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/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <asm/uaccess.h>
42 /* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
90 * [10-Sep-98 Alan Modra] Another symlink change.
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 static char *getname_flags(const char __user *filename, int flags, int *empty)
122 char *result = __getname(), *err;
125 if (unlikely(!result))
126 return ERR_PTR(-ENOMEM);
128 len = strncpy_from_user(result, filename, PATH_MAX);
130 if (unlikely(len < 0))
133 /* The empty path is special. */
134 if (unlikely(!len)) {
137 err = ERR_PTR(-ENOENT);
138 if (!(flags & LOOKUP_EMPTY))
142 err = ERR_PTR(-ENAMETOOLONG);
143 if (likely(len < PATH_MAX)) {
144 audit_getname(result);
153 char *getname(const char __user * filename)
155 return getname_flags(filename, 0, NULL);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname);
169 static int check_acl(struct inode *inode, int mask)
171 #ifdef CONFIG_FS_POSIX_ACL
172 struct posix_acl *acl;
174 if (mask & MAY_NOT_BLOCK) {
175 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
178 /* no ->get_acl() calls in RCU mode... */
179 if (acl == ACL_NOT_CACHED)
181 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
184 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
187 * A filesystem can force a ACL callback by just never filling the
188 * ACL cache. But normally you'd fill the cache either at inode
189 * instantiation time, or on the first ->get_acl call.
191 * If the filesystem doesn't have a get_acl() function at all, we'll
192 * just create the negative cache entry.
194 if (acl == ACL_NOT_CACHED) {
195 if (inode->i_op->get_acl) {
196 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
200 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
206 int error = posix_acl_permission(inode, acl, mask);
207 posix_acl_release(acl);
216 * This does the basic permission checking
218 static int acl_permission_check(struct inode *inode, int mask)
220 unsigned int mode = inode->i_mode;
222 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
225 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
226 int error = check_acl(inode, mask);
227 if (error != -EAGAIN)
231 if (in_group_p(inode->i_gid))
236 * If the DACs are ok we don't need any capability check.
238 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
244 * generic_permission - check for access rights on a Posix-like filesystem
245 * @inode: inode to check access rights for
246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
248 * Used to check for read/write/execute permissions on a file.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
251 * are used for other things.
253 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
254 * request cannot be satisfied (eg. requires blocking or too much complexity).
255 * It would then be called again in ref-walk mode.
257 int generic_permission(struct inode *inode, int mask)
262 * Do the basic permission checks.
264 ret = acl_permission_check(inode, mask);
268 if (S_ISDIR(inode->i_mode)) {
269 /* DACs are overridable for directories */
270 if (inode_capable(inode, CAP_DAC_OVERRIDE))
272 if (!(mask & MAY_WRITE))
273 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
278 * Read/write DACs are always overridable.
279 * Executable DACs are overridable when there is
280 * at least one exec bit set.
282 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
283 if (inode_capable(inode, CAP_DAC_OVERRIDE))
287 * Searching includes executable on directories, else just read.
289 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
290 if (mask == MAY_READ)
291 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
298 * We _really_ want to just do "generic_permission()" without
299 * even looking at the inode->i_op values. So we keep a cache
300 * flag in inode->i_opflags, that says "this has not special
301 * permission function, use the fast case".
303 static inline int do_inode_permission(struct inode *inode, int mask)
305 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
306 if (likely(inode->i_op->permission))
307 return inode->i_op->permission(inode, mask);
309 /* This gets set once for the inode lifetime */
310 spin_lock(&inode->i_lock);
311 inode->i_opflags |= IOP_FASTPERM;
312 spin_unlock(&inode->i_lock);
314 return generic_permission(inode, mask);
318 * __inode_permission - Check for access rights to a given inode
319 * @inode: Inode to check permission on
320 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
322 * Check for read/write/execute permissions on an inode.
324 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
326 * This does not check for a read-only file system. You probably want
327 * inode_permission().
329 int __inode_permission(struct inode *inode, int mask)
333 if (unlikely(mask & MAY_WRITE)) {
335 * Nobody gets write access to an immutable file.
337 if (IS_IMMUTABLE(inode))
341 retval = do_inode_permission(inode, mask);
345 retval = devcgroup_inode_permission(inode, mask);
349 return security_inode_permission(inode, mask);
353 * sb_permission - Check superblock-level permissions
354 * @sb: Superblock of inode to check permission on
355 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
357 * Separate out file-system wide checks from inode-specific permission checks.
359 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
361 if (unlikely(mask & MAY_WRITE)) {
362 umode_t mode = inode->i_mode;
364 /* Nobody gets write access to a read-only fs. */
365 if ((sb->s_flags & MS_RDONLY) &&
366 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
373 * inode_permission - Check for access rights to a given inode
374 * @inode: Inode to check permission on
375 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
377 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
378 * this, letting us set arbitrary permissions for filesystem access without
379 * changing the "normal" UIDs which are used for other things.
381 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
383 int inode_permission(struct inode *inode, int mask)
387 retval = sb_permission(inode->i_sb, inode, mask);
390 return __inode_permission(inode, mask);
394 * path_get - get a reference to a path
395 * @path: path to get the reference to
397 * Given a path increment the reference count to the dentry and the vfsmount.
399 void path_get(struct path *path)
404 EXPORT_SYMBOL(path_get);
407 * path_put - put a reference to a path
408 * @path: path to put the reference to
410 * Given a path decrement the reference count to the dentry and the vfsmount.
412 void path_put(struct path *path)
417 EXPORT_SYMBOL(path_put);
420 * Path walking has 2 modes, rcu-walk and ref-walk (see
421 * Documentation/filesystems/path-lookup.txt). In situations when we can't
422 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
423 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
424 * mode. Refcounts are grabbed at the last known good point before rcu-walk
425 * got stuck, so ref-walk may continue from there. If this is not successful
426 * (eg. a seqcount has changed), then failure is returned and it's up to caller
427 * to restart the path walk from the beginning in ref-walk mode.
431 * unlazy_walk - try to switch to ref-walk mode.
432 * @nd: nameidata pathwalk data
433 * @dentry: child of nd->path.dentry or NULL
434 * Returns: 0 on success, -ECHILD on failure
436 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
437 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
438 * @nd or NULL. Must be called from rcu-walk context.
440 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
442 struct fs_struct *fs = current->fs;
443 struct dentry *parent = nd->path.dentry;
446 BUG_ON(!(nd->flags & LOOKUP_RCU));
447 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
449 spin_lock(&fs->lock);
450 if (nd->root.mnt != fs->root.mnt ||
451 nd->root.dentry != fs->root.dentry)
454 spin_lock(&parent->d_lock);
456 if (!__d_rcu_to_refcount(parent, nd->seq))
458 BUG_ON(nd->inode != parent->d_inode);
460 if (dentry->d_parent != parent)
462 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
463 if (!__d_rcu_to_refcount(dentry, nd->seq))
466 * If the sequence check on the child dentry passed, then
467 * the child has not been removed from its parent. This
468 * means the parent dentry must be valid and able to take
469 * a reference at this point.
471 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
472 BUG_ON(!parent->d_count);
474 spin_unlock(&dentry->d_lock);
476 spin_unlock(&parent->d_lock);
479 spin_unlock(&fs->lock);
481 mntget(nd->path.mnt);
484 br_read_unlock(&vfsmount_lock);
485 nd->flags &= ~LOOKUP_RCU;
489 spin_unlock(&dentry->d_lock);
491 spin_unlock(&parent->d_lock);
494 spin_unlock(&fs->lock);
498 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
500 return dentry->d_op->d_revalidate(dentry, flags);
504 * complete_walk - successful completion of path walk
505 * @nd: pointer nameidata
507 * If we had been in RCU mode, drop out of it and legitimize nd->path.
508 * Revalidate the final result, unless we'd already done that during
509 * the path walk or the filesystem doesn't ask for it. Return 0 on
510 * success, -error on failure. In case of failure caller does not
511 * need to drop nd->path.
513 static int complete_walk(struct nameidata *nd)
515 struct dentry *dentry = nd->path.dentry;
518 if (nd->flags & LOOKUP_RCU) {
519 nd->flags &= ~LOOKUP_RCU;
520 if (!(nd->flags & LOOKUP_ROOT))
522 spin_lock(&dentry->d_lock);
523 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
524 spin_unlock(&dentry->d_lock);
526 br_read_unlock(&vfsmount_lock);
529 BUG_ON(nd->inode != dentry->d_inode);
530 spin_unlock(&dentry->d_lock);
531 mntget(nd->path.mnt);
533 br_read_unlock(&vfsmount_lock);
536 if (likely(!(nd->flags & LOOKUP_JUMPED)))
539 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
542 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
545 /* Note: we do not d_invalidate() */
546 status = d_revalidate(dentry, nd->flags);
557 static __always_inline void set_root(struct nameidata *nd)
560 get_fs_root(current->fs, &nd->root);
563 static int link_path_walk(const char *, struct nameidata *);
565 static __always_inline void set_root_rcu(struct nameidata *nd)
568 struct fs_struct *fs = current->fs;
572 seq = read_seqcount_begin(&fs->seq);
574 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
575 } while (read_seqcount_retry(&fs->seq, seq));
579 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
591 nd->flags |= LOOKUP_JUMPED;
593 nd->inode = nd->path.dentry->d_inode;
595 ret = link_path_walk(link, nd);
599 return PTR_ERR(link);
602 static void path_put_conditional(struct path *path, struct nameidata *nd)
605 if (path->mnt != nd->path.mnt)
609 static inline void path_to_nameidata(const struct path *path,
610 struct nameidata *nd)
612 if (!(nd->flags & LOOKUP_RCU)) {
613 dput(nd->path.dentry);
614 if (nd->path.mnt != path->mnt)
615 mntput(nd->path.mnt);
617 nd->path.mnt = path->mnt;
618 nd->path.dentry = path->dentry;
622 * Helper to directly jump to a known parsed path from ->follow_link,
623 * caller must have taken a reference to path beforehand.
