1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * Some corrections by tytso.
12 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
15 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
18 #include <linux/init.h>
19 #include <linux/export.h>
20 #include <linux/kernel.h>
21 #include <linux/slab.h>
23 #include <linux/filelock.h>
24 #include <linux/namei.h>
25 #include <linux/pagemap.h>
26 #include <linux/sched/mm.h>
27 #include <linux/fsnotify.h>
28 #include <linux/personality.h>
29 #include <linux/security.h>
30 #include <linux/ima.h>
31 #include <linux/syscalls.h>
32 #include <linux/mount.h>
33 #include <linux/audit.h>
34 #include <linux/capability.h>
35 #include <linux/file.h>
36 #include <linux/fcntl.h>
37 #include <linux/device_cgroup.h>
38 #include <linux/fs_struct.h>
39 #include <linux/posix_acl.h>
40 #include <linux/hash.h>
41 #include <linux/bitops.h>
42 #include <linux/init_task.h>
43 #include <linux/uaccess.h>
48 /* [Feb-1997 T. Schoebel-Theuer]
49 * Fundamental changes in the pathname lookup mechanisms (namei)
50 * were necessary because of omirr. The reason is that omirr needs
51 * to know the _real_ pathname, not the user-supplied one, in case
52 * of symlinks (and also when transname replacements occur).
54 * The new code replaces the old recursive symlink resolution with
55 * an iterative one (in case of non-nested symlink chains). It does
56 * this with calls to <fs>_follow_link().
57 * As a side effect, dir_namei(), _namei() and follow_link() are now
58 * replaced with a single function lookup_dentry() that can handle all
59 * the special cases of the former code.
61 * With the new dcache, the pathname is stored at each inode, at least as
62 * long as the refcount of the inode is positive. As a side effect, the
63 * size of the dcache depends on the inode cache and thus is dynamic.
65 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
66 * resolution to correspond with current state of the code.
68 * Note that the symlink resolution is not *completely* iterative.
69 * There is still a significant amount of tail- and mid- recursion in
70 * the algorithm. Also, note that <fs>_readlink() is not used in
71 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
72 * may return different results than <fs>_follow_link(). Many virtual
73 * filesystems (including /proc) exhibit this behavior.
76 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
77 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
78 * and the name already exists in form of a symlink, try to create the new
79 * name indicated by the symlink. The old code always complained that the
80 * name already exists, due to not following the symlink even if its target
81 * is nonexistent. The new semantics affects also mknod() and link() when
82 * the name is a symlink pointing to a non-existent name.
84 * I don't know which semantics is the right one, since I have no access
85 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
86 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
87 * "old" one. Personally, I think the new semantics is much more logical.
88 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
89 * file does succeed in both HP-UX and SunOs, but not in Solaris
90 * and in the old Linux semantics.
93 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
94 * semantics. See the comments in "open_namei" and "do_link" below.
96 * [10-Sep-98 Alan Modra] Another symlink change.
99 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
100 * inside the path - always follow.
101 * in the last component in creation/removal/renaming - never follow.
102 * if LOOKUP_FOLLOW passed - follow.
103 * if the pathname has trailing slashes - follow.
104 * otherwise - don't follow.
105 * (applied in that order).
107 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
108 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
109 * During the 2.4 we need to fix the userland stuff depending on it -
110 * hopefully we will be able to get rid of that wart in 2.5. So far only
111 * XEmacs seems to be relying on it...
114 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
115 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
116 * any extra contention...
119 /* In order to reduce some races, while at the same time doing additional
120 * checking and hopefully speeding things up, we copy filenames to the
121 * kernel data space before using them..
123 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
124 * PATH_MAX includes the nul terminator --RR.
127 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
130 getname_flags(const char __user *filename, int flags, int *empty)
132 struct filename *result;
136 result = audit_reusename(filename);
140 result = __getname();
141 if (unlikely(!result))
142 return ERR_PTR(-ENOMEM);
145 * First, try to embed the struct filename inside the names_cache
148 kname = (char *)result->iname;
149 result->name = kname;
151 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
152 if (unlikely(len < 0)) {
158 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
159 * separate struct filename so we can dedicate the entire
160 * names_cache allocation for the pathname, and re-do the copy from
163 if (unlikely(len == EMBEDDED_NAME_MAX)) {
164 const size_t size = offsetof(struct filename, iname[1]);
165 kname = (char *)result;
168 * size is chosen that way we to guarantee that
169 * result->iname[0] is within the same object and that
170 * kname can't be equal to result->iname, no matter what.
172 result = kzalloc(size, GFP_KERNEL);
173 if (unlikely(!result)) {
175 return ERR_PTR(-ENOMEM);
177 result->name = kname;
178 len = strncpy_from_user(kname, filename, PATH_MAX);
179 if (unlikely(len < 0)) {
184 if (unlikely(len == PATH_MAX)) {
187 return ERR_PTR(-ENAMETOOLONG);
191 atomic_set(&result->refcnt, 1);
192 /* The empty path is special. */
193 if (unlikely(!len)) {
196 if (!(flags & LOOKUP_EMPTY)) {
198 return ERR_PTR(-ENOENT);
202 result->uptr = filename;
203 result->aname = NULL;
204 audit_getname(result);
209 getname_uflags(const char __user *filename, int uflags)
211 int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
213 return getname_flags(filename, flags, NULL);
217 getname(const char __user * filename)
219 return getname_flags(filename, 0, NULL);
223 getname_kernel(const char * filename)
225 struct filename *result;
226 int len = strlen(filename) + 1;
228 result = __getname();
229 if (unlikely(!result))
230 return ERR_PTR(-ENOMEM);
232 if (len <= EMBEDDED_NAME_MAX) {
233 result->name = (char *)result->iname;
234 } else if (len <= PATH_MAX) {
235 const size_t size = offsetof(struct filename, iname[1]);
236 struct filename *tmp;
238 tmp = kmalloc(size, GFP_KERNEL);
239 if (unlikely(!tmp)) {
241 return ERR_PTR(-ENOMEM);
243 tmp->name = (char *)result;
247 return ERR_PTR(-ENAMETOOLONG);
249 memcpy((char *)result->name, filename, len);
251 result->aname = NULL;
252 atomic_set(&result->refcnt, 1);
253 audit_getname(result);
257 EXPORT_SYMBOL(getname_kernel);
259 void putname(struct filename *name)
264 if (WARN_ON_ONCE(!atomic_read(&name->refcnt)))
267 if (!atomic_dec_and_test(&name->refcnt))
270 if (name->name != name->iname) {
271 __putname(name->name);
276 EXPORT_SYMBOL(putname);
279 * check_acl - perform ACL permission checking
280 * @idmap: idmap of the mount the inode was found from
281 * @inode: inode to check permissions on
282 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
284 * This function performs the ACL permission checking. Since this function
285 * retrieve POSIX acls it needs to know whether it is called from a blocking or
286 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
288 * If the inode has been found through an idmapped mount the idmap of
289 * the vfsmount must be passed through @idmap. This function will then take
290 * care to map the inode according to @idmap before checking permissions.
291 * On non-idmapped mounts or if permission checking is to be performed on the
292 * raw inode simply passs @nop_mnt_idmap.
294 static int check_acl(struct mnt_idmap *idmap,
295 struct inode *inode, int mask)
297 #ifdef CONFIG_FS_POSIX_ACL
298 struct posix_acl *acl;
300 if (mask & MAY_NOT_BLOCK) {
301 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
304 /* no ->get_inode_acl() calls in RCU mode... */
305 if (is_uncached_acl(acl))
307 return posix_acl_permission(idmap, inode, acl, mask);
310 acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
314 int error = posix_acl_permission(idmap, inode, acl, mask);
315 posix_acl_release(acl);
324 * acl_permission_check - perform basic UNIX permission checking
325 * @idmap: idmap of the mount the inode was found from
326 * @inode: inode to check permissions on
327 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
329 * This function performs the basic UNIX permission checking. Since this
330 * function may retrieve POSIX acls it needs to know whether it is called from a
331 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
333 * If the inode has been found through an idmapped mount the idmap of
334 * the vfsmount must be passed through @idmap. This function will then take
335 * care to map the inode according to @idmap before checking permissions.
336 * On non-idmapped mounts or if permission checking is to be performed on the
337 * raw inode simply passs @nop_mnt_idmap.
339 static int acl_permission_check(struct mnt_idmap *idmap,
340 struct inode *inode, int mask)
342 unsigned int mode = inode->i_mode;
345 /* Are we the owner? If so, ACL's don't matter */
346 vfsuid = i_uid_into_vfsuid(idmap, inode);
347 if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) {
350 return (mask & ~mode) ? -EACCES : 0;
353 /* Do we have ACL's? */
354 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
355 int error = check_acl(idmap, inode, mask);
356 if (error != -EAGAIN)
360 /* Only RWX matters for group/other mode bits */
364 * Are the group permissions different from
365 * the other permissions in the bits we care
366 * about? Need to check group ownership if so.
368 if (mask & (mode ^ (mode >> 3))) {
369 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
370 if (vfsgid_in_group_p(vfsgid))
374 /* Bits in 'mode' clear that we require? */
375 return (mask & ~mode) ? -EACCES : 0;
379 * generic_permission - check for access rights on a Posix-like filesystem
380 * @idmap: idmap of the mount the inode was found from
381 * @inode: inode to check access rights for
382 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
383 * %MAY_NOT_BLOCK ...)
385 * Used to check for read/write/execute permissions on a file.
386 * We use "fsuid" for this, letting us set arbitrary permissions
387 * for filesystem access without changing the "normal" uids which
388 * are used for other things.
390 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
391 * request cannot be satisfied (eg. requires blocking or too much complexity).
392 * It would then be called again in ref-walk mode.
394 * If the inode has been found through an idmapped mount the idmap of
395 * the vfsmount must be passed through @idmap. This function will then take
396 * care to map the inode according to @idmap before checking permissions.
397 * On non-idmapped mounts or if permission checking is to be performed on the
398 * raw inode simply passs @nop_mnt_idmap.
400 int generic_permission(struct mnt_idmap *idmap, struct inode *inode,
406 * Do the basic permission checks.
408 ret = acl_permission_check(idmap, inode, mask);
412 if (S_ISDIR(inode->i_mode)) {
413 /* DACs are overridable for directories */
414 if (!(mask & MAY_WRITE))
415 if (capable_wrt_inode_uidgid(idmap, inode,
416 CAP_DAC_READ_SEARCH))
418 if (capable_wrt_inode_uidgid(idmap, inode,
425 * Searching includes executable on directories, else just read.
427 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
428 if (mask == MAY_READ)
429 if (capable_wrt_inode_uidgid(idmap, inode,
430 CAP_DAC_READ_SEARCH))
433 * Read/write DACs are always overridable.
434 * Executable DACs are overridable when there is
435 * at least one exec bit set.
437 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
438 if (capable_wrt_inode_uidgid(idmap, inode,
444 EXPORT_SYMBOL(generic_permission);
447 * do_inode_permission - UNIX permission checking
448 * @idmap: idmap of the mount the inode was found from
449 * @inode: inode to check permissions on
450 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
452 * We _really_ want to just do "generic_permission()" without
453 * even looking at the inode->i_op values. So we keep a cache
454 * flag in inode->i_opflags, that says "this has not special
455 * permission function, use the fast case".
457 static inline int do_inode_permission(struct mnt_idmap *idmap,
458 struct inode *inode, int mask)
460 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
461 if (likely(inode->i_op->permission))
462 return inode->i_op->permission(idmap, inode, mask);
464 /* This gets set once for the inode lifetime */
465 spin_lock(&inode->i_lock);
466 inode->i_opflags |= IOP_FASTPERM;
467 spin_unlock(&inode->i_lock);
469 return generic_permission(idmap, inode, mask);
473 * sb_permission - Check superblock-level permissions
474 * @sb: Superblock of inode to check permission on
475 * @inode: Inode to check permission on
476 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
478 * Separate out file-system wide checks from inode-specific permission checks.
480 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
482 if (unlikely(mask & MAY_WRITE)) {
483 umode_t mode = inode->i_mode;
485 /* Nobody gets write access to a read-only fs. */
486 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
493 * inode_permission - Check for access rights to a given inode
494 * @idmap: idmap of the mount the inode was found from
495 * @inode: Inode to check permission on
496 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
498 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
499 * this, letting us set arbitrary permissions for filesystem access without
500 * changing the "normal" UIDs which are used for other things.
502 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
504 int inode_permission(struct mnt_idmap *idmap,
505 struct inode *inode, int mask)
509 retval = sb_permission(inode->i_sb, inode, mask);
513 if (unlikely(mask & MAY_WRITE)) {
515 * Nobody gets write access to an immutable file.
517 if (IS_IMMUTABLE(inode))
521 * Updating mtime will likely cause i_uid and i_gid to be
522 * written back improperly if their true value is unknown
525 if (HAS_UNMAPPED_ID(idmap, inode))
529 retval = do_inode_permission(idmap, inode, mask);
533 retval = devcgroup_inode_permission(inode, mask);
537 return security_inode_permission(inode, mask);
539 EXPORT_SYMBOL(inode_permission);
542 * path_get - get a reference to a path
543 * @path: path to get the reference to
545 * Given a path increment the reference count to the dentry and the vfsmount.
547 void path_get(const struct path *path)
552 EXPORT_SYMBOL(path_get);
555 * path_put - put a reference to a path
556 * @path: path to put the reference to
558 * Given a path decrement the reference count to the dentry and the vfsmount.
560 void path_put(const struct path *path)
565 EXPORT_SYMBOL(path_put);
567 #define EMBEDDED_LEVELS 2
572 struct inode *inode; /* path.dentry.d_inode */
573 unsigned int flags, state;
574 unsigned seq, next_seq, m_seq, r_seq;
577 int total_link_count;
580 struct delayed_call done;
583 } *stack, internal[EMBEDDED_LEVELS];
584 struct filename *name;
585 struct nameidata *saved;
590 } __randomize_layout;
592 #define ND_ROOT_PRESET 1
593 #define ND_ROOT_GRABBED 2
596 static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
598 struct nameidata *old = current->nameidata;
599 p->stack = p->internal;
604 p->path.dentry = NULL;
605 p->total_link_count = old ? old->total_link_count : 0;
607 current->nameidata = p;
610 static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
611 const struct path *root)
613 __set_nameidata(p, dfd, name);
615 if (unlikely(root)) {
616 p->state = ND_ROOT_PRESET;
621 static void restore_nameidata(void)
623 struct nameidata *now = current->nameidata, *old = now->saved;
625 current->nameidata = old;
627 old->total_link_count = now->total_link_count;
628 if (now->stack != now->internal)
632 static bool nd_alloc_stack(struct nameidata *nd)
636 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
637 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
640 memcpy(p, nd->internal, sizeof(nd->internal));
646 * path_connected - Verify that a dentry is below mnt.mnt_root
647 * @mnt: The mountpoint to check.
648 * @dentry: The dentry to check.
650 * Rename can sometimes move a file or directory outside of a bind
651 * mount, path_connected allows those cases to be detected.
