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 != -ECHILD || !try_to_unlazy(nd))
1723 return inode_permission(idmap, nd->inode, MAY_EXEC);
1726 static int reserve_stack(struct nameidata *nd, struct path *link)
1728 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1731 if (likely(nd->depth != EMBEDDED_LEVELS))
1733 if (likely(nd->stack != nd->internal))
1735 if (likely(nd_alloc_stack(nd)))
1738 if (nd->flags & LOOKUP_RCU) {
1739 // we need to grab link before we do unlazy. And we can't skip
1740 // unlazy even if we fail to grab the link - cleanup needs it
1741 bool grabbed_link = legitimize_path(nd, link, nd->next_seq);
1743 if (!try_to_unlazy(nd) || !grabbed_link)
1746 if (nd_alloc_stack(nd))
1752 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1754 static const char *pick_link(struct nameidata *nd, struct path *link,
1755 struct inode *inode, int flags)
1759 int error = reserve_stack(nd, link);
1761 if (unlikely(error)) {
1762 if (!(nd->flags & LOOKUP_RCU))
1764 return ERR_PTR(error);
1766 last = nd->stack + nd->depth++;
1768 clear_delayed_call(&last->done);
1769 last->seq = nd->next_seq;
1771 if (flags & WALK_TRAILING) {
1772 error = may_follow_link(nd, inode);
1773 if (unlikely(error))
1774 return ERR_PTR(error);
1777 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1778 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1779 return ERR_PTR(-ELOOP);
1781 if (!(nd->flags & LOOKUP_RCU)) {
1782 touch_atime(&last->link);
1784 } else if (atime_needs_update(&last->link, inode)) {
1785 if (!try_to_unlazy(nd))
1786 return ERR_PTR(-ECHILD);
1787 touch_atime(&last->link);
1790 error = security_inode_follow_link(link->dentry, inode,
1791 nd->flags & LOOKUP_RCU);
1792 if (unlikely(error))
1793 return ERR_PTR(error);
1795 res = READ_ONCE(inode->i_link);
1797 const char * (*get)(struct dentry *, struct inode *,
1798 struct delayed_call *);
1799 get = inode->i_op->get_link;
1800 if (nd->flags & LOOKUP_RCU) {
1801 res = get(NULL, inode, &last->done);
1802 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1803 res = get(link->dentry, inode, &last->done);
1805 res = get(link->dentry, inode, &last->done);
1813 error = nd_jump_root(nd);
1814 if (unlikely(error))
1815 return ERR_PTR(error);
1816 while (unlikely(*++res == '/'))
1821 all_done: // pure jump
1827 * Do we need to follow links? We _really_ want to be able
1828 * to do this check without having to look at inode->i_op,
1829 * so we keep a cache of "no, this doesn't need follow_link"
1830 * for the common case.
1832 * NOTE: dentry must be what nd->next_seq had been sampled from.
1834 static const char *step_into(struct nameidata *nd, int flags,
1835 struct dentry *dentry)
1838 struct inode *inode;
1839 int err = handle_mounts(nd, dentry, &path);
1842 return ERR_PTR(err);
1843 inode = path.dentry->d_inode;
1844 if (likely(!d_is_symlink(path.dentry)) ||
1845 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1846 (flags & WALK_NOFOLLOW)) {
1847 /* not a symlink or should not follow */
1848 if (nd->flags & LOOKUP_RCU) {
1849 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1850 return ERR_PTR(-ECHILD);
1851 if (unlikely(!inode))
1852 return ERR_PTR(-ENOENT);
1854 dput(nd->path.dentry);
1855 if (nd->path.mnt != path.mnt)
1856 mntput(nd->path.mnt);
1860 nd->seq = nd->next_seq;
1863 if (nd->flags & LOOKUP_RCU) {
1864 /* make sure that d_is_symlink above matches inode */
1865 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1866 return ERR_PTR(-ECHILD);
1868 if (path.mnt == nd->path.mnt)
1871 return pick_link(nd, &path, inode, flags);
1874 static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1876 struct dentry *parent, *old;
1878 if (path_equal(&nd->path, &nd->root))
1880 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1883 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1884 &nd->root, &path, &seq))
1886 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1887 return ERR_PTR(-ECHILD);
1889 nd->inode = path.dentry->d_inode;
1891 // makes sure that non-RCU pathwalk could reach this state
1892 if (read_seqretry(&mount_lock, nd->m_seq))
1893 return ERR_PTR(-ECHILD);
1894 /* we know that mountpoint was pinned */
1896 old = nd->path.dentry;
1897 parent = old->d_parent;
1898 nd->next_seq = read_seqcount_begin(&parent->d_seq);
1899 // makes sure that non-RCU pathwalk could reach this state
1900 if (read_seqcount_retry(&old->d_seq, nd->seq))
1901 return ERR_PTR(-ECHILD);
1902 if (unlikely(!path_connected(nd->path.mnt, parent)))
1903 return ERR_PTR(-ECHILD);
1906 if (read_seqretry(&mount_lock, nd->m_seq))
1907 return ERR_PTR(-ECHILD);
1908 if (unlikely(nd->flags & LOOKUP_BENEATH))
1909 return ERR_PTR(-ECHILD);
1910 nd->next_seq = nd->seq;
1911 return nd->path.dentry;
1914 static struct dentry *follow_dotdot(struct nameidata *nd)
1916 struct dentry *parent;
1918 if (path_equal(&nd->path, &nd->root))
1920 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1923 if (!choose_mountpoint(real_mount(nd->path.mnt),
1926 path_put(&nd->path);
1928 nd->inode = path.dentry->d_inode;
1929 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1930 return ERR_PTR(-EXDEV);
1932 /* rare case of legitimate dget_parent()... */
1933 parent = dget_parent(nd->path.dentry);
1934 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1936 return ERR_PTR(-ENOENT);
1941 if (unlikely(nd->flags & LOOKUP_BENEATH))
1942 return ERR_PTR(-EXDEV);
1943 return dget(nd->path.dentry);
1946 static const char *handle_dots(struct nameidata *nd, int type)
1948 if (type == LAST_DOTDOT) {
1949 const char *error = NULL;
1950 struct dentry *parent;
1952 if (!nd->root.mnt) {
1953 error = ERR_PTR(set_root(nd));
1957 if (nd->flags & LOOKUP_RCU)
1958 parent = follow_dotdot_rcu(nd);
1960 parent = follow_dotdot(nd);
1962 return ERR_CAST(parent);
1963 error = step_into(nd, WALK_NOFOLLOW, parent);
1964 if (unlikely(error))
1967 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1969 * If there was a racing rename or mount along our
1970 * path, then we can't be sure that ".." hasn't jumped
1971 * above nd->root (and so userspace should retry or use
1975 if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
1976 return ERR_PTR(-EAGAIN);
1977 if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
1978 return ERR_PTR(-EAGAIN);
1984 static const char *walk_component(struct nameidata *nd, int flags)
1986 struct dentry *dentry;
1988 * "." and ".." are special - ".." especially so because it has
1989 * to be able to know about the current root directory and
1990 * parent relationships.
1992 if (unlikely(nd->last_type != LAST_NORM)) {
1993 if (!(flags & WALK_MORE) && nd->depth)
1995 return handle_dots(nd, nd->last_type);
1997 dentry = lookup_fast(nd);
1999 return ERR_CAST(dentry);
2000 if (unlikely(!dentry)) {
2001 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
2003 return ERR_CAST(dentry);
2005 if (!(flags & WALK_MORE) && nd->depth)
2007 return step_into(nd, flags, dentry);
2011 * We can do the critical dentry name comparison and hashing
2012 * operations one word at a time, but we are limited to:
2014 * - Architectures with fast unaligned word accesses. We could
2015 * do a "get_unaligned()" if this helps and is sufficiently
2018 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2019 * do not trap on the (extremely unlikely) case of a page
2020 * crossing operation.
2022 * - Furthermore, we need an efficient 64-bit compile for the
2023 * 64-bit case in order to generate the "number of bytes in
2024 * the final mask". Again, that could be replaced with a
2025 * efficient population count instruction or similar.
2027 #ifdef CONFIG_DCACHE_WORD_ACCESS
2029 #include <asm/word-at-a-time.h>
2033 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2035 #elif defined(CONFIG_64BIT)
2037 * Register pressure in the mixing function is an issue, particularly
2038 * on 32-bit x86, but almost any function requires one state value and
2039 * one temporary. Instead, use a function designed for two state values
2040 * and no temporaries.
2042 * This function cannot create a collision in only two iterations, so
2043 * we have two iterations to achieve avalanche. In those two iterations,
2044 * we have six layers of mixing, which is enough to spread one bit's
2045 * influence out to 2^6 = 64 state bits.
2047 * Rotate constants are scored by considering either 64 one-bit input
2048 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2049 * probability of that delta causing a change to each of the 128 output
2050 * bits, using a sample of random initial states.
2052 * The Shannon entropy of the computed probabilities is then summed
2053 * to produce a score. Ideally, any input change has a 50% chance of
2054 * toggling any given output bit.
2056 * Mixing scores (in bits) for (12,45):
2057 * Input delta: 1-bit 2-bit
2058 * 1 round: 713.3 42542.6
2059 * 2 rounds: 2753.7 140389.8
2060 * 3 rounds: 5954.1 233458.2
2061 * 4 rounds: 7862.6 256672.2
2062 * Perfect: 8192 258048
2063 * (64*128) (64*63/2 * 128)
2065 #define HASH_MIX(x, y, a) \
2067 y ^= x, x = rol64(x,12),\
2068 x += y, y = rol64(y,45),\
2072 * Fold two longs into one 32-bit hash value. This must be fast, but
2073 * latency isn't quite as critical, as there is a fair bit of additional
2074 * work done before the hash value is used.
2076 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2078 y ^= x * GOLDEN_RATIO_64;
2079 y *= GOLDEN_RATIO_64;
2083 #else /* 32-bit case */
2086 * Mixing scores (in bits) for (7,20):
2087 * Input delta: 1-bit 2-bit
2088 * 1 round: 330.3 9201.6
2089 * 2 rounds: 1246.4 25475.4
2090 * 3 rounds: 1907.1 31295.1
2091 * 4 rounds: 2042.3 31718.6
2092 * Perfect: 2048 31744
2093 * (32*64) (32*31/2 * 64)
2095 #define HASH_MIX(x, y, a) \
2097 y ^= x, x = rol32(x, 7),\
2098 x += y, y = rol32(y,20),\
2101 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2103 /* Use arch-optimized multiply if one exists */
2104 return __hash_32(y ^ __hash_32(x));
2110 * Return the hash of a string of known length. This is carfully
2111 * designed to match hash_name(), which is the more critical function.
