry);
3136 mutex_unlock(&target->i_mutex);
3141 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3142 struct inode *new_dir, struct dentry *new_dentry)
3145 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3146 const unsigned char *old_name;
3148 if (old_dentry->d_inode == new_dentry->d_inode)
3151 error = may_delete(old_dir, old_dentry, is_dir);
3155 if (!new_dentry->d_inode)
3156 error = may_create(new_dir, new_dentry);
3158 error = may_delete(new_dir, new_dentry, is_dir);
3162 if (!old_dir->i_op->rename)
3165 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3168 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3170 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3172 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3173 new_dentry->d_inode, old_dentry);
3174 fsnotify_oldname_free(old_name);
3179 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3180 int, newdfd, const char __user *, newname)
3182 struct dentry *old_dir, *new_dir;
3183 struct dentry *old_dentry, *new_dentry;
3184 struct dentry *trap;
3185 struct nameidata oldnd, newnd;
3190 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3194 error = user_path_parent(newdfd, newname, &newnd, &to);
3199 if (oldnd.path.mnt != newnd.path.mnt)
3202 old_dir = oldnd.path.dentry;
3204 if (oldnd.last_type != LAST_NORM)
3207 new_dir = newnd.path.dentry;
3208 if (newnd.last_type != LAST_NORM)
3211 oldnd.flags &= ~LOOKUP_PARENT;
3212 newnd.flags &= ~LOOKUP_PARENT;
3213 newnd.flags |= LOOKUP_RENAME_TARGET;
3215 trap = lock_rename(new_dir, old_dir);
3217 old_dentry = lookup_hash(&oldnd);
3218 error = PTR_ERR(old_dentry);
3219 if (IS_ERR(old_dentry))
3221 /* source must exist */
3223 if (!old_dentry->d_inode)
3225 /* unless the source is a directory trailing slashes give -ENOTDIR */
3226 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3228 if (oldnd.last.name[oldnd.last.len])
3230 if (newnd.last.name[newnd.last.len])
3233 /* source should not be ancestor of target */
3235 if (old_dentry == trap)
3237 new_dentry = lookup_hash(&newnd);
3238 error = PTR_ERR(new_dentry);
3239 if (IS_ERR(new_dentry))
3241 /* target should not be an ancestor of source */
3243 if (new_dentry == trap)
3246 error = mnt_want_write(oldnd.path.mnt);
3249 error = security_path_rename(&oldnd.path, old_dentry,
3250 &newnd.path, new_dentry);
3253 error = vfs_rename(old_dir->d_inode, old_dentry,
3254 new_dir->d_inode, new_dentry);
3256 mnt_drop_write(oldnd.path.mnt);
3262 unlock_rename(new_dir, old_dir);
3264 path_put(&newnd.path);
3267 path_put(&oldnd.path);
3273 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3275 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3278 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3282 len = PTR_ERR(link);
3287 if (len > (unsigned) buflen)
3289 if (copy_to_user(buffer, link, len))
3296 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3297 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3298 * using) it for any given inode is up to filesystem.
3300 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3302 struct nameidata nd;
3307 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3309 return PTR_ERR(cookie);
3311 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3312 if (dentry->d_inode->i_op->put_link)
3313 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3317 int vfs_follow_link(struct nameidata *nd, const char *link)
3319 return __vfs_follow_link(nd, link);
3322 /* get the link contents into pagecache */
3323 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3327 struct address_space *mapping = dentry->d_inode->i_mapping;
3328 page = read_mapping_page(mapping, 0, NULL);
3333 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3337 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3339 struct page *page = NULL;
3340 char *s = page_getlink(dentry, &page);
3341 int res = vfs_readlink(dentry,buffer,buflen,s);
3344 page_cache_release(page);
3349 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3351 struct page *page = NULL;
3352 nd_set_link(nd, page_getlink(dentry, &page));
3356 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3358 struct page *page = cookie;
3362 page_cache_release(page);
3367 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3369 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3371 struct address_space *mapping = inode->i_mapping;
3376 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3378 flags |= AOP_FLAG_NOFS;
3381 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3382 flags, &page, &fsdata);
3386 kaddr = kmap_atomic(page, KM_USER0);
3387 memcpy(kaddr, symname, len-1);
3388 kunmap_atomic(kaddr, KM_USER0);
3390 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3397 mark_inode_dirty(inode);
3403 int page_symlink(struct inode *inode, const char *symname, int len)
3405 return __page_symlink(inode, symname, len,
3406 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3409 const struct inode_operations page_symlink_inode_operations = {
3410 .readlink = generic_readlink,
3411 .follow_link = page_follow_link_light,
3412 .put_link = page_put_link,
3415 EXPORT_SYMBOL(user_path_at);
3416 EXPORT_SYMBOL(follow_down_one);
3417 EXPORT_SYMBOL(follow_down);
3418 EXPORT_SYMBOL(follow_up);
3419 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3420 EXPORT_SYMBOL(getname);
3421 EXPORT_SYMBOL(lock_rename);
3422 EXPORT_SYMBOL(lookup_one_len);
3423 EXPORT_SYMBOL(page_follow_link_light);
3424 EXPORT_SYMBOL(page_put_link);
3425 EXPORT_SYMBOL(page_readlink);
3426 EXPORT_SYMBOL(__page_symlink);
3427 EXPORT_SYMBOL(page_symlink);
3428 EXPORT_SYMBOL(page_symlink_inode_operations);
3429 EXPORT_SYMBOL(kern_path);
3430 EXPORT_SYMBOL(vfs_path_lookup);
3431 EXPORT_SYMBOL(inode_permission);
3432 EXPORT_SYMBOL(unlock_rename);
3433 EXPORT_SYMBOL(vfs_create);
3434 EXPORT_SYMBOL(vfs_follow_link);
3435 EXPORT_SYMBOL(vfs_link);
3436 EXPORT_SYMBOL(vfs_mkdir);
3437 EXPORT_SYMBOL(vfs_mknod);
3438 EXPORT_SYMBOL(generic_permission);
3439 EXPORT_SYMBOL(vfs_readlink);
3440 EXPORT_SYMBOL(vfs_rename);
3441 EXPORT_SYMBOL(vfs_rmdir);
3442 EXPORT_SYMBOL(vfs_symlink);
3443 EXPORT_SYMBOL(vfs_unlink);
3444 EXPORT_SYMBOL(dentry_unhash);
3445 EXPORT_SYMBOL(generic_readlink);