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 <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existent name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user *filename, char *page)
120 unsigned long len = PATH_MAX;
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
129 retval = strncpy_from_user(page, filename, len);
133 return -ENAMETOOLONG;
139 static char *getname_flags(const char __user * filename, int flags)
143 result = ERR_PTR(-ENOMEM);
146 int retval = do_getname(filename, tmp);
150 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
152 result = ERR_PTR(retval);
156 audit_getname(result);
160 char *getname(const char __user * filename)
162 return getname_flags(filename, 0);
165 #ifdef CONFIG_AUDITSYSCALL
166 void putname(const char *name)
168 if (unlikely(!audit_dummy_context()))
173 EXPORT_SYMBOL(putname);
177 * This does basic POSIX ACL permission checking
179 static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
180 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
182 unsigned int mode = inode->i_mode;
184 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
186 if (current_user_ns() != inode_userns(inode))
189 if (current_fsuid() == inode->i_uid)
192 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
193 int error = check_acl(inode, mask, flags);
194 if (error != -EAGAIN)
198 if (in_group_p(inode->i_gid))
204 * If the DACs are ok we don't need any capability check.
206 if ((mask & ~mode) == 0)
212 * generic_permission - check for access rights on a Posix-like filesystem
213 * @inode: inode to check access rights for
214 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
215 * @check_acl: optional callback to check for Posix ACLs
216 * @flags: IPERM_FLAG_ flags.
218 * Used to check for read/write/execute permissions on a file.
219 * We use "fsuid" for this, letting us set arbitrary permissions
220 * for filesystem access without changing the "normal" uids which
221 * are used for other things.
223 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
224 * request cannot be satisfied (eg. requires blocking or too much complexity).
225 * It would then be called again in ref-walk mode.
227 int generic_permission(struct inode *inode, int mask, unsigned int flags,
228 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
233 * Do the basic POSIX ACL permission checks.
235 ret = acl_permission_check(inode, mask, flags, check_acl);
240 * Read/write DACs are always overridable.
241 * Executable DACs are overridable if at least one exec bit is set.
243 if (!(mask & MAY_EXEC) || execute_ok(inode))
244 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
248 * Searching includes executable on directories, else just read.
250 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
251 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
252 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
259 * inode_permission - check for access rights to a given inode
260 * @inode: inode to check permission on
261 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
263 * Used to check for read/write/execute permissions on an inode.
264 * We use "fsuid" for this, letting us set arbitrary permissions
265 * for filesystem access without changing the "normal" uids which
266 * are used for other things.
268 int inode_permission(struct inode *inode, int mask)
272 if (mask & MAY_WRITE) {
273 umode_t mode = inode->i_mode;
276 * Nobody gets write access to a read-only fs.
278 if (IS_RDONLY(inode) &&
279 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
283 * Nobody gets write access to an immutable file.
285 if (IS_IMMUTABLE(inode))
289 if (inode->i_op->permission)
290 retval = inode->i_op->permission(inode, mask, 0);
292 retval = generic_permission(inode, mask, 0,
293 inode->i_op->check_acl);
298 retval = devcgroup_inode_permission(inode, mask);
302 return security_inode_permission(inode, mask);
306 * file_permission - check for additional access rights to a given file
307 * @file: file to check access rights for
308 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
310 * Used to check for read/write/execute permissions on an already opened
314 * Do not use this function in new code. All access checks should
315 * be done using inode_permission().
317 int file_permission(struct file *file, int mask)
319 return inode_permission(file->f_path.dentry->d_inode, mask);
323 * get_write_access() gets write permission for a file.
324 * put_write_access() releases this write permission.
325 * This is used for regular files.
326 * We cannot support write (and maybe mmap read-write shared) accesses and
327 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
328 * can have the following values:
329 * 0: no writers, no VM_DENYWRITE mappings
330 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
331 * > 0: (i_writecount) users are writing to the file.
333 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
334 * except for the cases where we don't hold i_writecount yet. Then we need to
335 * use {get,deny}_write_access() - these functions check the sign and refuse
336 * to do the change if sign is wrong. Exclusion between them is provided by
337 * the inode->i_lock spinlock.
340 int get_write_access(struct inode * inode)
342 spin_lock(&inode->i_lock);
343 if (atomic_read(&inode->i_writecount) < 0) {
344 spin_unlock(&inode->i_lock);
347 atomic_inc(&inode->i_writecount);
348 spin_unlock(&inode->i_lock);
353 int deny_write_access(struct file * file)
355 struct inode *inode = file->f_path.dentry->d_inode;
357 spin_lock(&inode->i_lock);
358 if (atomic_read(&inode->i_writecount) > 0) {
359 spin_unlock(&inode->i_lock);
362 atomic_dec(&inode->i_writecount);
363 spin_unlock(&inode->i_lock);
369 * path_get - get a reference to a path
370 * @path: path to get the reference to
372 * Given a path increment the reference count to the dentry and the vfsmount.
374 void path_get(struct path *path)
379 EXPORT_SYMBOL(path_get);
382 * path_put - put a reference to a path
383 * @path: path to put the reference to
385 * Given a path decrement the reference count to the dentry and the vfsmount.
387 void path_put(struct path *path)
392 EXPORT_SYMBOL(path_put);
395 * Path walking has 2 modes, rcu-walk and ref-walk (see
396 * Documentation/filesystems/path-lookup.txt). In situations when we can't
397 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
398 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
399 * mode. Refcounts are grabbed at the last known good point before rcu-walk
400 * got stuck, so ref-walk may continue from there. If this is not successful
401 * (eg. a seqcount has changed), then failure is returned and it's up to caller
402 * to restart the path walk from the beginning in ref-walk mode.
406 * unlazy_walk - try to switch to ref-walk mode.
407 * @nd: nameidata pathwalk data
408 * @dentry: child of nd->path.dentry or NULL
409 * Returns: 0 on success, -ECHILD on failure
411 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
412 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
413 * @nd or NULL. Must be called from rcu-walk context.
415 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
417 struct fs_struct *fs = current->fs;
418 struct dentry *parent = nd->path.dentry;
421 BUG_ON(!(nd->flags & LOOKUP_RCU));
422 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
424 spin_lock(&fs->lock);
425 if (nd->root.mnt != fs->root.mnt ||
426 nd->root.dentry != fs->root.dentry)
429 spin_lock(&parent->d_lock);
431 if (!__d_rcu_to_refcount(parent, nd->seq))
433 BUG_ON(nd->inode != parent->d_inode);
435 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
436 if (!__d_rcu_to_refcount(dentry, nd->seq))
439 * If the sequence check on the child dentry passed, then
440 * the child has not been removed from its parent. This
441 * means the parent dentry must be valid and able to take
442 * a reference at this point.
444 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
445 BUG_ON(!parent->d_count);
447 spin_unlock(&dentry->d_lock);
449 spin_unlock(&parent->d_lock);
452 spin_unlock(&fs->lock);
454 mntget(nd->path.mnt);
457 br_read_unlock(vfsmount_lock);
458 nd->flags &= ~LOOKUP_RCU;
462 spin_unlock(&dentry->d_lock);
464 spin_unlock(&parent->d_lock);
467 spin_unlock(&fs->lock);
472 * release_open_intent - free up open intent resources
473 * @nd: pointer to nameidata
475 void release_open_intent(struct nameidata *nd)
477 struct file *file = nd->intent.open.file;
479 if (file && !IS_ERR(file)) {
480 if (file->f_path.dentry == NULL)
487 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
489 return dentry->d_op->d_revalidate(dentry, nd);
492 static struct dentry *
493 do_revalidate(struct dentry *dentry, struct nameidata *nd)
495 int status = d_revalidate(dentry, nd);
496 if (unlikely(status <= 0)) {
498 * The dentry failed validation.
499 * If d_revalidate returned 0 attempt to invalidate
500 * the dentry otherwise d_revalidate is asking us
501 * to return a fail status.