653 static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
655 struct super_block *sb = mnt->mnt_sb;
657 /* Bind mounts can have disconnected paths */
658 if (mnt->mnt_root == sb->s_root)
661 return is_subdir(dentry, mnt->mnt_root);
664 static void drop_links(struct nameidata *nd)
668 struct saved *last = nd->stack + i;
669 do_delayed_call(&last->done);
670 clear_delayed_call(&last->done);
674 static void leave_rcu(struct nameidata *nd)
676 nd->flags &= ~LOOKUP_RCU;
677 nd->seq = nd->next_seq = 0;
681 static void terminate_walk(struct nameidata *nd)
684 if (!(nd->flags & LOOKUP_RCU)) {
687 for (i = 0; i < nd->depth; i++)
688 path_put(&nd->stack[i].link);
689 if (nd->state & ND_ROOT_GRABBED) {
691 nd->state &= ~ND_ROOT_GRABBED;
698 nd->path.dentry = NULL;
701 /* path_put is needed afterwards regardless of success or failure */
702 static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
704 int res = __legitimize_mnt(path->mnt, mseq);
711 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
715 return !read_seqcount_retry(&path->dentry->d_seq, seq);
718 static inline bool legitimize_path(struct nameidata *nd,
719 struct path *path, unsigned seq)
721 return __legitimize_path(path, seq, nd->m_seq);
724 static bool legitimize_links(struct nameidata *nd)
727 if (unlikely(nd->flags & LOOKUP_CACHED)) {
732 for (i = 0; i < nd->depth; i++) {
733 struct saved *last = nd->stack + i;
734 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
743 static bool legitimize_root(struct nameidata *nd)
745 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
746 if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
748 nd->state |= ND_ROOT_GRABBED;
749 return legitimize_path(nd, &nd->root, nd->root_seq);
753 * Path walking has 2 modes, rcu-walk and ref-walk (see
754 * Documentation/filesystems/path-lookup.txt). In situations when we can't
755 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
756 * normal reference counts on dentries and vfsmounts to transition to ref-walk
757 * mode. Refcounts are grabbed at the last known good point before rcu-walk
758 * got stuck, so ref-walk may continue from there. If this is not successful
759 * (eg. a seqcount has changed), then failure is returned and it's up to caller
760 * to restart the path walk from the beginning in ref-walk mode.
764 * try_to_unlazy - try to switch to ref-walk mode.
765 * @nd: nameidata pathwalk data
766 * Returns: true on success, false on failure
768 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
770 * Must be called from rcu-walk context.
771 * Nothing should touch nameidata between try_to_unlazy() failure and
774 static bool try_to_unlazy(struct nameidata *nd)
776 struct dentry *parent = nd->path.dentry;
778 BUG_ON(!(nd->flags & LOOKUP_RCU));
780 if (unlikely(!legitimize_links(nd)))
782 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
784 if (unlikely(!legitimize_root(nd)))
787 BUG_ON(nd->inode != parent->d_inode);
792 nd->path.dentry = NULL;
799 * try_to_unlazy_next - try to switch to ref-walk mode.
800 * @nd: nameidata pathwalk data
801 * @dentry: next dentry to step into
802 * Returns: true on success, false on failure
804 * Similar to try_to_unlazy(), but here we have the next dentry already
805 * picked by rcu-walk and want to legitimize that in addition to the current
806 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
807 * Nothing should touch nameidata between try_to_unlazy_next() failure and
810 static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry)
813 BUG_ON(!(nd->flags & LOOKUP_RCU));
815 if (unlikely(!legitimize_links(nd)))
817 res = __legitimize_mnt(nd->path.mnt, nd->m_seq);
823 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
827 * We need to move both the parent and the dentry from the RCU domain
828 * to be properly refcounted. And the sequence number in the dentry
829 * validates *both* dentry counters, since we checked the sequence
830 * number of the parent after we got the child sequence number. So we
831 * know the parent must still be valid if the child sequence number is
833 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
835 if (read_seqcount_retry(&dentry->d_seq, nd->next_seq))
838 * Sequence counts matched. Now make sure that the root is
839 * still valid and get it if required.
841 if (unlikely(!legitimize_root(nd)))
849 nd->path.dentry = NULL;
859 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
861 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
862 return dentry->d_op->d_revalidate(dentry, flags);
868 * complete_walk - successful completion of path walk
869 * @nd: pointer nameidata
871 * If we had been in RCU mode, drop out of it and legitimize nd->path.
872 * Revalidate the final result, unless we'd already done that during
873 * the path walk or the filesystem doesn't ask for it. Return 0 on
874 * success, -error on failure. In case of failure caller does not
875 * need to drop nd->path.
877 static int complete_walk(struct nameidata *nd)
879 struct dentry *dentry = nd->path.dentry;
882 if (nd->flags & LOOKUP_RCU) {
884 * We don't want to zero nd->root for scoped-lookups or
885 * externally-managed nd->root.
887 if (!(nd->state & ND_ROOT_PRESET))
888 if (!(nd->flags & LOOKUP_IS_SCOPED))
890 nd->flags &= ~LOOKUP_CACHED;
891 if (!try_to_unlazy(nd))
895 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
897 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
898 * ever step outside the root during lookup" and should already
899 * be guaranteed by the rest of namei, we want to avoid a namei
900 * BUG resulting in userspace being given a path that was not
901 * scoped within the root at some point during the lookup.
903 * So, do a final sanity-check to make sure that in the
904 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
905 * we won't silently return an fd completely outside of the
906 * requested root to userspace.
908 * Userspace could move the path outside the root after this
909 * check, but as discussed elsewhere this is not a concern (the
910 * resolved file was inside the root at some point).
912 if (!path_is_under(&nd->path, &nd->root))
916 if (likely(!(nd->state & ND_JUMPED)))
919 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
922 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
932 static int set_root(struct nameidata *nd)
934 struct fs_struct *fs = current->fs;
937 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
938 * still have to ensure it doesn't happen because it will cause a breakout
941 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
942 return -ENOTRECOVERABLE;
944 if (nd->flags & LOOKUP_RCU) {
948 seq = read_seqcount_begin(&fs->seq);
950 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
951 } while (read_seqcount_retry(&fs->seq, seq));
953 get_fs_root(fs, &nd->root);
954 nd->state |= ND_ROOT_GRABBED;
959 static int nd_jump_root(struct nameidata *nd)
961 if (unlikely(nd->flags & LOOKUP_BENEATH))
963 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
964 /* Absolute path arguments to path_init() are allowed. */
965 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
969 int error = set_root(nd);
973 if (nd->flags & LOOKUP_RCU) {
977 nd->inode = d->d_inode;
978 nd->seq = nd->root_seq;
979 if (read_seqcount_retry(&d->d_seq, nd->seq))
985 nd->inode = nd->path.dentry->d_inode;
987 nd->state |= ND_JUMPED;
992 * Helper to directly jump to a known parsed path from ->get_link,
993 * caller must have taken a reference to path beforehand.
995 int nd_jump_link(const struct path *path)
998 struct nameidata *nd = current->nameidata;
1000 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
1004 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
1005 if (nd->path.mnt != path->mnt)
1008 /* Not currently safe for scoped-lookups. */
1009 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1012 path_put(&nd->path);
1014 nd->inode = nd->path.dentry->d_inode;
1015 nd->state |= ND_JUMPED;
1023 static inline void put_link(struct nameidata *nd)
1025 struct saved *last = nd->stack + --nd->depth;
1026 do_delayed_call(&last->done);
1027 if (!(nd->flags & LOOKUP_RCU))
1028 path_put(&last->link);
1031 static int sysctl_protected_symlinks __read_mostly;
1032 static int sysctl_protected_hardlinks __read_mostly;
1033 static int sysctl_protected_fifos __read_mostly;
1034 static int sysctl_protected_regular __read_mostly;
1036 #ifdef CONFIG_SYSCTL
1037 static struct ctl_table namei_sysctls[] = {
1039 .procname = "protected_symlinks",
1040 .data = &sysctl_protected_symlinks,
1041 .maxlen = sizeof(int),
1043 .proc_handler = proc_dointvec_minmax,
1044 .extra1 = SYSCTL_ZERO,
1045 .extra2 = SYSCTL_ONE,
1048 .procname = "protected_hardlinks",
1049 .data = &sysctl_protected_hardlinks,
1050 .maxlen = sizeof(int),
1052 .proc_handler = proc_dointvec_minmax,
1053 .extra1 = SYSCTL_ZERO,
1054 .extra2 = SYSCTL_ONE,
1057 .procname = "protected_fifos",
1058 .data = &sysctl_protected_fifos,
1059 .maxlen = sizeof(int),
1061 .proc_handler = proc_dointvec_minmax,
1062 .extra1 = SYSCTL_ZERO,
1063 .extra2 = SYSCTL_TWO,
1066 .procname = "protected_regular",
1067 .data = &sysctl_protected_regular,
1068 .maxlen = sizeof(int),
1070 .proc_handler = proc_dointvec_minmax,
1071 .extra1 = SYSCTL_ZERO,
1072 .extra2 = SYSCTL_TWO,
1077 static int __init init_fs_namei_sysctls(void)
1079 register_sysctl_init("fs", namei_sysctls);
1082 fs_initcall(init_fs_namei_sysctls);
1084 #endif /* CONFIG_SYSCTL */
1087 * may_follow_link - Check symlink following for unsafe situations
1088 * @nd: nameidata pathwalk data
1089 * @inode: Used for idmapping.
1091 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1092 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1093 * in a sticky world-writable directory. This is to protect privileged
1094 * processes from failing races against path names that may change out
1095 * from under them by way of other users creating malicious symlinks.
1096 * It will permit symlinks to be followed only when outside a sticky
1097 * world-writable directory, or when the uid of the symlink and follower
1098 * match, or when the directory owner matches the symlink's owner.
1100 * Returns 0 if following the symlink is allowed, -ve on error.
1102 static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1104 struct mnt_idmap *idmap;
1107 if (!sysctl_protected_symlinks)
1110 idmap = mnt_idmap(nd->path.mnt);
1111 vfsuid = i_uid_into_vfsuid(idmap, inode);
1112 /* Allowed if owner and follower match. */
1113 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
1116 /* Allowed if parent directory not sticky and world-writable. */
1117 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1120 /* Allowed if parent directory and link owner match. */
1121 if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid))
1124 if (nd->flags & LOOKUP_RCU)
1127 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1128 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1133 * safe_hardlink_source - Check for safe hardlink conditions
1134 * @idmap: idmap of the mount the inode was found from
1135 * @inode: the source inode to hardlink from
1137 * Return false if at least one of the following conditions:
1138 * - inode is not a regular file
1140 * - inode is setgid and group-exec
1141 * - access failure for read and write
1143 * Otherwise returns true.
1145 static bool safe_hardlink_source(struct mnt_idmap *idmap,
1146 struct inode *inode)
1148 umode_t mode = inode->i_mode;
1150 /* Special files should not get pinned to the filesystem. */
1154 /* Setuid files should not get pinned to the filesystem. */
1158 /* Executable setgid files should not get pinned to the filesystem. */
1159 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1162 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1163 if (inode_permission(idmap, inode, MAY_READ | MAY_WRITE))
1170 * may_linkat - Check permissions for creating a hardlink
1171 * @idmap: idmap of the mount the inode was found from
1172 * @link: the source to hardlink from
1174 * Block hardlink when all of:
1175 * - sysctl_protected_hardlinks enabled
1176 * - fsuid does not match inode
1177 * - hardlink source is unsafe (see safe_hardlink_source() above)
1178 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1180 * If the inode has been found through an idmapped mount the idmap of
1181 * the vfsmount must be passed through @idmap. This function will then take
1182 * care to map the inode according to @idmap before checking permissions.
1183 * On non-idmapped mounts or if permission checking is to be performed on the
1184 * raw inode simply pass @nop_mnt_idmap.
1186 * Returns 0 if successful, -ve on error.
1188 int may_linkat(struct mnt_idmap *idmap, const struct path *link)
1190 struct inode *inode = link->dentry->d_inode;
1192 /* Inode writeback is not safe when the uid or gid are invalid. */
1193 if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
1194 !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
1197 if (!sysctl_protected_hardlinks)
1200 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1201 * otherwise, it must be a safe source.
1203 if (safe_hardlink_source(idmap, inode) ||
1204 inode_owner_or_capable(idmap, inode))
1207 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1212 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1213 * should be allowed, or not, on files that already
1215 * @idmap: idmap of the mount the inode was found from
1216 * @nd: nameidata pathwalk data
1217 * @inode: the inode of the file to open
1219 * Block an O_CREAT open of a FIFO (or a regular file) when:
1220 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1221 * - the file already exists
1222 * - we are in a sticky directory
1223 * - we don't own the file
1224 * - the owner of the directory doesn't own the file
1225 * - the directory is world writable
1226 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1227 * the directory doesn't have to be world writable: being group writable will
1230 * If the inode has been found through an idmapped mount the idmap of
1231 * the vfsmount must be passed through @idmap. This function will then take
1232 * care to map the inode according to @idmap before checking permissions.
1233 * On non-idmapped mounts or if permission checking is to be performed on the
1234 * raw inode simply pass @nop_mnt_idmap.
1236 * Returns 0 if the open is allowed, -ve on error.
1238 static int may_create_in_sticky(struct mnt_idmap *idmap,
1239 struct nameidata *nd, struct inode *const inode)
1241 umode_t dir_mode = nd->dir_mode;
1242 vfsuid_t dir_vfsuid = nd->dir_vfsuid;
1244 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1245 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1246 likely(!(dir_mode & S_ISVTX)) ||
1247 vfsuid_eq(i_uid_into_vfsuid(idmap, inode), dir_vfsuid) ||
1248 vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), current_fsuid()))
1251 if (likely(dir_mode & 0002) ||
1253 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1254 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1255 const char *operation = S_ISFIFO(inode->i_mode) ?
1256 "sticky_create_fifo" :
1257 "sticky_create_regular";
1258 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1265 * follow_up - Find the mountpoint of path's vfsmount
1267 * Given a path, find the mountpoint of its source file system.
1268 * Replace @path with the path of the mountpoint in the parent mount.
1271 * Return 1 if we went up a level and 0 if we were already at the
1274 int follow_up(struct path *path)
1276 struct mount *mnt = real_mount(path->mnt);
1277 struct mount *parent;
1278 struct dentry *mountpoint;
1280 read_seqlock_excl(&mount_lock);
1281 parent = mnt->mnt_parent;
1282 if (parent == mnt) {
1283 read_sequnlock_excl(&mount_lock);
1286 mntget(&parent->mnt);
1287 mountpoint = dget(mnt->mnt_mountpoint);
1288 read_sequnlock_excl(&mount_lock);
1290 path->dentry = mountpoint;
1292 path->mnt = &parent->mnt;
1295 EXPORT_SYMBOL(follow_up);
1297 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1298 struct path *path, unsigned *seqp)
1300 while (mnt_has_parent(m)) {
1301 struct dentry *mountpoint = m->mnt_mountpoint;
1304 if (unlikely(root->dentry == mountpoint &&
1305 root->mnt == &m->mnt))
1307 if (mountpoint != m->mnt.mnt_root) {
1308 path->mnt = &m->mnt;
1309 path->dentry = mountpoint;
1310 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1317 static bool choose_mountpoint(struct mount *m, const struct path *root,
1324 unsigned seq, mseq = read_seqbegin(&mount_lock);
1326 found = choose_mountpoint_rcu(m, root, path, &seq);
1327 if (unlikely(!found)) {
1328 if (!read_seqretry(&mount_lock, mseq))
1331 if (likely(__legitimize_path(path, seq, mseq)))
1343 * Perform an automount
1344 * - return -EISDIR to tell follow_managed() to stop and return the path we
1347 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1349 struct dentry *dentry = path->dentry;
1351 /* We don't want to mount if someone's just doing a stat -
1352 * unless they're stat'ing a directory and appended a '/' to
1355 * We do, however, want to mount if someone wants to open or
1356 * create a file of any type under the mountpoint, wants to
1357 * traverse through the mountpoint or wants to open the
1358 * mounted directory. Also, autofs may mark negative dentries
1359 * as being automount points. These will need the attentions
1360 * of the daemon to instantiate them before they can be used.
