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)
144 result = ERR_PTR(-ENOMEM);
147 int retval = do_getname(filename, tmp);
151 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
153 result = ERR_PTR(retval);
157 audit_getname(result);
161 char *getname(const char __user * filename)
163 return getname_flags(filename, 0);
166 #ifdef CONFIG_AUDITSYSCALL
167 void putname(const char *name)
169 if (unlikely(!audit_dummy_context()))
174 EXPORT_SYMBOL(putname);
177 static int check_acl(struct inode *inode, int mask)
179 #ifdef CONFIG_FS_POSIX_ACL
180 struct posix_acl *acl;
183 * Under RCU walk, we cannot even do a "get_cached_acl()",
184 * because that involves locking and getting a refcount on
187 * So the only case we handle during RCU walking is the
188 * case of a cached "no ACL at all", which needs no locks
191 if (mask & MAY_NOT_BLOCK) {
192 if (negative_cached_acl(inode, ACL_TYPE_ACCESS))
197 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
200 * A filesystem can force a ACL callback by just never filling the
201 * ACL cache. But normally you'd fill the cache either at inode
202 * instantiation time, or on the first ->get_acl call.
204 * If the filesystem doesn't have a get_acl() function at all, we'll
205 * just create the negative cache entry.
207 if (acl == ACL_NOT_CACHED) {
208 if (inode->i_op->get_acl) {
209 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
213 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
219 int error = posix_acl_permission(inode, acl, mask);
220 posix_acl_release(acl);
229 * This does basic POSIX ACL permission checking
231 static int acl_permission_check(struct inode *inode, int mask)
233 unsigned int mode = inode->i_mode;
235 mask &= MAY_READ | MAY_WRITE | MAY_EXEC | MAY_NOT_BLOCK;
237 if (current_user_ns() != inode_userns(inode))
240 if (likely(current_fsuid() == inode->i_uid))
243 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
244 int error = check_acl(inode, mask);
245 if (error != -EAGAIN)
249 if (in_group_p(inode->i_gid))
255 * If the DACs are ok we don't need any capability check.
257 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
263 * generic_permission - check for access rights on a Posix-like filesystem
264 * @inode: inode to check access rights for
265 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
267 * Used to check for read/write/execute permissions on a file.
268 * We use "fsuid" for this, letting us set arbitrary permissions
269 * for filesystem access without changing the "normal" uids which
270 * are used for other things.
272 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
273 * request cannot be satisfied (eg. requires blocking or too much complexity).
274 * It would then be called again in ref-walk mode.
276 int generic_permission(struct inode *inode, int mask)
281 * Do the basic POSIX ACL permission checks.
283 ret = acl_permission_check(inode, mask);
287 if (S_ISDIR(inode->i_mode)) {
288 /* DACs are overridable for directories */
289 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
291 if (!(mask & MAY_WRITE))
292 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
297 * Read/write DACs are always overridable.
298 * Executable DACs are overridable when there is
299 * at least one exec bit set.
301 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
302 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
306 * Searching includes executable on directories, else just read.
308 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
309 if (mask == MAY_READ)
310 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
317 * inode_permission - check for access rights to a given inode
318 * @inode: inode to check permission on
319 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
321 * Used to check for read/write/execute permissions on an inode.
322 * We use "fsuid" for this, letting us set arbitrary permissions
323 * for filesystem access without changing the "normal" uids which
324 * are used for other things.
326 int inode_permission(struct inode *inode, int mask)
330 if (mask & MAY_WRITE) {
331 umode_t mode = inode->i_mode;
334 * Nobody gets write access to a read-only fs.
336 if (IS_RDONLY(inode) &&
337 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
341 * Nobody gets write access to an immutable file.
343 if (IS_IMMUTABLE(inode))
347 if (inode->i_op->permission)
348 retval = inode->i_op->permission(inode, mask);
350 retval = generic_permission(inode, mask);
355 retval = devcgroup_inode_permission(inode, mask);
359 return security_inode_permission(inode, mask);
363 * path_get - get a reference to a path
364 * @path: path to get the reference to
366 * Given a path increment the reference count to the dentry and the vfsmount.
368 void path_get(struct path *path)
373 EXPORT_SYMBOL(path_get);
376 * path_put - put a reference to a path
377 * @path: path to put the reference to
379 * Given a path decrement the reference count to the dentry and the vfsmount.
381 void path_put(struct path *path)
386 EXPORT_SYMBOL(path_put);
389 * Path walking has 2 modes, rcu-walk and ref-walk (see
390 * Documentation/filesystems/path-lookup.txt). In situations when we can't
391 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
392 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
393 * mode. Refcounts are grabbed at the last known good point before rcu-walk
394 * got stuck, so ref-walk may continue from there. If this is not successful
395 * (eg. a seqcount has changed), then failure is returned and it's up to caller
396 * to restart the path walk from the beginning in ref-walk mode.
400 * unlazy_walk - try to switch to ref-walk mode.
401 * @nd: nameidata pathwalk data
402 * @dentry: child of nd->path.dentry or NULL
403 * Returns: 0 on success, -ECHILD on failure
405 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
406 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
407 * @nd or NULL. Must be called from rcu-walk context.
409 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
411 struct fs_struct *fs = current->fs;
412 struct dentry *parent = nd->path.dentry;
415 BUG_ON(!(nd->flags & LOOKUP_RCU));
416 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
418 spin_lock(&fs->lock);
419 if (nd->root.mnt != fs->root.mnt ||
420 nd->root.dentry != fs->root.dentry)
423 spin_lock(&parent->d_lock);
425 if (!__d_rcu_to_refcount(parent, nd->seq))
427 BUG_ON(nd->inode != parent->d_inode);
429 if (dentry->d_parent != parent)
431 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
432 if (!__d_rcu_to_refcount(dentry, nd->seq))
435 * If the sequence check on the child dentry passed, then
436 * the child has not been removed from its parent. This
437 * means the parent dentry must be valid and able to take
438 * a reference at this point.
440 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
441 BUG_ON(!parent->d_count);
443 spin_unlock(&dentry->d_lock);
445 spin_unlock(&parent->d_lock);
448 spin_unlock(&fs->lock);
450 mntget(nd->path.mnt);
453 br_read_unlock(vfsmount_lock);
454 nd->flags &= ~LOOKUP_RCU;
458 spin_unlock(&dentry->d_lock);
460 spin_unlock(&parent->d_lock);
463 spin_unlock(&fs->lock);
468 * release_open_intent - free up open intent resources
469 * @nd: pointer to nameidata
471 void release_open_intent(struct nameidata *nd)
473 struct file *file = nd->intent.open.file;
475 if (file && !IS_ERR(file)) {
476 if (file->f_path.dentry == NULL)
483 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
485 return dentry->d_op->d_revalidate(dentry, nd);
489 * complete_walk - successful completion of path walk
490 * @nd: pointer nameidata
492 * If we had been in RCU mode, drop out of it and legitimize nd->path.
493 * Revalidate the final result, unless we'd already done that during
494 * the path walk or the filesystem doesn't ask for it. Return 0 on
495 * success, -error on failure. In case of failure caller does not
496 * need to drop nd->path.
