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 pass @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 pass @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 pass @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,
1076 static int __init init_fs_namei_sysctls(void)
1078 register_sysctl_init("fs", namei_sysctls);
1081 fs_initcall(init_fs_namei_sysctls);
1083 #endif /* CONFIG_SYSCTL */
1086 * may_follow_link - Check symlink following for unsafe situations
1087 * @nd: nameidata pathwalk data
1088 * @inode: Used for idmapping.
1090 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1091 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1092 * in a sticky world-writable directory. This is to protect privileged
1093 * processes from failing races against path names that may change out
1094 * from under them by way of other users creating malicious symlinks.
1095 * It will permit symlinks to be followed only when outside a sticky
1096 * world-writable directory, or when the uid of the symlink and follower
1097 * match, or when the directory owner matches the symlink's owner.
1099 * Returns 0 if following the symlink is allowed, -ve on error.
1101 static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1103 struct mnt_idmap *idmap;
1106 if (!sysctl_protected_symlinks)
1109 idmap = mnt_idmap(nd->path.mnt);
1110 vfsuid = i_uid_into_vfsuid(idmap, inode);
1111 /* Allowed if owner and follower match. */
1112 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
1115 /* Allowed if parent directory not sticky and world-writable. */
1116 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1119 /* Allowed if parent directory and link owner match. */
1120 if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid))
1123 if (nd->flags & LOOKUP_RCU)
1126 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1127 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1132 * safe_hardlink_source - Check for safe hardlink conditions
1133 * @idmap: idmap of the mount the inode was found from
1134 * @inode: the source inode to hardlink from
1136 * Return false if at least one of the following conditions:
1137 * - inode is not a regular file
1139 * - inode is setgid and group-exec
1140 * - access failure for read and write
1142 * Otherwise returns true.
1144 static bool safe_hardlink_source(struct mnt_idmap *idmap,
1145 struct inode *inode)
1147 umode_t mode = inode->i_mode;
1149 /* Special files should not get pinned to the filesystem. */
1153 /* Setuid files should not get pinned to the filesystem. */
1157 /* Executable setgid files should not get pinned to the filesystem. */
1158 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1161 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1162 if (inode_permission(idmap, inode, MAY_READ | MAY_WRITE))
1169 * may_linkat - Check permissions for creating a hardlink
1170 * @idmap: idmap of the mount the inode was found from
1171 * @link: the source to hardlink from
1173 * Block hardlink when all of:
1174 * - sysctl_protected_hardlinks enabled
1175 * - fsuid does not match inode
1176 * - hardlink source is unsafe (see safe_hardlink_source() above)
1177 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1179 * If the inode has been found through an idmapped mount the idmap of
1180 * the vfsmount must be passed through @idmap. This function will then take
1181 * care to map the inode according to @idmap before checking permissions.
1182 * On non-idmapped mounts or if permission checking is to be performed on the
1183 * raw inode simply pass @nop_mnt_idmap.
1185 * Returns 0 if successful, -ve on error.
1187 int may_linkat(struct mnt_idmap *idmap, const struct path *link)
1189 struct inode *inode = link->dentry->d_inode;
1191 /* Inode writeback is not safe when the uid or gid are invalid. */
1192 if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
1193 !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
1196 if (!sysctl_protected_hardlinks)
1199 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1200 * otherwise, it must be a safe source.
1202 if (safe_hardlink_source(idmap, inode) ||
1203 inode_owner_or_capable(idmap, inode))
1206 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1211 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1212 * should be allowed, or not, on files that already
1214 * @idmap: idmap of the mount the inode was found from
1215 * @nd: nameidata pathwalk data
1216 * @inode: the inode of the file to open
1218 * Block an O_CREAT open of a FIFO (or a regular file) when:
1219 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1220 * - the file already exists
1221 * - we are in a sticky directory
1222 * - we don't own the file
1223 * - the owner of the directory doesn't own the file
1224 * - the directory is world writable
1225 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1226 * the directory doesn't have to be world writable: being group writable will
1229 * If the inode has been found through an idmapped mount the idmap of
1230 * the vfsmount must be passed through @idmap. This function will then take
1231 * care to map the inode according to @idmap before checking permissions.
1232 * On non-idmapped mounts or if permission checking is to be performed on the
1233 * raw inode simply pass @nop_mnt_idmap.
1235 * Returns 0 if the open is allowed, -ve on error.
1237 static int may_create_in_sticky(struct mnt_idmap *idmap,
1238 struct nameidata *nd, struct inode *const inode)
1240 umode_t dir_mode = nd->dir_mode;
1241 vfsuid_t dir_vfsuid = nd->dir_vfsuid;
1243 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1244 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1245 likely(!(dir_mode & S_ISVTX)) ||
1246 vfsuid_eq(i_uid_into_vfsuid(idmap, inode), dir_vfsuid) ||
1247 vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), current_fsuid()))
1250 if (likely(dir_mode & 0002) ||
1252 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1253 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1254 const char *operation = S_ISFIFO(inode->i_mode) ?
1255 "sticky_create_fifo" :
1256 "sticky_create_regular";
1257 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1264 * follow_up - Find the mountpoint of path's vfsmount
1266 * Given a path, find the mountpoint of its source file system.
1267 * Replace @path with the path of the mountpoint in the parent mount.
1270 * Return 1 if we went up a level and 0 if we were already at the
1273 int follow_up(struct path *path)
1275 struct mount *mnt = real_mount(path->mnt);
1276 struct mount *parent;
1277 struct dentry *mountpoint;
1279 read_seqlock_excl(&mount_lock);
1280 parent = mnt->mnt_parent;
1281 if (parent == mnt) {
1282 read_sequnlock_excl(&mount_lock);
1285 mntget(&parent->mnt);
1286 mountpoint = dget(mnt->mnt_mountpoint);
1287 read_sequnlock_excl(&mount_lock);
1289 path->dentry = mountpoint;
1291 path->mnt = &parent->mnt;
1294 EXPORT_SYMBOL(follow_up);
1296 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1297 struct path *path, unsigned *seqp)
1299 while (mnt_has_parent(m)) {
1300 struct dentry *mountpoint = m->mnt_mountpoint;
1303 if (unlikely(root->dentry == mountpoint &&
1304 root->mnt == &m->mnt))
1306 if (mountpoint != m->mnt.mnt_root) {
1307 path->mnt = &m->mnt;
1308 path->dentry = mountpoint;
1309 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1316 static bool choose_mountpoint(struct mount *m, const struct path *root,
1323 unsigned seq, mseq = read_seqbegin(&mount_lock);
1325 found = choose_mountpoint_rcu(m, root, path, &seq);
1326 if (unlikely(!found)) {
1327 if (!read_seqretry(&mount_lock, mseq))
1330 if (likely(__legitimize_path(path, seq, mseq)))
1342 * Perform an automount
1343 * - return -EISDIR to tell follow_managed() to stop and return the path we
1346 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1348 struct dentry *dentry = path->dentry;
1350 /* We don't want to mount if someone's just doing a stat -
1351 * unless they're stat'ing a directory and appended a '/' to
1354 * We do, however, want to mount if someone wants to open or
1355 * create a file of any type under the mountpoint, wants to
1356 * traverse through the mountpoint or wants to open the
1357 * mounted directory. Also, autofs may mark negative dentries
1358 * as being automount points. These will need the attentions
1359 * of the daemon to instantiate them before they can be used.
1361 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1362 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1366 if (count && (*count)++ >= MAXSYMLINKS)
1369 return finish_automount(dentry->d_op->d_automount(path), path);
1373 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1374 * dentries are pinned but not locked here, so negative dentry can go
1375 * positive right under us. Use of smp_load_acquire() provides a barrier
1376 * sufficient for ->d_inode and ->d_flags consistency.
1378 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1379 int *count, unsigned lookup_flags)
1381 struct vfsmount *mnt = path->mnt;
1382 bool need_mntput = false;
1385 while (flags & DCACHE_MANAGED_DENTRY) {
1386 /* Allow the filesystem to manage the transit without i_mutex
1388 if (flags & DCACHE_MANAGE_TRANSIT) {
1389 ret = path->dentry->d_op->d_manage(path, false);
1390 flags = smp_load_acquire(&path->dentry->d_flags);
1395 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1396 struct vfsmount *mounted = lookup_mnt(path);
1397 if (mounted) { // ... in our namespace
1401 path->mnt = mounted;
1402 path->dentry = dget(mounted->mnt_root);
1403 // here we know it's positive
1404 flags = path->dentry->d_flags;
1410 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1413 // uncovered automount point
1414 ret = follow_automount(path, count, lookup_flags);
1415 flags = smp_load_acquire(&path->dentry->d_flags);
1422 // possible if you race with several mount --move
1423 if (need_mntput && path->mnt == mnt)
1425 if (!ret && unlikely(d_flags_negative(flags)))
1427 *jumped = need_mntput;
1431 static inline int traverse_mounts(struct path *path, bool *jumped,
1432 int *count, unsigned lookup_flags)
1434 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1437 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1439 if (unlikely(d_flags_negative(flags)))
1443 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1446 int follow_down_one(struct path *path)
1448 struct vfsmount *mounted;
1450 mounted = lookup_mnt(path);
1454 path->mnt = mounted;
1455 path->dentry = dget(mounted->mnt_root);
1460 EXPORT_SYMBOL(follow_down_one);
1463 * Follow down to the covering mount currently visible to userspace. At each
1464 * point, the filesystem owning that dentry may be queried as to whether the
1465 * caller is permitted to proceed or not.
1467 int follow_down(struct path *path, unsigned int flags)
1469 struct vfsmount *mnt = path->mnt;
1471 int ret = traverse_mounts(path, &jumped, NULL, flags);
1473 if (path->mnt != mnt)
1477 EXPORT_SYMBOL(follow_down);
1480 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1481 * we meet a managed dentry that would need blocking.
1483 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1485 struct dentry *dentry = path->dentry;
1486 unsigned int flags = dentry->d_flags;
1488 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1491 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1496 * Don't forget we might have a non-mountpoint managed dentry
1497 * that wants to block transit.
1499 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1500 int res = dentry->d_op->d_manage(path, true);
1502 return res == -EISDIR;
1503 flags = dentry->d_flags;
1506 if (flags & DCACHE_MOUNTED) {
1507 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1509 path->mnt = &mounted->mnt;
1510 dentry = path->dentry = mounted->mnt.mnt_root;
1511 nd->state |= ND_JUMPED;
1512 nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1513 flags = dentry->d_flags;
1514 // makes sure that non-RCU pathwalk could reach
1516 if (read_seqretry(&mount_lock, nd->m_seq))
1520 if (read_seqretry(&mount_lock, nd->m_seq))
1523 return !(flags & DCACHE_NEED_AUTOMOUNT);
1527 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1533 path->mnt = nd->path.mnt;
1534 path->dentry = dentry;
1535 if (nd->flags & LOOKUP_RCU) {
1536 unsigned int seq = nd->next_seq;
1537 if (likely(__follow_mount_rcu(nd, path)))
1539 // *path and nd->next_seq might've been clobbered
1540 path->mnt = nd->path.mnt;
1541 path->dentry = dentry;
1543 if (!try_to_unlazy_next(nd, dentry))
1546 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1548 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1551 nd->state |= ND_JUMPED;
1553 if (unlikely(ret)) {
1555 if (path->mnt != nd->path.mnt)
1562 * This looks up the name in dcache and possibly revalidates the found dentry.
1563 * NULL is returned if the dentry does not exist in the cache.
1565 static struct dentry *lookup_dcache(const struct qstr *name,
1569 struct dentry *dentry = d_lookup(dir, name);
1571 int error = d_revalidate(dentry, flags);
1572 if (unlikely(error <= 0)) {
1574 d_invalidate(dentry);
1576 return ERR_PTR(error);
1583 * Parent directory has inode locked exclusive. This is one
1584 * and only case when ->lookup() gets called on non in-lookup
1585 * dentries - as the matter of fact, this only gets called
1586 * when directory is guaranteed to have no in-lookup children
1589 struct dentry *lookup_one_qstr_excl(const struct qstr *name,
1590 struct dentry *base,
1593 struct dentry *dentry = lookup_dcache(name, base, flags);
1595 struct inode *dir = base->d_inode;
1600 /* Don't create child dentry for a dead directory. */
1601 if (unlikely(IS_DEADDIR(dir)))
1602 return ERR_PTR(-ENOENT);
1604 dentry = d_alloc(base, name);
1605 if (unlikely(!dentry))
1606 return ERR_PTR(-ENOMEM);
1608 old = dir->i_op->lookup(dir, dentry, flags);
1609 if (unlikely(old)) {
1615 EXPORT_SYMBOL(lookup_one_qstr_excl);
1617 static struct dentry *lookup_fast(struct nameidata *nd)
1619 struct dentry *dentry, *parent = nd->path.dentry;
1623 * Rename seqlock is not required here because in the off chance
1624 * of a false negative due to a concurrent rename, the caller is
1625 * going to fall back to non-racy lookup.