2112 * In particular, we must end by hashing a final word containing 0..7
2113 * payload bytes, to match the way that hash_name() iterates until it
2114 * finds the delimiter after the name.
2116 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2118 unsigned long a, x = 0, y = (unsigned long)salt;
2123 a = load_unaligned_zeropad(name);
2124 if (len < sizeof(unsigned long))
2127 name += sizeof(unsigned long);
2128 len -= sizeof(unsigned long);
2130 x ^= a & bytemask_from_count(len);
2132 return fold_hash(x, y);
2134 EXPORT_SYMBOL(full_name_hash);
2136 /* Return the "hash_len" (hash and length) of a null-terminated string */
2137 u64 hashlen_string(const void *salt, const char *name)
2139 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2140 unsigned long adata, mask, len;
2141 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2148 len += sizeof(unsigned long);
2150 a = load_unaligned_zeropad(name+len);
2151 } while (!has_zero(a, &adata, &constants));
2153 adata = prep_zero_mask(a, adata, &constants);
2154 mask = create_zero_mask(adata);
2155 x ^= a & zero_bytemask(mask);
2157 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2159 EXPORT_SYMBOL(hashlen_string);
2162 * Calculate the length and hash of the path component, and
2163 * return the "hash_len" as the result.
2165 static inline u64 hash_name(const void *salt, const char *name)
2167 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2168 unsigned long adata, bdata, mask, len;
2169 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2176 len += sizeof(unsigned long);
2178 a = load_unaligned_zeropad(name+len);
2179 b = a ^ REPEAT_BYTE('/');
2180 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2182 adata = prep_zero_mask(a, adata, &constants);
2183 bdata = prep_zero_mask(b, bdata, &constants);
2184 mask = create_zero_mask(adata | bdata);
2185 x ^= a & zero_bytemask(mask);
2187 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2190 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2192 /* Return the hash of a string of known length */
2193 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2195 unsigned long hash = init_name_hash(salt);
2197 hash = partial_name_hash((unsigned char)*name++, hash);
2198 return end_name_hash(hash);
2200 EXPORT_SYMBOL(full_name_hash);
2202 /* Return the "hash_len" (hash and length) of a null-terminated string */
2203 u64 hashlen_string(const void *salt, const char *name)
2205 unsigned long hash = init_name_hash(salt);
2206 unsigned long len = 0, c;
2208 c = (unsigned char)*name;
2211 hash = partial_name_hash(c, hash);
2212 c = (unsigned char)name[len];
2214 return hashlen_create(end_name_hash(hash), len);
2216 EXPORT_SYMBOL(hashlen_string);
2219 * We know there's a real path component here of at least
2222 static inline u64 hash_name(const void *salt, const char *name)
2224 unsigned long hash = init_name_hash(salt);
2225 unsigned long len = 0, c;
2227 c = (unsigned char)*name;
2230 hash = partial_name_hash(c, hash);
2231 c = (unsigned char)name[len];
2232 } while (c && c != '/');
2233 return hashlen_create(end_name_hash(hash), len);
2240 * This is the basic name resolution function, turning a pathname into
2241 * the final dentry. We expect 'base' to be positive and a directory.
2243 * Returns 0 and nd will have valid dentry and mnt on success.
2244 * Returns error and drops reference to input namei data on failure.
2246 static int link_path_walk(const char *name, struct nameidata *nd)
2248 int depth = 0; // depth <= nd->depth
2251 nd->last_type = LAST_ROOT;
2252 nd->flags |= LOOKUP_PARENT;
2254 return PTR_ERR(name);
2258 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2262 /* At this point we know we have a real path component. */
2264 struct mnt_idmap *idmap;
2269 idmap = mnt_idmap(nd->path.mnt);
2270 err = may_lookup(idmap, nd);
2274 hash_len = hash_name(nd->path.dentry, name);
2277 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2279 if (name[1] == '.') {
2281 nd->state |= ND_JUMPED;
2287 if (likely(type == LAST_NORM)) {
2288 struct dentry *parent = nd->path.dentry;
2289 nd->state &= ~ND_JUMPED;
2290 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2291 struct qstr this = { { .hash_len = hash_len }, .name = name };
2292 err = parent->d_op->d_hash(parent, &this);
2295 hash_len = this.hash_len;
2300 nd->last.hash_len = hash_len;
2301 nd->last.name = name;
2302 nd->last_type = type;
2304 name += hashlen_len(hash_len);
2308 * If it wasn't NUL, we know it was '/'. Skip that
2309 * slash, and continue until no more slashes.
2313 } while (unlikely(*name == '/'));
2314 if (unlikely(!*name)) {
2316 /* pathname or trailing symlink, done */
2318 nd->dir_vfsuid = i_uid_into_vfsuid(idmap, nd->inode);
2319 nd->dir_mode = nd->inode->i_mode;
2320 nd->flags &= ~LOOKUP_PARENT;
2323 /* last component of nested symlink */
2324 name = nd->stack[--depth].name;
2325 link = walk_component(nd, 0);
2327 /* not the last component */
2328 link = walk_component(nd, WALK_MORE);
2330 if (unlikely(link)) {
2332 return PTR_ERR(link);
2333 /* a symlink to follow */
2334 nd->stack[depth++].name = name;
2338 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2339 if (nd->flags & LOOKUP_RCU) {
2340 if (!try_to_unlazy(nd))
2348 /* must be paired with terminate_walk() */
2349 static const char *path_init(struct nameidata *nd, unsigned flags)
2352 const char *s = nd->name->name;
2354 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2355 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2356 return ERR_PTR(-EAGAIN);
2359 flags &= ~LOOKUP_RCU;
2360 if (flags & LOOKUP_RCU)
2363 nd->seq = nd->next_seq = 0;
2366 nd->state |= ND_JUMPED;
2368 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2369 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2372 if (nd->state & ND_ROOT_PRESET) {
2373 struct dentry *root = nd->root.dentry;
2374 struct inode *inode = root->d_inode;
2375 if (*s && unlikely(!d_can_lookup(root)))
2376 return ERR_PTR(-ENOTDIR);
2377 nd->path = nd->root;
2379 if (flags & LOOKUP_RCU) {
2380 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2381 nd->root_seq = nd->seq;
2383 path_get(&nd->path);
2388 nd->root.mnt = NULL;
2390 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2391 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2392 error = nd_jump_root(nd);
2393 if (unlikely(error))
2394 return ERR_PTR(error);
2398 /* Relative pathname -- get the starting-point it is relative to. */
2399 if (nd->dfd == AT_FDCWD) {
2400 if (flags & LOOKUP_RCU) {
2401 struct fs_struct *fs = current->fs;
2405 seq = read_seqcount_begin(&fs->seq);
2407 nd->inode = nd->path.dentry->d_inode;
2408 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2409 } while (read_seqcount_retry(&fs->seq, seq));
2411 get_fs_pwd(current->fs, &nd->path);
2412 nd->inode = nd->path.dentry->d_inode;
2415 /* Caller must check execute permissions on the starting path component */
2416 struct fd f = fdget_raw(nd->dfd);
2417 struct dentry *dentry;
2420 return ERR_PTR(-EBADF);
2422 dentry = f.file->f_path.dentry;
2424 if (*s && unlikely(!d_can_lookup(dentry))) {
2426 return ERR_PTR(-ENOTDIR);
2429 nd->path = f.file->f_path;
2430 if (flags & LOOKUP_RCU) {
2431 nd->inode = nd->path.dentry->d_inode;
2432 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2434 path_get(&nd->path);
2435 nd->inode = nd->path.dentry->d_inode;
2440 /* For scoped-lookups we need to set the root to the dirfd as well. */
2441 if (flags & LOOKUP_IS_SCOPED) {
2442 nd->root = nd->path;
2443 if (flags & LOOKUP_RCU) {
2444 nd->root_seq = nd->seq;
2446 path_get(&nd->root);
2447 nd->state |= ND_ROOT_GRABBED;
2453 static inline const char *lookup_last(struct nameidata *nd)
2455 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2456 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2458 return walk_component(nd, WALK_TRAILING);
2461 static int handle_lookup_down(struct nameidata *nd)
2463 if (!(nd->flags & LOOKUP_RCU))
2464 dget(nd->path.dentry);
2465 nd->next_seq = nd->seq;
2466 return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry));
2469 /* Returns 0 and nd will be valid on success; Returns error, otherwise. */
2470 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2472 const char *s = path_init(nd, flags);
2475 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2476 err = handle_lookup_down(nd);
2477 if (unlikely(err < 0))
2481 while (!(err = link_path_walk(s, nd)) &&
2482 (s = lookup_last(nd)) != NULL)
2484 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2485 err = handle_lookup_down(nd);
2486 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2489 err = complete_walk(nd);
2491 if (!err && nd->flags & LOOKUP_DIRECTORY)
2492 if (!d_can_lookup(nd->path.dentry))
2496 nd->path.mnt = NULL;
2497 nd->path.dentry = NULL;
2503 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2504 struct path *path, struct path *root)
2507 struct nameidata nd;
2509 return PTR_ERR(name);
2510 set_nameidata(&nd, dfd, name, root);
2511 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2512 if (unlikely(retval == -ECHILD))
2513 retval = path_lookupat(&nd, flags, path);
2514 if (unlikely(retval == -ESTALE))
2515 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2517 if (likely(!retval))
2518 audit_inode(name, path->dentry,
2519 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2520 restore_nameidata();
2524 /* Returns 0 and nd will be valid on success; Returns error, otherwise. */
2525 static int path_parentat(struct nameidata *nd, unsigned flags,
2526 struct path *parent)
2528 const char *s = path_init(nd, flags);
2529 int err = link_path_walk(s, nd);
2531 err = complete_walk(nd);
2534 nd->path.mnt = NULL;
2535 nd->path.dentry = NULL;
2541 /* Note: this does not consume "name" */
2542 static int __filename_parentat(int dfd, struct filename *name,
2543 unsigned int flags, struct path *parent,
2544 struct qstr *last, int *type,
2545 const struct path *root)
2548 struct nameidata nd;
2551 return PTR_ERR(name);
2552 set_nameidata(&nd, dfd, name, root);
2553 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2554 if (unlikely(retval == -ECHILD))
2555 retval = path_parentat(&nd, flags, parent);
2556 if (unlikely(retval == -ESTALE))
2557 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2558 if (likely(!retval)) {
2560 *type = nd.last_type;
2561 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2563 restore_nameidata();
2567 static int filename_parentat(int dfd, struct filename *name,
2568 unsigned int flags, struct path *parent,
2569 struct qstr *last, int *type)
2571 return __filename_parentat(dfd, name, flags, parent, last, type, NULL);
2574 /* does lookup, returns the object with parent locked */
2575 static struct dentry *__kern_path_locked(struct filename *name, struct path *path)
2581 error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type);
2583 return ERR_PTR(error);
2584 if (unlikely(type != LAST_NORM)) {
2586 return ERR_PTR(-EINVAL);
2588 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2589 d = lookup_one_qstr_excl(&last, path->dentry, 0);
2591 inode_unlock(path->dentry->d_inode);
2597 struct dentry *kern_path_locked(const char *name, struct path *path)
2599 struct filename *filename = getname_kernel(name);
2600 struct dentry *res = __kern_path_locked(filename, path);
2606 int kern_path(const char *name, unsigned int flags, struct path *path)
2608 struct filename *filename = getname_kernel(name);
2609 int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2615 EXPORT_SYMBOL(kern_path);
2618 * vfs_path_parent_lookup - lookup a parent path relative to a dentry-vfsmount pair
2619 * @filename: filename structure
2620 * @flags: lookup flags
2621 * @parent: pointer to struct path to fill
2622 * @last: last component
2623 * @type: type of the last component
2624 * @root: pointer to struct path of the base directory
2626 int vfs_path_parent_lookup(struct filename *filename, unsigned int flags,
2627 struct path *parent, struct qstr *last, int *type,
2628 const struct path *root)
2630 return __filename_parentat(AT_FDCWD, filename, flags, parent, last,
2633 EXPORT_SYMBOL(vfs_path_parent_lookup);
2636 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2637 * @dentry: pointer to dentry of the base directory
2638 * @mnt: pointer to vfs mount of the base directory
2639 * @name: pointer to file name
2640 * @flags: lookup flags
2641 * @path: pointer to struct path to fill
2643 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2644 const char *name, unsigned int flags,
2647 struct filename *filename;
2648 struct path root = {.mnt = mnt, .dentry = dentry};
2651 filename = getname_kernel(name);
2652 /* the first argument of filename_lookup() is ignored with root */
2653 ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2657 EXPORT_SYMBOL(vfs_path_lookup);
2659 static int lookup_one_common(struct mnt_idmap *idmap,
2660 const char *name, struct dentry *base, int len,
2665 this->hash = full_name_hash(base, name, len);
2669 if (unlikely(name[0] == '.')) {
2670 if (len < 2 || (len == 2 && name[1] == '.'))