1362 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1363 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1367 if (count && (*count)++ >= MAXSYMLINKS)
1370 return finish_automount(dentry->d_op->d_automount(path), path);
1374 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1375 * dentries are pinned but not locked here, so negative dentry can go
1376 * positive right under us. Use of smp_load_acquire() provides a barrier
1377 * sufficient for ->d_inode and ->d_flags consistency.
1379 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1380 int *count, unsigned lookup_flags)
1382 struct vfsmount *mnt = path->mnt;
1383 bool need_mntput = false;
1386 while (flags & DCACHE_MANAGED_DENTRY) {
1387 /* Allow the filesystem to manage the transit without i_mutex
1389 if (flags & DCACHE_MANAGE_TRANSIT) {
1390 ret = path->dentry->d_op->d_manage(path, false);
1391 flags = smp_load_acquire(&path->dentry->d_flags);
1396 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1397 struct vfsmount *mounted = lookup_mnt(path);
1398 if (mounted) { // ... in our namespace
1402 path->mnt = mounted;
1403 path->dentry = dget(mounted->mnt_root);
1404 // here we know it's positive
1405 flags = path->dentry->d_flags;
1411 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1414 // uncovered automount point
1415 ret = follow_automount(path, count, lookup_flags);
1416 flags = smp_load_acquire(&path->dentry->d_flags);
1423 // possible if you race with several mount --move
1424 if (need_mntput && path->mnt == mnt)
1426 if (!ret && unlikely(d_flags_negative(flags)))
1428 *jumped = need_mntput;
1432 static inline int traverse_mounts(struct path *path, bool *jumped,
1433 int *count, unsigned lookup_flags)
1435 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1438 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1440 if (unlikely(d_flags_negative(flags)))
1444 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1447 int follow_down_one(struct path *path)
1449 struct vfsmount *mounted;
1451 mounted = lookup_mnt(path);
1455 path->mnt = mounted;
1456 path->dentry = dget(mounted->mnt_root);
1461 EXPORT_SYMBOL(follow_down_one);
1464 * Follow down to the covering mount currently visible to userspace. At each
1465 * point, the filesystem owning that dentry may be queried as to whether the
1466 * caller is permitted to proceed or not.
1468 int follow_down(struct path *path, unsigned int flags)
1470 struct vfsmount *mnt = path->mnt;
1472 int ret = traverse_mounts(path, &jumped, NULL, flags);
1474 if (path->mnt != mnt)
1478 EXPORT_SYMBOL(follow_down);
1481 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1482 * we meet a managed dentry that would need blocking.
1484 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1486 struct dentry *dentry = path->dentry;
1487 unsigned int flags = dentry->d_flags;
1489 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1492 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1497 * Don't forget we might have a non-mountpoint managed dentry
1498 * that wants to block transit.
1500 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1501 int res = dentry->d_op->d_manage(path, true);
1503 return res == -EISDIR;
1504 flags = dentry->d_flags;
1507 if (flags & DCACHE_MOUNTED) {
1508 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1510 path->mnt = &mounted->mnt;
1511 dentry = path->dentry = mounted->mnt.mnt_root;
1512 nd->state |= ND_JUMPED;
1513 nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1514 flags = dentry->d_flags;
1515 // makes sure that non-RCU pathwalk could reach
1517 if (read_seqretry(&mount_lock, nd->m_seq))
1521 if (read_seqretry(&mount_lock, nd->m_seq))
1524 return !(flags & DCACHE_NEED_AUTOMOUNT);
1528 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1534 path->mnt = nd->path.mnt;
1535 path->dentry = dentry;
1536 if (nd->flags & LOOKUP_RCU) {
1537 unsigned int seq = nd->next_seq;
1538 if (likely(__follow_mount_rcu(nd, path)))
1540 // *path and nd->next_seq might've been clobbered
1541 path->mnt = nd->path.mnt;
1542 path->dentry = dentry;
1544 if (!try_to_unlazy_next(nd, dentry))
1547 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1549 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1552 nd->state |= ND_JUMPED;
1554 if (unlikely(ret)) {
1556 if (path->mnt != nd->path.mnt)
1563 * This looks up the name in dcache and possibly revalidates the found dentry.
1564 * NULL is returned if the dentry does not exist in the cache.
1566 static struct dentry *lookup_dcache(const struct qstr *name,
1570 struct dentry *dentry = d_lookup(dir, name);
1572 int error = d_revalidate(dentry, flags);
1573 if (unlikely(error <= 0)) {
1575 d_invalidate(dentry);
1577 return ERR_PTR(error);
1584 * Parent directory has inode locked exclusive. This is one
1585 * and only case when ->lookup() gets called on non in-lookup
1586 * dentries - as the matter of fact, this only gets called
1587 * when directory is guaranteed to have no in-lookup children
1590 struct dentry *lookup_one_qstr_excl(const struct qstr *name,
1591 struct dentry *base,
1594 struct dentry *dentry = lookup_dcache(name, base, flags);
1596 struct inode *dir = base->d_inode;
1601 /* Don't create child dentry for a dead directory. */
1602 if (unlikely(IS_DEADDIR(dir)))
1603 return ERR_PTR(-ENOENT);
1605 dentry = d_alloc(base, name);
1606 if (unlikely(!dentry))
1607 return ERR_PTR(-ENOMEM);
1609 old = dir->i_op->lookup(dir, dentry, flags);
1610 if (unlikely(old)) {
1616 EXPORT_SYMBOL(lookup_one_qstr_excl);
1618 static struct dentry *lookup_fast(struct nameidata *nd)
1620 struct dentry *dentry, *parent = nd->path.dentry;
1624 * Rename seqlock is not required here because in the off chance
1625 * of a false negative due to a concurrent rename, the caller is
1626 * going to fall back to non-racy lookup.
1628 if (nd->flags & LOOKUP_RCU) {
1629 dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq);
1630 if (unlikely(!dentry)) {
1631 if (!try_to_unlazy(nd))
1632 return ERR_PTR(-ECHILD);
1637 * This sequence count validates that the parent had no
1638 * changes while we did the lookup of the dentry above.
1640 if (read_seqcount_retry(&parent->d_seq, nd->seq))
1641 return ERR_PTR(-ECHILD);
1643 status = d_revalidate(dentry, nd->flags);
1644 if (likely(status > 0))
1646 if (!try_to_unlazy_next(nd, dentry))
1647 return ERR_PTR(-ECHILD);
1648 if (status == -ECHILD)
1649 /* we'd been told to redo it in non-rcu mode */
1650 status = d_revalidate(dentry, nd->flags);
1652 dentry = __d_lookup(parent, &nd->last);
1653 if (unlikely(!dentry))
1655 status = d_revalidate(dentry, nd->flags);
1657 if (unlikely(status <= 0)) {
1659 d_invalidate(dentry);
1661 return ERR_PTR(status);
1666 /* Fast lookup failed, do it the slow way */
1667 static struct dentry *__lookup_slow(const struct qstr *name,
1671 struct dentry *dentry, *old;
1672 struct inode *inode = dir->d_inode;
1673 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1675 /* Don't go there if it's already dead */
1676 if (unlikely(IS_DEADDIR(inode)))
1677 return ERR_PTR(-ENOENT);
1679 dentry = d_alloc_parallel(dir, name, &wq);
1682 if (unlikely(!d_in_lookup(dentry))) {
1683 int error = d_revalidate(dentry, flags);
1684 if (unlikely(error <= 0)) {
1686 d_invalidate(dentry);
1691 dentry = ERR_PTR(error);
1694 old = inode->i_op->lookup(inode, dentry, flags);
1695 d_lookup_done(dentry);
1696 if (unlikely(old)) {
1704 static struct dentry *lookup_slow(const struct qstr *name,
1708 struct inode *inode = dir->d_inode;
1710 inode_lock_shared(inode);
1711 res = __lookup_slow(name, dir, flags);
1712 inode_unlock_shared(inode);
1716 static inline int may_lookup(struct mnt_idmap *idmap,
1717 struct nameidata *nd)
1719 if (nd->flags & LOOKUP_RCU) {
1720 int err = inode_permission(idmap, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1721 if (err != -ECHILD || !try_to_unlazy(nd))
1724 return inode_permission(idmap, nd->inode, MAY_EXEC);
1727 static int reserve_stack(struct nameidata *nd, struct path *link)
1729 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1732 if (likely(nd->depth != EMBEDDED_LEVELS))
1734 if (likely(nd->stack != nd->internal))
1736 if (likely(nd_alloc_stack(nd)))
1739 if (nd->flags & LOOKUP_RCU) {
1740 // we need to grab link before we do unlazy. And we can't skip
1741 // unlazy even if we fail to grab the link - cleanup needs it
1742 bool grabbed_link = legitimize_path(nd, link, nd->next_seq);
1744 if (!try_to_unlazy(nd) || !grabbed_link)
1747 if (nd_alloc_stack(nd))
1753 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1755 static const char *pick_link(struct nameidata *nd, struct path *link,
1756 struct inode *inode, int flags)
1760 int error = reserve_stack(nd, link);
1762 if (unlikely(error)) {
1763 if (!(nd->flags & LOOKUP_RCU))
1765 return ERR_PTR(error);
1767 last = nd->stack + nd->depth++;
1769 clear_delayed_call(&last->done);
1770 last->seq = nd->next_seq;
1772 if (flags & WALK_TRAILING) {
1773 error = may_follow_link(nd, inode);
1774 if (unlikely(error))
1775 return ERR_PTR(error);
1778 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1779 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1780 return ERR_PTR(-ELOOP);
1782 if (!(nd->flags & LOOKUP_RCU)) {
1783 touch_atime(&last->link);
1785 } else if (atime_needs_update(&last->link, inode)) {
1786 if (!try_to_unlazy(nd))
1787 return ERR_PTR(-ECHILD);
1788 touch_atime(&last->link);
1791 error = security_inode_follow_link(link->dentry, inode,
1792 nd->flags & LOOKUP_RCU);
1793 if (unlikely(error))
1794 return ERR_PTR(error);
1796 res = READ_ONCE(inode->i_link);
1798 const char * (*get)(struct dentry *, struct inode *,
1799 struct delayed_call *);
1800 get = inode->i_op->get_link;
1801 if (nd->flags & LOOKUP_RCU) {
1802 res = get(NULL, inode, &last->done);
1803 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1804 res = get(link->dentry, inode, &last->done);
1806 res = get(link->dentry, inode, &last->done);
1814 error = nd_jump_root(nd);
1815 if (unlikely(error))
1816 return ERR_PTR(error);
1817 while (unlikely(*++res == '/'))
1822 all_done: // pure jump
1828 * Do we need to follow links? We _really_ want to be able
1829 * to do this check without having to look at inode->i_op,
1830 * so we keep a cache of "no, this doesn't need follow_link"
1831 * for the common case.
1833 * NOTE: dentry must be what nd->next_seq had been sampled from.
1835 static const char *step_into(struct nameidata *nd, int flags,
1836 struct dentry *dentry)
1839 struct inode *inode;
1840 int err = handle_mounts(nd, dentry, &path);
1843 return ERR_PTR(err);
1844 inode = path.dentry->d_inode;
1845 if (likely(!d_is_symlink(path.dentry)) ||
1846 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1847 (flags & WALK_NOFOLLOW)) {
1848 /* not a symlink or should not follow */
1849 if (nd->flags & LOOKUP_RCU) {
1850 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1851 return ERR_PTR(-ECHILD);
1852 if (unlikely(!inode))
1853 return ERR_PTR(-ENOENT);
1855 dput(nd->path.dentry);
1856 if (nd->path.mnt != path.mnt)
1857 mntput(nd->path.mnt);
1861 nd->seq = nd->next_seq;
1864 if (nd->flags & LOOKUP_RCU) {
1865 /* make sure that d_is_symlink above matches inode */
1866 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1867 return ERR_PTR(-ECHILD);
1869 if (path.mnt == nd->path.mnt)
1872 return pick_link(nd, &path, inode, flags);
1875 static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1877 struct dentry *parent, *old;
1879 if (path_equal(&nd->path, &nd->root))
1881 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1884 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1885 &nd->root, &path, &seq))
1887 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1888 return ERR_PTR(-ECHILD);
1890 nd->inode = path.dentry->d_inode;
1892 // makes sure that non-RCU pathwalk could reach this state
1893 if (read_seqretry(&mount_lock, nd->m_seq))
1894 return ERR_PTR(-ECHILD);
1895 /* we know that mountpoint was pinned */
1897 old = nd->path.dentry;
1898 parent = old->d_parent;
1899 nd->next_seq = read_seqcount_begin(&parent->d_seq);
1900 // makes sure that non-RCU pathwalk could reach this state
1901 if (read_seqcount_retry(&old->d_seq, nd->seq))
1902 return ERR_PTR(-ECHILD);
1903 if (unlikely(!path_connected(nd->path.mnt, parent)))
1904 return ERR_PTR(-ECHILD);
1907 if (read_seqretry(&mount_lock, nd->m_seq))
1908 return ERR_PTR(-ECHILD);
1909 if (unlikely(nd->flags & LOOKUP_BENEATH))
1910 return ERR_PTR(-ECHILD);
1911 nd->next_seq = nd->seq;
1912 return nd->path.dentry;
1915 static struct dentry *follow_dotdot(struct nameidata *nd)
1917 struct dentry *parent;
1919 if (path_equal(&nd->path, &nd->root))
1921 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1924 if (!choose_mountpoint(real_mount(nd->path.mnt),
1927 path_put(&nd->path);
1929 nd->inode = path.dentry->d_inode;
1930 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1931 return ERR_PTR(-EXDEV);
1933 /* rare case of legitimate dget_parent()... */
1934 parent = dget_parent(nd->path.dentry);
1935 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1937 return ERR_PTR(-ENOENT);
1942 if (unlikely(nd->flags & LOOKUP_BENEATH))
1943 return ERR_PTR(-EXDEV);
1944 return dget(nd->path.dentry);
1947 static const char *handle_dots(struct nameidata *nd, int type)
1949 if (type == LAST_DOTDOT) {
1950 const char *error = NULL;
1951 struct dentry *parent;
1953 if (!nd->root.mnt) {
1954 error = ERR_PTR(set_root(nd));
1958 if (nd->flags & LOOKUP_RCU)
1959 parent = follow_dotdot_rcu(nd);
1961 parent = follow_dotdot(nd);
1963 return ERR_CAST(parent);
1964 error = step_into(nd, WALK_NOFOLLOW, parent);
1965 if (unlikely(error))
1968 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1970 * If there was a racing rename or mount along our
1971 * path, then we can't be sure that ".." hasn't jumped
1972 * above nd->root (and so userspace should retry or use
1976 if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
1977 return ERR_PTR(-EAGAIN);
1978 if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
1979 return ERR_PTR(-EAGAIN);
1985 static const char *walk_component(struct nameidata *nd, int flags)
1987 struct dentry *dentry;
1989 * "." and ".." are special - ".." especially so because it has
1990 * to be able to know about the current root directory and
1991 * parent relationships.