625 void nd_jump_link(struct nameidata *nd, struct path *path)
630 nd->inode = nd->path.dentry->d_inode;
631 nd->flags |= LOOKUP_JUMPED;
633 BUG_ON(nd->inode->i_op->follow_link);
636 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
638 struct inode *inode = link->dentry->d_inode;
639 if (inode->i_op->put_link)
640 inode->i_op->put_link(link->dentry, nd, cookie);
644 static __always_inline int
645 follow_link(struct path *link, struct nameidata *nd, void **p)
647 struct dentry *dentry = link->dentry;
651 BUG_ON(nd->flags & LOOKUP_RCU);
653 if (link->mnt == nd->path.mnt)
657 if (unlikely(current->total_link_count >= 40))
658 goto out_put_nd_path;
661 current->total_link_count++;
664 nd_set_link(nd, NULL);
666 error = security_inode_follow_link(link->dentry, nd);
668 goto out_put_nd_path;
670 nd->last_type = LAST_BIND;
671 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
674 goto out_put_nd_path;
679 error = __vfs_follow_link(nd, s);
681 put_link(nd, link, *p);
692 static int follow_up_rcu(struct path *path)
694 struct mount *mnt = real_mount(path->mnt);
695 struct mount *parent;
696 struct dentry *mountpoint;
698 parent = mnt->mnt_parent;
699 if (&parent->mnt == path->mnt)
701 mountpoint = mnt->mnt_mountpoint;
702 path->dentry = mountpoint;
703 path->mnt = &parent->mnt;
708 * follow_up - Find the mountpoint of path's vfsmount
710 * Given a path, find the mountpoint of its source file system.
711 * Replace @path with the path of the mountpoint in the parent mount.
714 * Return 1 if we went up a level and 0 if we were already at the
717 int follow_up(struct path *path)
719 struct mount *mnt = real_mount(path->mnt);
720 struct mount *parent;
721 struct dentry *mountpoint;
723 br_read_lock(&vfsmount_lock);
724 parent = mnt->mnt_parent;
725 if (&parent->mnt == path->mnt) {
726 br_read_unlock(&vfsmount_lock);
729 mntget(&parent->mnt);
730 mountpoint = dget(mnt->mnt_mountpoint);
731 br_read_unlock(&vfsmount_lock);
733 path->dentry = mountpoint;
735 path->mnt = &parent->mnt;
740 * Perform an automount
741 * - return -EISDIR to tell follow_managed() to stop and return the path we
744 static int follow_automount(struct path *path, unsigned flags,
747 struct vfsmount *mnt;
750 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
753 /* We don't want to mount if someone's just doing a stat -
754 * unless they're stat'ing a directory and appended a '/' to
757 * We do, however, want to mount if someone wants to open or
758 * create a file of any type under the mountpoint, wants to
759 * traverse through the mountpoint or wants to open the
760 * mounted directory. Also, autofs may mark negative dentries
761 * as being automount points. These will need the attentions
762 * of the daemon to instantiate them before they can be used.
764 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
765 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
766 path->dentry->d_inode)
769 current->total_link_count++;
770 if (current->total_link_count >= 40)
773 mnt = path->dentry->d_op->d_automount(path);
776 * The filesystem is allowed to return -EISDIR here to indicate
777 * it doesn't want to automount. For instance, autofs would do
778 * this so that its userspace daemon can mount on this dentry.
780 * However, we can only permit this if it's a terminal point in
781 * the path being looked up; if it wasn't then the remainder of
782 * the path is inaccessible and we should say so.
784 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
789 if (!mnt) /* mount collision */
793 /* lock_mount() may release path->mnt on error */
797 err = finish_automount(mnt, path);
801 /* Someone else made a mount here whilst we were busy */
806 path->dentry = dget(mnt->mnt_root);
815 * Handle a dentry that is managed in some way.
816 * - Flagged for transit management (autofs)
817 * - Flagged as mountpoint
818 * - Flagged as automount point
820 * This may only be called in refwalk mode.
822 * Serialization is taken care of in namespace.c
824 static int follow_managed(struct path *path, unsigned flags)
826 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
828 bool need_mntput = false;
831 /* Given that we're not holding a lock here, we retain the value in a
832 * local variable for each dentry as we look at it so that we don't see
833 * the components of that value change under us */
834 while (managed = ACCESS_ONCE(path->dentry->d_flags),
835 managed &= DCACHE_MANAGED_DENTRY,
836 unlikely(managed != 0)) {
837 /* Allow the filesystem to manage the transit without i_mutex
839 if (managed & DCACHE_MANAGE_TRANSIT) {
840 BUG_ON(!path->dentry->d_op);
841 BUG_ON(!path->dentry->d_op->d_manage);
842 ret = path->dentry->d_op->d_manage(path->dentry, false);
847 /* Transit to a mounted filesystem. */
848 if (managed & DCACHE_MOUNTED) {
849 struct vfsmount *mounted = lookup_mnt(path);
855 path->dentry = dget(mounted->mnt_root);
860 /* Something is mounted on this dentry in another
861 * namespace and/or whatever was mounted there in this
862 * namespace got unmounted before we managed to get the
866 /* Handle an automount point */
867 if (managed & DCACHE_NEED_AUTOMOUNT) {
868 ret = follow_automount(path, flags, &need_mntput);
874 /* We didn't change the current path point */
878 if (need_mntput && path->mnt == mnt)
882 return ret < 0 ? ret : need_mntput;
885 int follow_down_one(struct path *path)
887 struct vfsmount *mounted;
889 mounted = lookup_mnt(path);
894 path->dentry = dget(mounted->mnt_root);
900 static inline bool managed_dentry_might_block(struct dentry *dentry)
902 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
903 dentry->d_op->d_manage(dentry, true) < 0);
907 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
908 * we meet a managed dentry that would need blocking.
910 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
911 struct inode **inode)
914 struct mount *mounted;
916 * Don't forget we might have a non-mountpoint managed dentry
917 * that wants to block transit.
919 if (unlikely(managed_dentry_might_block(path->dentry)))
922 if (!d_mountpoint(path->dentry))
925 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
928 path->mnt = &mounted->mnt;
929 path->dentry = mounted->mnt.mnt_root;
930 nd->flags |= LOOKUP_JUMPED;
931 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
933 * Update the inode too. We don't need to re-check the
934 * dentry sequence number here after this d_inode read,
935 * because a mount-point is always pinned.
937 *inode = path->dentry->d_inode;
942 static void follow_mount_rcu(struct nameidata *nd)
944 while (d_mountpoint(nd->path.dentry)) {
945 struct mount *mounted;
946 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
949 nd->path.mnt = &mounted->mnt;
950 nd->path.dentry = mounted->mnt.mnt_root;
951 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
955 static int follow_dotdot_rcu(struct nameidata *nd)
960 if (nd->path.dentry == nd->root.dentry &&
961 nd->path.mnt == nd->root.mnt) {
964 if (nd->path.dentry != nd->path.mnt->mnt_root) {
965 struct dentry *old = nd->path.dentry;
966 struct dentry *parent = old->d_parent;
969 seq = read_seqcount_begin(&parent->d_seq);
970 if (read_seqcount_retry(&old->d_seq, nd->seq))
972 nd->path.dentry = parent;
976 if (!follow_up_rcu(&nd->path))
978 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
980 follow_mount_rcu(nd);
981 nd->inode = nd->path.dentry->d_inode;
985 nd->flags &= ~LOOKUP_RCU;
986 if (!(nd->flags & LOOKUP_ROOT))
989 br_read_unlock(&vfsmount_lock);
994 * Follow down to the covering mount currently visible to userspace. At each
995 * point, the filesystem owning that dentry may be queried as to whether the
996 * caller is permitted to proceed or not.
998 int follow_down(struct path *path)
1003 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1004 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1005 /* Allow the filesystem to manage the transit without i_mutex
1008 * We indicate to the filesystem if someone is trying to mount
1009 * something here. This gives autofs the chance to deny anyone
1010 * other than its daemon the right to mount on its
1013 * The filesystem may sleep at this point.
