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;
182 if (mask & MAY_NOT_BLOCK) {
183 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
186 /* no ->get_acl() calls in RCU mode... */
187 if (acl == ACL_NOT_CACHED)
189 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
192 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
195 * A filesystem can force a ACL callback by just never filling the
196 * ACL cache. But normally you'd fill the cache either at inode
197 * instantiation time, or on the first ->get_acl call.
199 * If the filesystem doesn't have a get_acl() function at all, we'll
200 * just create the negative cache entry.
202 if (acl == ACL_NOT_CACHED) {
203 if (inode->i_op->get_acl) {
204 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
208 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
214 int error = posix_acl_permission(inode, acl, mask);
215 posix_acl_release(acl);
224 * This does basic POSIX ACL permission checking
226 static int acl_permission_check(struct inode *inode, int mask)
228 unsigned int mode = inode->i_mode;
230 mask &= MAY_READ | MAY_WRITE | MAY_EXEC | MAY_NOT_BLOCK;
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 POSIX ACL 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 int inode_permission(struct inode *inode, int mask)
345 if (unlikely(mask & MAY_WRITE)) {
346 umode_t mode = inode->i_mode;
349 * Nobody gets write access to a read-only fs.
351 if (IS_RDONLY(inode) &&
352 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
356 * Nobody gets write access to an immutable file.
358 if (IS_IMMUTABLE(inode))
362 retval = do_inode_permission(inode, mask);
366 retval = devcgroup_inode_permission(inode, mask);
370 return security_inode_permission(inode, mask);
374 * path_get - get a reference to a path
375 * @path: path to get the reference to
377 * Given a path increment the reference count to the dentry and the vfsmount.
379 void path_get(struct path *path)
384 EXPORT_SYMBOL(path_get);
387 * path_put - put a reference to a path
388 * @path: path to put the reference to
390 * Given a path decrement the reference count to the dentry and the vfsmount.
392 void path_put(struct path *path)
397 EXPORT_SYMBOL(path_put);
400 * Path walking has 2 modes, rcu-walk and ref-walk (see
401 * Documentation/filesystems/path-lookup.txt). In situations when we can't
402 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
403 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
404 * mode. Refcounts are grabbed at the last known good point before rcu-walk
405 * got stuck, so ref-walk may continue from there. If this is not successful
406 * (eg. a seqcount has changed), then failure is returned and it's up to caller
407 * to restart the path walk from the beginning in ref-walk mode.
411 * unlazy_walk - try to switch to ref-walk mode.
412 * @nd: nameidata pathwalk data
413 * @dentry: child of nd->path.dentry or NULL
414 * Returns: 0 on success, -ECHILD on failure
416 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
417 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
418 * @nd or NULL. Must be called from rcu-walk context.
420 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
422 struct fs_struct *fs = current->fs;
423 struct dentry *parent = nd->path.dentry;
426 BUG_ON(!(nd->flags & LOOKUP_RCU));
427 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
429 spin_lock(&fs->lock);
430 if (nd->root.mnt != fs->root.mnt ||
431 nd->root.dentry != fs->root.dentry)
434 spin_lock(&parent->d_lock);
436 if (!__d_rcu_to_refcount(parent, nd->seq))
438 BUG_ON(nd->inode != parent->d_inode);
440 if (dentry->d_parent != parent)
442 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
443 if (!__d_rcu_to_refcount(dentry, nd->seq))
446 * If the sequence check on the child dentry passed, then
447 * the child has not been removed from its parent. This
448 * means the parent dentry must be valid and able to take
449 * a reference at this point.
451 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
452 BUG_ON(!parent->d_count);
454 spin_unlock(&dentry->d_lock);
456 spin_unlock(&parent->d_lock);
459 spin_unlock(&fs->lock);
461 mntget(nd->path.mnt);
464 br_read_unlock(vfsmount_lock);
465 nd->flags &= ~LOOKUP_RCU;
469 spin_unlock(&dentry->d_lock);
471 spin_unlock(&parent->d_lock);
474 spin_unlock(&fs->lock);
479 * release_open_intent - free up open intent resources
480 * @nd: pointer to nameidata
482 void release_open_intent(struct nameidata *nd)
484 struct file *file = nd->intent.open.file;
486 if (file && !IS_ERR(file)) {
487 if (file->f_path.dentry == NULL)
494 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
496 return dentry->d_op->d_revalidate(dentry, nd);
500 * complete_walk - successful completion of path walk
501 * @nd: pointer nameidata
503 * If we had been in RCU mode, drop out of it and legitimize nd->path.
504 * Revalidate the final result, unless we'd already done that during
505 * the path walk or the filesystem doesn't ask for it. Return 0 on
506 * success, -error on failure. In case of failure caller does not
507 * need to drop nd->path.
509 static int complete_walk(struct nameidata *nd)
511 struct dentry *dentry = nd->path.dentry;
514 if (nd->flags & LOOKUP_RCU) {
515 nd->flags &= ~LOOKUP_RCU;
516 if (!(nd->flags & LOOKUP_ROOT))
518 spin_lock(&dentry->d_lock);
519 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
520 spin_unlock(&dentry->d_lock);
522 br_read_unlock(vfsmount_lock);
525 BUG_ON(nd->inode != dentry->d_inode);
526 spin_unlock(&dentry->d_lock);
527 mntget(nd->path.mnt);
529 br_read_unlock(vfsmount_lock);
532 if (likely(!(nd->flags & LOOKUP_JUMPED)))
535 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
538 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
541 /* Note: we do not d_invalidate() */
542 status = d_revalidate(dentry, nd);
553 static __always_inline void set_root(struct nameidata *nd)
556 get_fs_root(current->fs, &nd->root);
559 static int link_path_walk(const char *, struct nameidata *);
561 static __always_inline void set_root_rcu(struct nameidata *nd)
564 struct fs_struct *fs = current->fs;
568 seq = read_seqcount_begin(&fs->seq);
570 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
571 } while (read_seqcount_retry(&fs->seq, seq));
575 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
587 nd->flags |= LOOKUP_JUMPED;
589 nd->inode = nd->path.dentry->d_inode;
591 ret = link_path_walk(link, nd);
595 return PTR_ERR(link);
598 static void path_put_conditional(struct path *path, struct nameidata *nd)
601 if (path->mnt != nd->path.mnt)
605 static inline void path_to_nameidata(const struct path *path,
606 struct nameidata *nd)
608 if (!(nd->flags & LOOKUP_RCU)) {
609 dput(nd->path.dentry);
610 if (nd->path.mnt != path->mnt)
611 mntput(nd->path.mnt);
613 nd->path.mnt = path->mnt;
614 nd->path.dentry = path->dentry;
617 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
619 struct inode *inode = link->dentry->d_inode;
620 if (!IS_ERR(cookie) && inode->i_op->put_link)
621 inode->i_op->put_link(link->dentry, nd, cookie);
625 static __always_inline int
626 follow_link(struct path *link, struct nameidata *nd, void **p)
629 struct dentry *dentry = link->dentry;
631 BUG_ON(nd->flags & LOOKUP_RCU);
633 if (link->mnt == nd->path.mnt)
636 if (unlikely(current->total_link_count >= 40)) {
637 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
642 current->total_link_count++;
644 touch_atime(link->mnt, dentry);
645 nd_set_link(nd, NULL);
647 error = security_inode_follow_link(link->dentry, nd);
649 *p = ERR_PTR(error); /* no ->put_link(), please */
654 nd->last_type = LAST_BIND;
655 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
658 char *s = nd_get_link(nd);
661 error = __vfs_follow_link(nd, s);
662 else if (nd->last_type == LAST_BIND) {
663 nd->flags |= LOOKUP_JUMPED;
664 nd->inode = nd->path.dentry->d_inode;
665 if (nd->inode->i_op->follow_link) {
666 /* stepped on a _really_ weird one */
675 static int follow_up_rcu(struct path *path)
677 struct vfsmount *parent;
678 struct dentry *mountpoint;
680 parent = path->mnt->mnt_parent;
681 if (parent == path->mnt)
683 mountpoint = path->mnt->mnt_mountpoint;
684 path->dentry = mountpoint;
689 int follow_up(struct path *path)
691 struct vfsmount *parent;
692 struct dentry *mountpoint;
694 br_read_lock(vfsmount_lock);
695 parent = path->mnt->mnt_parent;
696 if (parent == path->mnt) {
697 br_read_unlock(vfsmount_lock);
701 mountpoint = dget(path->mnt->mnt_mountpoint);
702 br_read_unlock(vfsmount_lock);
704 path->dentry = mountpoint;
711 * Perform an automount
712 * - return -EISDIR to tell follow_managed() to stop and return the path we
715 static int follow_automount(struct path *path, unsigned flags,
718 struct vfsmount *mnt;
721 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
724 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
725 * and this is the terminal part of the path.