2675 unsigned int c = *(const unsigned char *)name++;
2676 if (c == '/' || c == '\0')
2680 * See if the low-level filesystem might want
2681 * to use its own hash..
2683 if (base->d_flags & DCACHE_OP_HASH) {
2684 int err = base->d_op->d_hash(base, this);
2689 return inode_permission(idmap, base->d_inode, MAY_EXEC);
2693 * try_lookup_one_len - filesystem helper to lookup single pathname component
2694 * @name: pathname component to lookup
2695 * @base: base directory to lookup from
2696 * @len: maximum length @len should be interpreted to
2698 * Look up a dentry by name in the dcache, returning NULL if it does not
2699 * currently exist. The function does not try to create a dentry.
2701 * Note that this routine is purely a helper for filesystem usage and should
2702 * not be called by generic code.
2704 * The caller must hold base->i_mutex.
2706 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2711 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2713 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2715 return ERR_PTR(err);
2717 return lookup_dcache(&this, base, 0);
2719 EXPORT_SYMBOL(try_lookup_one_len);
2722 * lookup_one_len - filesystem helper to lookup single pathname component
2723 * @name: pathname component to lookup
2724 * @base: base directory to lookup from
2725 * @len: maximum length @len should be interpreted to
2727 * Note that this routine is purely a helper for filesystem usage and should
2728 * not be called by generic code.
2730 * The caller must hold base->i_mutex.
2732 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2734 struct dentry *dentry;
2738 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2740 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2742 return ERR_PTR(err);
2744 dentry = lookup_dcache(&this, base, 0);
2745 return dentry ? dentry : __lookup_slow(&this, base, 0);
2747 EXPORT_SYMBOL(lookup_one_len);
2750 * lookup_one - filesystem helper to lookup single pathname component
2751 * @idmap: idmap of the mount the lookup is performed from
2752 * @name: pathname component to lookup
2753 * @base: base directory to lookup from
2754 * @len: maximum length @len should be interpreted to
2756 * Note that this routine is purely a helper for filesystem usage and should
2757 * not be called by generic code.
2759 * The caller must hold base->i_mutex.
2761 struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name,
2762 struct dentry *base, int len)
2764 struct dentry *dentry;
2768 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2770 err = lookup_one_common(idmap, name, base, len, &this);
2772 return ERR_PTR(err);
2774 dentry = lookup_dcache(&this, base, 0);
2775 return dentry ? dentry : __lookup_slow(&this, base, 0);
2777 EXPORT_SYMBOL(lookup_one);
2780 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2781 * @idmap: idmap of the mount the lookup is performed from
2782 * @name: pathname component to lookup
2783 * @base: base directory to lookup from
2784 * @len: maximum length @len should be interpreted to
2786 * Note that this routine is purely a helper for filesystem usage and should
2787 * not be called by generic code.
2789 * Unlike lookup_one_len, it should be called without the parent
2790 * i_mutex held, and will take the i_mutex itself if necessary.
2792 struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap,
2793 const char *name, struct dentry *base,
2800 err = lookup_one_common(idmap, name, base, len, &this);
2802 return ERR_PTR(err);
2804 ret = lookup_dcache(&this, base, 0);
2806 ret = lookup_slow(&this, base, 0);
2809 EXPORT_SYMBOL(lookup_one_unlocked);
2812 * lookup_one_positive_unlocked - filesystem helper to lookup single
2813 * pathname component
2814 * @idmap: idmap of the mount the lookup is performed from
2815 * @name: pathname component to lookup
2816 * @base: base directory to lookup from
2817 * @len: maximum length @len should be interpreted to
2819 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2820 * known positive or ERR_PTR(). This is what most of the users want.
2822 * Note that pinned negative with unlocked parent _can_ become positive at any
2823 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2824 * positives have >d_inode stable, so this one avoids such problems.
2826 * Note that this routine is purely a helper for filesystem usage and should
2827 * not be called by generic code.
2829 * The helper should be called without i_mutex held.
2831 struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap,
2833 struct dentry *base, int len)
2835 struct dentry *ret = lookup_one_unlocked(idmap, name, base, len);
2837 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2839 ret = ERR_PTR(-ENOENT);
2843 EXPORT_SYMBOL(lookup_one_positive_unlocked);
2846 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2847 * @name: pathname component to lookup
2848 * @base: base directory to lookup from
2849 * @len: maximum length @len should be interpreted to
2851 * Note that this routine is purely a helper for filesystem usage and should
2852 * not be called by generic code.
2854 * Unlike lookup_one_len, it should be called without the parent
2855 * i_mutex held, and will take the i_mutex itself if necessary.
2857 struct dentry *lookup_one_len_unlocked(const char *name,
2858 struct dentry *base, int len)
2860 return lookup_one_unlocked(&nop_mnt_idmap, name, base, len);
2862 EXPORT_SYMBOL(lookup_one_len_unlocked);
2865 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2866 * on negatives. Returns known positive or ERR_PTR(); that's what
2867 * most of the users want. Note that pinned negative with unlocked parent
2868 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2869 * need to be very careful; pinned positives have ->d_inode stable, so
2870 * this one avoids such problems.
2872 struct dentry *lookup_positive_unlocked(const char *name,
2873 struct dentry *base, int len)
2875 return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len);
2877 EXPORT_SYMBOL(lookup_positive_unlocked);
2879 #ifdef CONFIG_UNIX98_PTYS
2880 int path_pts(struct path *path)
2882 /* Find something mounted on "pts" in the same directory as
2885 struct dentry *parent = dget_parent(path->dentry);
2886 struct dentry *child;
2887 struct qstr this = QSTR_INIT("pts", 3);
2889 if (unlikely(!path_connected(path->mnt, parent))) {
2894 path->dentry = parent;
2895 child = d_hash_and_lookup(parent, &this);
2896 if (IS_ERR_OR_NULL(child))
2899 path->dentry = child;
2901 follow_down(path, 0);
2906 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2907 struct path *path, int *empty)
2909 struct filename *filename = getname_flags(name, flags, empty);
2910 int ret = filename_lookup(dfd, filename, flags, path, NULL);
2915 EXPORT_SYMBOL(user_path_at_empty);
2917 int __check_sticky(struct mnt_idmap *idmap, struct inode *dir,
2918 struct inode *inode)
2920 kuid_t fsuid = current_fsuid();
2922 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), fsuid))
2924 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, dir), fsuid))
2926 return !capable_wrt_inode_uidgid(idmap, inode, CAP_FOWNER);
2928 EXPORT_SYMBOL(__check_sticky);
2931 * Check whether we can remove a link victim from directory dir, check
2932 * whether the type of victim is right.
2933 * 1. We can't do it if dir is read-only (done in permission())
2934 * 2. We should have write and exec permissions on dir
2935 * 3. We can't remove anything from append-only dir
2936 * 4. We can't do anything with immutable dir (done in permission())
2937 * 5. If the sticky bit on dir is set we should either
2938 * a. be owner of dir, or
2939 * b. be owner of victim, or
2940 * c. have CAP_FOWNER capability
2941 * 6. If the victim is append-only or immutable we can't do antyhing with
2942 * links pointing to it.
2943 * 7. If the victim has an unknown uid or gid we can't change the inode.
2944 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2945 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2946 * 10. We can't remove a root or mountpoint.
2947 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2948 * nfs_async_unlink().