1993 if (unlikely(nd->last_type != LAST_NORM)) {
1994 if (!(flags & WALK_MORE) && nd->depth)
1996 return handle_dots(nd, nd->last_type);
1998 dentry = lookup_fast(nd);
2000 return ERR_CAST(dentry);
2001 if (unlikely(!dentry)) {
2002 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
2004 return ERR_CAST(dentry);
2006 if (!(flags & WALK_MORE) && nd->depth)
2008 return step_into(nd, flags, dentry);
2012 * We can do the critical dentry name comparison and hashing
2013 * operations one word at a time, but we are limited to:
2015 * - Architectures with fast unaligned word accesses. We could
2016 * do a "get_unaligned()" if this helps and is sufficiently
2019 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2020 * do not trap on the (extremely unlikely) case of a page
2021 * crossing operation.
2023 * - Furthermore, we need an efficient 64-bit compile for the
2024 * 64-bit case in order to generate the "number of bytes in
2025 * the final mask". Again, that could be replaced with a
2026 * efficient population count instruction or similar.
2028 #ifdef CONFIG_DCACHE_WORD_ACCESS
2030 #include <asm/word-at-a-time.h>
2034 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2036 #elif defined(CONFIG_64BIT)
2038 * Register pressure in the mixing function is an issue, particularly
2039 * on 32-bit x86, but almost any function requires one state value and
2040 * one temporary. Instead, use a function designed for two state values
2041 * and no temporaries.
2043 * This function cannot create a collision in only two iterations, so
2044 * we have two iterations to achieve avalanche. In those two iterations,
2045 * we have six layers of mixing, which is enough to spread one bit's
2046 * influence out to 2^6 = 64 state bits.
2048 * Rotate constants are scored by considering either 64 one-bit input
2049 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2050 * probability of that delta causing a change to each of the 128 output
2051 * bits, using a sample of random initial states.
2053 * The Shannon entropy of the computed probabilities is then summed
2054 * to produce a score. Ideally, any input change has a 50% chance of
2055 * toggling any given output bit.
2057 * Mixing scores (in bits) for (12,45):
2058 * Input delta: 1-bit 2-bit
2059 * 1 round: 713.3 42542.6
2060 * 2 rounds: 2753.7 140389.8
2061 * 3 rounds: 5954.1 233458.2
2062 * 4 rounds: 7862.6 256672.2
2063 * Perfect: 8192 258048
2064 * (64*128) (64*63/2 * 128)
2066 #define HASH_MIX(x, y, a) \
2068 y ^= x, x = rol64(x,12),\
2069 x += y, y = rol64(y,45),\
2073 * Fold two longs into one 32-bit hash value. This must be fast, but
2074 * latency isn't quite as critical, as there is a fair bit of additional
2075 * work done before the hash value is used.
2077 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2079 y ^= x * GOLDEN_RATIO_64;
2080 y *= GOLDEN_RATIO_64;
2084 #else /* 32-bit case */
2087 * Mixing scores (in bits) for (7,20):
2088 * Input delta: 1-bit 2-bit
2089 * 1 round: 330.3 9201.6
2090 * 2 rounds: 1246.4 25475.4
2091 * 3 rounds: 1907.1 31295.1
2092 * 4 rounds: 2042.3 31718.6
2093 * Perfect: 2048 31744
2094 * (32*64) (32*31/2 * 64)
2096 #define HASH_MIX(x, y, a) \
2098 y ^= x, x = rol32(x, 7),\
2099 x += y, y = rol32(y,20),\
2102 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2104 /* Use arch-optimized multiply if one exists */
2105 return __hash_32(y ^ __hash_32(x));
2111 * Return the hash of a string of known length. This is carfully
2112 * designed to match hash_name(), which is the more critical function.
2113 * In particular, we must end by hashing a final word containing 0..7
2114 * payload bytes, to match the way that hash_name() iterates until it
2115 * finds the delimiter after the name.
2117 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2119 unsigned long a, x = 0, y = (unsigned long)salt;
2124 a = load_unaligned_zeropad(name);
2125 if (len < sizeof(unsigned long))
2128 name += sizeof(unsigned long);
2129 len -= sizeof(unsigned long);
2131 x ^= a & bytemask_from_count(len);
2133 return fold_hash(x, y);
2135 EXPORT_SYMBOL(full_name_hash);
2137 /* Return the "hash_len" (hash and length) of a null-terminated string */
2138 u64 hashlen_string(const void *salt, const char *name)
2140 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2141 unsigned long adata, mask, len;
2142 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2149 len += sizeof(unsigned long);
2151 a = load_unaligned_zeropad(name+len);
2152 } while (!has_zero(a, &adata, &constants));
2154 adata = prep_zero_mask(a, adata, &constants);
2155 mask = create_zero_mask(adata);
2156 x ^= a & zero_bytemask(mask);
2158 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2160 EXPORT_SYMBOL(hashlen_string);
2163 * Calculate the length and hash of the path component, and
2164 * return the "hash_len" as the result.
2166 static inline u64 hash_name(const void *salt, const char *name)
2168 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2169 unsigned long adata, bdata, mask, len;
2170 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2177 len += sizeof(unsigned long);
2179 a = load_unaligned_zeropad(name+len);
2180 b = a ^ REPEAT_BYTE('/');
2181 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2183 adata = prep_zero_mask(a, adata, &constants);
2184 bdata = prep_zero_mask(b, bdata, &constants);
2185 mask = create_zero_mask(adata | bdata);
2186 x ^= a & zero_bytemask(mask);
2188 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2191 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2193 /* Return the hash of a string of known length */
2194 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2196 unsigned long hash = init_name_hash(salt);
2198 hash = partial_name_hash((unsigned char)*name++, hash);
2199 return end_name_hash(hash);
2201 EXPORT_SYMBOL(full_name_hash);
2203 /* Return the "hash_len" (hash and length) of a null-terminated string */
2204 u64 hashlen_string(const void *salt, const char *name)
2206 unsigned long hash = init_name_hash(salt);
2207 unsigned long len = 0, c;
2209 c = (unsigned char)*name;
2212 hash = partial_name_hash(c, hash);
2213 c = (unsigned char)name[len];
2215 return hashlen_create(end_name_hash(hash), len);
2217 EXPORT_SYMBOL(hashlen_string);
2220 * We know there's a real path component here of at least
2223 static inline u64 hash_name(const void *salt, const char *name)
2225 unsigned long hash = init_name_hash(salt);
2226 unsigned long len = 0, c;
2228 c = (unsigned char)*name;
2231 hash = partial_name_hash(c, hash);
2232 c = (unsigned char)name[len];
2233 } while (c && c != '/');
2234 return hashlen_create(end_name_hash(hash), len);
2241 * This is the basic name resolution function, turning a pathname into
2242 * the final dentry. We expect 'base' to be positive and a directory.
2244 * Returns 0 and nd will have valid dentry and mnt on success.
2245 * Returns error and drops reference to input namei data on failure.
2247 static int link_path_walk(const char *name, struct nameidata *nd)
2249 int depth = 0; // depth <= nd->depth
2252 nd->last_type = LAST_ROOT;
2253 nd->flags |= LOOKUP_PARENT;
2255 return PTR_ERR(name);
2259 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2263 /* At this point we know we have a real path component. */
2265 struct mnt_idmap *idmap;
2270 idmap = mnt_idmap(nd->path.mnt);
2271 err = may_lookup(idmap, nd);
2275 hash_len = hash_name(nd->path.dentry, name);
2278 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2280 if (name[1] == '.') {
2282 nd->state |= ND_JUMPED;
2288 if (likely(type == LAST_NORM)) {
2289 struct dentry *parent = nd->path.dentry;
2290 nd->state &= ~ND_JUMPED;
2291 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2292 struct qstr this = { { .hash_len = hash_len }, .name = name };
2293 err = parent->d_op->d_hash(parent, &this);
2296 hash_len = this.hash_len;
2301 nd->last.hash_len = hash_len;
2302 nd->last.name = name;
2303 nd->last_type = type;
2305 name += hashlen_len(hash_len);
2309 * If it wasn't NUL, we know it was '/'. Skip that
2310 * slash, and continue until no more slashes.
2314 } while (unlikely(*name == '/'));
2315 if (unlikely(!*name)) {
2317 /* pathname or trailing symlink, done */
2319 nd->dir_vfsuid = i_uid_into_vfsuid(idmap, nd->inode);
2320 nd->dir_mode = nd->inode->i_mode;
2321 nd->flags &= ~LOOKUP_PARENT;
2324 /* last component of nested symlink */
2325 name = nd->stack[--depth].name;
2326 link = walk_component(nd, 0);
2328 /* not the last component */
2329 link = walk_component(nd, WALK_MORE);
2331 if (unlikely(link)) {
2333 return PTR_ERR(link);
2334 /* a symlink to follow */
2335 nd->stack[depth++].name = name;
2339 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2340 if (nd->flags & LOOKUP_RCU) {
2341 if (!try_to_unlazy(nd))
2349 /* must be paired with terminate_walk() */
2350 static const char *path_init(struct nameidata *nd, unsigned flags)
2353 const char *s = nd->name->name;
2355 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2356 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2357 return ERR_PTR(-EAGAIN);
2360 flags &= ~LOOKUP_RCU;
2361 if (flags & LOOKUP_RCU)
2364 nd->seq = nd->next_seq = 0;
2367 nd->state |= ND_JUMPED;
2369 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2370 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2373 if (nd->state & ND_ROOT_PRESET) {
2374 struct dentry *root = nd->root.dentry;
2375 struct inode *inode = root->d_inode;
2376 if (*s && unlikely(!d_can_lookup(root)))
2377 return ERR_PTR(-ENOTDIR);
2378 nd->path = nd->root;
2380 if (flags & LOOKUP_RCU) {
2381 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2382 nd->root_seq = nd->seq;
2384 path_get(&nd->path);
2389 nd->root.mnt = NULL;
2391 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2392 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2393 error = nd_jump_root(nd);
2394 if (unlikely(error))
2395 return ERR_PTR(error);
2399 /* Relative pathname -- get the starting-point it is relative to. */
2400 if (nd->dfd == AT_FDCWD) {
2401 if (flags & LOOKUP_RCU) {
2402 struct fs_struct *fs = current->fs;
2406 seq = read_seqcount_begin(&fs->seq);
2408 nd->inode = nd->path.dentry->d_inode;
2409 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2410 } while (read_seqcount_retry(&fs->seq, seq));
2412 get_fs_pwd(current->fs, &nd->path);
2413 nd->inode = nd->path.dentry->d_inode;
2416 /* Caller must check execute permissions on the starting path component */
2417 struct fd f = fdget_raw(nd->dfd);
2418 struct dentry *dentry;
2421 return ERR_PTR(-EBADF);
2423 dentry = f.file->f_path.dentry;
2425 if (*s && unlikely(!d_can_lookup(dentry))) {
2427 return ERR_PTR(-ENOTDIR);
2430 nd->path = f.file->f_path;
2431 if (flags & LOOKUP_RCU) {
2432 nd->inode = nd->path.dentry->d_inode;
2433 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2435 path_get(&nd->path);
2436 nd->inode = nd->path.dentry->d_inode;
2441 /* For scoped-lookups we need to set the root to the dirfd as well. */
2442 if (flags & LOOKUP_IS_SCOPED) {
2443 nd->root = nd->path;
2444 if (flags & LOOKUP_RCU) {
2445 nd->root_seq = nd->seq;
2447 path_get(&nd->root);
2448 nd->state |= ND_ROOT_GRABBED;
2454 static inline const char *lookup_last(struct nameidata *nd)
2456 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2457 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2459 return walk_component(nd, WALK_TRAILING);
2462 static int handle_lookup_down(struct nameidata *nd)
2464 if (!(nd->flags & LOOKUP_RCU))
2465 dget(nd->path.dentry);
2466 nd->next_seq = nd->seq;
2467 return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry));
2470 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2471 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2473 const char *s = path_init(nd, flags);
2476 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2477 err = handle_lookup_down(nd);
2478 if (unlikely(err < 0))
2482 while (!(err = link_path_walk(s, nd)) &&
2483 (s = lookup_last(nd)) != NULL)
2485 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2486 err = handle_lookup_down(nd);
2487 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2490 err = complete_walk(nd);
2492 if (!err && nd->flags & LOOKUP_DIRECTORY)
2493 if (!d_can_lookup(nd->path.dentry))
2497 nd->path.mnt = NULL;
2498 nd->path.dentry = NULL;
2504 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2505 struct path *path, struct path *root)
2508 struct nameidata nd;
2510 return PTR_ERR(name);
2511 set_nameidata(&nd, dfd, name, root);
2512 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2513 if (unlikely(retval == -ECHILD))
2514 retval = path_lookupat(&nd, flags, path);
2515 if (unlikely(retval == -ESTALE))
2516 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2518 if (likely(!retval))
2519 audit_inode(name, path->dentry,
2520 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2521 restore_nameidata();
2525 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2526 static int path_parentat(struct nameidata *nd, unsigned flags,
2527 struct path *parent)
2529 const char *s = path_init(nd, flags);
2530 int err = link_path_walk(s, nd);
2532 err = complete_walk(nd);
2535 nd->path.mnt = NULL;
2536 nd->path.dentry = NULL;
2542 /* Note: this does not consume "name" */
2543 static int __filename_parentat(int dfd, struct filename *name,
2544 unsigned int flags, struct path *parent,
2545 struct qstr *last, int *type,
2546 const struct path *root)
2549 struct nameidata nd;
2552 return PTR_ERR(name);
2553 set_nameidata(&nd, dfd, name, root);
2554 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2555 if (unlikely(retval == -ECHILD))
2556 retval = path_parentat(&nd, flags, parent);
2557 if (unlikely(retval == -ESTALE))
2558 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2559 if (likely(!retval)) {
2561 *type = nd.last_type;
2562 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2564 restore_nameidata();
2568 static int filename_parentat(int dfd, struct filename *name,
2569 unsigned int flags, struct path *parent,
2570 struct qstr *last, int *type)
2572 return __filename_parentat(dfd, name, flags, parent, last, type, NULL);
2575 /* does lookup, returns the object with parent locked */
2576 static struct dentry *__kern_path_locked(int dfd, struct filename *name, struct path *path)
2582 error = filename_parentat(dfd, name, 0, path, &last, &type);
2584 return ERR_PTR(error);
2585 if (unlikely(type != LAST_NORM)) {
2587 return ERR_PTR(-EINVAL);
2589 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2590 d = lookup_one_qstr_excl(&last, path->dentry, 0);
2592 inode_unlock(path->dentry->d_inode);
2598 struct dentry *kern_path_locked(const char *name, struct path *path)
2600 struct filename *filename = getname_kernel(name);
2601 struct dentry *res = __kern_path_locked(AT_FDCWD, filename, path);
2607 struct dentry *user_path_locked_at(int dfd, const char __user *name, struct path *path)
2609 struct filename *filename = getname(name);
2610 struct dentry *res = __kern_path_locked(dfd, filename, path);
2615 EXPORT_SYMBOL(user_path_locked_at);
2617 int kern_path(const char *name, unsigned int flags, struct path *path)
2619 struct filename *filename = getname_kernel(name);
2620 int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2626 EXPORT_SYMBOL(kern_path);
2629 * vfs_path_parent_lookup - lookup a parent path relative to a dentry-vfsmount pair
2630 * @filename: filename structure
2631 * @flags: lookup flags
2632 * @parent: pointer to struct path to fill
2633 * @last: last component
2634 * @type: type of the last component
2635 * @root: pointer to struct path of the base directory
2637 int vfs_path_parent_lookup(struct filename *filename, unsigned int flags,
2638 struct path *parent, struct qstr *last, int *type,
2639 const struct path *root)
2641 return __filename_parentat(AT_FDCWD, filename, flags, parent, last,
2644 EXPORT_SYMBOL(vfs_path_parent_lookup);
2647 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2648 * @dentry: pointer to dentry of the base directory
2649 * @mnt: pointer to vfs mount of the base directory
2650 * @name: pointer to file name
2651 * @flags: lookup flags
2652 * @path: pointer to struct path to fill
2654 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2655 const char *name, unsigned int flags,
2658 struct filename *filename;
2659 struct path root = {.mnt = mnt, .dentry = dentry};
2662 filename = getname_kernel(name);
2663 /* the first argument of filename_lookup() is ignored with root */
2664 ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2668 EXPORT_SYMBOL(vfs_path_lookup);
2670 static int lookup_one_common(struct mnt_idmap *idmap,
2671 const char *name, struct dentry *base, int len,
2676 this->hash = full_name_hash(base, name, len);
2680 if (unlikely(name[0] == '.')) {
2681 if (len < 2 || (len == 2 && name[1] == '.'))