727 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_PARENT))
728 return -EISDIR; /* we actually want to stop here */
730 /* We don't want to mount if someone's just doing a stat -
731 * unless they're stat'ing a directory and appended a '/' to
734 * We do, however, want to mount if someone wants to open or
735 * create a file of any type under the mountpoint, wants to
736 * traverse through the mountpoint or wants to open the
737 * mounted directory. Also, autofs may mark negative dentries
738 * as being automount points. These will need the attentions
739 * of the daemon to instantiate them before they can be used.
741 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
742 LOOKUP_OPEN | LOOKUP_CREATE)) &&
743 path->dentry->d_inode)
746 current->total_link_count++;
747 if (current->total_link_count >= 40)
750 mnt = path->dentry->d_op->d_automount(path);
753 * The filesystem is allowed to return -EISDIR here to indicate
754 * it doesn't want to automount. For instance, autofs would do
755 * this so that its userspace daemon can mount on this dentry.
757 * However, we can only permit this if it's a terminal point in
758 * the path being looked up; if it wasn't then the remainder of
759 * the path is inaccessible and we should say so.
761 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
766 if (!mnt) /* mount collision */
770 /* lock_mount() may release path->mnt on error */
774 err = finish_automount(mnt, path);
778 /* Someone else made a mount here whilst we were busy */
783 path->dentry = dget(mnt->mnt_root);
792 * Handle a dentry that is managed in some way.
793 * - Flagged for transit management (autofs)
794 * - Flagged as mountpoint
795 * - Flagged as automount point
797 * This may only be called in refwalk mode.
799 * Serialization is taken care of in namespace.c
801 static int follow_managed(struct path *path, unsigned flags)
803 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
805 bool need_mntput = false;
808 /* Given that we're not holding a lock here, we retain the value in a
809 * local variable for each dentry as we look at it so that we don't see
810 * the components of that value change under us */
811 while (managed = ACCESS_ONCE(path->dentry->d_flags),
812 managed &= DCACHE_MANAGED_DENTRY,
813 unlikely(managed != 0)) {
814 /* Allow the filesystem to manage the transit without i_mutex
816 if (managed & DCACHE_MANAGE_TRANSIT) {
817 BUG_ON(!path->dentry->d_op);
818 BUG_ON(!path->dentry->d_op->d_manage);
819 ret = path->dentry->d_op->d_manage(path->dentry, false);
824 /* Transit to a mounted filesystem. */
825 if (managed & DCACHE_MOUNTED) {
826 struct vfsmount *mounted = lookup_mnt(path);
832 path->dentry = dget(mounted->mnt_root);
837 /* Something is mounted on this dentry in another
838 * namespace and/or whatever was mounted there in this
839 * namespace got unmounted before we managed to get the
843 /* Handle an automount point */
844 if (managed & DCACHE_NEED_AUTOMOUNT) {
845 ret = follow_automount(path, flags, &need_mntput);
851 /* We didn't change the current path point */
855 if (need_mntput && path->mnt == mnt)
862 int follow_down_one(struct path *path)
864 struct vfsmount *mounted;
866 mounted = lookup_mnt(path);
871 path->dentry = dget(mounted->mnt_root);
877 static inline bool managed_dentry_might_block(struct dentry *dentry)
879 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
880 dentry->d_op->d_manage(dentry, true) < 0);
884 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
885 * we meet a managed dentry that would need blocking.
887 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
888 struct inode **inode)
891 struct vfsmount *mounted;
893 * Don't forget we might have a non-mountpoint managed dentry
894 * that wants to block transit.
896 if (unlikely(managed_dentry_might_block(path->dentry)))
899 if (!d_mountpoint(path->dentry))
902 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
906 path->dentry = mounted->mnt_root;
907 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
909 * Update the inode too. We don't need to re-check the
910 * dentry sequence number here after this d_inode read,
911 * because a mount-point is always pinned.
913 *inode = path->dentry->d_inode;
918 static void follow_mount_rcu(struct nameidata *nd)
920 while (d_mountpoint(nd->path.dentry)) {
921 struct vfsmount *mounted;
922 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
925 nd->path.mnt = mounted;
926 nd->path.dentry = mounted->mnt_root;
927 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
931 static int follow_dotdot_rcu(struct nameidata *nd)
936 if (nd->path.dentry == nd->root.dentry &&
937 nd->path.mnt == nd->root.mnt) {
940 if (nd->path.dentry != nd->path.mnt->mnt_root) {
941 struct dentry *old = nd->path.dentry;
942 struct dentry *parent = old->d_parent;
945 seq = read_seqcount_begin(&parent->d_seq);
946 if (read_seqcount_retry(&old->d_seq, nd->seq))
948 nd->path.dentry = parent;
952 if (!follow_up_rcu(&nd->path))
954 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
956 follow_mount_rcu(nd);
957 nd->inode = nd->path.dentry->d_inode;
961 nd->flags &= ~LOOKUP_RCU;
962 if (!(nd->flags & LOOKUP_ROOT))
965 br_read_unlock(vfsmount_lock);
970 * Follow down to the covering mount currently visible to userspace. At each
971 * point, the filesystem owning that dentry may be queried as to whether the
972 * caller is permitted to proceed or not.
974 int follow_down(struct path *path)
979 while (managed = ACCESS_ONCE(path->dentry->d_flags),
980 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
981 /* Allow the filesystem to manage the transit without i_mutex
984 * We indicate to the filesystem if someone is trying to mount
985 * something here. This gives autofs the chance to deny anyone
986 * other than its daemon the right to mount on its
989 * The filesystem may sleep at this point.
991 if (managed & DCACHE_MANAGE_TRANSIT) {
992 BUG_ON(!path->dentry->d_op);
993 BUG_ON(!path->dentry->d_op->d_manage);
994 ret = path->dentry->d_op->d_manage(
995 path->dentry, false);
997 return ret == -EISDIR ? 0 : ret;
1000 /* Transit to a mounted filesystem. */
1001 if (managed & DCACHE_MOUNTED) {
1002 struct vfsmount *mounted = lookup_mnt(path);
1007 path->mnt = mounted;
1008 path->dentry = dget(mounted->mnt_root);
1012 /* Don't handle automount points here */
1019 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1021 static void follow_mount(struct path *path)
1023 while (d_mountpoint(path->dentry)) {
1024 struct vfsmount *mounted = lookup_mnt(path);
1029 path->mnt = mounted;
1030 path->dentry = dget(mounted->mnt_root);
1034 static void follow_dotdot(struct nameidata *nd)
1039 struct dentry *old = nd->path.dentry;
1041 if (nd->path.dentry == nd->root.dentry &&
1042 nd->path.mnt == nd->root.mnt) {
1045 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1046 /* rare case of legitimate dget_parent()... */
1047 nd->path.dentry = dget_parent(nd->path.dentry);
1051 if (!follow_up(&nd->path))
1054 follow_mount(&nd->path);
1055 nd->inode = nd->path.dentry->d_inode;
1059 * Allocate a dentry with name and parent, and perform a parent
1060 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1061 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1062 * have verified that no child exists while under i_mutex.