505 dentry = ERR_PTR(status);
506 } else if (!d_invalidate(dentry)) {
515 * complete_walk - successful completion of path walk
516 * @nd: pointer nameidata
518 * If we had been in RCU mode, drop out of it and legitimize nd->path.
519 * Revalidate the final result, unless we'd already done that during
520 * the path walk or the filesystem doesn't ask for it. Return 0 on
521 * success, -error on failure. In case of failure caller does not
522 * need to drop nd->path.
524 static int complete_walk(struct nameidata *nd)
526 struct dentry *dentry = nd->path.dentry;
529 if (nd->flags & LOOKUP_RCU) {
530 nd->flags &= ~LOOKUP_RCU;
531 if (!(nd->flags & LOOKUP_ROOT))
533 spin_lock(&dentry->d_lock);
534 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
535 spin_unlock(&dentry->d_lock);
537 br_read_unlock(vfsmount_lock);
540 BUG_ON(nd->inode != dentry->d_inode);
541 spin_unlock(&dentry->d_lock);
542 mntget(nd->path.mnt);
544 br_read_unlock(vfsmount_lock);
547 if (likely(!(nd->flags & LOOKUP_JUMPED)))
550 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
553 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
556 /* Note: we do not d_invalidate() */
557 status = d_revalidate(dentry, nd);
569 * Short-cut version of permission(), for calling on directories
570 * during pathname resolution. Combines parts of permission()
571 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
573 * If appropriate, check DAC only. If not appropriate, or
574 * short-cut DAC fails, then call ->permission() to do more
575 * complete permission check.
577 static inline int exec_permission(struct inode *inode, unsigned int flags)
580 struct user_namespace *ns = inode_userns(inode);
582 if (inode->i_op->permission) {
583 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
585 ret = acl_permission_check(inode, MAY_EXEC, flags,
586 inode->i_op->check_acl);
593 if (ns_capable(ns, CAP_DAC_OVERRIDE) ||
594 ns_capable(ns, CAP_DAC_READ_SEARCH))
599 return security_inode_exec_permission(inode, flags);
602 static __always_inline void set_root(struct nameidata *nd)
605 get_fs_root(current->fs, &nd->root);
608 static int link_path_walk(const char *, struct nameidata *);
610 static __always_inline void set_root_rcu(struct nameidata *nd)
613 struct fs_struct *fs = current->fs;
617 seq = read_seqcount_begin(&fs->seq);
619 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
620 } while (read_seqcount_retry(&fs->seq, seq));
624 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
636 nd->flags |= LOOKUP_JUMPED;
638 nd->inode = nd->path.dentry->d_inode;
640 ret = link_path_walk(link, nd);
644 return PTR_ERR(link);
647 static void path_put_conditional(struct path *path, struct nameidata *nd)
650 if (path->mnt != nd->path.mnt)
654 static inline void path_to_nameidata(const struct path *path,
655 struct nameidata *nd)
657 if (!(nd->flags & LOOKUP_RCU)) {
658 dput(nd->path.dentry);
659 if (nd->path.mnt != path->mnt)
660 mntput(nd->path.mnt);
662 nd->path.mnt = path->mnt;
663 nd->path.dentry = path->dentry;
666 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
668 struct inode *inode = link->dentry->d_inode;
669 if (!IS_ERR(cookie) && inode->i_op->put_link)
670 inode->i_op->put_link(link->dentry, nd, cookie);
674 static __always_inline int
675 follow_link(struct path *link, struct nameidata *nd, void **p)
678 struct dentry *dentry = link->dentry;
680 BUG_ON(nd->flags & LOOKUP_RCU);
682 if (link->mnt == nd->path.mnt)
685 if (unlikely(current->total_link_count >= 40)) {
686 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
691 current->total_link_count++;
693 touch_atime(link->mnt, dentry);
694 nd_set_link(nd, NULL);
696 error = security_inode_follow_link(link->dentry, nd);
698 *p = ERR_PTR(error); /* no ->put_link(), please */
703 nd->last_type = LAST_BIND;
704 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
707 char *s = nd_get_link(nd);
710 error = __vfs_follow_link(nd, s);
711 else if (nd->last_type == LAST_BIND) {
712 nd->flags |= LOOKUP_JUMPED;
713 nd->inode = nd->path.dentry->d_inode;
714 if (nd->inode->i_op->follow_link) {
715 /* stepped on a _really_ weird one */
724 static int follow_up_rcu(struct path *path)
726 struct vfsmount *parent;
727 struct dentry *mountpoint;
729 parent = path->mnt->mnt_parent;
730 if (parent == path->mnt)
732 mountpoint = path->mnt->mnt_mountpoint;
733 path->dentry = mountpoint;
738 int follow_up(struct path *path)
740 struct vfsmount *parent;
741 struct dentry *mountpoint;
743 br_read_lock(vfsmount_lock);
744 parent = path->mnt->mnt_parent;
745 if (parent == path->mnt) {
746 br_read_unlock(vfsmount_lock);
750 mountpoint = dget(path->mnt->mnt_mountpoint);
751 br_read_unlock(vfsmount_lock);
753 path->dentry = mountpoint;
760 * Perform an automount
761 * - return -EISDIR to tell follow_managed() to stop and return the path we
764 static int follow_automount(struct path *path, unsigned flags,
767 struct vfsmount *mnt;
770 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
773 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
774 * and this is the terminal part of the path.
776 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
777 return -EISDIR; /* we actually want to stop here */
779 /* We want to mount if someone is trying to open/create a file of any
780 * type under the mountpoint, wants to traverse through the mountpoint
781 * or wants to open the mounted directory.
783 * We don't want to mount if someone's just doing a stat and they've
784 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
785 * appended a '/' to the name.
787 if (!(flags & LOOKUP_FOLLOW) &&
788 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
789 LOOKUP_OPEN | LOOKUP_CREATE)))
792 current->total_link_count++;
793 if (current->total_link_count >= 40)
796 mnt = path->dentry->d_op->d_automount(path);
799 * The filesystem is allowed to return -EISDIR here to indicate
800 * it doesn't want to automount. For instance, autofs would do
801 * this so that its userspace daemon can mount on this dentry.
803 * However, we can only permit this if it's a terminal point in
804 * the path being looked up; if it wasn't then the remainder of
805 * the path is inaccessible and we should say so.
807 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
812 if (!mnt) /* mount collision */
815 err = finish_automount(mnt, path);
819 /* Someone else made a mount here whilst we were busy */
826 path->dentry = dget(mnt->mnt_root);
836 * Handle a dentry that is managed in some way.
837 * - Flagged for transit management (autofs)
838 * - Flagged as mountpoint
839 * - Flagged as automount point
841 * This may only be called in refwalk mode.
843 * Serialization is taken care of in namespace.c
845 static int follow_managed(struct path *path, unsigned flags)
848 bool need_mntput = false;
851 /* Given that we're not holding a lock here, we retain the value in a
852 * local variable for each dentry as we look at it so that we don't see
853 * the components of that value change under us */
854 while (managed = ACCESS_ONCE(path->dentry->d_flags),
855 managed &= DCACHE_MANAGED_DENTRY,
856 unlikely(managed != 0)) {
857 /* Allow the filesystem to manage the transit without i_mutex
859 if (managed & DCACHE_MANAGE_TRANSIT) {
860 BUG_ON(!path->dentry->d_op);
861 BUG_ON(!path->dentry->d_op->d_manage);
862 ret = path->dentry->d_op->d_manage(path->dentry, false);
864 return ret == -EISDIR ? 0 : ret;
867 /* Transit to a mounted filesystem. */
868 if (managed & DCACHE_MOUNTED) {
869 struct vfsmount *mounted = lookup_mnt(path);
875 path->dentry = dget(mounted->mnt_root);
880 /* Something is mounted on this dentry in another
881 * namespace and/or whatever was mounted there in this
882 * namespace got unmounted before we managed to get the
886 /* Handle an automount point */
887 if (managed & DCACHE_NEED_AUTOMOUNT) {
888 ret = follow_automount(path, flags, &need_mntput);
890 return ret == -EISDIR ? 0 : ret;
894 /* We didn't change the current path point */
900 int follow_down_one(struct path *path)
902 struct vfsmount *mounted;
904 mounted = lookup_mnt(path);
909 path->dentry = dget(mounted->mnt_root);
915 static inline bool managed_dentry_might_block(struct dentry *dentry)
917 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
918 dentry->d_op->d_manage(dentry, true) < 0);
922 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
923 * we meet a managed dentry that would need blocking.