1015 if (managed & DCACHE_MANAGE_TRANSIT) {
1016 BUG_ON(!path->dentry->d_op);
1017 BUG_ON(!path->dentry->d_op->d_manage);
1018 ret = path->dentry->d_op->d_manage(
1019 path->dentry, false);
1021 return ret == -EISDIR ? 0 : ret;
1024 /* Transit to a mounted filesystem. */
1025 if (managed & DCACHE_MOUNTED) {
1026 struct vfsmount *mounted = lookup_mnt(path);
1031 path->mnt = mounted;
1032 path->dentry = dget(mounted->mnt_root);
1036 /* Don't handle automount points here */
1043 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1045 static void follow_mount(struct path *path)
1047 while (d_mountpoint(path->dentry)) {
1048 struct vfsmount *mounted = lookup_mnt(path);
1053 path->mnt = mounted;
1054 path->dentry = dget(mounted->mnt_root);
1058 static void follow_dotdot(struct nameidata *nd)
1063 struct dentry *old = nd->path.dentry;
1065 if (nd->path.dentry == nd->root.dentry &&
1066 nd->path.mnt == nd->root.mnt) {
1069 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1070 /* rare case of legitimate dget_parent()... */
1071 nd->path.dentry = dget_parent(nd->path.dentry);
1075 if (!follow_up(&nd->path))
1078 follow_mount(&nd->path);
1079 nd->inode = nd->path.dentry->d_inode;
1083 * This looks up the name in dcache, possibly revalidates the old dentry and
1084 * allocates a new one if not found or not valid. In the need_lookup argument
1085 * returns whether i_op->lookup is necessary.
1087 * dir->d_inode->i_mutex must be held
1089 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1090 unsigned int flags, bool *need_lookup)
1092 struct dentry *dentry;
1095 *need_lookup = false;
1096 dentry = d_lookup(dir, name);
1098 if (d_need_lookup(dentry)) {
1099 *need_lookup = true;
1100 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1101 error = d_revalidate(dentry, flags);
1102 if (unlikely(error <= 0)) {
1105 return ERR_PTR(error);
1106 } else if (!d_invalidate(dentry)) {
1115 dentry = d_alloc(dir, name);
1116 if (unlikely(!dentry))
1117 return ERR_PTR(-ENOMEM);
1119 *need_lookup = true;
1125 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1126 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1128 * dir->d_inode->i_mutex must be held
1130 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1135 /* Don't create child dentry for a dead directory. */
1136 if (unlikely(IS_DEADDIR(dir))) {
1138 return ERR_PTR(-ENOENT);
1141 old = dir->i_op->lookup(dir, dentry, flags);
1142 if (unlikely(old)) {
1149 static struct dentry *__lookup_hash(struct qstr *name,
1150 struct dentry *base, unsigned int flags)
1153 struct dentry *dentry;
1155 dentry = lookup_dcache(name, base, flags, &need_lookup);
1159 return lookup_real(base->d_inode, dentry, flags);
1163 * It's more convoluted than I'd like it to be, but... it's still fairly
1164 * small and for now I'd prefer to have fast path as straight as possible.
1165 * It _is_ time-critical.
1167 static int lookup_fast(struct nameidata *nd, struct qstr *name,
1168 struct path *path, struct inode **inode)
1170 struct vfsmount *mnt = nd->path.mnt;
1171 struct dentry *dentry, *parent = nd->path.dentry;
1177 * Rename seqlock is not required here because in the off chance
1178 * of a false negative due to a concurrent rename, we're going to
1179 * do the non-racy lookup, below.
1181 if (nd->flags & LOOKUP_RCU) {
1183 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
1188 * This sequence count validates that the inode matches
1189 * the dentry name information from lookup.
1191 *inode = dentry->d_inode;
1192 if (read_seqcount_retry(&dentry->d_seq, seq))
1196 * This sequence count validates that the parent had no
1197 * changes while we did the lookup of the dentry above.
1199 * The memory barrier in read_seqcount_begin of child is
1200 * enough, we can use __read_seqcount_retry here.
1202 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1206 if (unlikely(d_need_lookup(dentry)))
1208 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1209 status = d_revalidate(dentry, nd->flags);
1210 if (unlikely(status <= 0)) {
1211 if (status != -ECHILD)
1217 path->dentry = dentry;
1218 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1220 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1224 if (unlazy_walk(nd, dentry))
1227 dentry = __d_lookup(parent, name);
1230 if (unlikely(!dentry))
1233 if (unlikely(d_need_lookup(dentry))) {
1238 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1239 status = d_revalidate(dentry, nd->flags);
1240 if (unlikely(status <= 0)) {
1245 if (!d_invalidate(dentry)) {
1252 path->dentry = dentry;
1253 err = follow_managed(path, nd->flags);
1254 if (unlikely(err < 0)) {
1255 path_put_conditional(path, nd);
1259 nd->flags |= LOOKUP_JUMPED;
1260 *inode = path->dentry->d_inode;
1267 /* Fast lookup failed, do it the slow way */
1268 static int lookup_slow(struct nameidata *nd, struct qstr *name,
1271 struct dentry *dentry, *parent;
1274 parent = nd->path.dentry;
1275 BUG_ON(nd->inode != parent->d_inode);
1277 mutex_lock(&parent->d_inode->i_mutex);
1278 dentry = __lookup_hash(name, parent, nd->flags);
1279 mutex_unlock(&parent->d_inode->i_mutex);
1281 return PTR_ERR(dentry);
1282 path->mnt = nd->path.mnt;
1283 path->dentry = dentry;
1284 err = follow_managed(path, nd->flags);
1285 if (unlikely(err < 0)) {
1286 path_put_conditional(path, nd);
1290 nd->flags |= LOOKUP_JUMPED;
1294 static inline int may_lookup(struct nameidata *nd)
1296 if (nd->flags & LOOKUP_RCU) {
1297 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1300 if (unlazy_walk(nd, NULL))
1303 return inode_permission(nd->inode, MAY_EXEC);
1306 static inline int handle_dots(struct nameidata *nd, int type)
1308 if (type == LAST_DOTDOT) {
1309 if (nd->flags & LOOKUP_RCU) {
1310 if (follow_dotdot_rcu(nd))
1318 static void terminate_walk(struct nameidata *nd)
1320 if (!(nd->flags & LOOKUP_RCU)) {
1321 path_put(&nd->path);
1323 nd->flags &= ~LOOKUP_RCU;
1324 if (!(nd->flags & LOOKUP_ROOT))
1325 nd->root.mnt = NULL;
1327 br_read_unlock(&vfsmount_lock);
1332 * Do we need to follow links? We _really_ want to be able
1333 * to do this check without having to look at inode->i_op,
1334 * so we keep a cache of "no, this doesn't need follow_link"
1335 * for the common case.
1337 static inline int should_follow_link(struct inode *inode, int follow)
1339 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1340 if (likely(inode->i_op->follow_link))
1343 /* This gets set once for the inode lifetime */
1344 spin_lock(&inode->i_lock);
1345 inode->i_opflags |= IOP_NOFOLLOW;
1346 spin_unlock(&inode->i_lock);
1351 static inline int walk_component(struct nameidata *nd, struct path *path,
1352 struct qstr *name, int type, int follow)
1354 struct inode *inode;
1357 * "." and ".." are special - ".." especially so because it has
1358 * to be able to know about the current root directory and
1359 * parent relationships.
1361 if (unlikely(type != LAST_NORM))
1362 return handle_dots(nd, type);
1363 err = lookup_fast(nd, name, path, &inode);
1364 if (unlikely(err)) {
1368 err = lookup_slow(nd, name, path);
1372 inode = path->dentry->d_inode;
1378 if (should_follow_link(inode, follow)) {
1379 if (nd->flags & LOOKUP_RCU) {
1380 if (unlikely(unlazy_walk(nd, path->dentry))) {
1385 BUG_ON(inode != path->dentry->d_inode);
1388 path_to_nameidata(path, nd);
1393 path_to_nameidata(path, nd);
1400 * This limits recursive symlink follows to 8, while
1401 * limiting consecutive symlinks to 40.
1403 * Without that kind of total limit, nasty chains of consecutive
1404 * symlinks can cause almost arbitrarily long lookups.
1406 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1410 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1411 path_put_conditional(path, nd);
1412 path_put(&nd->path);
1415 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1418 current->link_count++;
1421 struct path link = *path;
1424 res = follow_link(&link, nd, &cookie);
1427 res = walk_component(nd, path, &nd->last,
1428 nd->last_type, LOOKUP_FOLLOW);
1429 put_link(nd, &link, cookie);
1432 current->link_count--;
1438 * We really don't want to look at inode->i_op->lookup
1439 * when we don't have to. So we keep a cache bit in
1440 * the inode ->i_opflags field that says "yes, we can
1441 * do lookup on this inode".
1443 static inline int can_lookup(struct inode *inode)
1445 if (likely(inode->i_opflags & IOP_LOOKUP))
1447 if (likely(!inode->i_op->lookup))
1450 /* We do this once for the lifetime of the inode */
1451 spin_lock(&inode->i_lock);
1452 inode->i_opflags |= IOP_LOOKUP;
1453 spin_unlock(&inode->i_lock);
1458 * We can do the critical dentry name comparison and hashing
1459 * operations one word at a time, but we are limited to:
1461 * - Architectures with fast unaligned word accesses. We could
1462 * do a "get_unaligned()" if this helps and is sufficiently
1465 * - Little-endian machines (so that we can generate the mask
1466 * of low bytes efficiently). Again, we *could* do a byte
1467 * swapping load on big-endian architectures if that is not
1468 * expensive enough to make the optimization worthless.
1470 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1471 * do not trap on the (extremely unlikely) case of a page
1472 * crossing operation.
