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.
430 static inline void lock_rcu_walk(void)
432 br_read_lock(&vfsmount_lock);
436 static inline void unlock_rcu_walk(void)
439 br_read_unlock(&vfsmount_lock);
443 * unlazy_walk - try to switch to ref-walk mode.
444 * @nd: nameidata pathwalk data
445 * @dentry: child of nd->path.dentry or NULL
446 * Returns: 0 on success, -ECHILD on failure
448 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
449 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
450 * @nd or NULL. Must be called from rcu-walk context.
452 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
454 struct fs_struct *fs = current->fs;
455 struct dentry *parent = nd->path.dentry;
458 BUG_ON(!(nd->flags & LOOKUP_RCU));
459 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
461 spin_lock(&fs->lock);
462 if (nd->root.mnt != fs->root.mnt ||
463 nd->root.dentry != fs->root.dentry)
466 spin_lock(&parent->d_lock);
468 if (!__d_rcu_to_refcount(parent, nd->seq))
470 BUG_ON(nd->inode != parent->d_inode);
472 if (dentry->d_parent != parent)
474 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
475 if (!__d_rcu_to_refcount(dentry, nd->seq))
478 * If the sequence check on the child dentry passed, then
479 * the child has not been removed from its parent. This
480 * means the parent dentry must be valid and able to take
481 * a reference at this point.
483 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
484 BUG_ON(!parent->d_count);
486 spin_unlock(&dentry->d_lock);
488 spin_unlock(&parent->d_lock);
491 spin_unlock(&fs->lock);
493 mntget(nd->path.mnt);
496 nd->flags &= ~LOOKUP_RCU;
500 spin_unlock(&dentry->d_lock);
502 spin_unlock(&parent->d_lock);
505 spin_unlock(&fs->lock);
509 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
511 return dentry->d_op->d_revalidate(dentry, flags);
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);
539 BUG_ON(nd->inode != dentry->d_inode);
540 spin_unlock(&dentry->d_lock);
541 mntget(nd->path.mnt);
545 if (likely(!(nd->flags & LOOKUP_JUMPED)))
548 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
551 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
554 /* Note: we do not d_invalidate() */
555 status = d_revalidate(dentry, nd->flags);
566 static __always_inline void set_root(struct nameidata *nd)
569 get_fs_root(current->fs, &nd->root);
572 static int link_path_walk(const char *, struct nameidata *);
574 static __always_inline void set_root_rcu(struct nameidata *nd)
577 struct fs_struct *fs = current->fs;
581 seq = read_seqcount_begin(&fs->seq);
583 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
584 } while (read_seqcount_retry(&fs->seq, seq));
588 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
600 nd->flags |= LOOKUP_JUMPED;
602 nd->inode = nd->path.dentry->d_inode;
604 ret = link_path_walk(link, nd);
608 return PTR_ERR(link);
611 static void path_put_conditional(struct path *path, struct nameidata *nd)
614 if (path->mnt != nd->path.mnt)
618 static inline void path_to_nameidata(const struct path *path,
619 struct nameidata *nd)
621 if (!(nd->flags & LOOKUP_RCU)) {
622 dput(nd->path.dentry);
623 if (nd->path.mnt != path->mnt)
624 mntput(nd->path.mnt);
626 nd->path.mnt = path->mnt;
627 nd->path.dentry = path->dentry;
631 * Helper to directly jump to a known parsed path from ->follow_link,
632 * caller must have taken a reference to path beforehand.
634 void nd_jump_link(struct nameidata *nd, struct path *path)
639 nd->inode = nd->path.dentry->d_inode;
640 nd->flags |= LOOKUP_JUMPED;
642 BUG_ON(nd->inode->i_op->follow_link);
645 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
647 struct inode *inode = link->dentry->d_inode;
648 if (inode->i_op->put_link)
649 inode->i_op->put_link(link->dentry, nd, cookie);
653 static __always_inline int
654 follow_link(struct path *link, struct nameidata *nd, void **p)
656 struct dentry *dentry = link->dentry;
660 BUG_ON(nd->flags & LOOKUP_RCU);
662 if (link->mnt == nd->path.mnt)
666 if (unlikely(current->total_link_count >= 40))
667 goto out_put_nd_path;
670 current->total_link_count++;
673 nd_set_link(nd, NULL);
675 error = security_inode_follow_link(link->dentry, nd);
677 goto out_put_nd_path;
679 nd->last_type = LAST_BIND;
680 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
683 goto out_put_nd_path;
688 error = __vfs_follow_link(nd, s);
690 put_link(nd, link, *p);
701 static int follow_up_rcu(struct path *path)
703 struct mount *mnt = real_mount(path->mnt);
704 struct mount *parent;
705 struct dentry *mountpoint;
707 parent = mnt->mnt_parent;
708 if (&parent->mnt == path->mnt)
710 mountpoint = mnt->mnt_mountpoint;
711 path->dentry = mountpoint;
712 path->mnt = &parent->mnt;
717 * follow_up - Find the mountpoint of path's vfsmount
719 * Given a path, find the mountpoint of its source file system.
720 * Replace @path with the path of the mountpoint in the parent mount.
723 * Return 1 if we went up a level and 0 if we were already at the
726 int follow_up(struct path *path)
728 struct mount *mnt = real_mount(path->mnt);
729 struct mount *parent;
730 struct dentry *mountpoint;
732 br_read_lock(&vfsmount_lock);
733 parent = mnt->mnt_parent;
735 br_read_unlock(&vfsmount_lock);
738 mntget(&parent->mnt);
739 mountpoint = dget(mnt->mnt_mountpoint);
740 br_read_unlock(&vfsmount_lock);
742 path->dentry = mountpoint;
744 path->mnt = &parent->mnt;
749 * Perform an automount
750 * - return -EISDIR to tell follow_managed() to stop and return the path we
753 static int follow_automount(struct path *path, unsigned flags,
756 struct vfsmount *mnt;
759 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
762 /* We don't want to mount if someone's just doing a stat -
763 * unless they're stat'ing a directory and appended a '/' to
766 * We do, however, want to mount if someone wants to open or
767 * create a file of any type under the mountpoint, wants to
768 * traverse through the mountpoint or wants to open the
769 * mounted directory. Also, autofs may mark negative dentries
770 * as being automount points. These will need the attentions
771 * of the daemon to instantiate them before they can be used.
773 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
774 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
775 path->dentry->d_inode)
778 current->total_link_count++;
779 if (current->total_link_count >= 40)
782 mnt = path->dentry->d_op->d_automount(path);
785 * The filesystem is allowed to return -EISDIR here to indicate
786 * it doesn't want to automount. For instance, autofs would do
787 * this so that its userspace daemon can mount on this dentry.
789 * However, we can only permit this if it's a terminal point in
790 * the path being looked up; if it wasn't then the remainder of
791 * the path is inaccessible and we should say so.
793 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
798 if (!mnt) /* mount collision */
802 /* lock_mount() may release path->mnt on error */
806 err = finish_automount(mnt, path);
810 /* Someone else made a mount here whilst we were busy */
815 path->dentry = dget(mnt->mnt_root);
824 * Handle a dentry that is managed in some way.
825 * - Flagged for transit management (autofs)
826 * - Flagged as mountpoint
827 * - Flagged as automount point
829 * This may only be called in refwalk mode.
831 * Serialization is taken care of in namespace.c
833 static int follow_managed(struct path *path, unsigned flags)
835 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
837 bool need_mntput = false;
840 /* Given that we're not holding a lock here, we retain the value in a
841 * local variable for each dentry as we look at it so that we don't see
842 * the components of that value change under us */
843 while (managed = ACCESS_ONCE(path->dentry->d_flags),
844 managed &= DCACHE_MANAGED_DENTRY,
845 unlikely(managed != 0)) {
846 /* Allow the filesystem to manage the transit without i_mutex
848 if (managed & DCACHE_MANAGE_TRANSIT) {
849 BUG_ON(!path->dentry->d_op);
850 BUG_ON(!path->dentry->d_op->d_manage);
851 ret = path->dentry->d_op->d_manage(path->dentry, false);
856 /* Transit to a mounted filesystem. */
857 if (managed & DCACHE_MOUNTED) {
858 struct vfsmount *mounted = lookup_mnt(path);
864 path->dentry = dget(mounted->mnt_root);
869 /* Something is mounted on this dentry in another
870 * namespace and/or whatever was mounted there in this
871 * namespace got unmounted before we managed to get the
875 /* Handle an automount point */
876 if (managed & DCACHE_NEED_AUTOMOUNT) {
877 ret = follow_automount(path, flags, &need_mntput);
883 /* We didn't change the current path point */
887 if (need_mntput && path->mnt == mnt)
891 return ret < 0 ? ret : need_mntput;
894 int follow_down_one(struct path *path)
896 struct vfsmount *mounted;
898 mounted = lookup_mnt(path);
903 path->dentry = dget(mounted->mnt_root);
909 static inline bool managed_dentry_might_block(struct dentry *dentry)
911 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
912 dentry->d_op->d_manage(dentry, true) < 0);
916 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
917 * we meet a managed dentry that would need blocking.