2950 static int may_delete(struct mnt_idmap *idmap, struct inode *dir,
2951 struct dentry *victim, bool isdir)
2953 struct inode *inode = d_backing_inode(victim);
2956 if (d_is_negative(victim))
2960 BUG_ON(victim->d_parent->d_inode != dir);
2962 /* Inode writeback is not safe when the uid or gid are invalid. */
2963 if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
2964 !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
2967 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2969 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
2975 if (check_sticky(idmap, dir, inode) || IS_APPEND(inode) ||
2976 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2977 HAS_UNMAPPED_ID(idmap, inode))
2980 if (!d_is_dir(victim))
2982 if (IS_ROOT(victim))
2984 } else if (d_is_dir(victim))
2986 if (IS_DEADDIR(dir))
2988 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2993 /* Check whether we can create an object with dentry child in directory
2995 * 1. We can't do it if child already exists (open has special treatment for
2996 * this case, but since we are inlined it's OK)
2997 * 2. We can't do it if dir is read-only (done in permission())
2998 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2999 * 4. We should have write and exec permissions on dir
3000 * 5. We can't do it if dir is immutable (done in permission())
3002 static inline int may_create(struct mnt_idmap *idmap,
3003 struct inode *dir, struct dentry *child)
3005 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
3008 if (IS_DEADDIR(dir))
3010 if (!fsuidgid_has_mapping(dir->i_sb, idmap))
3013 return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3016 // p1 != p2, both are on the same filesystem, ->s_vfs_rename_mutex is held
3017 static struct dentry *lock_two_directories(struct dentry *p1, struct dentry *p2)
3019 struct dentry *p = p1, *q = p2, *r;
3021 while ((r = p->d_parent) != p2 && r != p)
3024 // p is a child of p2 and an ancestor of p1 or p1 itself
3025 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3026 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT2);
3029 // p is the root of connected component that contains p1
3030 // p2 does not occur on the path from p to p1
3031 while ((r = q->d_parent) != p1 && r != p && r != q)
3034 // q is a child of p1 and an ancestor of p2 or p2 itself
3035 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3036 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3038 } else if (likely(r == p)) {
3039 // both p2 and p1 are descendents of p
3040 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3041 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3043 } else { // no common ancestor at the time we'd been called
3044 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3045 return ERR_PTR(-EXDEV);
3050 * p1 and p2 should be directories on the same fs.
3052 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
3055 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3059 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
3060 return lock_two_directories(p1, p2);
3062 EXPORT_SYMBOL(lock_rename);
3065 * c1 and p2 should be on the same fs.
3067 struct dentry *lock_rename_child(struct dentry *c1, struct dentry *p2)
3069 if (READ_ONCE(c1->d_parent) == p2) {
3071 * hopefully won't need to touch ->s_vfs_rename_mutex at all.
3073 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3075 * now that p2 is locked, nobody can move in or out of it,
3076 * so the test below is safe.
3078 if (likely(c1->d_parent == p2))
3082 * c1 got moved out of p2 while we'd been taking locks;
3083 * unlock and fall back to slow case.
3085 inode_unlock(p2->d_inode);
3088 mutex_lock(&c1->d_sb->s_vfs_rename_mutex);
3090 * nobody can move out of any directories on this fs.
3092 if (likely(c1->d_parent != p2))
3093 return lock_two_directories(c1->d_parent, p2);
3096 * c1 got moved into p2 while we were taking locks;
3097 * we need p2 locked and ->s_vfs_rename_mutex unlocked,
3098 * for consistency with lock_rename().
3100 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3101 mutex_unlock(&c1->d_sb->s_vfs_rename_mutex);
3104 EXPORT_SYMBOL(lock_rename_child);
3106 void unlock_rename(struct dentry *p1, struct dentry *p2)
3108 inode_unlock(p1->d_inode);
3110 inode_unlock(p2->d_inode);
3111 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3114 EXPORT_SYMBOL(unlock_rename);
3117 * vfs_prepare_mode - prepare the mode to be used for a new inode
3118 * @idmap: idmap of the mount the inode was found from
3119 * @dir: parent directory of the new inode
3120 * @mode: mode of the new inode
3121 * @mask_perms: allowed permission by the vfs
3122 * @type: type of file to be created
3124 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3125 * object to be created.
3127 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3128 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3129 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3130 * POSIX ACL supporting filesystems.
3132 * Note that it's currently valid for @type to be 0 if a directory is created.
3133 * Filesystems raise that flag individually and we need to check whether each
3134 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3137 * Returns: mode to be passed to the filesystem
3139 static inline umode_t vfs_prepare_mode(struct mnt_idmap *idmap,
3140 const struct inode *dir, umode_t mode,
3141 umode_t mask_perms, umode_t type)
3143 mode = mode_strip_sgid(idmap, dir, mode);
3144 mode = mode_strip_umask(dir, mode);
3147 * Apply the vfs mandated allowed permission mask and set the type of
3148 * file to be created before we call into the filesystem.
3150 mode &= (mask_perms & ~S_IFMT);
3151 mode |= (type & S_IFMT);
3157 * vfs_create - create new file
3158 * @idmap: idmap of the mount the inode was found from
3159 * @dir: inode of @dentry
3160 * @dentry: pointer to dentry of the base directory
3161 * @mode: mode of the new file
3162 * @want_excl: whether the file must not yet exist
3164 * Create a new file.
3166 * If the inode has been found through an idmapped mount the idmap of
3167 * the vfsmount must be passed through @idmap. This function will then take
3168 * care to map the inode according to @idmap before checking permissions.
3169 * On non-idmapped mounts or if permission checking is to be performed on the
3170 * raw inode simply pass @nop_mnt_idmap.
3172 int vfs_create(struct mnt_idmap *idmap, struct inode *dir,
3173 struct dentry *dentry, umode_t mode, bool want_excl)
3177 error = may_create(idmap, dir, dentry);
3181 if (!dir->i_op->create)
3182 return -EACCES; /* shouldn't it be ENOSYS? */
3184 mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG);
3185 error = security_inode_create(dir, dentry, mode);
3188 error = dir->i_op->create(idmap, dir, dentry, mode, want_excl);
3190 fsnotify_create(dir, dentry);
3193 EXPORT_SYMBOL(vfs_create);
3195 int vfs_mkobj(struct dentry *dentry, umode_t mode,
3196 int (*f)(struct dentry *, umode_t, void *),
3199 struct inode *dir = dentry->d_parent->d_inode;
3200 int error = may_create(&nop_mnt_idmap, dir, dentry);
3206 error = security_inode_create(dir, dentry, mode);
3209 error = f(dentry, mode, arg);
3211 fsnotify_create(dir, dentry);
3214 EXPORT_SYMBOL(vfs_mkobj);
3216 bool may_open_dev(const struct path *path)
3218 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3219 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3222 static int may_open(struct mnt_idmap *idmap, const struct path *path,
3223 int acc_mode, int flag)
3225 struct dentry *dentry = path->dentry;
3226 struct inode *inode = dentry->d_inode;
3232 switch (inode->i_mode & S_IFMT) {
3236 if (acc_mode & MAY_WRITE)
3238 if (acc_mode & MAY_EXEC)
3243 if (!may_open_dev(path))
3248 if (acc_mode & MAY_EXEC)
3253 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3258 error = inode_permission(idmap, inode, MAY_OPEN | acc_mode);
3263 * An append-only file must be opened in append mode for writing.
3265 if (IS_APPEND(inode)) {
3266 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3272 /* O_NOATIME can only be set by the owner or superuser */
3273 if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode))
3279 static int handle_truncate(struct mnt_idmap *idmap, struct file *filp)
3281 const struct path *path = &filp->f_path;
3282 struct inode *inode = path->dentry->d_inode;
3283 int error = get_write_access(inode);
3287 error = security_file_truncate(filp);
3289 error = do_truncate(idmap, path->dentry, 0,
3290 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3293 put_write_access(inode);
3297 static inline int open_to_namei_flags(int flag)
3299 if ((flag & O_ACCMODE) == 3)
3304 static int may_o_create(struct mnt_idmap *idmap,
3305 const struct path *dir, struct dentry *dentry,
3308 int error = security_path_mknod(dir, dentry, mode, 0);
3312 if (!fsuidgid_has_mapping(dir->dentry->d_sb, idmap))
3315 error = inode_permission(idmap, dir->dentry->d_inode,
3316 MAY_WRITE | MAY_EXEC);
3320 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3324 * Attempt to atomically look up, create and open a file from a negative
3327 * Returns 0 if successful. The file will have been created and attached to
3328 * @file by the filesystem calling finish_open().
3330 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3331 * be set. The caller will need to perform the open themselves. @path will
3332 * have been updated to point to the new dentry. This may be negative.
3334 * Returns an error code otherwise.
3336 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3338 int open_flag, umode_t mode)
3340 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3341 struct inode *dir = nd->path.dentry->d_inode;
3344 if (nd->flags & LOOKUP_DIRECTORY)
3345 open_flag |= O_DIRECTORY;
3347 file->f_path.dentry = DENTRY_NOT_SET;
3348 file->f_path.mnt = nd->path.mnt;
3349 error = dir->i_op->atomic_open(dir, dentry, file,
3350 open_to_namei_flags(open_flag), mode);
3351 d_lookup_done(dentry);
3353 if (file->f_mode & FMODE_OPENED) {
3354 if (unlikely(dentry != file->f_path.dentry)) {
3356 dentry = dget(file->f_path.dentry);
3358 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3361 if (file->f_path.dentry) {
3363 dentry = file->f_path.dentry;
3365 if (unlikely(d_is_negative(dentry)))
3371 dentry = ERR_PTR(error);
3377 * Look up and maybe create and open the last component.
3379 * Must be called with parent locked (exclusive in O_CREAT case).
3381 * Returns 0 on success, that is, if
3382 * the file was successfully atomically created (if necessary) and opened, or
3383 * the file was not completely opened at this time, though lookups and
3384 * creations were performed.
3385 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3386 * In the latter case dentry returned in @path might be negative if O_CREAT
3387 * hadn't been specified.
3389 * An error code is returned on failure.