1474 * - Furthermore, we need an efficient 64-bit compile for the
1475 * 64-bit case in order to generate the "number of bytes in
1476 * the final mask". Again, that could be replaced with a
1477 * efficient population count instruction or similar.
1479 #ifdef CONFIG_DCACHE_WORD_ACCESS
1481 #include <asm/word-at-a-time.h>
1485 static inline unsigned int fold_hash(unsigned long hash)
1487 hash += hash >> (8*sizeof(int));
1491 #else /* 32-bit case */
1493 #define fold_hash(x) (x)
1497 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1499 unsigned long a, mask;
1500 unsigned long hash = 0;
1503 a = load_unaligned_zeropad(name);
1504 if (len < sizeof(unsigned long))
1508 name += sizeof(unsigned long);
1509 len -= sizeof(unsigned long);
1513 mask = ~(~0ul << len*8);
1516 return fold_hash(hash);
1518 EXPORT_SYMBOL(full_name_hash);
1521 * Calculate the length and hash of the path component, and
1522 * return the length of the component;
1524 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1526 unsigned long a, b, adata, bdata, mask, hash, len;
1527 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1530 len = -sizeof(unsigned long);
1532 hash = (hash + a) * 9;
1533 len += sizeof(unsigned long);
1534 a = load_unaligned_zeropad(name+len);
1535 b = a ^ REPEAT_BYTE('/');
1536 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1538 adata = prep_zero_mask(a, adata, &constants);
1539 bdata = prep_zero_mask(b, bdata, &constants);
1541 mask = create_zero_mask(adata | bdata);
1543 hash += a & zero_bytemask(mask);
1544 *hashp = fold_hash(hash);
1546 return len + find_zero(mask);
1551 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1553 unsigned long hash = init_name_hash();
1555 hash = partial_name_hash(*name++, hash);
1556 return end_name_hash(hash);
1558 EXPORT_SYMBOL(full_name_hash);
1561 * We know there's a real path component here of at least
1564 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1566 unsigned long hash = init_name_hash();
1567 unsigned long len = 0, c;
1569 c = (unsigned char)*name;
1572 hash = partial_name_hash(c, hash);
1573 c = (unsigned char)name[len];
1574 } while (c && c != '/');
1575 *hashp = end_name_hash(hash);
1583 * This is the basic name resolution function, turning a pathname into
1584 * the final dentry. We expect 'base' to be positive and a directory.
1586 * Returns 0 and nd will have valid dentry and mnt on success.
1587 * Returns error and drops reference to input namei data on failure.
1589 static int link_path_walk(const char *name, struct nameidata *nd)
1599 /* At this point we know we have a real path component. */
1605 err = may_lookup(nd);
1609 len = hash_name(name, &this.hash);
1614 if (name[0] == '.') switch (len) {
1616 if (name[1] == '.') {
1618 nd->flags |= LOOKUP_JUMPED;
1624 if (likely(type == LAST_NORM)) {
1625 struct dentry *parent = nd->path.dentry;
1626 nd->flags &= ~LOOKUP_JUMPED;
1627 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1628 err = parent->d_op->d_hash(parent, nd->inode,
1636 goto last_component;
1638 * If it wasn't NUL, we know it was '/'. Skip that
1639 * slash, and continue until no more slashes.
1643 } while (unlikely(name[len] == '/'));
1645 goto last_component;
1648 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1653 err = nested_symlink(&next, nd);
1657 if (can_lookup(nd->inode))
1661 /* here ends the main loop */
1665 nd->last_type = type;
1672 static int path_init(int dfd, const char *name, unsigned int flags,
1673 struct nameidata *nd, struct file **fp)
1679 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1680 nd->flags = flags | LOOKUP_JUMPED;
1682 if (flags & LOOKUP_ROOT) {
1683 struct inode *inode = nd->root.dentry->d_inode;
1685 if (!inode->i_op->lookup)
1687 retval = inode_permission(inode, MAY_EXEC);
1691 nd->path = nd->root;
1693 if (flags & LOOKUP_RCU) {
1694 br_read_lock(&vfsmount_lock);
1696 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1698 path_get(&nd->path);
1703 nd->root.mnt = NULL;
1706 if (flags & LOOKUP_RCU) {
1707 br_read_lock(&vfsmount_lock);
1712 path_get(&nd->root);
1714 nd->path = nd->root;
1715 } else if (dfd == AT_FDCWD) {
1716 if (flags & LOOKUP_RCU) {
1717 struct fs_struct *fs = current->fs;
1720 br_read_lock(&vfsmount_lock);
1724 seq = read_seqcount_begin(&fs->seq);
1726 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1727 } while (read_seqcount_retry(&fs->seq, seq));
1729 get_fs_pwd(current->fs, &nd->path);
1732 struct dentry *dentry;
1734 file = fget_raw_light(dfd, &fput_needed);
1739 dentry = file->f_path.dentry;
1743 if (!S_ISDIR(dentry->d_inode->i_mode))
1746 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1751 nd->path = file->f_path;
1752 if (flags & LOOKUP_RCU) {
1755 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1756 br_read_lock(&vfsmount_lock);
1759 path_get(&file->f_path);
1760 fput_light(file, fput_needed);
1764 nd->inode = nd->path.dentry->d_inode;
1768 fput_light(file, fput_needed);
1773 static inline int lookup_last(struct nameidata *nd, struct path *path)
1775 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1776 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1778 nd->flags &= ~LOOKUP_PARENT;
1779 return walk_component(nd, path, &nd->last, nd->last_type,
1780 nd->flags & LOOKUP_FOLLOW);
1783 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1784 static int path_lookupat(int dfd, const char *name,
1785 unsigned int flags, struct nameidata *nd)
1787 struct file *base = NULL;
1792 * Path walking is largely split up into 2 different synchronisation
1793 * schemes, rcu-walk and ref-walk (explained in
1794 * Documentation/filesystems/path-lookup.txt). These share much of the
1795 * path walk code, but some things particularly setup, cleanup, and
1796 * following mounts are sufficiently divergent that functions are
1797 * duplicated. Typically there is a function foo(), and its RCU
1798 * analogue, foo_rcu().
1800 * -ECHILD is the error number of choice (just to avoid clashes) that
1801 * is returned if some aspect of an rcu-walk fails. Such an error must
1802 * be handled by restarting a traditional ref-walk (which will always
1803 * be able to complete).
1805 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1810 current->total_link_count = 0;
1811 err = link_path_walk(name, nd);
1813 if (!err && !(flags & LOOKUP_PARENT)) {
1814 err = lookup_last(nd, &path);
1817 struct path link = path;
1818 nd->flags |= LOOKUP_PARENT;
1819 err = follow_link(&link, nd, &cookie);
1822 err = lookup_last(nd, &path);
1823 put_link(nd, &link, cookie);
1828 err = complete_walk(nd);
1830 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1831 if (!nd->inode->i_op->lookup) {
1832 path_put(&nd->path);
1840 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1841 path_put(&nd->root);
1842 nd->root.mnt = NULL;
1847 static int do_path_lookup(int dfd, const char *name,
1848 unsigned int flags, struct nameidata *nd)
1850 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1851 if (unlikely(retval == -ECHILD))
1852 retval = path_lookupat(dfd, name, flags, nd);
1853 if (unlikely(retval == -ESTALE))
1854 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1856 if (likely(!retval)) {
1857 if (unlikely(!audit_dummy_context())) {
1858 if (nd->path.dentry && nd->inode)
1859 audit_inode(name, nd->path.dentry);
1865 /* does lookup, returns the object with parent locked */
1866 struct dentry *kern_path_locked(const char *name, struct path *path)
1868 struct nameidata nd;
1870 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
1872 return ERR_PTR(err);
1873 if (nd.last_type != LAST_NORM) {
1875 return ERR_PTR(-EINVAL);
1877 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1878 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
1880 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1888 int kern_path(const char *name, unsigned int flags, struct path *path)
1890 struct nameidata nd;
1891 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1898 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1899 * @dentry: pointer to dentry of the base directory
1900 * @mnt: pointer to vfs mount of the base directory
1901 * @name: pointer to file name
1902 * @flags: lookup flags
1903 * @path: pointer to struct path to fill
1905 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1906 const char *name, unsigned int flags,
1909 struct nameidata nd;
1911 nd.root.dentry = dentry;
1913 BUG_ON(flags & LOOKUP_PARENT);
1914 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1915 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1922 * Restricted form of lookup. Doesn't follow links, single-component only,
1923 * needs parent already locked. Doesn't follow mounts.
1926 static struct dentry *lookup_hash(struct nameidata *nd)
1928 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
1932 * lookup_one_len - filesystem helper to lookup single pathname component
1933 * @name: pathname component to lookup
1934 * @base: base directory to lookup from
1935 * @len: maximum length @len should be interpreted to
1937 * Note that this routine is purely a helper for filesystem usage and should
1938 * not be called by generic code. Also note that by using this function the
1939 * nameidata argument is passed to the filesystem methods and a filesystem
1940 * using this helper needs to be prepared for that.
1942 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1948 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1952 this.hash = full_name_hash(name, len);
1954 return ERR_PTR(-EACCES);
1957 c = *(const unsigned char *)name++;
1958 if (c == '/' || c == '\0')
1959 return ERR_PTR(-EACCES);
1962 * See if the low-level filesystem might want
1963 * to use its own hash..