2686 unsigned int c = *(const unsigned char *)name++;
2687 if (c == '/' || c == '\0')
2691 * See if the low-level filesystem might want
2692 * to use its own hash..
2694 if (base->d_flags & DCACHE_OP_HASH) {
2695 int err = base->d_op->d_hash(base, this);
2700 return inode_permission(idmap, base->d_inode, MAY_EXEC);
2704 * try_lookup_one_len - filesystem helper to lookup single pathname component
2705 * @name: pathname component to lookup
2706 * @base: base directory to lookup from
2707 * @len: maximum length @len should be interpreted to
2709 * Look up a dentry by name in the dcache, returning NULL if it does not
2710 * currently exist. The function does not try to create a dentry.
2712 * Note that this routine is purely a helper for filesystem usage and should
2713 * not be called by generic code.
2715 * The caller must hold base->i_mutex.
2717 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2722 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2724 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2726 return ERR_PTR(err);
2728 return lookup_dcache(&this, base, 0);
2730 EXPORT_SYMBOL(try_lookup_one_len);
2733 * lookup_one_len - filesystem helper to lookup single pathname component
2734 * @name: pathname component to lookup
2735 * @base: base directory to lookup from
2736 * @len: maximum length @len should be interpreted to
2738 * Note that this routine is purely a helper for filesystem usage and should
2739 * not be called by generic code.
2741 * The caller must hold base->i_mutex.
2743 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2745 struct dentry *dentry;
2749 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2751 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2753 return ERR_PTR(err);
2755 dentry = lookup_dcache(&this, base, 0);
2756 return dentry ? dentry : __lookup_slow(&this, base, 0);
2758 EXPORT_SYMBOL(lookup_one_len);
2761 * lookup_one - filesystem helper to lookup single pathname component
2762 * @idmap: idmap of the mount the lookup is performed from
2763 * @name: pathname component to lookup
2764 * @base: base directory to lookup from
2765 * @len: maximum length @len should be interpreted to
2767 * Note that this routine is purely a helper for filesystem usage and should
2768 * not be called by generic code.
2770 * The caller must hold base->i_mutex.
2772 struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name,
2773 struct dentry *base, int len)
2775 struct dentry *dentry;
2779 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2781 err = lookup_one_common(idmap, name, base, len, &this);
2783 return ERR_PTR(err);
2785 dentry = lookup_dcache(&this, base, 0);
2786 return dentry ? dentry : __lookup_slow(&this, base, 0);
2788 EXPORT_SYMBOL(lookup_one);
2791 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2792 * @idmap: idmap of the mount the lookup is performed from
2793 * @name: pathname component to lookup
2794 * @base: base directory to lookup from
2795 * @len: maximum length @len should be interpreted to
2797 * Note that this routine is purely a helper for filesystem usage and should
2798 * not be called by generic code.
2800 * Unlike lookup_one_len, it should be called without the parent
2801 * i_mutex held, and will take the i_mutex itself if necessary.
2803 struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap,
2804 const char *name, struct dentry *base,
2811 err = lookup_one_common(idmap, name, base, len, &this);
2813 return ERR_PTR(err);
2815 ret = lookup_dcache(&this, base, 0);
2817 ret = lookup_slow(&this, base, 0);
2820 EXPORT_SYMBOL(lookup_one_unlocked);
2823 * lookup_one_positive_unlocked - filesystem helper to lookup single
2824 * pathname component
2825 * @idmap: idmap of the mount the lookup is performed from
2826 * @name: pathname component to lookup
2827 * @base: base directory to lookup from
2828 * @len: maximum length @len should be interpreted to
2830 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2831 * known positive or ERR_PTR(). This is what most of the users want.
2833 * Note that pinned negative with unlocked parent _can_ become positive at any
2834 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2835 * positives have >d_inode stable, so this one avoids such problems.
2837 * Note that this routine is purely a helper for filesystem usage and should
2838 * not be called by generic code.
2840 * The helper should be called without i_mutex held.
2842 struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap,
2844 struct dentry *base, int len)
2846 struct dentry *ret = lookup_one_unlocked(idmap, name, base, len);
2848 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2850 ret = ERR_PTR(-ENOENT);
2854 EXPORT_SYMBOL(lookup_one_positive_unlocked);
2857 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2858 * @name: pathname component to lookup
2859 * @base: base directory to lookup from
2860 * @len: maximum length @len should be interpreted to
2862 * Note that this routine is purely a helper for filesystem usage and should
2863 * not be called by generic code.
2865 * Unlike lookup_one_len, it should be called without the parent
2866 * i_mutex held, and will take the i_mutex itself if necessary.
2868 struct dentry *lookup_one_len_unlocked(const char *name,
2869 struct dentry *base, int len)
2871 return lookup_one_unlocked(&nop_mnt_idmap, name, base, len);
2873 EXPORT_SYMBOL(lookup_one_len_unlocked);
2876 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2877 * on negatives. Returns known positive or ERR_PTR(); that's what
2878 * most of the users want. Note that pinned negative with unlocked parent
2879 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2880 * need to be very careful; pinned positives have ->d_inode stable, so
2881 * this one avoids such problems.
2883 struct dentry *lookup_positive_unlocked(const char *name,
2884 struct dentry *base, int len)
2886 return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len);
2888 EXPORT_SYMBOL(lookup_positive_unlocked);
2890 #ifdef CONFIG_UNIX98_PTYS
2891 int path_pts(struct path *path)
2893 /* Find something mounted on "pts" in the same directory as
2896 struct dentry *parent = dget_parent(path->dentry);
2897 struct dentry *child;
2898 struct qstr this = QSTR_INIT("pts", 3);
2900 if (unlikely(!path_connected(path->mnt, parent))) {
2905 path->dentry = parent;
2906 child = d_hash_and_lookup(parent, &this);
2907 if (IS_ERR_OR_NULL(child))
2910 path->dentry = child;
2912 follow_down(path, 0);
2917 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2918 struct path *path, int *empty)
2920 struct filename *filename = getname_flags(name, flags, empty);
2921 int ret = filename_lookup(dfd, filename, flags, path, NULL);
2926 EXPORT_SYMBOL(user_path_at_empty);
2928 int __check_sticky(struct mnt_idmap *idmap, struct inode *dir,
2929 struct inode *inode)
2931 kuid_t fsuid = current_fsuid();
2933 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), fsuid))
2935 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, dir), fsuid))
2937 return !capable_wrt_inode_uidgid(idmap, inode, CAP_FOWNER);
2939 EXPORT_SYMBOL(__check_sticky);
2942 * Check whether we can remove a link victim from directory dir, check
2943 * whether the type of victim is right.
2944 * 1. We can't do it if dir is read-only (done in permission())
2945 * 2. We should have write and exec permissions on dir
2946 * 3. We can't remove anything from append-only dir
2947 * 4. We can't do anything with immutable dir (done in permission())
2948 * 5. If the sticky bit on dir is set we should either
2949 * a. be owner of dir, or
2950 * b. be owner of victim, or
2951 * c. have CAP_FOWNER capability
2952 * 6. If the victim is append-only or immutable we can't do antyhing with
2953 * links pointing to it.
2954 * 7. If the victim has an unknown uid or gid we can't change the inode.
2955 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2956 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2957 * 10. We can't remove a root or mountpoint.
2958 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2959 * nfs_async_unlink().
2961 static int may_delete(struct mnt_idmap *idmap, struct inode *dir,
2962 struct dentry *victim, bool isdir)
2964 struct inode *inode = d_backing_inode(victim);
2967 if (d_is_negative(victim))
2971 BUG_ON(victim->d_parent->d_inode != dir);
2973 /* Inode writeback is not safe when the uid or gid are invalid. */
2974 if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
2975 !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
2978 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2980 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
2986 if (check_sticky(idmap, dir, inode) || IS_APPEND(inode) ||
2987 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2988 HAS_UNMAPPED_ID(idmap, inode))
2991 if (!d_is_dir(victim))
2993 if (IS_ROOT(victim))
2995 } else if (d_is_dir(victim))
2997 if (IS_DEADDIR(dir))
2999 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
3004 /* Check whether we can create an object with dentry child in directory
3006 * 1. We can't do it if child already exists (open has special treatment for
3007 * this case, but since we are inlined it's OK)
3008 * 2. We can't do it if dir is read-only (done in permission())
3009 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
3010 * 4. We should have write and exec permissions on dir
3011 * 5. We can't do it if dir is immutable (done in permission())
3013 static inline int may_create(struct mnt_idmap *idmap,
3014 struct inode *dir, struct dentry *child)
3016 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
3019 if (IS_DEADDIR(dir))
3021 if (!fsuidgid_has_mapping(dir->i_sb, idmap))
3024 return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3027 static struct dentry *lock_two_directories(struct dentry *p1, struct dentry *p2)
3031 p = d_ancestor(p2, p1);
3033 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3034 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT2);
3038 p = d_ancestor(p1, p2);
3039 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3040 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3045 * p1 and p2 should be directories on the same fs.
3047 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
3050 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3054 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
3055 return lock_two_directories(p1, p2);
3057 EXPORT_SYMBOL(lock_rename);
3060 * c1 and p2 should be on the same fs.
3062 struct dentry *lock_rename_child(struct dentry *c1, struct dentry *p2)
3064 if (READ_ONCE(c1->d_parent) == p2) {
3066 * hopefully won't need to touch ->s_vfs_rename_mutex at all.
3068 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3070 * now that p2 is locked, nobody can move in or out of it,
3071 * so the test below is safe.
3073 if (likely(c1->d_parent == p2))
3077 * c1 got moved out of p2 while we'd been taking locks;
3078 * unlock and fall back to slow case.
3080 inode_unlock(p2->d_inode);
3083 mutex_lock(&c1->d_sb->s_vfs_rename_mutex);
3085 * nobody can move out of any directories on this fs.
3087 if (likely(c1->d_parent != p2))
3088 return lock_two_directories(c1->d_parent, p2);
3091 * c1 got moved into p2 while we were taking locks;
3092 * we need p2 locked and ->s_vfs_rename_mutex unlocked,
3093 * for consistency with lock_rename().
3095 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3096 mutex_unlock(&c1->d_sb->s_vfs_rename_mutex);
3099 EXPORT_SYMBOL(lock_rename_child);
3101 void unlock_rename(struct dentry *p1, struct dentry *p2)
3103 inode_unlock(p1->d_inode);
3105 inode_unlock(p2->d_inode);
3106 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3109 EXPORT_SYMBOL(unlock_rename);
3112 * vfs_prepare_mode - prepare the mode to be used for a new inode
3113 * @idmap: idmap of the mount the inode was found from
3114 * @dir: parent directory of the new inode
3115 * @mode: mode of the new inode
3116 * @mask_perms: allowed permission by the vfs
3117 * @type: type of file to be created
3119 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3120 * object to be created.
3122 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3123 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3124 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3125 * POSIX ACL supporting filesystems.
3127 * Note that it's currently valid for @type to be 0 if a directory is created.
3128 * Filesystems raise that flag individually and we need to check whether each
3129 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3132 * Returns: mode to be passed to the filesystem
3134 static inline umode_t vfs_prepare_mode(struct mnt_idmap *idmap,
3135 const struct inode *dir, umode_t mode,
3136 umode_t mask_perms, umode_t type)
3138 mode = mode_strip_sgid(idmap, dir, mode);
3139 mode = mode_strip_umask(dir, mode);
3142 * Apply the vfs mandated allowed permission mask and set the type of
3143 * file to be created before we call into the filesystem.
3145 mode &= (mask_perms & ~S_IFMT);
3146 mode |= (type & S_IFMT);
3152 * vfs_create - create new file
3153 * @idmap: idmap of the mount the inode was found from
3154 * @dir: inode of @dentry
3155 * @dentry: pointer to dentry of the base directory
3156 * @mode: mode of the new file
3157 * @want_excl: whether the file must not yet exist
3159 * Create a new file.
3161 * If the inode has been found through an idmapped mount the idmap of
3162 * the vfsmount must be passed through @idmap. This function will then take
3163 * care to map the inode according to @idmap before checking permissions.
3164 * On non-idmapped mounts or if permission checking is to be performed on the
3165 * raw inode simply passs @nop_mnt_idmap.