3391 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3392 const struct open_flags *op,
3395 struct mnt_idmap *idmap;
3396 struct dentry *dir = nd->path.dentry;
3397 struct inode *dir_inode = dir->d_inode;
3398 int open_flag = op->open_flag;
3399 struct dentry *dentry;
3400 int error, create_error = 0;
3401 umode_t mode = op->mode;
3402 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3404 if (unlikely(IS_DEADDIR(dir_inode)))
3405 return ERR_PTR(-ENOENT);
3407 file->f_mode &= ~FMODE_CREATED;
3408 dentry = d_lookup(dir, &nd->last);
3411 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3415 if (d_in_lookup(dentry))
3418 error = d_revalidate(dentry, nd->flags);
3419 if (likely(error > 0))
3423 d_invalidate(dentry);
3427 if (dentry->d_inode) {
3428 /* Cached positive dentry: will open in f_op->open */
3433 * Checking write permission is tricky, bacuse we don't know if we are
3434 * going to actually need it: O_CREAT opens should work as long as the
3435 * file exists. But checking existence breaks atomicity. The trick is
3436 * to check access and if not granted clear O_CREAT from the flags.
3438 * Another problem is returing the "right" error value (e.g. for an
3439 * O_EXCL open we want to return EEXIST not EROFS).
3441 if (unlikely(!got_write))
3442 open_flag &= ~O_TRUNC;
3443 idmap = mnt_idmap(nd->path.mnt);
3444 if (open_flag & O_CREAT) {
3445 if (open_flag & O_EXCL)
3446 open_flag &= ~O_TRUNC;
3447 mode = vfs_prepare_mode(idmap, dir->d_inode, mode, mode, mode);
3448 if (likely(got_write))
3449 create_error = may_o_create(idmap, &nd->path,
3452 create_error = -EROFS;
3455 open_flag &= ~O_CREAT;
3456 if (dir_inode->i_op->atomic_open) {
3457 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3458 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3459 dentry = ERR_PTR(create_error);
3463 if (d_in_lookup(dentry)) {
3464 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3466 d_lookup_done(dentry);
3467 if (unlikely(res)) {
3469 error = PTR_ERR(res);
3477 /* Negative dentry, just create the file */
3478 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3479 file->f_mode |= FMODE_CREATED;
3480 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3481 if (!dir_inode->i_op->create) {
3486 error = dir_inode->i_op->create(idmap, dir_inode, dentry,
3487 mode, open_flag & O_EXCL);
3491 if (unlikely(create_error) && !dentry->d_inode) {
3492 error = create_error;
3499 return ERR_PTR(error);
3502 static const char *open_last_lookups(struct nameidata *nd,
3503 struct file *file, const struct open_flags *op)
3505 struct dentry *dir = nd->path.dentry;
3506 int open_flag = op->open_flag;
3507 bool got_write = false;
3508 struct dentry *dentry;
3511 nd->flags |= op->intent;
3513 if (nd->last_type != LAST_NORM) {
3516 return handle_dots(nd, nd->last_type);
3519 if (!(open_flag & O_CREAT)) {
3520 if (nd->last.name[nd->last.len])
3521 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3522 /* we _can_ be in RCU mode here */
3523 dentry = lookup_fast(nd);
3525 return ERR_CAST(dentry);
3529 if (WARN_ON_ONCE(nd->flags & LOOKUP_RCU))
3530 return ERR_PTR(-ECHILD);
3532 /* create side of things */
3533 if (nd->flags & LOOKUP_RCU) {
3534 if (!try_to_unlazy(nd))
3535 return ERR_PTR(-ECHILD);
3537 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3538 /* trailing slashes? */
3539 if (unlikely(nd->last.name[nd->last.len]))
3540 return ERR_PTR(-EISDIR);
3543 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3544 got_write = !mnt_want_write(nd->path.mnt);
3546 * do _not_ fail yet - we might not need that or fail with
3547 * a different error; let lookup_open() decide; we'll be
3548 * dropping this one anyway.
3551 if (open_flag & O_CREAT)
3552 inode_lock(dir->d_inode);
3554 inode_lock_shared(dir->d_inode);
3555 dentry = lookup_open(nd, file, op, got_write);
3556 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3557 fsnotify_create(dir->d_inode, dentry);
3558 if (open_flag & O_CREAT)
3559 inode_unlock(dir->d_inode);
3561 inode_unlock_shared(dir->d_inode);
3564 mnt_drop_write(nd->path.mnt);
3567 return ERR_CAST(dentry);
3569 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3570 dput(nd->path.dentry);
3571 nd->path.dentry = dentry;
3578 res = step_into(nd, WALK_TRAILING, dentry);
3580 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3585 * Handle the last step of open()
3587 static int do_open(struct nameidata *nd,
3588 struct file *file, const struct open_flags *op)
3590 struct mnt_idmap *idmap;
3591 int open_flag = op->open_flag;
3596 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3597 error = complete_walk(nd);
3601 if (!(file->f_mode & FMODE_CREATED))
3602 audit_inode(nd->name, nd->path.dentry, 0);
3603 idmap = mnt_idmap(nd->path.mnt);
3604 if (open_flag & O_CREAT) {
3605 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3607 if (d_is_dir(nd->path.dentry))
3609 error = may_create_in_sticky(idmap, nd,
3610 d_backing_inode(nd->path.dentry));
3611 if (unlikely(error))
3614 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3617 do_truncate = false;
3618 acc_mode = op->acc_mode;
3619 if (file->f_mode & FMODE_CREATED) {
3620 /* Don't check for write permission, don't truncate */
3621 open_flag &= ~O_TRUNC;
3623 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3624 error = mnt_want_write(nd->path.mnt);
3629 error = may_open(idmap, &nd->path, acc_mode, open_flag);
3630 if (!error && !(file->f_mode & FMODE_OPENED))
3631 error = vfs_open(&nd->path, file);
3633 error = ima_file_check(file, op->acc_mode);
3634 if (!error && do_truncate)
3635 error = handle_truncate(idmap, file);
3636 if (unlikely(error > 0)) {
3641 mnt_drop_write(nd->path.mnt);
3646 * vfs_tmpfile - create tmpfile
3647 * @idmap: idmap of the mount the inode was found from
3648 * @parentpath: pointer to the path of the base directory
3649 * @file: file descriptor of the new tmpfile
3650 * @mode: mode of the new tmpfile
3652 * Create a temporary file.
3654 * If the inode has been found through an idmapped mount the idmap of
3655 * the vfsmount must be passed through @idmap. This function will then take
3656 * care to map the inode according to @idmap before checking permissions.
3657 * On non-idmapped mounts or if permission checking is to be performed on the
3658 * raw inode simply pass @nop_mnt_idmap.
3660 static int vfs_tmpfile(struct mnt_idmap *idmap,
3661 const struct path *parentpath,
3662 struct file *file, umode_t mode)
3664 struct dentry *child;
3665 struct inode *dir = d_inode(parentpath->dentry);
3666 struct inode *inode;
3668 int open_flag = file->f_flags;
3670 /* we want directory to be writable */
3671 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3674 if (!dir->i_op->tmpfile)
3676 child = d_alloc(parentpath->dentry, &slash_name);
3677 if (unlikely(!child))
3679 file->f_path.mnt = parentpath->mnt;
3680 file->f_path.dentry = child;
3681 mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3682 error = dir->i_op->tmpfile(idmap, dir, file, mode);
3686 /* Don't check for other permissions, the inode was just created */
3687 error = may_open(idmap, &file->f_path, 0, file->f_flags);
3690 inode = file_inode(file);
3691 if (!(open_flag & O_EXCL)) {
3692 spin_lock(&inode->i_lock);
3693 inode->i_state |= I_LINKABLE;
3694 spin_unlock(&inode->i_lock);
3696 ima_post_create_tmpfile(idmap, inode);
3701 * kernel_tmpfile_open - open a tmpfile for kernel internal use
3702 * @idmap: idmap of the mount the inode was found from
3703 * @parentpath: path of the base directory
3704 * @mode: mode of the new tmpfile
3706 * @cred: credentials for open
3708 * Create and open a temporary file. The file is not accounted in nr_files,
3709 * hence this is only for kernel internal use, and must not be installed into
3710 * file tables or such.
3712 struct file *kernel_tmpfile_open(struct mnt_idmap *idmap,
3713 const struct path *parentpath,
3714 umode_t mode, int open_flag,
3715 const struct cred *cred)
3720 file = alloc_empty_file_noaccount(open_flag, cred);
3724 error = vfs_tmpfile(idmap, parentpath, file, mode);
3727 file = ERR_PTR(error);
3731 EXPORT_SYMBOL(kernel_tmpfile_open);
3733 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3734 const struct open_flags *op,
3738 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3740 if (unlikely(error))
3742 error = mnt_want_write(path.mnt);
3743 if (unlikely(error))
3745 error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode);
3748 audit_inode(nd->name, file->f_path.dentry, 0);
3750 mnt_drop_write(path.mnt);
3756 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3759 int error = path_lookupat(nd, flags, &path);
3761 audit_inode(nd->name, path.dentry, 0);
3762 error = vfs_open(&path, file);
3768 static struct file *path_openat(struct nameidata *nd,
3769 const struct open_flags *op, unsigned flags)
3774 file = alloc_empty_file(op->open_flag, current_cred());
3778 if (unlikely(file->f_flags & __O_TMPFILE)) {
3779 error = do_tmpfile(nd, flags, op, file);
3780 } else if (unlikely(file->f_flags & O_PATH)) {
3781 error = do_o_path(nd, flags, file);
3783 const char *s = path_init(nd, flags);
3784 while (!(error = link_path_walk(s, nd)) &&
3785 (s = open_last_lookups(nd, file, op)) != NULL)
3788 error = do_open(nd, file, op);
3791 if (likely(!error)) {
3792 if (likely(file->f_mode & FMODE_OPENED))
3798 if (error == -EOPENSTALE) {
3799 if (flags & LOOKUP_RCU)
3804 return ERR_PTR(error);
3807 struct file *do_filp_open(int dfd, struct filename *pathname,
3808 const struct open_flags *op)
3810 struct nameidata nd;
3811 int flags = op->lookup_flags;
3814 set_nameidata(&nd, dfd, pathname, NULL);
3815 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3816 if (unlikely(filp == ERR_PTR(-ECHILD)))
3817 filp = path_openat(&nd, op, flags);
3818 if (unlikely(filp == ERR_PTR(-ESTALE)))
3819 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3820 restore_nameidata();
3824 struct file *do_file_open_root(const struct path *root,
3825 const char *name, const struct open_flags *op)
3827 struct nameidata nd;
3829 struct filename *filename;
3830 int flags = op->lookup_flags;
3832 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3833 return ERR_PTR(-ELOOP);
3835 filename = getname_kernel(name);
3836 if (IS_ERR(filename))
3837 return ERR_CAST(filename);
3839 set_nameidata(&nd, -1, filename, root);
3840 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3841 if (unlikely(file == ERR_PTR(-ECHILD)))
3842 file = path_openat(&nd, op, flags);
3843 if (unlikely(file == ERR_PTR(-ESTALE)))
3844 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3845 restore_nameidata();
3850 static struct dentry *filename_create(int dfd, struct filename *name,
3851 struct path *path, unsigned int lookup_flags)
3853 struct dentry *dentry = ERR_PTR(-EEXIST);
3855 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3856 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3857 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3862 error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3864 return ERR_PTR(error);
3867 * Yucky last component or no last component at all?
