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/namei.h>
24 #include <linux/pagemap.h>
25 #include <linux/fsnotify.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/ima.h>
29 #include <linux/syscalls.h>
30 #include <linux/mount.h>
31 #include <linux/audit.h>
32 #include <linux/capability.h>
33 #include <linux/file.h>
34 #include <linux/fcntl.h>
35 #include <linux/device_cgroup.h>
36 #include <linux/fs_struct.h>
37 #include <linux/posix_acl.h>
38 #include <linux/hash.h>
39 #include <linux/bitops.h>
40 #include <linux/init_task.h>
41 #include <linux/uaccess.h>
46 /* [Feb-1997 T. Schoebel-Theuer]
47 * Fundamental changes in the pathname lookup mechanisms (namei)
48 * were necessary because of omirr. The reason is that omirr needs
49 * to know the _real_ pathname, not the user-supplied one, in case
50 * of symlinks (and also when transname replacements occur).
52 * The new code replaces the old recursive symlink resolution with
53 * an iterative one (in case of non-nested symlink chains). It does
54 * this with calls to <fs>_follow_link().
55 * As a side effect, dir_namei(), _namei() and follow_link() are now
56 * replaced with a single function lookup_dentry() that can handle all
57 * the special cases of the former code.
59 * With the new dcache, the pathname is stored at each inode, at least as
60 * long as the refcount of the inode is positive. As a side effect, the
61 * size of the dcache depends on the inode cache and thus is dynamic.
63 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 * resolution to correspond with current state of the code.
66 * Note that the symlink resolution is not *completely* iterative.
67 * There is still a significant amount of tail- and mid- recursion in
68 * the algorithm. Also, note that <fs>_readlink() is not used in
69 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 * may return different results than <fs>_follow_link(). Many virtual
71 * filesystems (including /proc) exhibit this behavior.
74 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 * and the name already exists in form of a symlink, try to create the new
77 * name indicated by the symlink. The old code always complained that the
78 * name already exists, due to not following the symlink even if its target
79 * is nonexistent. The new semantics affects also mknod() and link() when
80 * the name is a symlink pointing to a non-existent name.
82 * I don't know which semantics is the right one, since I have no access
83 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 * "old" one. Personally, I think the new semantics is much more logical.
86 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 * file does succeed in both HP-UX and SunOs, but not in Solaris
88 * and in the old Linux semantics.
91 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 * semantics. See the comments in "open_namei" and "do_link" below.
94 * [10-Sep-98 Alan Modra] Another symlink change.
97 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 * inside the path - always follow.
99 * in the last component in creation/removal/renaming - never follow.
100 * if LOOKUP_FOLLOW passed - follow.
101 * if the pathname has trailing slashes - follow.
102 * otherwise - don't follow.
103 * (applied in that order).
105 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 * During the 2.4 we need to fix the userland stuff depending on it -
108 * hopefully we will be able to get rid of that wart in 2.5. So far only
109 * XEmacs seems to be relying on it...
112 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 * any extra contention...
117 /* In order to reduce some races, while at the same time doing additional
118 * checking and hopefully speeding things up, we copy filenames to the
119 * kernel data space before using them..
121 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 * PATH_MAX includes the nul terminator --RR.
125 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
128 getname_flags(const char __user *filename, int flags, int *empty)
130 struct filename *result;
134 result = audit_reusename(filename);
138 result = __getname();
139 if (unlikely(!result))
140 return ERR_PTR(-ENOMEM);
143 * First, try to embed the struct filename inside the names_cache
146 kname = (char *)result->iname;
147 result->name = kname;
149 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
150 if (unlikely(len < 0)) {
156 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 * separate struct filename so we can dedicate the entire
158 * names_cache allocation for the pathname, and re-do the copy from
161 if (unlikely(len == EMBEDDED_NAME_MAX)) {
162 const size_t size = offsetof(struct filename, iname[1]);
163 kname = (char *)result;
166 * size is chosen that way we to guarantee that
167 * result->iname[0] is within the same object and that
168 * kname can't be equal to result->iname, no matter what.
170 result = kzalloc(size, GFP_KERNEL);
171 if (unlikely(!result)) {
173 return ERR_PTR(-ENOMEM);
175 result->name = kname;
176 len = strncpy_from_user(kname, filename, PATH_MAX);
177 if (unlikely(len < 0)) {
182 if (unlikely(len == PATH_MAX)) {
185 return ERR_PTR(-ENAMETOOLONG);
190 /* The empty path is special. */
191 if (unlikely(!len)) {
194 if (!(flags & LOOKUP_EMPTY)) {
196 return ERR_PTR(-ENOENT);
200 result->uptr = filename;
201 result->aname = NULL;
202 audit_getname(result);
207 getname(const char __user * filename)
209 return getname_flags(filename, 0, NULL);
213 getname_kernel(const char * filename)
215 struct filename *result;
216 int len = strlen(filename) + 1;
218 result = __getname();
219 if (unlikely(!result))
220 return ERR_PTR(-ENOMEM);
222 if (len <= EMBEDDED_NAME_MAX) {
223 result->name = (char *)result->iname;
224 } else if (len <= PATH_MAX) {
225 const size_t size = offsetof(struct filename, iname[1]);
226 struct filename *tmp;
228 tmp = kmalloc(size, GFP_KERNEL);
229 if (unlikely(!tmp)) {
231 return ERR_PTR(-ENOMEM);
233 tmp->name = (char *)result;
237 return ERR_PTR(-ENAMETOOLONG);
239 memcpy((char *)result->name, filename, len);
241 result->aname = NULL;
243 audit_getname(result);
248 void putname(struct filename *name)
250 if (IS_ERR_OR_NULL(name))
253 BUG_ON(name->refcnt <= 0);
255 if (--name->refcnt > 0)
258 if (name->name != name->iname) {
259 __putname(name->name);
266 * check_acl - perform ACL permission checking
267 * @mnt_userns: user namespace of the mount the inode was found from
268 * @inode: inode to check permissions on
269 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
271 * This function performs the ACL permission checking. Since this function
272 * retrieve POSIX acls it needs to know whether it is called from a blocking or
273 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
275 * If the inode has been found through an idmapped mount the user namespace of
276 * the vfsmount must be passed through @mnt_userns. This function will then take
277 * care to map the inode according to @mnt_userns before checking permissions.
278 * On non-idmapped mounts or if permission checking is to be performed on the
279 * raw inode simply passs init_user_ns.
281 static int check_acl(struct user_namespace *mnt_userns,
282 struct inode *inode, int mask)
284 #ifdef CONFIG_FS_POSIX_ACL
285 struct posix_acl *acl;
287 if (mask & MAY_NOT_BLOCK) {
288 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
291 /* no ->get_acl() calls in RCU mode... */
292 if (is_uncached_acl(acl))
294 return posix_acl_permission(mnt_userns, inode, acl, mask);
297 acl = get_acl(inode, ACL_TYPE_ACCESS);
301 int error = posix_acl_permission(mnt_userns, inode, acl, mask);
302 posix_acl_release(acl);
311 * acl_permission_check - perform basic UNIX permission checking
312 * @mnt_userns: user namespace of the mount the inode was found from
313 * @inode: inode to check permissions on
314 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
316 * This function performs the basic UNIX permission checking. Since this
317 * function may retrieve POSIX acls it needs to know whether it is called from a
318 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
320 * If the inode has been found through an idmapped mount the user namespace of
321 * the vfsmount must be passed through @mnt_userns. This function will then take
322 * care to map the inode according to @mnt_userns before checking permissions.
323 * On non-idmapped mounts or if permission checking is to be performed on the
324 * raw inode simply passs init_user_ns.
326 static int acl_permission_check(struct user_namespace *mnt_userns,
327 struct inode *inode, int mask)
329 unsigned int mode = inode->i_mode;
332 /* Are we the owner? If so, ACL's don't matter */
333 i_uid = i_uid_into_mnt(mnt_userns, inode);
334 if (likely(uid_eq(current_fsuid(), i_uid))) {
337 return (mask & ~mode) ? -EACCES : 0;
340 /* Do we have ACL's? */
341 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
342 int error = check_acl(mnt_userns, inode, mask);
343 if (error != -EAGAIN)
347 /* Only RWX matters for group/other mode bits */
351 * Are the group permissions different from
352 * the other permissions in the bits we care
353 * about? Need to check group ownership if so.
355 if (mask & (mode ^ (mode >> 3))) {
356 kgid_t kgid = i_gid_into_mnt(mnt_userns, inode);
357 if (in_group_p(kgid))
361 /* Bits in 'mode' clear that we require? */
362 return (mask & ~mode) ? -EACCES : 0;
366 * generic_permission - check for access rights on a Posix-like filesystem
367 * @mnt_userns: user namespace of the mount the inode was found from
368 * @inode: inode to check access rights for
369 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
370 * %MAY_NOT_BLOCK ...)
372 * Used to check for read/write/execute permissions on a file.
373 * We use "fsuid" for this, letting us set arbitrary permissions
374 * for filesystem access without changing the "normal" uids which
375 * are used for other things.
377 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
378 * request cannot be satisfied (eg. requires blocking or too much complexity).
379 * It would then be called again in ref-walk mode.
381 * If the inode has been found through an idmapped mount the user namespace of
382 * the vfsmount must be passed through @mnt_userns. This function will then take
383 * care to map the inode according to @mnt_userns before checking permissions.
384 * On non-idmapped mounts or if permission checking is to be performed on the
385 * raw inode simply passs init_user_ns.
387 int generic_permission(struct user_namespace *mnt_userns, struct inode *inode,
393 * Do the basic permission checks.
395 ret = acl_permission_check(mnt_userns, inode, mask);
399 if (S_ISDIR(inode->i_mode)) {
400 /* DACs are overridable for directories */
401 if (!(mask & MAY_WRITE))
402 if (capable_wrt_inode_uidgid(mnt_userns, inode,
403 CAP_DAC_READ_SEARCH))
405 if (capable_wrt_inode_uidgid(mnt_userns, inode,
412 * Searching includes executable on directories, else just read.
414 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
415 if (mask == MAY_READ)
416 if (capable_wrt_inode_uidgid(mnt_userns, inode,
417 CAP_DAC_READ_SEARCH))
420 * Read/write DACs are always overridable.
421 * Executable DACs are overridable when there is
422 * at least one exec bit set.
424 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
425 if (capable_wrt_inode_uidgid(mnt_userns, inode,
431 EXPORT_SYMBOL(generic_permission);
434 * do_inode_permission - UNIX permission checking
435 * @mnt_userns: user namespace of the mount the inode was found from
436 * @inode: inode to check permissions on
437 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
439 * We _really_ want to just do "generic_permission()" without
440 * even looking at the inode->i_op values. So we keep a cache
441 * flag in inode->i_opflags, that says "this has not special
442 * permission function, use the fast case".
444 static inline int do_inode_permission(struct user_namespace *mnt_userns,
445 struct inode *inode, int mask)
447 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
448 if (likely(inode->i_op->permission))
449 return inode->i_op->permission(mnt_userns, inode, mask);
451 /* This gets set once for the inode lifetime */
452 spin_lock(&inode->i_lock);
453 inode->i_opflags |= IOP_FASTPERM;
454 spin_unlock(&inode->i_lock);
456 return generic_permission(mnt_userns, inode, mask);
460 * sb_permission - Check superblock-level permissions
461 * @sb: Superblock of inode to check permission on
462 * @inode: Inode to check permission on
463 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
465 * Separate out file-system wide checks from inode-specific permission checks.
467 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
469 if (unlikely(mask & MAY_WRITE)) {
470 umode_t mode = inode->i_mode;
472 /* Nobody gets write access to a read-only fs. */
473 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
480 * inode_permission - Check for access rights to a given inode
481 * @mnt_userns: User namespace of the mount the inode was found from
482 * @inode: Inode to check permission on
483 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
485 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
486 * this, letting us set arbitrary permissions for filesystem access without
487 * changing the "normal" UIDs which are used for other things.
489 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
491 int inode_permission(struct user_namespace *mnt_userns,
492 struct inode *inode, int mask)
496 retval = sb_permission(inode->i_sb, inode, mask);
500 if (unlikely(mask & MAY_WRITE)) {
502 * Nobody gets write access to an immutable file.
504 if (IS_IMMUTABLE(inode))
508 * Updating mtime will likely cause i_uid and i_gid to be
509 * written back improperly if their true value is unknown
512 if (HAS_UNMAPPED_ID(mnt_userns, inode))
516 retval = do_inode_permission(mnt_userns, inode, mask);
520 retval = devcgroup_inode_permission(inode, mask);
524 return security_inode_permission(inode, mask);
526 EXPORT_SYMBOL(inode_permission);
529 * path_get - get a reference to a path
530 * @path: path to get the reference to
532 * Given a path increment the reference count to the dentry and the vfsmount.