919 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
920 struct inode **inode)
923 struct mount *mounted;
925 * Don't forget we might have a non-mountpoint managed dentry
926 * that wants to block transit.
928 if (unlikely(managed_dentry_might_block(path->dentry)))
931 if (!d_mountpoint(path->dentry))
934 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
937 path->mnt = &mounted->mnt;
938 path->dentry = mounted->mnt.mnt_root;
939 nd->flags |= LOOKUP_JUMPED;
940 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
942 * Update the inode too. We don't need to re-check the
943 * dentry sequence number here after this d_inode read,
944 * because a mount-point is always pinned.
946 *inode = path->dentry->d_inode;
951 static void follow_mount_rcu(struct nameidata *nd)
953 while (d_mountpoint(nd->path.dentry)) {
954 struct mount *mounted;
955 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
958 nd->path.mnt = &mounted->mnt;
959 nd->path.dentry = mounted->mnt.mnt_root;
960 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
964 static int follow_dotdot_rcu(struct nameidata *nd)
969 if (nd->path.dentry == nd->root.dentry &&
970 nd->path.mnt == nd->root.mnt) {
973 if (nd->path.dentry != nd->path.mnt->mnt_root) {
974 struct dentry *old = nd->path.dentry;
975 struct dentry *parent = old->d_parent;
978 seq = read_seqcount_begin(&parent->d_seq);
979 if (read_seqcount_retry(&old->d_seq, nd->seq))
981 nd->path.dentry = parent;
985 if (!follow_up_rcu(&nd->path))
987 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
989 follow_mount_rcu(nd);
990 nd->inode = nd->path.dentry->d_inode;
994 nd->flags &= ~LOOKUP_RCU;
995 if (!(nd->flags & LOOKUP_ROOT))
1002 * Follow down to the covering mount currently visible to userspace. At each
1003 * point, the filesystem owning that dentry may be queried as to whether the
1004 * caller is permitted to proceed or not.
1006 int follow_down(struct path *path)
1011 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1012 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1013 /* Allow the filesystem to manage the transit without i_mutex
1016 * We indicate to the filesystem if someone is trying to mount
1017 * something here. This gives autofs the chance to deny anyone
1018 * other than its daemon the right to mount on its
1021 * The filesystem may sleep at this point.
1023 if (managed & DCACHE_MANAGE_TRANSIT) {
1024 BUG_ON(!path->dentry->d_op);
1025 BUG_ON(!path->dentry->d_op->d_manage);
1026 ret = path->dentry->d_op->d_manage(
1027 path->dentry, false);
1029 return ret == -EISDIR ? 0 : ret;
1032 /* Transit to a mounted filesystem. */
1033 if (managed & DCACHE_MOUNTED) {
1034 struct vfsmount *mounted = lookup_mnt(path);
1039 path->mnt = mounted;
1040 path->dentry = dget(mounted->mnt_root);
1044 /* Don't handle automount points here */
1051 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1053 static void follow_mount(struct path *path)
1055 while (d_mountpoint(path->dentry)) {
1056 struct vfsmount *mounted = lookup_mnt(path);
1061 path->mnt = mounted;
1062 path->dentry = dget(mounted->mnt_root);
1066 static void follow_dotdot(struct nameidata *nd)
1071 struct dentry *old = nd->path.dentry;
1073 if (nd->path.dentry == nd->root.dentry &&
1074 nd->path.mnt == nd->root.mnt) {
1077 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1078 /* rare case of legitimate dget_parent()... */
1079 nd->path.dentry = dget_parent(nd->path.dentry);
1083 if (!follow_up(&nd->path))
1086 follow_mount(&nd->path);
1087 nd->inode = nd->path.dentry->d_inode;
1091 * This looks up the name in dcache, possibly revalidates the old dentry and
1092 * allocates a new one if not found or not valid. In the need_lookup argument
1093 * returns whether i_op->lookup is necessary.
1095 * dir->d_inode->i_mutex must be held
1097 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1098 unsigned int flags, bool *need_lookup)
1100 struct dentry *dentry;
1103 *need_lookup = false;
1104 dentry = d_lookup(dir, name);
1106 if (d_need_lookup(dentry)) {
1107 *need_lookup = true;
1108 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1109 error = d_revalidate(dentry, flags);
1110 if (unlikely(error <= 0)) {
1113 return ERR_PTR(error);
1114 } else if (!d_invalidate(dentry)) {
1123 dentry = d_alloc(dir, name);
1124 if (unlikely(!dentry))
1125 return ERR_PTR(-ENOMEM);
1127 *need_lookup = true;
1133 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1134 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1136 * dir->d_inode->i_mutex must be held
1138 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1143 /* Don't create child dentry for a dead directory. */
1144 if (unlikely(IS_DEADDIR(dir))) {
1146 return ERR_PTR(-ENOENT);
1149 old = dir->i_op->lookup(dir, dentry, flags);
1150 if (unlikely(old)) {
1157 static struct dentry *__lookup_hash(struct qstr *name,
1158 struct dentry *base, unsigned int flags)
1161 struct dentry *dentry;
1163 dentry = lookup_dcache(name, base, flags, &need_lookup);
1167 return lookup_real(base->d_inode, dentry, flags);
1171 * It's more convoluted than I'd like it to be, but... it's still fairly
1172 * small and for now I'd prefer to have fast path as straight as possible.
1173 * It _is_ time-critical.
1175 static int lookup_fast(struct nameidata *nd, struct qstr *name,
1176 struct path *path, struct inode **inode)
1178 struct vfsmount *mnt = nd->path.mnt;
1179 struct dentry *dentry, *parent = nd->path.dentry;
1185 * Rename seqlock is not required here because in the off chance
1186 * of a false negative due to a concurrent rename, we're going to
1187 * do the non-racy lookup, below.
1189 if (nd->flags & LOOKUP_RCU) {
1191 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
1196 * This sequence count validates that the inode matches
1197 * the dentry name information from lookup.
1199 *inode = dentry->d_inode;
1200 if (read_seqcount_retry(&dentry->d_seq, seq))
1204 * This sequence count validates that the parent had no
1205 * changes while we did the lookup of the dentry above.
1207 * The memory barrier in read_seqcount_begin of child is
1208 * enough, we can use __read_seqcount_retry here.
1210 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1214 if (unlikely(d_need_lookup(dentry)))
1216 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1217 status = d_revalidate(dentry, nd->flags);
1218 if (unlikely(status <= 0)) {
1219 if (status != -ECHILD)
1225 path->dentry = dentry;
1226 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1228 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1232 if (unlazy_walk(nd, dentry))
1235 dentry = __d_lookup(parent, name);
1238 if (unlikely(!dentry))
1241 if (unlikely(d_need_lookup(dentry))) {
1246 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1247 status = d_revalidate(dentry, nd->flags);
1248 if (unlikely(status <= 0)) {
1253 if (!d_invalidate(dentry)) {
1260 path->dentry = dentry;
1261 err = follow_managed(path, nd->flags);
1262 if (unlikely(err < 0)) {
1263 path_put_conditional(path, nd);
1267 nd->flags |= LOOKUP_JUMPED;
1268 *inode = path->dentry->d_inode;
1275 /* Fast lookup failed, do it the slow way */
1276 static int lookup_slow(struct nameidata *nd, struct qstr *name,
1279 struct dentry *dentry, *parent;
1282 parent = nd->path.dentry;
1283 BUG_ON(nd->inode != parent->d_inode);
1285 mutex_lock(&parent->d_inode->i_mutex);
1286 dentry = __lookup_hash(name, parent, nd->flags);
1287 mutex_unlock(&parent->d_inode->i_mutex);
1289 return PTR_ERR(dentry);
1290 path->mnt = nd->path.mnt;
1291 path->dentry = dentry;
1292 err = follow_managed(path, nd->flags);
1293 if (unlikely(err < 0)) {
1294 path_put_conditional(path, nd);
1298 nd->flags |= LOOKUP_JUMPED;
1302 static inline int may_lookup(struct nameidata *nd)
1304 if (nd->flags & LOOKUP_RCU) {
1305 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1308 if (unlazy_walk(nd, NULL))
1311 return inode_permission(nd->inode, MAY_EXEC);
1314 static inline int handle_dots(struct nameidata *nd, int type)
1316 if (type == LAST_DOTDOT) {
1317 if (nd->flags & LOOKUP_RCU) {
1318 if (follow_dotdot_rcu(nd))
1326 static void terminate_walk(struct nameidata *nd)
1328 if (!(nd->flags & LOOKUP_RCU)) {
1329 path_put(&nd->path);
1331 nd->flags &= ~LOOKUP_RCU;
1332 if (!(nd->flags & LOOKUP_ROOT))
1333 nd->root.mnt = NULL;
1339 * Do we need to follow links? We _really_ want to be able
1340 * to do this check without having to look at inode->i_op,
1341 * so we keep a cache of "no, this doesn't need follow_link"
1342 * for the common case.