1064 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1065 struct qstr *name, struct nameidata *nd)
1067 struct inode *inode = parent->d_inode;
1068 struct dentry *dentry;
1071 /* Don't create child dentry for a dead directory. */
1072 if (unlikely(IS_DEADDIR(inode)))
1073 return ERR_PTR(-ENOENT);
1075 dentry = d_alloc(parent, name);
1076 if (unlikely(!dentry))
1077 return ERR_PTR(-ENOMEM);
1079 old = inode->i_op->lookup(inode, dentry, nd);
1080 if (unlikely(old)) {
1088 * We already have a dentry, but require a lookup to be performed on the parent
1089 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1090 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1091 * child exists while under i_mutex.
1093 static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1094 struct nameidata *nd)
1096 struct inode *inode = parent->d_inode;
1099 /* Don't create child dentry for a dead directory. */
1100 if (unlikely(IS_DEADDIR(inode)))
1101 return ERR_PTR(-ENOENT);
1103 old = inode->i_op->lookup(inode, dentry, nd);
1104 if (unlikely(old)) {
1112 * It's more convoluted than I'd like it to be, but... it's still fairly
1113 * small and for now I'd prefer to have fast path as straight as possible.
1114 * It _is_ time-critical.
1116 static int do_lookup(struct nameidata *nd, struct qstr *name,
1117 struct path *path, struct inode **inode)
1119 struct vfsmount *mnt = nd->path.mnt;
1120 struct dentry *dentry, *parent = nd->path.dentry;
1126 * Rename seqlock is not required here because in the off chance
1127 * of a false negative due to a concurrent rename, we're going to
1128 * do the non-racy lookup, below.
1130 if (nd->flags & LOOKUP_RCU) {
1133 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1137 /* Memory barrier in read_seqcount_begin of child is enough */
1138 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1142 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1143 status = d_revalidate(dentry, nd);
1144 if (unlikely(status <= 0)) {
1145 if (status != -ECHILD)
1150 if (unlikely(d_need_lookup(dentry)))
1153 path->dentry = dentry;
1154 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1156 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1160 if (unlazy_walk(nd, dentry))
1163 dentry = __d_lookup(parent, name);
1166 if (dentry && unlikely(d_need_lookup(dentry))) {
1171 if (unlikely(!dentry)) {
1172 struct inode *dir = parent->d_inode;
1173 BUG_ON(nd->inode != dir);
1175 mutex_lock(&dir->i_mutex);
1176 dentry = d_lookup(parent, name);
1177 if (likely(!dentry)) {
1178 dentry = d_alloc_and_lookup(parent, name, nd);
1179 if (IS_ERR(dentry)) {
1180 mutex_unlock(&dir->i_mutex);
1181 return PTR_ERR(dentry);
1186 } else if (unlikely(d_need_lookup(dentry))) {
1187 dentry = d_inode_lookup(parent, dentry, nd);
1188 if (IS_ERR(dentry)) {
1189 mutex_unlock(&dir->i_mutex);
1190 return PTR_ERR(dentry);
1196 mutex_unlock(&dir->i_mutex);
1198 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1199 status = d_revalidate(dentry, nd);
1200 if (unlikely(status <= 0)) {
1205 if (!d_invalidate(dentry)) {
1214 path->dentry = dentry;
1215 err = follow_managed(path, nd->flags);
1216 if (unlikely(err < 0)) {
1217 path_put_conditional(path, nd);
1220 *inode = path->dentry->d_inode;
1224 static inline int may_lookup(struct nameidata *nd)
1226 if (nd->flags & LOOKUP_RCU) {
1227 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1230 if (unlazy_walk(nd, NULL))
1233 return inode_permission(nd->inode, MAY_EXEC);
1236 static inline int handle_dots(struct nameidata *nd, int type)
1238 if (type == LAST_DOTDOT) {
1239 if (nd->flags & LOOKUP_RCU) {
1240 if (follow_dotdot_rcu(nd))
1248 static void terminate_walk(struct nameidata *nd)
1250 if (!(nd->flags & LOOKUP_RCU)) {
1251 path_put(&nd->path);
1253 nd->flags &= ~LOOKUP_RCU;
1254 if (!(nd->flags & LOOKUP_ROOT))
1255 nd->root.mnt = NULL;
1257 br_read_unlock(vfsmount_lock);
1262 * Do we need to follow links? We _really_ want to be able
1263 * to do this check without having to look at inode->i_op,
1264 * so we keep a cache of "no, this doesn't need follow_link"
1265 * for the common case.
1267 static inline int should_follow_link(struct inode *inode, int follow)
1269 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1270 if (likely(inode->i_op->follow_link))
1273 /* This gets set once for the inode lifetime */
1274 spin_lock(&inode->i_lock);
1275 inode->i_opflags |= IOP_NOFOLLOW;
1276 spin_unlock(&inode->i_lock);
1281 static inline int walk_component(struct nameidata *nd, struct path *path,
1282 struct qstr *name, int type, int follow)
1284 struct inode *inode;
1287 * "." and ".." are special - ".." especially so because it has
1288 * to be able to know about the current root directory and
1289 * parent relationships.
1291 if (unlikely(type != LAST_NORM))
1292 return handle_dots(nd, type);
1293 err = do_lookup(nd, name, path, &inode);
1294 if (unlikely(err)) {
1299 path_to_nameidata(path, nd);
1303 if (should_follow_link(inode, follow)) {
1304 if (nd->flags & LOOKUP_RCU) {
1305 if (unlikely(unlazy_walk(nd, path->dentry))) {
1310 BUG_ON(inode != path->dentry->d_inode);
1313 path_to_nameidata(path, nd);
1319 * This limits recursive symlink follows to 8, while
1320 * limiting consecutive symlinks to 40.
1322 * Without that kind of total limit, nasty chains of consecutive
1323 * symlinks can cause almost arbitrarily long lookups.
1325 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1329 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1330 path_put_conditional(path, nd);
1331 path_put(&nd->path);
1334 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1337 current->link_count++;
1340 struct path link = *path;
1343 res = follow_link(&link, nd, &cookie);
1345 res = walk_component(nd, path, &nd->last,
1346 nd->last_type, LOOKUP_FOLLOW);
1347 put_link(nd, &link, cookie);
1350 current->link_count--;
1356 * We really don't want to look at inode->i_op->lookup
1357 * when we don't have to. So we keep a cache bit in
1358 * the inode ->i_opflags field that says "yes, we can
1359 * do lookup on this inode".
1361 static inline int can_lookup(struct inode *inode)
1363 if (likely(inode->i_opflags & IOP_LOOKUP))
1365 if (likely(!inode->i_op->lookup))
1368 /* We do this once for the lifetime of the inode */
1369 spin_lock(&inode->i_lock);
1370 inode->i_opflags |= IOP_LOOKUP;
1371 spin_unlock(&inode->i_lock);
1377 * This is the basic name resolution function, turning a pathname into
1378 * the final dentry. We expect 'base' to be positive and a directory.
1380 * Returns 0 and nd will have valid dentry and mnt on success.
1381 * Returns error and drops reference to input namei data on failure.