1627 if (nd->flags & LOOKUP_RCU) {
1628 dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq);
1629 if (unlikely(!dentry)) {
1630 if (!try_to_unlazy(nd))
1631 return ERR_PTR(-ECHILD);
1636 * This sequence count validates that the parent had no
1637 * changes while we did the lookup of the dentry above.
1639 if (read_seqcount_retry(&parent->d_seq, nd->seq))
1640 return ERR_PTR(-ECHILD);
1642 status = d_revalidate(dentry, nd->flags);
1643 if (likely(status > 0))
1645 if (!try_to_unlazy_next(nd, dentry))
1646 return ERR_PTR(-ECHILD);
1647 if (status == -ECHILD)
1648 /* we'd been told to redo it in non-rcu mode */
1649 status = d_revalidate(dentry, nd->flags);
1651 dentry = __d_lookup(parent, &nd->last);
1652 if (unlikely(!dentry))
1654 status = d_revalidate(dentry, nd->flags);
1656 if (unlikely(status <= 0)) {
1658 d_invalidate(dentry);
1660 return ERR_PTR(status);
1665 /* Fast lookup failed, do it the slow way */
1666 static struct dentry *__lookup_slow(const struct qstr *name,
1670 struct dentry *dentry, *old;
1671 struct inode *inode = dir->d_inode;
1672 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1674 /* Don't go there if it's already dead */
1675 if (unlikely(IS_DEADDIR(inode)))
1676 return ERR_PTR(-ENOENT);
1678 dentry = d_alloc_parallel(dir, name, &wq);
1681 if (unlikely(!d_in_lookup(dentry))) {
1682 int error = d_revalidate(dentry, flags);
1683 if (unlikely(error <= 0)) {
1685 d_invalidate(dentry);
1690 dentry = ERR_PTR(error);
1693 old = inode->i_op->lookup(inode, dentry, flags);
1694 d_lookup_done(dentry);
1695 if (unlikely(old)) {
1703 static struct dentry *lookup_slow(const struct qstr *name,
1707 struct inode *inode = dir->d_inode;
1709 inode_lock_shared(inode);
1710 res = __lookup_slow(name, dir, flags);
1711 inode_unlock_shared(inode);
1715 static inline int may_lookup(struct mnt_idmap *idmap,
1716 struct nameidata *nd)
1718 if (nd->flags & LOOKUP_RCU) {
1719 int err = inode_permission(idmap, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1720 if (!err) // success, keep going
1722 if (!try_to_unlazy(nd))
1723 return -ECHILD; // redo it all non-lazy
1724 if (err != -ECHILD) // hard error
1727 return inode_permission(idmap, nd->inode, MAY_EXEC);
1730 static int reserve_stack(struct nameidata *nd, struct path *link)
1732 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1735 if (likely(nd->depth != EMBEDDED_LEVELS))
1737 if (likely(nd->stack != nd->internal))
1739 if (likely(nd_alloc_stack(nd)))
1742 if (nd->flags & LOOKUP_RCU) {
1743 // we need to grab link before we do unlazy. And we can't skip
1744 // unlazy even if we fail to grab the link - cleanup needs it
1745 bool grabbed_link = legitimize_path(nd, link, nd->next_seq);
1747 if (!try_to_unlazy(nd) || !grabbed_link)
1750 if (nd_alloc_stack(nd))
1756 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1758 static const char *pick_link(struct nameidata *nd, struct path *link,
1759 struct inode *inode, int flags)
1763 int error = reserve_stack(nd, link);
1765 if (unlikely(error)) {
1766 if (!(nd->flags & LOOKUP_RCU))
1768 return ERR_PTR(error);
1770 last = nd->stack + nd->depth++;
1772 clear_delayed_call(&last->done);
1773 last->seq = nd->next_seq;
1775 if (flags & WALK_TRAILING) {
1776 error = may_follow_link(nd, inode);
1777 if (unlikely(error))
1778 return ERR_PTR(error);
1781 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1782 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1783 return ERR_PTR(-ELOOP);
1785 if (!(nd->flags & LOOKUP_RCU)) {
1786 touch_atime(&last->link);
1788 } else if (atime_needs_update(&last->link, inode)) {
1789 if (!try_to_unlazy(nd))
1790 return ERR_PTR(-ECHILD);
1791 touch_atime(&last->link);
1794 error = security_inode_follow_link(link->dentry, inode,
1795 nd->flags & LOOKUP_RCU);
1796 if (unlikely(error))
1797 return ERR_PTR(error);
1799 res = READ_ONCE(inode->i_link);
1801 const char * (*get)(struct dentry *, struct inode *,
1802 struct delayed_call *);
1803 get = inode->i_op->get_link;
1804 if (nd->flags & LOOKUP_RCU) {
1805 res = get(NULL, inode, &last->done);
1806 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1807 res = get(link->dentry, inode, &last->done);
1809 res = get(link->dentry, inode, &last->done);
1817 error = nd_jump_root(nd);
1818 if (unlikely(error))
1819 return ERR_PTR(error);
1820 while (unlikely(*++res == '/'))
1825 all_done: // pure jump
1831 * Do we need to follow links? We _really_ want to be able
1832 * to do this check without having to look at inode->i_op,
1833 * so we keep a cache of "no, this doesn't need follow_link"
1834 * for the common case.
1836 * NOTE: dentry must be what nd->next_seq had been sampled from.
1838 static const char *step_into(struct nameidata *nd, int flags,
1839 struct dentry *dentry)
1842 struct inode *inode;
1843 int err = handle_mounts(nd, dentry, &path);
1846 return ERR_PTR(err);
1847 inode = path.dentry->d_inode;
1848 if (likely(!d_is_symlink(path.dentry)) ||
1849 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1850 (flags & WALK_NOFOLLOW)) {
1851 /* not a symlink or should not follow */
1852 if (nd->flags & LOOKUP_RCU) {
1853 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1854 return ERR_PTR(-ECHILD);
1855 if (unlikely(!inode))
1856 return ERR_PTR(-ENOENT);
1858 dput(nd->path.dentry);
1859 if (nd->path.mnt != path.mnt)
1860 mntput(nd->path.mnt);
1864 nd->seq = nd->next_seq;
1867 if (nd->flags & LOOKUP_RCU) {
1868 /* make sure that d_is_symlink above matches inode */
1869 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1870 return ERR_PTR(-ECHILD);
1872 if (path.mnt == nd->path.mnt)
1875 return pick_link(nd, &path, inode, flags);
1878 static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1880 struct dentry *parent, *old;
1882 if (path_equal(&nd->path, &nd->root))
1884 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1887 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1888 &nd->root, &path, &seq))
1890 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1891 return ERR_PTR(-ECHILD);
1893 nd->inode = path.dentry->d_inode;
1895 // makes sure that non-RCU pathwalk could reach this state
1896 if (read_seqretry(&mount_lock, nd->m_seq))
1897 return ERR_PTR(-ECHILD);
1898 /* we know that mountpoint was pinned */
1900 old = nd->path.dentry;
1901 parent = old->d_parent;
1902 nd->next_seq = read_seqcount_begin(&parent->d_seq);
1903 // makes sure that non-RCU pathwalk could reach this state
1904 if (read_seqcount_retry(&old->d_seq, nd->seq))
1905 return ERR_PTR(-ECHILD);
1906 if (unlikely(!path_connected(nd->path.mnt, parent)))
1907 return ERR_PTR(-ECHILD);
1910 if (read_seqretry(&mount_lock, nd->m_seq))
1911 return ERR_PTR(-ECHILD);
1912 if (unlikely(nd->flags & LOOKUP_BENEATH))
1913 return ERR_PTR(-ECHILD);
1914 nd->next_seq = nd->seq;
1915 return nd->path.dentry;
1918 static struct dentry *follow_dotdot(struct nameidata *nd)
1920 struct dentry *parent;
1922 if (path_equal(&nd->path, &nd->root))
1924 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1927 if (!choose_mountpoint(real_mount(nd->path.mnt),
1930 path_put(&nd->path);
1932 nd->inode = path.dentry->d_inode;
1933 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1934 return ERR_PTR(-EXDEV);
1936 /* rare case of legitimate dget_parent()... */
1937 parent = dget_parent(nd->path.dentry);
1938 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1940 return ERR_PTR(-ENOENT);
1945 if (unlikely(nd->flags & LOOKUP_BENEATH))
1946 return ERR_PTR(-EXDEV);
1947 return dget(nd->path.dentry);
1950 static const char *handle_dots(struct nameidata *nd, int type)
1952 if (type == LAST_DOTDOT) {
1953 const char *error = NULL;
1954 struct dentry *parent;
1956 if (!nd->root.mnt) {
1957 error = ERR_PTR(set_root(nd));
1961 if (nd->flags & LOOKUP_RCU)
1962 parent = follow_dotdot_rcu(nd);
1964 parent = follow_dotdot(nd);
1966 return ERR_CAST(parent);
1967 error = step_into(nd, WALK_NOFOLLOW, parent);
1968 if (unlikely(error))
1971 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1973 * If there was a racing rename or mount along our
1974 * path, then we can't be sure that ".." hasn't jumped
1975 * above nd->root (and so userspace should retry or use
1979 if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
1980 return ERR_PTR(-EAGAIN);
1981 if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
1982 return ERR_PTR(-EAGAIN);
1988 static const char *walk_component(struct nameidata *nd, int flags)
1990 struct dentry *dentry;
1992 * "." and ".." are special - ".." especially so because it has
1993 * to be able to know about the current root directory and
1994 * parent relationships.
1996 if (unlikely(nd->last_type != LAST_NORM)) {
1997 if (!(flags & WALK_MORE) && nd->depth)
1999 return handle_dots(nd, nd->last_type);
2001 dentry = lookup_fast(nd);
2003 return ERR_CAST(dentry);
2004 if (unlikely(!dentry)) {
2005 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
2007 return ERR_CAST(dentry);
2009 if (!(flags & WALK_MORE) && nd->depth)
2011 return step_into(nd, flags, dentry);
2015 * We can do the critical dentry name comparison and hashing
2016 * operations one word at a time, but we are limited to:
2018 * - Architectures with fast unaligned word accesses. We could
2019 * do a "get_unaligned()" if this helps and is sufficiently
2022 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2023 * do not trap on the (extremely unlikely) case of a page
2024 * crossing operation.
2026 * - Furthermore, we need an efficient 64-bit compile for the
2027 * 64-bit case in order to generate the "number of bytes in
2028 * the final mask". Again, that could be replaced with a
2029 * efficient population count instruction or similar.
2031 #ifdef CONFIG_DCACHE_WORD_ACCESS
2033 #include <asm/word-at-a-time.h>
2037 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2039 #elif defined(CONFIG_64BIT)
2041 * Register pressure in the mixing function is an issue, particularly
2042 * on 32-bit x86, but almost any function requires one state value and
2043 * one temporary. Instead, use a function designed for two state values
2044 * and no temporaries.
2046 * This function cannot create a collision in only two iterations, so
2047 * we have two iterations to achieve avalanche. In those two iterations,
2048 * we have six layers of mixing, which is enough to spread one bit's
2049 * influence out to 2^6 = 64 state bits.
2051 * Rotate constants are scored by considering either 64 one-bit input
2052 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2053 * probability of that delta causing a change to each of the 128 output
2054 * bits, using a sample of random initial states.
2056 * The Shannon entropy of the computed probabilities is then summed
2057 * to produce a score. Ideally, any input change has a 50% chance of
2058 * toggling any given output bit.