925 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
926 struct inode **inode)
929 struct vfsmount *mounted;
931 * Don't forget we might have a non-mountpoint managed dentry
932 * that wants to block transit.
934 *inode = path->dentry->d_inode;
935 if (unlikely(managed_dentry_might_block(path->dentry)))
938 if (!d_mountpoint(path->dentry))
941 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
945 path->dentry = mounted->mnt_root;
946 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
949 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
954 static void follow_mount_rcu(struct nameidata *nd, struct path *path,
955 struct inode **inode)
958 struct vfsmount *mounted;
959 *inode = path->dentry->d_inode;
961 if (!d_mountpoint(path->dentry))
964 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
968 path->dentry = mounted->mnt_root;
969 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
973 static int follow_dotdot_rcu(struct nameidata *nd)
975 struct inode *inode = nd->inode;
980 if (nd->path.dentry == nd->root.dentry &&
981 nd->path.mnt == nd->root.mnt) {
984 if (nd->path.dentry != nd->path.mnt->mnt_root) {
985 struct dentry *old = nd->path.dentry;
986 struct dentry *parent = old->d_parent;
989 seq = read_seqcount_begin(&parent->d_seq);
990 if (read_seqcount_retry(&old->d_seq, nd->seq))
992 inode = parent->d_inode;
993 nd->path.dentry = parent;
997 if (!follow_up_rcu(&nd->path))
999 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1000 inode = nd->path.dentry->d_inode;
1002 follow_mount_rcu(nd, &nd->path, &inode);
1007 nd->flags &= ~LOOKUP_RCU;
1008 if (!(nd->flags & LOOKUP_ROOT))
1009 nd->root.mnt = NULL;
1011 br_read_unlock(vfsmount_lock);
1016 * Follow down to the covering mount currently visible to userspace. At each
1017 * point, the filesystem owning that dentry may be queried as to whether the
1018 * caller is permitted to proceed or not.
1020 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1023 int follow_down(struct path *path)
1028 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1029 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1030 /* Allow the filesystem to manage the transit without i_mutex
1033 * We indicate to the filesystem if someone is trying to mount
1034 * something here. This gives autofs the chance to deny anyone
1035 * other than its daemon the right to mount on its
1038 * The filesystem may sleep at this point.
1040 if (managed & DCACHE_MANAGE_TRANSIT) {
1041 BUG_ON(!path->dentry->d_op);
1042 BUG_ON(!path->dentry->d_op->d_manage);
1043 ret = path->dentry->d_op->d_manage(
1044 path->dentry, false);
1046 return ret == -EISDIR ? 0 : ret;
1049 /* Transit to a mounted filesystem. */
1050 if (managed & DCACHE_MOUNTED) {
1051 struct vfsmount *mounted = lookup_mnt(path);
1056 path->mnt = mounted;
1057 path->dentry = dget(mounted->mnt_root);
1061 /* Don't handle automount points here */
1068 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1070 static void follow_mount(struct path *path)
1072 while (d_mountpoint(path->dentry)) {
1073 struct vfsmount *mounted = lookup_mnt(path);
1078 path->mnt = mounted;
1079 path->dentry = dget(mounted->mnt_root);
1083 static void follow_dotdot(struct nameidata *nd)
1088 struct dentry *old = nd->path.dentry;
1090 if (nd->path.dentry == nd->root.dentry &&
1091 nd->path.mnt == nd->root.mnt) {
1094 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1095 /* rare case of legitimate dget_parent()... */
1096 nd->path.dentry = dget_parent(nd->path.dentry);
1100 if (!follow_up(&nd->path))
1103 follow_mount(&nd->path);
1104 nd->inode = nd->path.dentry->d_inode;
1108 * Allocate a dentry with name and parent, and perform a parent
1109 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1110 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1111 * have verified that no child exists while under i_mutex.
1113 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1114 struct qstr *name, struct nameidata *nd)
1116 struct inode *inode = parent->d_inode;
1117 struct dentry *dentry;
1120 /* Don't create child dentry for a dead directory. */
1121 if (unlikely(IS_DEADDIR(inode)))
1122 return ERR_PTR(-ENOENT);
1124 dentry = d_alloc(parent, name);
1125 if (unlikely(!dentry))
1126 return ERR_PTR(-ENOMEM);
1128 old = inode->i_op->lookup(inode, dentry, nd);
1129 if (unlikely(old)) {
1137 * It's more convoluted than I'd like it to be, but... it's still fairly
1138 * small and for now I'd prefer to have fast path as straight as possible.
1139 * It _is_ time-critical.
1141 static int do_lookup(struct nameidata *nd, struct qstr *name,
1142 struct path *path, struct inode **inode)
1144 struct vfsmount *mnt = nd->path.mnt;
1145 struct dentry *dentry, *parent = nd->path.dentry;
1151 * Rename seqlock is not required here because in the off chance
1152 * of a false negative due to a concurrent rename, we're going to
1153 * do the non-racy lookup, below.
1155 if (nd->flags & LOOKUP_RCU) {
1158 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1162 /* Memory barrier in read_seqcount_begin of child is enough */
1163 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1167 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1168 status = d_revalidate(dentry, nd);
1169 if (unlikely(status <= 0)) {
1170 if (status != -ECHILD)
1176 path->dentry = dentry;
1177 if (likely(__follow_mount_rcu(nd, path, inode)))
1180 if (unlazy_walk(nd, dentry))
1183 dentry = __d_lookup(parent, name);
1187 if (unlikely(!dentry)) {
1188 struct inode *dir = parent->d_inode;
1189 BUG_ON(nd->inode != dir);
1191 mutex_lock(&dir->i_mutex);
1192 dentry = d_lookup(parent, name);
1193 if (likely(!dentry)) {
1194 dentry = d_alloc_and_lookup(parent, name, nd);
1195 if (IS_ERR(dentry)) {
1196 mutex_unlock(&dir->i_mutex);
1197 return PTR_ERR(dentry);
1203 mutex_unlock(&dir->i_mutex);
1205 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1206 status = d_revalidate(dentry, nd);
1207 if (unlikely(status <= 0)) {
1212 if (!d_invalidate(dentry)) {
1221 path->dentry = dentry;
1222 err = follow_managed(path, nd->flags);
1223 if (unlikely(err < 0)) {
1224 path_put_conditional(path, nd);
1227 *inode = path->dentry->d_inode;
1231 static inline int may_lookup(struct nameidata *nd)
1233 if (nd->flags & LOOKUP_RCU) {
1234 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1237 if (unlazy_walk(nd, NULL))
1240 return exec_permission(nd->inode, 0);
1243 static inline int handle_dots(struct nameidata *nd, int type)
1245 if (type == LAST_DOTDOT) {
1246 if (nd->flags & LOOKUP_RCU) {
1247 if (follow_dotdot_rcu(nd))
1255 static void terminate_walk(struct nameidata *nd)
1257 if (!(nd->flags & LOOKUP_RCU)) {
1258 path_put(&nd->path);
1260 nd->flags &= ~LOOKUP_RCU;
1261 if (!(nd->flags & LOOKUP_ROOT))
1262 nd->root.mnt = NULL;
1264 br_read_unlock(vfsmount_lock);
1268 static inline int walk_component(struct nameidata *nd, struct path *path,
1269 struct qstr *name, int type, int follow)
1271 struct inode *inode;
1274 * "." and ".." are special - ".." especially so because it has
1275 * to be able to know about the current root directory and
1276 * parent relationships.