534 void path_get(const struct path *path)
539 EXPORT_SYMBOL(path_get);
542 * path_put - put a reference to a path
543 * @path: path to put the reference to
545 * Given a path decrement the reference count to the dentry and the vfsmount.
547 void path_put(const struct path *path)
552 EXPORT_SYMBOL(path_put);
554 #define EMBEDDED_LEVELS 2
559 struct inode *inode; /* path.dentry.d_inode */
560 unsigned int flags, state;
561 unsigned seq, m_seq, r_seq;
564 int total_link_count;
567 struct delayed_call done;
570 } *stack, internal[EMBEDDED_LEVELS];
571 struct filename *name;
572 struct nameidata *saved;
577 } __randomize_layout;
579 #define ND_ROOT_PRESET 1
580 #define ND_ROOT_GRABBED 2
583 static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
585 struct nameidata *old = current->nameidata;
586 p->stack = p->internal;
591 p->path.dentry = NULL;
592 p->total_link_count = old ? old->total_link_count : 0;
594 current->nameidata = p;
597 static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
598 const struct path *root)
600 __set_nameidata(p, dfd, name);
602 if (unlikely(root)) {
603 p->state = ND_ROOT_PRESET;
608 static void restore_nameidata(void)
610 struct nameidata *now = current->nameidata, *old = now->saved;
612 current->nameidata = old;
614 old->total_link_count = now->total_link_count;
615 if (now->stack != now->internal)
619 static bool nd_alloc_stack(struct nameidata *nd)
623 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
624 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
627 memcpy(p, nd->internal, sizeof(nd->internal));
633 * path_connected - Verify that a dentry is below mnt.mnt_root
635 * Rename can sometimes move a file or directory outside of a bind
636 * mount, path_connected allows those cases to be detected.
638 static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
640 struct super_block *sb = mnt->mnt_sb;
642 /* Bind mounts can have disconnected paths */
643 if (mnt->mnt_root == sb->s_root)
646 return is_subdir(dentry, mnt->mnt_root);
649 static void drop_links(struct nameidata *nd)
653 struct saved *last = nd->stack + i;
654 do_delayed_call(&last->done);
655 clear_delayed_call(&last->done);
659 static void terminate_walk(struct nameidata *nd)
662 if (!(nd->flags & LOOKUP_RCU)) {
665 for (i = 0; i < nd->depth; i++)
666 path_put(&nd->stack[i].link);
667 if (nd->state & ND_ROOT_GRABBED) {
669 nd->state &= ~ND_ROOT_GRABBED;
672 nd->flags &= ~LOOKUP_RCU;
677 nd->path.dentry = NULL;
680 /* path_put is needed afterwards regardless of success or failure */
681 static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
683 int res = __legitimize_mnt(path->mnt, mseq);
690 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
694 return !read_seqcount_retry(&path->dentry->d_seq, seq);
697 static inline bool legitimize_path(struct nameidata *nd,
698 struct path *path, unsigned seq)
700 return __legitimize_path(path, seq, nd->m_seq);
703 static bool legitimize_links(struct nameidata *nd)
706 if (unlikely(nd->flags & LOOKUP_CACHED)) {
711 for (i = 0; i < nd->depth; i++) {
712 struct saved *last = nd->stack + i;
713 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
722 static bool legitimize_root(struct nameidata *nd)
725 * For scoped-lookups (where nd->root has been zeroed), we need to
726 * restart the whole lookup from scratch -- because set_root() is wrong
727 * for these lookups (nd->dfd is the root, not the filesystem root).
729 if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED))
731 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
732 if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
734 nd->state |= ND_ROOT_GRABBED;
735 return legitimize_path(nd, &nd->root, nd->root_seq);
739 * Path walking has 2 modes, rcu-walk and ref-walk (see
740 * Documentation/filesystems/path-lookup.txt). In situations when we can't
741 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
742 * normal reference counts on dentries and vfsmounts to transition to ref-walk
743 * mode. Refcounts are grabbed at the last known good point before rcu-walk
744 * got stuck, so ref-walk may continue from there. If this is not successful
745 * (eg. a seqcount has changed), then failure is returned and it's up to caller
746 * to restart the path walk from the beginning in ref-walk mode.
750 * try_to_unlazy - try to switch to ref-walk mode.
751 * @nd: nameidata pathwalk data
752 * Returns: true on success, false on failure
754 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
756 * Must be called from rcu-walk context.
757 * Nothing should touch nameidata between try_to_unlazy() failure and
760 static bool try_to_unlazy(struct nameidata *nd)
762 struct dentry *parent = nd->path.dentry;
764 BUG_ON(!(nd->flags & LOOKUP_RCU));
766 nd->flags &= ~LOOKUP_RCU;
767 if (unlikely(!legitimize_links(nd)))
769 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
771 if (unlikely(!legitimize_root(nd)))
774 BUG_ON(nd->inode != parent->d_inode);
779 nd->path.dentry = NULL;
786 * try_to_unlazy_next - try to switch to ref-walk mode.
787 * @nd: nameidata pathwalk data
788 * @dentry: next dentry to step into
789 * @seq: seq number to check @dentry against
790 * Returns: true on success, false on failure
792 * Similar to to try_to_unlazy(), but here we have the next dentry already
793 * picked by rcu-walk and want to legitimize that in addition to the current
794 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
795 * Nothing should touch nameidata between try_to_unlazy_next() failure and
798 static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq)
800 BUG_ON(!(nd->flags & LOOKUP_RCU));
802 nd->flags &= ~LOOKUP_RCU;
803 if (unlikely(!legitimize_links(nd)))
805 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
807 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
811 * We need to move both the parent and the dentry from the RCU domain
812 * to be properly refcounted. And the sequence number in the dentry
813 * validates *both* dentry counters, since we checked the sequence
814 * number of the parent after we got the child sequence number. So we
815 * know the parent must still be valid if the child sequence number is
817 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
819 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
822 * Sequence counts matched. Now make sure that the root is
823 * still valid and get it if required.
825 if (unlikely(!legitimize_root(nd)))
833 nd->path.dentry = NULL;
843 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
845 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
846 return dentry->d_op->d_revalidate(dentry, flags);
852 * complete_walk - successful completion of path walk
853 * @nd: pointer nameidata
855 * If we had been in RCU mode, drop out of it and legitimize nd->path.
856 * Revalidate the final result, unless we'd already done that during
857 * the path walk or the filesystem doesn't ask for it. Return 0 on
858 * success, -error on failure. In case of failure caller does not
859 * need to drop nd->path.
861 static int complete_walk(struct nameidata *nd)
863 struct dentry *dentry = nd->path.dentry;
866 if (nd->flags & LOOKUP_RCU) {
868 * We don't want to zero nd->root for scoped-lookups or
869 * externally-managed nd->root.
871 if (!(nd->state & ND_ROOT_PRESET))
872 if (!(nd->flags & LOOKUP_IS_SCOPED))
874 nd->flags &= ~LOOKUP_CACHED;
875 if (!try_to_unlazy(nd))
879 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
881 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
882 * ever step outside the root during lookup" and should already
883 * be guaranteed by the rest of namei, we want to avoid a namei
884 * BUG resulting in userspace being given a path that was not
885 * scoped within the root at some point during the lookup.
887 * So, do a final sanity-check to make sure that in the
888 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
889 * we won't silently return an fd completely outside of the
890 * requested root to userspace.
892 * Userspace could move the path outside the root after this
893 * check, but as discussed elsewhere this is not a concern (the
894 * resolved file was inside the root at some point).
896 if (!path_is_under(&nd->path, &nd->root))
900 if (likely(!(nd->state & ND_JUMPED)))
903 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
906 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
916 static int set_root(struct nameidata *nd)
918 struct fs_struct *fs = current->fs;
921 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
922 * still have to ensure it doesn't happen because it will cause a breakout
925 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
926 return -ENOTRECOVERABLE;
928 if (nd->flags & LOOKUP_RCU) {
932 seq = read_seqcount_begin(&fs->seq);
934 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
935 } while (read_seqcount_retry(&fs->seq, seq));
937 get_fs_root(fs, &nd->root);
938 nd->state |= ND_ROOT_GRABBED;
943 static int nd_jump_root(struct nameidata *nd)
945 if (unlikely(nd->flags & LOOKUP_BENEATH))
947 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
948 /* Absolute path arguments to path_init() are allowed. */
949 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
953 int error = set_root(nd);
957 if (nd->flags & LOOKUP_RCU) {
961 nd->inode = d->d_inode;
962 nd->seq = nd->root_seq;
963 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
969 nd->inode = nd->path.dentry->d_inode;
971 nd->state |= ND_JUMPED;
976 * Helper to directly jump to a known parsed path from ->get_link,
977 * caller must have taken a reference to path beforehand.
979 int nd_jump_link(struct path *path)
982 struct nameidata *nd = current->nameidata;
984 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
988 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
989 if (nd->path.mnt != path->mnt)
992 /* Not currently safe for scoped-lookups. */
993 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
998 nd->inode = nd->path.dentry->d_inode;
999 nd->state |= ND_JUMPED;
1007 static inline void put_link(struct nameidata *nd)
1009 struct saved *last = nd->stack + --nd->depth;
1010 do_delayed_call(&last->done);
1011 if (!(nd->flags & LOOKUP_RCU))
1012 path_put(&last->link);
1015 int sysctl_protected_symlinks __read_mostly = 0;
1016 int sysctl_protected_hardlinks __read_mostly = 0;
1017 int sysctl_protected_fifos __read_mostly;
1018 int sysctl_protected_regular __read_mostly;
1021 * may_follow_link - Check symlink following for unsafe situations
1022 * @nd: nameidata pathwalk data
1024 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1025 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1026 * in a sticky world-writable directory. This is to protect privileged
1027 * processes from failing races against path names that may change out
1028 * from under them by way of other users creating malicious symlinks.
1029 * It will permit symlinks to be followed only when outside a sticky
1030 * world-writable directory, or when the uid of the symlink and follower
1031 * match, or when the directory owner matches the symlink's owner.
1033 * Returns 0 if following the symlink is allowed, -ve on error.
1035 static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1037 struct user_namespace *mnt_userns;
1040 if (!sysctl_protected_symlinks)
1043 mnt_userns = mnt_user_ns(nd->path.mnt);
1044 i_uid = i_uid_into_mnt(mnt_userns, inode);
1045 /* Allowed if owner and follower match. */
1046 if (uid_eq(current_cred()->fsuid, i_uid))
1049 /* Allowed if parent directory not sticky and world-writable. */
1050 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1053 /* Allowed if parent directory and link owner match. */
1054 if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid))
1057 if (nd->flags & LOOKUP_RCU)
1060 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1061 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1066 * safe_hardlink_source - Check for safe hardlink conditions
1067 * @mnt_userns: user namespace of the mount the inode was found from
1068 * @inode: the source inode to hardlink from
1070 * Return false if at least one of the following conditions:
1071 * - inode is not a regular file
1073 * - inode is setgid and group-exec
1074 * - access failure for read and write
1076 * Otherwise returns true.
1078 static bool safe_hardlink_source(struct user_namespace *mnt_userns,
1079 struct inode *inode)
1081 umode_t mode = inode->i_mode;
1083 /* Special files should not get pinned to the filesystem. */
1087 /* Setuid files should not get pinned to the filesystem. */
1091 /* Executable setgid files should not get pinned to the filesystem. */
1092 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1095 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1096 if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE))
1103 * may_linkat - Check permissions for creating a hardlink
1104 * @mnt_userns: user namespace of the mount the inode was found from
1105 * @link: the source to hardlink from
1107 * Block hardlink when all of:
1108 * - sysctl_protected_hardlinks enabled
1109 * - fsuid does not match inode
1110 * - hardlink source is unsafe (see safe_hardlink_source() above)
1111 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1113 * If the inode has been found through an idmapped mount the user namespace of
1114 * the vfsmount must be passed through @mnt_userns. This function will then take
1115 * care to map the inode according to @mnt_userns before checking permissions.
1116 * On non-idmapped mounts or if permission checking is to be performed on the
1117 * raw inode simply passs init_user_ns.
1119 * Returns 0 if successful, -ve on error.
1121 int may_linkat(struct user_namespace *mnt_userns, struct path *link)
1123 struct inode *inode = link->dentry->d_inode;
1125 /* Inode writeback is not safe when the uid or gid are invalid. */
1126 if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
1127 !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
1130 if (!sysctl_protected_hardlinks)
1133 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1134 * otherwise, it must be a safe source.