2060 * Mixing scores (in bits) for (12,45):
2061 * Input delta: 1-bit 2-bit
2062 * 1 round: 713.3 42542.6
2063 * 2 rounds: 2753.7 140389.8
2064 * 3 rounds: 5954.1 233458.2
2065 * 4 rounds: 7862.6 256672.2
2066 * Perfect: 8192 258048
2067 * (64*128) (64*63/2 * 128)
2069 #define HASH_MIX(x, y, a) \
2071 y ^= x, x = rol64(x,12),\
2072 x += y, y = rol64(y,45),\
2076 * Fold two longs into one 32-bit hash value. This must be fast, but
2077 * latency isn't quite as critical, as there is a fair bit of additional
2078 * work done before the hash value is used.
2080 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2082 y ^= x * GOLDEN_RATIO_64;
2083 y *= GOLDEN_RATIO_64;
2087 #else /* 32-bit case */
2090 * Mixing scores (in bits) for (7,20):
2091 * Input delta: 1-bit 2-bit
2092 * 1 round: 330.3 9201.6
2093 * 2 rounds: 1246.4 25475.4
2094 * 3 rounds: 1907.1 31295.1
2095 * 4 rounds: 2042.3 31718.6
2096 * Perfect: 2048 31744
2097 * (32*64) (32*31/2 * 64)
2099 #define HASH_MIX(x, y, a) \
2101 y ^= x, x = rol32(x, 7),\
2102 x += y, y = rol32(y,20),\
2105 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2107 /* Use arch-optimized multiply if one exists */
2108 return __hash_32(y ^ __hash_32(x));
2114 * Return the hash of a string of known length. This is carfully
2115 * designed to match hash_name(), which is the more critical function.
2116 * In particular, we must end by hashing a final word containing 0..7
2117 * payload bytes, to match the way that hash_name() iterates until it
2118 * finds the delimiter after the name.
2120 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2122 unsigned long a, x = 0, y = (unsigned long)salt;
2127 a = load_unaligned_zeropad(name);
2128 if (len < sizeof(unsigned long))
2131 name += sizeof(unsigned long);
2132 len -= sizeof(unsigned long);
2134 x ^= a & bytemask_from_count(len);
2136 return fold_hash(x, y);
2138 EXPORT_SYMBOL(full_name_hash);
2140 /* Return the "hash_len" (hash and length) of a null-terminated string */
2141 u64 hashlen_string(const void *salt, const char *name)
2143 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2144 unsigned long adata, mask, len;
2145 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2152 len += sizeof(unsigned long);
2154 a = load_unaligned_zeropad(name+len);
2155 } while (!has_zero(a, &adata, &constants));
2157 adata = prep_zero_mask(a, adata, &constants);
2158 mask = create_zero_mask(adata);
2159 x ^= a & zero_bytemask(mask);
2161 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2163 EXPORT_SYMBOL(hashlen_string);
2166 * Calculate the length and hash of the path component, and
2167 * return the "hash_len" as the result.
2169 static inline u64 hash_name(const void *salt, const char *name)
2171 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2172 unsigned long adata, bdata, mask, len;
2173 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2180 len += sizeof(unsigned long);
2182 a = load_unaligned_zeropad(name+len);
2183 b = a ^ REPEAT_BYTE('/');
2184 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2186 adata = prep_zero_mask(a, adata, &constants);
2187 bdata = prep_zero_mask(b, bdata, &constants);
2188 mask = create_zero_mask(adata | bdata);
2189 x ^= a & zero_bytemask(mask);
2191 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2194 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2196 /* Return the hash of a string of known length */
2197 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2199 unsigned long hash = init_name_hash(salt);
2201 hash = partial_name_hash((unsigned char)*name++, hash);
2202 return end_name_hash(hash);
2204 EXPORT_SYMBOL(full_name_hash);
2206 /* Return the "hash_len" (hash and length) of a null-terminated string */
2207 u64 hashlen_string(const void *salt, const char *name)
2209 unsigned long hash = init_name_hash(salt);
2210 unsigned long len = 0, c;
2212 c = (unsigned char)*name;
2215 hash = partial_name_hash(c, hash);
2216 c = (unsigned char)name[len];
2218 return hashlen_create(end_name_hash(hash), len);
2220 EXPORT_SYMBOL(hashlen_string);
2223 * We know there's a real path component here of at least
2226 static inline u64 hash_name(const void *salt, const char *name)
2228 unsigned long hash = init_name_hash(salt);
2229 unsigned long len = 0, c;
2231 c = (unsigned char)*name;
2234 hash = partial_name_hash(c, hash);
2235 c = (unsigned char)name[len];
2236 } while (c && c != '/');
2237 return hashlen_create(end_name_hash(hash), len);
2244 * This is the basic name resolution function, turning a pathname into
2245 * the final dentry. We expect 'base' to be positive and a directory.
2247 * Returns 0 and nd will have valid dentry and mnt on success.
2248 * Returns error and drops reference to input namei data on failure.
2250 static int link_path_walk(const char *name, struct nameidata *nd)
2252 int depth = 0; // depth <= nd->depth
2255 nd->last_type = LAST_ROOT;
2256 nd->flags |= LOOKUP_PARENT;
2258 return PTR_ERR(name);
2262 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2266 /* At this point we know we have a real path component. */
2268 struct mnt_idmap *idmap;
2273 idmap = mnt_idmap(nd->path.mnt);
2274 err = may_lookup(idmap, nd);
2278 hash_len = hash_name(nd->path.dentry, name);
2281 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2283 if (name[1] == '.') {
2285 nd->state |= ND_JUMPED;
2291 if (likely(type == LAST_NORM)) {
2292 struct dentry *parent = nd->path.dentry;
2293 nd->state &= ~ND_JUMPED;
2294 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2295 struct qstr this = { { .hash_len = hash_len }, .name = name };
2296 err = parent->d_op->d_hash(parent, &this);
2299 hash_len = this.hash_len;
2304 nd->last.hash_len = hash_len;
2305 nd->last.name = name;
2306 nd->last_type = type;
2308 name += hashlen_len(hash_len);
2312 * If it wasn't NUL, we know it was '/'. Skip that
2313 * slash, and continue until no more slashes.
2317 } while (unlikely(*name == '/'));
2318 if (unlikely(!*name)) {
2320 /* pathname or trailing symlink, done */
2322 nd->dir_vfsuid = i_uid_into_vfsuid(idmap, nd->inode);
2323 nd->dir_mode = nd->inode->i_mode;
2324 nd->flags &= ~LOOKUP_PARENT;
2327 /* last component of nested symlink */
2328 name = nd->stack[--depth].name;
2329 link = walk_component(nd, 0);
2331 /* not the last component */
2332 link = walk_component(nd, WALK_MORE);
2334 if (unlikely(link)) {
2336 return PTR_ERR(link);
2337 /* a symlink to follow */
2338 nd->stack[depth++].name = name;
2342 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2343 if (nd->flags & LOOKUP_RCU) {
2344 if (!try_to_unlazy(nd))
2352 /* must be paired with terminate_walk() */
2353 static const char *path_init(struct nameidata *nd, unsigned flags)
2356 const char *s = nd->name->name;
2358 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2359 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2360 return ERR_PTR(-EAGAIN);
2363 flags &= ~LOOKUP_RCU;
2364 if (flags & LOOKUP_RCU)
2367 nd->seq = nd->next_seq = 0;
2370 nd->state |= ND_JUMPED;
2372 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2373 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2376 if (nd->state & ND_ROOT_PRESET) {
2377 struct dentry *root = nd->root.dentry;
2378 struct inode *inode = root->d_inode;
2379 if (*s && unlikely(!d_can_lookup(root)))
2380 return ERR_PTR(-ENOTDIR);
2381 nd->path = nd->root;
2383 if (flags & LOOKUP_RCU) {
2384 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2385 nd->root_seq = nd->seq;
2387 path_get(&nd->path);
2392 nd->root.mnt = NULL;
2394 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2395 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2396 error = nd_jump_root(nd);
2397 if (unlikely(error))
2398 return ERR_PTR(error);
2402 /* Relative pathname -- get the starting-point it is relative to. */
2403 if (nd->dfd == AT_FDCWD) {
2404 if (flags & LOOKUP_RCU) {
2405 struct fs_struct *fs = current->fs;
2409 seq = read_seqcount_begin(&fs->seq);
2411 nd->inode = nd->path.dentry->d_inode;
2412 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2413 } while (read_seqcount_retry(&fs->seq, seq));
2415 get_fs_pwd(current->fs, &nd->path);
2416 nd->inode = nd->path.dentry->d_inode;
2419 /* Caller must check execute permissions on the starting path component */
2420 struct fd f = fdget_raw(nd->dfd);
2421 struct dentry *dentry;
2424 return ERR_PTR(-EBADF);
2426 dentry = f.file->f_path.dentry;
2428 if (*s && unlikely(!d_can_lookup(dentry))) {
2430 return ERR_PTR(-ENOTDIR);
2433 nd->path = f.file->f_path;
2434 if (flags & LOOKUP_RCU) {
2435 nd->inode = nd->path.dentry->d_inode;
2436 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2438 path_get(&nd->path);
2439 nd->inode = nd->path.dentry->d_inode;
2444 /* For scoped-lookups we need to set the root to the dirfd as well. */
2445 if (flags & LOOKUP_IS_SCOPED) {
2446 nd->root = nd->path;
2447 if (flags & LOOKUP_RCU) {
2448 nd->root_seq = nd->seq;
2450 path_get(&nd->root);
2451 nd->state |= ND_ROOT_GRABBED;
2457 static inline const char *lookup_last(struct nameidata *nd)
2459 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2460 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2462 return walk_component(nd, WALK_TRAILING);
2465 static int handle_lookup_down(struct nameidata *nd)
2467 if (!(nd->flags & LOOKUP_RCU))
2468 dget(nd->path.dentry);
2469 nd->next_seq = nd->seq;
2470 return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry));
2473 /* Returns 0 and nd will be valid on success; Returns error, otherwise. */
2474 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2476 const char *s = path_init(nd, flags);
2479 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2480 err = handle_lookup_down(nd);
2481 if (unlikely(err < 0))
2485 while (!(err = link_path_walk(s, nd)) &&
2486 (s = lookup_last(nd)) != NULL)
2488 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2489 err = handle_lookup_down(nd);
2490 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2493 err = complete_walk(nd);
2495 if (!err && nd->flags & LOOKUP_DIRECTORY)
2496 if (!d_can_lookup(nd->path.dentry))
2500 nd->path.mnt = NULL;
2501 nd->path.dentry = NULL;
2507 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2508 struct path *path, struct path *root)
2511 struct nameidata nd;
2513 return PTR_ERR(name);
2514 set_nameidata(&nd, dfd, name, root);
2515 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2516 if (unlikely(retval == -ECHILD))
2517 retval = path_lookupat(&nd, flags, path);
2518 if (unlikely(retval == -ESTALE))
2519 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2521 if (likely(!retval))
2522 audit_inode(name, path->dentry,
2523 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2524 restore_nameidata();
2528 /* Returns 0 and nd will be valid on success; Returns error, otherwise. */
2529 static int path_parentat(struct nameidata *nd, unsigned flags,
2530 struct path *parent)
2532 const char *s = path_init(nd, flags);
2533 int err = link_path_walk(s, nd);
2535 err = complete_walk(nd);
2538 nd->path.mnt = NULL;
2539 nd->path.dentry = NULL;
2545 /* Note: this does not consume "name" */
2546 static int __filename_parentat(int dfd, struct filename *name,
2547 unsigned int flags, struct path *parent,
2548 struct qstr *last, int *type,
2549 const struct path *root)
2552 struct nameidata nd;
2555 return PTR_ERR(name);
2556 set_nameidata(&nd, dfd, name, root);
2557 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2558 if (unlikely(retval == -ECHILD))
2559 retval = path_parentat(&nd, flags, parent);
2560 if (unlikely(retval == -ESTALE))
2561 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2562 if (likely(!retval)) {
2564 *type = nd.last_type;
2565 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2567 restore_nameidata();
2571 static int filename_parentat(int dfd, struct filename *name,
2572 unsigned int flags, struct path *parent,
2573 struct qstr *last, int *type)
2575 return __filename_parentat(dfd, name, flags, parent, last, type, NULL);
2578 /* does lookup, returns the object with parent locked */
2579 static struct dentry *__kern_path_locked(int dfd, struct filename *name, struct path *path)
2585 error = filename_parentat(dfd, name, 0, path, &last, &type);
2587 return ERR_PTR(error);
2588 if (unlikely(type != LAST_NORM)) {
2590 return ERR_PTR(-EINVAL);
2592 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2593 d = lookup_one_qstr_excl(&last, path->dentry, 0);
2595 inode_unlock(path->dentry->d_inode);
2601 struct dentry *kern_path_locked(const char *name, struct path *path)
2603 struct filename *filename = getname_kernel(name);
2604 struct dentry *res = __kern_path_locked(AT_FDCWD, filename, path);
2610 struct dentry *user_path_locked_at(int dfd, const char __user *name, struct path *path)
2612 struct filename *filename = getname(name);
2613 struct dentry *res = __kern_path_locked(dfd, filename, path);
2618 EXPORT_SYMBOL(user_path_locked_at);
2620 int kern_path(const char *name, unsigned int flags, struct path *path)
2622 struct filename *filename = getname_kernel(name);
2623 int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2629 EXPORT_SYMBOL(kern_path);
2632 * vfs_path_parent_lookup - lookup a parent path relative to a dentry-vfsmount pair
2633 * @filename: filename structure
2634 * @flags: lookup flags
2635 * @parent: pointer to struct path to fill
2636 * @last: last component
2637 * @type: type of the last component
2638 * @root: pointer to struct path of the base directory
2640 int vfs_path_parent_lookup(struct filename *filename, unsigned int flags,
2641 struct path *parent, struct qstr *last, int *type,
2642 const struct path *root)
2644 return __filename_parentat(AT_FDCWD, filename, flags, parent, last,
2647 EXPORT_SYMBOL(vfs_path_parent_lookup);
2650 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2651 * @dentry: pointer to dentry of the base directory
2652 * @mnt: pointer to vfs mount of the base directory
2653 * @name: pointer to file name
2654 * @flags: lookup flags
2655 * @path: pointer to struct path to fill
2657 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2658 const char *name, unsigned int flags,
2661 struct filename *filename;
2662 struct path root = {.mnt = mnt, .dentry = dentry};
2665 filename = getname_kernel(name);
2666 /* the first argument of filename_lookup() is ignored with root */
2667 ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2671 EXPORT_SYMBOL(vfs_path_lookup);
2673 static int lookup_one_common(struct mnt_idmap *idmap,
2674 const char *name, struct dentry *base, int len,
2679 this->hash = full_name_hash(base, name, len);
2683 if (is_dot_dotdot(name, len))
2687 unsigned int c = *(const unsigned char *)name++;
2688 if (c == '/' || c == '\0')
2692 * See if the low-level filesystem might want
2693 * to use its own hash..