3167 int vfs_create(struct mnt_idmap *idmap, struct inode *dir,
3168 struct dentry *dentry, umode_t mode, bool want_excl)
3172 error = may_create(idmap, dir, dentry);
3176 if (!dir->i_op->create)
3177 return -EACCES; /* shouldn't it be ENOSYS? */
3179 mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG);
3180 error = security_inode_create(dir, dentry, mode);
3183 error = dir->i_op->create(idmap, dir, dentry, mode, want_excl);
3185 fsnotify_create(dir, dentry);
3188 EXPORT_SYMBOL(vfs_create);
3190 int vfs_mkobj(struct dentry *dentry, umode_t mode,
3191 int (*f)(struct dentry *, umode_t, void *),
3194 struct inode *dir = dentry->d_parent->d_inode;
3195 int error = may_create(&nop_mnt_idmap, dir, dentry);
3201 error = security_inode_create(dir, dentry, mode);
3204 error = f(dentry, mode, arg);
3206 fsnotify_create(dir, dentry);
3209 EXPORT_SYMBOL(vfs_mkobj);
3211 bool may_open_dev(const struct path *path)
3213 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3214 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3217 static int may_open(struct mnt_idmap *idmap, const struct path *path,
3218 int acc_mode, int flag)
3220 struct dentry *dentry = path->dentry;
3221 struct inode *inode = dentry->d_inode;
3227 switch (inode->i_mode & S_IFMT) {
3231 if (acc_mode & MAY_WRITE)
3233 if (acc_mode & MAY_EXEC)
3238 if (!may_open_dev(path))
3243 if (acc_mode & MAY_EXEC)
3248 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3253 error = inode_permission(idmap, inode, MAY_OPEN | acc_mode);
3258 * An append-only file must be opened in append mode for writing.
3260 if (IS_APPEND(inode)) {
3261 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3267 /* O_NOATIME can only be set by the owner or superuser */
3268 if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode))
3274 static int handle_truncate(struct mnt_idmap *idmap, struct file *filp)
3276 const struct path *path = &filp->f_path;
3277 struct inode *inode = path->dentry->d_inode;
3278 int error = get_write_access(inode);
3282 error = security_file_truncate(filp);
3284 error = do_truncate(idmap, path->dentry, 0,
3285 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3288 put_write_access(inode);
3292 static inline int open_to_namei_flags(int flag)
3294 if ((flag & O_ACCMODE) == 3)
3299 static int may_o_create(struct mnt_idmap *idmap,
3300 const struct path *dir, struct dentry *dentry,
3303 int error = security_path_mknod(dir, dentry, mode, 0);
3307 if (!fsuidgid_has_mapping(dir->dentry->d_sb, idmap))
3310 error = inode_permission(idmap, dir->dentry->d_inode,
3311 MAY_WRITE | MAY_EXEC);
3315 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3319 * Attempt to atomically look up, create and open a file from a negative
3322 * Returns 0 if successful. The file will have been created and attached to
3323 * @file by the filesystem calling finish_open().
3325 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3326 * be set. The caller will need to perform the open themselves. @path will
3327 * have been updated to point to the new dentry. This may be negative.
3329 * Returns an error code otherwise.
3331 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3333 int open_flag, umode_t mode)
3335 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3336 struct inode *dir = nd->path.dentry->d_inode;
3339 if (nd->flags & LOOKUP_DIRECTORY)
3340 open_flag |= O_DIRECTORY;
3342 file->f_path.dentry = DENTRY_NOT_SET;
3343 file->f_path.mnt = nd->path.mnt;
3344 error = dir->i_op->atomic_open(dir, dentry, file,
3345 open_to_namei_flags(open_flag), mode);
3346 d_lookup_done(dentry);
3348 if (file->f_mode & FMODE_OPENED) {
3349 if (unlikely(dentry != file->f_path.dentry)) {
3351 dentry = dget(file->f_path.dentry);
3353 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3356 if (file->f_path.dentry) {
3358 dentry = file->f_path.dentry;
3360 if (unlikely(d_is_negative(dentry)))
3366 dentry = ERR_PTR(error);
3372 * Look up and maybe create and open the last component.
3374 * Must be called with parent locked (exclusive in O_CREAT case).
3376 * Returns 0 on success, that is, if
3377 * the file was successfully atomically created (if necessary) and opened, or
3378 * the file was not completely opened at this time, though lookups and
3379 * creations were performed.
3380 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3381 * In the latter case dentry returned in @path might be negative if O_CREAT
3382 * hadn't been specified.
3384 * An error code is returned on failure.
3386 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3387 const struct open_flags *op,
3390 struct mnt_idmap *idmap;
3391 struct dentry *dir = nd->path.dentry;
3392 struct inode *dir_inode = dir->d_inode;
3393 int open_flag = op->open_flag;
3394 struct dentry *dentry;
3395 int error, create_error = 0;
3396 umode_t mode = op->mode;
3397 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3399 if (unlikely(IS_DEADDIR(dir_inode)))
3400 return ERR_PTR(-ENOENT);
3402 file->f_mode &= ~FMODE_CREATED;
3403 dentry = d_lookup(dir, &nd->last);
3406 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3410 if (d_in_lookup(dentry))
3413 error = d_revalidate(dentry, nd->flags);
3414 if (likely(error > 0))
3418 d_invalidate(dentry);
3422 if (dentry->d_inode) {
3423 /* Cached positive dentry: will open in f_op->open */
3428 * Checking write permission is tricky, bacuse we don't know if we are
3429 * going to actually need it: O_CREAT opens should work as long as the
3430 * file exists. But checking existence breaks atomicity. The trick is
3431 * to check access and if not granted clear O_CREAT from the flags.
3433 * Another problem is returing the "right" error value (e.g. for an
3434 * O_EXCL open we want to return EEXIST not EROFS).
3436 if (unlikely(!got_write))
3437 open_flag &= ~O_TRUNC;
3438 idmap = mnt_idmap(nd->path.mnt);
3439 if (open_flag & O_CREAT) {
3440 if (open_flag & O_EXCL)
3441 open_flag &= ~O_TRUNC;
3442 mode = vfs_prepare_mode(idmap, dir->d_inode, mode, mode, mode);
3443 if (likely(got_write))
3444 create_error = may_o_create(idmap, &nd->path,
3447 create_error = -EROFS;
3450 open_flag &= ~O_CREAT;
3451 if (dir_inode->i_op->atomic_open) {
3452 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3453 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3454 dentry = ERR_PTR(create_error);
3458 if (d_in_lookup(dentry)) {
3459 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3461 d_lookup_done(dentry);
3462 if (unlikely(res)) {
3464 error = PTR_ERR(res);
3472 /* Negative dentry, just create the file */
3473 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3474 file->f_mode |= FMODE_CREATED;
3475 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3476 if (!dir_inode->i_op->create) {
3481 error = dir_inode->i_op->create(idmap, dir_inode, dentry,
3482 mode, open_flag & O_EXCL);
3486 if (unlikely(create_error) && !dentry->d_inode) {
3487 error = create_error;
3494 return ERR_PTR(error);
3497 static const char *open_last_lookups(struct nameidata *nd,
3498 struct file *file, const struct open_flags *op)
3500 struct dentry *dir = nd->path.dentry;
3501 int open_flag = op->open_flag;
3502 bool got_write = false;
3503 struct dentry *dentry;
3506 nd->flags |= op->intent;
3508 if (nd->last_type != LAST_NORM) {
3511 return handle_dots(nd, nd->last_type);
3514 if (!(open_flag & O_CREAT)) {
3515 if (nd->last.name[nd->last.len])
3516 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3517 /* we _can_ be in RCU mode here */
3518 dentry = lookup_fast(nd);
3520 return ERR_CAST(dentry);
3524 if (WARN_ON_ONCE(nd->flags & LOOKUP_RCU))
3525 return ERR_PTR(-ECHILD);
3527 /* create side of things */
3528 if (nd->flags & LOOKUP_RCU) {
3529 if (!try_to_unlazy(nd))
3530 return ERR_PTR(-ECHILD);
3532 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3533 /* trailing slashes? */
3534 if (unlikely(nd->last.name[nd->last.len]))
3535 return ERR_PTR(-EISDIR);
3538 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3539 got_write = !mnt_want_write(nd->path.mnt);
3541 * do _not_ fail yet - we might not need that or fail with
3542 * a different error; let lookup_open() decide; we'll be
3543 * dropping this one anyway.
3546 if (open_flag & O_CREAT)
3547 inode_lock(dir->d_inode);
3549 inode_lock_shared(dir->d_inode);
3550 dentry = lookup_open(nd, file, op, got_write);
3551 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3552 fsnotify_create(dir->d_inode, dentry);
3553 if (open_flag & O_CREAT)
3554 inode_unlock(dir->d_inode);
3556 inode_unlock_shared(dir->d_inode);
3559 mnt_drop_write(nd->path.mnt);
3562 return ERR_CAST(dentry);
3564 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3565 dput(nd->path.dentry);
3566 nd->path.dentry = dentry;
3573 res = step_into(nd, WALK_TRAILING, dentry);
3575 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3580 * Handle the last step of open()
3582 static int do_open(struct nameidata *nd,
3583 struct file *file, const struct open_flags *op)
3585 struct mnt_idmap *idmap;
3586 int open_flag = op->open_flag;
3591 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3592 error = complete_walk(nd);
3596 if (!(file->f_mode & FMODE_CREATED))
3597 audit_inode(nd->name, nd->path.dentry, 0);
3598 idmap = mnt_idmap(nd->path.mnt);
3599 if (open_flag & O_CREAT) {
3600 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3602 if (d_is_dir(nd->path.dentry))
3604 error = may_create_in_sticky(idmap, nd,
3605 d_backing_inode(nd->path.dentry));
3606 if (unlikely(error))
3609 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3612 do_truncate = false;
3613 acc_mode = op->acc_mode;
3614 if (file->f_mode & FMODE_CREATED) {
3615 /* Don't check for write permission, don't truncate */
3616 open_flag &= ~O_TRUNC;
3618 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3619 error = mnt_want_write(nd->path.mnt);
3624 error = may_open(idmap, &nd->path, acc_mode, open_flag);
3625 if (!error && !(file->f_mode & FMODE_OPENED))
3626 error = vfs_open(&nd->path, file);
3628 error = ima_file_check(file, op->acc_mode);
3629 if (!error && do_truncate)
3630 error = handle_truncate(idmap, file);
3631 if (unlikely(error > 0)) {
3636 mnt_drop_write(nd->path.mnt);
3641 * vfs_tmpfile - create tmpfile
3642 * @idmap: idmap of the mount the inode was found from
3643 * @parentpath: pointer to the path of the base directory
3644 * @file: file descriptor of the new tmpfile
3645 * @mode: mode of the new tmpfile
3647 * Create a temporary file.
3649 * If the inode has been found through an idmapped mount the idmap of
3650 * the vfsmount must be passed through @idmap. This function will then take
3651 * care to map the inode according to @idmap before checking permissions.
3652 * On non-idmapped mounts or if permission checking is to be performed on the
3653 * raw inode simply passs @nop_mnt_idmap.
3655 static int vfs_tmpfile(struct mnt_idmap *idmap,
3656 const struct path *parentpath,
3657 struct file *file, umode_t mode)
3659 struct dentry *child;
3660 struct inode *dir = d_inode(parentpath->dentry);
3661 struct inode *inode;
3663 int open_flag = file->f_flags;
3665 /* we want directory to be writable */
3666 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3669 if (!dir->i_op->tmpfile)
3671 child = d_alloc(parentpath->dentry, &slash_name);
3672 if (unlikely(!child))
3674 file->f_path.mnt = parentpath->mnt;
3675 file->f_path.dentry = child;
3676 mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3677 error = dir->i_op->tmpfile(idmap, dir, file, mode);
3681 /* Don't check for other permissions, the inode was just created */
3682 error = may_open(idmap, &file->f_path, 0, file->f_flags);
3685 inode = file_inode(file);
3686 if (!(open_flag & O_EXCL)) {
3687 spin_lock(&inode->i_lock);
3688 inode->i_state |= I_LINKABLE;
3689 spin_unlock(&inode->i_lock);
3691 ima_post_create_tmpfile(idmap, inode);
3696 * kernel_tmpfile_open - open a tmpfile for kernel internal use
3697 * @idmap: idmap of the mount the inode was found from
3698 * @parentpath: path of the base directory
3699 * @mode: mode of the new tmpfile
3701 * @cred: credentials for open
3703 * Create and open a temporary file. The file is not accounted in nr_files,
3704 * hence this is only for kernel internal use, and must not be installed into
3705 * file tables or such.
3707 struct file *kernel_tmpfile_open(struct mnt_idmap *idmap,
3708 const struct path *parentpath,
3709 umode_t mode, int open_flag,
3710 const struct cred *cred)
3715 file = alloc_empty_file_noaccount(open_flag, cred);
3719 error = vfs_tmpfile(idmap, parentpath, file, mode);
3722 file = ERR_PTR(error);
3726 EXPORT_SYMBOL(kernel_tmpfile_open);
3728 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3729 const struct open_flags *op,
3733 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3735 if (unlikely(error))
3737 error = mnt_want_write(path.mnt);
3738 if (unlikely(error))
3740 error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode);
3743 audit_inode(nd->name, file->f_path.dentry, 0);
3745 mnt_drop_write(path.mnt);
3751 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3754 int error = path_lookupat(nd, flags, &path);
3756 audit_inode(nd->name, path.dentry, 0);
3757 error = vfs_open(&path, file);
3763 static struct file *path_openat(struct nameidata *nd,
3764 const struct open_flags *op, unsigned flags)
3769 file = alloc_empty_file(op->open_flag, current_cred());
3773 if (unlikely(file->f_flags & __O_TMPFILE)) {
3774 error = do_tmpfile(nd, flags, op, file);
3775 } else if (unlikely(file->f_flags & O_PATH)) {
3776 error = do_o_path(nd, flags, file);
3778 const char *s = path_init(nd, flags);
3779 while (!(error = link_path_walk(s, nd)) &&
3780 (s = open_last_lookups(nd, file, op)) != NULL)
3783 error = do_open(nd, file, op);
3786 if (likely(!error)) {
3787 if (likely(file->f_mode & FMODE_OPENED))
3792 if (unlikely(file->f_mode & FMODE_OPENED))
3795 release_empty_file(file);
3796 if (error == -EOPENSTALE) {
3797 if (flags & LOOKUP_RCU)
3802 return ERR_PTR(error);
3805 struct file *do_filp_open(int dfd, struct filename *pathname,
3806 const struct open_flags *op)
3808 struct nameidata nd;
3809 int flags = op->lookup_flags;
3812 set_nameidata(&nd, dfd, pathname, NULL);
3813 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3814 if (unlikely(filp == ERR_PTR(-ECHILD)))
3815 filp = path_openat(&nd, op, flags);
3816 if (unlikely(filp == ERR_PTR(-ESTALE)))
3817 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3818 restore_nameidata();
3822 struct file *do_file_open_root(const struct path *root,
3823 const char *name, const struct open_flags *op)
3825 struct nameidata nd;
3827 struct filename *filename;
3828 int flags = op->lookup_flags;
3830 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3831 return ERR_PTR(-ELOOP);
3833 filename = getname_kernel(name);
3834 if (IS_ERR(filename))
3835 return ERR_CAST(filename);
3837 set_nameidata(&nd, -1, filename, root);
3838 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3839 if (unlikely(file == ERR_PTR(-ECHILD)))
3840 file = path_openat(&nd, op, flags);
3841 if (unlikely(file == ERR_PTR(-ESTALE)))
3842 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3843 restore_nameidata();
3848 static struct dentry *filename_create(int dfd, struct filename *name,
3849 struct path *path, unsigned int lookup_flags)
3851 struct dentry *dentry = ERR_PTR(-EEXIST);
3853 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3854 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3855 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3860 error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3862 return ERR_PTR(error);
3865 * Yucky last component or no last component at all?