1965 if (base->d_flags & DCACHE_OP_HASH) {
1966 int err = base->d_op->d_hash(base, base->d_inode, &this);
1968 return ERR_PTR(err);
1971 err = inode_permission(base->d_inode, MAY_EXEC);
1973 return ERR_PTR(err);
1975 return __lookup_hash(&this, base, 0);
1978 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1979 struct path *path, int *empty)
1981 struct nameidata nd;
1982 char *tmp = getname_flags(name, flags, empty);
1983 int err = PTR_ERR(tmp);
1986 BUG_ON(flags & LOOKUP_PARENT);
1988 err = do_path_lookup(dfd, tmp, flags, &nd);
1996 int user_path_at(int dfd, const char __user *name, unsigned flags,
1999 return user_path_at_empty(dfd, name, flags, path, NULL);
2002 static int user_path_parent(int dfd, const char __user *path,
2003 struct nameidata *nd, char **name)
2005 char *s = getname(path);
2011 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
2021 * It's inline, so penalty for filesystems that don't use sticky bit is
2024 static inline int check_sticky(struct inode *dir, struct inode *inode)
2026 kuid_t fsuid = current_fsuid();
2028 if (!(dir->i_mode & S_ISVTX))
2030 if (uid_eq(inode->i_uid, fsuid))
2032 if (uid_eq(dir->i_uid, fsuid))
2034 return !inode_capable(inode, CAP_FOWNER);
2038 * Check whether we can remove a link victim from directory dir, check
2039 * whether the type of victim is right.
2040 * 1. We can't do it if dir is read-only (done in permission())
2041 * 2. We should have write and exec permissions on dir
2042 * 3. We can't remove anything from append-only dir
2043 * 4. We can't do anything with immutable dir (done in permission())
2044 * 5. If the sticky bit on dir is set we should either
2045 * a. be owner of dir, or
2046 * b. be owner of victim, or
2047 * c. have CAP_FOWNER capability
2048 * 6. If the victim is append-only or immutable we can't do antyhing with
2049 * links pointing to it.
2050 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2051 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2052 * 9. We can't remove a root or mountpoint.
2053 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2054 * nfs_async_unlink().
2056 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2060 if (!victim->d_inode)
2063 BUG_ON(victim->d_parent->d_inode != dir);
2064 audit_inode_child(victim, dir);
2066 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2071 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2072 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2075 if (!S_ISDIR(victim->d_inode->i_mode))
2077 if (IS_ROOT(victim))
2079 } else if (S_ISDIR(victim->d_inode->i_mode))
2081 if (IS_DEADDIR(dir))
2083 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2088 /* Check whether we can create an object with dentry child in directory
2090 * 1. We can't do it if child already exists (open has special treatment for
2091 * this case, but since we are inlined it's OK)
2092 * 2. We can't do it if dir is read-only (done in permission())
2093 * 3. We should have write and exec permissions on dir
2094 * 4. We can't do it if dir is immutable (done in permission())
2096 static inline int may_create(struct inode *dir, struct dentry *child)
2100 if (IS_DEADDIR(dir))
2102 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2106 * p1 and p2 should be directories on the same fs.
2108 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2113 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2117 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2119 p = d_ancestor(p2, p1);
2121 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2122 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2126 p = d_ancestor(p1, p2);
2128 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2129 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2133 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2134 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2138 void unlock_rename(struct dentry *p1, struct dentry *p2)
2140 mutex_unlock(&p1->d_inode->i_mutex);
2142 mutex_unlock(&p2->d_inode->i_mutex);
2143 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2147 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2150 int error = may_create(dir, dentry);
2154 if (!dir->i_op->create)
2155 return -EACCES; /* shouldn't it be ENOSYS? */
2158 error = security_inode_create(dir, dentry, mode);
2161 error = dir->i_op->create(dir, dentry, mode, want_excl);
2163 fsnotify_create(dir, dentry);
2167 static int may_open(struct path *path, int acc_mode, int flag)
2169 struct dentry *dentry = path->dentry;
2170 struct inode *inode = dentry->d_inode;
2180 switch (inode->i_mode & S_IFMT) {
2184 if (acc_mode & MAY_WRITE)
2189 if (path->mnt->mnt_flags & MNT_NODEV)
2198 error = inode_permission(inode, acc_mode);
2203 * An append-only file must be opened in append mode for writing.
2205 if (IS_APPEND(inode)) {
2206 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2212 /* O_NOATIME can only be set by the owner or superuser */
2213 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2219 static int handle_truncate(struct file *filp)
2221 struct path *path = &filp->f_path;
2222 struct inode *inode = path->dentry->d_inode;
2223 int error = get_write_access(inode);
2227 * Refuse to truncate files with mandatory locks held on them.
2229 error = locks_verify_locked(inode);
2231 error = security_path_truncate(path);
2233 error = do_truncate(path->dentry, 0,
2234 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2237 put_write_access(inode);
2241 static inline int open_to_namei_flags(int flag)
2243 if ((flag & O_ACCMODE) == 3)
2248 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2250 int error = security_path_mknod(dir, dentry, mode, 0);
2254 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2258 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2262 * Attempt to atomically look up, create and open a file from a negative
2265 * Returns 0 if successful. The file will have been created and attached to
2266 * @file by the filesystem calling finish_open().
2268 * Returns 1 if the file was looked up only or didn't need creating. The
2269 * caller will need to perform the open themselves. @path will have been
2270 * updated to point to the new dentry. This may be negative.
2272 * Returns an error code otherwise.
2274 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2275 struct path *path, struct file *file,
2276 const struct open_flags *op,
2277 bool *want_write, bool need_lookup,
2280 struct inode *dir = nd->path.dentry->d_inode;
2281 unsigned open_flag = open_to_namei_flags(op->open_flag);
2285 int create_error = 0;
2286 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2288 BUG_ON(dentry->d_inode);
2290 /* Don't create child dentry for a dead directory. */
2291 if (unlikely(IS_DEADDIR(dir))) {
2296 mode = op->mode & S_IALLUGO;
2297 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2298 mode &= ~current_umask();
2300 if (open_flag & O_EXCL) {
2301 open_flag &= ~O_TRUNC;
2302 *opened |= FILE_CREATED;
2306 * Checking write permission is tricky, bacuse we don't know if we are
2307 * going to actually need it: O_CREAT opens should work as long as the
2308 * file exists. But checking existence breaks atomicity. The trick is
2309 * to check access and if not granted clear O_CREAT from the flags.
2311 * Another problem is returing the "right" error value (e.g. for an
2312 * O_EXCL open we want to return EEXIST not EROFS).
2314 if ((open_flag & (O_CREAT | O_TRUNC)) ||
2315 (open_flag & O_ACCMODE) != O_RDONLY) {
2316 error = mnt_want_write(nd->path.mnt);
2319 } else if (!(open_flag & O_CREAT)) {
2321 * No O_CREATE -> atomicity not a requirement -> fall
2322 * back to lookup + open
2325 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2326 /* Fall back and fail with the right error */
2327 create_error = error;
2330 /* No side effects, safe to clear O_CREAT */
2331 create_error = error;
2332 open_flag &= ~O_CREAT;
2336 if (open_flag & O_CREAT) {
2337 error = may_o_create(&nd->path, dentry, op->mode);
2339 create_error = error;
2340 if (open_flag & O_EXCL)
2342 open_flag &= ~O_CREAT;
2346 if (nd->flags & LOOKUP_DIRECTORY)
2347 open_flag |= O_DIRECTORY;
2349 file->f_path.dentry = DENTRY_NOT_SET;
2350 file->f_path.mnt = nd->path.mnt;
2351 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2354 if (create_error && error == -ENOENT)
2355 error = create_error;
2359 acc_mode = op->acc_mode;
2360 if (*opened & FILE_CREATED) {
2361 fsnotify_create(dir, dentry);
2362 acc_mode = MAY_OPEN;
2365 if (error) { /* returned 1, that is */
2366 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2370 if (file->f_path.dentry) {
2372 dentry = file->f_path.dentry;
2378 * We didn't have the inode before the open, so check open permission
2381 error = may_open(&file->f_path, acc_mode, open_flag);
2391 dentry = lookup_real(dir, dentry, nd->flags);
2393 return PTR_ERR(dentry);
2396 int open_flag = op->open_flag;
2398 error = create_error;
2399 if ((open_flag & O_EXCL)) {
2400 if (!dentry->d_inode)
2402 } else if (!dentry->d_inode) {
2404 } else if ((open_flag & O_TRUNC) &&
2405 S_ISREG(dentry->d_inode->i_mode)) {
2408 /* will fail later, go on to get the right error */
2412 path->dentry = dentry;
2413 path->mnt = nd->path.mnt;
2418 * Look up and maybe create and open the last component.
2420 * Must be called with i_mutex held on parent.
2422 * Returns 0 if the file was successfully atomically created (if necessary) and
2423 * opened. In this case the file will be returned attached to @file.
2425 * Returns 1 if the file was not completely opened at this time, though lookups
2426 * and creations will have been performed and the dentry returned in @path will
2427 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2428 * specified then a negative dentry may be returned.
2430 * An error code is returned otherwise.