1136 if (safe_hardlink_source(mnt_userns, inode) ||
1137 inode_owner_or_capable(mnt_userns, inode))
1140 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1145 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1146 * should be allowed, or not, on files that already
1148 * @mnt_userns: user namespace of the mount the inode was found from
1149 * @nd: nameidata pathwalk data
1150 * @inode: the inode of the file to open
1152 * Block an O_CREAT open of a FIFO (or a regular file) when:
1153 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1154 * - the file already exists
1155 * - we are in a sticky directory
1156 * - we don't own the file
1157 * - the owner of the directory doesn't own the file
1158 * - the directory is world writable
1159 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1160 * the directory doesn't have to be world writable: being group writable will
1163 * If the inode has been found through an idmapped mount the user namespace of
1164 * the vfsmount must be passed through @mnt_userns. This function will then take
1165 * care to map the inode according to @mnt_userns before checking permissions.
1166 * On non-idmapped mounts or if permission checking is to be performed on the
1167 * raw inode simply passs init_user_ns.
1169 * Returns 0 if the open is allowed, -ve on error.
1171 static int may_create_in_sticky(struct user_namespace *mnt_userns,
1172 struct nameidata *nd, struct inode *const inode)
1174 umode_t dir_mode = nd->dir_mode;
1175 kuid_t dir_uid = nd->dir_uid;
1177 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1178 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1179 likely(!(dir_mode & S_ISVTX)) ||
1180 uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) ||
1181 uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode)))
1184 if (likely(dir_mode & 0002) ||
1186 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1187 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1188 const char *operation = S_ISFIFO(inode->i_mode) ?
1189 "sticky_create_fifo" :
1190 "sticky_create_regular";
1191 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1198 * follow_up - Find the mountpoint of path's vfsmount
1200 * Given a path, find the mountpoint of its source file system.
1201 * Replace @path with the path of the mountpoint in the parent mount.
1204 * Return 1 if we went up a level and 0 if we were already at the
1207 int follow_up(struct path *path)
1209 struct mount *mnt = real_mount(path->mnt);
1210 struct mount *parent;
1211 struct dentry *mountpoint;
1213 read_seqlock_excl(&mount_lock);
1214 parent = mnt->mnt_parent;
1215 if (parent == mnt) {
1216 read_sequnlock_excl(&mount_lock);
1219 mntget(&parent->mnt);
1220 mountpoint = dget(mnt->mnt_mountpoint);
1221 read_sequnlock_excl(&mount_lock);
1223 path->dentry = mountpoint;
1225 path->mnt = &parent->mnt;
1228 EXPORT_SYMBOL(follow_up);
1230 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1231 struct path *path, unsigned *seqp)
1233 while (mnt_has_parent(m)) {
1234 struct dentry *mountpoint = m->mnt_mountpoint;
1237 if (unlikely(root->dentry == mountpoint &&
1238 root->mnt == &m->mnt))
1240 if (mountpoint != m->mnt.mnt_root) {
1241 path->mnt = &m->mnt;
1242 path->dentry = mountpoint;
1243 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1250 static bool choose_mountpoint(struct mount *m, const struct path *root,
1257 unsigned seq, mseq = read_seqbegin(&mount_lock);
1259 found = choose_mountpoint_rcu(m, root, path, &seq);
1260 if (unlikely(!found)) {
1261 if (!read_seqretry(&mount_lock, mseq))
1264 if (likely(__legitimize_path(path, seq, mseq)))
1276 * Perform an automount
1277 * - return -EISDIR to tell follow_managed() to stop and return the path we
1280 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1282 struct dentry *dentry = path->dentry;
1284 /* We don't want to mount if someone's just doing a stat -
1285 * unless they're stat'ing a directory and appended a '/' to
1288 * We do, however, want to mount if someone wants to open or
1289 * create a file of any type under the mountpoint, wants to
1290 * traverse through the mountpoint or wants to open the
1291 * mounted directory. Also, autofs may mark negative dentries
1292 * as being automount points. These will need the attentions
1293 * of the daemon to instantiate them before they can be used.
1295 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1296 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1300 if (count && (*count)++ >= MAXSYMLINKS)
1303 return finish_automount(dentry->d_op->d_automount(path), path);
1307 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1308 * dentries are pinned but not locked here, so negative dentry can go
1309 * positive right under us. Use of smp_load_acquire() provides a barrier
1310 * sufficient for ->d_inode and ->d_flags consistency.
1312 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1313 int *count, unsigned lookup_flags)
1315 struct vfsmount *mnt = path->mnt;
1316 bool need_mntput = false;
1319 while (flags & DCACHE_MANAGED_DENTRY) {
1320 /* Allow the filesystem to manage the transit without i_mutex
1322 if (flags & DCACHE_MANAGE_TRANSIT) {
1323 ret = path->dentry->d_op->d_manage(path, false);
1324 flags = smp_load_acquire(&path->dentry->d_flags);
1329 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1330 struct vfsmount *mounted = lookup_mnt(path);
1331 if (mounted) { // ... in our namespace
1335 path->mnt = mounted;
1336 path->dentry = dget(mounted->mnt_root);
1337 // here we know it's positive
1338 flags = path->dentry->d_flags;
1344 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1347 // uncovered automount point
1348 ret = follow_automount(path, count, lookup_flags);
1349 flags = smp_load_acquire(&path->dentry->d_flags);
1356 // possible if you race with several mount --move
1357 if (need_mntput && path->mnt == mnt)
1359 if (!ret && unlikely(d_flags_negative(flags)))
1361 *jumped = need_mntput;
1365 static inline int traverse_mounts(struct path *path, bool *jumped,
1366 int *count, unsigned lookup_flags)
1368 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1371 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1373 if (unlikely(d_flags_negative(flags)))
1377 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1380 int follow_down_one(struct path *path)
1382 struct vfsmount *mounted;
1384 mounted = lookup_mnt(path);
1388 path->mnt = mounted;
1389 path->dentry = dget(mounted->mnt_root);
1394 EXPORT_SYMBOL(follow_down_one);
1397 * Follow down to the covering mount currently visible to userspace. At each
1398 * point, the filesystem owning that dentry may be queried as to whether the
1399 * caller is permitted to proceed or not.
1401 int follow_down(struct path *path)
1403 struct vfsmount *mnt = path->mnt;
1405 int ret = traverse_mounts(path, &jumped, NULL, 0);
1407 if (path->mnt != mnt)
1411 EXPORT_SYMBOL(follow_down);
1414 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1415 * we meet a managed dentry that would need blocking.
1417 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1418 struct inode **inode, unsigned *seqp)
1420 struct dentry *dentry = path->dentry;
1421 unsigned int flags = dentry->d_flags;
1423 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1426 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1431 * Don't forget we might have a non-mountpoint managed dentry
1432 * that wants to block transit.
1434 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1435 int res = dentry->d_op->d_manage(path, true);
1437 return res == -EISDIR;
1438 flags = dentry->d_flags;
1441 if (flags & DCACHE_MOUNTED) {
1442 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1444 path->mnt = &mounted->mnt;
1445 dentry = path->dentry = mounted->mnt.mnt_root;
1446 nd->state |= ND_JUMPED;
1447 *seqp = read_seqcount_begin(&dentry->d_seq);
1448 *inode = dentry->d_inode;
1450 * We don't need to re-check ->d_seq after this
1451 * ->d_inode read - there will be an RCU delay
1452 * between mount hash removal and ->mnt_root
1453 * becoming unpinned.
1455 flags = dentry->d_flags;
1458 if (read_seqretry(&mount_lock, nd->m_seq))
1461 return !(flags & DCACHE_NEED_AUTOMOUNT);
1465 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1466 struct path *path, struct inode **inode,
1472 path->mnt = nd->path.mnt;
1473 path->dentry = dentry;
1474 if (nd->flags & LOOKUP_RCU) {
1475 unsigned int seq = *seqp;
1476 if (unlikely(!*inode))
1478 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1480 if (!try_to_unlazy_next(nd, dentry, seq))
1482 // *path might've been clobbered by __follow_mount_rcu()
1483 path->mnt = nd->path.mnt;
1484 path->dentry = dentry;
1486 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1488 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1491 nd->state |= ND_JUMPED;
1493 if (unlikely(ret)) {
1495 if (path->mnt != nd->path.mnt)
1498 *inode = d_backing_inode(path->dentry);
1499 *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1505 * This looks up the name in dcache and possibly revalidates the found dentry.
1506 * NULL is returned if the dentry does not exist in the cache.
1508 static struct dentry *lookup_dcache(const struct qstr *name,
1512 struct dentry *dentry = d_lookup(dir, name);
1514 int error = d_revalidate(dentry, flags);
1515 if (unlikely(error <= 0)) {
1517 d_invalidate(dentry);
1519 return ERR_PTR(error);
1526 * Parent directory has inode locked exclusive. This is one
1527 * and only case when ->lookup() gets called on non in-lookup
1528 * dentries - as the matter of fact, this only gets called
1529 * when directory is guaranteed to have no in-lookup children
1532 static struct dentry *__lookup_hash(const struct qstr *name,
1533 struct dentry *base, unsigned int flags)
1535 struct dentry *dentry = lookup_dcache(name, base, flags);
1537 struct inode *dir = base->d_inode;
1542 /* Don't create child dentry for a dead directory. */
1543 if (unlikely(IS_DEADDIR(dir)))
1544 return ERR_PTR(-ENOENT);
1546 dentry = d_alloc(base, name);
1547 if (unlikely(!dentry))
1548 return ERR_PTR(-ENOMEM);
1550 old = dir->i_op->lookup(dir, dentry, flags);
1551 if (unlikely(old)) {
1558 static struct dentry *lookup_fast(struct nameidata *nd,
1559 struct inode **inode,
1562 struct dentry *dentry, *parent = nd->path.dentry;
1566 * Rename seqlock is not required here because in the off chance
1567 * of a false negative due to a concurrent rename, the caller is
1568 * going to fall back to non-racy lookup.
1570 if (nd->flags & LOOKUP_RCU) {
1572 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1573 if (unlikely(!dentry)) {
1574 if (!try_to_unlazy(nd))
1575 return ERR_PTR(-ECHILD);
1580 * This sequence count validates that the inode matches
1581 * the dentry name information from lookup.
1583 *inode = d_backing_inode(dentry);
1584 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1585 return ERR_PTR(-ECHILD);
1588 * This sequence count validates that the parent had no
1589 * changes while we did the lookup of the dentry above.
1591 * The memory barrier in read_seqcount_begin of child is
1592 * enough, we can use __read_seqcount_retry here.