498 static int complete_walk(struct nameidata *nd)
500 struct dentry *dentry = nd->path.dentry;
503 if (nd->flags & LOOKUP_RCU) {
504 nd->flags &= ~LOOKUP_RCU;
505 if (!(nd->flags & LOOKUP_ROOT))
507 spin_lock(&dentry->d_lock);
508 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
509 spin_unlock(&dentry->d_lock);
511 br_read_unlock(vfsmount_lock);
514 BUG_ON(nd->inode != dentry->d_inode);
515 spin_unlock(&dentry->d_lock);
516 mntget(nd->path.mnt);
518 br_read_unlock(vfsmount_lock);
521 if (likely(!(nd->flags & LOOKUP_JUMPED)))
524 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
527 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
530 /* Note: we do not d_invalidate() */
531 status = d_revalidate(dentry, nd);
542 static __always_inline void set_root(struct nameidata *nd)
545 get_fs_root(current->fs, &nd->root);
548 static int link_path_walk(const char *, struct nameidata *);
550 static __always_inline void set_root_rcu(struct nameidata *nd)
553 struct fs_struct *fs = current->fs;
557 seq = read_seqcount_begin(&fs->seq);
559 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
560 } while (read_seqcount_retry(&fs->seq, seq));
564 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
576 nd->flags |= LOOKUP_JUMPED;
578 nd->inode = nd->path.dentry->d_inode;
580 ret = link_path_walk(link, nd);
584 return PTR_ERR(link);
587 static void path_put_conditional(struct path *path, struct nameidata *nd)
590 if (path->mnt != nd->path.mnt)
594 static inline void path_to_nameidata(const struct path *path,
595 struct nameidata *nd)
597 if (!(nd->flags & LOOKUP_RCU)) {
598 dput(nd->path.dentry);
599 if (nd->path.mnt != path->mnt)
600 mntput(nd->path.mnt);
602 nd->path.mnt = path->mnt;
603 nd->path.dentry = path->dentry;
606 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
608 struct inode *inode = link->dentry->d_inode;
609 if (!IS_ERR(cookie) && inode->i_op->put_link)
610 inode->i_op->put_link(link->dentry, nd, cookie);
614 static __always_inline int
615 follow_link(struct path *link, struct nameidata *nd, void **p)
618 struct dentry *dentry = link->dentry;
620 BUG_ON(nd->flags & LOOKUP_RCU);
622 if (link->mnt == nd->path.mnt)
625 if (unlikely(current->total_link_count >= 40)) {
626 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
631 current->total_link_count++;
633 touch_atime(link->mnt, dentry);
634 nd_set_link(nd, NULL);
636 error = security_inode_follow_link(link->dentry, nd);
638 *p = ERR_PTR(error); /* no ->put_link(), please */
643 nd->last_type = LAST_BIND;
644 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
647 char *s = nd_get_link(nd);
650 error = __vfs_follow_link(nd, s);
651 else if (nd->last_type == LAST_BIND) {
652 nd->flags |= LOOKUP_JUMPED;
653 nd->inode = nd->path.dentry->d_inode;
654 if (nd->inode->i_op->follow_link) {
655 /* stepped on a _really_ weird one */
664 static int follow_up_rcu(struct path *path)
666 struct vfsmount *parent;
667 struct dentry *mountpoint;
669 parent = path->mnt->mnt_parent;
670 if (parent == path->mnt)
672 mountpoint = path->mnt->mnt_mountpoint;
673 path->dentry = mountpoint;
678 int follow_up(struct path *path)
680 struct vfsmount *parent;
681 struct dentry *mountpoint;
683 br_read_lock(vfsmount_lock);
684 parent = path->mnt->mnt_parent;
685 if (parent == path->mnt) {
686 br_read_unlock(vfsmount_lock);
690 mountpoint = dget(path->mnt->mnt_mountpoint);
691 br_read_unlock(vfsmount_lock);
693 path->dentry = mountpoint;
700 * Perform an automount
701 * - return -EISDIR to tell follow_managed() to stop and return the path we
704 static int follow_automount(struct path *path, unsigned flags,
707 struct vfsmount *mnt;
710 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
713 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
714 * and this is the terminal part of the path.
716 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_PARENT))
717 return -EISDIR; /* we actually want to stop here */
720 * We don't want to mount if someone's just doing a stat and they've
721 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
722 * appended a '/' to the name.
724 if (!(flags & LOOKUP_FOLLOW)) {
725 /* We do, however, want to mount if someone wants to open or
726 * create a file of any type under the mountpoint, wants to
727 * traverse through the mountpoint or wants to open the mounted
729 * Also, autofs may mark negative dentries as being automount
730 * points. These will need the attentions of the daemon to
731 * instantiate them before they can be used.
733 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
734 LOOKUP_OPEN | LOOKUP_CREATE)) &&
735 path->dentry->d_inode)
738 current->total_link_count++;
739 if (current->total_link_count >= 40)
742 mnt = path->dentry->d_op->d_automount(path);
745 * The filesystem is allowed to return -EISDIR here to indicate
746 * it doesn't want to automount. For instance, autofs would do
747 * this so that its userspace daemon can mount on this dentry.
749 * However, we can only permit this if it's a terminal point in
750 * the path being looked up; if it wasn't then the remainder of
751 * the path is inaccessible and we should say so.
753 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
758 if (!mnt) /* mount collision */
762 /* lock_mount() may release path->mnt on error */
766 err = finish_automount(mnt, path);
770 /* Someone else made a mount here whilst we were busy */
775 path->dentry = dget(mnt->mnt_root);
784 * Handle a dentry that is managed in some way.
785 * - Flagged for transit management (autofs)
786 * - Flagged as mountpoint
787 * - Flagged as automount point
789 * This may only be called in refwalk mode.
791 * Serialization is taken care of in namespace.c
793 static int follow_managed(struct path *path, unsigned flags)
795 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
797 bool need_mntput = false;
800 /* Given that we're not holding a lock here, we retain the value in a
801 * local variable for each dentry as we look at it so that we don't see
802 * the components of that value change under us */
803 while (managed = ACCESS_ONCE(path->dentry->d_flags),
804 managed &= DCACHE_MANAGED_DENTRY,
805 unlikely(managed != 0)) {
806 /* Allow the filesystem to manage the transit without i_mutex
808 if (managed & DCACHE_MANAGE_TRANSIT) {
809 BUG_ON(!path->dentry->d_op);
810 BUG_ON(!path->dentry->d_op->d_manage);
811 ret = path->dentry->d_op->d_manage(path->dentry, false);
816 /* Transit to a mounted filesystem. */
817 if (managed & DCACHE_MOUNTED) {
818 struct vfsmount *mounted = lookup_mnt(path);
824 path->dentry = dget(mounted->mnt_root);
829 /* Something is mounted on this dentry in another
830 * namespace and/or whatever was mounted there in this
831 * namespace got unmounted before we managed to get the
835 /* Handle an automount point */
836 if (managed & DCACHE_NEED_AUTOMOUNT) {
837 ret = follow_automount(path, flags, &need_mntput);
843 /* We didn't change the current path point */
847 if (need_mntput && path->mnt == mnt)
854 int follow_down_one(struct path *path)
856 struct vfsmount *mounted;
858 mounted = lookup_mnt(path);
863 path->dentry = dget(mounted->mnt_root);
869 static inline bool managed_dentry_might_block(struct dentry *dentry)
871 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
872 dentry->d_op->d_manage(dentry, true) < 0);
876 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
877 * we meet a managed dentry that would need blocking.
879 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
880 struct inode **inode)
883 struct vfsmount *mounted;
885 * Don't forget we might have a non-mountpoint managed dentry
886 * that wants to block transit.
888 if (unlikely(managed_dentry_might_block(path->dentry)))
891 if (!d_mountpoint(path->dentry))
894 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
898 path->dentry = mounted->mnt_root;
899 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
901 * Update the inode too. We don't need to re-check the
902 * dentry sequence number here after this d_inode read,
903 * because a mount-point is always pinned.