1278 if (unlikely(type != LAST_NORM))
1279 return handle_dots(nd, type);
1280 err = do_lookup(nd, name, path, &inode);
1281 if (unlikely(err)) {
1286 path_to_nameidata(path, nd);
1290 if (unlikely(inode->i_op->follow_link) && follow) {
1291 if (nd->flags & LOOKUP_RCU) {
1292 if (unlikely(unlazy_walk(nd, path->dentry))) {
1297 BUG_ON(inode != path->dentry->d_inode);
1300 path_to_nameidata(path, nd);
1306 * This limits recursive symlink follows to 8, while
1307 * limiting consecutive symlinks to 40.
1309 * Without that kind of total limit, nasty chains of consecutive
1310 * symlinks can cause almost arbitrarily long lookups.
1312 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1316 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1317 path_put_conditional(path, nd);
1318 path_put(&nd->path);
1321 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1324 current->link_count++;
1327 struct path link = *path;
1330 res = follow_link(&link, nd, &cookie);
1332 res = walk_component(nd, path, &nd->last,
1333 nd->last_type, LOOKUP_FOLLOW);
1334 put_link(nd, &link, cookie);
1337 current->link_count--;
1344 * This is the basic name resolution function, turning a pathname into
1345 * the final dentry. We expect 'base' to be positive and a directory.
1347 * Returns 0 and nd will have valid dentry and mnt on success.
1348 * Returns error and drops reference to input namei data on failure.
1350 static int link_path_walk(const char *name, struct nameidata *nd)
1354 unsigned int lookup_flags = nd->flags;
1361 /* At this point we know we have a real path component. */
1368 nd->flags |= LOOKUP_CONTINUE;
1370 err = may_lookup(nd);
1375 c = *(const unsigned char *)name;
1377 hash = init_name_hash();
1380 hash = partial_name_hash(c, hash);
1381 c = *(const unsigned char *)name;
1382 } while (c && (c != '/'));
1383 this.len = name - (const char *) this.name;
1384 this.hash = end_name_hash(hash);
1387 if (this.name[0] == '.') switch (this.len) {
1389 if (this.name[1] == '.') {
1391 nd->flags |= LOOKUP_JUMPED;
1397 if (likely(type == LAST_NORM)) {
1398 struct dentry *parent = nd->path.dentry;
1399 nd->flags &= ~LOOKUP_JUMPED;
1400 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1401 err = parent->d_op->d_hash(parent, nd->inode,
1408 /* remove trailing slashes? */
1410 goto last_component;
1411 while (*++name == '/');
1413 goto last_component;
1415 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1420 err = nested_symlink(&next, nd);
1425 if (!nd->inode->i_op->lookup)
1428 /* here ends the main loop */
1431 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1432 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1434 nd->last_type = type;
1441 static int path_init(int dfd, const char *name, unsigned int flags,
1442 struct nameidata *nd, struct file **fp)
1448 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1449 nd->flags = flags | LOOKUP_JUMPED;
1451 if (flags & LOOKUP_ROOT) {
1452 struct inode *inode = nd->root.dentry->d_inode;
1454 if (!inode->i_op->lookup)
1456 retval = inode_permission(inode, MAY_EXEC);
1460 nd->path = nd->root;
1462 if (flags & LOOKUP_RCU) {
1463 br_read_lock(vfsmount_lock);
1465 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1467 path_get(&nd->path);
1472 nd->root.mnt = NULL;
1475 if (flags & LOOKUP_RCU) {
1476 br_read_lock(vfsmount_lock);
1481 path_get(&nd->root);
1483 nd->path = nd->root;
1484 } else if (dfd == AT_FDCWD) {
1485 if (flags & LOOKUP_RCU) {
1486 struct fs_struct *fs = current->fs;
1489 br_read_lock(vfsmount_lock);
1493 seq = read_seqcount_begin(&fs->seq);
1495 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1496 } while (read_seqcount_retry(&fs->seq, seq));
1498 get_fs_pwd(current->fs, &nd->path);
1501 struct dentry *dentry;
1503 file = fget_raw_light(dfd, &fput_needed);
1508 dentry = file->f_path.dentry;
1512 if (!S_ISDIR(dentry->d_inode->i_mode))
1515 retval = file_permission(file, MAY_EXEC);
1520 nd->path = file->f_path;
1521 if (flags & LOOKUP_RCU) {
1524 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1525 br_read_lock(vfsmount_lock);
1528 path_get(&file->f_path);
1529 fput_light(file, fput_needed);
1533 nd->inode = nd->path.dentry->d_inode;
1537 fput_light(file, fput_needed);
1542 static inline int lookup_last(struct nameidata *nd, struct path *path)
1544 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1545 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1547 nd->flags &= ~LOOKUP_PARENT;
1548 return walk_component(nd, path, &nd->last, nd->last_type,
1549 nd->flags & LOOKUP_FOLLOW);
1552 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1553 static int path_lookupat(int dfd, const char *name,
1554 unsigned int flags, struct nameidata *nd)
1556 struct file *base = NULL;
1561 * Path walking is largely split up into 2 different synchronisation
1562 * schemes, rcu-walk and ref-walk (explained in
1563 * Documentation/filesystems/path-lookup.txt). These share much of the
1564 * path walk code, but some things particularly setup, cleanup, and
1565 * following mounts are sufficiently divergent that functions are
1566 * duplicated. Typically there is a function foo(), and its RCU
1567 * analogue, foo_rcu().
1569 * -ECHILD is the error number of choice (just to avoid clashes) that
1570 * is returned if some aspect of an rcu-walk fails. Such an error must
1571 * be handled by restarting a traditional ref-walk (which will always
1572 * be able to complete).
1574 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1579 current->total_link_count = 0;
1580 err = link_path_walk(name, nd);
1582 if (!err && !(flags & LOOKUP_PARENT)) {
1583 err = lookup_last(nd, &path);
1586 struct path link = path;
1587 nd->flags |= LOOKUP_PARENT;
1588 err = follow_link(&link, nd, &cookie);
1590 err = lookup_last(nd, &path);
1591 put_link(nd, &link, cookie);
1596 err = complete_walk(nd);
1598 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1599 if (!nd->inode->i_op->lookup) {
1600 path_put(&nd->path);
1608 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1609 path_put(&nd->root);
1610 nd->root.mnt = NULL;
1615 static int do_path_lookup(int dfd, const char *name,
1616 unsigned int flags, struct nameidata *nd)
1618 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1619 if (unlikely(retval == -ECHILD))
1620 retval = path_lookupat(dfd, name, flags, nd);
1621 if (unlikely(retval == -ESTALE))
1622 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1624 if (likely(!retval)) {
1625 if (unlikely(!audit_dummy_context())) {
1626 if (nd->path.dentry && nd->inode)
1627 audit_inode(name, nd->path.dentry);
1633 int kern_path_parent(const char *name, struct nameidata *nd)
1635 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1638 int kern_path(const char *name, unsigned int flags, struct path *path)
1640 struct nameidata nd;
1641 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1648 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1649 * @dentry: pointer to dentry of the base directory
1650 * @mnt: pointer to vfs mount of the base directory
1651 * @name: pointer to file name
1652 * @flags: lookup flags
1653 * @nd: pointer to nameidata
1655 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1656 const char *name, unsigned int flags,
1657 struct nameidata *nd)
1659 nd->root.dentry = dentry;
1661 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1662 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
1665 static struct dentry *__lookup_hash(struct qstr *name,
1666 struct dentry *base, struct nameidata *nd)
1668 struct inode *inode = base->d_inode;
1669 struct dentry *dentry;
1672 err = exec_permission(inode, 0);
1674 return ERR_PTR(err);
1677 * Don't bother with __d_lookup: callers are for creat as
1678 * well as unlink, so a lot of the time it would cost
1681 dentry = d_lookup(base, name);
1683 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
1684 dentry = do_revalidate(dentry, nd);
1687 dentry = d_alloc_and_lookup(base, name, nd);
1693 * Restricted form of lookup. Doesn't follow links, single-component only,
1694 * needs parent already locked. Doesn't follow mounts.