1344 static inline int should_follow_link(struct inode *inode, int follow)
1346 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1347 if (likely(inode->i_op->follow_link))
1350 /* This gets set once for the inode lifetime */
1351 spin_lock(&inode->i_lock);
1352 inode->i_opflags |= IOP_NOFOLLOW;
1353 spin_unlock(&inode->i_lock);
1358 static inline int walk_component(struct nameidata *nd, struct path *path,
1359 struct qstr *name, int type, int follow)
1361 struct inode *inode;
1364 * "." and ".." are special - ".." especially so because it has
1365 * to be able to know about the current root directory and
1366 * parent relationships.
1368 if (unlikely(type != LAST_NORM))
1369 return handle_dots(nd, type);
1370 err = lookup_fast(nd, name, path, &inode);
1371 if (unlikely(err)) {
1375 err = lookup_slow(nd, name, path);
1379 inode = path->dentry->d_inode;
1385 if (should_follow_link(inode, follow)) {
1386 if (nd->flags & LOOKUP_RCU) {
1387 if (unlikely(unlazy_walk(nd, path->dentry))) {
1392 BUG_ON(inode != path->dentry->d_inode);
1395 path_to_nameidata(path, nd);
1400 path_to_nameidata(path, nd);
1407 * This limits recursive symlink follows to 8, while
1408 * limiting consecutive symlinks to 40.
1410 * Without that kind of total limit, nasty chains of consecutive
1411 * symlinks can cause almost arbitrarily long lookups.
1413 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1417 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1418 path_put_conditional(path, nd);
1419 path_put(&nd->path);
1422 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1425 current->link_count++;
1428 struct path link = *path;
1431 res = follow_link(&link, nd, &cookie);
1434 res = walk_component(nd, path, &nd->last,
1435 nd->last_type, LOOKUP_FOLLOW);
1436 put_link(nd, &link, cookie);
1439 current->link_count--;
1445 * We really don't want to look at inode->i_op->lookup
1446 * when we don't have to. So we keep a cache bit in
1447 * the inode ->i_opflags field that says "yes, we can
1448 * do lookup on this inode".
1450 static inline int can_lookup(struct inode *inode)
1452 if (likely(inode->i_opflags & IOP_LOOKUP))
1454 if (likely(!inode->i_op->lookup))
1457 /* We do this once for the lifetime of the inode */
1458 spin_lock(&inode->i_lock);
1459 inode->i_opflags |= IOP_LOOKUP;
1460 spin_unlock(&inode->i_lock);
1465 * We can do the critical dentry name comparison and hashing
1466 * operations one word at a time, but we are limited to:
1468 * - Architectures with fast unaligned word accesses. We could
1469 * do a "get_unaligned()" if this helps and is sufficiently
1472 * - Little-endian machines (so that we can generate the mask
1473 * of low bytes efficiently). Again, we *could* do a byte
1474 * swapping load on big-endian architectures if that is not
1475 * expensive enough to make the optimization worthless.
1477 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1478 * do not trap on the (extremely unlikely) case of a page
1479 * crossing operation.
1481 * - Furthermore, we need an efficient 64-bit compile for the
1482 * 64-bit case in order to generate the "number of bytes in
1483 * the final mask". Again, that could be replaced with a
1484 * efficient population count instruction or similar.
1486 #ifdef CONFIG_DCACHE_WORD_ACCESS
1488 #include <asm/word-at-a-time.h>
1492 static inline unsigned int fold_hash(unsigned long hash)
1494 hash += hash >> (8*sizeof(int));
1498 #else /* 32-bit case */
1500 #define fold_hash(x) (x)
1504 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1506 unsigned long a, mask;
1507 unsigned long hash = 0;
1510 a = load_unaligned_zeropad(name);
1511 if (len < sizeof(unsigned long))
1515 name += sizeof(unsigned long);
1516 len -= sizeof(unsigned long);
1520 mask = ~(~0ul << len*8);
1523 return fold_hash(hash);
1525 EXPORT_SYMBOL(full_name_hash);
1528 * Calculate the length and hash of the path component, and
1529 * return the length of the component;
1531 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1533 unsigned long a, b, adata, bdata, mask, hash, len;
1534 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1537 len = -sizeof(unsigned long);
1539 hash = (hash + a) * 9;
1540 len += sizeof(unsigned long);
1541 a = load_unaligned_zeropad(name+len);
1542 b = a ^ REPEAT_BYTE('/');
1543 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1545 adata = prep_zero_mask(a, adata, &constants);
1546 bdata = prep_zero_mask(b, bdata, &constants);
1548 mask = create_zero_mask(adata | bdata);
1550 hash += a & zero_bytemask(mask);
1551 *hashp = fold_hash(hash);
1553 return len + find_zero(mask);
1558 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1560 unsigned long hash = init_name_hash();
1562 hash = partial_name_hash(*name++, hash);
1563 return end_name_hash(hash);
1565 EXPORT_SYMBOL(full_name_hash);
1568 * We know there's a real path component here of at least
1571 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1573 unsigned long hash = init_name_hash();
1574 unsigned long len = 0, c;
1576 c = (unsigned char)*name;
1579 hash = partial_name_hash(c, hash);
1580 c = (unsigned char)name[len];
1581 } while (c && c != '/');
1582 *hashp = end_name_hash(hash);
1590 * This is the basic name resolution function, turning a pathname into
1591 * the final dentry. We expect 'base' to be positive and a directory.
1593 * Returns 0 and nd will have valid dentry and mnt on success.
1594 * Returns error and drops reference to input namei data on failure.
1596 static int link_path_walk(const char *name, struct nameidata *nd)
1606 /* At this point we know we have a real path component. */
1612 err = may_lookup(nd);
1616 len = hash_name(name, &this.hash);
1621 if (name[0] == '.') switch (len) {
1623 if (name[1] == '.') {
1625 nd->flags |= LOOKUP_JUMPED;
1631 if (likely(type == LAST_NORM)) {
1632 struct dentry *parent = nd->path.dentry;
1633 nd->flags &= ~LOOKUP_JUMPED;
1634 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1635 err = parent->d_op->d_hash(parent, nd->inode,
1643 goto last_component;
1645 * If it wasn't NUL, we know it was '/'. Skip that
1646 * slash, and continue until no more slashes.