3868 * (foo/., foo/.., /////)
3870 if (unlikely(type != LAST_NORM))
3873 /* don't fail immediately if it's r/o, at least try to report other errors */
3874 err2 = mnt_want_write(path->mnt);
3876 * Do the final lookup. Suppress 'create' if there is a trailing
3877 * '/', and a directory wasn't requested.
3879 if (last.name[last.len] && !want_dir)
3881 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3882 dentry = lookup_one_qstr_excl(&last, path->dentry,
3883 reval_flag | create_flags);
3888 if (d_is_positive(dentry))
3892 * Special case - lookup gave negative, but... we had foo/bar/
3893 * From the vfs_mknod() POV we just have a negative dentry -
3894 * all is fine. Let's be bastards - you had / on the end, you've
3895 * been asking for (non-existent) directory. -ENOENT for you.
3897 if (unlikely(!create_flags)) {
3901 if (unlikely(err2)) {
3908 dentry = ERR_PTR(error);
3910 inode_unlock(path->dentry->d_inode);
3912 mnt_drop_write(path->mnt);
3918 struct dentry *kern_path_create(int dfd, const char *pathname,
3919 struct path *path, unsigned int lookup_flags)
3921 struct filename *filename = getname_kernel(pathname);
3922 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3927 EXPORT_SYMBOL(kern_path_create);
3929 void done_path_create(struct path *path, struct dentry *dentry)
3932 inode_unlock(path->dentry->d_inode);
3933 mnt_drop_write(path->mnt);
3936 EXPORT_SYMBOL(done_path_create);
3938 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3939 struct path *path, unsigned int lookup_flags)
3941 struct filename *filename = getname(pathname);
3942 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3947 EXPORT_SYMBOL(user_path_create);
3950 * vfs_mknod - create device node or file
3951 * @idmap: idmap of the mount the inode was found from
3952 * @dir: inode of @dentry
3953 * @dentry: pointer to dentry of the base directory
3954 * @mode: mode of the new device node or file
3955 * @dev: device number of device to create
3957 * Create a device node or file.
3959 * If the inode has been found through an idmapped mount the idmap of
3960 * the vfsmount must be passed through @idmap. This function will then take
3961 * care to map the inode according to @idmap before checking permissions.
3962 * On non-idmapped mounts or if permission checking is to be performed on the
3963 * raw inode simply pass @nop_mnt_idmap.
3965 int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
3966 struct dentry *dentry, umode_t mode, dev_t dev)
3968 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3969 int error = may_create(idmap, dir, dentry);
3974 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3975 !capable(CAP_MKNOD))
3978 if (!dir->i_op->mknod)
3981 mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3982 error = devcgroup_inode_mknod(mode, dev);
3986 error = security_inode_mknod(dir, dentry, mode, dev);
3990 error = dir->i_op->mknod(idmap, dir, dentry, mode, dev);
3992 fsnotify_create(dir, dentry);
3995 EXPORT_SYMBOL(vfs_mknod);
3997 static int may_mknod(umode_t mode)
3999 switch (mode & S_IFMT) {
4005 case 0: /* zero mode translates to S_IFREG */
4014 static int do_mknodat(int dfd, struct filename *name, umode_t mode,
4017 struct mnt_idmap *idmap;
4018 struct dentry *dentry;
4021 unsigned int lookup_flags = 0;
4023 error = may_mknod(mode);
4027 dentry = filename_create(dfd, name, &path, lookup_flags);
4028 error = PTR_ERR(dentry);
4032 error = security_path_mknod(&path, dentry,
4033 mode_strip_umask(path.dentry->d_inode, mode), dev);
4037 idmap = mnt_idmap(path.mnt);
4038 switch (mode & S_IFMT) {
4039 case 0: case S_IFREG:
4040 error = vfs_create(idmap, path.dentry->d_inode,
4041 dentry, mode, true);
4043 ima_post_path_mknod(idmap, dentry);
4045 case S_IFCHR: case S_IFBLK:
4046 error = vfs_mknod(idmap, path.dentry->d_inode,
4047 dentry, mode, new_decode_dev(dev));
4049 case S_IFIFO: case S_IFSOCK:
4050 error = vfs_mknod(idmap, path.dentry->d_inode,
4055 done_path_create(&path, dentry);
4056 if (retry_estale(error, lookup_flags)) {
4057 lookup_flags |= LOOKUP_REVAL;
4065 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
4068 return do_mknodat(dfd, getname(filename), mode, dev);
4071 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
4073 return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
4077 * vfs_mkdir - create directory
4078 * @idmap: idmap of the mount the inode was found from
4079 * @dir: inode of @dentry
4080 * @dentry: pointer to dentry of the base directory
4081 * @mode: mode of the new directory
4083 * Create a directory.
4085 * If the inode has been found through an idmapped mount the idmap of
4086 * the vfsmount must be passed through @idmap. This function will then take
4087 * care to map the inode according to @idmap before checking permissions.
4088 * On non-idmapped mounts or if permission checking is to be performed on the
4089 * raw inode simply pass @nop_mnt_idmap.
4091 int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
4092 struct dentry *dentry, umode_t mode)
4095 unsigned max_links = dir->i_sb->s_max_links;
4097 error = may_create(idmap, dir, dentry);
4101 if (!dir->i_op->mkdir)
4104 mode = vfs_prepare_mode(idmap, dir, mode, S_IRWXUGO | S_ISVTX, 0);
4105 error = security_inode_mkdir(dir, dentry, mode);
4109 if (max_links && dir->i_nlink >= max_links)
4112 error = dir->i_op->mkdir(idmap, dir, dentry, mode);
4114 fsnotify_mkdir(dir, dentry);
4117 EXPORT_SYMBOL(vfs_mkdir);
4119 int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4121 struct dentry *dentry;
4124 unsigned int lookup_flags = LOOKUP_DIRECTORY;
4127 dentry = filename_create(dfd, name, &path, lookup_flags);
4128 error = PTR_ERR(dentry);
4132 error = security_path_mkdir(&path, dentry,
4133 mode_strip_umask(path.dentry->d_inode, mode));
4135 error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode,
4138 done_path_create(&path, dentry);
4139 if (retry_estale(error, lookup_flags)) {
4140 lookup_flags |= LOOKUP_REVAL;
4148 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4150 return do_mkdirat(dfd, getname(pathname), mode);
4153 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4155 return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4159 * vfs_rmdir - remove directory
4160 * @idmap: idmap of the mount the inode was found from
4161 * @dir: inode of @dentry
4162 * @dentry: pointer to dentry of the base directory
4164 * Remove a directory.
4166 * If the inode has been found through an idmapped mount the idmap of
4167 * the vfsmount must be passed through @idmap. This function will then take
4168 * care to map the inode according to @idmap before checking permissions.
4169 * On non-idmapped mounts or if permission checking is to be performed on the
4170 * raw inode simply pass @nop_mnt_idmap.
4172 int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir,
4173 struct dentry *dentry)
4175 int error = may_delete(idmap, dir, dentry, 1);
4180 if (!dir->i_op->rmdir)
4184 inode_lock(dentry->d_inode);
4187 if (is_local_mountpoint(dentry) ||
4188 (dentry->d_inode->i_flags & S_KERNEL_FILE))
4191 error = security_inode_rmdir(dir, dentry);
4195 error = dir->i_op->rmdir(dir, dentry);
4199 shrink_dcache_parent(dentry);
4200 dentry->d_inode->i_flags |= S_DEAD;
4202 detach_mounts(dentry);
4205 inode_unlock(dentry->d_inode);
4208 d_delete_notify(dir, dentry);
4211 EXPORT_SYMBOL(vfs_rmdir);
4213 int do_rmdir(int dfd, struct filename *name)
4216 struct dentry *dentry;
4220 unsigned int lookup_flags = 0;
4222 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4238 error = mnt_want_write(path.mnt);
4242 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4243 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4244 error = PTR_ERR(dentry);
4247 if (!dentry->d_inode) {
4251 error = security_path_rmdir(&path, dentry);
4254 error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry);
4258 inode_unlock(path.dentry->d_inode);
4259 mnt_drop_write(path.mnt);
4262 if (retry_estale(error, lookup_flags)) {
4263 lookup_flags |= LOOKUP_REVAL;
4271 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4273 return do_rmdir(AT_FDCWD, getname(pathname));
4277 * vfs_unlink - unlink a filesystem object
4278 * @idmap: idmap of the mount the inode was found from
4279 * @dir: parent directory
4281 * @delegated_inode: returns victim inode, if the inode is delegated.
4283 * The caller must hold dir->i_mutex.
4285 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4286 * return a reference to the inode in delegated_inode. The caller
4287 * should then break the delegation on that inode and retry. Because
4288 * breaking a delegation may take a long time, the caller should drop
4289 * dir->i_mutex before doing so.
4291 * Alternatively, a caller may pass NULL for delegated_inode. This may
4292 * be appropriate for callers that expect the underlying filesystem not
4293 * to be NFS exported.
4295 * If the inode has been found through an idmapped mount the idmap of
4296 * the vfsmount must be passed through @idmap. This function will then take
4297 * care to map the inode according to @idmap before checking permissions.
4298 * On non-idmapped mounts or if permission checking is to be performed on the
4299 * raw inode simply pass @nop_mnt_idmap.
4301 int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir,
4302 struct dentry *dentry, struct inode **delegated_inode)
4304 struct inode *target = dentry->d_inode;
4305 int error = may_delete(idmap, dir, dentry, 0);
4310 if (!dir->i_op->unlink)
4314 if (IS_SWAPFILE(target))
4316 else if (is_local_mountpoint(dentry))
4319 error = security_inode_unlink(dir, dentry);
4321 error = try_break_deleg(target, delegated_inode);
4324 error = dir->i_op->unlink(dir, dentry);
4327 detach_mounts(dentry);
4332 inode_unlock(target);
4334 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4335 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4336 fsnotify_unlink(dir, dentry);
4337 } else if (!error) {
4338 fsnotify_link_count(target);
4339 d_delete_notify(dir, dentry);
4344 EXPORT_SYMBOL(vfs_unlink);
4347 * Make sure that the actual truncation of the file will occur outside its
4348 * directory's i_mutex. Truncate can take a long time if there is a lot of
4349 * writeout happening, and we don't want to prevent access to the directory
4350 * while waiting on the I/O.