905 *inode = path->dentry->d_inode;
910 static void follow_mount_rcu(struct nameidata *nd)
912 while (d_mountpoint(nd->path.dentry)) {
913 struct vfsmount *mounted;
914 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
917 nd->path.mnt = mounted;
918 nd->path.dentry = mounted->mnt_root;
919 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
923 static int follow_dotdot_rcu(struct nameidata *nd)
928 if (nd->path.dentry == nd->root.dentry &&
929 nd->path.mnt == nd->root.mnt) {
932 if (nd->path.dentry != nd->path.mnt->mnt_root) {
933 struct dentry *old = nd->path.dentry;
934 struct dentry *parent = old->d_parent;
937 seq = read_seqcount_begin(&parent->d_seq);
938 if (read_seqcount_retry(&old->d_seq, nd->seq))
940 nd->path.dentry = parent;
944 if (!follow_up_rcu(&nd->path))
946 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
948 follow_mount_rcu(nd);
949 nd->inode = nd->path.dentry->d_inode;
953 nd->flags &= ~LOOKUP_RCU;
954 if (!(nd->flags & LOOKUP_ROOT))
957 br_read_unlock(vfsmount_lock);
962 * Follow down to the covering mount currently visible to userspace. At each
963 * point, the filesystem owning that dentry may be queried as to whether the
964 * caller is permitted to proceed or not.
966 int follow_down(struct path *path)
971 while (managed = ACCESS_ONCE(path->dentry->d_flags),
972 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
973 /* Allow the filesystem to manage the transit without i_mutex
976 * We indicate to the filesystem if someone is trying to mount
977 * something here. This gives autofs the chance to deny anyone
978 * other than its daemon the right to mount on its
981 * The filesystem may sleep at this point.
983 if (managed & DCACHE_MANAGE_TRANSIT) {
984 BUG_ON(!path->dentry->d_op);
985 BUG_ON(!path->dentry->d_op->d_manage);
986 ret = path->dentry->d_op->d_manage(
987 path->dentry, false);
989 return ret == -EISDIR ? 0 : ret;
992 /* Transit to a mounted filesystem. */
993 if (managed & DCACHE_MOUNTED) {
994 struct vfsmount *mounted = lookup_mnt(path);
1000 path->dentry = dget(mounted->mnt_root);
1004 /* Don't handle automount points here */
1011 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1013 static void follow_mount(struct path *path)
1015 while (d_mountpoint(path->dentry)) {
1016 struct vfsmount *mounted = lookup_mnt(path);
1021 path->mnt = mounted;
1022 path->dentry = dget(mounted->mnt_root);
1026 static void follow_dotdot(struct nameidata *nd)
1031 struct dentry *old = nd->path.dentry;
1033 if (nd->path.dentry == nd->root.dentry &&
1034 nd->path.mnt == nd->root.mnt) {
1037 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1038 /* rare case of legitimate dget_parent()... */
1039 nd->path.dentry = dget_parent(nd->path.dentry);
1043 if (!follow_up(&nd->path))
1046 follow_mount(&nd->path);
1047 nd->inode = nd->path.dentry->d_inode;
1051 * Allocate a dentry with name and parent, and perform a parent
1052 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1053 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1054 * have verified that no child exists while under i_mutex.
1056 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1057 struct qstr *name, struct nameidata *nd)
1059 struct inode *inode = parent->d_inode;
1060 struct dentry *dentry;
1063 /* Don't create child dentry for a dead directory. */
1064 if (unlikely(IS_DEADDIR(inode)))
1065 return ERR_PTR(-ENOENT);
1067 dentry = d_alloc(parent, name);
1068 if (unlikely(!dentry))
1069 return ERR_PTR(-ENOMEM);
1071 old = inode->i_op->lookup(inode, dentry, nd);
1072 if (unlikely(old)) {
1080 * We already have a dentry, but require a lookup to be performed on the parent
1081 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1082 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1083 * child exists while under i_mutex.
1085 static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1086 struct nameidata *nd)
1088 struct inode *inode = parent->d_inode;
1091 /* Don't create child dentry for a dead directory. */
1092 if (unlikely(IS_DEADDIR(inode)))
1093 return ERR_PTR(-ENOENT);
1095 old = inode->i_op->lookup(inode, dentry, nd);
1096 if (unlikely(old)) {
1104 * It's more convoluted than I'd like it to be, but... it's still fairly
1105 * small and for now I'd prefer to have fast path as straight as possible.
1106 * It _is_ time-critical.
1108 static int do_lookup(struct nameidata *nd, struct qstr *name,
1109 struct path *path, struct inode **inode)
1111 struct vfsmount *mnt = nd->path.mnt;
1112 struct dentry *dentry, *parent = nd->path.dentry;
1118 * Rename seqlock is not required here because in the off chance
1119 * of a false negative due to a concurrent rename, we're going to
1120 * do the non-racy lookup, below.
1122 if (nd->flags & LOOKUP_RCU) {
1125 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1129 /* Memory barrier in read_seqcount_begin of child is enough */
1130 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1134 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1135 status = d_revalidate(dentry, nd);
1136 if (unlikely(status <= 0)) {
1137 if (status != -ECHILD)
1142 if (unlikely(d_need_lookup(dentry)))
1145 path->dentry = dentry;
1146 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1148 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1152 if (unlazy_walk(nd, dentry))
1155 dentry = __d_lookup(parent, name);
1158 if (dentry && unlikely(d_need_lookup(dentry))) {
1163 if (unlikely(!dentry)) {
1164 struct inode *dir = parent->d_inode;
1165 BUG_ON(nd->inode != dir);
1167 mutex_lock(&dir->i_mutex);
1168 dentry = d_lookup(parent, name);
1169 if (likely(!dentry)) {
1170 dentry = d_alloc_and_lookup(parent, name, nd);
1171 if (IS_ERR(dentry)) {
1172 mutex_unlock(&dir->i_mutex);
1173 return PTR_ERR(dentry);
1178 } else if (unlikely(d_need_lookup(dentry))) {
1179 dentry = d_inode_lookup(parent, dentry, nd);
1180 if (IS_ERR(dentry)) {
1181 mutex_unlock(&dir->i_mutex);
1182 return PTR_ERR(dentry);
1188 mutex_unlock(&dir->i_mutex);
1190 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1191 status = d_revalidate(dentry, nd);
1192 if (unlikely(status <= 0)) {
1197 if (!d_invalidate(dentry)) {
1206 path->dentry = dentry;
1207 err = follow_managed(path, nd->flags);
1208 if (unlikely(err < 0)) {
1209 path_put_conditional(path, nd);
1212 *inode = path->dentry->d_inode;
1216 static inline int may_lookup(struct nameidata *nd)
1218 if (nd->flags & LOOKUP_RCU) {
1219 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1222 if (unlazy_walk(nd, NULL))
1225 return inode_permission(nd->inode, MAY_EXEC);
1228 static inline int handle_dots(struct nameidata *nd, int type)
1230 if (type == LAST_DOTDOT) {
1231 if (nd->flags & LOOKUP_RCU) {
1232 if (follow_dotdot_rcu(nd))
1240 static void terminate_walk(struct nameidata *nd)
1242 if (!(nd->flags & LOOKUP_RCU)) {
1243 path_put(&nd->path);
1245 nd->flags &= ~LOOKUP_RCU;
1246 if (!(nd->flags & LOOKUP_ROOT))
1247 nd->root.mnt = NULL;
1249 br_read_unlock(vfsmount_lock);
1253 static inline int walk_component(struct nameidata *nd, struct path *path,
1254 struct qstr *name, int type, int follow)
1256 struct inode *inode;
1259 * "." and ".." are special - ".." especially so because it has
1260 * to be able to know about the current root directory and
1261 * parent relationships.
1263 if (unlikely(type != LAST_NORM))
1264 return handle_dots(nd, type);
1265 err = do_lookup(nd, name, path, &inode);
1266 if (unlikely(err)) {
1271 path_to_nameidata(path, nd);
1275 if (unlikely(inode->i_op->follow_link) && follow) {
1276 if (nd->flags & LOOKUP_RCU) {
1277 if (unlikely(unlazy_walk(nd, path->dentry))) {
1282 BUG_ON(inode != path->dentry->d_inode);
1285 path_to_nameidata(path, nd);
1291 * This limits recursive symlink follows to 8, while
1292 * limiting consecutive symlinks to 40.
1294 * Without that kind of total limit, nasty chains of consecutive
1295 * symlinks can cause almost arbitrarily long lookups.