1650 } while (unlikely(name[len] == '/'));
1652 goto last_component;
1655 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1660 err = nested_symlink(&next, nd);
1664 if (can_lookup(nd->inode))
1668 /* here ends the main loop */
1672 nd->last_type = type;
1679 static int path_init(int dfd, const char *name, unsigned int flags,
1680 struct nameidata *nd, struct file **fp)
1686 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1687 nd->flags = flags | LOOKUP_JUMPED;
1689 if (flags & LOOKUP_ROOT) {
1690 struct inode *inode = nd->root.dentry->d_inode;
1692 if (!inode->i_op->lookup)
1694 retval = inode_permission(inode, MAY_EXEC);
1698 nd->path = nd->root;
1700 if (flags & LOOKUP_RCU) {
1702 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1704 path_get(&nd->path);
1709 nd->root.mnt = NULL;
1712 if (flags & LOOKUP_RCU) {
1717 path_get(&nd->root);
1719 nd->path = nd->root;
1720 } else if (dfd == AT_FDCWD) {
1721 if (flags & LOOKUP_RCU) {
1722 struct fs_struct *fs = current->fs;
1728 seq = read_seqcount_begin(&fs->seq);
1730 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1731 } while (read_seqcount_retry(&fs->seq, seq));
1733 get_fs_pwd(current->fs, &nd->path);
1736 struct dentry *dentry;
1738 file = fget_raw_light(dfd, &fput_needed);
1743 dentry = file->f_path.dentry;
1747 if (!S_ISDIR(dentry->d_inode->i_mode))
1750 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1755 nd->path = file->f_path;
1756 if (flags & LOOKUP_RCU) {
1759 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1762 path_get(&file->f_path);
1763 fput_light(file, fput_needed);
1767 nd->inode = nd->path.dentry->d_inode;
1771 fput_light(file, fput_needed);
1776 static inline int lookup_last(struct nameidata *nd, struct path *path)
1778 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1779 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1781 nd->flags &= ~LOOKUP_PARENT;
1782 return walk_component(nd, path, &nd->last, nd->last_type,
1783 nd->flags & LOOKUP_FOLLOW);
1786 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1787 static int path_lookupat(int dfd, const char *name,
1788 unsigned int flags, struct nameidata *nd)
1790 struct file *base = NULL;
1795 * Path walking is largely split up into 2 different synchronisation
1796 * schemes, rcu-walk and ref-walk (explained in
1797 * Documentation/filesystems/path-lookup.txt). These share much of the
1798 * path walk code, but some things particularly setup, cleanup, and
1799 * following mounts are sufficiently divergent that functions are
1800 * duplicated. Typically there is a function foo(), and its RCU
1801 * analogue, foo_rcu().
1803 * -ECHILD is the error number of choice (just to avoid clashes) that
1804 * is returned if some aspect of an rcu-walk fails. Such an error must
1805 * be handled by restarting a traditional ref-walk (which will always
1806 * be able to complete).
1808 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1813 current->total_link_count = 0;
1814 err = link_path_walk(name, nd);
1816 if (!err && !(flags & LOOKUP_PARENT)) {
1817 err = lookup_last(nd, &path);
1820 struct path link = path;
1821 nd->flags |= LOOKUP_PARENT;
1822 err = follow_link(&link, nd, &cookie);
1825 err = lookup_last(nd, &path);
1826 put_link(nd, &link, cookie);
1831 err = complete_walk(nd);
1833 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1834 if (!nd->inode->i_op->lookup) {
1835 path_put(&nd->path);
1843 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1844 path_put(&nd->root);
1845 nd->root.mnt = NULL;
1850 static int do_path_lookup(int dfd, const char *name,
1851 unsigned int flags, struct nameidata *nd)
1853 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1854 if (unlikely(retval == -ECHILD))
1855 retval = path_lookupat(dfd, name, flags, nd);
1856 if (unlikely(retval == -ESTALE))
1857 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1859 if (likely(!retval)) {
1860 if (unlikely(!audit_dummy_context())) {
1861 if (nd->path.dentry && nd->inode)
1862 audit_inode(name, nd->path.dentry);
1868 /* does lookup, returns the object with parent locked */
1869 struct dentry *kern_path_locked(const char *name, struct path *path)
1871 struct nameidata nd;
1873 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
1875 return ERR_PTR(err);
1876 if (nd.last_type != LAST_NORM) {
1878 return ERR_PTR(-EINVAL);
1880 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1881 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
1883 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1891 int kern_path(const char *name, unsigned int flags, struct path *path)
1893 struct nameidata nd;
1894 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1901 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1902 * @dentry: pointer to dentry of the base directory
1903 * @mnt: pointer to vfs mount of the base directory
1904 * @name: pointer to file name
1905 * @flags: lookup flags
1906 * @path: pointer to struct path to fill
1908 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1909 const char *name, unsigned int flags,
1912 struct nameidata nd;
1914 nd.root.dentry = dentry;
1916 BUG_ON(flags & LOOKUP_PARENT);
1917 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1918 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1925 * Restricted form of lookup. Doesn't follow links, single-component only,
1926 * needs parent already locked. Doesn't follow mounts.
1929 static struct dentry *lookup_hash(struct nameidata *nd)
1931 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
1935 * lookup_one_len - filesystem helper to lookup single pathname component
1936 * @name: pathname component to lookup
1937 * @base: base directory to lookup from
1938 * @len: maximum length @len should be interpreted to
1940 * Note that this routine is purely a helper for filesystem usage and should
1941 * not be called by generic code. Also note that by using this function the
1942 * nameidata argument is passed to the filesystem methods and a filesystem
1943 * using this helper needs to be prepared for that.
1945 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1951 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1955 this.hash = full_name_hash(name, len);
1957 return ERR_PTR(-EACCES);
1960 c = *(const unsigned char *)name++;
1961 if (c == '/' || c == '\0')
1962 return ERR_PTR(-EACCES);
1965 * See if the low-level filesystem might want
1966 * to use its own hash..
1968 if (base->d_flags & DCACHE_OP_HASH) {
1969 int err = base->d_op->d_hash(base, base->d_inode, &this);
1971 return ERR_PTR(err);
1974 err = inode_permission(base->d_inode, MAY_EXEC);
1976 return ERR_PTR(err);
1978 return __lookup_hash(&this, base, 0);
1981 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1982 struct path *path, int *empty)
1984 struct nameidata nd;
1985 char *tmp = getname_flags(name, flags, empty);
1986 int err = PTR_ERR(tmp);
1989 BUG_ON(flags & LOOKUP_PARENT);
1991 err = do_path_lookup(dfd, tmp, flags, &nd);
1999 int user_path_at(int dfd, const char __user *name, unsigned flags,
2002 return user_path_at_empty(dfd, name, flags, path, NULL);
2005 static int user_path_parent(int dfd, const char __user *path,
2006 struct nameidata *nd, char **name)
2008 char *s = getname(path);
2014 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
2024 * It's inline, so penalty for filesystems that don't use sticky bit is
2027 static inline int check_sticky(struct inode *dir, struct inode *inode)
2029 kuid_t fsuid = current_fsuid();
2031 if (!(dir->i_mode & S_ISVTX))
2033 if (uid_eq(inode->i_uid, fsuid))
2035 if (uid_eq(dir->i_uid, fsuid))
2037 return !inode_capable(inode, CAP_FOWNER);
2041 * Check whether we can remove a link victim from directory dir, check
2042 * whether the type of victim is right.
2043 * 1. We can't do it if dir is read-only (done in permission())
2044 * 2. We should have write and exec permissions on dir
2045 * 3. We can't remove anything from append-only dir
2046 * 4. We can't do anything with immutable dir (done in permission())
2047 * 5. If the sticky bit on dir is set we should either
2048 * a. be owner of dir, or
2049 * b. be owner of victim, or
2050 * c. have CAP_FOWNER capability
2051 * 6. If the victim is append-only or immutable we can't do antyhing with
2052 * links pointing to it.
2053 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2054 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2055 * 9. We can't remove a root or mountpoint.
2056 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2057 * nfs_async_unlink().
2059 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2063 if (!victim->d_inode)
2066 BUG_ON(victim->d_parent->d_inode != dir);
2067 audit_inode_child(victim, dir);
2069 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2074 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2075 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2078 if (!S_ISDIR(victim->d_inode->i_mode))
2080 if (IS_ROOT(victim))
2082 } else if (S_ISDIR(victim->d_inode->i_mode))
2084 if (IS_DEADDIR(dir))
2086 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2091 /* Check whether we can create an object with dentry child in directory
2093 * 1. We can't do it if child already exists (open has special treatment for
2094 * this case, but since we are inlined it's OK)
2095 * 2. We can't do it if dir is read-only (done in permission())
2096 * 3. We should have write and exec permissions on dir
2097 * 4. We can't do it if dir is immutable (done in permission())
2099 static inline int may_create(struct inode *dir, struct dentry *child)
2103 if (IS_DEADDIR(dir))
2105 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2109 * p1 and p2 should be directories on the same fs.