1594 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1595 return ERR_PTR(-ECHILD);
1598 status = d_revalidate(dentry, nd->flags);
1599 if (likely(status > 0))
1601 if (!try_to_unlazy_next(nd, dentry, seq))
1602 return ERR_PTR(-ECHILD);
1603 if (status == -ECHILD)
1604 /* we'd been told to redo it in non-rcu mode */
1605 status = d_revalidate(dentry, nd->flags);
1607 dentry = __d_lookup(parent, &nd->last);
1608 if (unlikely(!dentry))
1610 status = d_revalidate(dentry, nd->flags);
1612 if (unlikely(status <= 0)) {
1614 d_invalidate(dentry);
1616 return ERR_PTR(status);
1621 /* Fast lookup failed, do it the slow way */
1622 static struct dentry *__lookup_slow(const struct qstr *name,
1626 struct dentry *dentry, *old;
1627 struct inode *inode = dir->d_inode;
1628 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1630 /* Don't go there if it's already dead */
1631 if (unlikely(IS_DEADDIR(inode)))
1632 return ERR_PTR(-ENOENT);
1634 dentry = d_alloc_parallel(dir, name, &wq);
1637 if (unlikely(!d_in_lookup(dentry))) {
1638 int error = d_revalidate(dentry, flags);
1639 if (unlikely(error <= 0)) {
1641 d_invalidate(dentry);
1646 dentry = ERR_PTR(error);
1649 old = inode->i_op->lookup(inode, dentry, flags);
1650 d_lookup_done(dentry);
1651 if (unlikely(old)) {
1659 static struct dentry *lookup_slow(const struct qstr *name,
1663 struct inode *inode = dir->d_inode;
1665 inode_lock_shared(inode);
1666 res = __lookup_slow(name, dir, flags);
1667 inode_unlock_shared(inode);
1671 static inline int may_lookup(struct user_namespace *mnt_userns,
1672 struct nameidata *nd)
1674 if (nd->flags & LOOKUP_RCU) {
1675 int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1676 if (err != -ECHILD || !try_to_unlazy(nd))
1679 return inode_permission(mnt_userns, nd->inode, MAY_EXEC);
1682 static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq)
1684 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1687 if (likely(nd->depth != EMBEDDED_LEVELS))
1689 if (likely(nd->stack != nd->internal))
1691 if (likely(nd_alloc_stack(nd)))
1694 if (nd->flags & LOOKUP_RCU) {
1695 // we need to grab link before we do unlazy. And we can't skip
1696 // unlazy even if we fail to grab the link - cleanup needs it
1697 bool grabbed_link = legitimize_path(nd, link, seq);
1699 if (!try_to_unlazy(nd) != 0 || !grabbed_link)
1702 if (nd_alloc_stack(nd))
1708 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1710 static const char *pick_link(struct nameidata *nd, struct path *link,
1711 struct inode *inode, unsigned seq, int flags)
1715 int error = reserve_stack(nd, link, seq);
1717 if (unlikely(error)) {
1718 if (!(nd->flags & LOOKUP_RCU))
1720 return ERR_PTR(error);
1722 last = nd->stack + nd->depth++;
1724 clear_delayed_call(&last->done);
1727 if (flags & WALK_TRAILING) {
1728 error = may_follow_link(nd, inode);
1729 if (unlikely(error))
1730 return ERR_PTR(error);
1733 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1734 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1735 return ERR_PTR(-ELOOP);
1737 if (!(nd->flags & LOOKUP_RCU)) {
1738 touch_atime(&last->link);
1740 } else if (atime_needs_update(&last->link, inode)) {
1741 if (!try_to_unlazy(nd))
1742 return ERR_PTR(-ECHILD);
1743 touch_atime(&last->link);
1746 error = security_inode_follow_link(link->dentry, inode,
1747 nd->flags & LOOKUP_RCU);
1748 if (unlikely(error))
1749 return ERR_PTR(error);
1751 res = READ_ONCE(inode->i_link);
1753 const char * (*get)(struct dentry *, struct inode *,
1754 struct delayed_call *);
1755 get = inode->i_op->get_link;
1756 if (nd->flags & LOOKUP_RCU) {
1757 res = get(NULL, inode, &last->done);
1758 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1759 res = get(link->dentry, inode, &last->done);
1761 res = get(link->dentry, inode, &last->done);
1769 error = nd_jump_root(nd);
1770 if (unlikely(error))
1771 return ERR_PTR(error);
1772 while (unlikely(*++res == '/'))
1777 all_done: // pure jump
1783 * Do we need to follow links? We _really_ want to be able
1784 * to do this check without having to look at inode->i_op,
1785 * so we keep a cache of "no, this doesn't need follow_link"
1786 * for the common case.
1788 static const char *step_into(struct nameidata *nd, int flags,
1789 struct dentry *dentry, struct inode *inode, unsigned seq)
1792 int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1795 return ERR_PTR(err);
1796 if (likely(!d_is_symlink(path.dentry)) ||
1797 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1798 (flags & WALK_NOFOLLOW)) {
1799 /* not a symlink or should not follow */
1800 if (!(nd->flags & LOOKUP_RCU)) {
1801 dput(nd->path.dentry);
1802 if (nd->path.mnt != path.mnt)
1803 mntput(nd->path.mnt);
1810 if (nd->flags & LOOKUP_RCU) {
1811 /* make sure that d_is_symlink above matches inode */
1812 if (read_seqcount_retry(&path.dentry->d_seq, seq))
1813 return ERR_PTR(-ECHILD);
1815 if (path.mnt == nd->path.mnt)
1818 return pick_link(nd, &path, inode, seq, flags);
1821 static struct dentry *follow_dotdot_rcu(struct nameidata *nd,
1822 struct inode **inodep,
1825 struct dentry *parent, *old;
1827 if (path_equal(&nd->path, &nd->root))
1829 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1832 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1833 &nd->root, &path, &seq))
1835 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1836 return ERR_PTR(-ECHILD);
1838 nd->inode = path.dentry->d_inode;
1840 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1841 return ERR_PTR(-ECHILD);
1842 /* we know that mountpoint was pinned */
1844 old = nd->path.dentry;
1845 parent = old->d_parent;
1846 *inodep = parent->d_inode;
1847 *seqp = read_seqcount_begin(&parent->d_seq);
1848 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1849 return ERR_PTR(-ECHILD);
1850 if (unlikely(!path_connected(nd->path.mnt, parent)))
1851 return ERR_PTR(-ECHILD);
1854 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1855 return ERR_PTR(-ECHILD);
1856 if (unlikely(nd->flags & LOOKUP_BENEATH))
1857 return ERR_PTR(-ECHILD);
1861 static struct dentry *follow_dotdot(struct nameidata *nd,
1862 struct inode **inodep,
1865 struct dentry *parent;
1867 if (path_equal(&nd->path, &nd->root))
1869 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1872 if (!choose_mountpoint(real_mount(nd->path.mnt),
1875 path_put(&nd->path);
1877 nd->inode = path.dentry->d_inode;
1878 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1879 return ERR_PTR(-EXDEV);
1881 /* rare case of legitimate dget_parent()... */
1882 parent = dget_parent(nd->path.dentry);
1883 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1885 return ERR_PTR(-ENOENT);
1888 *inodep = parent->d_inode;
1892 if (unlikely(nd->flags & LOOKUP_BENEATH))
1893 return ERR_PTR(-EXDEV);
1894 dget(nd->path.dentry);
1898 static const char *handle_dots(struct nameidata *nd, int type)
1900 if (type == LAST_DOTDOT) {
1901 const char *error = NULL;
1902 struct dentry *parent;
1903 struct inode *inode;
1906 if (!nd->root.mnt) {
1907 error = ERR_PTR(set_root(nd));
1911 if (nd->flags & LOOKUP_RCU)
1912 parent = follow_dotdot_rcu(nd, &inode, &seq);
1914 parent = follow_dotdot(nd, &inode, &seq);
1916 return ERR_CAST(parent);
1917 if (unlikely(!parent))
1918 error = step_into(nd, WALK_NOFOLLOW,
1919 nd->path.dentry, nd->inode, nd->seq);
1921 error = step_into(nd, WALK_NOFOLLOW,
1922 parent, inode, seq);
1923 if (unlikely(error))
1926 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1928 * If there was a racing rename or mount along our
1929 * path, then we can't be sure that ".." hasn't jumped
1930 * above nd->root (and so userspace should retry or use
1934 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1935 return ERR_PTR(-EAGAIN);
1936 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1937 return ERR_PTR(-EAGAIN);
1943 static const char *walk_component(struct nameidata *nd, int flags)
1945 struct dentry *dentry;
1946 struct inode *inode;
1949 * "." and ".." are special - ".." especially so because it has
1950 * to be able to know about the current root directory and
1951 * parent relationships.
1953 if (unlikely(nd->last_type != LAST_NORM)) {
1954 if (!(flags & WALK_MORE) && nd->depth)
1956 return handle_dots(nd, nd->last_type);
1958 dentry = lookup_fast(nd, &inode, &seq);
1960 return ERR_CAST(dentry);
1961 if (unlikely(!dentry)) {
1962 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1964 return ERR_CAST(dentry);
1966 if (!(flags & WALK_MORE) && nd->depth)
1968 return step_into(nd, flags, dentry, inode, seq);
1972 * We can do the critical dentry name comparison and hashing
1973 * operations one word at a time, but we are limited to:
1975 * - Architectures with fast unaligned word accesses. We could
1976 * do a "get_unaligned()" if this helps and is sufficiently
1979 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1980 * do not trap on the (extremely unlikely) case of a page
1981 * crossing operation.
1983 * - Furthermore, we need an efficient 64-bit compile for the
1984 * 64-bit case in order to generate the "number of bytes in
1985 * the final mask". Again, that could be replaced with a
1986 * efficient population count instruction or similar.
1988 #ifdef CONFIG_DCACHE_WORD_ACCESS
1990 #include <asm/word-at-a-time.h>
1994 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1996 #elif defined(CONFIG_64BIT)
1998 * Register pressure in the mixing function is an issue, particularly
1999 * on 32-bit x86, but almost any function requires one state value and
2000 * one temporary. Instead, use a function designed for two state values
2001 * and no temporaries.
2003 * This function cannot create a collision in only two iterations, so
2004 * we have two iterations to achieve avalanche. In those two iterations,
2005 * we have six layers of mixing, which is enough to spread one bit's
2006 * influence out to 2^6 = 64 state bits.
2008 * Rotate constants are scored by considering either 64 one-bit input
2009 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2010 * probability of that delta causing a change to each of the 128 output
2011 * bits, using a sample of random initial states.
2013 * The Shannon entropy of the computed probabilities is then summed
2014 * to produce a score. Ideally, any input change has a 50% chance of
2015 * toggling any given output bit.
2017 * Mixing scores (in bits) for (12,45):
2018 * Input delta: 1-bit 2-bit
2019 * 1 round: 713.3 42542.6
2020 * 2 rounds: 2753.7 140389.8
2021 * 3 rounds: 5954.1 233458.2
2022 * 4 rounds: 7862.6 256672.2
2023 * Perfect: 8192 258048
2024 * (64*128) (64*63/2 * 128)
2026 #define HASH_MIX(x, y, a) \
2028 y ^= x, x = rol64(x,12),\
2029 x += y, y = rol64(y,45),\
2033 * Fold two longs into one 32-bit hash value. This must be fast, but
2034 * latency isn't quite as critical, as there is a fair bit of additional
2035 * work done before the hash value is used.
2037 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2039 y ^= x * GOLDEN_RATIO_64;
2040 y *= GOLDEN_RATIO_64;
2044 #else /* 32-bit case */
2047 * Mixing scores (in bits) for (7,20):
2048 * Input delta: 1-bit 2-bit
2049 * 1 round: 330.3 9201.6
2050 * 2 rounds: 1246.4 25475.4
2051 * 3 rounds: 1907.1 31295.1
2052 * 4 rounds: 2042.3 31718.6
2053 * Perfect: 2048 31744
2054 * (32*64) (32*31/2 * 64)
2056 #define HASH_MIX(x, y, a) \
2058 y ^= x, x = rol32(x, 7),\
2059 x += y, y = rol32(y,20),\
2062 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2064 /* Use arch-optimized multiply if one exists */
2065 return __hash_32(y ^ __hash_32(x));
2071 * Return the hash of a string of known length. This is carfully
2072 * designed to match hash_name(), which is the more critical function.
2073 * In particular, we must end by hashing a final word containing 0..7
2074 * payload bytes, to match the way that hash_name() iterates until it
2075 * finds the delimiter after the name.
2077 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2079 unsigned long a, x = 0, y = (unsigned long)salt;
2084 a = load_unaligned_zeropad(name);
2085 if (len < sizeof(unsigned long))
2088 name += sizeof(unsigned long);
2089 len -= sizeof(unsigned long);
2091 x ^= a & bytemask_from_count(len);
2093 return fold_hash(x, y);
2095 EXPORT_SYMBOL(full_name_hash);
2097 /* Return the "hash_len" (hash and length) of a null-terminated string */
2098 u64 hashlen_string(const void *salt, const char *name)
2100 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2101 unsigned long adata, mask, len;
2102 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2109 len += sizeof(unsigned long);
2111 a = load_unaligned_zeropad(name+len);
2112 } while (!has_zero(a, &adata, &constants));
2114 adata = prep_zero_mask(a, adata, &constants);
2115 mask = create_zero_mask(adata);
2116 x ^= a & zero_bytemask(mask);
2118 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2120 EXPORT_SYMBOL(hashlen_string);
2123 * Calculate the length and hash of the path component, and
2124 * return the "hash_len" as the result.
2126 static inline u64 hash_name(const void *salt, const char *name)
2128 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2129 unsigned long adata, bdata, mask, len;
2130 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2137 len += sizeof(unsigned long);
2139 a = load_unaligned_zeropad(name+len);
2140 b = a ^ REPEAT_BYTE('/');
2141 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2143 adata = prep_zero_mask(a, adata, &constants);
2144 bdata = prep_zero_mask(b, bdata, &constants);
2145 mask = create_zero_mask(adata | bdata);
2146 x ^= a & zero_bytemask(mask);
2148 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2151 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2153 /* Return the hash of a string of known length */
2154 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2156 unsigned long hash = init_name_hash(salt);
2158 hash = partial_name_hash((unsigned char)*name++, hash);
2159 return end_name_hash(hash);
2161 EXPORT_SYMBOL(full_name_hash);
2163 /* Return the "hash_len" (hash and length) of a null-terminated string */
2164 u64 hashlen_string(const void *salt, const char *name)
2166 unsigned long hash = init_name_hash(salt);
2167 unsigned long len = 0, c;
2169 c = (unsigned char)*name;
2172 hash = partial_name_hash(c, hash);
2173 c = (unsigned char)name[len];
2175 return hashlen_create(end_name_hash(hash), len);
2177 EXPORT_SYMBOL(hashlen_string);
2180 * We know there's a real path component here of at least
2183 static inline u64 hash_name(const void *salt, const char *name)
2185 unsigned long hash = init_name_hash(salt);
2186 unsigned long len = 0, c;
2188 c = (unsigned char)*name;
2191 hash = partial_name_hash(c, hash);
2192 c = (unsigned char)name[len];
2193 } while (c && c != '/');
2194 return hashlen_create(end_name_hash(hash), len);
2201 * This is the basic name resolution function, turning a pathname into
2202 * the final dentry. We expect 'base' to be positive and a directory.