3866 * (foo/., foo/.., /////)
3868 if (unlikely(type != LAST_NORM))
3871 /* don't fail immediately if it's r/o, at least try to report other errors */
3872 err2 = mnt_want_write(path->mnt);
3874 * Do the final lookup. Suppress 'create' if there is a trailing
3875 * '/', and a directory wasn't requested.
3877 if (last.name[last.len] && !want_dir)
3879 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3880 dentry = lookup_one_qstr_excl(&last, path->dentry,
3881 reval_flag | create_flags);
3886 if (d_is_positive(dentry))
3890 * Special case - lookup gave negative, but... we had foo/bar/
3891 * From the vfs_mknod() POV we just have a negative dentry -
3892 * all is fine. Let's be bastards - you had / on the end, you've
3893 * been asking for (non-existent) directory. -ENOENT for you.
3895 if (unlikely(!create_flags)) {
3899 if (unlikely(err2)) {
3906 dentry = ERR_PTR(error);
3908 inode_unlock(path->dentry->d_inode);
3910 mnt_drop_write(path->mnt);
3916 struct dentry *kern_path_create(int dfd, const char *pathname,
3917 struct path *path, unsigned int lookup_flags)
3919 struct filename *filename = getname_kernel(pathname);
3920 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3925 EXPORT_SYMBOL(kern_path_create);
3927 void done_path_create(struct path *path, struct dentry *dentry)
3930 inode_unlock(path->dentry->d_inode);
3931 mnt_drop_write(path->mnt);
3934 EXPORT_SYMBOL(done_path_create);
3936 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3937 struct path *path, unsigned int lookup_flags)
3939 struct filename *filename = getname(pathname);
3940 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3945 EXPORT_SYMBOL(user_path_create);
3948 * vfs_mknod - create device node or file
3949 * @idmap: idmap of the mount the inode was found from
3950 * @dir: inode of @dentry
3951 * @dentry: pointer to dentry of the base directory
3952 * @mode: mode of the new device node or file
3953 * @dev: device number of device to create
3955 * Create a device node or file.
3957 * If the inode has been found through an idmapped mount the idmap of
3958 * the vfsmount must be passed through @idmap. This function will then take
3959 * care to map the inode according to @idmap before checking permissions.
3960 * On non-idmapped mounts or if permission checking is to be performed on the
3961 * raw inode simply passs @nop_mnt_idmap.
3963 int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
3964 struct dentry *dentry, umode_t mode, dev_t dev)
3966 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3967 int error = may_create(idmap, dir, dentry);
3972 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3973 !capable(CAP_MKNOD))
3976 if (!dir->i_op->mknod)
3979 mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3980 error = devcgroup_inode_mknod(mode, dev);
3984 error = security_inode_mknod(dir, dentry, mode, dev);
3988 error = dir->i_op->mknod(idmap, dir, dentry, mode, dev);
3990 fsnotify_create(dir, dentry);
3993 EXPORT_SYMBOL(vfs_mknod);
3995 static int may_mknod(umode_t mode)
3997 switch (mode & S_IFMT) {
4003 case 0: /* zero mode translates to S_IFREG */
4012 static int do_mknodat(int dfd, struct filename *name, umode_t mode,
4015 struct mnt_idmap *idmap;
4016 struct dentry *dentry;
4019 unsigned int lookup_flags = 0;
4021 error = may_mknod(mode);
4025 dentry = filename_create(dfd, name, &path, lookup_flags);
4026 error = PTR_ERR(dentry);
4030 error = security_path_mknod(&path, dentry,
4031 mode_strip_umask(path.dentry->d_inode, mode), dev);
4035 idmap = mnt_idmap(path.mnt);
4036 switch (mode & S_IFMT) {
4037 case 0: case S_IFREG:
4038 error = vfs_create(idmap, path.dentry->d_inode,
4039 dentry, mode, true);
4041 ima_post_path_mknod(idmap, dentry);
4043 case S_IFCHR: case S_IFBLK:
4044 error = vfs_mknod(idmap, path.dentry->d_inode,
4045 dentry, mode, new_decode_dev(dev));
4047 case S_IFIFO: case S_IFSOCK:
4048 error = vfs_mknod(idmap, path.dentry->d_inode,
4053 done_path_create(&path, dentry);
4054 if (retry_estale(error, lookup_flags)) {
4055 lookup_flags |= LOOKUP_REVAL;
4063 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
4066 return do_mknodat(dfd, getname(filename), mode, dev);
4069 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
4071 return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
4075 * vfs_mkdir - create directory
4076 * @idmap: idmap of the mount the inode was found from
4077 * @dir: inode of @dentry
4078 * @dentry: pointer to dentry of the base directory
4079 * @mode: mode of the new directory
4081 * Create a directory.
4083 * If the inode has been found through an idmapped mount the idmap of
4084 * the vfsmount must be passed through @idmap. This function will then take
4085 * care to map the inode according to @idmap before checking permissions.
4086 * On non-idmapped mounts or if permission checking is to be performed on the
4087 * raw inode simply passs @nop_mnt_idmap.
4089 int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
4090 struct dentry *dentry, umode_t mode)
4093 unsigned max_links = dir->i_sb->s_max_links;
4095 error = may_create(idmap, dir, dentry);
4099 if (!dir->i_op->mkdir)
4102 mode = vfs_prepare_mode(idmap, dir, mode, S_IRWXUGO | S_ISVTX, 0);
4103 error = security_inode_mkdir(dir, dentry, mode);
4107 if (max_links && dir->i_nlink >= max_links)
4110 error = dir->i_op->mkdir(idmap, dir, dentry, mode);
4112 fsnotify_mkdir(dir, dentry);
4115 EXPORT_SYMBOL(vfs_mkdir);
4117 int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4119 struct dentry *dentry;
4122 unsigned int lookup_flags = LOOKUP_DIRECTORY;
4125 dentry = filename_create(dfd, name, &path, lookup_flags);
4126 error = PTR_ERR(dentry);
4130 error = security_path_mkdir(&path, dentry,
4131 mode_strip_umask(path.dentry->d_inode, mode));
4133 error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode,
4136 done_path_create(&path, dentry);
4137 if (retry_estale(error, lookup_flags)) {
4138 lookup_flags |= LOOKUP_REVAL;
4146 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4148 return do_mkdirat(dfd, getname(pathname), mode);
4151 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4153 return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4157 * vfs_rmdir - remove directory
4158 * @idmap: idmap of the mount the inode was found from
4159 * @dir: inode of @dentry
4160 * @dentry: pointer to dentry of the base directory
4162 * Remove a directory.
4164 * If the inode has been found through an idmapped mount the idmap of
4165 * the vfsmount must be passed through @idmap. This function will then take
4166 * care to map the inode according to @idmap before checking permissions.
4167 * On non-idmapped mounts or if permission checking is to be performed on the
4168 * raw inode simply passs @nop_mnt_idmap.
4170 int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir,
4171 struct dentry *dentry)
4173 int error = may_delete(idmap, dir, dentry, 1);
4178 if (!dir->i_op->rmdir)
4182 inode_lock(dentry->d_inode);
4185 if (is_local_mountpoint(dentry) ||
4186 (dentry->d_inode->i_flags & S_KERNEL_FILE))
4189 error = security_inode_rmdir(dir, dentry);
4193 error = dir->i_op->rmdir(dir, dentry);
4197 shrink_dcache_parent(dentry);
4198 dentry->d_inode->i_flags |= S_DEAD;
4200 detach_mounts(dentry);
4203 inode_unlock(dentry->d_inode);
4206 d_delete_notify(dir, dentry);
4209 EXPORT_SYMBOL(vfs_rmdir);
4211 int do_rmdir(int dfd, struct filename *name)
4214 struct dentry *dentry;
4218 unsigned int lookup_flags = 0;
4220 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4236 error = mnt_want_write(path.mnt);
4240 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4241 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4242 error = PTR_ERR(dentry);
4245 if (!dentry->d_inode) {
4249 error = security_path_rmdir(&path, dentry);
4252 error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry);
4256 inode_unlock(path.dentry->d_inode);
4257 mnt_drop_write(path.mnt);
4260 if (retry_estale(error, lookup_flags)) {
4261 lookup_flags |= LOOKUP_REVAL;
4269 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4271 return do_rmdir(AT_FDCWD, getname(pathname));
4275 * vfs_unlink - unlink a filesystem object
4276 * @idmap: idmap of the mount the inode was found from
4277 * @dir: parent directory
4279 * @delegated_inode: returns victim inode, if the inode is delegated.
4281 * The caller must hold dir->i_mutex.
4283 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4284 * return a reference to the inode in delegated_inode. The caller
4285 * should then break the delegation on that inode and retry. Because
4286 * breaking a delegation may take a long time, the caller should drop
4287 * dir->i_mutex before doing so.
4289 * Alternatively, a caller may pass NULL for delegated_inode. This may
4290 * be appropriate for callers that expect the underlying filesystem not
4291 * to be NFS exported.
4293 * If the inode has been found through an idmapped mount the idmap of
4294 * the vfsmount must be passed through @idmap. This function will then take
4295 * care to map the inode according to @idmap before checking permissions.
4296 * On non-idmapped mounts or if permission checking is to be performed on the
4297 * raw inode simply passs @nop_mnt_idmap.
4299 int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir,
4300 struct dentry *dentry, struct inode **delegated_inode)
4302 struct inode *target = dentry->d_inode;
4303 int error = may_delete(idmap, dir, dentry, 0);
4308 if (!dir->i_op->unlink)
4312 if (IS_SWAPFILE(target))
4314 else if (is_local_mountpoint(dentry))
4317 error = security_inode_unlink(dir, dentry);
4319 error = try_break_deleg(target, delegated_inode);
4322 error = dir->i_op->unlink(dir, dentry);
4325 detach_mounts(dentry);
4330 inode_unlock(target);
4332 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4333 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4334 fsnotify_unlink(dir, dentry);
4335 } else if (!error) {
4336 fsnotify_link_count(target);
4337 d_delete_notify(dir, dentry);
4342 EXPORT_SYMBOL(vfs_unlink);
4345 * Make sure that the actual truncation of the file will occur outside its
4346 * directory's i_mutex. Truncate can take a long time if there is a lot of
4347 * writeout happening, and we don't want to prevent access to the directory
4348 * while waiting on the I/O.
4350 int do_unlinkat(int dfd, struct filename *name)
4353 struct dentry *dentry;
4357 struct inode *inode = NULL;
4358 struct inode *delegated_inode = NULL;
4359 unsigned int lookup_flags = 0;
4361 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4366 if (type != LAST_NORM)
4369 error = mnt_want_write(path.mnt);
4373 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4374 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4375 error = PTR_ERR(dentry);
4376 if (!IS_ERR(dentry)) {
4378 /* Why not before? Because we want correct error value */
4379 if (last.name[last.len] || d_is_negative(dentry))
4381 inode = dentry->d_inode;
4383 error = security_path_unlink(&path, dentry);
4386 error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4387 dentry, &delegated_inode);
4391 inode_unlock(path.dentry->d_inode);
4393 iput(inode); /* truncate the inode here */
4395 if (delegated_inode) {
4396 error = break_deleg_wait(&delegated_inode);
4400 mnt_drop_write(path.mnt);
4403 if (retry_estale(error, lookup_flags)) {
4404 lookup_flags |= LOOKUP_REVAL;
4413 if (d_is_negative(dentry))
4415 else if (d_is_dir(dentry))
4422 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4424 if ((flag & ~AT_REMOVEDIR) != 0)
4427 if (flag & AT_REMOVEDIR)
4428 return do_rmdir(dfd, getname(pathname));
4429 return do_unlinkat(dfd, getname(pathname));
4432 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4434 return do_unlinkat(AT_FDCWD, getname(pathname));
4438 * vfs_symlink - create symlink
4439 * @idmap: idmap of the mount the inode was found from
4440 * @dir: inode of @dentry
4441 * @dentry: pointer to dentry of the base directory
4442 * @oldname: name of the file to link to
4446 * If the inode has been found through an idmapped mount the idmap of
4447 * the vfsmount must be passed through @idmap. This function will then take
4448 * care to map the inode according to @idmap before checking permissions.
4449 * On non-idmapped mounts or if permission checking is to be performed on the
4450 * raw inode simply passs @nop_mnt_idmap.
4452 int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
4453 struct dentry *dentry, const char *oldname)
4457 error = may_create(idmap, dir, dentry);
4461 if (!dir->i_op->symlink)
4464 error = security_inode_symlink(dir, dentry, oldname);
4468 error = dir->i_op->symlink(idmap, dir, dentry, oldname);
4470 fsnotify_create(dir, dentry);
4473 EXPORT_SYMBOL(vfs_symlink);
4475 int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4478 struct dentry *dentry;
4480 unsigned int lookup_flags = 0;
4483 error = PTR_ERR(from);
4487 dentry = filename_create(newdfd, to, &path, lookup_flags);
4488 error = PTR_ERR(dentry);
4492 error = security_path_symlink(&path, dentry, from->name);
4494 error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4495 dentry, from->name);
4496 done_path_create(&path, dentry);
4497 if (retry_estale(error, lookup_flags)) {
4498 lookup_flags |= LOOKUP_REVAL;
4507 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4508 int, newdfd, const char __user *, newname)
4510 return do_symlinkat(getname(oldname), newdfd, getname(newname));
4513 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4515 return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4519 * vfs_link - create a new link
4520 * @old_dentry: object to be linked
4521 * @idmap: idmap of the mount
4523 * @new_dentry: where to create the new link
4524 * @delegated_inode: returns inode needing a delegation break
4526 * The caller must hold dir->i_mutex
4528 * If vfs_link discovers a delegation on the to-be-linked file in need
4529 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4530 * inode in delegated_inode. The caller should then break the delegation
4531 * and retry. Because breaking a delegation may take a long time, the
4532 * caller should drop the i_mutex before doing so.
4534 * Alternatively, a caller may pass NULL for delegated_inode. This may
4535 * be appropriate for callers that expect the underlying filesystem not
4536 * to be NFS exported.
4538 * If the inode has been found through an idmapped mount the idmap of
4539 * the vfsmount must be passed through @idmap. This function will then take
4540 * care to map the inode according to @idmap before checking permissions.
4541 * On non-idmapped mounts or if permission checking is to be performed on the
4542 * raw inode simply passs @nop_mnt_idmap.
4544 int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap,
4545 struct inode *dir, struct dentry *new_dentry,
4546 struct inode **delegated_inode)
4548 struct inode *inode = old_dentry->d_inode;
4549 unsigned max_links = dir->i_sb->s_max_links;
4555 error = may_create(idmap, dir, new_dentry);
4559 if (dir->i_sb != inode->i_sb)
4563 * A link to an append-only or immutable file cannot be created.