2111 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2116 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2120 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2122 p = d_ancestor(p2, p1);
2124 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2125 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2129 p = d_ancestor(p1, p2);
2131 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2132 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2136 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2137 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2141 void unlock_rename(struct dentry *p1, struct dentry *p2)
2143 mutex_unlock(&p1->d_inode->i_mutex);
2145 mutex_unlock(&p2->d_inode->i_mutex);
2146 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2150 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2153 int error = may_create(dir, dentry);
2157 if (!dir->i_op->create)
2158 return -EACCES; /* shouldn't it be ENOSYS? */
2161 error = security_inode_create(dir, dentry, mode);
2164 error = dir->i_op->create(dir, dentry, mode, want_excl);
2166 fsnotify_create(dir, dentry);
2170 static int may_open(struct path *path, int acc_mode, int flag)
2172 struct dentry *dentry = path->dentry;
2173 struct inode *inode = dentry->d_inode;
2183 switch (inode->i_mode & S_IFMT) {
2187 if (acc_mode & MAY_WRITE)
2192 if (path->mnt->mnt_flags & MNT_NODEV)
2201 error = inode_permission(inode, acc_mode);
2206 * An append-only file must be opened in append mode for writing.
2208 if (IS_APPEND(inode)) {
2209 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2215 /* O_NOATIME can only be set by the owner or superuser */
2216 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2222 static int handle_truncate(struct file *filp)
2224 struct path *path = &filp->f_path;
2225 struct inode *inode = path->dentry->d_inode;
2226 int error = get_write_access(inode);
2230 * Refuse to truncate files with mandatory locks held on them.
2232 error = locks_verify_locked(inode);
2234 error = security_path_truncate(path);
2236 error = do_truncate(path->dentry, 0,
2237 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2240 put_write_access(inode);
2244 static inline int open_to_namei_flags(int flag)
2246 if ((flag & O_ACCMODE) == 3)
2251 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2253 int error = security_path_mknod(dir, dentry, mode, 0);
2257 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2261 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2265 * Attempt to atomically look up, create and open a file from a negative
2268 * Returns 0 if successful. The file will have been created and attached to
2269 * @file by the filesystem calling finish_open().
2271 * Returns 1 if the file was looked up only or didn't need creating. The
2272 * caller will need to perform the open themselves. @path will have been
2273 * updated to point to the new dentry. This may be negative.
2275 * Returns an error code otherwise.
2277 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2278 struct path *path, struct file *file,
2279 const struct open_flags *op,
2280 bool *want_write, bool need_lookup,
2283 struct inode *dir = nd->path.dentry->d_inode;
2284 unsigned open_flag = open_to_namei_flags(op->open_flag);
2288 int create_error = 0;
2289 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2291 BUG_ON(dentry->d_inode);
2293 /* Don't create child dentry for a dead directory. */
2294 if (unlikely(IS_DEADDIR(dir))) {
2299 mode = op->mode & S_IALLUGO;
2300 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2301 mode &= ~current_umask();
2303 if (open_flag & O_EXCL) {
2304 open_flag &= ~O_TRUNC;
2305 *opened |= FILE_CREATED;
2309 * Checking write permission is tricky, bacuse we don't know if we are
2310 * going to actually need it: O_CREAT opens should work as long as the
2311 * file exists. But checking existence breaks atomicity. The trick is
2312 * to check access and if not granted clear O_CREAT from the flags.
2314 * Another problem is returing the "right" error value (e.g. for an
2315 * O_EXCL open we want to return EEXIST not EROFS).
2317 if ((open_flag & (O_CREAT | O_TRUNC)) ||
2318 (open_flag & O_ACCMODE) != O_RDONLY) {
2319 error = mnt_want_write(nd->path.mnt);
2322 } else if (!(open_flag & O_CREAT)) {
2324 * No O_CREATE -> atomicity not a requirement -> fall
2325 * back to lookup + open
2328 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2329 /* Fall back and fail with the right error */
2330 create_error = error;
2333 /* No side effects, safe to clear O_CREAT */
2334 create_error = error;
2335 open_flag &= ~O_CREAT;
2339 if (open_flag & O_CREAT) {
2340 error = may_o_create(&nd->path, dentry, op->mode);
2342 create_error = error;
2343 if (open_flag & O_EXCL)
2345 open_flag &= ~O_CREAT;
2349 if (nd->flags & LOOKUP_DIRECTORY)
2350 open_flag |= O_DIRECTORY;
2352 file->f_path.dentry = DENTRY_NOT_SET;
2353 file->f_path.mnt = nd->path.mnt;
2354 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2357 if (create_error && error == -ENOENT)
2358 error = create_error;
2362 acc_mode = op->acc_mode;
2363 if (*opened & FILE_CREATED) {
2364 fsnotify_create(dir, dentry);
2365 acc_mode = MAY_OPEN;
2368 if (error) { /* returned 1, that is */
2369 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2373 if (file->f_path.dentry) {
2375 dentry = file->f_path.dentry;
2381 * We didn't have the inode before the open, so check open permission
2384 error = may_open(&file->f_path, acc_mode, open_flag);
2394 dentry = lookup_real(dir, dentry, nd->flags);
2396 return PTR_ERR(dentry);
2399 int open_flag = op->open_flag;
2401 error = create_error;
2402 if ((open_flag & O_EXCL)) {
2403 if (!dentry->d_inode)
2405 } else if (!dentry->d_inode) {
2407 } else if ((open_flag & O_TRUNC) &&
2408 S_ISREG(dentry->d_inode->i_mode)) {
2411 /* will fail later, go on to get the right error */
2415 path->dentry = dentry;
2416 path->mnt = nd->path.mnt;
2421 * Look up and maybe create and open the last component.
2423 * Must be called with i_mutex held on parent.
2425 * Returns 0 if the file was successfully atomically created (if necessary) and
2426 * opened. In this case the file will be returned attached to @file.
2428 * Returns 1 if the file was not completely opened at this time, though lookups
2429 * and creations will have been performed and the dentry returned in @path will
2430 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2431 * specified then a negative dentry may be returned.
2433 * An error code is returned otherwise.
2435 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2436 * cleared otherwise prior to returning.
2438 static int lookup_open(struct nameidata *nd, struct path *path,
2440 const struct open_flags *op,
2441 bool *want_write, int *opened)
2443 struct dentry *dir = nd->path.dentry;
2444 struct inode *dir_inode = dir->d_inode;
2445 struct dentry *dentry;
2449 *opened &= ~FILE_CREATED;
2450 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2452 return PTR_ERR(dentry);
2454 /* Cached positive dentry: will open in f_op->open */
2455 if (!need_lookup && dentry->d_inode)
2458 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2459 return atomic_open(nd, dentry, path, file, op, want_write,
2460 need_lookup, opened);
2464 BUG_ON(dentry->d_inode);
2466 dentry = lookup_real(dir_inode, dentry, nd->flags);
2468 return PTR_ERR(dentry);
2471 /* Negative dentry, just create the file */
2472 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2473 umode_t mode = op->mode;
2474 if (!IS_POSIXACL(dir->d_inode))
2475 mode &= ~current_umask();
2477 * This write is needed to ensure that a
2478 * rw->ro transition does not occur between
2479 * the time when the file is created and when
2480 * a permanent write count is taken through
2481 * the 'struct file' in finish_open().
2483 error = mnt_want_write(nd->path.mnt);
2487 *opened |= FILE_CREATED;
2488 error = security_path_mknod(&nd->path, dentry, mode, 0);
2491 error = vfs_create(dir->d_inode, dentry, mode,
2492 nd->flags & LOOKUP_EXCL);
2497 path->dentry = dentry;
2498 path->mnt = nd->path.mnt;
2507 * Handle the last step of open()
2509 static int do_last(struct nameidata *nd, struct path *path,
2510 struct file *file, const struct open_flags *op,
2511 int *opened, const char *pathname)
2513 struct dentry *dir = nd->path.dentry;
2514 int open_flag = op->open_flag;
2515 bool will_truncate = (open_flag & O_TRUNC) != 0;
2516 bool want_write = false;
2517 int acc_mode = op->acc_mode;
2518 struct inode *inode;
2519 bool symlink_ok = false;
2520 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2521 bool retried = false;
2524 nd->flags &= ~LOOKUP_PARENT;
2525 nd->flags |= op->intent;
2527 switch (nd->last_type) {
2530 error = handle_dots(nd, nd->last_type);
2535 error = complete_walk(nd);
2538 audit_inode(pathname, nd->path.dentry);
2539 if (open_flag & O_CREAT) {
2545 error = complete_walk(nd);
2548 audit_inode(pathname, dir);
2552 if (!(open_flag & O_CREAT)) {
2553 if (nd->last.name[nd->last.len])
2554 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2555 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2557 /* we _can_ be in RCU mode here */
2558 error = lookup_fast(nd, &nd->last, path, &inode);
2565 BUG_ON(nd->inode != dir->d_inode);
2567 /* create side of things */
2569 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2570 * has been cleared when we got to the last component we are
2573 error = complete_walk(nd);
2577 audit_inode(pathname, dir);
2579 /* trailing slashes? */
2580 if (nd->last.name[nd->last.len])
2585 mutex_lock(&dir->d_inode->i_mutex);
2586 error = lookup_open(nd, path, file, op, &want_write, opened);
2587 mutex_unlock(&dir->d_inode->i_mutex);
2593 if ((*opened & FILE_CREATED) ||
2594 !S_ISREG(file->f_path.dentry->d_inode->i_mode))
2595 will_truncate = false;
2597 audit_inode(pathname, file->f_path.dentry);
2601 if (*opened & FILE_CREATED) {
2602 /* Don't check for write permission, don't truncate */
2603 open_flag &= ~O_TRUNC;
2604 will_truncate = false;
2605 acc_mode = MAY_OPEN;
2606 path_to_nameidata(path, nd);
2607 goto finish_open_created;
2611 * It already exists.