4352 int do_unlinkat(int dfd, struct filename *name)
4355 struct dentry *dentry;
4359 struct inode *inode = NULL;
4360 struct inode *delegated_inode = NULL;
4361 unsigned int lookup_flags = 0;
4363 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4368 if (type != LAST_NORM)
4371 error = mnt_want_write(path.mnt);
4375 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4376 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4377 error = PTR_ERR(dentry);
4378 if (!IS_ERR(dentry)) {
4380 /* Why not before? Because we want correct error value */
4381 if (last.name[last.len] || d_is_negative(dentry))
4383 inode = dentry->d_inode;
4385 error = security_path_unlink(&path, dentry);
4388 error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4389 dentry, &delegated_inode);
4393 inode_unlock(path.dentry->d_inode);
4395 iput(inode); /* truncate the inode here */
4397 if (delegated_inode) {
4398 error = break_deleg_wait(&delegated_inode);
4402 mnt_drop_write(path.mnt);
4405 if (retry_estale(error, lookup_flags)) {
4406 lookup_flags |= LOOKUP_REVAL;
4415 if (d_is_negative(dentry))
4417 else if (d_is_dir(dentry))
4424 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4426 if ((flag & ~AT_REMOVEDIR) != 0)
4429 if (flag & AT_REMOVEDIR)
4430 return do_rmdir(dfd, getname(pathname));
4431 return do_unlinkat(dfd, getname(pathname));
4434 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4436 return do_unlinkat(AT_FDCWD, getname(pathname));
4440 * vfs_symlink - create symlink
4441 * @idmap: idmap of the mount the inode was found from
4442 * @dir: inode of @dentry
4443 * @dentry: pointer to dentry of the base directory
4444 * @oldname: name of the file to link to
4448 * If the inode has been found through an idmapped mount the idmap of
4449 * the vfsmount must be passed through @idmap. This function will then take
4450 * care to map the inode according to @idmap before checking permissions.
4451 * On non-idmapped mounts or if permission checking is to be performed on the
4452 * raw inode simply pass @nop_mnt_idmap.
4454 int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
4455 struct dentry *dentry, const char *oldname)
4459 error = may_create(idmap, dir, dentry);
4463 if (!dir->i_op->symlink)
4466 error = security_inode_symlink(dir, dentry, oldname);
4470 error = dir->i_op->symlink(idmap, dir, dentry, oldname);
4472 fsnotify_create(dir, dentry);
4475 EXPORT_SYMBOL(vfs_symlink);
4477 int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4480 struct dentry *dentry;
4482 unsigned int lookup_flags = 0;
4485 error = PTR_ERR(from);
4489 dentry = filename_create(newdfd, to, &path, lookup_flags);
4490 error = PTR_ERR(dentry);
4494 error = security_path_symlink(&path, dentry, from->name);
4496 error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4497 dentry, from->name);
4498 done_path_create(&path, dentry);
4499 if (retry_estale(error, lookup_flags)) {
4500 lookup_flags |= LOOKUP_REVAL;
4509 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4510 int, newdfd, const char __user *, newname)
4512 return do_symlinkat(getname(oldname), newdfd, getname(newname));
4515 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4517 return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4521 * vfs_link - create a new link
4522 * @old_dentry: object to be linked
4523 * @idmap: idmap of the mount
4525 * @new_dentry: where to create the new link
4526 * @delegated_inode: returns inode needing a delegation break
4528 * The caller must hold dir->i_mutex
4530 * If vfs_link discovers a delegation on the to-be-linked file in need
4531 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4532 * inode in delegated_inode. The caller should then break the delegation
4533 * and retry. Because breaking a delegation may take a long time, the
4534 * caller should drop the i_mutex before doing so.
4536 * Alternatively, a caller may pass NULL for delegated_inode. This may
4537 * be appropriate for callers that expect the underlying filesystem not
4538 * to be NFS exported.
4540 * If the inode has been found through an idmapped mount the idmap of
4541 * the vfsmount must be passed through @idmap. This function will then take
4542 * care to map the inode according to @idmap before checking permissions.
4543 * On non-idmapped mounts or if permission checking is to be performed on the
4544 * raw inode simply pass @nop_mnt_idmap.
4546 int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap,
4547 struct inode *dir, struct dentry *new_dentry,
4548 struct inode **delegated_inode)
4550 struct inode *inode = old_dentry->d_inode;
4551 unsigned max_links = dir->i_sb->s_max_links;
4557 error = may_create(idmap, dir, new_dentry);
4561 if (dir->i_sb != inode->i_sb)
4565 * A link to an append-only or immutable file cannot be created.
4567 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4570 * Updating the link count will likely cause i_uid and i_gid to
4571 * be writen back improperly if their true value is unknown to
4574 if (HAS_UNMAPPED_ID(idmap, inode))
4576 if (!dir->i_op->link)
4578 if (S_ISDIR(inode->i_mode))
4581 error = security_inode_link(old_dentry, dir, new_dentry);
4586 /* Make sure we don't allow creating hardlink to an unlinked file */
4587 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4589 else if (max_links && inode->i_nlink >= max_links)
4592 error = try_break_deleg(inode, delegated_inode);
4594 error = dir->i_op->link(old_dentry, dir, new_dentry);
4597 if (!error && (inode->i_state & I_LINKABLE)) {
4598 spin_lock(&inode->i_lock);
4599 inode->i_state &= ~I_LINKABLE;
4600 spin_unlock(&inode->i_lock);
4602 inode_unlock(inode);
4604 fsnotify_link(dir, inode, new_dentry);
4607 EXPORT_SYMBOL(vfs_link);
4610 * Hardlinks are often used in delicate situations. We avoid
4611 * security-related surprises by not following symlinks on the
4614 * We don't follow them on the oldname either to be compatible
4615 * with linux 2.0, and to avoid hard-linking to directories
4616 * and other special files. --ADM
4618 int do_linkat(int olddfd, struct filename *old, int newdfd,
4619 struct filename *new, int flags)
4621 struct mnt_idmap *idmap;
4622 struct dentry *new_dentry;
4623 struct path old_path, new_path;
4624 struct inode *delegated_inode = NULL;
4628 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4633 * To use null names we require CAP_DAC_READ_SEARCH
4634 * This ensures that not everyone will be able to create
4635 * handlink using the passed filedescriptor.
4637 if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4642 if (flags & AT_SYMLINK_FOLLOW)
4643 how |= LOOKUP_FOLLOW;
4645 error = filename_lookup(olddfd, old, how, &old_path, NULL);
4649 new_dentry = filename_create(newdfd, new, &new_path,
4650 (how & LOOKUP_REVAL));
4651 error = PTR_ERR(new_dentry);
4652 if (IS_ERR(new_dentry))
4656 if (old_path.mnt != new_path.mnt)
4658 idmap = mnt_idmap(new_path.mnt);
4659 error = may_linkat(idmap, &old_path);
4660 if (unlikely(error))
4662 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4665 error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode,
4666 new_dentry, &delegated_inode);
4668 done_path_create(&new_path, new_dentry);
4669 if (delegated_inode) {
4670 error = break_deleg_wait(&delegated_inode);
4672 path_put(&old_path);
4676 if (retry_estale(error, how)) {
4677 path_put(&old_path);
4678 how |= LOOKUP_REVAL;
4682 path_put(&old_path);
4690 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4691 int, newdfd, const char __user *, newname, int, flags)
4693 return do_linkat(olddfd, getname_uflags(oldname, flags),
4694 newdfd, getname(newname), flags);
4697 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4699 return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4703 * vfs_rename - rename a filesystem object
4704 * @rd: pointer to &struct renamedata info
4706 * The caller must hold multiple mutexes--see lock_rename()).
4708 * If vfs_rename discovers a delegation in need of breaking at either
4709 * the source or destination, it will return -EWOULDBLOCK and return a
4710 * reference to the inode in delegated_inode. The caller should then
4711 * break the delegation and retry. Because breaking a delegation may
4712 * take a long time, the caller should drop all locks before doing
4715 * Alternatively, a caller may pass NULL for delegated_inode. This may
4716 * be appropriate for callers that expect the underlying filesystem not
4717 * to be NFS exported.
4719 * The worst of all namespace operations - renaming directory. "Perverted"
4720 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4723 * a) we can get into loop creation.
4724 * b) race potential - two innocent renames can create a loop together.
4725 * That's where 4.4BSD screws up. Current fix: serialization on
4726 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4728 * c) we may have to lock up to _four_ objects - parents and victim (if it exists),
4729 * and source (if it's a non-directory or a subdirectory that moves to
4730 * different parent).
4731 * And that - after we got ->i_mutex on parents (until then we don't know
4732 * whether the target exists). Solution: try to be smart with locking
4733 * order for inodes. We rely on the fact that tree topology may change
4734 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4735 * move will be locked. Thus we can rank directories by the tree
4736 * (ancestors first) and rank all non-directories after them.
4737 * That works since everybody except rename does "lock parent, lookup,
4738 * lock child" and rename is under ->s_vfs_rename_mutex.
4739 * HOWEVER, it relies on the assumption that any object with ->lookup()
4740 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4741 * we'd better make sure that there's no link(2) for them.
4742 * d) conversion from fhandle to dentry may come in the wrong moment - when
4743 * we are removing the target. Solution: we will have to grab ->i_mutex
4744 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4745 * ->i_mutex on parents, which works but leads to some truly excessive
4748 int vfs_rename(struct renamedata *rd)
4751 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4752 struct dentry *old_dentry = rd->old_dentry;
4753 struct dentry *new_dentry = rd->new_dentry;
4754 struct inode **delegated_inode = rd->delegated_inode;
4755 unsigned int flags = rd->flags;
4756 bool is_dir = d_is_dir(old_dentry);
4757 struct inode *source = old_dentry->d_inode;
4758 struct inode *target = new_dentry->d_inode;
4759 bool new_is_dir = false;
4760 unsigned max_links = new_dir->i_sb->s_max_links;
4761 struct name_snapshot old_name;
4762 bool lock_old_subdir, lock_new_subdir;
4764 if (source == target)
4767 error = may_delete(rd->old_mnt_idmap, old_dir, old_dentry, is_dir);
4772 error = may_create(rd->new_mnt_idmap, new_dir, new_dentry);
4774 new_is_dir = d_is_dir(new_dentry);
4776 if (!(flags & RENAME_EXCHANGE))
4777 error = may_delete(rd->new_mnt_idmap, new_dir,
4778 new_dentry, is_dir);
4780 error = may_delete(rd->new_mnt_idmap, new_dir,
4781 new_dentry, new_is_dir);
4786 if (!old_dir->i_op->rename)
4790 * If we are going to change the parent - check write permissions,
4791 * we'll need to flip '..'.