2695 if (base->d_flags & DCACHE_OP_HASH) {
2696 int err = base->d_op->d_hash(base, this);
2701 return inode_permission(idmap, base->d_inode, MAY_EXEC);
2705 * try_lookup_one_len - filesystem helper to lookup single pathname component
2706 * @name: pathname component to lookup
2707 * @base: base directory to lookup from
2708 * @len: maximum length @len should be interpreted to
2710 * Look up a dentry by name in the dcache, returning NULL if it does not
2711 * currently exist. The function does not try to create a dentry.
2713 * Note that this routine is purely a helper for filesystem usage and should
2714 * not be called by generic code.
2716 * The caller must hold base->i_mutex.
2718 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2723 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2725 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2727 return ERR_PTR(err);
2729 return lookup_dcache(&this, base, 0);
2731 EXPORT_SYMBOL(try_lookup_one_len);
2734 * lookup_one_len - filesystem helper to lookup single pathname component
2735 * @name: pathname component to lookup
2736 * @base: base directory to lookup from
2737 * @len: maximum length @len should be interpreted to
2739 * Note that this routine is purely a helper for filesystem usage and should
2740 * not be called by generic code.
2742 * The caller must hold base->i_mutex.
2744 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2746 struct dentry *dentry;
2750 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2752 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2754 return ERR_PTR(err);
2756 dentry = lookup_dcache(&this, base, 0);
2757 return dentry ? dentry : __lookup_slow(&this, base, 0);
2759 EXPORT_SYMBOL(lookup_one_len);
2762 * lookup_one - filesystem helper to lookup single pathname component
2763 * @idmap: idmap of the mount the lookup is performed from
2764 * @name: pathname component to lookup
2765 * @base: base directory to lookup from
2766 * @len: maximum length @len should be interpreted to
2768 * Note that this routine is purely a helper for filesystem usage and should
2769 * not be called by generic code.
2771 * The caller must hold base->i_mutex.
2773 struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name,
2774 struct dentry *base, int len)
2776 struct dentry *dentry;
2780 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2782 err = lookup_one_common(idmap, name, base, len, &this);
2784 return ERR_PTR(err);
2786 dentry = lookup_dcache(&this, base, 0);
2787 return dentry ? dentry : __lookup_slow(&this, base, 0);
2789 EXPORT_SYMBOL(lookup_one);
2792 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2793 * @idmap: idmap of the mount the lookup is performed from
2794 * @name: pathname component to lookup
2795 * @base: base directory to lookup from
2796 * @len: maximum length @len should be interpreted to
2798 * Note that this routine is purely a helper for filesystem usage and should
2799 * not be called by generic code.
2801 * Unlike lookup_one_len, it should be called without the parent
2802 * i_mutex held, and will take the i_mutex itself if necessary.
2804 struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap,
2805 const char *name, struct dentry *base,
2812 err = lookup_one_common(idmap, name, base, len, &this);
2814 return ERR_PTR(err);
2816 ret = lookup_dcache(&this, base, 0);
2818 ret = lookup_slow(&this, base, 0);
2821 EXPORT_SYMBOL(lookup_one_unlocked);
2824 * lookup_one_positive_unlocked - filesystem helper to lookup single
2825 * pathname component
2826 * @idmap: idmap of the mount the lookup is performed from
2827 * @name: pathname component to lookup
2828 * @base: base directory to lookup from
2829 * @len: maximum length @len should be interpreted to
2831 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2832 * known positive or ERR_PTR(). This is what most of the users want.
2834 * Note that pinned negative with unlocked parent _can_ become positive at any
2835 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2836 * positives have >d_inode stable, so this one avoids such problems.
2838 * Note that this routine is purely a helper for filesystem usage and should
2839 * not be called by generic code.
2841 * The helper should be called without i_mutex held.
2843 struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap,
2845 struct dentry *base, int len)
2847 struct dentry *ret = lookup_one_unlocked(idmap, name, base, len);
2849 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2851 ret = ERR_PTR(-ENOENT);
2855 EXPORT_SYMBOL(lookup_one_positive_unlocked);
2858 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2859 * @name: pathname component to lookup
2860 * @base: base directory to lookup from
2861 * @len: maximum length @len should be interpreted to
2863 * Note that this routine is purely a helper for filesystem usage and should
2864 * not be called by generic code.
2866 * Unlike lookup_one_len, it should be called without the parent
2867 * i_mutex held, and will take the i_mutex itself if necessary.
2869 struct dentry *lookup_one_len_unlocked(const char *name,
2870 struct dentry *base, int len)
2872 return lookup_one_unlocked(&nop_mnt_idmap, name, base, len);
2874 EXPORT_SYMBOL(lookup_one_len_unlocked);
2877 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2878 * on negatives. Returns known positive or ERR_PTR(); that's what
2879 * most of the users want. Note that pinned negative with unlocked parent
2880 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2881 * need to be very careful; pinned positives have ->d_inode stable, so
2882 * this one avoids such problems.
2884 struct dentry *lookup_positive_unlocked(const char *name,
2885 struct dentry *base, int len)
2887 return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len);
2889 EXPORT_SYMBOL(lookup_positive_unlocked);
2891 #ifdef CONFIG_UNIX98_PTYS
2892 int path_pts(struct path *path)
2894 /* Find something mounted on "pts" in the same directory as
2897 struct dentry *parent = dget_parent(path->dentry);
2898 struct dentry *child;
2899 struct qstr this = QSTR_INIT("pts", 3);
2901 if (unlikely(!path_connected(path->mnt, parent))) {
2906 path->dentry = parent;
2907 child = d_hash_and_lookup(parent, &this);
2908 if (IS_ERR_OR_NULL(child))
2911 path->dentry = child;
2913 follow_down(path, 0);
2918 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2919 struct path *path, int *empty)
2921 struct filename *filename = getname_flags(name, flags, empty);
2922 int ret = filename_lookup(dfd, filename, flags, path, NULL);
2927 EXPORT_SYMBOL(user_path_at_empty);
2929 int __check_sticky(struct mnt_idmap *idmap, struct inode *dir,
2930 struct inode *inode)
2932 kuid_t fsuid = current_fsuid();
2934 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), fsuid))
2936 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, dir), fsuid))
2938 return !capable_wrt_inode_uidgid(idmap, inode, CAP_FOWNER);
2940 EXPORT_SYMBOL(__check_sticky);
2943 * Check whether we can remove a link victim from directory dir, check
2944 * whether the type of victim is right.
2945 * 1. We can't do it if dir is read-only (done in permission())
2946 * 2. We should have write and exec permissions on dir
2947 * 3. We can't remove anything from append-only dir
2948 * 4. We can't do anything with immutable dir (done in permission())
2949 * 5. If the sticky bit on dir is set we should either
2950 * a. be owner of dir, or
2951 * b. be owner of victim, or
2952 * c. have CAP_FOWNER capability
2953 * 6. If the victim is append-only or immutable we can't do antyhing with
2954 * links pointing to it.
2955 * 7. If the victim has an unknown uid or gid we can't change the inode.
2956 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2957 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2958 * 10. We can't remove a root or mountpoint.
2959 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2960 * nfs_async_unlink().
2962 static int may_delete(struct mnt_idmap *idmap, struct inode *dir,
2963 struct dentry *victim, bool isdir)
2965 struct inode *inode = d_backing_inode(victim);
2968 if (d_is_negative(victim))
2972 BUG_ON(victim->d_parent->d_inode != dir);
2974 /* Inode writeback is not safe when the uid or gid are invalid. */
2975 if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
2976 !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
2979 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2981 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
2987 if (check_sticky(idmap, dir, inode) || IS_APPEND(inode) ||
2988 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2989 HAS_UNMAPPED_ID(idmap, inode))
2992 if (!d_is_dir(victim))
2994 if (IS_ROOT(victim))
2996 } else if (d_is_dir(victim))
2998 if (IS_DEADDIR(dir))
3000 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
3005 /* Check whether we can create an object with dentry child in directory
3007 * 1. We can't do it if child already exists (open has special treatment for
3008 * this case, but since we are inlined it's OK)
3009 * 2. We can't do it if dir is read-only (done in permission())
3010 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
3011 * 4. We should have write and exec permissions on dir
3012 * 5. We can't do it if dir is immutable (done in permission())
3014 static inline int may_create(struct mnt_idmap *idmap,
3015 struct inode *dir, struct dentry *child)
3017 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
3020 if (IS_DEADDIR(dir))
3022 if (!fsuidgid_has_mapping(dir->i_sb, idmap))
3025 return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3028 // p1 != p2, both are on the same filesystem, ->s_vfs_rename_mutex is held
3029 static struct dentry *lock_two_directories(struct dentry *p1, struct dentry *p2)
3031 struct dentry *p = p1, *q = p2, *r;
3033 while ((r = p->d_parent) != p2 && r != p)
3036 // p is a child of p2 and an ancestor of p1 or p1 itself
3037 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3038 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT2);
3041 // p is the root of connected component that contains p1
3042 // p2 does not occur on the path from p to p1
3043 while ((r = q->d_parent) != p1 && r != p && r != q)
3046 // q is a child of p1 and an ancestor of p2 or p2 itself
3047 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3048 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3050 } else if (likely(r == p)) {
3051 // both p2 and p1 are descendents of p
3052 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3053 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3055 } else { // no common ancestor at the time we'd been called
3056 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3057 return ERR_PTR(-EXDEV);
3062 * p1 and p2 should be directories on the same fs.
3064 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
3067 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3071 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
3072 return lock_two_directories(p1, p2);
3074 EXPORT_SYMBOL(lock_rename);
3077 * c1 and p2 should be on the same fs.
3079 struct dentry *lock_rename_child(struct dentry *c1, struct dentry *p2)
3081 if (READ_ONCE(c1->d_parent) == p2) {
3083 * hopefully won't need to touch ->s_vfs_rename_mutex at all.