1697 static struct dentry *lookup_hash(struct nameidata *nd)
1699 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1703 * lookup_one_len - filesystem helper to lookup single pathname component
1704 * @name: pathname component to lookup
1705 * @base: base directory to lookup from
1706 * @len: maximum length @len should be interpreted to
1708 * Note that this routine is purely a helper for filesystem usage and should
1709 * not be called by generic code. Also note that by using this function the
1710 * nameidata argument is passed to the filesystem methods and a filesystem
1711 * using this helper needs to be prepared for that.
1713 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1719 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1724 return ERR_PTR(-EACCES);
1726 hash = init_name_hash();
1728 c = *(const unsigned char *)name++;
1729 if (c == '/' || c == '\0')
1730 return ERR_PTR(-EACCES);
1731 hash = partial_name_hash(c, hash);
1733 this.hash = end_name_hash(hash);
1735 * See if the low-level filesystem might want
1736 * to use its own hash..
1738 if (base->d_flags & DCACHE_OP_HASH) {
1739 int err = base->d_op->d_hash(base, base->d_inode, &this);
1741 return ERR_PTR(err);
1744 return __lookup_hash(&this, base, NULL);
1747 int user_path_at(int dfd, const char __user *name, unsigned flags,
1750 struct nameidata nd;
1751 char *tmp = getname_flags(name, flags);
1752 int err = PTR_ERR(tmp);
1755 BUG_ON(flags & LOOKUP_PARENT);
1757 err = do_path_lookup(dfd, tmp, flags, &nd);
1765 static int user_path_parent(int dfd, const char __user *path,
1766 struct nameidata *nd, char **name)
1768 char *s = getname(path);
1774 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1784 * It's inline, so penalty for filesystems that don't use sticky bit is
1787 static inline int check_sticky(struct inode *dir, struct inode *inode)
1789 uid_t fsuid = current_fsuid();
1791 if (!(dir->i_mode & S_ISVTX))
1793 if (current_user_ns() != inode_userns(inode))
1795 if (inode->i_uid == fsuid)
1797 if (dir->i_uid == fsuid)
1801 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1805 * Check whether we can remove a link victim from directory dir, check
1806 * whether the type of victim is right.
1807 * 1. We can't do it if dir is read-only (done in permission())
1808 * 2. We should have write and exec permissions on dir
1809 * 3. We can't remove anything from append-only dir
1810 * 4. We can't do anything with immutable dir (done in permission())
1811 * 5. If the sticky bit on dir is set we should either
1812 * a. be owner of dir, or
1813 * b. be owner of victim, or
1814 * c. have CAP_FOWNER capability
1815 * 6. If the victim is append-only or immutable we can't do antyhing with
1816 * links pointing to it.
1817 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1818 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1819 * 9. We can't remove a root or mountpoint.
1820 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1821 * nfs_async_unlink().
1823 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1827 if (!victim->d_inode)
1830 BUG_ON(victim->d_parent->d_inode != dir);
1831 audit_inode_child(victim, dir);
1833 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1838 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1839 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1842 if (!S_ISDIR(victim->d_inode->i_mode))
1844 if (IS_ROOT(victim))
1846 } else if (S_ISDIR(victim->d_inode->i_mode))
1848 if (IS_DEADDIR(dir))
1850 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1855 /* Check whether we can create an object with dentry child in directory
1857 * 1. We can't do it if child already exists (open has special treatment for
1858 * this case, but since we are inlined it's OK)
1859 * 2. We can't do it if dir is read-only (done in permission())
1860 * 3. We should have write and exec permissions on dir
1861 * 4. We can't do it if dir is immutable (done in permission())
1863 static inline int may_create(struct inode *dir, struct dentry *child)
1867 if (IS_DEADDIR(dir))
1869 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1873 * p1 and p2 should be directories on the same fs.
1875 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1880 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1884 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1886 p = d_ancestor(p2, p1);
1888 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1889 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1893 p = d_ancestor(p1, p2);
1895 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1896 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1900 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1901 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1905 void unlock_rename(struct dentry *p1, struct dentry *p2)
1907 mutex_unlock(&p1->d_inode->i_mutex);
1909 mutex_unlock(&p2->d_inode->i_mutex);
1910 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1914 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1915 struct nameidata *nd)
1917 int error = may_create(dir, dentry);
1922 if (!dir->i_op->create)
1923 return -EACCES; /* shouldn't it be ENOSYS? */
1926 error = security_inode_create(dir, dentry, mode);
1929 error = dir->i_op->create(dir, dentry, mode, nd);
1931 fsnotify_create(dir, dentry);
1935 static int may_open(struct path *path, int acc_mode, int flag)
1937 struct dentry *dentry = path->dentry;
1938 struct inode *inode = dentry->d_inode;
1948 switch (inode->i_mode & S_IFMT) {
1952 if (acc_mode & MAY_WRITE)
1957 if (path->mnt->mnt_flags & MNT_NODEV)
1966 error = inode_permission(inode, acc_mode);
1971 * An append-only file must be opened in append mode for writing.
1973 if (IS_APPEND(inode)) {
1974 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
1980 /* O_NOATIME can only be set by the owner or superuser */
1981 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
1985 * Ensure there are no outstanding leases on the file.
1987 return break_lease(inode, flag);
1990 static int handle_truncate(struct file *filp)
1992 struct path *path = &filp->f_path;
1993 struct inode *inode = path->dentry->d_inode;
1994 int error = get_write_access(inode);
1998 * Refuse to truncate files with mandatory locks held on them.
2000 error = locks_verify_locked(inode);
2002 error = security_path_truncate(path);
2004 error = do_truncate(path->dentry, 0,
2005 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2008 put_write_access(inode);
2013 * Note that while the flag value (low two bits) for sys_open means:
2018 * it is changed into
2019 * 00 - no permissions needed
2020 * 01 - read-permission
2021 * 10 - write-permission
2023 * for the internal routines (ie open_namei()/follow_link() etc)
2024 * This is more logical, and also allows the 00 "no perm needed"
2025 * to be used for symlinks (where the permissions are checked
2029 static inline int open_to_namei_flags(int flag)
2031 if ((flag+1) & O_ACCMODE)
2037 * Handle the last step of open()
2039 static struct file *do_last(struct nameidata *nd, struct path *path,
2040 const struct open_flags *op, const char *pathname)
2042 struct dentry *dir = nd->path.dentry;
2043 struct dentry *dentry;
2044 int open_flag = op->open_flag;
2045 int will_truncate = open_flag & O_TRUNC;
2047 int acc_mode = op->acc_mode;
2051 nd->flags &= ~LOOKUP_PARENT;
2052 nd->flags |= op->intent;
2054 switch (nd->last_type) {
2057 error = handle_dots(nd, nd->last_type);
2059 return ERR_PTR(error);
2062 error = complete_walk(nd);
2064 return ERR_PTR(error);
2065 audit_inode(pathname, nd->path.dentry);
2066 if (open_flag & O_CREAT) {
2072 error = complete_walk(nd);
2074 return ERR_PTR(error);
2075 audit_inode(pathname, dir);
2079 if (!(open_flag & O_CREAT)) {
2081 if (nd->last.name[nd->last.len])
2082 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2083 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2085 /* we _can_ be in RCU mode here */
2086 error = walk_component(nd, path, &nd->last, LAST_NORM,
2089 return ERR_PTR(error);
2090 if (error) /* symlink */
2093 error = complete_walk(nd);
2095 return ERR_PTR(-ECHILD);
2098 if (nd->flags & LOOKUP_DIRECTORY) {
2099 if (!nd->inode->i_op->lookup)
2102 audit_inode(pathname, nd->path.dentry);
2106 /* create side of things */
2107 error = complete_walk(nd);
2109 return ERR_PTR(error);
2111 audit_inode(pathname, dir);
2113 /* trailing slashes? */
2114 if (nd->last.name[nd->last.len])
2117 mutex_lock(&dir->d_inode->i_mutex);
2119 dentry = lookup_hash(nd);
2120 error = PTR_ERR(dentry);
2121 if (IS_ERR(dentry)) {
2122 mutex_unlock(&dir->d_inode->i_mutex);
2126 path->dentry = dentry;
2127 path->mnt = nd->path.mnt;
2129 /* Negative dentry, just create the file */
2130 if (!dentry->d_inode) {
2131 int mode = op->mode;
2132 if (!IS_POSIXACL(dir->d_inode))
2133 mode &= ~current_umask();
2135 * This write is needed to ensure that a
2136 * rw->ro transition does not occur between
2137 * the time when the file is created and when
2138 * a permanent write count is taken through
2139 * the 'struct file' in nameidata_to_filp().