2613 audit_inode(pathname, path->dentry);
2616 * If atomic_open() acquired write access it is dropped now due to
2617 * possible mount and symlink following (this might be optimized away if
2621 mnt_drop_write(nd->path.mnt);
2626 if (open_flag & O_EXCL)
2629 error = follow_managed(path, nd->flags);
2634 nd->flags |= LOOKUP_JUMPED;
2636 BUG_ON(nd->flags & LOOKUP_RCU);
2637 inode = path->dentry->d_inode;
2639 /* we _can_ be in RCU mode here */
2642 path_to_nameidata(path, nd);
2646 if (should_follow_link(inode, !symlink_ok)) {
2647 if (nd->flags & LOOKUP_RCU) {
2648 if (unlikely(unlazy_walk(nd, path->dentry))) {
2653 BUG_ON(inode != path->dentry->d_inode);
2657 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2658 path_to_nameidata(path, nd);
2660 save_parent.dentry = nd->path.dentry;
2661 save_parent.mnt = mntget(path->mnt);
2662 nd->path.dentry = path->dentry;
2666 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2667 error = complete_walk(nd);
2669 path_put(&save_parent);
2673 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
2676 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
2678 audit_inode(pathname, nd->path.dentry);
2680 if (!S_ISREG(nd->inode->i_mode))
2681 will_truncate = false;
2683 if (will_truncate) {
2684 error = mnt_want_write(nd->path.mnt);
2689 finish_open_created:
2690 error = may_open(&nd->path, acc_mode, open_flag);
2693 file->f_path.mnt = nd->path.mnt;
2694 error = finish_open(file, nd->path.dentry, NULL, opened);
2696 if (error == -EOPENSTALE)
2701 error = open_check_o_direct(file);
2704 error = ima_file_check(file, op->acc_mode);
2708 if (will_truncate) {
2709 error = handle_truncate(file);
2715 mnt_drop_write(nd->path.mnt);
2716 path_put(&save_parent);
2721 path_put_conditional(path, nd);
2728 /* If no saved parent or already retried then can't retry */
2729 if (!save_parent.dentry || retried)
2732 BUG_ON(save_parent.dentry != dir);
2733 path_put(&nd->path);
2734 nd->path = save_parent;
2735 nd->inode = dir->d_inode;
2736 save_parent.mnt = NULL;
2737 save_parent.dentry = NULL;
2739 mnt_drop_write(nd->path.mnt);
2746 static struct file *path_openat(int dfd, const char *pathname,
2747 struct nameidata *nd, const struct open_flags *op, int flags)
2749 struct file *base = NULL;
2755 file = get_empty_filp();
2757 return ERR_PTR(-ENFILE);
2759 file->f_flags = op->open_flag;
2761 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2762 if (unlikely(error))
2765 current->total_link_count = 0;
2766 error = link_path_walk(pathname, nd);
2767 if (unlikely(error))
2770 error = do_last(nd, &path, file, op, &opened, pathname);
2771 while (unlikely(error > 0)) { /* trailing symlink */
2772 struct path link = path;
2774 if (!(nd->flags & LOOKUP_FOLLOW)) {
2775 path_put_conditional(&path, nd);
2776 path_put(&nd->path);
2780 nd->flags |= LOOKUP_PARENT;
2781 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2782 error = follow_link(&link, nd, &cookie);
2783 if (unlikely(error))
2785 error = do_last(nd, &path, file, op, &opened, pathname);
2786 put_link(nd, &link, cookie);
2789 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2790 path_put(&nd->root);
2793 if (!(opened & FILE_OPENED)) {
2797 if (unlikely(error)) {
2798 if (error == -EOPENSTALE) {
2799 if (flags & LOOKUP_RCU)
2804 file = ERR_PTR(error);
2809 struct file *do_filp_open(int dfd, const char *pathname,
2810 const struct open_flags *op, int flags)
2812 struct nameidata nd;
2815 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2816 if (unlikely(filp == ERR_PTR(-ECHILD)))
2817 filp = path_openat(dfd, pathname, &nd, op, flags);
2818 if (unlikely(filp == ERR_PTR(-ESTALE)))
2819 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2823 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2824 const char *name, const struct open_flags *op, int flags)
2826 struct nameidata nd;
2830 nd.root.dentry = dentry;
2832 flags |= LOOKUP_ROOT;
2834 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2835 return ERR_PTR(-ELOOP);
2837 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2838 if (unlikely(file == ERR_PTR(-ECHILD)))
2839 file = path_openat(-1, name, &nd, op, flags);
2840 if (unlikely(file == ERR_PTR(-ESTALE)))
2841 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2845 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2847 struct dentry *dentry = ERR_PTR(-EEXIST);
2848 struct nameidata nd;
2849 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2851 return ERR_PTR(error);
2854 * Yucky last component or no last component at all?
2855 * (foo/., foo/.., /////)
2857 if (nd.last_type != LAST_NORM)
2859 nd.flags &= ~LOOKUP_PARENT;
2860 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2863 * Do the final lookup.
2865 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2866 dentry = lookup_hash(&nd);
2870 if (dentry->d_inode)
2873 * Special case - lookup gave negative, but... we had foo/bar/
2874 * From the vfs_mknod() POV we just have a negative dentry -
2875 * all is fine. Let's be bastards - you had / on the end, you've
2876 * been asking for (non-existent) directory. -ENOENT for you.
2878 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2880 dentry = ERR_PTR(-ENOENT);
2887 dentry = ERR_PTR(-EEXIST);
2889 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2894 EXPORT_SYMBOL(kern_path_create);
2896 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2898 char *tmp = getname(pathname);
2901 return ERR_CAST(tmp);
2902 res = kern_path_create(dfd, tmp, path, is_dir);
2906 EXPORT_SYMBOL(user_path_create);
2908 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2910 int error = may_create(dir, dentry);
2915 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2918 if (!dir->i_op->mknod)
2921 error = devcgroup_inode_mknod(mode, dev);
2925 error = security_inode_mknod(dir, dentry, mode, dev);
2929 error = dir->i_op->mknod(dir, dentry, mode, dev);
2931 fsnotify_create(dir, dentry);
2935 static int may_mknod(umode_t mode)
2937 switch (mode & S_IFMT) {
2943 case 0: /* zero mode translates to S_IFREG */
2952 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2955 struct dentry *dentry;
2962 dentry = user_path_create(dfd, filename, &path, 0);
2964 return PTR_ERR(dentry);
2966 if (!IS_POSIXACL(path.dentry->d_inode))
2967 mode &= ~current_umask();
2968 error = may_mknod(mode);
2971 error = mnt_want_write(path.mnt);
2974 error = security_path_mknod(&path, dentry, mode, dev);
2976 goto out_drop_write;
2977 switch (mode & S_IFMT) {
2978 case 0: case S_IFREG:
2979 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
2981 case S_IFCHR: case S_IFBLK:
2982 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2983 new_decode_dev(dev));
2985 case S_IFIFO: case S_IFSOCK:
2986 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2990 mnt_drop_write(path.mnt);
2993 mutex_unlock(&path.dentry->d_inode->i_mutex);
2999 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3001 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3004 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3006 int error = may_create(dir, dentry);
3007 unsigned max_links = dir->i_sb->s_max_links;
3012 if (!dir->i_op->mkdir)
3015 mode &= (S_IRWXUGO|S_ISVTX);
3016 error = security_inode_mkdir(dir, dentry, mode);
3020 if (max_links && dir->i_nlink >= max_links)
3023 error = dir->i_op->mkdir(dir, dentry, mode);
3025 fsnotify_mkdir(dir, dentry);
3029 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3031 struct dentry *dentry;
3035 dentry = user_path_create(dfd, pathname, &path, 1);
3037 return PTR_ERR(dentry);
3039 if (!IS_POSIXACL(path.dentry->d_inode))
3040 mode &= ~current_umask();
3041 error = mnt_want_write(path.mnt);
3044 error = security_path_mkdir(&path, dentry, mode);
3046 goto out_drop_write;
3047 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3049 mnt_drop_write(path.mnt);
3052 mutex_unlock(&path.dentry->d_inode->i_mutex);
3057 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3059 return sys_mkdirat(AT_FDCWD, pathname, mode);
3063 * The dentry_unhash() helper will try to drop the dentry early: we
3064 * should have a usage count of 1 if we're the only user of this
3065 * dentry, and if that is true (possibly after pruning the dcache),
3066 * then we drop the dentry now.