2432 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2433 * cleared otherwise prior to returning.
2435 static int lookup_open(struct nameidata *nd, struct path *path,
2437 const struct open_flags *op,
2438 bool *want_write, int *opened)
2440 struct dentry *dir = nd->path.dentry;
2441 struct inode *dir_inode = dir->d_inode;
2442 struct dentry *dentry;
2446 *opened &= ~FILE_CREATED;
2447 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2449 return PTR_ERR(dentry);
2451 /* Cached positive dentry: will open in f_op->open */
2452 if (!need_lookup && dentry->d_inode)
2455 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2456 return atomic_open(nd, dentry, path, file, op, want_write,
2457 need_lookup, opened);
2461 BUG_ON(dentry->d_inode);
2463 dentry = lookup_real(dir_inode, dentry, nd->flags);
2465 return PTR_ERR(dentry);
2468 /* Negative dentry, just create the file */
2469 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2470 umode_t mode = op->mode;
2471 if (!IS_POSIXACL(dir->d_inode))
2472 mode &= ~current_umask();
2474 * This write is needed to ensure that a
2475 * rw->ro transition does not occur between
2476 * the time when the file is created and when
2477 * a permanent write count is taken through
2478 * the 'struct file' in finish_open().
2480 error = mnt_want_write(nd->path.mnt);
2484 *opened |= FILE_CREATED;
2485 error = security_path_mknod(&nd->path, dentry, mode, 0);
2488 error = vfs_create(dir->d_inode, dentry, mode,
2489 nd->flags & LOOKUP_EXCL);
2494 path->dentry = dentry;
2495 path->mnt = nd->path.mnt;
2504 * Handle the last step of open()
2506 static int do_last(struct nameidata *nd, struct path *path,
2507 struct file *file, const struct open_flags *op,
2508 int *opened, const char *pathname)
2510 struct dentry *dir = nd->path.dentry;
2511 int open_flag = op->open_flag;
2512 bool will_truncate = (open_flag & O_TRUNC) != 0;
2513 bool want_write = false;
2514 int acc_mode = op->acc_mode;
2515 struct inode *inode;
2516 bool symlink_ok = false;
2517 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2518 bool retried = false;
2521 nd->flags &= ~LOOKUP_PARENT;
2522 nd->flags |= op->intent;
2524 switch (nd->last_type) {
2527 error = handle_dots(nd, nd->last_type);
2532 error = complete_walk(nd);
2535 audit_inode(pathname, nd->path.dentry);
2536 if (open_flag & O_CREAT) {
2542 error = complete_walk(nd);
2545 audit_inode(pathname, dir);
2549 if (!(open_flag & O_CREAT)) {
2550 if (nd->last.name[nd->last.len])
2551 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2552 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2554 /* we _can_ be in RCU mode here */
2555 error = lookup_fast(nd, &nd->last, path, &inode);
2562 BUG_ON(nd->inode != dir->d_inode);
2564 /* create side of things */
2566 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2567 * has been cleared when we got to the last component we are
2570 error = complete_walk(nd);
2574 audit_inode(pathname, dir);
2576 /* trailing slashes? */
2577 if (nd->last.name[nd->last.len])
2582 mutex_lock(&dir->d_inode->i_mutex);
2583 error = lookup_open(nd, path, file, op, &want_write, opened);
2584 mutex_unlock(&dir->d_inode->i_mutex);
2590 if ((*opened & FILE_CREATED) ||
2591 !S_ISREG(file->f_path.dentry->d_inode->i_mode))
2592 will_truncate = false;
2594 audit_inode(pathname, file->f_path.dentry);
2598 if (*opened & FILE_CREATED) {
2599 /* Don't check for write permission, don't truncate */
2600 open_flag &= ~O_TRUNC;
2601 will_truncate = false;
2602 acc_mode = MAY_OPEN;
2603 path_to_nameidata(path, nd);
2604 goto finish_open_created;
2608 * It already exists.
2610 audit_inode(pathname, path->dentry);
2613 * If atomic_open() acquired write access it is dropped now due to
2614 * possible mount and symlink following (this might be optimized away if
2618 mnt_drop_write(nd->path.mnt);
2623 if (open_flag & O_EXCL)
2626 error = follow_managed(path, nd->flags);
2631 nd->flags |= LOOKUP_JUMPED;
2633 BUG_ON(nd->flags & LOOKUP_RCU);
2634 inode = path->dentry->d_inode;
2636 /* we _can_ be in RCU mode here */
2639 path_to_nameidata(path, nd);
2643 if (should_follow_link(inode, !symlink_ok)) {
2644 if (nd->flags & LOOKUP_RCU) {
2645 if (unlikely(unlazy_walk(nd, path->dentry))) {
2650 BUG_ON(inode != path->dentry->d_inode);
2654 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2655 path_to_nameidata(path, nd);
2657 save_parent.dentry = nd->path.dentry;
2658 save_parent.mnt = mntget(path->mnt);
2659 nd->path.dentry = path->dentry;
2663 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2664 error = complete_walk(nd);
2666 path_put(&save_parent);
2670 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
2673 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
2675 audit_inode(pathname, nd->path.dentry);
2677 if (!S_ISREG(nd->inode->i_mode))
2678 will_truncate = false;
2680 if (will_truncate) {
2681 error = mnt_want_write(nd->path.mnt);
2686 finish_open_created:
2687 error = may_open(&nd->path, acc_mode, open_flag);
2690 file->f_path.mnt = nd->path.mnt;
2691 error = finish_open(file, nd->path.dentry, NULL, opened);
2693 if (error == -EOPENSTALE)
2698 error = open_check_o_direct(file);
2701 error = ima_file_check(file, op->acc_mode);
2705 if (will_truncate) {
2706 error = handle_truncate(file);
2712 mnt_drop_write(nd->path.mnt);
2713 path_put(&save_parent);
2718 path_put_conditional(path, nd);
2725 /* If no saved parent or already retried then can't retry */
2726 if (!save_parent.dentry || retried)
2729 BUG_ON(save_parent.dentry != dir);
2730 path_put(&nd->path);
2731 nd->path = save_parent;
2732 nd->inode = dir->d_inode;
2733 save_parent.mnt = NULL;
2734 save_parent.dentry = NULL;
2736 mnt_drop_write(nd->path.mnt);
2743 static struct file *path_openat(int dfd, const char *pathname,
2744 struct nameidata *nd, const struct open_flags *op, int flags)
2746 struct file *base = NULL;
2752 file = get_empty_filp();
2754 return ERR_PTR(-ENFILE);
2756 file->f_flags = op->open_flag;
2758 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2759 if (unlikely(error))
2762 current->total_link_count = 0;
2763 error = link_path_walk(pathname, nd);
2764 if (unlikely(error))
2767 error = do_last(nd, &path, file, op, &opened, pathname);
2768 while (unlikely(error > 0)) { /* trailing symlink */
2769 struct path link = path;
2771 if (!(nd->flags & LOOKUP_FOLLOW)) {
2772 path_put_conditional(&path, nd);
2773 path_put(&nd->path);
2777 nd->flags |= LOOKUP_PARENT;
2778 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2779 error = follow_link(&link, nd, &cookie);
2780 if (unlikely(error))
2782 error = do_last(nd, &path, file, op, &opened, pathname);
2783 put_link(nd, &link, cookie);
2786 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2787 path_put(&nd->root);
2790 if (!(opened & FILE_OPENED)) {
2794 if (unlikely(error)) {
2795 if (error == -EOPENSTALE) {
2796 if (flags & LOOKUP_RCU)
2801 file = ERR_PTR(error);
2806 struct file *do_filp_open(int dfd, const char *pathname,
2807 const struct open_flags *op, int flags)
2809 struct nameidata nd;
2812 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2813 if (unlikely(filp == ERR_PTR(-ECHILD)))
2814 filp = path_openat(dfd, pathname, &nd, op, flags);
2815 if (unlikely(filp == ERR_PTR(-ESTALE)))
2816 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2820 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2821 const char *name, const struct open_flags *op, int flags)
2823 struct nameidata nd;
2827 nd.root.dentry = dentry;
2829 flags |= LOOKUP_ROOT;
2831 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2832 return ERR_PTR(-ELOOP);
2834 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2835 if (unlikely(file == ERR_PTR(-ECHILD)))
2836 file = path_openat(-1, name, &nd, op, flags);
2837 if (unlikely(file == ERR_PTR(-ESTALE)))
2838 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2842 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2844 struct dentry *dentry = ERR_PTR(-EEXIST);
2845 struct nameidata nd;
2846 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2848 return ERR_PTR(error);
2851 * Yucky last component or no last component at all?
2852 * (foo/., foo/.., /////)
2854 if (nd.last_type != LAST_NORM)
2856 nd.flags &= ~LOOKUP_PARENT;
2857 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2860 * Do the final lookup.
2862 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2863 dentry = lookup_hash(&nd);
2867 if (dentry->d_inode)
2870 * Special case - lookup gave negative, but... we had foo/bar/
2871 * From the vfs_mknod() POV we just have a negative dentry -
2872 * all is fine. Let's be bastards - you had / on the end, you've
2873 * been asking for (non-existent) directory. -ENOENT for you.