1383 static int link_path_walk(const char *name, struct nameidata *nd)
1393 /* At this point we know we have a real path component. */
1400 err = may_lookup(nd);
1405 c = *(const unsigned char *)name;
1407 hash = init_name_hash();
1410 hash = partial_name_hash(c, hash);
1411 c = *(const unsigned char *)name;
1412 } while (c && (c != '/'));
1413 this.len = name - (const char *) this.name;
1414 this.hash = end_name_hash(hash);
1417 if (this.name[0] == '.') switch (this.len) {
1419 if (this.name[1] == '.') {
1421 nd->flags |= LOOKUP_JUMPED;
1427 if (likely(type == LAST_NORM)) {
1428 struct dentry *parent = nd->path.dentry;
1429 nd->flags &= ~LOOKUP_JUMPED;
1430 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1431 err = parent->d_op->d_hash(parent, nd->inode,
1438 /* remove trailing slashes? */
1440 goto last_component;
1441 while (*++name == '/');
1443 goto last_component;
1445 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1450 err = nested_symlink(&next, nd);
1454 if (can_lookup(nd->inode))
1458 /* here ends the main loop */
1462 nd->last_type = type;
1469 static int path_init(int dfd, const char *name, unsigned int flags,
1470 struct nameidata *nd, struct file **fp)
1476 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1477 nd->flags = flags | LOOKUP_JUMPED;
1479 if (flags & LOOKUP_ROOT) {
1480 struct inode *inode = nd->root.dentry->d_inode;
1482 if (!inode->i_op->lookup)
1484 retval = inode_permission(inode, MAY_EXEC);
1488 nd->path = nd->root;
1490 if (flags & LOOKUP_RCU) {
1491 br_read_lock(vfsmount_lock);
1493 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1495 path_get(&nd->path);
1500 nd->root.mnt = NULL;
1503 if (flags & LOOKUP_RCU) {
1504 br_read_lock(vfsmount_lock);
1509 path_get(&nd->root);
1511 nd->path = nd->root;
1512 } else if (dfd == AT_FDCWD) {
1513 if (flags & LOOKUP_RCU) {
1514 struct fs_struct *fs = current->fs;
1517 br_read_lock(vfsmount_lock);
1521 seq = read_seqcount_begin(&fs->seq);
1523 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1524 } while (read_seqcount_retry(&fs->seq, seq));
1526 get_fs_pwd(current->fs, &nd->path);
1529 struct dentry *dentry;
1531 file = fget_raw_light(dfd, &fput_needed);
1536 dentry = file->f_path.dentry;
1540 if (!S_ISDIR(dentry->d_inode->i_mode))
1543 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1548 nd->path = file->f_path;
1549 if (flags & LOOKUP_RCU) {
1552 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1553 br_read_lock(vfsmount_lock);
1556 path_get(&file->f_path);
1557 fput_light(file, fput_needed);
1561 nd->inode = nd->path.dentry->d_inode;
1565 fput_light(file, fput_needed);
1570 static inline int lookup_last(struct nameidata *nd, struct path *path)
1572 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1573 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1575 nd->flags &= ~LOOKUP_PARENT;
1576 return walk_component(nd, path, &nd->last, nd->last_type,
1577 nd->flags & LOOKUP_FOLLOW);
1580 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1581 static int path_lookupat(int dfd, const char *name,
1582 unsigned int flags, struct nameidata *nd)
1584 struct file *base = NULL;
1589 * Path walking is largely split up into 2 different synchronisation
1590 * schemes, rcu-walk and ref-walk (explained in
1591 * Documentation/filesystems/path-lookup.txt). These share much of the
1592 * path walk code, but some things particularly setup, cleanup, and
1593 * following mounts are sufficiently divergent that functions are
1594 * duplicated. Typically there is a function foo(), and its RCU
1595 * analogue, foo_rcu().
1597 * -ECHILD is the error number of choice (just to avoid clashes) that
1598 * is returned if some aspect of an rcu-walk fails. Such an error must
1599 * be handled by restarting a traditional ref-walk (which will always
1600 * be able to complete).
1602 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1607 current->total_link_count = 0;
1608 err = link_path_walk(name, nd);
1610 if (!err && !(flags & LOOKUP_PARENT)) {
1611 err = lookup_last(nd, &path);
1614 struct path link = path;
1615 nd->flags |= LOOKUP_PARENT;
1616 err = follow_link(&link, nd, &cookie);
1618 err = lookup_last(nd, &path);
1619 put_link(nd, &link, cookie);
1624 err = complete_walk(nd);
1626 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1627 if (!nd->inode->i_op->lookup) {
1628 path_put(&nd->path);
1636 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1637 path_put(&nd->root);
1638 nd->root.mnt = NULL;
1643 static int do_path_lookup(int dfd, const char *name,
1644 unsigned int flags, struct nameidata *nd)
1646 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1647 if (unlikely(retval == -ECHILD))
1648 retval = path_lookupat(dfd, name, flags, nd);
1649 if (unlikely(retval == -ESTALE))
1650 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1652 if (likely(!retval)) {
1653 if (unlikely(!audit_dummy_context())) {
1654 if (nd->path.dentry && nd->inode)
1655 audit_inode(name, nd->path.dentry);
1661 int kern_path_parent(const char *name, struct nameidata *nd)
1663 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1666 int kern_path(const char *name, unsigned int flags, struct path *path)
1668 struct nameidata nd;
1669 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1676 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1677 * @dentry: pointer to dentry of the base directory
1678 * @mnt: pointer to vfs mount of the base directory
1679 * @name: pointer to file name
1680 * @flags: lookup flags
1681 * @path: pointer to struct path to fill
1683 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1684 const char *name, unsigned int flags,
1687 struct nameidata nd;
1689 nd.root.dentry = dentry;
1691 BUG_ON(flags & LOOKUP_PARENT);
1692 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1693 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1699 static struct dentry *__lookup_hash(struct qstr *name,
1700 struct dentry *base, struct nameidata *nd)
1702 struct inode *inode = base->d_inode;
1703 struct dentry *dentry;
1706 err = inode_permission(inode, MAY_EXEC);
1708 return ERR_PTR(err);
1711 * Don't bother with __d_lookup: callers are for creat as
1712 * well as unlink, so a lot of the time it would cost
1715 dentry = d_lookup(base, name);
1717 if (dentry && d_need_lookup(dentry)) {
1719 * __lookup_hash is called with the parent dir's i_mutex already
1720 * held, so we are good to go here.
1722 dentry = d_inode_lookup(base, dentry, nd);
1727 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1728 int status = d_revalidate(dentry, nd);
1729 if (unlikely(status <= 0)) {
1731 * The dentry failed validation.
1732 * If d_revalidate returned 0 attempt to invalidate
1733 * the dentry otherwise d_revalidate is asking us
1734 * to return a fail status.
1738 return ERR_PTR(status);
1739 } else if (!d_invalidate(dentry)) {
1747 dentry = d_alloc_and_lookup(base, name, nd);
1753 * Restricted form of lookup. Doesn't follow links, single-component only,
1754 * needs parent already locked. Doesn't follow mounts.
1757 static struct dentry *lookup_hash(struct nameidata *nd)
1759 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1763 * lookup_one_len - filesystem helper to lookup single pathname component
1764 * @name: pathname component to lookup
1765 * @base: base directory to lookup from
1766 * @len: maximum length @len should be interpreted to
1768 * Note that this routine is purely a helper for filesystem usage and should
1769 * not be called by generic code. Also note that by using this function the
1770 * nameidata argument is passed to the filesystem methods and a filesystem
1771 * using this helper needs to be prepared for that.
1773 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1779 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1784 return ERR_PTR(-EACCES);
1786 hash = init_name_hash();
1788 c = *(const unsigned char *)name++;
1789 if (c == '/' || c == '\0')
1790 return ERR_PTR(-EACCES);
1791 hash = partial_name_hash(c, hash);
1793 this.hash = end_name_hash(hash);
1795 * See if the low-level filesystem might want
1796 * to use its own hash..