4793 if (new_dir != old_dir) {
4795 error = inode_permission(rd->old_mnt_idmap, source,
4800 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4801 error = inode_permission(rd->new_mnt_idmap, target,
4808 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4813 take_dentry_name_snapshot(&old_name, old_dentry);
4817 * The source subdirectory needs to be locked on cross-directory
4818 * rename or cross-directory exchange since its parent changes.
4819 * The target subdirectory needs to be locked on cross-directory
4820 * exchange due to parent change and on any rename due to becoming
4822 * Non-directories need locking in all cases (for NFS reasons);
4823 * they get locked after any subdirectories (in inode address order).
4825 * NOTE: WE ONLY LOCK UNRELATED DIRECTORIES IN CROSS-DIRECTORY CASE.
4826 * NEVER, EVER DO THAT WITHOUT ->s_vfs_rename_mutex.
4828 lock_old_subdir = new_dir != old_dir;
4829 lock_new_subdir = new_dir != old_dir || !(flags & RENAME_EXCHANGE);
4831 if (lock_old_subdir)
4832 inode_lock_nested(source, I_MUTEX_CHILD);
4833 if (target && (!new_is_dir || lock_new_subdir))
4835 } else if (new_is_dir) {
4836 if (lock_new_subdir)
4837 inode_lock_nested(target, I_MUTEX_CHILD);
4840 lock_two_nondirectories(source, target);
4844 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4848 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4851 if (max_links && new_dir != old_dir) {
4853 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4855 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4856 old_dir->i_nlink >= max_links)
4860 error = try_break_deleg(source, delegated_inode);
4864 if (target && !new_is_dir) {
4865 error = try_break_deleg(target, delegated_inode);
4869 error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry,
4870 new_dir, new_dentry, flags);
4874 if (!(flags & RENAME_EXCHANGE) && target) {
4876 shrink_dcache_parent(new_dentry);
4877 target->i_flags |= S_DEAD;
4879 dont_mount(new_dentry);
4880 detach_mounts(new_dentry);
4882 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4883 if (!(flags & RENAME_EXCHANGE))
4884 d_move(old_dentry, new_dentry);
4886 d_exchange(old_dentry, new_dentry);
4889 if (!is_dir || lock_old_subdir)
4890 inode_unlock(source);
4891 if (target && (!new_is_dir || lock_new_subdir))
4892 inode_unlock(target);
4895 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4896 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4897 if (flags & RENAME_EXCHANGE) {
4898 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4899 new_is_dir, NULL, new_dentry);
4902 release_dentry_name_snapshot(&old_name);
4906 EXPORT_SYMBOL(vfs_rename);
4908 int do_renameat2(int olddfd, struct filename *from, int newdfd,
4909 struct filename *to, unsigned int flags)
4911 struct renamedata rd;
4912 struct dentry *old_dentry, *new_dentry;
4913 struct dentry *trap;
4914 struct path old_path, new_path;
4915 struct qstr old_last, new_last;
4916 int old_type, new_type;
4917 struct inode *delegated_inode = NULL;
4918 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4919 bool should_retry = false;
4920 int error = -EINVAL;
4922 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4925 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4926 (flags & RENAME_EXCHANGE))
4929 if (flags & RENAME_EXCHANGE)
4933 error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4934 &old_last, &old_type);
4938 error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4944 if (old_path.mnt != new_path.mnt)
4948 if (old_type != LAST_NORM)
4951 if (flags & RENAME_NOREPLACE)
4953 if (new_type != LAST_NORM)
4956 error = mnt_want_write(old_path.mnt);
4961 trap = lock_rename(new_path.dentry, old_path.dentry);
4963 error = PTR_ERR(trap);
4964 goto exit_lock_rename;
4967 old_dentry = lookup_one_qstr_excl(&old_last, old_path.dentry,
4969 error = PTR_ERR(old_dentry);
4970 if (IS_ERR(old_dentry))
4972 /* source must exist */
4974 if (d_is_negative(old_dentry))
4976 new_dentry = lookup_one_qstr_excl(&new_last, new_path.dentry,
4977 lookup_flags | target_flags);
4978 error = PTR_ERR(new_dentry);
4979 if (IS_ERR(new_dentry))
4982 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4984 if (flags & RENAME_EXCHANGE) {
4986 if (d_is_negative(new_dentry))
4989 if (!d_is_dir(new_dentry)) {
4991 if (new_last.name[new_last.len])
4995 /* unless the source is a directory trailing slashes give -ENOTDIR */
4996 if (!d_is_dir(old_dentry)) {
4998 if (old_last.name[old_last.len])
5000 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
5003 /* source should not be ancestor of target */
5005 if (old_dentry == trap)
5007 /* target should not be an ancestor of source */
5008 if (!(flags & RENAME_EXCHANGE))
5010 if (new_dentry == trap)
5013 error = security_path_rename(&old_path, old_dentry,
5014 &new_path, new_dentry, flags);
5018 rd.old_dir = old_path.dentry->d_inode;
5019 rd.old_dentry = old_dentry;
5020 rd.old_mnt_idmap = mnt_idmap(old_path.mnt);
5021 rd.new_dir = new_path.dentry->d_inode;
5022 rd.new_dentry = new_dentry;
5023 rd.new_mnt_idmap = mnt_idmap(new_path.mnt);
5024 rd.delegated_inode = &delegated_inode;
5026 error = vfs_rename(&rd);
5032 unlock_rename(new_path.dentry, old_path.dentry);
5034 if (delegated_inode) {
5035 error = break_deleg_wait(&delegated_inode);
5039 mnt_drop_write(old_path.mnt);
5041 if (retry_estale(error, lookup_flags))
5042 should_retry = true;
5043 path_put(&new_path);
5045 path_put(&old_path);
5047 should_retry = false;
5048 lookup_flags |= LOOKUP_REVAL;
5057 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
5058 int, newdfd, const char __user *, newname, unsigned int, flags)
5060 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5064 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
5065 int, newdfd, const char __user *, newname)
5067 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5071 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5073 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
5074 getname(newname), 0);
5077 int readlink_copy(char __user *buffer, int buflen, const char *link)
5079 int len = PTR_ERR(link);
5084 if (len > (unsigned) buflen)
5086 if (copy_to_user(buffer, link, len))
5093 * vfs_readlink - copy symlink body into userspace buffer
5094 * @dentry: dentry on which to get symbolic link
5095 * @buffer: user memory pointer
5096 * @buflen: size of buffer
5098 * Does not touch atime. That's up to the caller if necessary
5100 * Does not call security hook.
5102 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5104 struct inode *inode = d_inode(dentry);
5105 DEFINE_DELAYED_CALL(done);
5109 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5110 if (unlikely(inode->i_op->readlink))
5111 return inode->i_op->readlink(dentry, buffer, buflen);
5113 if (!d_is_symlink(dentry))
5116 spin_lock(&inode->i_lock);
5117 inode->i_opflags |= IOP_DEFAULT_READLINK;
5118 spin_unlock(&inode->i_lock);
5121 link = READ_ONCE(inode->i_link);
5123 link = inode->i_op->get_link(dentry, inode, &done);
5125 return PTR_ERR(link);
5127 res = readlink_copy(buffer, buflen, link);
5128 do_delayed_call(&done);
5131 EXPORT_SYMBOL(vfs_readlink);
5134 * vfs_get_link - get symlink body
5135 * @dentry: dentry on which to get symbolic link
5136 * @done: caller needs to free returned data with this
5138 * Calls security hook and i_op->get_link() on the supplied inode.
5140 * It does not touch atime. That's up to the caller if necessary.
5142 * Does not work on "special" symlinks like /proc/$$/fd/N
5144 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5146 const char *res = ERR_PTR(-EINVAL);
5147 struct inode *inode = d_inode(dentry);
5149 if (d_is_symlink(dentry)) {
5150 res = ERR_PTR(security_inode_readlink(dentry));
5152 res = inode->i_op->get_link(dentry, inode, done);
5156 EXPORT_SYMBOL(vfs_get_link);
5158 /* get the link contents into pagecache */
5159 const char *page_get_link(struct dentry *dentry, struct inode *inode,
5160 struct delayed_call *callback)
5164 struct address_space *mapping = inode->i_mapping;
5167 page = find_get_page(mapping, 0);
5169 return ERR_PTR(-ECHILD);
5170 if (!PageUptodate(page)) {
5172 return ERR_PTR(-ECHILD);
5175 page = read_mapping_page(mapping, 0, NULL);
5179 set_delayed_call(callback, page_put_link, page);
5180 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5181 kaddr = page_address(page);
5182 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5186 EXPORT_SYMBOL(page_get_link);
5188 void page_put_link(void *arg)
5192 EXPORT_SYMBOL(page_put_link);
5194 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5196 DEFINE_DELAYED_CALL(done);
5197 int res = readlink_copy(buffer, buflen,
5198 page_get_link(dentry, d_inode(dentry),
5200 do_delayed_call(&done);
5203 EXPORT_SYMBOL(page_readlink);
5205 int page_symlink(struct inode *inode, const char *symname, int len)
5207 struct address_space *mapping = inode->i_mapping;
5208 const struct address_space_operations *aops = mapping->a_ops;
5209 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5211 void *fsdata = NULL;
5217 flags = memalloc_nofs_save();
5218 err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5220 memalloc_nofs_restore(flags);
5224 memcpy(page_address(page), symname, len-1);
5226 err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5233 mark_inode_dirty(inode);
5238 EXPORT_SYMBOL(page_symlink);
5240 const struct inode_operations page_symlink_inode_operations = {
5241 .get_link = page_get_link,
5243 EXPORT_SYMBOL(page_symlink_inode_operations);