2204 * Returns 0 and nd will have valid dentry and mnt on success.
2205 * Returns error and drops reference to input namei data on failure.
2207 static int link_path_walk(const char *name, struct nameidata *nd)
2209 int depth = 0; // depth <= nd->depth
2212 nd->last_type = LAST_ROOT;
2213 nd->flags |= LOOKUP_PARENT;
2215 return PTR_ERR(name);
2219 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2223 /* At this point we know we have a real path component. */
2225 struct user_namespace *mnt_userns;
2230 mnt_userns = mnt_user_ns(nd->path.mnt);
2231 err = may_lookup(mnt_userns, nd);
2235 hash_len = hash_name(nd->path.dentry, name);
2238 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2240 if (name[1] == '.') {
2242 nd->state |= ND_JUMPED;
2248 if (likely(type == LAST_NORM)) {
2249 struct dentry *parent = nd->path.dentry;
2250 nd->state &= ~ND_JUMPED;
2251 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2252 struct qstr this = { { .hash_len = hash_len }, .name = name };
2253 err = parent->d_op->d_hash(parent, &this);
2256 hash_len = this.hash_len;
2261 nd->last.hash_len = hash_len;
2262 nd->last.name = name;
2263 nd->last_type = type;
2265 name += hashlen_len(hash_len);
2269 * If it wasn't NUL, we know it was '/'. Skip that
2270 * slash, and continue until no more slashes.
2274 } while (unlikely(*name == '/'));
2275 if (unlikely(!*name)) {
2277 /* pathname or trailing symlink, done */
2279 nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode);
2280 nd->dir_mode = nd->inode->i_mode;
2281 nd->flags &= ~LOOKUP_PARENT;
2284 /* last component of nested symlink */
2285 name = nd->stack[--depth].name;
2286 link = walk_component(nd, 0);
2288 /* not the last component */
2289 link = walk_component(nd, WALK_MORE);
2291 if (unlikely(link)) {
2293 return PTR_ERR(link);
2294 /* a symlink to follow */
2295 nd->stack[depth++].name = name;
2299 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2300 if (nd->flags & LOOKUP_RCU) {
2301 if (!try_to_unlazy(nd))
2309 /* must be paired with terminate_walk() */
2310 static const char *path_init(struct nameidata *nd, unsigned flags)
2313 const char *s = nd->name->name;
2315 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2316 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2317 return ERR_PTR(-EAGAIN);
2320 flags &= ~LOOKUP_RCU;
2321 if (flags & LOOKUP_RCU)
2325 nd->state |= ND_JUMPED;
2327 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2328 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2331 if (nd->state & ND_ROOT_PRESET) {
2332 struct dentry *root = nd->root.dentry;
2333 struct inode *inode = root->d_inode;
2334 if (*s && unlikely(!d_can_lookup(root)))
2335 return ERR_PTR(-ENOTDIR);
2336 nd->path = nd->root;
2338 if (flags & LOOKUP_RCU) {
2339 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2340 nd->root_seq = nd->seq;
2342 path_get(&nd->path);
2347 nd->root.mnt = NULL;
2349 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2350 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2351 error = nd_jump_root(nd);
2352 if (unlikely(error))
2353 return ERR_PTR(error);
2357 /* Relative pathname -- get the starting-point it is relative to. */
2358 if (nd->dfd == AT_FDCWD) {
2359 if (flags & LOOKUP_RCU) {
2360 struct fs_struct *fs = current->fs;
2364 seq = read_seqcount_begin(&fs->seq);
2366 nd->inode = nd->path.dentry->d_inode;
2367 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2368 } while (read_seqcount_retry(&fs->seq, seq));
2370 get_fs_pwd(current->fs, &nd->path);
2371 nd->inode = nd->path.dentry->d_inode;
2374 /* Caller must check execute permissions on the starting path component */
2375 struct fd f = fdget_raw(nd->dfd);
2376 struct dentry *dentry;
2379 return ERR_PTR(-EBADF);
2381 dentry = f.file->f_path.dentry;
2383 if (*s && unlikely(!d_can_lookup(dentry))) {
2385 return ERR_PTR(-ENOTDIR);
2388 nd->path = f.file->f_path;
2389 if (flags & LOOKUP_RCU) {
2390 nd->inode = nd->path.dentry->d_inode;
2391 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2393 path_get(&nd->path);
2394 nd->inode = nd->path.dentry->d_inode;
2399 /* For scoped-lookups we need to set the root to the dirfd as well. */
2400 if (flags & LOOKUP_IS_SCOPED) {
2401 nd->root = nd->path;
2402 if (flags & LOOKUP_RCU) {
2403 nd->root_seq = nd->seq;
2405 path_get(&nd->root);
2406 nd->state |= ND_ROOT_GRABBED;
2412 static inline const char *lookup_last(struct nameidata *nd)
2414 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2415 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2417 return walk_component(nd, WALK_TRAILING);
2420 static int handle_lookup_down(struct nameidata *nd)
2422 if (!(nd->flags & LOOKUP_RCU))
2423 dget(nd->path.dentry);
2424 return PTR_ERR(step_into(nd, WALK_NOFOLLOW,
2425 nd->path.dentry, nd->inode, nd->seq));
2428 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2429 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2431 const char *s = path_init(nd, flags);
2434 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2435 err = handle_lookup_down(nd);
2436 if (unlikely(err < 0))
2440 while (!(err = link_path_walk(s, nd)) &&
2441 (s = lookup_last(nd)) != NULL)
2443 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2444 err = handle_lookup_down(nd);
2445 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2448 err = complete_walk(nd);
2450 if (!err && nd->flags & LOOKUP_DIRECTORY)
2451 if (!d_can_lookup(nd->path.dentry))
2455 nd->path.mnt = NULL;
2456 nd->path.dentry = NULL;
2462 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2463 struct path *path, struct path *root)
2466 struct nameidata nd;
2468 return PTR_ERR(name);
2469 set_nameidata(&nd, dfd, name, root);
2470 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2471 if (unlikely(retval == -ECHILD))
2472 retval = path_lookupat(&nd, flags, path);
2473 if (unlikely(retval == -ESTALE))
2474 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2476 if (likely(!retval))
2477 audit_inode(name, path->dentry,
2478 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2479 restore_nameidata();
2484 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2485 static int path_parentat(struct nameidata *nd, unsigned flags,
2486 struct path *parent)
2488 const char *s = path_init(nd, flags);
2489 int err = link_path_walk(s, nd);
2491 err = complete_walk(nd);
2494 nd->path.mnt = NULL;
2495 nd->path.dentry = NULL;
2501 static int __filename_parentat(int dfd, struct filename *name,
2502 unsigned int flags, struct path *parent,
2503 struct qstr *last, int *type)
2506 struct nameidata nd;
2509 return PTR_ERR(name);
2510 set_nameidata(&nd, dfd, name, NULL);
2511 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2512 if (unlikely(retval == -ECHILD))
2513 retval = path_parentat(&nd, flags, parent);
2514 if (unlikely(retval == -ESTALE))
2515 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2516 if (likely(!retval)) {
2518 *type = nd.last_type;
2519 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2521 restore_nameidata();
2525 static int filename_parentat(int dfd, struct filename *name,
2526 unsigned int flags, struct path *parent,
2527 struct qstr *last, int *type)
2529 int retval = __filename_parentat(dfd, name, flags, parent, last, type);
2535 /* does lookup, returns the object with parent locked */
2536 struct dentry *kern_path_locked(const char *name, struct path *path)
2542 error = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2545 return ERR_PTR(error);
2546 if (unlikely(type != LAST_NORM)) {
2548 return ERR_PTR(-EINVAL);
2550 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2551 d = __lookup_hash(&last, path->dentry, 0);
2553 inode_unlock(path->dentry->d_inode);
2559 int kern_path(const char *name, unsigned int flags, struct path *path)
2561 return filename_lookup(AT_FDCWD, getname_kernel(name),
2564 EXPORT_SYMBOL(kern_path);
2567 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2568 * @dentry: pointer to dentry of the base directory
2569 * @mnt: pointer to vfs mount of the base directory
2570 * @name: pointer to file name
2571 * @flags: lookup flags
2572 * @path: pointer to struct path to fill
2574 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2575 const char *name, unsigned int flags,
2578 struct path root = {.mnt = mnt, .dentry = dentry};
2579 /* the first argument of filename_lookup() is ignored with root */
2580 return filename_lookup(AT_FDCWD, getname_kernel(name),
2581 flags , path, &root);
2583 EXPORT_SYMBOL(vfs_path_lookup);
2585 static int lookup_one_len_common(const char *name, struct dentry *base,
2586 int len, struct qstr *this)
2590 this->hash = full_name_hash(base, name, len);
2594 if (unlikely(name[0] == '.')) {
2595 if (len < 2 || (len == 2 && name[1] == '.'))
2600 unsigned int c = *(const unsigned char *)name++;
2601 if (c == '/' || c == '\0')
2605 * See if the low-level filesystem might want
2606 * to use its own hash..
2608 if (base->d_flags & DCACHE_OP_HASH) {
2609 int err = base->d_op->d_hash(base, this);
2614 return inode_permission(&init_user_ns, base->d_inode, MAY_EXEC);
2618 * try_lookup_one_len - filesystem helper to lookup single pathname component
2619 * @name: pathname component to lookup
2620 * @base: base directory to lookup from
2621 * @len: maximum length @len should be interpreted to
2623 * Look up a dentry by name in the dcache, returning NULL if it does not
2624 * currently exist. The function does not try to create a dentry.
2626 * Note that this routine is purely a helper for filesystem usage and should
2627 * not be called by generic code.
2629 * The caller must hold base->i_mutex.
2631 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2636 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2638 err = lookup_one_len_common(name, base, len, &this);
2640 return ERR_PTR(err);
2642 return lookup_dcache(&this, base, 0);
2644 EXPORT_SYMBOL(try_lookup_one_len);
2647 * lookup_one_len - filesystem helper to lookup single pathname component
2648 * @name: pathname component to lookup
2649 * @base: base directory to lookup from
2650 * @len: maximum length @len should be interpreted to
2652 * Note that this routine is purely a helper for filesystem usage and should
2653 * not be called by generic code.
2655 * The caller must hold base->i_mutex.
2657 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2659 struct dentry *dentry;
2663 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2665 err = lookup_one_len_common(name, base, len, &this);
2667 return ERR_PTR(err);
2669 dentry = lookup_dcache(&this, base, 0);
2670 return dentry ? dentry : __lookup_slow(&this, base, 0);
2672 EXPORT_SYMBOL(lookup_one_len);
2675 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2676 * @name: pathname component to lookup
2677 * @base: base directory to lookup from
2678 * @len: maximum length @len should be interpreted to
2680 * Note that this routine is purely a helper for filesystem usage and should
2681 * not be called by generic code.
2683 * Unlike lookup_one_len, it should be called without the parent
2684 * i_mutex held, and will take the i_mutex itself if necessary.
2686 struct dentry *lookup_one_len_unlocked(const char *name,
2687 struct dentry *base, int len)
2693 err = lookup_one_len_common(name, base, len, &this);
2695 return ERR_PTR(err);
2697 ret = lookup_dcache(&this, base, 0);
2699 ret = lookup_slow(&this, base, 0);
2702 EXPORT_SYMBOL(lookup_one_len_unlocked);
2705 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2706 * on negatives. Returns known positive or ERR_PTR(); that's what
2707 * most of the users want. Note that pinned negative with unlocked parent
2708 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2709 * need to be very careful; pinned positives have ->d_inode stable, so
2710 * this one avoids such problems.
2712 struct dentry *lookup_positive_unlocked(const char *name,
2713 struct dentry *base, int len)
2715 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2716 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2718 ret = ERR_PTR(-ENOENT);
2722 EXPORT_SYMBOL(lookup_positive_unlocked);
2724 #ifdef CONFIG_UNIX98_PTYS
2725 int path_pts(struct path *path)
2727 /* Find something mounted on "pts" in the same directory as
2730 struct dentry *parent = dget_parent(path->dentry);
2731 struct dentry *child;
2732 struct qstr this = QSTR_INIT("pts", 3);
2734 if (unlikely(!path_connected(path->mnt, parent))) {
2739 path->dentry = parent;
2740 child = d_hash_and_lookup(parent, &this);
2744 path->dentry = child;
2751 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2752 struct path *path, int *empty)
2754 return filename_lookup(dfd, getname_flags(name, flags, empty),
2757 EXPORT_SYMBOL(user_path_at_empty);
2759 int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir,
2760 struct inode *inode)
2762 kuid_t fsuid = current_fsuid();
2764 if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid))
2766 if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid))
2768 return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER);
2770 EXPORT_SYMBOL(__check_sticky);
2773 * Check whether we can remove a link victim from directory dir, check
2774 * whether the type of victim is right.