1297 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1301 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1302 path_put_conditional(path, nd);
1303 path_put(&nd->path);
1306 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1309 current->link_count++;
1312 struct path link = *path;
1315 res = follow_link(&link, nd, &cookie);
1317 res = walk_component(nd, path, &nd->last,
1318 nd->last_type, LOOKUP_FOLLOW);
1319 put_link(nd, &link, cookie);
1322 current->link_count--;
1329 * This is the basic name resolution function, turning a pathname into
1330 * the final dentry. We expect 'base' to be positive and a directory.
1332 * Returns 0 and nd will have valid dentry and mnt on success.
1333 * Returns error and drops reference to input namei data on failure.
1335 static int link_path_walk(const char *name, struct nameidata *nd)
1345 /* At this point we know we have a real path component. */
1352 err = may_lookup(nd);
1357 c = *(const unsigned char *)name;
1359 hash = init_name_hash();
1362 hash = partial_name_hash(c, hash);
1363 c = *(const unsigned char *)name;
1364 } while (c && (c != '/'));
1365 this.len = name - (const char *) this.name;
1366 this.hash = end_name_hash(hash);
1369 if (this.name[0] == '.') switch (this.len) {
1371 if (this.name[1] == '.') {
1373 nd->flags |= LOOKUP_JUMPED;
1379 if (likely(type == LAST_NORM)) {
1380 struct dentry *parent = nd->path.dentry;
1381 nd->flags &= ~LOOKUP_JUMPED;
1382 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1383 err = parent->d_op->d_hash(parent, nd->inode,
1390 /* remove trailing slashes? */
1392 goto last_component;
1393 while (*++name == '/');
1395 goto last_component;
1397 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1402 err = nested_symlink(&next, nd);
1407 if (!nd->inode->i_op->lookup)
1410 /* here ends the main loop */
1414 nd->last_type = type;
1421 static int path_init(int dfd, const char *name, unsigned int flags,
1422 struct nameidata *nd, struct file **fp)
1428 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1429 nd->flags = flags | LOOKUP_JUMPED;
1431 if (flags & LOOKUP_ROOT) {
1432 struct inode *inode = nd->root.dentry->d_inode;
1434 if (!inode->i_op->lookup)
1436 retval = inode_permission(inode, MAY_EXEC);
1440 nd->path = nd->root;
1442 if (flags & LOOKUP_RCU) {
1443 br_read_lock(vfsmount_lock);
1445 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1447 path_get(&nd->path);
1452 nd->root.mnt = NULL;
1455 if (flags & LOOKUP_RCU) {
1456 br_read_lock(vfsmount_lock);
1461 path_get(&nd->root);
1463 nd->path = nd->root;
1464 } else if (dfd == AT_FDCWD) {
1465 if (flags & LOOKUP_RCU) {
1466 struct fs_struct *fs = current->fs;
1469 br_read_lock(vfsmount_lock);
1473 seq = read_seqcount_begin(&fs->seq);
1475 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1476 } while (read_seqcount_retry(&fs->seq, seq));
1478 get_fs_pwd(current->fs, &nd->path);
1481 struct dentry *dentry;
1483 file = fget_raw_light(dfd, &fput_needed);
1488 dentry = file->f_path.dentry;
1492 if (!S_ISDIR(dentry->d_inode->i_mode))
1495 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1500 nd->path = file->f_path;
1501 if (flags & LOOKUP_RCU) {
1504 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1505 br_read_lock(vfsmount_lock);
1508 path_get(&file->f_path);
1509 fput_light(file, fput_needed);
1513 nd->inode = nd->path.dentry->d_inode;
1517 fput_light(file, fput_needed);
1522 static inline int lookup_last(struct nameidata *nd, struct path *path)
1524 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1525 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1527 nd->flags &= ~LOOKUP_PARENT;
1528 return walk_component(nd, path, &nd->last, nd->last_type,
1529 nd->flags & LOOKUP_FOLLOW);
1532 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1533 static int path_lookupat(int dfd, const char *name,
1534 unsigned int flags, struct nameidata *nd)
1536 struct file *base = NULL;
1541 * Path walking is largely split up into 2 different synchronisation
1542 * schemes, rcu-walk and ref-walk (explained in
1543 * Documentation/filesystems/path-lookup.txt). These share much of the
1544 * path walk code, but some things particularly setup, cleanup, and
1545 * following mounts are sufficiently divergent that functions are
1546 * duplicated. Typically there is a function foo(), and its RCU
1547 * analogue, foo_rcu().
1549 * -ECHILD is the error number of choice (just to avoid clashes) that
1550 * is returned if some aspect of an rcu-walk fails. Such an error must
1551 * be handled by restarting a traditional ref-walk (which will always
1552 * be able to complete).
1554 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1559 current->total_link_count = 0;
1560 err = link_path_walk(name, nd);
1562 if (!err && !(flags & LOOKUP_PARENT)) {
1563 err = lookup_last(nd, &path);
1566 struct path link = path;
1567 nd->flags |= LOOKUP_PARENT;
1568 err = follow_link(&link, nd, &cookie);
1570 err = lookup_last(nd, &path);
1571 put_link(nd, &link, cookie);
1576 err = complete_walk(nd);
1578 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1579 if (!nd->inode->i_op->lookup) {
1580 path_put(&nd->path);
1588 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1589 path_put(&nd->root);
1590 nd->root.mnt = NULL;
1595 static int do_path_lookup(int dfd, const char *name,
1596 unsigned int flags, struct nameidata *nd)
1598 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1599 if (unlikely(retval == -ECHILD))
1600 retval = path_lookupat(dfd, name, flags, nd);
1601 if (unlikely(retval == -ESTALE))
1602 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1604 if (likely(!retval)) {
1605 if (unlikely(!audit_dummy_context())) {
1606 if (nd->path.dentry && nd->inode)
1607 audit_inode(name, nd->path.dentry);
1613 int kern_path_parent(const char *name, struct nameidata *nd)
1615 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1618 int kern_path(const char *name, unsigned int flags, struct path *path)
1620 struct nameidata nd;
1621 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1628 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1629 * @dentry: pointer to dentry of the base directory
1630 * @mnt: pointer to vfs mount of the base directory
1631 * @name: pointer to file name
1632 * @flags: lookup flags
1633 * @path: pointer to struct path to fill
1635 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1636 const char *name, unsigned int flags,
1639 struct nameidata nd;
1641 nd.root.dentry = dentry;
1643 BUG_ON(flags & LOOKUP_PARENT);
1644 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1645 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1651 static struct dentry *__lookup_hash(struct qstr *name,
1652 struct dentry *base, struct nameidata *nd)
1654 struct inode *inode = base->d_inode;
1655 struct dentry *dentry;
1658 err = inode_permission(inode, MAY_EXEC);
1660 return ERR_PTR(err);
1663 * Don't bother with __d_lookup: callers are for creat as
1664 * well as unlink, so a lot of the time it would cost
1667 dentry = d_lookup(base, name);
1669 if (dentry && d_need_lookup(dentry)) {
1671 * __lookup_hash is called with the parent dir's i_mutex already
1672 * held, so we are good to go here.
1674 dentry = d_inode_lookup(base, dentry, nd);
1679 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1680 int status = d_revalidate(dentry, nd);
1681 if (unlikely(status <= 0)) {
1683 * The dentry failed validation.
1684 * If d_revalidate returned 0 attempt to invalidate
1685 * the dentry otherwise d_revalidate is asking us
1686 * to return a fail status.
1690 return ERR_PTR(status);
1691 } else if (!d_invalidate(dentry)) {
1699 dentry = d_alloc_and_lookup(base, name, nd);
1705 * Restricted form of lookup. Doesn't follow links, single-component only,
1706 * needs parent already locked. Doesn't follow mounts.