3085 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3087 * now that p2 is locked, nobody can move in or out of it,
3088 * so the test below is safe.
3090 if (likely(c1->d_parent == p2))
3094 * c1 got moved out of p2 while we'd been taking locks;
3095 * unlock and fall back to slow case.
3097 inode_unlock(p2->d_inode);
3100 mutex_lock(&c1->d_sb->s_vfs_rename_mutex);
3102 * nobody can move out of any directories on this fs.
3104 if (likely(c1->d_parent != p2))
3105 return lock_two_directories(c1->d_parent, p2);
3108 * c1 got moved into p2 while we were taking locks;
3109 * we need p2 locked and ->s_vfs_rename_mutex unlocked,
3110 * for consistency with lock_rename().
3112 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3113 mutex_unlock(&c1->d_sb->s_vfs_rename_mutex);
3116 EXPORT_SYMBOL(lock_rename_child);
3118 void unlock_rename(struct dentry *p1, struct dentry *p2)
3120 inode_unlock(p1->d_inode);
3122 inode_unlock(p2->d_inode);
3123 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3126 EXPORT_SYMBOL(unlock_rename);
3129 * vfs_prepare_mode - prepare the mode to be used for a new inode
3130 * @idmap: idmap of the mount the inode was found from
3131 * @dir: parent directory of the new inode
3132 * @mode: mode of the new inode
3133 * @mask_perms: allowed permission by the vfs
3134 * @type: type of file to be created
3136 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3137 * object to be created.
3139 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3140 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3141 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3142 * POSIX ACL supporting filesystems.
3144 * Note that it's currently valid for @type to be 0 if a directory is created.
3145 * Filesystems raise that flag individually and we need to check whether each
3146 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3149 * Returns: mode to be passed to the filesystem
3151 static inline umode_t vfs_prepare_mode(struct mnt_idmap *idmap,
3152 const struct inode *dir, umode_t mode,
3153 umode_t mask_perms, umode_t type)
3155 mode = mode_strip_sgid(idmap, dir, mode);
3156 mode = mode_strip_umask(dir, mode);
3159 * Apply the vfs mandated allowed permission mask and set the type of
3160 * file to be created before we call into the filesystem.
3162 mode &= (mask_perms & ~S_IFMT);
3163 mode |= (type & S_IFMT);
3169 * vfs_create - create new file
3170 * @idmap: idmap of the mount the inode was found from
3171 * @dir: inode of @dentry
3172 * @dentry: pointer to dentry of the base directory
3173 * @mode: mode of the new file
3174 * @want_excl: whether the file must not yet exist
3176 * Create a new file.
3178 * If the inode has been found through an idmapped mount the idmap of
3179 * the vfsmount must be passed through @idmap. This function will then take
3180 * care to map the inode according to @idmap before checking permissions.
3181 * On non-idmapped mounts or if permission checking is to be performed on the
3182 * raw inode simply pass @nop_mnt_idmap.
3184 int vfs_create(struct mnt_idmap *idmap, struct inode *dir,
3185 struct dentry *dentry, umode_t mode, bool want_excl)
3189 error = may_create(idmap, dir, dentry);
3193 if (!dir->i_op->create)
3194 return -EACCES; /* shouldn't it be ENOSYS? */
3196 mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG);
3197 error = security_inode_create(dir, dentry, mode);
3200 error = dir->i_op->create(idmap, dir, dentry, mode, want_excl);
3202 fsnotify_create(dir, dentry);
3205 EXPORT_SYMBOL(vfs_create);
3207 int vfs_mkobj(struct dentry *dentry, umode_t mode,
3208 int (*f)(struct dentry *, umode_t, void *),
3211 struct inode *dir = dentry->d_parent->d_inode;
3212 int error = may_create(&nop_mnt_idmap, dir, dentry);
3218 error = security_inode_create(dir, dentry, mode);
3221 error = f(dentry, mode, arg);
3223 fsnotify_create(dir, dentry);
3226 EXPORT_SYMBOL(vfs_mkobj);
3228 bool may_open_dev(const struct path *path)
3230 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3231 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3234 static int may_open(struct mnt_idmap *idmap, const struct path *path,
3235 int acc_mode, int flag)
3237 struct dentry *dentry = path->dentry;
3238 struct inode *inode = dentry->d_inode;
3244 switch (inode->i_mode & S_IFMT) {
3248 if (acc_mode & MAY_WRITE)
3250 if (acc_mode & MAY_EXEC)
3255 if (!may_open_dev(path))
3260 if (acc_mode & MAY_EXEC)
3265 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3270 error = inode_permission(idmap, inode, MAY_OPEN | acc_mode);
3275 * An append-only file must be opened in append mode for writing.
3277 if (IS_APPEND(inode)) {
3278 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3284 /* O_NOATIME can only be set by the owner or superuser */
3285 if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode))
3291 static int handle_truncate(struct mnt_idmap *idmap, struct file *filp)
3293 const struct path *path = &filp->f_path;
3294 struct inode *inode = path->dentry->d_inode;
3295 int error = get_write_access(inode);
3299 error = security_file_truncate(filp);
3301 error = do_truncate(idmap, path->dentry, 0,
3302 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3305 put_write_access(inode);
3309 static inline int open_to_namei_flags(int flag)
3311 if ((flag & O_ACCMODE) == 3)
3316 static int may_o_create(struct mnt_idmap *idmap,
3317 const struct path *dir, struct dentry *dentry,
3320 int error = security_path_mknod(dir, dentry, mode, 0);
3324 if (!fsuidgid_has_mapping(dir->dentry->d_sb, idmap))
3327 error = inode_permission(idmap, dir->dentry->d_inode,
3328 MAY_WRITE | MAY_EXEC);
3332 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3336 * Attempt to atomically look up, create and open a file from a negative
3339 * Returns 0 if successful. The file will have been created and attached to
3340 * @file by the filesystem calling finish_open().
3342 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3343 * be set. The caller will need to perform the open themselves. @path will
3344 * have been updated to point to the new dentry. This may be negative.
3346 * Returns an error code otherwise.
3348 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3350 int open_flag, umode_t mode)
3352 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3353 struct inode *dir = nd->path.dentry->d_inode;
3356 if (nd->flags & LOOKUP_DIRECTORY)
3357 open_flag |= O_DIRECTORY;
3359 file->f_path.dentry = DENTRY_NOT_SET;
3360 file->f_path.mnt = nd->path.mnt;
3361 error = dir->i_op->atomic_open(dir, dentry, file,
3362 open_to_namei_flags(open_flag), mode);
3363 d_lookup_done(dentry);
3365 if (file->f_mode & FMODE_OPENED) {
3366 if (unlikely(dentry != file->f_path.dentry)) {
3368 dentry = dget(file->f_path.dentry);
3370 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3373 if (file->f_path.dentry) {
3375 dentry = file->f_path.dentry;
3377 if (unlikely(d_is_negative(dentry)))
3383 dentry = ERR_PTR(error);
3389 * Look up and maybe create and open the last component.
3391 * Must be called with parent locked (exclusive in O_CREAT case).
3393 * Returns 0 on success, that is, if
3394 * the file was successfully atomically created (if necessary) and opened, or
3395 * the file was not completely opened at this time, though lookups and
3396 * creations were performed.
3397 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3398 * In the latter case dentry returned in @path might be negative if O_CREAT
3399 * hadn't been specified.
3401 * An error code is returned on failure.
3403 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3404 const struct open_flags *op,
3407 struct mnt_idmap *idmap;
3408 struct dentry *dir = nd->path.dentry;
3409 struct inode *dir_inode = dir->d_inode;
3410 int open_flag = op->open_flag;
3411 struct dentry *dentry;
3412 int error, create_error = 0;
3413 umode_t mode = op->mode;
3414 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3416 if (unlikely(IS_DEADDIR(dir_inode)))
3417 return ERR_PTR(-ENOENT);
3419 file->f_mode &= ~FMODE_CREATED;
3420 dentry = d_lookup(dir, &nd->last);
3423 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3427 if (d_in_lookup(dentry))
3430 error = d_revalidate(dentry, nd->flags);
3431 if (likely(error > 0))
3435 d_invalidate(dentry);
3439 if (dentry->d_inode) {
3440 /* Cached positive dentry: will open in f_op->open */
3445 * Checking write permission is tricky, bacuse we don't know if we are
3446 * going to actually need it: O_CREAT opens should work as long as the
3447 * file exists. But checking existence breaks atomicity. The trick is
3448 * to check access and if not granted clear O_CREAT from the flags.
3450 * Another problem is returing the "right" error value (e.g. for an
3451 * O_EXCL open we want to return EEXIST not EROFS).
3453 if (unlikely(!got_write))
3454 open_flag &= ~O_TRUNC;
3455 idmap = mnt_idmap(nd->path.mnt);
3456 if (open_flag & O_CREAT) {
3457 if (open_flag & O_EXCL)
3458 open_flag &= ~O_TRUNC;
3459 mode = vfs_prepare_mode(idmap, dir->d_inode, mode, mode, mode);
3460 if (likely(got_write))
3461 create_error = may_o_create(idmap, &nd->path,
3464 create_error = -EROFS;
3467 open_flag &= ~O_CREAT;
3468 if (dir_inode->i_op->atomic_open) {
3469 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3470 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3471 dentry = ERR_PTR(create_error);
3475 if (d_in_lookup(dentry)) {
3476 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3478 d_lookup_done(dentry);
3479 if (unlikely(res)) {
3481 error = PTR_ERR(res);
3489 /* Negative dentry, just create the file */
3490 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3491 file->f_mode |= FMODE_CREATED;
3492 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3493 if (!dir_inode->i_op->create) {
3498 error = dir_inode->i_op->create(idmap, dir_inode, dentry,
3499 mode, open_flag & O_EXCL);
3503 if (unlikely(create_error) && !dentry->d_inode) {
3504 error = create_error;
3511 return ERR_PTR(error);
3514 static const char *open_last_lookups(struct nameidata *nd,
3515 struct file *file, const struct open_flags *op)
3517 struct dentry *dir = nd->path.dentry;
3518 int open_flag = op->open_flag;
3519 bool got_write = false;
3520 struct dentry *dentry;
3523 nd->flags |= op->intent;
3525 if (nd->last_type != LAST_NORM) {
3528 return handle_dots(nd, nd->last_type);
3531 if (!(open_flag & O_CREAT)) {
3532 if (nd->last.name[nd->last.len])
3533 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3534 /* we _can_ be in RCU mode here */
3535 dentry = lookup_fast(nd);
3537 return ERR_CAST(dentry);
3541 if (WARN_ON_ONCE(nd->flags & LOOKUP_RCU))
3542 return ERR_PTR(-ECHILD);
3544 /* create side of things */
3545 if (nd->flags & LOOKUP_RCU) {
3546 if (!try_to_unlazy(nd))
3547 return ERR_PTR(-ECHILD);
3549 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3550 /* trailing slashes? */
3551 if (unlikely(nd->last.name[nd->last.len]))
3552 return ERR_PTR(-EISDIR);
3555 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3556 got_write = !mnt_want_write(nd->path.mnt);
3558 * do _not_ fail yet - we might not need that or fail with
3559 * a different error; let lookup_open() decide; we'll be
3560 * dropping this one anyway.