2775 * 1. We can't do it if dir is read-only (done in permission())
2776 * 2. We should have write and exec permissions on dir
2777 * 3. We can't remove anything from append-only dir
2778 * 4. We can't do anything with immutable dir (done in permission())
2779 * 5. If the sticky bit on dir is set we should either
2780 * a. be owner of dir, or
2781 * b. be owner of victim, or
2782 * c. have CAP_FOWNER capability
2783 * 6. If the victim is append-only or immutable we can't do antyhing with
2784 * links pointing to it.
2785 * 7. If the victim has an unknown uid or gid we can't change the inode.
2786 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2787 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2788 * 10. We can't remove a root or mountpoint.
2789 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2790 * nfs_async_unlink().
2792 static int may_delete(struct user_namespace *mnt_userns, struct inode *dir,
2793 struct dentry *victim, bool isdir)
2795 struct inode *inode = d_backing_inode(victim);
2798 if (d_is_negative(victim))
2802 BUG_ON(victim->d_parent->d_inode != dir);
2804 /* Inode writeback is not safe when the uid or gid are invalid. */
2805 if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
2806 !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
2809 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2811 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2817 if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) ||
2818 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2819 HAS_UNMAPPED_ID(mnt_userns, inode))
2822 if (!d_is_dir(victim))
2824 if (IS_ROOT(victim))
2826 } else if (d_is_dir(victim))
2828 if (IS_DEADDIR(dir))
2830 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2835 /* Check whether we can create an object with dentry child in directory
2837 * 1. We can't do it if child already exists (open has special treatment for
2838 * this case, but since we are inlined it's OK)
2839 * 2. We can't do it if dir is read-only (done in permission())
2840 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2841 * 4. We should have write and exec permissions on dir
2842 * 5. We can't do it if dir is immutable (done in permission())
2844 static inline int may_create(struct user_namespace *mnt_userns,
2845 struct inode *dir, struct dentry *child)
2847 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2850 if (IS_DEADDIR(dir))
2852 if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
2855 return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2859 * p1 and p2 should be directories on the same fs.
2861 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2866 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2870 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2872 p = d_ancestor(p2, p1);
2874 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2875 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2879 p = d_ancestor(p1, p2);
2881 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2882 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2886 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2887 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2890 EXPORT_SYMBOL(lock_rename);
2892 void unlock_rename(struct dentry *p1, struct dentry *p2)
2894 inode_unlock(p1->d_inode);
2896 inode_unlock(p2->d_inode);
2897 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2900 EXPORT_SYMBOL(unlock_rename);
2903 * vfs_create - create new file
2904 * @mnt_userns: user namespace of the mount the inode was found from
2905 * @dir: inode of @dentry
2906 * @dentry: pointer to dentry of the base directory
2907 * @mode: mode of the new file
2908 * @want_excl: whether the file must not yet exist
2910 * Create a new file.
2912 * If the inode has been found through an idmapped mount the user namespace of
2913 * the vfsmount must be passed through @mnt_userns. This function will then take
2914 * care to map the inode according to @mnt_userns before checking permissions.
2915 * On non-idmapped mounts or if permission checking is to be performed on the
2916 * raw inode simply passs init_user_ns.
2918 int vfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2919 struct dentry *dentry, umode_t mode, bool want_excl)
2921 int error = may_create(mnt_userns, dir, dentry);
2925 if (!dir->i_op->create)
2926 return -EACCES; /* shouldn't it be ENOSYS? */
2929 error = security_inode_create(dir, dentry, mode);
2932 error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl);
2934 fsnotify_create(dir, dentry);
2937 EXPORT_SYMBOL(vfs_create);
2939 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2940 int (*f)(struct dentry *, umode_t, void *),
2943 struct inode *dir = dentry->d_parent->d_inode;
2944 int error = may_create(&init_user_ns, dir, dentry);
2950 error = security_inode_create(dir, dentry, mode);
2953 error = f(dentry, mode, arg);
2955 fsnotify_create(dir, dentry);
2958 EXPORT_SYMBOL(vfs_mkobj);
2960 bool may_open_dev(const struct path *path)
2962 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2963 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2966 static int may_open(struct user_namespace *mnt_userns, const struct path *path,
2967 int acc_mode, int flag)
2969 struct dentry *dentry = path->dentry;
2970 struct inode *inode = dentry->d_inode;
2976 switch (inode->i_mode & S_IFMT) {
2980 if (acc_mode & MAY_WRITE)
2982 if (acc_mode & MAY_EXEC)
2987 if (!may_open_dev(path))
2992 if (acc_mode & MAY_EXEC)
2997 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3002 error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode);
3007 * An append-only file must be opened in append mode for writing.
3009 if (IS_APPEND(inode)) {
3010 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3016 /* O_NOATIME can only be set by the owner or superuser */
3017 if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode))
3023 static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp)
3025 const struct path *path = &filp->f_path;
3026 struct inode *inode = path->dentry->d_inode;
3027 int error = get_write_access(inode);
3031 * Refuse to truncate files with mandatory locks held on them.
3033 error = locks_verify_locked(filp);
3035 error = security_path_truncate(path);
3037 error = do_truncate(mnt_userns, path->dentry, 0,
3038 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3041 put_write_access(inode);
3045 static inline int open_to_namei_flags(int flag)
3047 if ((flag & O_ACCMODE) == 3)
3052 static int may_o_create(struct user_namespace *mnt_userns,
3053 const struct path *dir, struct dentry *dentry,
3056 int error = security_path_mknod(dir, dentry, mode, 0);
3060 if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns))
3063 error = inode_permission(mnt_userns, dir->dentry->d_inode,
3064 MAY_WRITE | MAY_EXEC);
3068 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3072 * Attempt to atomically look up, create and open a file from a negative
3075 * Returns 0 if successful. The file will have been created and attached to
3076 * @file by the filesystem calling finish_open().
3078 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3079 * be set. The caller will need to perform the open themselves. @path will
3080 * have been updated to point to the new dentry. This may be negative.
3082 * Returns an error code otherwise.
3084 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3086 int open_flag, umode_t mode)
3088 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3089 struct inode *dir = nd->path.dentry->d_inode;
3092 if (nd->flags & LOOKUP_DIRECTORY)
3093 open_flag |= O_DIRECTORY;
3095 file->f_path.dentry = DENTRY_NOT_SET;
3096 file->f_path.mnt = nd->path.mnt;
3097 error = dir->i_op->atomic_open(dir, dentry, file,
3098 open_to_namei_flags(open_flag), mode);
3099 d_lookup_done(dentry);
3101 if (file->f_mode & FMODE_OPENED) {
3102 if (unlikely(dentry != file->f_path.dentry)) {
3104 dentry = dget(file->f_path.dentry);
3106 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3109 if (file->f_path.dentry) {
3111 dentry = file->f_path.dentry;
3113 if (unlikely(d_is_negative(dentry)))
3119 dentry = ERR_PTR(error);
3125 * Look up and maybe create and open the last component.
3127 * Must be called with parent locked (exclusive in O_CREAT case).
3129 * Returns 0 on success, that is, if
3130 * the file was successfully atomically created (if necessary) and opened, or
3131 * the file was not completely opened at this time, though lookups and
3132 * creations were performed.
3133 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3134 * In the latter case dentry returned in @path might be negative if O_CREAT
3135 * hadn't been specified.
3137 * An error code is returned on failure.
3139 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3140 const struct open_flags *op,
3143 struct user_namespace *mnt_userns;
3144 struct dentry *dir = nd->path.dentry;
3145 struct inode *dir_inode = dir->d_inode;
3146 int open_flag = op->open_flag;
3147 struct dentry *dentry;
3148 int error, create_error = 0;
3149 umode_t mode = op->mode;
3150 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3152 if (unlikely(IS_DEADDIR(dir_inode)))
3153 return ERR_PTR(-ENOENT);
3155 file->f_mode &= ~FMODE_CREATED;
3156 dentry = d_lookup(dir, &nd->last);
3159 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3163 if (d_in_lookup(dentry))
3166 error = d_revalidate(dentry, nd->flags);
3167 if (likely(error > 0))
3171 d_invalidate(dentry);
3175 if (dentry->d_inode) {
3176 /* Cached positive dentry: will open in f_op->open */
3181 * Checking write permission is tricky, bacuse we don't know if we are
3182 * going to actually need it: O_CREAT opens should work as long as the
3183 * file exists. But checking existence breaks atomicity. The trick is
3184 * to check access and if not granted clear O_CREAT from the flags.
3186 * Another problem is returing the "right" error value (e.g. for an
3187 * O_EXCL open we want to return EEXIST not EROFS).
3189 if (unlikely(!got_write))
3190 open_flag &= ~O_TRUNC;
3191 mnt_userns = mnt_user_ns(nd->path.mnt);
3192 if (open_flag & O_CREAT) {
3193 if (open_flag & O_EXCL)
3194 open_flag &= ~O_TRUNC;
3195 if (!IS_POSIXACL(dir->d_inode))
3196 mode &= ~current_umask();
3197 if (likely(got_write))
3198 create_error = may_o_create(mnt_userns, &nd->path,
3201 create_error = -EROFS;
3204 open_flag &= ~O_CREAT;
3205 if (dir_inode->i_op->atomic_open) {
3206 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3207 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3208 dentry = ERR_PTR(create_error);
3212 if (d_in_lookup(dentry)) {
3213 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3215 d_lookup_done(dentry);
3216 if (unlikely(res)) {
3218 error = PTR_ERR(res);
3226 /* Negative dentry, just create the file */
3227 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3228 file->f_mode |= FMODE_CREATED;
3229 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3230 if (!dir_inode->i_op->create) {
3235 error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry,
3236 mode, open_flag & O_EXCL);
3240 if (unlikely(create_error) && !dentry->d_inode) {
3241 error = create_error;
3248 return ERR_PTR(error);
3251 static const char *open_last_lookups(struct nameidata *nd,
3252 struct file *file, const struct open_flags *op)
3254 struct dentry *dir = nd->path.dentry;
3255 int open_flag = op->open_flag;
3256 bool got_write = false;
3258 struct inode *inode;
3259 struct dentry *dentry;
3262 nd->flags |= op->intent;
3264 if (nd->last_type != LAST_NORM) {
3267 return handle_dots(nd, nd->last_type);
3270 if (!(open_flag & O_CREAT)) {
3271 if (nd->last.name[nd->last.len])
3272 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3273 /* we _can_ be in RCU mode here */
3274 dentry = lookup_fast(nd, &inode, &seq);
3276 return ERR_CAST(dentry);
3280 BUG_ON(nd->flags & LOOKUP_RCU);
3282 /* create side of things */
3283 if (nd->flags & LOOKUP_RCU) {
3284 if (!try_to_unlazy(nd))
3285 return ERR_PTR(-ECHILD);
3287 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3288 /* trailing slashes? */
3289 if (unlikely(nd->last.name[nd->last.len]))
3290 return ERR_PTR(-EISDIR);
3293 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3294 got_write = !mnt_want_write(nd->path.mnt);
3296 * do _not_ fail yet - we might not need that or fail with
3297 * a different error; let lookup_open() decide; we'll be
3298 * dropping this one anyway.