1709 static struct dentry *lookup_hash(struct nameidata *nd)
1711 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1715 * lookup_one_len - filesystem helper to lookup single pathname component
1716 * @name: pathname component to lookup
1717 * @base: base directory to lookup from
1718 * @len: maximum length @len should be interpreted to
1720 * Note that this routine is purely a helper for filesystem usage and should
1721 * not be called by generic code. Also note that by using this function the
1722 * nameidata argument is passed to the filesystem methods and a filesystem
1723 * using this helper needs to be prepared for that.
1725 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1731 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1736 return ERR_PTR(-EACCES);
1738 hash = init_name_hash();
1740 c = *(const unsigned char *)name++;
1741 if (c == '/' || c == '\0')
1742 return ERR_PTR(-EACCES);
1743 hash = partial_name_hash(c, hash);
1745 this.hash = end_name_hash(hash);
1747 * See if the low-level filesystem might want
1748 * to use its own hash..
1750 if (base->d_flags & DCACHE_OP_HASH) {
1751 int err = base->d_op->d_hash(base, base->d_inode, &this);
1753 return ERR_PTR(err);
1756 return __lookup_hash(&this, base, NULL);
1759 int user_path_at(int dfd, const char __user *name, unsigned flags,
1762 struct nameidata nd;
1763 char *tmp = getname_flags(name, flags);
1764 int err = PTR_ERR(tmp);
1767 BUG_ON(flags & LOOKUP_PARENT);
1769 err = do_path_lookup(dfd, tmp, flags, &nd);
1777 static int user_path_parent(int dfd, const char __user *path,
1778 struct nameidata *nd, char **name)
1780 char *s = getname(path);
1786 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1796 * It's inline, so penalty for filesystems that don't use sticky bit is
1799 static inline int check_sticky(struct inode *dir, struct inode *inode)
1801 uid_t fsuid = current_fsuid();
1803 if (!(dir->i_mode & S_ISVTX))
1805 if (current_user_ns() != inode_userns(inode))
1807 if (inode->i_uid == fsuid)
1809 if (dir->i_uid == fsuid)
1813 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1817 * Check whether we can remove a link victim from directory dir, check
1818 * whether the type of victim is right.
1819 * 1. We can't do it if dir is read-only (done in permission())
1820 * 2. We should have write and exec permissions on dir
1821 * 3. We can't remove anything from append-only dir
1822 * 4. We can't do anything with immutable dir (done in permission())
1823 * 5. If the sticky bit on dir is set we should either
1824 * a. be owner of dir, or
1825 * b. be owner of victim, or
1826 * c. have CAP_FOWNER capability
1827 * 6. If the victim is append-only or immutable we can't do antyhing with
1828 * links pointing to it.
1829 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1830 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1831 * 9. We can't remove a root or mountpoint.
1832 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1833 * nfs_async_unlink().
1835 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1839 if (!victim->d_inode)
1842 BUG_ON(victim->d_parent->d_inode != dir);
1843 audit_inode_child(victim, dir);
1845 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1850 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1851 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1854 if (!S_ISDIR(victim->d_inode->i_mode))
1856 if (IS_ROOT(victim))
1858 } else if (S_ISDIR(victim->d_inode->i_mode))
1860 if (IS_DEADDIR(dir))
1862 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1867 /* Check whether we can create an object with dentry child in directory
1869 * 1. We can't do it if child already exists (open has special treatment for
1870 * this case, but since we are inlined it's OK)
1871 * 2. We can't do it if dir is read-only (done in permission())
1872 * 3. We should have write and exec permissions on dir
1873 * 4. We can't do it if dir is immutable (done in permission())
1875 static inline int may_create(struct inode *dir, struct dentry *child)
1879 if (IS_DEADDIR(dir))
1881 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1885 * p1 and p2 should be directories on the same fs.
1887 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1892 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1896 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1898 p = d_ancestor(p2, p1);
1900 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1901 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1905 p = d_ancestor(p1, p2);
1907 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1908 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1912 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1913 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1917 void unlock_rename(struct dentry *p1, struct dentry *p2)
1919 mutex_unlock(&p1->d_inode->i_mutex);
1921 mutex_unlock(&p2->d_inode->i_mutex);
1922 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1926 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1927 struct nameidata *nd)
1929 int error = may_create(dir, dentry);
1934 if (!dir->i_op->create)
1935 return -EACCES; /* shouldn't it be ENOSYS? */
1938 error = security_inode_create(dir, dentry, mode);
1941 error = dir->i_op->create(dir, dentry, mode, nd);
1943 fsnotify_create(dir, dentry);
1947 static int may_open(struct path *path, int acc_mode, int flag)
1949 struct dentry *dentry = path->dentry;
1950 struct inode *inode = dentry->d_inode;
1960 switch (inode->i_mode & S_IFMT) {
1964 if (acc_mode & MAY_WRITE)
1969 if (path->mnt->mnt_flags & MNT_NODEV)
1978 error = inode_permission(inode, acc_mode);
1983 * An append-only file must be opened in append mode for writing.
1985 if (IS_APPEND(inode)) {
1986 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
1992 /* O_NOATIME can only be set by the owner or superuser */
1993 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
1997 * Ensure there are no outstanding leases on the file.
1999 return break_lease(inode, flag);
2002 static int handle_truncate(struct file *filp)
2004 struct path *path = &filp->f_path;
2005 struct inode *inode = path->dentry->d_inode;
2006 int error = get_write_access(inode);
2010 * Refuse to truncate files with mandatory locks held on them.
2012 error = locks_verify_locked(inode);
2014 error = security_path_truncate(path);
2016 error = do_truncate(path->dentry, 0,
2017 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2020 put_write_access(inode);
2024 static inline int open_to_namei_flags(int flag)
2026 if ((flag & O_ACCMODE) == 3)
2032 * Handle the last step of open()
2034 static struct file *do_last(struct nameidata *nd, struct path *path,
2035 const struct open_flags *op, const char *pathname)
2037 struct dentry *dir = nd->path.dentry;
2038 struct dentry *dentry;
2039 int open_flag = op->open_flag;
2040 int will_truncate = open_flag & O_TRUNC;
2042 int acc_mode = op->acc_mode;
2046 nd->flags &= ~LOOKUP_PARENT;
2047 nd->flags |= op->intent;
2049 switch (nd->last_type) {
2052 error = handle_dots(nd, nd->last_type);
2054 return ERR_PTR(error);
2057 error = complete_walk(nd);
2059 return ERR_PTR(error);
2060 audit_inode(pathname, nd->path.dentry);
2061 if (open_flag & O_CREAT) {
2067 error = complete_walk(nd);
2069 return ERR_PTR(error);
2070 audit_inode(pathname, dir);
2074 if (!(open_flag & O_CREAT)) {
2076 if (nd->last.name[nd->last.len])
2077 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2078 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2080 /* we _can_ be in RCU mode here */
2081 error = walk_component(nd, path, &nd->last, LAST_NORM,
2084 return ERR_PTR(error);
2085 if (error) /* symlink */
2088 error = complete_walk(nd);
2090 return ERR_PTR(-ECHILD);
2093 if (nd->flags & LOOKUP_DIRECTORY) {
2094 if (!nd->inode->i_op->lookup)
2097 audit_inode(pathname, nd->path.dentry);
2101 /* create side of things */
2102 error = complete_walk(nd);
2104 return ERR_PTR(error);
2106 audit_inode(pathname, dir);
2108 /* trailing slashes? */
2109 if (nd->last.name[nd->last.len])
2112 mutex_lock(&dir->d_inode->i_mutex);
2114 dentry = lookup_hash(nd);
2115 error = PTR_ERR(dentry);
2116 if (IS_ERR(dentry)) {
2117 mutex_unlock(&dir->d_inode->i_mutex);
2121 path->dentry = dentry;
2122 path->mnt = nd->path.mnt;
2124 /* Negative dentry, just create the file */
2125 if (!dentry->d_inode) {
2126 int mode = op->mode;
2127 if (!IS_POSIXACL(dir->d_inode))
2128 mode &= ~current_umask();
2130 * This write is needed to ensure that a
2131 * rw->ro transition does not occur between
2132 * the time when the file is created and when
2133 * a permanent write count is taken through
2134 * the 'struct file' in nameidata_to_filp().