4565 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4568 * Updating the link count will likely cause i_uid and i_gid to
4569 * be writen back improperly if their true value is unknown to
4572 if (HAS_UNMAPPED_ID(idmap, inode))
4574 if (!dir->i_op->link)
4576 if (S_ISDIR(inode->i_mode))
4579 error = security_inode_link(old_dentry, dir, new_dentry);
4584 /* Make sure we don't allow creating hardlink to an unlinked file */
4585 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4587 else if (max_links && inode->i_nlink >= max_links)
4590 error = try_break_deleg(inode, delegated_inode);
4592 error = dir->i_op->link(old_dentry, dir, new_dentry);
4595 if (!error && (inode->i_state & I_LINKABLE)) {
4596 spin_lock(&inode->i_lock);
4597 inode->i_state &= ~I_LINKABLE;
4598 spin_unlock(&inode->i_lock);
4600 inode_unlock(inode);
4602 fsnotify_link(dir, inode, new_dentry);
4605 EXPORT_SYMBOL(vfs_link);
4608 * Hardlinks are often used in delicate situations. We avoid
4609 * security-related surprises by not following symlinks on the
4612 * We don't follow them on the oldname either to be compatible
4613 * with linux 2.0, and to avoid hard-linking to directories
4614 * and other special files. --ADM
4616 int do_linkat(int olddfd, struct filename *old, int newdfd,
4617 struct filename *new, int flags)
4619 struct mnt_idmap *idmap;
4620 struct dentry *new_dentry;
4621 struct path old_path, new_path;
4622 struct inode *delegated_inode = NULL;
4626 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4631 * To use null names we require CAP_DAC_READ_SEARCH
4632 * This ensures that not everyone will be able to create
4633 * handlink using the passed filedescriptor.
4635 if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4640 if (flags & AT_SYMLINK_FOLLOW)
4641 how |= LOOKUP_FOLLOW;
4643 error = filename_lookup(olddfd, old, how, &old_path, NULL);
4647 new_dentry = filename_create(newdfd, new, &new_path,
4648 (how & LOOKUP_REVAL));
4649 error = PTR_ERR(new_dentry);
4650 if (IS_ERR(new_dentry))
4654 if (old_path.mnt != new_path.mnt)
4656 idmap = mnt_idmap(new_path.mnt);
4657 error = may_linkat(idmap, &old_path);
4658 if (unlikely(error))
4660 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4663 error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode,
4664 new_dentry, &delegated_inode);
4666 done_path_create(&new_path, new_dentry);
4667 if (delegated_inode) {
4668 error = break_deleg_wait(&delegated_inode);
4670 path_put(&old_path);
4674 if (retry_estale(error, how)) {
4675 path_put(&old_path);
4676 how |= LOOKUP_REVAL;
4680 path_put(&old_path);
4688 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4689 int, newdfd, const char __user *, newname, int, flags)
4691 return do_linkat(olddfd, getname_uflags(oldname, flags),
4692 newdfd, getname(newname), flags);
4695 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4697 return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4701 * vfs_rename - rename a filesystem object
4702 * @rd: pointer to &struct renamedata info
4704 * The caller must hold multiple mutexes--see lock_rename()).
4706 * If vfs_rename discovers a delegation in need of breaking at either
4707 * the source or destination, it will return -EWOULDBLOCK and return a
4708 * reference to the inode in delegated_inode. The caller should then
4709 * break the delegation and retry. Because breaking a delegation may
4710 * take a long time, the caller should drop all locks before doing
4713 * Alternatively, a caller may pass NULL for delegated_inode. This may
4714 * be appropriate for callers that expect the underlying filesystem not
4715 * to be NFS exported.
4717 * The worst of all namespace operations - renaming directory. "Perverted"
4718 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4721 * a) we can get into loop creation.
4722 * b) race potential - two innocent renames can create a loop together.
4723 * That's where 4.4BSD screws up. Current fix: serialization on
4724 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4726 * c) we may have to lock up to _four_ objects - parents and victim (if it exists),
4727 * and source (if it's a non-directory or a subdirectory that moves to
4728 * different parent).
4729 * And that - after we got ->i_mutex on parents (until then we don't know
4730 * whether the target exists). Solution: try to be smart with locking
4731 * order for inodes. We rely on the fact that tree topology may change
4732 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4733 * move will be locked. Thus we can rank directories by the tree
4734 * (ancestors first) and rank all non-directories after them.
4735 * That works since everybody except rename does "lock parent, lookup,
4736 * lock child" and rename is under ->s_vfs_rename_mutex.
4737 * HOWEVER, it relies on the assumption that any object with ->lookup()
4738 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4739 * we'd better make sure that there's no link(2) for them.
4740 * d) conversion from fhandle to dentry may come in the wrong moment - when
4741 * we are removing the target. Solution: we will have to grab ->i_mutex
4742 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4743 * ->i_mutex on parents, which works but leads to some truly excessive
4746 int vfs_rename(struct renamedata *rd)
4749 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4750 struct dentry *old_dentry = rd->old_dentry;
4751 struct dentry *new_dentry = rd->new_dentry;
4752 struct inode **delegated_inode = rd->delegated_inode;
4753 unsigned int flags = rd->flags;
4754 bool is_dir = d_is_dir(old_dentry);
4755 struct inode *source = old_dentry->d_inode;
4756 struct inode *target = new_dentry->d_inode;
4757 bool new_is_dir = false;
4758 unsigned max_links = new_dir->i_sb->s_max_links;
4759 struct name_snapshot old_name;
4760 bool lock_old_subdir, lock_new_subdir;
4762 if (source == target)
4765 error = may_delete(rd->old_mnt_idmap, old_dir, old_dentry, is_dir);
4770 error = may_create(rd->new_mnt_idmap, new_dir, new_dentry);
4772 new_is_dir = d_is_dir(new_dentry);
4774 if (!(flags & RENAME_EXCHANGE))
4775 error = may_delete(rd->new_mnt_idmap, new_dir,
4776 new_dentry, is_dir);
4778 error = may_delete(rd->new_mnt_idmap, new_dir,
4779 new_dentry, new_is_dir);
4784 if (!old_dir->i_op->rename)
4788 * If we are going to change the parent - check write permissions,
4789 * we'll need to flip '..'.
4791 if (new_dir != old_dir) {
4793 error = inode_permission(rd->old_mnt_idmap, source,
4798 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4799 error = inode_permission(rd->new_mnt_idmap, target,
4806 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4811 take_dentry_name_snapshot(&old_name, old_dentry);
4815 * The source subdirectory needs to be locked on cross-directory
4816 * rename or cross-directory exchange since its parent changes.
4817 * The target subdirectory needs to be locked on cross-directory
4818 * exchange due to parent change and on any rename due to becoming
4820 * Non-directories need locking in all cases (for NFS reasons);
4821 * they get locked after any subdirectories (in inode address order).
4823 * NOTE: WE ONLY LOCK UNRELATED DIRECTORIES IN CROSS-DIRECTORY CASE.
4824 * NEVER, EVER DO THAT WITHOUT ->s_vfs_rename_mutex.
4826 lock_old_subdir = new_dir != old_dir;
4827 lock_new_subdir = new_dir != old_dir || !(flags & RENAME_EXCHANGE);
4829 if (lock_old_subdir)
4830 inode_lock_nested(source, I_MUTEX_CHILD);
4831 if (target && (!new_is_dir || lock_new_subdir))
4833 } else if (new_is_dir) {
4834 if (lock_new_subdir)
4835 inode_lock_nested(target, I_MUTEX_CHILD);
4838 lock_two_nondirectories(source, target);
4842 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4846 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4849 if (max_links && new_dir != old_dir) {
4851 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4853 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4854 old_dir->i_nlink >= max_links)
4858 error = try_break_deleg(source, delegated_inode);
4862 if (target && !new_is_dir) {
4863 error = try_break_deleg(target, delegated_inode);
4867 error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry,
4868 new_dir, new_dentry, flags);
4872 if (!(flags & RENAME_EXCHANGE) && target) {
4874 shrink_dcache_parent(new_dentry);
4875 target->i_flags |= S_DEAD;
4877 dont_mount(new_dentry);
4878 detach_mounts(new_dentry);
4880 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4881 if (!(flags & RENAME_EXCHANGE))
4882 d_move(old_dentry, new_dentry);
4884 d_exchange(old_dentry, new_dentry);
4887 if (!is_dir || lock_old_subdir)
4888 inode_unlock(source);
4889 if (target && (!new_is_dir || lock_new_subdir))
4890 inode_unlock(target);
4893 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4894 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4895 if (flags & RENAME_EXCHANGE) {
4896 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4897 new_is_dir, NULL, new_dentry);
4900 release_dentry_name_snapshot(&old_name);
4904 EXPORT_SYMBOL(vfs_rename);
4906 int do_renameat2(int olddfd, struct filename *from, int newdfd,
4907 struct filename *to, unsigned int flags)
4909 struct renamedata rd;
4910 struct dentry *old_dentry, *new_dentry;
4911 struct dentry *trap;
4912 struct path old_path, new_path;
4913 struct qstr old_last, new_last;
4914 int old_type, new_type;
4915 struct inode *delegated_inode = NULL;
4916 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4917 bool should_retry = false;
4918 int error = -EINVAL;
4920 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4923 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4924 (flags & RENAME_EXCHANGE))
4927 if (flags & RENAME_EXCHANGE)
4931 error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4932 &old_last, &old_type);
4936 error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4942 if (old_path.mnt != new_path.mnt)
4946 if (old_type != LAST_NORM)
4949 if (flags & RENAME_NOREPLACE)
4951 if (new_type != LAST_NORM)
4954 error = mnt_want_write(old_path.mnt);
4959 trap = lock_rename(new_path.dentry, old_path.dentry);
4961 old_dentry = lookup_one_qstr_excl(&old_last, old_path.dentry,
4963 error = PTR_ERR(old_dentry);
4964 if (IS_ERR(old_dentry))
4966 /* source must exist */
4968 if (d_is_negative(old_dentry))
4970 new_dentry = lookup_one_qstr_excl(&new_last, new_path.dentry,
4971 lookup_flags | target_flags);
4972 error = PTR_ERR(new_dentry);
4973 if (IS_ERR(new_dentry))
4976 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4978 if (flags & RENAME_EXCHANGE) {
4980 if (d_is_negative(new_dentry))
4983 if (!d_is_dir(new_dentry)) {
4985 if (new_last.name[new_last.len])
4989 /* unless the source is a directory trailing slashes give -ENOTDIR */
4990 if (!d_is_dir(old_dentry)) {
4992 if (old_last.name[old_last.len])
4994 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4997 /* source should not be ancestor of target */
4999 if (old_dentry == trap)
5001 /* target should not be an ancestor of source */
5002 if (!(flags & RENAME_EXCHANGE))
5004 if (new_dentry == trap)
5007 error = security_path_rename(&old_path, old_dentry,
5008 &new_path, new_dentry, flags);
5012 rd.old_dir = old_path.dentry->d_inode;
5013 rd.old_dentry = old_dentry;
5014 rd.old_mnt_idmap = mnt_idmap(old_path.mnt);
5015 rd.new_dir = new_path.dentry->d_inode;
5016 rd.new_dentry = new_dentry;
5017 rd.new_mnt_idmap = mnt_idmap(new_path.mnt);
5018 rd.delegated_inode = &delegated_inode;
5020 error = vfs_rename(&rd);
5026 unlock_rename(new_path.dentry, old_path.dentry);
5027 if (delegated_inode) {
5028 error = break_deleg_wait(&delegated_inode);
5032 mnt_drop_write(old_path.mnt);
5034 if (retry_estale(error, lookup_flags))
5035 should_retry = true;
5036 path_put(&new_path);
5038 path_put(&old_path);
5040 should_retry = false;
5041 lookup_flags |= LOOKUP_REVAL;
5050 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
5051 int, newdfd, const char __user *, newname, unsigned int, flags)
5053 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5057 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
5058 int, newdfd, const char __user *, newname)
5060 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5064 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5066 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
5067 getname(newname), 0);
5070 int readlink_copy(char __user *buffer, int buflen, const char *link)
5072 int len = PTR_ERR(link);
5077 if (len > (unsigned) buflen)
5079 if (copy_to_user(buffer, link, len))
5086 * vfs_readlink - copy symlink body into userspace buffer
5087 * @dentry: dentry on which to get symbolic link
5088 * @buffer: user memory pointer
5089 * @buflen: size of buffer
5091 * Does not touch atime. That's up to the caller if necessary
5093 * Does not call security hook.
5095 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5097 struct inode *inode = d_inode(dentry);
5098 DEFINE_DELAYED_CALL(done);
5102 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5103 if (unlikely(inode->i_op->readlink))
5104 return inode->i_op->readlink(dentry, buffer, buflen);
5106 if (!d_is_symlink(dentry))
5109 spin_lock(&inode->i_lock);
5110 inode->i_opflags |= IOP_DEFAULT_READLINK;
5111 spin_unlock(&inode->i_lock);
5114 link = READ_ONCE(inode->i_link);
5116 link = inode->i_op->get_link(dentry, inode, &done);
5118 return PTR_ERR(link);
5120 res = readlink_copy(buffer, buflen, link);
5121 do_delayed_call(&done);
5124 EXPORT_SYMBOL(vfs_readlink);
5127 * vfs_get_link - get symlink body
5128 * @dentry: dentry on which to get symbolic link
5129 * @done: caller needs to free returned data with this
5131 * Calls security hook and i_op->get_link() on the supplied inode.
5133 * It does not touch atime. That's up to the caller if necessary.
5135 * Does not work on "special" symlinks like /proc/$$/fd/N
5137 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5139 const char *res = ERR_PTR(-EINVAL);
5140 struct inode *inode = d_inode(dentry);
5142 if (d_is_symlink(dentry)) {
5143 res = ERR_PTR(security_inode_readlink(dentry));
5145 res = inode->i_op->get_link(dentry, inode, done);
5149 EXPORT_SYMBOL(vfs_get_link);
5151 /* get the link contents into pagecache */
5152 const char *page_get_link(struct dentry *dentry, struct inode *inode,
5153 struct delayed_call *callback)
5157 struct address_space *mapping = inode->i_mapping;
5160 page = find_get_page(mapping, 0);
5162 return ERR_PTR(-ECHILD);
5163 if (!PageUptodate(page)) {
5165 return ERR_PTR(-ECHILD);
5168 page = read_mapping_page(mapping, 0, NULL);
5172 set_delayed_call(callback, page_put_link, page);
5173 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5174 kaddr = page_address(page);
5175 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5179 EXPORT_SYMBOL(page_get_link);
5181 void page_put_link(void *arg)
5185 EXPORT_SYMBOL(page_put_link);
5187 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5189 DEFINE_DELAYED_CALL(done);
5190 int res = readlink_copy(buffer, buflen,
5191 page_get_link(dentry, d_inode(dentry),
5193 do_delayed_call(&done);
5196 EXPORT_SYMBOL(page_readlink);
5198 int page_symlink(struct inode *inode, const char *symname, int len)
5200 struct address_space *mapping = inode->i_mapping;
5201 const struct address_space_operations *aops = mapping->a_ops;
5202 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5204 void *fsdata = NULL;
5210 flags = memalloc_nofs_save();
5211 err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5213 memalloc_nofs_restore(flags);
5217 memcpy(page_address(page), symname, len-1);
5219 err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5226 mark_inode_dirty(inode);
5231 EXPORT_SYMBOL(page_symlink);
5233 const struct inode_operations page_symlink_inode_operations = {
5234 .get_link = page_get_link,
5236 EXPORT_SYMBOL(page_symlink_inode_operations);