1798 if (base->d_flags & DCACHE_OP_HASH) {
1799 int err = base->d_op->d_hash(base, base->d_inode, &this);
1801 return ERR_PTR(err);
1804 return __lookup_hash(&this, base, NULL);
1807 int user_path_at(int dfd, const char __user *name, unsigned flags,
1810 struct nameidata nd;
1811 char *tmp = getname_flags(name, flags);
1812 int err = PTR_ERR(tmp);
1815 BUG_ON(flags & LOOKUP_PARENT);
1817 err = do_path_lookup(dfd, tmp, flags, &nd);
1825 static int user_path_parent(int dfd, const char __user *path,
1826 struct nameidata *nd, char **name)
1828 char *s = getname(path);
1834 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1844 * It's inline, so penalty for filesystems that don't use sticky bit is
1847 static inline int check_sticky(struct inode *dir, struct inode *inode)
1849 uid_t fsuid = current_fsuid();
1851 if (!(dir->i_mode & S_ISVTX))
1853 if (current_user_ns() != inode_userns(inode))
1855 if (inode->i_uid == fsuid)
1857 if (dir->i_uid == fsuid)
1861 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1865 * Check whether we can remove a link victim from directory dir, check
1866 * whether the type of victim is right.
1867 * 1. We can't do it if dir is read-only (done in permission())
1868 * 2. We should have write and exec permissions on dir
1869 * 3. We can't remove anything from append-only dir
1870 * 4. We can't do anything with immutable dir (done in permission())
1871 * 5. If the sticky bit on dir is set we should either
1872 * a. be owner of dir, or
1873 * b. be owner of victim, or
1874 * c. have CAP_FOWNER capability
1875 * 6. If the victim is append-only or immutable we can't do antyhing with
1876 * links pointing to it.
1877 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1878 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1879 * 9. We can't remove a root or mountpoint.
1880 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1881 * nfs_async_unlink().
1883 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1887 if (!victim->d_inode)
1890 BUG_ON(victim->d_parent->d_inode != dir);
1891 audit_inode_child(victim, dir);
1893 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1898 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1899 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1902 if (!S_ISDIR(victim->d_inode->i_mode))
1904 if (IS_ROOT(victim))
1906 } else if (S_ISDIR(victim->d_inode->i_mode))
1908 if (IS_DEADDIR(dir))
1910 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1915 /* Check whether we can create an object with dentry child in directory
1917 * 1. We can't do it if child already exists (open has special treatment for
1918 * this case, but since we are inlined it's OK)
1919 * 2. We can't do it if dir is read-only (done in permission())
1920 * 3. We should have write and exec permissions on dir
1921 * 4. We can't do it if dir is immutable (done in permission())
1923 static inline int may_create(struct inode *dir, struct dentry *child)
1927 if (IS_DEADDIR(dir))
1929 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1933 * p1 and p2 should be directories on the same fs.
1935 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1940 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1944 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1946 p = d_ancestor(p2, p1);
1948 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1949 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1953 p = d_ancestor(p1, p2);
1955 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1956 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1960 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1961 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1965 void unlock_rename(struct dentry *p1, struct dentry *p2)
1967 mutex_unlock(&p1->d_inode->i_mutex);
1969 mutex_unlock(&p2->d_inode->i_mutex);
1970 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1974 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1975 struct nameidata *nd)
1977 int error = may_create(dir, dentry);
1982 if (!dir->i_op->create)
1983 return -EACCES; /* shouldn't it be ENOSYS? */
1986 error = security_inode_create(dir, dentry, mode);
1989 error = dir->i_op->create(dir, dentry, mode, nd);
1991 fsnotify_create(dir, dentry);
1995 static int may_open(struct path *path, int acc_mode, int flag)
1997 struct dentry *dentry = path->dentry;
1998 struct inode *inode = dentry->d_inode;
2008 switch (inode->i_mode & S_IFMT) {
2012 if (acc_mode & MAY_WRITE)
2017 if (path->mnt->mnt_flags & MNT_NODEV)
2026 error = inode_permission(inode, acc_mode);
2031 * An append-only file must be opened in append mode for writing.
2033 if (IS_APPEND(inode)) {
2034 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2040 /* O_NOATIME can only be set by the owner or superuser */
2041 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2045 * Ensure there are no outstanding leases on the file.
2047 return break_lease(inode, flag);
2050 static int handle_truncate(struct file *filp)
2052 struct path *path = &filp->f_path;
2053 struct inode *inode = path->dentry->d_inode;
2054 int error = get_write_access(inode);
2058 * Refuse to truncate files with mandatory locks held on them.
2060 error = locks_verify_locked(inode);
2062 error = security_path_truncate(path);
2064 error = do_truncate(path->dentry, 0,
2065 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2068 put_write_access(inode);
2072 static inline int open_to_namei_flags(int flag)
2074 if ((flag & O_ACCMODE) == 3)
2080 * Handle the last step of open()
2082 static struct file *do_last(struct nameidata *nd, struct path *path,
2083 const struct open_flags *op, const char *pathname)
2085 struct dentry *dir = nd->path.dentry;
2086 struct dentry *dentry;
2087 int open_flag = op->open_flag;
2088 int will_truncate = open_flag & O_TRUNC;
2090 int acc_mode = op->acc_mode;
2094 nd->flags &= ~LOOKUP_PARENT;
2095 nd->flags |= op->intent;
2097 switch (nd->last_type) {
2100 error = handle_dots(nd, nd->last_type);
2102 return ERR_PTR(error);
2105 error = complete_walk(nd);
2107 return ERR_PTR(error);
2108 audit_inode(pathname, nd->path.dentry);
2109 if (open_flag & O_CREAT) {
2115 error = complete_walk(nd);
2117 return ERR_PTR(error);
2118 audit_inode(pathname, dir);
2122 if (!(open_flag & O_CREAT)) {
2124 if (nd->last.name[nd->last.len])
2125 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2126 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2128 /* we _can_ be in RCU mode here */
2129 error = walk_component(nd, path, &nd->last, LAST_NORM,
2132 return ERR_PTR(error);
2133 if (error) /* symlink */
2136 error = complete_walk(nd);
2138 return ERR_PTR(-ECHILD);
2141 if (nd->flags & LOOKUP_DIRECTORY) {
2142 if (!nd->inode->i_op->lookup)
2145 audit_inode(pathname, nd->path.dentry);
2149 /* create side of things */
2150 error = complete_walk(nd);
2152 return ERR_PTR(error);
2154 audit_inode(pathname, dir);
2156 /* trailing slashes? */
2157 if (nd->last.name[nd->last.len])
2160 mutex_lock(&dir->d_inode->i_mutex);
2162 dentry = lookup_hash(nd);
2163 error = PTR_ERR(dentry);
2164 if (IS_ERR(dentry)) {
2165 mutex_unlock(&dir->d_inode->i_mutex);
2169 path->dentry = dentry;
2170 path->mnt = nd->path.mnt;
2172 /* Negative dentry, just create the file */
2173 if (!dentry->d_inode) {
2174 int mode = op->mode;
2175 if (!IS_POSIXACL(dir->d_inode))
2176 mode &= ~current_umask();
2178 * This write is needed to ensure that a
2179 * rw->ro transition does not occur between
2180 * the time when the file is created and when
2181 * a permanent write count is taken through
2182 * the 'struct file' in nameidata_to_filp().