2136 error = mnt_want_write(nd->path.mnt);
2138 goto exit_mutex_unlock;
2140 /* Don't check for write permission, don't truncate */
2141 open_flag &= ~O_TRUNC;
2143 acc_mode = MAY_OPEN;
2144 error = security_path_mknod(&nd->path, dentry, mode, 0);
2146 goto exit_mutex_unlock;
2147 error = vfs_create(dir->d_inode, dentry, mode, nd);
2149 goto exit_mutex_unlock;
2150 mutex_unlock(&dir->d_inode->i_mutex);
2151 dput(nd->path.dentry);
2152 nd->path.dentry = dentry;
2157 * It already exists.
2159 mutex_unlock(&dir->d_inode->i_mutex);
2160 audit_inode(pathname, path->dentry);
2163 if (open_flag & O_EXCL)
2166 error = follow_managed(path, nd->flags);
2171 if (!path->dentry->d_inode)
2174 if (path->dentry->d_inode->i_op->follow_link)
2177 path_to_nameidata(path, nd);
2178 nd->inode = path->dentry->d_inode;
2180 if (S_ISDIR(nd->inode->i_mode))
2183 if (!S_ISREG(nd->inode->i_mode))
2186 if (will_truncate) {
2187 error = mnt_want_write(nd->path.mnt);
2193 error = may_open(&nd->path, acc_mode, open_flag);
2196 filp = nameidata_to_filp(nd);
2197 if (!IS_ERR(filp)) {
2198 error = ima_file_check(filp, op->acc_mode);
2201 filp = ERR_PTR(error);
2204 if (!IS_ERR(filp)) {
2205 if (will_truncate) {
2206 error = handle_truncate(filp);
2209 filp = ERR_PTR(error);
2215 mnt_drop_write(nd->path.mnt);
2216 path_put(&nd->path);
2220 mutex_unlock(&dir->d_inode->i_mutex);
2222 path_put_conditional(path, nd);
2224 filp = ERR_PTR(error);
2228 static struct file *path_openat(int dfd, const char *pathname,
2229 struct nameidata *nd, const struct open_flags *op, int flags)
2231 struct file *base = NULL;
2236 filp = get_empty_filp();
2238 return ERR_PTR(-ENFILE);
2240 filp->f_flags = op->open_flag;
2241 nd->intent.open.file = filp;
2242 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2243 nd->intent.open.create_mode = op->mode;
2245 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2246 if (unlikely(error))
2249 current->total_link_count = 0;
2250 error = link_path_walk(pathname, nd);
2251 if (unlikely(error))
2254 filp = do_last(nd, &path, op, pathname);
2255 while (unlikely(!filp)) { /* trailing symlink */
2256 struct path link = path;
2258 if (!(nd->flags & LOOKUP_FOLLOW)) {
2259 path_put_conditional(&path, nd);
2260 path_put(&nd->path);
2261 filp = ERR_PTR(-ELOOP);
2264 nd->flags |= LOOKUP_PARENT;
2265 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2266 error = follow_link(&link, nd, &cookie);
2267 if (unlikely(error))
2268 filp = ERR_PTR(error);
2270 filp = do_last(nd, &path, op, pathname);
2271 put_link(nd, &link, cookie);
2274 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2275 path_put(&nd->root);
2278 release_open_intent(nd);
2282 filp = ERR_PTR(error);
2286 struct file *do_filp_open(int dfd, const char *pathname,
2287 const struct open_flags *op, int flags)
2289 struct nameidata nd;
2292 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2293 if (unlikely(filp == ERR_PTR(-ECHILD)))
2294 filp = path_openat(dfd, pathname, &nd, op, flags);
2295 if (unlikely(filp == ERR_PTR(-ESTALE)))
2296 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2300 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2301 const char *name, const struct open_flags *op, int flags)
2303 struct nameidata nd;
2307 nd.root.dentry = dentry;
2309 flags |= LOOKUP_ROOT;
2311 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2312 return ERR_PTR(-ELOOP);
2314 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2315 if (unlikely(file == ERR_PTR(-ECHILD)))
2316 file = path_openat(-1, name, &nd, op, flags);
2317 if (unlikely(file == ERR_PTR(-ESTALE)))
2318 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2322 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2324 struct dentry *dentry = ERR_PTR(-EEXIST);
2325 struct nameidata nd;
2326 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2328 return ERR_PTR(error);
2331 * Yucky last component or no last component at all?
2332 * (foo/., foo/.., /////)
2334 if (nd.last_type != LAST_NORM)
2336 nd.flags &= ~LOOKUP_PARENT;
2337 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2338 nd.intent.open.flags = O_EXCL;
2341 * Do the final lookup.
2343 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2344 dentry = lookup_hash(&nd);
2348 if (dentry->d_inode)
2351 * Special case - lookup gave negative, but... we had foo/bar/
2352 * From the vfs_mknod() POV we just have a negative dentry -
2353 * all is fine. Let's be bastards - you had / on the end, you've
2354 * been asking for (non-existent) directory. -ENOENT for you.
2356 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2358 dentry = ERR_PTR(-ENOENT);
2365 dentry = ERR_PTR(-EEXIST);
2367 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2372 EXPORT_SYMBOL(kern_path_create);
2374 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2376 char *tmp = getname(pathname);
2379 return ERR_CAST(tmp);
2380 res = kern_path_create(dfd, tmp, path, is_dir);
2384 EXPORT_SYMBOL(user_path_create);
2386 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2388 int error = may_create(dir, dentry);
2393 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2394 !ns_capable(inode_userns(dir), CAP_MKNOD))
2397 if (!dir->i_op->mknod)
2400 error = devcgroup_inode_mknod(mode, dev);
2404 error = security_inode_mknod(dir, dentry, mode, dev);
2408 error = dir->i_op->mknod(dir, dentry, mode, dev);
2410 fsnotify_create(dir, dentry);
2414 static int may_mknod(mode_t mode)
2416 switch (mode & S_IFMT) {
2422 case 0: /* zero mode translates to S_IFREG */
2431 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2434 struct dentry *dentry;
2441 dentry = user_path_create(dfd, filename, &path, 0);
2443 return PTR_ERR(dentry);
2445 if (!IS_POSIXACL(path.dentry->d_inode))
2446 mode &= ~current_umask();
2447 error = may_mknod(mode);
2450 error = mnt_want_write(path.mnt);
2453 error = security_path_mknod(&path, dentry, mode, dev);
2455 goto out_drop_write;
2456 switch (mode & S_IFMT) {
2457 case 0: case S_IFREG:
2458 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2460 case S_IFCHR: case S_IFBLK:
2461 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2462 new_decode_dev(dev));
2464 case S_IFIFO: case S_IFSOCK:
2465 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2469 mnt_drop_write(path.mnt);
2472 mutex_unlock(&path.dentry->d_inode->i_mutex);
2478 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2480 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2483 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2485 int error = may_create(dir, dentry);
2490 if (!dir->i_op->mkdir)
2493 mode &= (S_IRWXUGO|S_ISVTX);
2494 error = security_inode_mkdir(dir, dentry, mode);
2498 error = dir->i_op->mkdir(dir, dentry, mode);
2500 fsnotify_mkdir(dir, dentry);
2504 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2506 struct dentry *dentry;
2510 dentry = user_path_create(dfd, pathname, &path, 1);
2512 return PTR_ERR(dentry);
2514 if (!IS_POSIXACL(path.dentry->d_inode))
2515 mode &= ~current_umask();
2516 error = mnt_want_write(path.mnt);
2519 error = security_path_mkdir(&path, dentry, mode);
2521 goto out_drop_write;
2522 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2524 mnt_drop_write(path.mnt);
2527 mutex_unlock(&path.dentry->d_inode->i_mutex);
2532 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2534 return sys_mkdirat(AT_FDCWD, pathname, mode);
2538 * The dentry_unhash() helper will try to drop the dentry early: we
2539 * should have a usage count of 2 if we're the only user of this
2540 * dentry, and if that is true (possibly after pruning the dcache),
2541 * then we drop the dentry now.