2141 error = mnt_want_write(nd->path.mnt);
2143 goto exit_mutex_unlock;
2145 /* Don't check for write permission, don't truncate */
2146 open_flag &= ~O_TRUNC;
2148 acc_mode = MAY_OPEN;
2149 error = security_path_mknod(&nd->path, dentry, mode, 0);
2151 goto exit_mutex_unlock;
2152 error = vfs_create(dir->d_inode, dentry, mode, nd);
2154 goto exit_mutex_unlock;
2155 mutex_unlock(&dir->d_inode->i_mutex);
2156 dput(nd->path.dentry);
2157 nd->path.dentry = dentry;
2162 * It already exists.
2164 mutex_unlock(&dir->d_inode->i_mutex);
2165 audit_inode(pathname, path->dentry);
2168 if (open_flag & O_EXCL)
2171 error = follow_managed(path, nd->flags);
2176 if (!path->dentry->d_inode)
2179 if (path->dentry->d_inode->i_op->follow_link)
2182 path_to_nameidata(path, nd);
2183 nd->inode = path->dentry->d_inode;
2185 if (S_ISDIR(nd->inode->i_mode))
2188 if (!S_ISREG(nd->inode->i_mode))
2191 if (will_truncate) {
2192 error = mnt_want_write(nd->path.mnt);
2198 error = may_open(&nd->path, acc_mode, open_flag);
2201 filp = nameidata_to_filp(nd);
2202 if (!IS_ERR(filp)) {
2203 error = ima_file_check(filp, op->acc_mode);
2206 filp = ERR_PTR(error);
2209 if (!IS_ERR(filp)) {
2210 if (will_truncate) {
2211 error = handle_truncate(filp);
2214 filp = ERR_PTR(error);
2220 mnt_drop_write(nd->path.mnt);
2221 path_put(&nd->path);
2225 mutex_unlock(&dir->d_inode->i_mutex);
2227 path_put_conditional(path, nd);
2229 filp = ERR_PTR(error);
2233 static struct file *path_openat(int dfd, const char *pathname,
2234 struct nameidata *nd, const struct open_flags *op, int flags)
2236 struct file *base = NULL;
2241 filp = get_empty_filp();
2243 return ERR_PTR(-ENFILE);
2245 filp->f_flags = op->open_flag;
2246 nd->intent.open.file = filp;
2247 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2248 nd->intent.open.create_mode = op->mode;
2250 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2251 if (unlikely(error))
2254 current->total_link_count = 0;
2255 error = link_path_walk(pathname, nd);
2256 if (unlikely(error))
2259 filp = do_last(nd, &path, op, pathname);
2260 while (unlikely(!filp)) { /* trailing symlink */
2261 struct path link = path;
2263 if (!(nd->flags & LOOKUP_FOLLOW)) {
2264 path_put_conditional(&path, nd);
2265 path_put(&nd->path);
2266 filp = ERR_PTR(-ELOOP);
2269 nd->flags |= LOOKUP_PARENT;
2270 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2271 error = follow_link(&link, nd, &cookie);
2272 if (unlikely(error))
2273 filp = ERR_PTR(error);
2275 filp = do_last(nd, &path, op, pathname);
2276 put_link(nd, &link, cookie);
2279 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2280 path_put(&nd->root);
2283 release_open_intent(nd);
2287 filp = ERR_PTR(error);
2291 struct file *do_filp_open(int dfd, const char *pathname,
2292 const struct open_flags *op, int flags)
2294 struct nameidata nd;
2297 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2298 if (unlikely(filp == ERR_PTR(-ECHILD)))
2299 filp = path_openat(dfd, pathname, &nd, op, flags);
2300 if (unlikely(filp == ERR_PTR(-ESTALE)))
2301 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2305 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2306 const char *name, const struct open_flags *op, int flags)
2308 struct nameidata nd;
2312 nd.root.dentry = dentry;
2314 flags |= LOOKUP_ROOT;
2316 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2317 return ERR_PTR(-ELOOP);
2319 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2320 if (unlikely(file == ERR_PTR(-ECHILD)))
2321 file = path_openat(-1, name, &nd, op, flags);
2322 if (unlikely(file == ERR_PTR(-ESTALE)))
2323 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2328 * lookup_create - lookup a dentry, creating it if it doesn't exist
2329 * @nd: nameidata info
2330 * @is_dir: directory flag
2332 * Simple function to lookup and return a dentry and create it
2333 * if it doesn't exist. Is SMP-safe.
2335 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2337 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2339 struct dentry *dentry = ERR_PTR(-EEXIST);
2341 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2343 * Yucky last component or no last component at all?
2344 * (foo/., foo/.., /////)
2346 if (nd->last_type != LAST_NORM)
2348 nd->flags &= ~LOOKUP_PARENT;
2349 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2350 nd->intent.open.flags = O_EXCL;
2353 * Do the final lookup.
2355 dentry = lookup_hash(nd);
2359 if (dentry->d_inode)
2362 * Special case - lookup gave negative, but... we had foo/bar/
2363 * From the vfs_mknod() POV we just have a negative dentry -
2364 * all is fine. Let's be bastards - you had / on the end, you've
2365 * been asking for (non-existent) directory. -ENOENT for you.
2367 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2369 dentry = ERR_PTR(-ENOENT);
2374 dentry = ERR_PTR(-EEXIST);
2378 EXPORT_SYMBOL_GPL(lookup_create);
2380 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2382 int error = may_create(dir, dentry);
2387 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2388 !ns_capable(inode_userns(dir), CAP_MKNOD))
2391 if (!dir->i_op->mknod)
2394 error = devcgroup_inode_mknod(mode, dev);
2398 error = security_inode_mknod(dir, dentry, mode, dev);
2402 error = dir->i_op->mknod(dir, dentry, mode, dev);
2404 fsnotify_create(dir, dentry);
2408 static int may_mknod(mode_t mode)
2410 switch (mode & S_IFMT) {
2416 case 0: /* zero mode translates to S_IFREG */
2425 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2430 struct dentry *dentry;
2431 struct nameidata nd;
2436 error = user_path_parent(dfd, filename, &nd, &tmp);
2440 dentry = lookup_create(&nd, 0);
2441 if (IS_ERR(dentry)) {
2442 error = PTR_ERR(dentry);
2445 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2446 mode &= ~current_umask();
2447 error = may_mknod(mode);
2450 error = mnt_want_write(nd.path.mnt);
2453 error = security_path_mknod(&nd.path, dentry, mode, dev);
2455 goto out_drop_write;
2456 switch (mode & S_IFMT) {
2457 case 0: case S_IFREG:
2458 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2460 case S_IFCHR: case S_IFBLK:
2461 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2462 new_decode_dev(dev));
2464 case S_IFIFO: case S_IFSOCK:
2465 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2469 mnt_drop_write(nd.path.mnt);
2473 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2480 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2482 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2485 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2487 int error = may_create(dir, dentry);
2492 if (!dir->i_op->mkdir)
2495 mode &= (S_IRWXUGO|S_ISVTX);
2496 error = security_inode_mkdir(dir, dentry, mode);
2500 error = dir->i_op->mkdir(dir, dentry, mode);
2502 fsnotify_mkdir(dir, dentry);
2506 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2510 struct dentry *dentry;
2511 struct nameidata nd;
2513 error = user_path_parent(dfd, pathname, &nd, &tmp);
2517 dentry = lookup_create(&nd, 1);
2518 error = PTR_ERR(dentry);
2522 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2523 mode &= ~current_umask();
2524 error = mnt_want_write(nd.path.mnt);
2527 error = security_path_mkdir(&nd.path, dentry, mode);
2529 goto out_drop_write;
2530 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2532 mnt_drop_write(nd.path.mnt);
2536 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2543 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2545 return sys_mkdirat(AT_FDCWD, pathname, mode);
2549 * The dentry_unhash() helper will try to drop the dentry early: we
2550 * should have a usage count of 2 if we're the only user of this
2551 * dentry, and if that is true (possibly after pruning the dcache),
2552 * then we drop the dentry now.