2184 error = mnt_want_write(nd->path.mnt);
2186 goto exit_mutex_unlock;
2188 /* Don't check for write permission, don't truncate */
2189 open_flag &= ~O_TRUNC;
2191 acc_mode = MAY_OPEN;
2192 error = security_path_mknod(&nd->path, dentry, mode, 0);
2194 goto exit_mutex_unlock;
2195 error = vfs_create(dir->d_inode, dentry, mode, nd);
2197 goto exit_mutex_unlock;
2198 mutex_unlock(&dir->d_inode->i_mutex);
2199 dput(nd->path.dentry);
2200 nd->path.dentry = dentry;
2205 * It already exists.
2207 mutex_unlock(&dir->d_inode->i_mutex);
2208 audit_inode(pathname, path->dentry);
2211 if (open_flag & O_EXCL)
2214 error = follow_managed(path, nd->flags);
2219 if (!path->dentry->d_inode)
2222 if (path->dentry->d_inode->i_op->follow_link)
2225 path_to_nameidata(path, nd);
2226 nd->inode = path->dentry->d_inode;
2228 if (S_ISDIR(nd->inode->i_mode))
2231 if (!S_ISREG(nd->inode->i_mode))
2234 if (will_truncate) {
2235 error = mnt_want_write(nd->path.mnt);
2241 error = may_open(&nd->path, acc_mode, open_flag);
2244 filp = nameidata_to_filp(nd);
2245 if (!IS_ERR(filp)) {
2246 error = ima_file_check(filp, op->acc_mode);
2249 filp = ERR_PTR(error);
2252 if (!IS_ERR(filp)) {
2253 if (will_truncate) {
2254 error = handle_truncate(filp);
2257 filp = ERR_PTR(error);
2263 mnt_drop_write(nd->path.mnt);
2264 path_put(&nd->path);
2268 mutex_unlock(&dir->d_inode->i_mutex);
2270 path_put_conditional(path, nd);
2272 filp = ERR_PTR(error);
2276 static struct file *path_openat(int dfd, const char *pathname,
2277 struct nameidata *nd, const struct open_flags *op, int flags)
2279 struct file *base = NULL;
2284 filp = get_empty_filp();
2286 return ERR_PTR(-ENFILE);
2288 filp->f_flags = op->open_flag;
2289 nd->intent.open.file = filp;
2290 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2291 nd->intent.open.create_mode = op->mode;
2293 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2294 if (unlikely(error))
2297 current->total_link_count = 0;
2298 error = link_path_walk(pathname, nd);
2299 if (unlikely(error))
2302 filp = do_last(nd, &path, op, pathname);
2303 while (unlikely(!filp)) { /* trailing symlink */
2304 struct path link = path;
2306 if (!(nd->flags & LOOKUP_FOLLOW)) {
2307 path_put_conditional(&path, nd);
2308 path_put(&nd->path);
2309 filp = ERR_PTR(-ELOOP);
2312 nd->flags |= LOOKUP_PARENT;
2313 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2314 error = follow_link(&link, nd, &cookie);
2315 if (unlikely(error))
2316 filp = ERR_PTR(error);
2318 filp = do_last(nd, &path, op, pathname);
2319 put_link(nd, &link, cookie);
2322 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2323 path_put(&nd->root);
2326 release_open_intent(nd);
2330 filp = ERR_PTR(error);
2334 struct file *do_filp_open(int dfd, const char *pathname,
2335 const struct open_flags *op, int flags)
2337 struct nameidata nd;
2340 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2341 if (unlikely(filp == ERR_PTR(-ECHILD)))
2342 filp = path_openat(dfd, pathname, &nd, op, flags);
2343 if (unlikely(filp == ERR_PTR(-ESTALE)))
2344 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2348 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2349 const char *name, const struct open_flags *op, int flags)
2351 struct nameidata nd;
2355 nd.root.dentry = dentry;
2357 flags |= LOOKUP_ROOT;
2359 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2360 return ERR_PTR(-ELOOP);
2362 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2363 if (unlikely(file == ERR_PTR(-ECHILD)))
2364 file = path_openat(-1, name, &nd, op, flags);
2365 if (unlikely(file == ERR_PTR(-ESTALE)))
2366 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2370 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2372 struct dentry *dentry = ERR_PTR(-EEXIST);
2373 struct nameidata nd;
2374 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2376 return ERR_PTR(error);
2379 * Yucky last component or no last component at all?
2380 * (foo/., foo/.., /////)
2382 if (nd.last_type != LAST_NORM)
2384 nd.flags &= ~LOOKUP_PARENT;
2385 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2386 nd.intent.open.flags = O_EXCL;
2389 * Do the final lookup.
2391 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2392 dentry = lookup_hash(&nd);
2396 if (dentry->d_inode)
2399 * Special case - lookup gave negative, but... we had foo/bar/
2400 * From the vfs_mknod() POV we just have a negative dentry -
2401 * all is fine. Let's be bastards - you had / on the end, you've
2402 * been asking for (non-existent) directory. -ENOENT for you.
2404 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2406 dentry = ERR_PTR(-ENOENT);
2413 dentry = ERR_PTR(-EEXIST);
2415 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2420 EXPORT_SYMBOL(kern_path_create);
2422 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2424 char *tmp = getname(pathname);
2427 return ERR_CAST(tmp);
2428 res = kern_path_create(dfd, tmp, path, is_dir);
2432 EXPORT_SYMBOL(user_path_create);
2434 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2436 int error = may_create(dir, dentry);
2441 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2442 !ns_capable(inode_userns(dir), CAP_MKNOD))
2445 if (!dir->i_op->mknod)
2448 error = devcgroup_inode_mknod(mode, dev);
2452 error = security_inode_mknod(dir, dentry, mode, dev);
2456 error = dir->i_op->mknod(dir, dentry, mode, dev);
2458 fsnotify_create(dir, dentry);
2462 static int may_mknod(mode_t mode)
2464 switch (mode & S_IFMT) {
2470 case 0: /* zero mode translates to S_IFREG */
2479 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2482 struct dentry *dentry;
2489 dentry = user_path_create(dfd, filename, &path, 0);
2491 return PTR_ERR(dentry);
2493 if (!IS_POSIXACL(path.dentry->d_inode))
2494 mode &= ~current_umask();
2495 error = may_mknod(mode);
2498 error = mnt_want_write(path.mnt);
2501 error = security_path_mknod(&path, dentry, mode, dev);
2503 goto out_drop_write;
2504 switch (mode & S_IFMT) {
2505 case 0: case S_IFREG:
2506 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2508 case S_IFCHR: case S_IFBLK:
2509 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2510 new_decode_dev(dev));
2512 case S_IFIFO: case S_IFSOCK:
2513 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2517 mnt_drop_write(path.mnt);
2520 mutex_unlock(&path.dentry->d_inode->i_mutex);
2526 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2528 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2531 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2533 int error = may_create(dir, dentry);
2538 if (!dir->i_op->mkdir)
2541 mode &= (S_IRWXUGO|S_ISVTX);
2542 error = security_inode_mkdir(dir, dentry, mode);
2546 error = dir->i_op->mkdir(dir, dentry, mode);
2548 fsnotify_mkdir(dir, dentry);
2552 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2554 struct dentry *dentry;
2558 dentry = user_path_create(dfd, pathname, &path, 1);
2560 return PTR_ERR(dentry);
2562 if (!IS_POSIXACL(path.dentry->d_inode))
2563 mode &= ~current_umask();
2564 error = mnt_want_write(path.mnt);
2567 error = security_path_mkdir(&path, dentry, mode);
2569 goto out_drop_write;
2570 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2572 mnt_drop_write(path.mnt);
2575 mutex_unlock(&path.dentry->d_inode->i_mutex);
2580 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2582 return sys_mkdirat(AT_FDCWD, pathname, mode);
2586 * The dentry_unhash() helper will try to drop the dentry early: we
2587 * should have a usage count of 2 if we're the only user of this
2588 * dentry, and if that is true (possibly after pruning the dcache),
2589 * then we drop the dentry now.