2875 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2877 dentry = ERR_PTR(-ENOENT);
2884 dentry = ERR_PTR(-EEXIST);
2886 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2891 EXPORT_SYMBOL(kern_path_create);
2893 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2895 char *tmp = getname(pathname);
2898 return ERR_CAST(tmp);
2899 res = kern_path_create(dfd, tmp, path, is_dir);
2903 EXPORT_SYMBOL(user_path_create);
2905 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2907 int error = may_create(dir, dentry);
2912 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2915 if (!dir->i_op->mknod)
2918 error = devcgroup_inode_mknod(mode, dev);
2922 error = security_inode_mknod(dir, dentry, mode, dev);
2926 error = dir->i_op->mknod(dir, dentry, mode, dev);
2928 fsnotify_create(dir, dentry);
2932 static int may_mknod(umode_t mode)
2934 switch (mode & S_IFMT) {
2940 case 0: /* zero mode translates to S_IFREG */
2949 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2952 struct dentry *dentry;
2959 dentry = user_path_create(dfd, filename, &path, 0);
2961 return PTR_ERR(dentry);
2963 if (!IS_POSIXACL(path.dentry->d_inode))
2964 mode &= ~current_umask();
2965 error = may_mknod(mode);
2968 error = mnt_want_write(path.mnt);
2971 error = security_path_mknod(&path, dentry, mode, dev);
2973 goto out_drop_write;
2974 switch (mode & S_IFMT) {
2975 case 0: case S_IFREG:
2976 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
2978 case S_IFCHR: case S_IFBLK:
2979 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2980 new_decode_dev(dev));
2982 case S_IFIFO: case S_IFSOCK:
2983 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2987 mnt_drop_write(path.mnt);
2990 mutex_unlock(&path.dentry->d_inode->i_mutex);
2996 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
2998 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3001 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3003 int error = may_create(dir, dentry);
3004 unsigned max_links = dir->i_sb->s_max_links;
3009 if (!dir->i_op->mkdir)
3012 mode &= (S_IRWXUGO|S_ISVTX);
3013 error = security_inode_mkdir(dir, dentry, mode);
3017 if (max_links && dir->i_nlink >= max_links)
3020 error = dir->i_op->mkdir(dir, dentry, mode);
3022 fsnotify_mkdir(dir, dentry);
3026 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3028 struct dentry *dentry;
3032 dentry = user_path_create(dfd, pathname, &path, 1);
3034 return PTR_ERR(dentry);
3036 if (!IS_POSIXACL(path.dentry->d_inode))
3037 mode &= ~current_umask();
3038 error = mnt_want_write(path.mnt);
3041 error = security_path_mkdir(&path, dentry, mode);
3043 goto out_drop_write;
3044 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3046 mnt_drop_write(path.mnt);
3049 mutex_unlock(&path.dentry->d_inode->i_mutex);
3054 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3056 return sys_mkdirat(AT_FDCWD, pathname, mode);
3060 * The dentry_unhash() helper will try to drop the dentry early: we
3061 * should have a usage count of 1 if we're the only user of this
3062 * dentry, and if that is true (possibly after pruning the dcache),
3063 * then we drop the dentry now.
3065 * A low-level filesystem can, if it choses, legally
3068 * if (!d_unhashed(dentry))
3071 * if it cannot handle the case of removing a directory
3072 * that is still in use by something else..
3074 void dentry_unhash(struct dentry *dentry)
3076 shrink_dcache_parent(dentry);
3077 spin_lock(&dentry->d_lock);
3078 if (dentry->d_count == 1)
3080 spin_unlock(&dentry->d_lock);
3083 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3085 int error = may_delete(dir, dentry, 1);
3090 if (!dir->i_op->rmdir)
3094 mutex_lock(&dentry->d_inode->i_mutex);
3097 if (d_mountpoint(dentry))
3100 error = security_inode_rmdir(dir, dentry);
3104 shrink_dcache_parent(dentry);
3105 error = dir->i_op->rmdir(dir, dentry);
3109 dentry->d_inode->i_flags |= S_DEAD;
3113 mutex_unlock(&dentry->d_inode->i_mutex);
3120 static long do_rmdir(int dfd, const char __user *pathname)
3124 struct dentry *dentry;
3125 struct nameidata nd;
3127 error = user_path_parent(dfd, pathname, &nd, &name);
3131 switch(nd.last_type) {
3143 nd.flags &= ~LOOKUP_PARENT;
3145 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3146 dentry = lookup_hash(&nd);
3147 error = PTR_ERR(dentry);
3150 if (!dentry->d_inode) {
3154 error = mnt_want_write(nd.path.mnt);
3157 error = security_path_rmdir(&nd.path, dentry);
3160 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3162 mnt_drop_write(nd.path.mnt);
3166 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3173 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3175 return do_rmdir(AT_FDCWD, pathname);
3178 int vfs_unlink(struct inode *dir, struct dentry *dentry)
3180 int error = may_delete(dir, dentry, 0);
3185 if (!dir->i_op->unlink)
3188 mutex_lock(&dentry->d_inode->i_mutex);
3189 if (d_mountpoint(dentry))
3192 error = security_inode_unlink(dir, dentry);
3194 error = dir->i_op->unlink(dir, dentry);
3199 mutex_unlock(&dentry->d_inode->i_mutex);
3201 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3202 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3203 fsnotify_link_count(dentry->d_inode);
3211 * Make sure that the actual truncation of the file will occur outside its
3212 * directory's i_mutex. Truncate can take a long time if there is a lot of
3213 * writeout happening, and we don't want to prevent access to the directory
3214 * while waiting on the I/O.
3216 static long do_unlinkat(int dfd, const char __user *pathname)
3220 struct dentry *dentry;
3221 struct nameidata nd;
3222 struct inode *inode = NULL;
3224 error = user_path_parent(dfd, pathname, &nd, &name);
3229 if (nd.last_type != LAST_NORM)
3232 nd.flags &= ~LOOKUP_PARENT;
3234 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3235 dentry = lookup_hash(&nd);
3236 error = PTR_ERR(dentry);
3237 if (!IS_ERR(dentry)) {
3238 /* Why not before? Because we want correct error value */
3239 if (nd.last.name[nd.last.len])
3241 inode = dentry->d_inode;
3245 error = mnt_want_write(nd.path.mnt);
3248 error = security_path_unlink(&nd.path, dentry);
3251 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
3253 mnt_drop_write(nd.path.mnt);
3257 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3259 iput(inode); /* truncate the inode here */
3266 error = !dentry->d_inode ? -ENOENT :
3267 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3271 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3273 if ((flag & ~AT_REMOVEDIR) != 0)
3276 if (flag & AT_REMOVEDIR)
3277 return do_rmdir(dfd, pathname);
3279 return do_unlinkat(dfd, pathname);
3282 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3284 return do_unlinkat(AT_FDCWD, pathname);
3287 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3289 int error = may_create(dir, dentry);
3294 if (!dir->i_op->symlink)
3297 error = security_inode_symlink(dir, dentry, oldname);
3301 error = dir->i_op->symlink(dir, dentry, oldname);
3303 fsnotify_create(dir, dentry);
3307 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3308 int, newdfd, const char __user *, newname)
3312 struct dentry *dentry;
3315 from = getname(oldname);
3317 return PTR_ERR(from);
3319 dentry = user_path_create(newdfd, newname, &path, 0);
3320 error = PTR_ERR(dentry);
3324 error = mnt_want_write(path.mnt);
3327 error = security_path_symlink(&path, dentry, from);
3329 goto out_drop_write;
3330 error = vfs_symlink(path.dentry->d_inode, dentry, from);
3332 mnt_drop_write(path.mnt);
3335 mutex_unlock(&path.dentry->d_inode->i_mutex);
3342 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3344 return sys_symlinkat(oldname, AT_FDCWD, newname);
3347 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3349 struct inode *inode = old_dentry->d_inode;
3350 unsigned max_links = dir->i_sb->s_max_links;
3356 error = may_create(dir, new_dentry);
3360 if (dir->i_sb != inode->i_sb)
3364 * A link to an append-only or immutable file cannot be created.
3366 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3368 if (!dir->i_op->link)
3370 if (S_ISDIR(inode->i_mode))
3373 error = security_inode_link(old_dentry, dir, new_dentry);
3377 mutex_lock(&inode->i_mutex);
3378 /* Make sure we don't allow creating hardlink to an unlinked file */
3379 if (inode->i_nlink == 0)
3381 else if (max_links && inode->i_nlink >= max_links)
3384 error = dir->i_op->link(old_dentry, dir, new_dentry);
3385 mutex_unlock(&inode->i_mutex);
3387 fsnotify_link(dir, inode, new_dentry);
3392 * Hardlinks are often used in delicate situations. We avoid
3393 * security-related surprises by not following symlinks on the
3396 * We don't follow them on the oldname either to be compatible
3397 * with linux 2.0, and to avoid hard-linking to directories
3398 * and other special files. --ADM
3400 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3401 int, newdfd, const char __user *, newname, int, flags)
3403 struct dentry *new_dentry;
3404 struct path old_path, new_path;
3408 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3411 * To use null names we require CAP_DAC_READ_SEARCH
3412 * This ensures that not everyone will be able to create
3413 * handlink using the passed filedescriptor.