3301 if (open_flag & O_CREAT)
3302 inode_lock(dir->d_inode);
3304 inode_lock_shared(dir->d_inode);
3305 dentry = lookup_open(nd, file, op, got_write);
3306 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3307 fsnotify_create(dir->d_inode, dentry);
3308 if (open_flag & O_CREAT)
3309 inode_unlock(dir->d_inode);
3311 inode_unlock_shared(dir->d_inode);
3314 mnt_drop_write(nd->path.mnt);
3317 return ERR_CAST(dentry);
3319 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3320 dput(nd->path.dentry);
3321 nd->path.dentry = dentry;
3328 res = step_into(nd, WALK_TRAILING, dentry, inode, seq);
3330 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3335 * Handle the last step of open()
3337 static int do_open(struct nameidata *nd,
3338 struct file *file, const struct open_flags *op)
3340 struct user_namespace *mnt_userns;
3341 int open_flag = op->open_flag;
3346 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3347 error = complete_walk(nd);
3351 if (!(file->f_mode & FMODE_CREATED))
3352 audit_inode(nd->name, nd->path.dentry, 0);
3353 mnt_userns = mnt_user_ns(nd->path.mnt);
3354 if (open_flag & O_CREAT) {
3355 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3357 if (d_is_dir(nd->path.dentry))
3359 error = may_create_in_sticky(mnt_userns, nd,
3360 d_backing_inode(nd->path.dentry));
3361 if (unlikely(error))
3364 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3367 do_truncate = false;
3368 acc_mode = op->acc_mode;
3369 if (file->f_mode & FMODE_CREATED) {
3370 /* Don't check for write permission, don't truncate */
3371 open_flag &= ~O_TRUNC;
3373 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3374 error = mnt_want_write(nd->path.mnt);
3379 error = may_open(mnt_userns, &nd->path, acc_mode, open_flag);
3380 if (!error && !(file->f_mode & FMODE_OPENED))
3381 error = vfs_open(&nd->path, file);
3383 error = ima_file_check(file, op->acc_mode);
3384 if (!error && do_truncate)
3385 error = handle_truncate(mnt_userns, file);
3386 if (unlikely(error > 0)) {
3391 mnt_drop_write(nd->path.mnt);
3396 * vfs_tmpfile - create tmpfile
3397 * @mnt_userns: user namespace of the mount the inode was found from
3398 * @dentry: pointer to dentry of the base directory
3399 * @mode: mode of the new tmpfile
3402 * Create a temporary file.
3404 * If the inode has been found through an idmapped mount the user namespace of
3405 * the vfsmount must be passed through @mnt_userns. This function will then take
3406 * care to map the inode according to @mnt_userns before checking permissions.
3407 * On non-idmapped mounts or if permission checking is to be performed on the
3408 * raw inode simply passs init_user_ns.
3410 struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns,
3411 struct dentry *dentry, umode_t mode, int open_flag)
3413 struct dentry *child = NULL;
3414 struct inode *dir = dentry->d_inode;
3415 struct inode *inode;
3418 /* we want directory to be writable */
3419 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
3422 error = -EOPNOTSUPP;
3423 if (!dir->i_op->tmpfile)
3426 child = d_alloc(dentry, &slash_name);
3427 if (unlikely(!child))
3429 error = dir->i_op->tmpfile(mnt_userns, dir, child, mode);
3433 inode = child->d_inode;
3434 if (unlikely(!inode))
3436 if (!(open_flag & O_EXCL)) {
3437 spin_lock(&inode->i_lock);
3438 inode->i_state |= I_LINKABLE;
3439 spin_unlock(&inode->i_lock);
3441 ima_post_create_tmpfile(mnt_userns, inode);
3446 return ERR_PTR(error);
3448 EXPORT_SYMBOL(vfs_tmpfile);
3450 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3451 const struct open_flags *op,
3454 struct user_namespace *mnt_userns;
3455 struct dentry *child;
3457 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3458 if (unlikely(error))
3460 error = mnt_want_write(path.mnt);
3461 if (unlikely(error))
3463 mnt_userns = mnt_user_ns(path.mnt);
3464 child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag);
3465 error = PTR_ERR(child);
3469 path.dentry = child;
3470 audit_inode(nd->name, child, 0);
3471 /* Don't check for other permissions, the inode was just created */
3472 error = may_open(mnt_userns, &path, 0, op->open_flag);
3474 error = vfs_open(&path, file);
3476 mnt_drop_write(path.mnt);
3482 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3485 int error = path_lookupat(nd, flags, &path);
3487 audit_inode(nd->name, path.dentry, 0);
3488 error = vfs_open(&path, file);
3494 static struct file *path_openat(struct nameidata *nd,
3495 const struct open_flags *op, unsigned flags)
3500 file = alloc_empty_file(op->open_flag, current_cred());
3504 if (unlikely(file->f_flags & __O_TMPFILE)) {
3505 error = do_tmpfile(nd, flags, op, file);
3506 } else if (unlikely(file->f_flags & O_PATH)) {
3507 error = do_o_path(nd, flags, file);
3509 const char *s = path_init(nd, flags);
3510 while (!(error = link_path_walk(s, nd)) &&
3511 (s = open_last_lookups(nd, file, op)) != NULL)
3514 error = do_open(nd, file, op);
3517 if (likely(!error)) {
3518 if (likely(file->f_mode & FMODE_OPENED))
3524 if (error == -EOPENSTALE) {
3525 if (flags & LOOKUP_RCU)
3530 return ERR_PTR(error);
3533 struct file *do_filp_open(int dfd, struct filename *pathname,
3534 const struct open_flags *op)
3536 struct nameidata nd;
3537 int flags = op->lookup_flags;
3540 set_nameidata(&nd, dfd, pathname, NULL);
3541 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3542 if (unlikely(filp == ERR_PTR(-ECHILD)))
3543 filp = path_openat(&nd, op, flags);
3544 if (unlikely(filp == ERR_PTR(-ESTALE)))
3545 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3546 restore_nameidata();
3550 struct file *do_file_open_root(const struct path *root,
3551 const char *name, const struct open_flags *op)
3553 struct nameidata nd;
3555 struct filename *filename;
3556 int flags = op->lookup_flags;
3558 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3559 return ERR_PTR(-ELOOP);
3561 filename = getname_kernel(name);
3562 if (IS_ERR(filename))
3563 return ERR_CAST(filename);
3565 set_nameidata(&nd, -1, filename, root);
3566 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3567 if (unlikely(file == ERR_PTR(-ECHILD)))
3568 file = path_openat(&nd, op, flags);
3569 if (unlikely(file == ERR_PTR(-ESTALE)))
3570 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3571 restore_nameidata();
3576 static struct dentry *filename_create(int dfd, struct filename *name,
3577 struct path *path, unsigned int lookup_flags)
3579 struct dentry *dentry = ERR_PTR(-EEXIST);
3584 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3587 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3588 * other flags passed in are ignored!
3590 lookup_flags &= LOOKUP_REVAL;
3592 error = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3594 return ERR_PTR(error);
3597 * Yucky last component or no last component at all?
3598 * (foo/., foo/.., /////)
3600 if (unlikely(type != LAST_NORM))
3603 /* don't fail immediately if it's r/o, at least try to report other errors */
3604 err2 = mnt_want_write(path->mnt);
3606 * Do the final lookup.
3608 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3609 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3610 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3615 if (d_is_positive(dentry))
3619 * Special case - lookup gave negative, but... we had foo/bar/
3620 * From the vfs_mknod() POV we just have a negative dentry -
3621 * all is fine. Let's be bastards - you had / on the end, you've
3622 * been asking for (non-existent) directory. -ENOENT for you.
3624 if (unlikely(!is_dir && last.name[last.len])) {
3628 if (unlikely(err2)) {
3635 dentry = ERR_PTR(error);
3637 inode_unlock(path->dentry->d_inode);
3639 mnt_drop_write(path->mnt);
3645 struct dentry *kern_path_create(int dfd, const char *pathname,
3646 struct path *path, unsigned int lookup_flags)
3648 return filename_create(dfd, getname_kernel(pathname),
3649 path, lookup_flags);
3651 EXPORT_SYMBOL(kern_path_create);
3653 void done_path_create(struct path *path, struct dentry *dentry)
3656 inode_unlock(path->dentry->d_inode);
3657 mnt_drop_write(path->mnt);
3660 EXPORT_SYMBOL(done_path_create);
3662 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3663 struct path *path, unsigned int lookup_flags)
3665 return filename_create(dfd, getname(pathname), path, lookup_flags);
3667 EXPORT_SYMBOL(user_path_create);
3670 * vfs_mknod - create device node or file
3671 * @mnt_userns: user namespace of the mount the inode was found from
3672 * @dir: inode of @dentry
3673 * @dentry: pointer to dentry of the base directory
3674 * @mode: mode of the new device node or file
3675 * @dev: device number of device to create
3677 * Create a device node or file.
3679 * If the inode has been found through an idmapped mount the user namespace of
3680 * the vfsmount must be passed through @mnt_userns. This function will then take
3681 * care to map the inode according to @mnt_userns before checking permissions.
3682 * On non-idmapped mounts or if permission checking is to be performed on the
3683 * raw inode simply passs init_user_ns.
3685 int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
3686 struct dentry *dentry, umode_t mode, dev_t dev)
3688 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3689 int error = may_create(mnt_userns, dir, dentry);
3694 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3695 !capable(CAP_MKNOD))
3698 if (!dir->i_op->mknod)
3701 error = devcgroup_inode_mknod(mode, dev);
3705 error = security_inode_mknod(dir, dentry, mode, dev);
3709 error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev);
3711 fsnotify_create(dir, dentry);
3714 EXPORT_SYMBOL(vfs_mknod);
3716 static int may_mknod(umode_t mode)
3718 switch (mode & S_IFMT) {
3724 case 0: /* zero mode translates to S_IFREG */
3733 static long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3736 struct user_namespace *mnt_userns;
3737 struct dentry *dentry;
3740 unsigned int lookup_flags = 0;
3742 error = may_mknod(mode);
3746 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3748 return PTR_ERR(dentry);
3750 if (!IS_POSIXACL(path.dentry->d_inode))
3751 mode &= ~current_umask();
3752 error = security_path_mknod(&path, dentry, mode, dev);
3756 mnt_userns = mnt_user_ns(path.mnt);
3757 switch (mode & S_IFMT) {
3758 case 0: case S_IFREG:
3759 error = vfs_create(mnt_userns, path.dentry->d_inode,
3760 dentry, mode, true);
3762 ima_post_path_mknod(mnt_userns, dentry);
3764 case S_IFCHR: case S_IFBLK:
3765 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3766 dentry, mode, new_decode_dev(dev));
3768 case S_IFIFO: case S_IFSOCK:
3769 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3774 done_path_create(&path, dentry);
3775 if (retry_estale(error, lookup_flags)) {
3776 lookup_flags |= LOOKUP_REVAL;
3782 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3785 return do_mknodat(dfd, filename, mode, dev);
3788 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3790 return do_mknodat(AT_FDCWD, filename, mode, dev);
3794 * vfs_mkdir - create directory
3795 * @mnt_userns: user namespace of the mount the inode was found from
3796 * @dir: inode of @dentry
3797 * @dentry: pointer to dentry of the base directory
3798 * @mode: mode of the new directory
3800 * Create a directory.
3802 * If the inode has been found through an idmapped mount the user namespace of
3803 * the vfsmount must be passed through @mnt_userns. This function will then take
3804 * care to map the inode according to @mnt_userns before checking permissions.
3805 * On non-idmapped mounts or if permission checking is to be performed on the
3806 * raw inode simply passs init_user_ns.
3808 int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
3809 struct dentry *dentry, umode_t mode)
3811 int error = may_create(mnt_userns, dir, dentry);
3812 unsigned max_links = dir->i_sb->s_max_links;
3817 if (!dir->i_op->mkdir)
3820 mode &= (S_IRWXUGO|S_ISVTX);
3821 error = security_inode_mkdir(dir, dentry, mode);
3825 if (max_links && dir->i_nlink >= max_links)
3828 error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode);
3830 fsnotify_mkdir(dir, dentry);
3833 EXPORT_SYMBOL(vfs_mkdir);
3835 static long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3837 struct dentry *dentry;
3840 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3843 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3845 return PTR_ERR(dentry);
3847 if (!IS_POSIXACL(path.dentry->d_inode))
3848 mode &= ~current_umask();
3849 error = security_path_mkdir(&path, dentry, mode);
3851 struct user_namespace *mnt_userns;
3852 mnt_userns = mnt_user_ns(path.mnt);
3853 error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry,
3856 done_path_create(&path, dentry);
3857 if (retry_estale(error, lookup_flags)) {
3858 lookup_flags |= LOOKUP_REVAL;
3864 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3866 return do_mkdirat(dfd, pathname, mode);
3869 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3871 return do_mkdirat(AT_FDCWD, pathname, mode);
3875 * vfs_rmdir - remove directory
3876 * @mnt_userns: user namespace of the mount the inode was found from
3877 * @dir: inode of @dentry
3878 * @dentry: pointer to dentry of the base directory
3880 * Remove a directory.
3882 * If the inode has been found through an idmapped mount the user namespace of
3883 * the vfsmount must be passed through @mnt_userns. This function will then take
3884 * care to map the inode according to @mnt_userns before checking permissions.
3885 * On non-idmapped mounts or if permission checking is to be performed on the
3886 * raw inode simply passs init_user_ns.