2591 * A low-level filesystem can, if it choses, legally
2594 * if (!d_unhashed(dentry))
2597 * if it cannot handle the case of removing a directory
2598 * that is still in use by something else..
2600 void dentry_unhash(struct dentry *dentry)
2602 shrink_dcache_parent(dentry);
2603 spin_lock(&dentry->d_lock);
2604 if (dentry->d_count == 1)
2606 spin_unlock(&dentry->d_lock);
2609 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2611 int error = may_delete(dir, dentry, 1);
2616 if (!dir->i_op->rmdir)
2620 mutex_lock(&dentry->d_inode->i_mutex);
2623 if (d_mountpoint(dentry))
2626 error = security_inode_rmdir(dir, dentry);
2630 shrink_dcache_parent(dentry);
2631 error = dir->i_op->rmdir(dir, dentry);
2635 dentry->d_inode->i_flags |= S_DEAD;
2639 mutex_unlock(&dentry->d_inode->i_mutex);
2646 static long do_rmdir(int dfd, const char __user *pathname)
2650 struct dentry *dentry;
2651 struct nameidata nd;
2653 error = user_path_parent(dfd, pathname, &nd, &name);
2657 switch(nd.last_type) {
2669 nd.flags &= ~LOOKUP_PARENT;
2671 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2672 dentry = lookup_hash(&nd);
2673 error = PTR_ERR(dentry);
2676 if (!dentry->d_inode) {
2680 error = mnt_want_write(nd.path.mnt);
2683 error = security_path_rmdir(&nd.path, dentry);
2686 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2688 mnt_drop_write(nd.path.mnt);
2692 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2699 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2701 return do_rmdir(AT_FDCWD, pathname);
2704 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2706 int error = may_delete(dir, dentry, 0);
2711 if (!dir->i_op->unlink)
2714 mutex_lock(&dentry->d_inode->i_mutex);
2715 if (d_mountpoint(dentry))
2718 error = security_inode_unlink(dir, dentry);
2720 error = dir->i_op->unlink(dir, dentry);
2725 mutex_unlock(&dentry->d_inode->i_mutex);
2727 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2728 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2729 fsnotify_link_count(dentry->d_inode);
2737 * Make sure that the actual truncation of the file will occur outside its
2738 * directory's i_mutex. Truncate can take a long time if there is a lot of
2739 * writeout happening, and we don't want to prevent access to the directory
2740 * while waiting on the I/O.
2742 static long do_unlinkat(int dfd, const char __user *pathname)
2746 struct dentry *dentry;
2747 struct nameidata nd;
2748 struct inode *inode = NULL;
2750 error = user_path_parent(dfd, pathname, &nd, &name);
2755 if (nd.last_type != LAST_NORM)
2758 nd.flags &= ~LOOKUP_PARENT;
2760 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2761 dentry = lookup_hash(&nd);
2762 error = PTR_ERR(dentry);
2763 if (!IS_ERR(dentry)) {
2764 /* Why not before? Because we want correct error value */
2765 if (nd.last.name[nd.last.len])
2767 inode = dentry->d_inode;
2771 error = mnt_want_write(nd.path.mnt);
2774 error = security_path_unlink(&nd.path, dentry);
2777 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2779 mnt_drop_write(nd.path.mnt);
2783 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2785 iput(inode); /* truncate the inode here */
2792 error = !dentry->d_inode ? -ENOENT :
2793 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2797 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2799 if ((flag & ~AT_REMOVEDIR) != 0)
2802 if (flag & AT_REMOVEDIR)
2803 return do_rmdir(dfd, pathname);
2805 return do_unlinkat(dfd, pathname);
2808 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2810 return do_unlinkat(AT_FDCWD, pathname);
2813 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2815 int error = may_create(dir, dentry);
2820 if (!dir->i_op->symlink)
2823 error = security_inode_symlink(dir, dentry, oldname);
2827 error = dir->i_op->symlink(dir, dentry, oldname);
2829 fsnotify_create(dir, dentry);
2833 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2834 int, newdfd, const char __user *, newname)
2838 struct dentry *dentry;
2841 from = getname(oldname);
2843 return PTR_ERR(from);
2845 dentry = user_path_create(newdfd, newname, &path, 0);
2846 error = PTR_ERR(dentry);
2850 error = mnt_want_write(path.mnt);
2853 error = security_path_symlink(&path, dentry, from);
2855 goto out_drop_write;
2856 error = vfs_symlink(path.dentry->d_inode, dentry, from);
2858 mnt_drop_write(path.mnt);
2861 mutex_unlock(&path.dentry->d_inode->i_mutex);
2868 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2870 return sys_symlinkat(oldname, AT_FDCWD, newname);
2873 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2875 struct inode *inode = old_dentry->d_inode;
2881 error = may_create(dir, new_dentry);
2885 if (dir->i_sb != inode->i_sb)
2889 * A link to an append-only or immutable file cannot be created.
2891 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2893 if (!dir->i_op->link)
2895 if (S_ISDIR(inode->i_mode))
2898 error = security_inode_link(old_dentry, dir, new_dentry);
2902 mutex_lock(&inode->i_mutex);
2903 /* Make sure we don't allow creating hardlink to an unlinked file */
2904 if (inode->i_nlink == 0)
2907 error = dir->i_op->link(old_dentry, dir, new_dentry);
2908 mutex_unlock(&inode->i_mutex);
2910 fsnotify_link(dir, inode, new_dentry);
2915 * Hardlinks are often used in delicate situations. We avoid
2916 * security-related surprises by not following symlinks on the
2919 * We don't follow them on the oldname either to be compatible
2920 * with linux 2.0, and to avoid hard-linking to directories
2921 * and other special files. --ADM
2923 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2924 int, newdfd, const char __user *, newname, int, flags)
2926 struct dentry *new_dentry;
2927 struct path old_path, new_path;
2931 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
2934 * To use null names we require CAP_DAC_READ_SEARCH
2935 * This ensures that not everyone will be able to create
2936 * handlink using the passed filedescriptor.
2938 if (flags & AT_EMPTY_PATH) {
2939 if (!capable(CAP_DAC_READ_SEARCH))
2944 if (flags & AT_SYMLINK_FOLLOW)
2945 how |= LOOKUP_FOLLOW;
2947 error = user_path_at(olddfd, oldname, how, &old_path);
2951 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
2952 error = PTR_ERR(new_dentry);
2953 if (IS_ERR(new_dentry))
2957 if (old_path.mnt != new_path.mnt)
2959 error = mnt_want_write(new_path.mnt);
2962 error = security_path_link(old_path.dentry, &new_path, new_dentry);
2964 goto out_drop_write;
2965 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
2967 mnt_drop_write(new_path.mnt);
2970 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
2971 path_put(&new_path);
2973 path_put(&old_path);
2978 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2980 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2984 * The worst of all namespace operations - renaming directory. "Perverted"
2985 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2987 * a) we can get into loop creation. Check is done in is_subdir().
2988 * b) race potential - two innocent renames can create a loop together.
2989 * That's where 4.4 screws up. Current fix: serialization on
2990 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2992 * c) we have to lock _three_ objects - parents and victim (if it exists).
2993 * And that - after we got ->i_mutex on parents (until then we don't know
2994 * whether the target exists). Solution: try to be smart with locking
2995 * order for inodes. We rely on the fact that tree topology may change
2996 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2997 * move will be locked. Thus we can rank directories by the tree
2998 * (ancestors first) and rank all non-directories after them.
2999 * That works since everybody except rename does "lock parent, lookup,
3000 * lock child" and rename is under ->s_vfs_rename_mutex.