3068 * A low-level filesystem can, if it choses, legally
3071 * if (!d_unhashed(dentry))
3074 * if it cannot handle the case of removing a directory
3075 * that is still in use by something else..
3077 void dentry_unhash(struct dentry *dentry)
3079 shrink_dcache_parent(dentry);
3080 spin_lock(&dentry->d_lock);
3081 if (dentry->d_count == 1)
3083 spin_unlock(&dentry->d_lock);
3086 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3088 int error = may_delete(dir, dentry, 1);
3093 if (!dir->i_op->rmdir)
3097 mutex_lock(&dentry->d_inode->i_mutex);
3100 if (d_mountpoint(dentry))
3103 error = security_inode_rmdir(dir, dentry);
3107 shrink_dcache_parent(dentry);
3108 error = dir->i_op->rmdir(dir, dentry);
3112 dentry->d_inode->i_flags |= S_DEAD;
3116 mutex_unlock(&dentry->d_inode->i_mutex);
3123 static long do_rmdir(int dfd, const char __user *pathname)
3127 struct dentry *dentry;
3128 struct nameidata nd;
3130 error = user_path_parent(dfd, pathname, &nd, &name);
3134 switch(nd.last_type) {
3146 nd.flags &= ~LOOKUP_PARENT;
3148 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3149 dentry = lookup_hash(&nd);
3150 error = PTR_ERR(dentry);
3153 if (!dentry->d_inode) {
3157 error = mnt_want_write(nd.path.mnt);
3160 error = security_path_rmdir(&nd.path, dentry);
3163 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3165 mnt_drop_write(nd.path.mnt);
3169 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3176 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3178 return do_rmdir(AT_FDCWD, pathname);
3181 int vfs_unlink(struct inode *dir, struct dentry *dentry)
3183 int error = may_delete(dir, dentry, 0);
3188 if (!dir->i_op->unlink)
3191 mutex_lock(&dentry->d_inode->i_mutex);
3192 if (d_mountpoint(dentry))
3195 error = security_inode_unlink(dir, dentry);
3197 error = dir->i_op->unlink(dir, dentry);
3202 mutex_unlock(&dentry->d_inode->i_mutex);
3204 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3205 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3206 fsnotify_link_count(dentry->d_inode);
3214 * Make sure that the actual truncation of the file will occur outside its
3215 * directory's i_mutex. Truncate can take a long time if there is a lot of
3216 * writeout happening, and we don't want to prevent access to the directory
3217 * while waiting on the I/O.
3219 static long do_unlinkat(int dfd, const char __user *pathname)
3223 struct dentry *dentry;
3224 struct nameidata nd;
3225 struct inode *inode = NULL;
3227 error = user_path_parent(dfd, pathname, &nd, &name);
3232 if (nd.last_type != LAST_NORM)
3235 nd.flags &= ~LOOKUP_PARENT;
3237 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3238 dentry = lookup_hash(&nd);
3239 error = PTR_ERR(dentry);
3240 if (!IS_ERR(dentry)) {
3241 /* Why not before? Because we want correct error value */
3242 if (nd.last.name[nd.last.len])
3244 inode = dentry->d_inode;
3248 error = mnt_want_write(nd.path.mnt);
3251 error = security_path_unlink(&nd.path, dentry);
3254 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
3256 mnt_drop_write(nd.path.mnt);
3260 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3262 iput(inode); /* truncate the inode here */
3269 error = !dentry->d_inode ? -ENOENT :
3270 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3274 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3276 if ((flag & ~AT_REMOVEDIR) != 0)
3279 if (flag & AT_REMOVEDIR)
3280 return do_rmdir(dfd, pathname);
3282 return do_unlinkat(dfd, pathname);
3285 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3287 return do_unlinkat(AT_FDCWD, pathname);
3290 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3292 int error = may_create(dir, dentry);
3297 if (!dir->i_op->symlink)
3300 error = security_inode_symlink(dir, dentry, oldname);
3304 error = dir->i_op->symlink(dir, dentry, oldname);
3306 fsnotify_create(dir, dentry);
3310 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3311 int, newdfd, const char __user *, newname)
3315 struct dentry *dentry;
3318 from = getname(oldname);
3320 return PTR_ERR(from);
3322 dentry = user_path_create(newdfd, newname, &path, 0);
3323 error = PTR_ERR(dentry);
3327 error = mnt_want_write(path.mnt);
3330 error = security_path_symlink(&path, dentry, from);
3332 goto out_drop_write;
3333 error = vfs_symlink(path.dentry->d_inode, dentry, from);
3335 mnt_drop_write(path.mnt);
3338 mutex_unlock(&path.dentry->d_inode->i_mutex);
3345 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3347 return sys_symlinkat(oldname, AT_FDCWD, newname);
3350 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3352 struct inode *inode = old_dentry->d_inode;
3353 unsigned max_links = dir->i_sb->s_max_links;
3359 error = may_create(dir, new_dentry);
3363 if (dir->i_sb != inode->i_sb)
3367 * A link to an append-only or immutable file cannot be created.
3369 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3371 if (!dir->i_op->link)
3373 if (S_ISDIR(inode->i_mode))
3376 error = security_inode_link(old_dentry, dir, new_dentry);
3380 mutex_lock(&inode->i_mutex);
3381 /* Make sure we don't allow creating hardlink to an unlinked file */
3382 if (inode->i_nlink == 0)
3384 else if (max_links && inode->i_nlink >= max_links)
3387 error = dir->i_op->link(old_dentry, dir, new_dentry);
3388 mutex_unlock(&inode->i_mutex);
3390 fsnotify_link(dir, inode, new_dentry);
3395 * Hardlinks are often used in delicate situations. We avoid
3396 * security-related surprises by not following symlinks on the
3399 * We don't follow them on the oldname either to be compatible
3400 * with linux 2.0, and to avoid hard-linking to directories
3401 * and other special files. --ADM
3403 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3404 int, newdfd, const char __user *, newname, int, flags)
3406 struct dentry *new_dentry;
3407 struct path old_path, new_path;
3411 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3414 * To use null names we require CAP_DAC_READ_SEARCH
3415 * This ensures that not everyone will be able to create
3416 * handlink using the passed filedescriptor.
3418 if (flags & AT_EMPTY_PATH) {
3419 if (!capable(CAP_DAC_READ_SEARCH))
3424 if (flags & AT_SYMLINK_FOLLOW)
3425 how |= LOOKUP_FOLLOW;
3427 error = user_path_at(olddfd, oldname, how, &old_path);
3431 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3432 error = PTR_ERR(new_dentry);
3433 if (IS_ERR(new_dentry))
3437 if (old_path.mnt != new_path.mnt)
3439 error = mnt_want_write(new_path.mnt);
3442 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3444 goto out_drop_write;
3445 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3447 mnt_drop_write(new_path.mnt);
3450 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3451 path_put(&new_path);
3453 path_put(&old_path);
3458 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3460 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3464 * The worst of all namespace operations - renaming directory. "Perverted"
3465 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3467 * a) we can get into loop creation. Check is done in is_subdir().
3468 * b) race potential - two innocent renames can create a loop together.