2543 * A low-level filesystem can, if it choses, legally
2546 * if (!d_unhashed(dentry))
2549 * if it cannot handle the case of removing a directory
2550 * that is still in use by something else..
2552 void dentry_unhash(struct dentry *dentry)
2554 shrink_dcache_parent(dentry);
2555 spin_lock(&dentry->d_lock);
2556 if (dentry->d_count == 1)
2558 spin_unlock(&dentry->d_lock);
2561 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2563 int error = may_delete(dir, dentry, 1);
2568 if (!dir->i_op->rmdir)
2571 mutex_lock(&dentry->d_inode->i_mutex);
2574 if (d_mountpoint(dentry))
2577 error = security_inode_rmdir(dir, dentry);
2581 shrink_dcache_parent(dentry);
2582 error = dir->i_op->rmdir(dir, dentry);
2586 dentry->d_inode->i_flags |= S_DEAD;
2590 mutex_unlock(&dentry->d_inode->i_mutex);
2596 static long do_rmdir(int dfd, const char __user *pathname)
2600 struct dentry *dentry;
2601 struct nameidata nd;
2603 error = user_path_parent(dfd, pathname, &nd, &name);
2607 switch(nd.last_type) {
2619 nd.flags &= ~LOOKUP_PARENT;
2621 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2622 dentry = lookup_hash(&nd);
2623 error = PTR_ERR(dentry);
2626 if (!dentry->d_inode) {
2630 error = mnt_want_write(nd.path.mnt);
2633 error = security_path_rmdir(&nd.path, dentry);
2636 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2638 mnt_drop_write(nd.path.mnt);
2642 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2649 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2651 return do_rmdir(AT_FDCWD, pathname);
2654 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2656 int error = may_delete(dir, dentry, 0);
2661 if (!dir->i_op->unlink)
2664 mutex_lock(&dentry->d_inode->i_mutex);
2665 if (d_mountpoint(dentry))
2668 error = security_inode_unlink(dir, dentry);
2670 error = dir->i_op->unlink(dir, dentry);
2675 mutex_unlock(&dentry->d_inode->i_mutex);
2677 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2678 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2679 fsnotify_link_count(dentry->d_inode);
2687 * Make sure that the actual truncation of the file will occur outside its
2688 * directory's i_mutex. Truncate can take a long time if there is a lot of
2689 * writeout happening, and we don't want to prevent access to the directory
2690 * while waiting on the I/O.
2692 static long do_unlinkat(int dfd, const char __user *pathname)
2696 struct dentry *dentry;
2697 struct nameidata nd;
2698 struct inode *inode = NULL;
2700 error = user_path_parent(dfd, pathname, &nd, &name);
2705 if (nd.last_type != LAST_NORM)
2708 nd.flags &= ~LOOKUP_PARENT;
2710 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2711 dentry = lookup_hash(&nd);
2712 error = PTR_ERR(dentry);
2713 if (!IS_ERR(dentry)) {
2714 /* Why not before? Because we want correct error value */
2715 if (nd.last.name[nd.last.len])
2717 inode = dentry->d_inode;
2721 error = mnt_want_write(nd.path.mnt);
2724 error = security_path_unlink(&nd.path, dentry);
2727 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2729 mnt_drop_write(nd.path.mnt);
2733 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2735 iput(inode); /* truncate the inode here */
2742 error = !dentry->d_inode ? -ENOENT :
2743 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2747 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2749 if ((flag & ~AT_REMOVEDIR) != 0)
2752 if (flag & AT_REMOVEDIR)
2753 return do_rmdir(dfd, pathname);
2755 return do_unlinkat(dfd, pathname);
2758 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2760 return do_unlinkat(AT_FDCWD, pathname);
2763 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2765 int error = may_create(dir, dentry);
2770 if (!dir->i_op->symlink)
2773 error = security_inode_symlink(dir, dentry, oldname);
2777 error = dir->i_op->symlink(dir, dentry, oldname);
2779 fsnotify_create(dir, dentry);
2783 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2784 int, newdfd, const char __user *, newname)
2788 struct dentry *dentry;
2791 from = getname(oldname);
2793 return PTR_ERR(from);
2795 dentry = user_path_create(newdfd, newname, &path, 0);
2796 error = PTR_ERR(dentry);
2800 error = mnt_want_write(path.mnt);
2803 error = security_path_symlink(&path, dentry, from);
2805 goto out_drop_write;
2806 error = vfs_symlink(path.dentry->d_inode, dentry, from);
2808 mnt_drop_write(path.mnt);
2811 mutex_unlock(&path.dentry->d_inode->i_mutex);
2818 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2820 return sys_symlinkat(oldname, AT_FDCWD, newname);
2823 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2825 struct inode *inode = old_dentry->d_inode;
2831 error = may_create(dir, new_dentry);
2835 if (dir->i_sb != inode->i_sb)
2839 * A link to an append-only or immutable file cannot be created.
2841 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2843 if (!dir->i_op->link)
2845 if (S_ISDIR(inode->i_mode))
2848 error = security_inode_link(old_dentry, dir, new_dentry);
2852 mutex_lock(&inode->i_mutex);
2853 /* Make sure we don't allow creating hardlink to an unlinked file */
2854 if (inode->i_nlink == 0)
2857 error = dir->i_op->link(old_dentry, dir, new_dentry);
2858 mutex_unlock(&inode->i_mutex);
2860 fsnotify_link(dir, inode, new_dentry);
2865 * Hardlinks are often used in delicate situations. We avoid
2866 * security-related surprises by not following symlinks on the
2869 * We don't follow them on the oldname either to be compatible
2870 * with linux 2.0, and to avoid hard-linking to directories
2871 * and other special files. --ADM
2873 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2874 int, newdfd, const char __user *, newname, int, flags)
2876 struct dentry *new_dentry;
2877 struct path old_path, new_path;
2881 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
2884 * To use null names we require CAP_DAC_READ_SEARCH
2885 * This ensures that not everyone will be able to create
2886 * handlink using the passed filedescriptor.
2888 if (flags & AT_EMPTY_PATH) {
2889 if (!capable(CAP_DAC_READ_SEARCH))
2894 if (flags & AT_SYMLINK_FOLLOW)
2895 how |= LOOKUP_FOLLOW;
2897 error = user_path_at(olddfd, oldname, how, &old_path);
2901 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
2902 error = PTR_ERR(new_dentry);
2903 if (IS_ERR(new_dentry))
2907 if (old_path.mnt != new_path.mnt)
2909 error = mnt_want_write(new_path.mnt);
2912 error = security_path_link(old_path.dentry, &new_path, new_dentry);
2914 goto out_drop_write;
2915 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
2917 mnt_drop_write(new_path.mnt);
2920 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
2921 path_put(&new_path);
2923 path_put(&old_path);
2928 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2930 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2934 * The worst of all namespace operations - renaming directory. "Perverted"
2935 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2937 * a) we can get into loop creation. Check is done in is_subdir().
2938 * b) race potential - two innocent renames can create a loop together.
2939 * That's where 4.4 screws up. Current fix: serialization on
2940 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2942 * c) we have to lock _three_ objects - parents and victim (if it exists).
2943 * And that - after we got ->i_mutex on parents (until then we don't know
2944 * whether the target exists). Solution: try to be smart with locking
2945 * order for inodes. We rely on the fact that tree topology may change
2946 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2947 * move will be locked. Thus we can rank directories by the tree
2948 * (ancestors first) and rank all non-directories after them.