2554 * A low-level filesystem can, if it choses, legally
2557 * if (!d_unhashed(dentry))
2560 * if it cannot handle the case of removing a directory
2561 * that is still in use by something else..
2563 void dentry_unhash(struct dentry *dentry)
2565 shrink_dcache_parent(dentry);
2566 spin_lock(&dentry->d_lock);
2567 if (dentry->d_count == 1)
2569 spin_unlock(&dentry->d_lock);
2572 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2574 int error = may_delete(dir, dentry, 1);
2579 if (!dir->i_op->rmdir)
2582 mutex_lock(&dentry->d_inode->i_mutex);
2585 if (d_mountpoint(dentry))
2588 error = security_inode_rmdir(dir, dentry);
2592 error = dir->i_op->rmdir(dir, dentry);
2596 dentry->d_inode->i_flags |= S_DEAD;
2600 mutex_unlock(&dentry->d_inode->i_mutex);
2606 static long do_rmdir(int dfd, const char __user *pathname)
2610 struct dentry *dentry;
2611 struct nameidata nd;
2613 error = user_path_parent(dfd, pathname, &nd, &name);
2617 switch(nd.last_type) {
2629 nd.flags &= ~LOOKUP_PARENT;
2631 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2632 dentry = lookup_hash(&nd);
2633 error = PTR_ERR(dentry);
2636 error = mnt_want_write(nd.path.mnt);
2639 error = security_path_rmdir(&nd.path, dentry);
2642 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2644 mnt_drop_write(nd.path.mnt);
2648 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2655 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2657 return do_rmdir(AT_FDCWD, pathname);
2660 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2662 int error = may_delete(dir, dentry, 0);
2667 if (!dir->i_op->unlink)
2670 mutex_lock(&dentry->d_inode->i_mutex);
2671 if (d_mountpoint(dentry))
2674 error = security_inode_unlink(dir, dentry);
2676 error = dir->i_op->unlink(dir, dentry);
2681 mutex_unlock(&dentry->d_inode->i_mutex);
2683 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2684 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2685 fsnotify_link_count(dentry->d_inode);
2693 * Make sure that the actual truncation of the file will occur outside its
2694 * directory's i_mutex. Truncate can take a long time if there is a lot of
2695 * writeout happening, and we don't want to prevent access to the directory
2696 * while waiting on the I/O.
2698 static long do_unlinkat(int dfd, const char __user *pathname)
2702 struct dentry *dentry;
2703 struct nameidata nd;
2704 struct inode *inode = NULL;
2706 error = user_path_parent(dfd, pathname, &nd, &name);
2711 if (nd.last_type != LAST_NORM)
2714 nd.flags &= ~LOOKUP_PARENT;
2716 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2717 dentry = lookup_hash(&nd);
2718 error = PTR_ERR(dentry);
2719 if (!IS_ERR(dentry)) {
2720 /* Why not before? Because we want correct error value */
2721 if (nd.last.name[nd.last.len])
2723 inode = dentry->d_inode;
2726 error = mnt_want_write(nd.path.mnt);
2729 error = security_path_unlink(&nd.path, dentry);
2732 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2734 mnt_drop_write(nd.path.mnt);
2738 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2740 iput(inode); /* truncate the inode here */
2747 error = !dentry->d_inode ? -ENOENT :
2748 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2752 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2754 if ((flag & ~AT_REMOVEDIR) != 0)
2757 if (flag & AT_REMOVEDIR)
2758 return do_rmdir(dfd, pathname);
2760 return do_unlinkat(dfd, pathname);
2763 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2765 return do_unlinkat(AT_FDCWD, pathname);
2768 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2770 int error = may_create(dir, dentry);
2775 if (!dir->i_op->symlink)
2778 error = security_inode_symlink(dir, dentry, oldname);
2782 error = dir->i_op->symlink(dir, dentry, oldname);
2784 fsnotify_create(dir, dentry);
2788 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2789 int, newdfd, const char __user *, newname)
2794 struct dentry *dentry;
2795 struct nameidata nd;
2797 from = getname(oldname);
2799 return PTR_ERR(from);
2801 error = user_path_parent(newdfd, newname, &nd, &to);
2805 dentry = lookup_create(&nd, 0);
2806 error = PTR_ERR(dentry);
2810 error = mnt_want_write(nd.path.mnt);
2813 error = security_path_symlink(&nd.path, dentry, from);
2815 goto out_drop_write;
2816 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2818 mnt_drop_write(nd.path.mnt);
2822 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2830 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2832 return sys_symlinkat(oldname, AT_FDCWD, newname);
2835 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2837 struct inode *inode = old_dentry->d_inode;
2843 error = may_create(dir, new_dentry);
2847 if (dir->i_sb != inode->i_sb)
2851 * A link to an append-only or immutable file cannot be created.
2853 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2855 if (!dir->i_op->link)
2857 if (S_ISDIR(inode->i_mode))
2860 error = security_inode_link(old_dentry, dir, new_dentry);
2864 mutex_lock(&inode->i_mutex);
2865 /* Make sure we don't allow creating hardlink to an unlinked file */
2866 if (inode->i_nlink == 0)
2869 error = dir->i_op->link(old_dentry, dir, new_dentry);
2870 mutex_unlock(&inode->i_mutex);
2872 fsnotify_link(dir, inode, new_dentry);
2877 * Hardlinks are often used in delicate situations. We avoid
2878 * security-related surprises by not following symlinks on the
2881 * We don't follow them on the oldname either to be compatible
2882 * with linux 2.0, and to avoid hard-linking to directories
2883 * and other special files. --ADM
2885 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2886 int, newdfd, const char __user *, newname, int, flags)
2888 struct dentry *new_dentry;
2889 struct nameidata nd;
2890 struct path old_path;
2895 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
2898 * To use null names we require CAP_DAC_READ_SEARCH
2899 * This ensures that not everyone will be able to create
2900 * handlink using the passed filedescriptor.
2902 if (flags & AT_EMPTY_PATH) {
2903 if (!capable(CAP_DAC_READ_SEARCH))
2908 if (flags & AT_SYMLINK_FOLLOW)
2909 how |= LOOKUP_FOLLOW;
2911 error = user_path_at(olddfd, oldname, how, &old_path);
2915 error = user_path_parent(newdfd, newname, &nd, &to);
2919 if (old_path.mnt != nd.path.mnt)
2921 new_dentry = lookup_create(&nd, 0);
2922 error = PTR_ERR(new_dentry);
2923 if (IS_ERR(new_dentry))
2925 error = mnt_want_write(nd.path.mnt);
2928 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2930 goto out_drop_write;
2931 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2933 mnt_drop_write(nd.path.mnt);
2937 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2942 path_put(&old_path);
2947 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2949 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2953 * The worst of all namespace operations - renaming directory. "Perverted"
2954 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2956 * a) we can get into loop creation. Check is done in is_subdir().
2957 * b) race potential - two innocent renames can create a loop together.
2958 * That's where 4.4 screws up. Current fix: serialization on
2959 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2961 * c) we have to lock _three_ objects - parents and victim (if it exists).
2962 * And that - after we got ->i_mutex on parents (until then we don't know
2963 * whether the target exists). Solution: try to be smart with locking
2964 * order for inodes. We rely on the fact that tree topology may change
2965 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2966 * move will be locked. Thus we can rank directories by the tree
2967 * (ancestors first) and rank all non-directories after them.