3469 * That's where 4.4 screws up. Current fix: serialization on
3470 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3472 * c) we have to lock _three_ objects - parents and victim (if it exists).
3473 * And that - after we got ->i_mutex on parents (until then we don't know
3474 * whether the target exists). Solution: try to be smart with locking
3475 * order for inodes. We rely on the fact that tree topology may change
3476 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3477 * move will be locked. Thus we can rank directories by the tree
3478 * (ancestors first) and rank all non-directories after them.
3479 * That works since everybody except rename does "lock parent, lookup,
3480 * lock child" and rename is under ->s_vfs_rename_mutex.
3481 * HOWEVER, it relies on the assumption that any object with ->lookup()
3482 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3483 * we'd better make sure that there's no link(2) for them.
3484 * d) conversion from fhandle to dentry may come in the wrong moment - when
3485 * we are removing the target. Solution: we will have to grab ->i_mutex
3486 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3487 * ->i_mutex on parents, which works but leads to some truly excessive
3490 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3491 struct inode *new_dir, struct dentry *new_dentry)
3494 struct inode *target = new_dentry->d_inode;
3495 unsigned max_links = new_dir->i_sb->s_max_links;
3498 * If we are going to change the parent - check write permissions,
3499 * we'll need to flip '..'.
3501 if (new_dir != old_dir) {
3502 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3507 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3513 mutex_lock(&target->i_mutex);
3516 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3520 if (max_links && !target && new_dir != old_dir &&
3521 new_dir->i_nlink >= max_links)
3525 shrink_dcache_parent(new_dentry);
3526 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3531 target->i_flags |= S_DEAD;
3532 dont_mount(new_dentry);
3536 mutex_unlock(&target->i_mutex);
3539 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3540 d_move(old_dentry,new_dentry);
3544 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3545 struct inode *new_dir, struct dentry *new_dentry)
3547 struct inode *target = new_dentry->d_inode;
3550 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3556 mutex_lock(&target->i_mutex);
3559 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3562 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3567 dont_mount(new_dentry);
3568 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3569 d_move(old_dentry, new_dentry);
3572 mutex_unlock(&target->i_mutex);
3577 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3578 struct inode *new_dir, struct dentry *new_dentry)
3581 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3582 const unsigned char *old_name;
3584 if (old_dentry->d_inode == new_dentry->d_inode)
3587 error = may_delete(old_dir, old_dentry, is_dir);
3591 if (!new_dentry->d_inode)
3592 error = may_create(new_dir, new_dentry);
3594 error = may_delete(new_dir, new_dentry, is_dir);
3598 if (!old_dir->i_op->rename)
3601 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3604 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3606 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3608 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3609 new_dentry->d_inode, old_dentry);
3610 fsnotify_oldname_free(old_name);
3615 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3616 int, newdfd, const char __user *, newname)
3618 struct dentry *old_dir, *new_dir;
3619 struct dentry *old_dentry, *new_dentry;
3620 struct dentry *trap;
3621 struct nameidata oldnd, newnd;
3626 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3630 error = user_path_parent(newdfd, newname, &newnd, &to);
3635 if (oldnd.path.mnt != newnd.path.mnt)
3638 old_dir = oldnd.path.dentry;
3640 if (oldnd.last_type != LAST_NORM)
3643 new_dir = newnd.path.dentry;
3644 if (newnd.last_type != LAST_NORM)
3647 oldnd.flags &= ~LOOKUP_PARENT;
3648 newnd.flags &= ~LOOKUP_PARENT;
3649 newnd.flags |= LOOKUP_RENAME_TARGET;
3651 trap = lock_rename(new_dir, old_dir);
3653 old_dentry = lookup_hash(&oldnd);
3654 error = PTR_ERR(old_dentry);
3655 if (IS_ERR(old_dentry))
3657 /* source must exist */
3659 if (!old_dentry->d_inode)
3661 /* unless the source is a directory trailing slashes give -ENOTDIR */
3662 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3664 if (oldnd.last.name[oldnd.last.len])
3666 if (newnd.last.name[newnd.last.len])
3669 /* source should not be ancestor of target */
3671 if (old_dentry == trap)
3673 new_dentry = lookup_hash(&newnd);
3674 error = PTR_ERR(new_dentry);
3675 if (IS_ERR(new_dentry))
3677 /* target should not be an ancestor of source */
3679 if (new_dentry == trap)
3682 error = mnt_want_write(oldnd.path.mnt);
3685 error = security_path_rename(&oldnd.path, old_dentry,
3686 &newnd.path, new_dentry);
3689 error = vfs_rename(old_dir->d_inode, old_dentry,
3690 new_dir->d_inode, new_dentry);
3692 mnt_drop_write(oldnd.path.mnt);
3698 unlock_rename(new_dir, old_dir);
3700 path_put(&newnd.path);
3703 path_put(&oldnd.path);
3709 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3711 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3714 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3718 len = PTR_ERR(link);
3723 if (len > (unsigned) buflen)
3725 if (copy_to_user(buffer, link, len))
3732 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3733 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3734 * using) it for any given inode is up to filesystem.
3736 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3738 struct nameidata nd;
3743 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3745 return PTR_ERR(cookie);
3747 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3748 if (dentry->d_inode->i_op->put_link)
3749 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3753 int vfs_follow_link(struct nameidata *nd, const char *link)
3755 return __vfs_follow_link(nd, link);
3758 /* get the link contents into pagecache */
3759 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3763 struct address_space *mapping = dentry->d_inode->i_mapping;
3764 page = read_mapping_page(mapping, 0, NULL);
3769 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3773 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3775 struct page *page = NULL;
3776 char *s = page_getlink(dentry, &page);
3777 int res = vfs_readlink(dentry,buffer,buflen,s);
3780 page_cache_release(page);
3785 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3787 struct page *page = NULL;
3788 nd_set_link(nd, page_getlink(dentry, &page));
3792 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3794 struct page *page = cookie;
3798 page_cache_release(page);
3803 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3805 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3807 struct address_space *mapping = inode->i_mapping;
3812 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3814 flags |= AOP_FLAG_NOFS;
3817 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3818 flags, &page, &fsdata);
3822 kaddr = kmap_atomic(page);
3823 memcpy(kaddr, symname, len-1);
3824 kunmap_atomic(kaddr);
3826 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3833 mark_inode_dirty(inode);
3839 int page_symlink(struct inode *inode, const char *symname, int len)
3841 return __page_symlink(inode, symname, len,
3842 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3845 const struct inode_operations page_symlink_inode_operations = {
3846 .readlink = generic_readlink,
3847 .follow_link = page_follow_link_light,
3848 .put_link = page_put_link,
3851 EXPORT_SYMBOL(user_path_at);
3852 EXPORT_SYMBOL(follow_down_one);
3853 EXPORT_SYMBOL(follow_down);
3854 EXPORT_SYMBOL(follow_up);
3855 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3856 EXPORT_SYMBOL(getname);
3857 EXPORT_SYMBOL(lock_rename);
3858 EXPORT_SYMBOL(lookup_one_len);
3859 EXPORT_SYMBOL(page_follow_link_light);
3860 EXPORT_SYMBOL(page_put_link);
3861 EXPORT_SYMBOL(page_readlink);
3862 EXPORT_SYMBOL(__page_symlink);
3863 EXPORT_SYMBOL(page_symlink);
3864 EXPORT_SYMBOL(page_symlink_inode_operations);
3865 EXPORT_SYMBOL(kern_path);
3866 EXPORT_SYMBOL(vfs_path_lookup);
3867 EXPORT_SYMBOL(inode_permission);
3868 EXPORT_SYMBOL(unlock_rename);
3869 EXPORT_SYMBOL(vfs_create);
3870 EXPORT_SYMBOL(vfs_follow_link);
3871 EXPORT_SYMBOL(vfs_link);
3872 EXPORT_SYMBOL(vfs_mkdir);
3873 EXPORT_SYMBOL(vfs_mknod);
3874 EXPORT_SYMBOL(generic_permission);
3875 EXPORT_SYMBOL(vfs_readlink);
3876 EXPORT_SYMBOL(vfs_rename);
3877 EXPORT_SYMBOL(vfs_rmdir);
3878 EXPORT_SYMBOL(vfs_symlink);
3879 EXPORT_SYMBOL(vfs_unlink);
3880 EXPORT_SYMBOL(dentry_unhash);
3881 EXPORT_SYMBOL(generic_readlink);