2949 * That works since everybody except rename does "lock parent, lookup,
2950 * lock child" and rename is under ->s_vfs_rename_mutex.
2951 * HOWEVER, it relies on the assumption that any object with ->lookup()
2952 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2953 * we'd better make sure that there's no link(2) for them.
2954 * d) conversion from fhandle to dentry may come in the wrong moment - when
2955 * we are removing the target. Solution: we will have to grab ->i_mutex
2956 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2957 * ->i_mutex on parents, which works but leads to some truly excessive
2960 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2961 struct inode *new_dir, struct dentry *new_dentry)
2964 struct inode *target = new_dentry->d_inode;
2967 * If we are going to change the parent - check write permissions,
2968 * we'll need to flip '..'.
2970 if (new_dir != old_dir) {
2971 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2976 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2981 mutex_lock(&target->i_mutex);
2984 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
2988 shrink_dcache_parent(new_dentry);
2989 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2994 target->i_flags |= S_DEAD;
2995 dont_mount(new_dentry);
2999 mutex_unlock(&target->i_mutex);
3001 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3002 d_move(old_dentry,new_dentry);
3006 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3007 struct inode *new_dir, struct dentry *new_dentry)
3009 struct inode *target = new_dentry->d_inode;
3012 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3018 mutex_lock(&target->i_mutex);
3021 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3024 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3029 dont_mount(new_dentry);
3030 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3031 d_move(old_dentry, new_dentry);
3034 mutex_unlock(&target->i_mutex);
3039 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3040 struct inode *new_dir, struct dentry *new_dentry)
3043 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3044 const unsigned char *old_name;
3046 if (old_dentry->d_inode == new_dentry->d_inode)
3049 error = may_delete(old_dir, old_dentry, is_dir);
3053 if (!new_dentry->d_inode)
3054 error = may_create(new_dir, new_dentry);
3056 error = may_delete(new_dir, new_dentry, is_dir);
3060 if (!old_dir->i_op->rename)
3063 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3066 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3068 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3070 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3071 new_dentry->d_inode, old_dentry);
3072 fsnotify_oldname_free(old_name);
3077 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3078 int, newdfd, const char __user *, newname)
3080 struct dentry *old_dir, *new_dir;
3081 struct dentry *old_dentry, *new_dentry;
3082 struct dentry *trap;
3083 struct nameidata oldnd, newnd;
3088 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3092 error = user_path_parent(newdfd, newname, &newnd, &to);
3097 if (oldnd.path.mnt != newnd.path.mnt)
3100 old_dir = oldnd.path.dentry;
3102 if (oldnd.last_type != LAST_NORM)
3105 new_dir = newnd.path.dentry;
3106 if (newnd.last_type != LAST_NORM)
3109 oldnd.flags &= ~LOOKUP_PARENT;
3110 newnd.flags &= ~LOOKUP_PARENT;
3111 newnd.flags |= LOOKUP_RENAME_TARGET;
3113 trap = lock_rename(new_dir, old_dir);
3115 old_dentry = lookup_hash(&oldnd);
3116 error = PTR_ERR(old_dentry);
3117 if (IS_ERR(old_dentry))
3119 /* source must exist */
3121 if (!old_dentry->d_inode)
3123 /* unless the source is a directory trailing slashes give -ENOTDIR */
3124 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3126 if (oldnd.last.name[oldnd.last.len])
3128 if (newnd.last.name[newnd.last.len])
3131 /* source should not be ancestor of target */
3133 if (old_dentry == trap)
3135 new_dentry = lookup_hash(&newnd);
3136 error = PTR_ERR(new_dentry);
3137 if (IS_ERR(new_dentry))
3139 /* target should not be an ancestor of source */
3141 if (new_dentry == trap)
3144 error = mnt_want_write(oldnd.path.mnt);
3147 error = security_path_rename(&oldnd.path, old_dentry,
3148 &newnd.path, new_dentry);
3151 error = vfs_rename(old_dir->d_inode, old_dentry,
3152 new_dir->d_inode, new_dentry);
3154 mnt_drop_write(oldnd.path.mnt);
3160 unlock_rename(new_dir, old_dir);
3162 path_put(&newnd.path);
3165 path_put(&oldnd.path);
3171 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3173 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3176 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3180 len = PTR_ERR(link);
3185 if (len > (unsigned) buflen)
3187 if (copy_to_user(buffer, link, len))
3194 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3195 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3196 * using) it for any given inode is up to filesystem.
3198 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3200 struct nameidata nd;
3205 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3207 return PTR_ERR(cookie);
3209 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3210 if (dentry->d_inode->i_op->put_link)
3211 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3215 int vfs_follow_link(struct nameidata *nd, const char *link)
3217 return __vfs_follow_link(nd, link);
3220 /* get the link contents into pagecache */
3221 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3225 struct address_space *mapping = dentry->d_inode->i_mapping;
3226 page = read_mapping_page(mapping, 0, NULL);
3231 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3235 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3237 struct page *page = NULL;
3238 char *s = page_getlink(dentry, &page);
3239 int res = vfs_readlink(dentry,buffer,buflen,s);
3242 page_cache_release(page);
3247 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3249 struct page *page = NULL;
3250 nd_set_link(nd, page_getlink(dentry, &page));
3254 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3256 struct page *page = cookie;
3260 page_cache_release(page);
3265 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3267 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3269 struct address_space *mapping = inode->i_mapping;
3274 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3276 flags |= AOP_FLAG_NOFS;
3279 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3280 flags, &page, &fsdata);
3284 kaddr = kmap_atomic(page, KM_USER0);
3285 memcpy(kaddr, symname, len-1);
3286 kunmap_atomic(kaddr, KM_USER0);
3288 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3295 mark_inode_dirty(inode);
3301 int page_symlink(struct inode *inode, const char *symname, int len)
3303 return __page_symlink(inode, symname, len,
3304 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3307 const struct inode_operations page_symlink_inode_operations = {
3308 .readlink = generic_readlink,
3309 .follow_link = page_follow_link_light,
3310 .put_link = page_put_link,
3313 EXPORT_SYMBOL(user_path_at);
3314 EXPORT_SYMBOL(follow_down_one);
3315 EXPORT_SYMBOL(follow_down);
3316 EXPORT_SYMBOL(follow_up);
3317 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3318 EXPORT_SYMBOL(getname);
3319 EXPORT_SYMBOL(lock_rename);
3320 EXPORT_SYMBOL(lookup_one_len);
3321 EXPORT_SYMBOL(page_follow_link_light);
3322 EXPORT_SYMBOL(page_put_link);
3323 EXPORT_SYMBOL(page_readlink);
3324 EXPORT_SYMBOL(__page_symlink);
3325 EXPORT_SYMBOL(page_symlink);
3326 EXPORT_SYMBOL(page_symlink_inode_operations);
3327 EXPORT_SYMBOL(kern_path);
3328 EXPORT_SYMBOL(vfs_path_lookup);
3329 EXPORT_SYMBOL(inode_permission);
3330 EXPORT_SYMBOL(unlock_rename);
3331 EXPORT_SYMBOL(vfs_create);
3332 EXPORT_SYMBOL(vfs_follow_link);
3333 EXPORT_SYMBOL(vfs_link);
3334 EXPORT_SYMBOL(vfs_mkdir);
3335 EXPORT_SYMBOL(vfs_mknod);
3336 EXPORT_SYMBOL(generic_permission);
3337 EXPORT_SYMBOL(vfs_readlink);
3338 EXPORT_SYMBOL(vfs_rename);
3339 EXPORT_SYMBOL(vfs_rmdir);
3340 EXPORT_SYMBOL(vfs_symlink);
3341 EXPORT_SYMBOL(vfs_unlink);
3342 EXPORT_SYMBOL(dentry_unhash);
3343 EXPORT_SYMBOL(generic_readlink);