2968 * That works since everybody except rename does "lock parent, lookup,
2969 * lock child" and rename is under ->s_vfs_rename_mutex.
2970 * HOWEVER, it relies on the assumption that any object with ->lookup()
2971 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2972 * we'd better make sure that there's no link(2) for them.
2973 * d) conversion from fhandle to dentry may come in the wrong moment - when
2974 * we are removing the target. Solution: we will have to grab ->i_mutex
2975 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2976 * ->i_mutex on parents, which works but leads to some truly excessive
2979 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2980 struct inode *new_dir, struct dentry *new_dentry)
2983 struct inode *target = new_dentry->d_inode;
2986 * If we are going to change the parent - check write permissions,
2987 * we'll need to flip '..'.
2989 if (new_dir != old_dir) {
2990 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2995 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3000 mutex_lock(&target->i_mutex);
3003 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3006 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3011 target->i_flags |= S_DEAD;
3012 dont_mount(new_dentry);
3016 mutex_unlock(&target->i_mutex);
3018 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3019 d_move(old_dentry,new_dentry);
3023 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3024 struct inode *new_dir, struct dentry *new_dentry)
3026 struct inode *target = new_dentry->d_inode;
3029 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3035 mutex_lock(&target->i_mutex);
3038 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3041 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3046 dont_mount(new_dentry);
3047 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3048 d_move(old_dentry, new_dentry);
3051 mutex_unlock(&target->i_mutex);
3056 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3057 struct inode *new_dir, struct dentry *new_dentry)
3060 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3061 const unsigned char *old_name;
3063 if (old_dentry->d_inode == new_dentry->d_inode)
3066 error = may_delete(old_dir, old_dentry, is_dir);
3070 if (!new_dentry->d_inode)
3071 error = may_create(new_dir, new_dentry);
3073 error = may_delete(new_dir, new_dentry, is_dir);
3077 if (!old_dir->i_op->rename)
3080 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3083 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3085 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3087 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3088 new_dentry->d_inode, old_dentry);
3089 fsnotify_oldname_free(old_name);
3094 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3095 int, newdfd, const char __user *, newname)
3097 struct dentry *old_dir, *new_dir;
3098 struct dentry *old_dentry, *new_dentry;
3099 struct dentry *trap;
3100 struct nameidata oldnd, newnd;
3105 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3109 error = user_path_parent(newdfd, newname, &newnd, &to);
3114 if (oldnd.path.mnt != newnd.path.mnt)
3117 old_dir = oldnd.path.dentry;
3119 if (oldnd.last_type != LAST_NORM)
3122 new_dir = newnd.path.dentry;
3123 if (newnd.last_type != LAST_NORM)
3126 oldnd.flags &= ~LOOKUP_PARENT;
3127 newnd.flags &= ~LOOKUP_PARENT;
3128 newnd.flags |= LOOKUP_RENAME_TARGET;
3130 trap = lock_rename(new_dir, old_dir);
3132 old_dentry = lookup_hash(&oldnd);
3133 error = PTR_ERR(old_dentry);
3134 if (IS_ERR(old_dentry))
3136 /* source must exist */
3138 if (!old_dentry->d_inode)
3140 /* unless the source is a directory trailing slashes give -ENOTDIR */
3141 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3143 if (oldnd.last.name[oldnd.last.len])
3145 if (newnd.last.name[newnd.last.len])
3148 /* source should not be ancestor of target */
3150 if (old_dentry == trap)
3152 new_dentry = lookup_hash(&newnd);
3153 error = PTR_ERR(new_dentry);
3154 if (IS_ERR(new_dentry))
3156 /* target should not be an ancestor of source */
3158 if (new_dentry == trap)
3161 error = mnt_want_write(oldnd.path.mnt);
3164 error = security_path_rename(&oldnd.path, old_dentry,
3165 &newnd.path, new_dentry);
3168 error = vfs_rename(old_dir->d_inode, old_dentry,
3169 new_dir->d_inode, new_dentry);
3171 mnt_drop_write(oldnd.path.mnt);
3177 unlock_rename(new_dir, old_dir);
3179 path_put(&newnd.path);
3182 path_put(&oldnd.path);
3188 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3190 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3193 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3197 len = PTR_ERR(link);
3202 if (len > (unsigned) buflen)
3204 if (copy_to_user(buffer, link, len))
3211 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3212 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3213 * using) it for any given inode is up to filesystem.
3215 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3217 struct nameidata nd;
3222 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3224 return PTR_ERR(cookie);
3226 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3227 if (dentry->d_inode->i_op->put_link)
3228 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3232 int vfs_follow_link(struct nameidata *nd, const char *link)
3234 return __vfs_follow_link(nd, link);
3237 /* get the link contents into pagecache */
3238 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3242 struct address_space *mapping = dentry->d_inode->i_mapping;
3243 page = read_mapping_page(mapping, 0, NULL);
3248 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3252 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3254 struct page *page = NULL;
3255 char *s = page_getlink(dentry, &page);
3256 int res = vfs_readlink(dentry,buffer,buflen,s);
3259 page_cache_release(page);
3264 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3266 struct page *page = NULL;
3267 nd_set_link(nd, page_getlink(dentry, &page));
3271 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3273 struct page *page = cookie;
3277 page_cache_release(page);
3282 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3284 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3286 struct address_space *mapping = inode->i_mapping;
3291 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3293 flags |= AOP_FLAG_NOFS;
3296 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3297 flags, &page, &fsdata);
3301 kaddr = kmap_atomic(page, KM_USER0);
3302 memcpy(kaddr, symname, len-1);
3303 kunmap_atomic(kaddr, KM_USER0);
3305 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3312 mark_inode_dirty(inode);
3318 int page_symlink(struct inode *inode, const char *symname, int len)
3320 return __page_symlink(inode, symname, len,
3321 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3324 const struct inode_operations page_symlink_inode_operations = {
3325 .readlink = generic_readlink,
3326 .follow_link = page_follow_link_light,
3327 .put_link = page_put_link,
3330 EXPORT_SYMBOL(user_path_at);
3331 EXPORT_SYMBOL(follow_down_one);
3332 EXPORT_SYMBOL(follow_down);
3333 EXPORT_SYMBOL(follow_up);
3334 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3335 EXPORT_SYMBOL(getname);
3336 EXPORT_SYMBOL(lock_rename);
3337 EXPORT_SYMBOL(lookup_one_len);
3338 EXPORT_SYMBOL(page_follow_link_light);
3339 EXPORT_SYMBOL(page_put_link);
3340 EXPORT_SYMBOL(page_readlink);
3341 EXPORT_SYMBOL(__page_symlink);
3342 EXPORT_SYMBOL(page_symlink);
3343 EXPORT_SYMBOL(page_symlink_inode_operations);
3344 EXPORT_SYMBOL(kern_path_parent);
3345 EXPORT_SYMBOL(kern_path);
3346 EXPORT_SYMBOL(vfs_path_lookup);
3347 EXPORT_SYMBOL(inode_permission);
3348 EXPORT_SYMBOL(file_permission);
3349 EXPORT_SYMBOL(unlock_rename);
3350 EXPORT_SYMBOL(vfs_create);
3351 EXPORT_SYMBOL(vfs_follow_link);
3352 EXPORT_SYMBOL(vfs_link);
3353 EXPORT_SYMBOL(vfs_mkdir);
3354 EXPORT_SYMBOL(vfs_mknod);
3355 EXPORT_SYMBOL(generic_permission);
3356 EXPORT_SYMBOL(vfs_readlink);
3357 EXPORT_SYMBOL(vfs_rename);
3358 EXPORT_SYMBOL(vfs_rmdir);
3359 EXPORT_SYMBOL(vfs_symlink);
3360 EXPORT_SYMBOL(vfs_unlink);
3361 EXPORT_SYMBOL(dentry_unhash);
3362 EXPORT_SYMBOL(generic_readlink);