3415 if (flags & AT_EMPTY_PATH) {
3416 if (!capable(CAP_DAC_READ_SEARCH))
3421 if (flags & AT_SYMLINK_FOLLOW)
3422 how |= LOOKUP_FOLLOW;
3424 error = user_path_at(olddfd, oldname, how, &old_path);
3428 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3429 error = PTR_ERR(new_dentry);
3430 if (IS_ERR(new_dentry))
3434 if (old_path.mnt != new_path.mnt)
3436 error = mnt_want_write(new_path.mnt);
3439 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3441 goto out_drop_write;
3442 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3444 mnt_drop_write(new_path.mnt);
3447 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3448 path_put(&new_path);
3450 path_put(&old_path);
3455 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3457 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3461 * The worst of all namespace operations - renaming directory. "Perverted"
3462 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3464 * a) we can get into loop creation. Check is done in is_subdir().
3465 * b) race potential - two innocent renames can create a loop together.
3466 * That's where 4.4 screws up. Current fix: serialization on
3467 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3469 * c) we have to lock _three_ objects - parents and victim (if it exists).
3470 * And that - after we got ->i_mutex on parents (until then we don't know
3471 * whether the target exists). Solution: try to be smart with locking
3472 * order for inodes. We rely on the fact that tree topology may change
3473 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3474 * move will be locked. Thus we can rank directories by the tree
3475 * (ancestors first) and rank all non-directories after them.
3476 * That works since everybody except rename does "lock parent, lookup,
3477 * lock child" and rename is under ->s_vfs_rename_mutex.
3478 * HOWEVER, it relies on the assumption that any object with ->lookup()
3479 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3480 * we'd better make sure that there's no link(2) for them.
3481 * d) conversion from fhandle to dentry may come in the wrong moment - when
3482 * we are removing the target. Solution: we will have to grab ->i_mutex
3483 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3484 * ->i_mutex on parents, which works but leads to some truly excessive
3487 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3488 struct inode *new_dir, struct dentry *new_dentry)
3491 struct inode *target = new_dentry->d_inode;
3492 unsigned max_links = new_dir->i_sb->s_max_links;
3495 * If we are going to change the parent - check write permissions,
3496 * we'll need to flip '..'.
3498 if (new_dir != old_dir) {
3499 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3504 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3510 mutex_lock(&target->i_mutex);
3513 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3517 if (max_links && !target && new_dir != old_dir &&
3518 new_dir->i_nlink >= max_links)
3522 shrink_dcache_parent(new_dentry);
3523 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3528 target->i_flags |= S_DEAD;
3529 dont_mount(new_dentry);
3533 mutex_unlock(&target->i_mutex);
3536 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3537 d_move(old_dentry,new_dentry);
3541 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3542 struct inode *new_dir, struct dentry *new_dentry)
3544 struct inode *target = new_dentry->d_inode;
3547 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3553 mutex_lock(&target->i_mutex);
3556 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3559 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3564 dont_mount(new_dentry);
3565 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3566 d_move(old_dentry, new_dentry);
3569 mutex_unlock(&target->i_mutex);
3574 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3575 struct inode *new_dir, struct dentry *new_dentry)
3578 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3579 const unsigned char *old_name;
3581 if (old_dentry->d_inode == new_dentry->d_inode)
3584 error = may_delete(old_dir, old_dentry, is_dir);
3588 if (!new_dentry->d_inode)
3589 error = may_create(new_dir, new_dentry);
3591 error = may_delete(new_dir, new_dentry, is_dir);
3595 if (!old_dir->i_op->rename)
3598 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3601 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3603 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3605 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3606 new_dentry->d_inode, old_dentry);
3607 fsnotify_oldname_free(old_name);
3612 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3613 int, newdfd, const char __user *, newname)
3615 struct dentry *old_dir, *new_dir;
3616 struct dentry *old_dentry, *new_dentry;
3617 struct dentry *trap;
3618 struct nameidata oldnd, newnd;
3623 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3627 error = user_path_parent(newdfd, newname, &newnd, &to);
3632 if (oldnd.path.mnt != newnd.path.mnt)
3635 old_dir = oldnd.path.dentry;
3637 if (oldnd.last_type != LAST_NORM)
3640 new_dir = newnd.path.dentry;
3641 if (newnd.last_type != LAST_NORM)
3644 oldnd.flags &= ~LOOKUP_PARENT;
3645 newnd.flags &= ~LOOKUP_PARENT;
3646 newnd.flags |= LOOKUP_RENAME_TARGET;
3648 trap = lock_rename(new_dir, old_dir);
3650 old_dentry = lookup_hash(&oldnd);
3651 error = PTR_ERR(old_dentry);
3652 if (IS_ERR(old_dentry))
3654 /* source must exist */
3656 if (!old_dentry->d_inode)
3658 /* unless the source is a directory trailing slashes give -ENOTDIR */
3659 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3661 if (oldnd.last.name[oldnd.last.len])
3663 if (newnd.last.name[newnd.last.len])
3666 /* source should not be ancestor of target */
3668 if (old_dentry == trap)
3670 new_dentry = lookup_hash(&newnd);
3671 error = PTR_ERR(new_dentry);
3672 if (IS_ERR(new_dentry))
3674 /* target should not be an ancestor of source */
3676 if (new_dentry == trap)
3679 error = mnt_want_write(oldnd.path.mnt);
3682 error = security_path_rename(&oldnd.path, old_dentry,
3683 &newnd.path, new_dentry);
3686 error = vfs_rename(old_dir->d_inode, old_dentry,
3687 new_dir->d_inode, new_dentry);
3689 mnt_drop_write(oldnd.path.mnt);
3695 unlock_rename(new_dir, old_dir);
3697 path_put(&newnd.path);
3700 path_put(&oldnd.path);
3706 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3708 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3711 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3715 len = PTR_ERR(link);
3720 if (len > (unsigned) buflen)
3722 if (copy_to_user(buffer, link, len))
3729 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3730 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3731 * using) it for any given inode is up to filesystem.
3733 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3735 struct nameidata nd;
3740 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3742 return PTR_ERR(cookie);
3744 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3745 if (dentry->d_inode->i_op->put_link)
3746 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3750 int vfs_follow_link(struct nameidata *nd, const char *link)
3752 return __vfs_follow_link(nd, link);
3755 /* get the link contents into pagecache */
3756 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3760 struct address_space *mapping = dentry->d_inode->i_mapping;
3761 page = read_mapping_page(mapping, 0, NULL);
3766 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3770 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3772 struct page *page = NULL;
3773 char *s = page_getlink(dentry, &page);
3774 int res = vfs_readlink(dentry,buffer,buflen,s);
3777 page_cache_release(page);
3782 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3784 struct page *page = NULL;
3785 nd_set_link(nd, page_getlink(dentry, &page));
3789 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3791 struct page *page = cookie;
3795 page_cache_release(page);
3800 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3802 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3804 struct address_space *mapping = inode->i_mapping;
3809 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3811 flags |= AOP_FLAG_NOFS;
3814 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3815 flags, &page, &fsdata);
3819 kaddr = kmap_atomic(page);
3820 memcpy(kaddr, symname, len-1);
3821 kunmap_atomic(kaddr);
3823 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3830 mark_inode_dirty(inode);
3836 int page_symlink(struct inode *inode, const char *symname, int len)
3838 return __page_symlink(inode, symname, len,
3839 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3842 const struct inode_operations page_symlink_inode_operations = {
3843 .readlink = generic_readlink,
3844 .follow_link = page_follow_link_light,
3845 .put_link = page_put_link,
3848 EXPORT_SYMBOL(user_path_at);
3849 EXPORT_SYMBOL(follow_down_one);
3850 EXPORT_SYMBOL(follow_down);
3851 EXPORT_SYMBOL(follow_up);
3852 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3853 EXPORT_SYMBOL(getname);
3854 EXPORT_SYMBOL(lock_rename);
3855 EXPORT_SYMBOL(lookup_one_len);
3856 EXPORT_SYMBOL(page_follow_link_light);
3857 EXPORT_SYMBOL(page_put_link);
3858 EXPORT_SYMBOL(page_readlink);
3859 EXPORT_SYMBOL(__page_symlink);
3860 EXPORT_SYMBOL(page_symlink);
3861 EXPORT_SYMBOL(page_symlink_inode_operations);
3862 EXPORT_SYMBOL(kern_path);
3863 EXPORT_SYMBOL(vfs_path_lookup);
3864 EXPORT_SYMBOL(inode_permission);
3865 EXPORT_SYMBOL(unlock_rename);
3866 EXPORT_SYMBOL(vfs_create);
3867 EXPORT_SYMBOL(vfs_follow_link);
3868 EXPORT_SYMBOL(vfs_link);
3869 EXPORT_SYMBOL(vfs_mkdir);
3870 EXPORT_SYMBOL(vfs_mknod);
3871 EXPORT_SYMBOL(generic_permission);
3872 EXPORT_SYMBOL(vfs_readlink);
3873 EXPORT_SYMBOL(vfs_rename);
3874 EXPORT_SYMBOL(vfs_rmdir);
3875 EXPORT_SYMBOL(vfs_symlink);
3876 EXPORT_SYMBOL(vfs_unlink);
3877 EXPORT_SYMBOL(dentry_unhash);
3878 EXPORT_SYMBOL(generic_readlink);