3563 if (open_flag & O_CREAT)
3564 inode_lock(dir->d_inode);
3566 inode_lock_shared(dir->d_inode);
3567 dentry = lookup_open(nd, file, op, got_write);
3568 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3569 fsnotify_create(dir->d_inode, dentry);
3570 if (open_flag & O_CREAT)
3571 inode_unlock(dir->d_inode);
3573 inode_unlock_shared(dir->d_inode);
3576 mnt_drop_write(nd->path.mnt);
3579 return ERR_CAST(dentry);
3581 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3582 dput(nd->path.dentry);
3583 nd->path.dentry = dentry;
3590 res = step_into(nd, WALK_TRAILING, dentry);
3592 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3597 * Handle the last step of open()
3599 static int do_open(struct nameidata *nd,
3600 struct file *file, const struct open_flags *op)
3602 struct mnt_idmap *idmap;
3603 int open_flag = op->open_flag;
3608 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3609 error = complete_walk(nd);
3613 if (!(file->f_mode & FMODE_CREATED))
3614 audit_inode(nd->name, nd->path.dentry, 0);
3615 idmap = mnt_idmap(nd->path.mnt);
3616 if (open_flag & O_CREAT) {
3617 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3619 if (d_is_dir(nd->path.dentry))
3621 error = may_create_in_sticky(idmap, nd,
3622 d_backing_inode(nd->path.dentry));
3623 if (unlikely(error))
3626 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3629 do_truncate = false;
3630 acc_mode = op->acc_mode;
3631 if (file->f_mode & FMODE_CREATED) {
3632 /* Don't check for write permission, don't truncate */
3633 open_flag &= ~O_TRUNC;
3635 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3636 error = mnt_want_write(nd->path.mnt);
3641 error = may_open(idmap, &nd->path, acc_mode, open_flag);
3642 if (!error && !(file->f_mode & FMODE_OPENED))
3643 error = vfs_open(&nd->path, file);
3645 error = ima_file_check(file, op->acc_mode);
3646 if (!error && do_truncate)
3647 error = handle_truncate(idmap, file);
3648 if (unlikely(error > 0)) {
3653 mnt_drop_write(nd->path.mnt);
3658 * vfs_tmpfile - create tmpfile
3659 * @idmap: idmap of the mount the inode was found from
3660 * @parentpath: pointer to the path of the base directory
3661 * @file: file descriptor of the new tmpfile
3662 * @mode: mode of the new tmpfile
3664 * Create a temporary file.
3666 * If the inode has been found through an idmapped mount the idmap of
3667 * the vfsmount must be passed through @idmap. This function will then take
3668 * care to map the inode according to @idmap before checking permissions.
3669 * On non-idmapped mounts or if permission checking is to be performed on the
3670 * raw inode simply pass @nop_mnt_idmap.
3672 static int vfs_tmpfile(struct mnt_idmap *idmap,
3673 const struct path *parentpath,
3674 struct file *file, umode_t mode)
3676 struct dentry *child;
3677 struct inode *dir = d_inode(parentpath->dentry);
3678 struct inode *inode;
3680 int open_flag = file->f_flags;
3682 /* we want directory to be writable */
3683 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3686 if (!dir->i_op->tmpfile)
3688 child = d_alloc(parentpath->dentry, &slash_name);
3689 if (unlikely(!child))
3691 file->f_path.mnt = parentpath->mnt;
3692 file->f_path.dentry = child;
3693 mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3694 error = dir->i_op->tmpfile(idmap, dir, file, mode);
3698 /* Don't check for other permissions, the inode was just created */
3699 error = may_open(idmap, &file->f_path, 0, file->f_flags);
3702 inode = file_inode(file);
3703 if (!(open_flag & O_EXCL)) {
3704 spin_lock(&inode->i_lock);
3705 inode->i_state |= I_LINKABLE;
3706 spin_unlock(&inode->i_lock);
3708 ima_post_create_tmpfile(idmap, inode);
3713 * kernel_tmpfile_open - open a tmpfile for kernel internal use
3714 * @idmap: idmap of the mount the inode was found from
3715 * @parentpath: path of the base directory
3716 * @mode: mode of the new tmpfile
3718 * @cred: credentials for open
3720 * Create and open a temporary file. The file is not accounted in nr_files,
3721 * hence this is only for kernel internal use, and must not be installed into
3722 * file tables or such.
3724 struct file *kernel_tmpfile_open(struct mnt_idmap *idmap,
3725 const struct path *parentpath,
3726 umode_t mode, int open_flag,
3727 const struct cred *cred)
3732 file = alloc_empty_file_noaccount(open_flag, cred);
3736 error = vfs_tmpfile(idmap, parentpath, file, mode);
3739 file = ERR_PTR(error);
3743 EXPORT_SYMBOL(kernel_tmpfile_open);
3745 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3746 const struct open_flags *op,
3750 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3752 if (unlikely(error))
3754 error = mnt_want_write(path.mnt);
3755 if (unlikely(error))
3757 error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode);
3760 audit_inode(nd->name, file->f_path.dentry, 0);
3762 mnt_drop_write(path.mnt);
3768 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3771 int error = path_lookupat(nd, flags, &path);
3773 audit_inode(nd->name, path.dentry, 0);
3774 error = vfs_open(&path, file);
3780 static struct file *path_openat(struct nameidata *nd,
3781 const struct open_flags *op, unsigned flags)
3786 file = alloc_empty_file(op->open_flag, current_cred());
3790 if (unlikely(file->f_flags & __O_TMPFILE)) {
3791 error = do_tmpfile(nd, flags, op, file);
3792 } else if (unlikely(file->f_flags & O_PATH)) {
3793 error = do_o_path(nd, flags, file);
3795 const char *s = path_init(nd, flags);
3796 while (!(error = link_path_walk(s, nd)) &&
3797 (s = open_last_lookups(nd, file, op)) != NULL)
3800 error = do_open(nd, file, op);
3803 if (likely(!error)) {
3804 if (likely(file->f_mode & FMODE_OPENED))
3810 if (error == -EOPENSTALE) {
3811 if (flags & LOOKUP_RCU)
3816 return ERR_PTR(error);
3819 struct file *do_filp_open(int dfd, struct filename *pathname,
3820 const struct open_flags *op)
3822 struct nameidata nd;
3823 int flags = op->lookup_flags;
3826 set_nameidata(&nd, dfd, pathname, NULL);
3827 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3828 if (unlikely(filp == ERR_PTR(-ECHILD)))
3829 filp = path_openat(&nd, op, flags);
3830 if (unlikely(filp == ERR_PTR(-ESTALE)))
3831 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3832 restore_nameidata();
3836 struct file *do_file_open_root(const struct path *root,
3837 const char *name, const struct open_flags *op)
3839 struct nameidata nd;
3841 struct filename *filename;
3842 int flags = op->lookup_flags;
3844 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3845 return ERR_PTR(-ELOOP);
3847 filename = getname_kernel(name);
3848 if (IS_ERR(filename))
3849 return ERR_CAST(filename);
3851 set_nameidata(&nd, -1, filename, root);
3852 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3853 if (unlikely(file == ERR_PTR(-ECHILD)))
3854 file = path_openat(&nd, op, flags);
3855 if (unlikely(file == ERR_PTR(-ESTALE)))
3856 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3857 restore_nameidata();
3862 static struct dentry *filename_create(int dfd, struct filename *name,
3863 struct path *path, unsigned int lookup_flags)
3865 struct dentry *dentry = ERR_PTR(-EEXIST);
3867 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3868 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3869 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3874 error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3876 return ERR_PTR(error);
3879 * Yucky last component or no last component at all?
3880 * (foo/., foo/.., /////)
3882 if (unlikely(type != LAST_NORM))
3885 /* don't fail immediately if it's r/o, at least try to report other errors */
3886 err2 = mnt_want_write(path->mnt);
3888 * Do the final lookup. Suppress 'create' if there is a trailing
3889 * '/', and a directory wasn't requested.
3891 if (last.name[last.len] && !want_dir)
3893 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3894 dentry = lookup_one_qstr_excl(&last, path->dentry,
3895 reval_flag | create_flags);
3900 if (d_is_positive(dentry))
3904 * Special case - lookup gave negative, but... we had foo/bar/
3905 * From the vfs_mknod() POV we just have a negative dentry -
3906 * all is fine. Let's be bastards - you had / on the end, you've
3907 * been asking for (non-existent) directory. -ENOENT for you.
3909 if (unlikely(!create_flags)) {
3913 if (unlikely(err2)) {
3920 dentry = ERR_PTR(error);
3922 inode_unlock(path->dentry->d_inode);
3924 mnt_drop_write(path->mnt);
3930 struct dentry *kern_path_create(int dfd, const char *pathname,
3931 struct path *path, unsigned int lookup_flags)
3933 struct filename *filename = getname_kernel(pathname);
3934 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3939 EXPORT_SYMBOL(kern_path_create);
3941 void done_path_create(struct path *path, struct dentry *dentry)
3944 inode_unlock(path->dentry->d_inode);
3945 mnt_drop_write(path->mnt);
3948 EXPORT_SYMBOL(done_path_create);
3950 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3951 struct path *path, unsigned int lookup_flags)
3953 struct filename *filename = getname(pathname);
3954 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3959 EXPORT_SYMBOL(user_path_create);
3962 * vfs_mknod - create device node or file
3963 * @idmap: idmap of the mount the inode was found from
3964 * @dir: inode of @dentry
3965 * @dentry: pointer to dentry of the base directory
3966 * @mode: mode of the new device node or file
3967 * @dev: device number of device to create
3969 * Create a device node or file.
3971 * If the inode has been found through an idmapped mount the idmap of
3972 * the vfsmount must be passed through @idmap. This function will then take
3973 * care to map the inode according to @idmap before checking permissions.
3974 * On non-idmapped mounts or if permission checking is to be performed on the
3975 * raw inode simply pass @nop_mnt_idmap.
3977 int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
3978 struct dentry *dentry, umode_t mode, dev_t dev)
3980 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3981 int error = may_create(idmap, dir, dentry);
3986 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3987 !capable(CAP_MKNOD))
3990 if (!dir->i_op->mknod)
3993 mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3994 error = devcgroup_inode_mknod(mode, dev);
3998 error = security_inode_mknod(dir, dentry, mode, dev);
4002 error = dir->i_op->mknod(idmap, dir, dentry, mode, dev);
4004 fsnotify_create(dir, dentry);
4007 EXPORT_SYMBOL(vfs_mknod);
4009 static int may_mknod(umode_t mode)
4011 switch (mode & S_IFMT) {
4017 case 0: /* zero mode translates to S_IFREG */
4026 static int do_mknodat(int dfd, struct filename *name, umode_t mode,
4029 struct mnt_idmap *idmap;
4030 struct dentry *dentry;
4033 unsigned int lookup_flags = 0;
4035 error = may_mknod(mode);
4039 dentry = filename_create(dfd, name, &path, lookup_flags);
4040 error = PTR_ERR(dentry);
4044 error = security_path_mknod(&path, dentry,
4045 mode_strip_umask(path.dentry->d_inode, mode), dev);
4049 idmap = mnt_idmap(path.mnt);
4050 switch (mode & S_IFMT) {
4051 case 0: case S_IFREG:
4052 error = vfs_create(idmap, path.dentry->d_inode,
4053 dentry, mode, true);
4055 ima_post_path_mknod(idmap, dentry);
4057 case S_IFCHR: case S_IFBLK:
4058 error = vfs_mknod(idmap, path.dentry->d_inode,
4059 dentry, mode, new_decode_dev(dev));
4061 case S_IFIFO: case S_IFSOCK:
4062 error = vfs_mknod(idmap, path.dentry->d_inode,
4067 done_path_create(&path, dentry);
4068 if (retry_estale(error, lookup_flags)) {
4069 lookup_flags |= LOOKUP_REVAL;
4077 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
4080 return do_mknodat(dfd, getname(filename), mode, dev);
4083 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
4085 return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
4089 * vfs_mkdir - create directory
4090 * @idmap: idmap of the mount the inode was found from
4091 * @dir: inode of @dentry
4092 * @dentry: pointer to dentry of the base directory
4093 * @mode: mode of the new directory
4095 * Create a directory.
4097 * If the inode has been found through an idmapped mount the idmap of
4098 * the vfsmount must be passed through @idmap. This function will then take
4099 * care to map the inode according to @idmap before checking permissions.
4100 * On non-idmapped mounts or if permission checking is to be performed on the
4101 * raw inode simply pass @nop_mnt_idmap.
4103 int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
4104 struct dentry *dentry, umode_t mode)
4107 unsigned max_links = dir->i_sb->s_max_links;
4109 error = may_create(idmap, dir, dentry);
4113 if (!dir->i_op->mkdir)
4116 mode = vfs_prepare_mode(idmap, dir, mode, S_IRWXUGO | S_ISVTX, 0);
4117 error = security_inode_mkdir(dir, dentry, mode);
4121 if (max_links && dir->i_nlink >= max_links)
4124 error = dir->i_op->mkdir(idmap, dir, dentry, mode);
4126 fsnotify_mkdir(dir, dentry);
4129 EXPORT_SYMBOL(vfs_mkdir);
4131 int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4133 struct dentry *dentry;
4136 unsigned int lookup_flags = LOOKUP_DIRECTORY;
4139 dentry = filename_create(dfd, name, &path, lookup_flags);
4140 error = PTR_ERR(dentry);
4144 error = security_path_mkdir(&path, dentry,
4145 mode_strip_umask(path.dentry->d_inode, mode));
4147 error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode,
4150 done_path_create(&path, dentry);
4151 if (retry_estale(error, lookup_flags)) {
4152 lookup_flags |= LOOKUP_REVAL;
4160 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4162 return do_mkdirat(dfd, getname(pathname), mode);
4165 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4167 return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4171 * vfs_rmdir - remove directory
4172 * @idmap: idmap of the mount the inode was found from
4173 * @dir: inode of @dentry
4174 * @dentry: pointer to dentry of the base directory
4176 * Remove a directory.