3001 * HOWEVER, it relies on the assumption that any object with ->lookup()
3002 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3003 * we'd better make sure that there's no link(2) for them.
3004 * d) conversion from fhandle to dentry may come in the wrong moment - when
3005 * we are removing the target. Solution: we will have to grab ->i_mutex
3006 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3007 * ->i_mutex on parents, which works but leads to some truly excessive
3010 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3011 struct inode *new_dir, struct dentry *new_dentry)
3014 struct inode *target = new_dentry->d_inode;
3017 * If we are going to change the parent - check write permissions,
3018 * we'll need to flip '..'.
3020 if (new_dir != old_dir) {
3021 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3026 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3032 mutex_lock(&target->i_mutex);
3035 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3039 shrink_dcache_parent(new_dentry);
3040 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3045 target->i_flags |= S_DEAD;
3046 dont_mount(new_dentry);
3050 mutex_unlock(&target->i_mutex);
3053 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3054 d_move(old_dentry,new_dentry);
3058 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3059 struct inode *new_dir, struct dentry *new_dentry)
3061 struct inode *target = new_dentry->d_inode;
3064 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3070 mutex_lock(&target->i_mutex);
3073 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3076 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3081 dont_mount(new_dentry);
3082 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3083 d_move(old_dentry, new_dentry);
3086 mutex_unlock(&target->i_mutex);
3091 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3092 struct inode *new_dir, struct dentry *new_dentry)
3095 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3096 const unsigned char *old_name;
3098 if (old_dentry->d_inode == new_dentry->d_inode)
3101 error = may_delete(old_dir, old_dentry, is_dir);
3105 if (!new_dentry->d_inode)
3106 error = may_create(new_dir, new_dentry);
3108 error = may_delete(new_dir, new_dentry, is_dir);
3112 if (!old_dir->i_op->rename)
3115 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3118 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3120 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3122 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3123 new_dentry->d_inode, old_dentry);
3124 fsnotify_oldname_free(old_name);
3129 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3130 int, newdfd, const char __user *, newname)
3132 struct dentry *old_dir, *new_dir;
3133 struct dentry *old_dentry, *new_dentry;
3134 struct dentry *trap;
3135 struct nameidata oldnd, newnd;
3140 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3144 error = user_path_parent(newdfd, newname, &newnd, &to);
3149 if (oldnd.path.mnt != newnd.path.mnt)
3152 old_dir = oldnd.path.dentry;
3154 if (oldnd.last_type != LAST_NORM)
3157 new_dir = newnd.path.dentry;
3158 if (newnd.last_type != LAST_NORM)
3161 oldnd.flags &= ~LOOKUP_PARENT;
3162 newnd.flags &= ~LOOKUP_PARENT;
3163 newnd.flags |= LOOKUP_RENAME_TARGET;
3165 trap = lock_rename(new_dir, old_dir);
3167 old_dentry = lookup_hash(&oldnd);
3168 error = PTR_ERR(old_dentry);
3169 if (IS_ERR(old_dentry))
3171 /* source must exist */
3173 if (!old_dentry->d_inode)
3175 /* unless the source is a directory trailing slashes give -ENOTDIR */
3176 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3178 if (oldnd.last.name[oldnd.last.len])
3180 if (newnd.last.name[newnd.last.len])
3183 /* source should not be ancestor of target */
3185 if (old_dentry == trap)
3187 new_dentry = lookup_hash(&newnd);
3188 error = PTR_ERR(new_dentry);
3189 if (IS_ERR(new_dentry))
3191 /* target should not be an ancestor of source */
3193 if (new_dentry == trap)
3196 error = mnt_want_write(oldnd.path.mnt);
3199 error = security_path_rename(&oldnd.path, old_dentry,
3200 &newnd.path, new_dentry);
3203 error = vfs_rename(old_dir->d_inode, old_dentry,
3204 new_dir->d_inode, new_dentry);
3206 mnt_drop_write(oldnd.path.mnt);
3212 unlock_rename(new_dir, old_dir);
3214 path_put(&newnd.path);
3217 path_put(&oldnd.path);
3223 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3225 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3228 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3232 len = PTR_ERR(link);
3237 if (len > (unsigned) buflen)
3239 if (copy_to_user(buffer, link, len))
3246 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3247 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3248 * using) it for any given inode is up to filesystem.
3250 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3252 struct nameidata nd;
3257 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3259 return PTR_ERR(cookie);
3261 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3262 if (dentry->d_inode->i_op->put_link)
3263 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3267 int vfs_follow_link(struct nameidata *nd, const char *link)
3269 return __vfs_follow_link(nd, link);
3272 /* get the link contents into pagecache */
3273 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3277 struct address_space *mapping = dentry->d_inode->i_mapping;
3278 page = read_mapping_page(mapping, 0, NULL);
3283 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3287 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3289 struct page *page = NULL;
3290 char *s = page_getlink(dentry, &page);
3291 int res = vfs_readlink(dentry,buffer,buflen,s);
3294 page_cache_release(page);
3299 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3301 struct page *page = NULL;
3302 nd_set_link(nd, page_getlink(dentry, &page));
3306 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3308 struct page *page = cookie;
3312 page_cache_release(page);
3317 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3319 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3321 struct address_space *mapping = inode->i_mapping;
3326 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3328 flags |= AOP_FLAG_NOFS;
3331 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3332 flags, &page, &fsdata);
3336 kaddr = kmap_atomic(page, KM_USER0);
3337 memcpy(kaddr, symname, len-1);
3338 kunmap_atomic(kaddr, KM_USER0);
3340 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3347 mark_inode_dirty(inode);
3353 int page_symlink(struct inode *inode, const char *symname, int len)
3355 return __page_symlink(inode, symname, len,
3356 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3359 const struct inode_operations page_symlink_inode_operations = {
3360 .readlink = generic_readlink,
3361 .follow_link = page_follow_link_light,
3362 .put_link = page_put_link,
3365 EXPORT_SYMBOL(user_path_at);
3366 EXPORT_SYMBOL(follow_down_one);
3367 EXPORT_SYMBOL(follow_down);
3368 EXPORT_SYMBOL(follow_up);
3369 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3370 EXPORT_SYMBOL(getname);
3371 EXPORT_SYMBOL(lock_rename);
3372 EXPORT_SYMBOL(lookup_one_len);
3373 EXPORT_SYMBOL(page_follow_link_light);
3374 EXPORT_SYMBOL(page_put_link);
3375 EXPORT_SYMBOL(page_readlink);
3376 EXPORT_SYMBOL(__page_symlink);
3377 EXPORT_SYMBOL(page_symlink);
3378 EXPORT_SYMBOL(page_symlink_inode_operations);
3379 EXPORT_SYMBOL(kern_path);
3380 EXPORT_SYMBOL(vfs_path_lookup);
3381 EXPORT_SYMBOL(inode_permission);
3382 EXPORT_SYMBOL(unlock_rename);
3383 EXPORT_SYMBOL(vfs_create);
3384 EXPORT_SYMBOL(vfs_follow_link);
3385 EXPORT_SYMBOL(vfs_link);
3386 EXPORT_SYMBOL(vfs_mkdir);
3387 EXPORT_SYMBOL(vfs_mknod);
3388 EXPORT_SYMBOL(generic_permission);
3389 EXPORT_SYMBOL(vfs_readlink);
3390 EXPORT_SYMBOL(vfs_rename);
3391 EXPORT_SYMBOL(vfs_rmdir);
3392 EXPORT_SYMBOL(vfs_symlink);
3393 EXPORT_SYMBOL(vfs_unlink);
3394 EXPORT_SYMBOL(dentry_unhash);
3395 EXPORT_SYMBOL(generic_readlink);