4178 * If the inode has been found through an idmapped mount the idmap of
4179 * the vfsmount must be passed through @idmap. This function will then take
4180 * care to map the inode according to @idmap before checking permissions.
4181 * On non-idmapped mounts or if permission checking is to be performed on the
4182 * raw inode simply pass @nop_mnt_idmap.
4184 int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir,
4185 struct dentry *dentry)
4187 int error = may_delete(idmap, dir, dentry, 1);
4192 if (!dir->i_op->rmdir)
4196 inode_lock(dentry->d_inode);
4199 if (is_local_mountpoint(dentry) ||
4200 (dentry->d_inode->i_flags & S_KERNEL_FILE))
4203 error = security_inode_rmdir(dir, dentry);
4207 error = dir->i_op->rmdir(dir, dentry);
4211 shrink_dcache_parent(dentry);
4212 dentry->d_inode->i_flags |= S_DEAD;
4214 detach_mounts(dentry);
4217 inode_unlock(dentry->d_inode);
4220 d_delete_notify(dir, dentry);
4223 EXPORT_SYMBOL(vfs_rmdir);
4225 int do_rmdir(int dfd, struct filename *name)
4228 struct dentry *dentry;
4232 unsigned int lookup_flags = 0;
4234 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4250 error = mnt_want_write(path.mnt);
4254 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4255 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4256 error = PTR_ERR(dentry);
4259 if (!dentry->d_inode) {
4263 error = security_path_rmdir(&path, dentry);
4266 error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry);
4270 inode_unlock(path.dentry->d_inode);
4271 mnt_drop_write(path.mnt);
4274 if (retry_estale(error, lookup_flags)) {
4275 lookup_flags |= LOOKUP_REVAL;
4283 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4285 return do_rmdir(AT_FDCWD, getname(pathname));
4289 * vfs_unlink - unlink a filesystem object
4290 * @idmap: idmap of the mount the inode was found from
4291 * @dir: parent directory
4293 * @delegated_inode: returns victim inode, if the inode is delegated.
4295 * The caller must hold dir->i_mutex.
4297 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4298 * return a reference to the inode in delegated_inode. The caller
4299 * should then break the delegation on that inode and retry. Because
4300 * breaking a delegation may take a long time, the caller should drop
4301 * dir->i_mutex before doing so.
4303 * Alternatively, a caller may pass NULL for delegated_inode. This may
4304 * be appropriate for callers that expect the underlying filesystem not
4305 * to be NFS exported.
4307 * If the inode has been found through an idmapped mount the idmap of
4308 * the vfsmount must be passed through @idmap. This function will then take
4309 * care to map the inode according to @idmap before checking permissions.
4310 * On non-idmapped mounts or if permission checking is to be performed on the
4311 * raw inode simply pass @nop_mnt_idmap.
4313 int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir,
4314 struct dentry *dentry, struct inode **delegated_inode)
4316 struct inode *target = dentry->d_inode;
4317 int error = may_delete(idmap, dir, dentry, 0);
4322 if (!dir->i_op->unlink)
4326 if (IS_SWAPFILE(target))
4328 else if (is_local_mountpoint(dentry))
4331 error = security_inode_unlink(dir, dentry);
4333 error = try_break_deleg(target, delegated_inode);
4336 error = dir->i_op->unlink(dir, dentry);
4339 detach_mounts(dentry);
4344 inode_unlock(target);
4346 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4347 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4348 fsnotify_unlink(dir, dentry);
4349 } else if (!error) {
4350 fsnotify_link_count(target);
4351 d_delete_notify(dir, dentry);
4356 EXPORT_SYMBOL(vfs_unlink);
4359 * Make sure that the actual truncation of the file will occur outside its
4360 * directory's i_mutex. Truncate can take a long time if there is a lot of
4361 * writeout happening, and we don't want to prevent access to the directory
4362 * while waiting on the I/O.
4364 int do_unlinkat(int dfd, struct filename *name)
4367 struct dentry *dentry;
4371 struct inode *inode = NULL;
4372 struct inode *delegated_inode = NULL;
4373 unsigned int lookup_flags = 0;
4375 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4380 if (type != LAST_NORM)
4383 error = mnt_want_write(path.mnt);
4387 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4388 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4389 error = PTR_ERR(dentry);
4390 if (!IS_ERR(dentry)) {
4392 /* Why not before? Because we want correct error value */
4393 if (last.name[last.len] || d_is_negative(dentry))
4395 inode = dentry->d_inode;
4397 error = security_path_unlink(&path, dentry);
4400 error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4401 dentry, &delegated_inode);
4405 inode_unlock(path.dentry->d_inode);
4407 iput(inode); /* truncate the inode here */
4409 if (delegated_inode) {
4410 error = break_deleg_wait(&delegated_inode);
4414 mnt_drop_write(path.mnt);
4417 if (retry_estale(error, lookup_flags)) {
4418 lookup_flags |= LOOKUP_REVAL;
4427 if (d_is_negative(dentry))
4429 else if (d_is_dir(dentry))
4436 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4438 if ((flag & ~AT_REMOVEDIR) != 0)
4441 if (flag & AT_REMOVEDIR)
4442 return do_rmdir(dfd, getname(pathname));
4443 return do_unlinkat(dfd, getname(pathname));
4446 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4448 return do_unlinkat(AT_FDCWD, getname(pathname));
4452 * vfs_symlink - create symlink
4453 * @idmap: idmap of the mount the inode was found from
4454 * @dir: inode of @dentry
4455 * @dentry: pointer to dentry of the base directory
4456 * @oldname: name of the file to link to
4460 * If the inode has been found through an idmapped mount the idmap of
4461 * the vfsmount must be passed through @idmap. This function will then take
4462 * care to map the inode according to @idmap before checking permissions.
4463 * On non-idmapped mounts or if permission checking is to be performed on the
4464 * raw inode simply pass @nop_mnt_idmap.
4466 int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
4467 struct dentry *dentry, const char *oldname)
4471 error = may_create(idmap, dir, dentry);
4475 if (!dir->i_op->symlink)
4478 error = security_inode_symlink(dir, dentry, oldname);
4482 error = dir->i_op->symlink(idmap, dir, dentry, oldname);
4484 fsnotify_create(dir, dentry);
4487 EXPORT_SYMBOL(vfs_symlink);
4489 int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4492 struct dentry *dentry;
4494 unsigned int lookup_flags = 0;
4497 error = PTR_ERR(from);
4501 dentry = filename_create(newdfd, to, &path, lookup_flags);
4502 error = PTR_ERR(dentry);
4506 error = security_path_symlink(&path, dentry, from->name);
4508 error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4509 dentry, from->name);
4510 done_path_create(&path, dentry);
4511 if (retry_estale(error, lookup_flags)) {
4512 lookup_flags |= LOOKUP_REVAL;
4521 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4522 int, newdfd, const char __user *, newname)
4524 return do_symlinkat(getname(oldname), newdfd, getname(newname));
4527 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4529 return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4533 * vfs_link - create a new link
4534 * @old_dentry: object to be linked
4535 * @idmap: idmap of the mount
4537 * @new_dentry: where to create the new link
4538 * @delegated_inode: returns inode needing a delegation break
4540 * The caller must hold dir->i_mutex
4542 * If vfs_link discovers a delegation on the to-be-linked file in need
4543 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4544 * inode in delegated_inode. The caller should then break the delegation
4545 * and retry. Because breaking a delegation may take a long time, the
4546 * caller should drop the i_mutex before doing so.
4548 * Alternatively, a caller may pass NULL for delegated_inode. This may
4549 * be appropriate for callers that expect the underlying filesystem not
4550 * to be NFS exported.
4552 * If the inode has been found through an idmapped mount the idmap of
4553 * the vfsmount must be passed through @idmap. This function will then take
4554 * care to map the inode according to @idmap before checking permissions.
4555 * On non-idmapped mounts or if permission checking is to be performed on the
4556 * raw inode simply pass @nop_mnt_idmap.
4558 int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap,
4559 struct inode *dir, struct dentry *new_dentry,
4560 struct inode **delegated_inode)
4562 struct inode *inode = old_dentry->d_inode;
4563 unsigned max_links = dir->i_sb->s_max_links;
4569 error = may_create(idmap, dir, new_dentry);
4573 if (dir->i_sb != inode->i_sb)
4577 * A link to an append-only or immutable file cannot be created.
4579 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4582 * Updating the link count will likely cause i_uid and i_gid to
4583 * be writen back improperly if their true value is unknown to
4586 if (HAS_UNMAPPED_ID(idmap, inode))
4588 if (!dir->i_op->link)
4590 if (S_ISDIR(inode->i_mode))
4593 error = security_inode_link(old_dentry, dir, new_dentry);
4598 /* Make sure we don't allow creating hardlink to an unlinked file */
4599 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4601 else if (max_links && inode->i_nlink >= max_links)
4604 error = try_break_deleg(inode, delegated_inode);
4606 error = dir->i_op->link(old_dentry, dir, new_dentry);
4609 if (!error && (inode->i_state & I_LINKABLE)) {
4610 spin_lock(&inode->i_lock);
4611 inode->i_state &= ~I_LINKABLE;
4612 spin_unlock(&inode->i_lock);
4614 inode_unlock(inode);
4616 fsnotify_link(dir, inode, new_dentry);
4619 EXPORT_SYMBOL(vfs_link);
4622 * Hardlinks are often used in delicate situations. We avoid
4623 * security-related surprises by not following symlinks on the
4626 * We don't follow them on the oldname either to be compatible
4627 * with linux 2.0, and to avoid hard-linking to directories
4628 * and other special files. --ADM
4630 int do_linkat(int olddfd, struct filename *old, int newdfd,
4631 struct filename *new, int flags)
4633 struct mnt_idmap *idmap;
4634 struct dentry *new_dentry;
4635 struct path old_path, new_path;
4636 struct inode *delegated_inode = NULL;
4640 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4645 * To use null names we require CAP_DAC_READ_SEARCH
4646 * This ensures that not everyone will be able to create
4647 * handlink using the passed filedescriptor.
4649 if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4654 if (flags & AT_SYMLINK_FOLLOW)
4655 how |= LOOKUP_FOLLOW;
4657 error = filename_lookup(olddfd, old, how, &old_path, NULL);
4661 new_dentry = filename_create(newdfd, new, &new_path,
4662 (how & LOOKUP_REVAL));
4663 error = PTR_ERR(new_dentry);
4664 if (IS_ERR(new_dentry))
4668 if (old_path.mnt != new_path.mnt)
4670 idmap = mnt_idmap(new_path.mnt);
4671 error = may_linkat(idmap, &old_path);
4672 if (unlikely(error))
4674 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4677 error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode,
4678 new_dentry, &delegated_inode);
4680 done_path_create(&new_path, new_dentry);
4681 if (delegated_inode) {
4682 error = break_deleg_wait(&delegated_inode);
4684 path_put(&old_path);
4688 if (retry_estale(error, how)) {
4689 path_put(&old_path);
4690 how |= LOOKUP_REVAL;
4694 path_put(&old_path);
4702 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4703 int, newdfd, const char __user *, newname, int, flags)
4705 return do_linkat(olddfd, getname_uflags(oldname, flags),
4706 newdfd, getname(newname), flags);
4709 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4711 return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4715 * vfs_rename - rename a filesystem object
4716 * @rd: pointer to &struct renamedata info
4718 * The caller must hold multiple mutexes--see lock_rename()).
4720 * If vfs_rename discovers a delegation in need of breaking at either
4721 * the source or destination, it will return -EWOULDBLOCK and return a
4722 * reference to the inode in delegated_inode. The caller should then
4723 * break the delegation and retry. Because breaking a delegation may
4724 * take a long time, the caller should drop all locks before doing
4727 * Alternatively, a caller may pass NULL for delegated_inode. This may
4728 * be appropriate for callers that expect the underlying filesystem not
4729 * to be NFS exported.
4731 * The worst of all namespace operations - renaming directory. "Perverted"
4732 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4735 * a) we can get into loop creation.
4736 * b) race potential - two innocent renames can create a loop together.
4737 * That's where 4.4BSD screws up. Current fix: serialization on
4738 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4740 * c) we may have to lock up to _four_ objects - parents and victim (if it exists),
4741 * and source (if it's a non-directory or a subdirectory that moves to
4742 * different parent).
4743 * And that - after we got ->i_mutex on parents (until then we don't know
4744 * whether the target exists). Solution: try to be smart with locking
4745 * order for inodes. We rely on the fact that tree topology may change
4746 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4747 * move will be locked. Thus we can rank directories by the tree
4748 * (ancestors first) and rank all non-directories after them.
4749 * That works since everybody except rename does "lock parent, lookup,
4750 * lock child" and rename is under ->s_vfs_rename_mutex.
4751 * HOWEVER, it relies on the assumption that any object with ->lookup()
4752 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4753 * we'd better make sure that there's no link(2) for them.
4754 * d) conversion from fhandle to dentry may come in the wrong moment - when
4755 * we are removing the target. Solution: we will have to grab ->i_mutex
4756 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4757 * ->i_mutex on parents, which works but leads to some truly excessive
4760 int vfs_rename(struct renamedata *rd)
4763 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4764 struct dentry *old_dentry = rd->old_dentry;
4765 struct dentry *new_dentry = rd->new_dentry;
4766 struct inode **delegated_inode = rd->delegated_inode;
4767 unsigned int flags = rd->flags;
4768 bool is_dir = d_is_dir(old_dentry);
4769 struct inode *source = old_dentry->d_inode;
4770 struct inode *target = new_dentry->d_inode;
4771 bool new_is_dir = false;
4772 unsigned max_links = new_dir->i_sb->s_max_links;
4773 struct name_snapshot old_name;
4774 bool lock_old_subdir, lock_new_subdir;
4776 if (source == target)
4779 error = may_delete(rd->old_mnt_idmap, old_dir, old_dentry, is_dir);
4784 error = may_create(rd->new_mnt_idmap, new_dir, new_dentry);
4786 new_is_dir = d_is_dir(new_dentry);
4788 if (!(flags & RENAME_EXCHANGE))
4789 error = may_delete(rd->new_mnt_idmap, new_dir,
4790 new_dentry, is_dir);
4792 error = may_delete(rd->new_mnt_idmap, new_dir,
4793 new_dentry, new_is_dir);
4798 if (!old_dir->i_op->rename)
4802 * If we are going to change the parent - check write permissions,
4803 * we'll need to flip '..'.
4805 if (new_dir != old_dir) {
4807 error = inode_permission(rd->old_mnt_idmap, source,
4812 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4813 error = inode_permission(rd->new_mnt_idmap, target,
4820 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4825 take_dentry_name_snapshot(&old_name, old_dentry);
4829 * The source subdirectory needs to be locked on cross-directory
4830 * rename or cross-directory exchange since its parent changes.
4831 * The target subdirectory needs to be locked on cross-directory
4832 * exchange due to parent change and on any rename due to becoming
4834 * Non-directories need locking in all cases (for NFS reasons);
4835 * they get locked after any subdirectories (in inode address order).
4837 * NOTE: WE ONLY LOCK UNRELATED DIRECTORIES IN CROSS-DIRECTORY CASE.
4838 * NEVER, EVER DO THAT WITHOUT ->s_vfs_rename_mutex.
4840 lock_old_subdir = new_dir != old_dir;
4841 lock_new_subdir = new_dir != old_dir || !(flags & RENAME_EXCHANGE);
4843 if (lock_old_subdir)
4844 inode_lock_nested(source, I_MUTEX_CHILD);
4845 if (target && (!new_is_dir || lock_new_subdir))
4847 } else if (new_is_dir) {
4848 if (lock_new_subdir)
4849 inode_lock_nested(target, I_MUTEX_CHILD);
4852 lock_two_nondirectories(source, target);
4856 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4860 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4863 if (max_links && new_dir != old_dir) {
4865 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4867 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4868 old_dir->i_nlink >= max_links)
4872 error = try_break_deleg(source, delegated_inode);
4876 if (target && !new_is_dir) {
4877 error = try_break_deleg(target, delegated_inode);
4881 error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry,
4882 new_dir, new_dentry, flags);
4886 if (!(flags & RENAME_EXCHANGE) && target) {
4888 shrink_dcache_parent(new_dentry);
4889 target->i_flags |= S_DEAD;
4891 dont_mount(new_dentry);
4892 detach_mounts(new_dentry);
4894 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4895 if (!(flags & RENAME_EXCHANGE))
4896 d_move(old_dentry, new_dentry);
4898 d_exchange(old_dentry, new_dentry);
4901 if (!is_dir || lock_old_subdir)
4902 inode_unlock(source);
4903 if (target && (!new_is_dir || lock_new_subdir))
4904 inode_unlock(target);
4907 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4908 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4909 if (flags & RENAME_EXCHANGE) {
4910 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4911 new_is_dir, NULL, new_dentry);
4914 release_dentry_name_snapshot(&old_name);
4918 EXPORT_SYMBOL(vfs_rename);
4920 int do_renameat2(int olddfd, struct filename *from, int newdfd,
4921 struct filename *to, unsigned int flags)
4923 struct renamedata rd;
4924 struct dentry *old_dentry, *new_dentry;
4925 struct dentry *trap;
4926 struct path old_path, new_path;
4927 struct qstr old_last, new_last;
4928 int old_type, new_type;
4929 struct inode *delegated_inode = NULL;
4930 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4931 bool should_retry = false;
4932 int error = -EINVAL;
4934 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4937 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4938 (flags & RENAME_EXCHANGE))
4941 if (flags & RENAME_EXCHANGE)
4945 error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4946 &old_last, &old_type);
4950 error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4956 if (old_path.mnt != new_path.mnt)
4960 if (old_type != LAST_NORM)
4963 if (flags & RENAME_NOREPLACE)
4965 if (new_type != LAST_NORM)
4968 error = mnt_want_write(old_path.mnt);
4973 trap = lock_rename(new_path.dentry, old_path.dentry);
4975 error = PTR_ERR(trap);
4976 goto exit_lock_rename;
4979 old_dentry = lookup_one_qstr_excl(&old_last, old_path.dentry,
4981 error = PTR_ERR(old_dentry);
4982 if (IS_ERR(old_dentry))
4984 /* source must exist */
4986 if (d_is_negative(old_dentry))
4988 new_dentry = lookup_one_qstr_excl(&new_last, new_path.dentry,
4989 lookup_flags | target_flags);
4990 error = PTR_ERR(new_dentry);
4991 if (IS_ERR(new_dentry))
4994 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4996 if (flags & RENAME_EXCHANGE) {
4998 if (d_is_negative(new_dentry))
5001 if (!d_is_dir(new_dentry)) {
5003 if (new_last.name[new_last.len])
5007 /* unless the source is a directory trailing slashes give -ENOTDIR */
5008 if (!d_is_dir(old_dentry)) {
5010 if (old_last.name[old_last.len])
5012 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
5015 /* source should not be ancestor of target */
5017 if (old_dentry == trap)
5019 /* target should not be an ancestor of source */
5020 if (!(flags & RENAME_EXCHANGE))
5022 if (new_dentry == trap)
5025 error = security_path_rename(&old_path, old_dentry,
5026 &new_path, new_dentry, flags);
5030 rd.old_dir = old_path.dentry->d_inode;
5031 rd.old_dentry = old_dentry;
5032 rd.old_mnt_idmap = mnt_idmap(old_path.mnt);
5033 rd.new_dir = new_path.dentry->d_inode;
5034 rd.new_dentry = new_dentry;
5035 rd.new_mnt_idmap = mnt_idmap(new_path.mnt);
5036 rd.delegated_inode = &delegated_inode;
5038 error = vfs_rename(&rd);
5044 unlock_rename(new_path.dentry, old_path.dentry);
5046 if (delegated_inode) {
5047 error = break_deleg_wait(&delegated_inode);
5051 mnt_drop_write(old_path.mnt);
5053 if (retry_estale(error, lookup_flags))
5054 should_retry = true;
5055 path_put(&new_path);
5057 path_put(&old_path);
5059 should_retry = false;
5060 lookup_flags |= LOOKUP_REVAL;
5069 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
5070 int, newdfd, const char __user *, newname, unsigned int, flags)
5072 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5076 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
5077 int, newdfd, const char __user *, newname)
5079 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5083 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5085 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
5086 getname(newname), 0);
5089 int readlink_copy(char __user *buffer, int buflen, const char *link)
5091 int len = PTR_ERR(link);
5096 if (len > (unsigned) buflen)
5098 if (copy_to_user(buffer, link, len))
5105 * vfs_readlink - copy symlink body into userspace buffer
5106 * @dentry: dentry on which to get symbolic link
5107 * @buffer: user memory pointer
5108 * @buflen: size of buffer
5110 * Does not touch atime. That's up to the caller if necessary
5112 * Does not call security hook.
5114 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5116 struct inode *inode = d_inode(dentry);
5117 DEFINE_DELAYED_CALL(done);
5121 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5122 if (unlikely(inode->i_op->readlink))
5123 return inode->i_op->readlink(dentry, buffer, buflen);
5125 if (!d_is_symlink(dentry))
5128 spin_lock(&inode->i_lock);
5129 inode->i_opflags |= IOP_DEFAULT_READLINK;
5130 spin_unlock(&inode->i_lock);
5133 link = READ_ONCE(inode->i_link);
5135 link = inode->i_op->get_link(dentry, inode, &done);
5137 return PTR_ERR(link);
5139 res = readlink_copy(buffer, buflen, link);
5140 do_delayed_call(&done);
5143 EXPORT_SYMBOL(vfs_readlink);
5146 * vfs_get_link - get symlink body
5147 * @dentry: dentry on which to get symbolic link
5148 * @done: caller needs to free returned data with this
5150 * Calls security hook and i_op->get_link() on the supplied inode.
5152 * It does not touch atime. That's up to the caller if necessary.
5154 * Does not work on "special" symlinks like /proc/$$/fd/N
5156 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5158 const char *res = ERR_PTR(-EINVAL);
5159 struct inode *inode = d_inode(dentry);
5161 if (d_is_symlink(dentry)) {
5162 res = ERR_PTR(security_inode_readlink(dentry));
5164 res = inode->i_op->get_link(dentry, inode, done);
5168 EXPORT_SYMBOL(vfs_get_link);
5170 /* get the link contents into pagecache */
5171 const char *page_get_link(struct dentry *dentry, struct inode *inode,
5172 struct delayed_call *callback)
5176 struct address_space *mapping = inode->i_mapping;
5179 page = find_get_page(mapping, 0);
5181 return ERR_PTR(-ECHILD);
5182 if (!PageUptodate(page)) {
5184 return ERR_PTR(-ECHILD);
5187 page = read_mapping_page(mapping, 0, NULL);
5191 set_delayed_call(callback, page_put_link, page);
5192 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5193 kaddr = page_address(page);
5194 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5198 EXPORT_SYMBOL(page_get_link);
5200 void page_put_link(void *arg)
5204 EXPORT_SYMBOL(page_put_link);
5206 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5208 DEFINE_DELAYED_CALL(done);
5209 int res = readlink_copy(buffer, buflen,
5210 page_get_link(dentry, d_inode(dentry),
5212 do_delayed_call(&done);
5215 EXPORT_SYMBOL(page_readlink);
5217 int page_symlink(struct inode *inode, const char *symname, int len)
5219 struct address_space *mapping = inode->i_mapping;
5220 const struct address_space_operations *aops = mapping->a_ops;
5221 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5223 void *fsdata = NULL;
5229 flags = memalloc_nofs_save();
5230 err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5232 memalloc_nofs_restore(flags);
5236 memcpy(page_address(page), symname, len-1);
5238 err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5245 mark_inode_dirty(inode);
5250 EXPORT_SYMBOL(page_symlink);
5252 const struct inode_operations page_symlink_inode_operations = {
5253 .get_link = page_get_link,
5255 EXPORT_SYMBOL(page_symlink_inode_operations);