3888 int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir,
3889 struct dentry *dentry)
3891 int error = may_delete(mnt_userns, dir, dentry, 1);
3896 if (!dir->i_op->rmdir)
3900 inode_lock(dentry->d_inode);
3903 if (is_local_mountpoint(dentry))
3906 error = security_inode_rmdir(dir, dentry);
3910 error = dir->i_op->rmdir(dir, dentry);
3914 shrink_dcache_parent(dentry);
3915 dentry->d_inode->i_flags |= S_DEAD;
3917 detach_mounts(dentry);
3918 fsnotify_rmdir(dir, dentry);
3921 inode_unlock(dentry->d_inode);
3927 EXPORT_SYMBOL(vfs_rmdir);
3929 long do_rmdir(int dfd, struct filename *name)
3931 struct user_namespace *mnt_userns;
3933 struct dentry *dentry;
3937 unsigned int lookup_flags = 0;
3939 error = __filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
3955 error = mnt_want_write(path.mnt);
3959 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3960 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3961 error = PTR_ERR(dentry);
3964 if (!dentry->d_inode) {
3968 error = security_path_rmdir(&path, dentry);
3971 mnt_userns = mnt_user_ns(path.mnt);
3972 error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry);
3976 inode_unlock(path.dentry->d_inode);
3977 mnt_drop_write(path.mnt);
3980 if (retry_estale(error, lookup_flags)) {
3981 lookup_flags |= LOOKUP_REVAL;
3989 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3991 return do_rmdir(AT_FDCWD, getname(pathname));
3995 * vfs_unlink - unlink a filesystem object
3996 * @mnt_userns: user namespace of the mount the inode was found from
3997 * @dir: parent directory
3999 * @delegated_inode: returns victim inode, if the inode is delegated.
4001 * The caller must hold dir->i_mutex.
4003 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4004 * return a reference to the inode in delegated_inode. The caller
4005 * should then break the delegation on that inode and retry. Because
4006 * breaking a delegation may take a long time, the caller should drop
4007 * dir->i_mutex before doing so.
4009 * Alternatively, a caller may pass NULL for delegated_inode. This may
4010 * be appropriate for callers that expect the underlying filesystem not
4011 * to be NFS exported.
4013 * If the inode has been found through an idmapped mount the user namespace of
4014 * the vfsmount must be passed through @mnt_userns. This function will then take
4015 * care to map the inode according to @mnt_userns before checking permissions.
4016 * On non-idmapped mounts or if permission checking is to be performed on the
4017 * raw inode simply passs init_user_ns.
4019 int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir,
4020 struct dentry *dentry, struct inode **delegated_inode)
4022 struct inode *target = dentry->d_inode;
4023 int error = may_delete(mnt_userns, dir, dentry, 0);
4028 if (!dir->i_op->unlink)
4032 if (is_local_mountpoint(dentry))
4035 error = security_inode_unlink(dir, dentry);
4037 error = try_break_deleg(target, delegated_inode);
4040 error = dir->i_op->unlink(dir, dentry);
4043 detach_mounts(dentry);
4044 fsnotify_unlink(dir, dentry);
4049 inode_unlock(target);
4051 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4052 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
4053 fsnotify_link_count(target);
4059 EXPORT_SYMBOL(vfs_unlink);
4062 * Make sure that the actual truncation of the file will occur outside its
4063 * directory's i_mutex. Truncate can take a long time if there is a lot of
4064 * writeout happening, and we don't want to prevent access to the directory
4065 * while waiting on the I/O.
4067 long do_unlinkat(int dfd, struct filename *name)
4070 struct dentry *dentry;
4074 struct inode *inode = NULL;
4075 struct inode *delegated_inode = NULL;
4076 unsigned int lookup_flags = 0;
4078 error = __filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4083 if (type != LAST_NORM)
4086 error = mnt_want_write(path.mnt);
4090 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4091 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4092 error = PTR_ERR(dentry);
4093 if (!IS_ERR(dentry)) {
4094 struct user_namespace *mnt_userns;
4096 /* Why not before? Because we want correct error value */
4097 if (last.name[last.len])
4099 inode = dentry->d_inode;
4100 if (d_is_negative(dentry))
4103 error = security_path_unlink(&path, dentry);
4106 mnt_userns = mnt_user_ns(path.mnt);
4107 error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry,
4112 inode_unlock(path.dentry->d_inode);
4114 iput(inode); /* truncate the inode here */
4116 if (delegated_inode) {
4117 error = break_deleg_wait(&delegated_inode);
4121 mnt_drop_write(path.mnt);
4124 if (retry_estale(error, lookup_flags)) {
4125 lookup_flags |= LOOKUP_REVAL;
4134 if (d_is_negative(dentry))
4136 else if (d_is_dir(dentry))
4143 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4145 if ((flag & ~AT_REMOVEDIR) != 0)
4148 if (flag & AT_REMOVEDIR)
4149 return do_rmdir(dfd, getname(pathname));
4150 return do_unlinkat(dfd, getname(pathname));
4153 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4155 return do_unlinkat(AT_FDCWD, getname(pathname));
4159 * vfs_symlink - create symlink
4160 * @mnt_userns: user namespace of the mount the inode was found from
4161 * @dir: inode of @dentry
4162 * @dentry: pointer to dentry of the base directory
4163 * @oldname: name of the file to link to
4167 * If the inode has been found through an idmapped mount the user namespace of
4168 * the vfsmount must be passed through @mnt_userns. This function will then take
4169 * care to map the inode according to @mnt_userns before checking permissions.
4170 * On non-idmapped mounts or if permission checking is to be performed on the
4171 * raw inode simply passs init_user_ns.
4173 int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
4174 struct dentry *dentry, const char *oldname)
4176 int error = may_create(mnt_userns, dir, dentry);
4181 if (!dir->i_op->symlink)
4184 error = security_inode_symlink(dir, dentry, oldname);
4188 error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname);
4190 fsnotify_create(dir, dentry);
4193 EXPORT_SYMBOL(vfs_symlink);
4195 static long do_symlinkat(const char __user *oldname, int newdfd,
4196 const char __user *newname)
4199 struct filename *from;
4200 struct dentry *dentry;
4202 unsigned int lookup_flags = 0;
4204 from = getname(oldname);
4206 return PTR_ERR(from);
4208 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4209 error = PTR_ERR(dentry);
4213 error = security_path_symlink(&path, dentry, from->name);
4215 struct user_namespace *mnt_userns;
4217 mnt_userns = mnt_user_ns(path.mnt);
4218 error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry,
4221 done_path_create(&path, dentry);
4222 if (retry_estale(error, lookup_flags)) {
4223 lookup_flags |= LOOKUP_REVAL;
4231 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4232 int, newdfd, const char __user *, newname)
4234 return do_symlinkat(oldname, newdfd, newname);
4237 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4239 return do_symlinkat(oldname, AT_FDCWD, newname);
4243 * vfs_link - create a new link
4244 * @old_dentry: object to be linked
4245 * @mnt_userns: the user namespace of the mount
4247 * @new_dentry: where to create the new link
4248 * @delegated_inode: returns inode needing a delegation break
4250 * The caller must hold dir->i_mutex
4252 * If vfs_link discovers a delegation on the to-be-linked file in need
4253 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4254 * inode in delegated_inode. The caller should then break the delegation
4255 * and retry. Because breaking a delegation may take a long time, the
4256 * caller should drop the i_mutex before doing so.
4258 * Alternatively, a caller may pass NULL for delegated_inode. This may
4259 * be appropriate for callers that expect the underlying filesystem not
4260 * to be NFS exported.
4262 * If the inode has been found through an idmapped mount the user namespace of
4263 * the vfsmount must be passed through @mnt_userns. This function will then take
4264 * care to map the inode according to @mnt_userns before checking permissions.
4265 * On non-idmapped mounts or if permission checking is to be performed on the
4266 * raw inode simply passs init_user_ns.
4268 int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns,
4269 struct inode *dir, struct dentry *new_dentry,
4270 struct inode **delegated_inode)
4272 struct inode *inode = old_dentry->d_inode;
4273 unsigned max_links = dir->i_sb->s_max_links;
4279 error = may_create(mnt_userns, dir, new_dentry);
4283 if (dir->i_sb != inode->i_sb)
4287 * A link to an append-only or immutable file cannot be created.
4289 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4292 * Updating the link count will likely cause i_uid and i_gid to
4293 * be writen back improperly if their true value is unknown to
4296 if (HAS_UNMAPPED_ID(mnt_userns, inode))
4298 if (!dir->i_op->link)
4300 if (S_ISDIR(inode->i_mode))
4303 error = security_inode_link(old_dentry, dir, new_dentry);
4308 /* Make sure we don't allow creating hardlink to an unlinked file */
4309 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4311 else if (max_links && inode->i_nlink >= max_links)
4314 error = try_break_deleg(inode, delegated_inode);
4316 error = dir->i_op->link(old_dentry, dir, new_dentry);
4319 if (!error && (inode->i_state & I_LINKABLE)) {
4320 spin_lock(&inode->i_lock);
4321 inode->i_state &= ~I_LINKABLE;
4322 spin_unlock(&inode->i_lock);
4324 inode_unlock(inode);
4326 fsnotify_link(dir, inode, new_dentry);
4329 EXPORT_SYMBOL(vfs_link);
4332 * Hardlinks are often used in delicate situations. We avoid
4333 * security-related surprises by not following symlinks on the
4336 * We don't follow them on the oldname either to be compatible
4337 * with linux 2.0, and to avoid hard-linking to directories
4338 * and other special files. --ADM
4340 static int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4341 const char __user *newname, int flags)
4343 struct user_namespace *mnt_userns;
4344 struct dentry *new_dentry;
4345 struct path old_path, new_path;
4346 struct inode *delegated_inode = NULL;
4350 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4353 * To use null names we require CAP_DAC_READ_SEARCH
4354 * This ensures that not everyone will be able to create
4355 * handlink using the passed filedescriptor.
4357 if (flags & AT_EMPTY_PATH) {
4358 if (!capable(CAP_DAC_READ_SEARCH))
4363 if (flags & AT_SYMLINK_FOLLOW)
4364 how |= LOOKUP_FOLLOW;
4366 error = user_path_at(olddfd, oldname, how, &old_path);
4370 new_dentry = user_path_create(newdfd, newname, &new_path,
4371 (how & LOOKUP_REVAL));
4372 error = PTR_ERR(new_dentry);
4373 if (IS_ERR(new_dentry))
4377 if (old_path.mnt != new_path.mnt)
4379 mnt_userns = mnt_user_ns(new_path.mnt);
4380 error = may_linkat(mnt_userns, &old_path);
4381 if (unlikely(error))
4383 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4386 error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode,
4387 new_dentry, &delegated_inode);
4389 done_path_create(&new_path, new_dentry);
4390 if (delegated_inode) {
4391 error = break_deleg_wait(&delegated_inode);
4393 path_put(&old_path);
4397 if (retry_estale(error, how)) {
4398 path_put(&old_path);
4399 how |= LOOKUP_REVAL;
4403 path_put(&old_path);
4408 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4409 int, newdfd, const char __user *, newname, int, flags)
4411 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4414 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4416 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4420 * vfs_rename - rename a filesystem object
4421 * @rd: pointer to &struct renamedata info
4423 * The caller must hold multiple mutexes--see lock_rename()).
4425 * If vfs_rename discovers a delegation in need of breaking at either
4426 * the source or destination, it will return -EWOULDBLOCK and return a
4427 * reference to the inode in delegated_inode. The caller should then
4428 * break the delegation and retry. Because breaking a delegation may
4429 * take a long time, the caller should drop all locks before doing
4432 * Alternatively, a caller may pass NULL for delegated_inode. This may
4433 * be appropriate for callers that expect the underlying filesystem not
4434 * to be NFS exported.
4436 * The worst of all namespace operations - renaming directory. "Perverted"
4437 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4440 * a) we can get into loop creation.
4441 * b) race potential - two innocent renames can create a loop together.
4442 * That's where 4.4 screws up. Current fix: serialization on
4443 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4445 * c) we have to lock _four_ objects - parents and victim (if it exists),
4446 * and source (if it is not a directory).
4447 * And that - after we got ->i_mutex on parents (until then we don't know
4448 * whether the target exists). Solution: try to be smart with locking
4449 * order for inodes. We rely on the fact that tree topology may change
4450 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4451 * move will be locked. Thus we can rank directories by the tree
4452 * (ancestors first) and rank all non-directories after them.
4453 * That works since everybody except rename does "lock parent, lookup,
4454 * lock child" and rename is under ->s_vfs_rename_mutex.
4455 * HOWEVER, it relies on the assumption that any object with ->lookup()
4456 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4457 * we'd better make sure that there's no link(2) for them.
4458 * d) conversion from fhandle to dentry may come in the wrong moment - when
4459 * we are removing the target. Solution: we will have to grab ->i_mutex
4460 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4461 * ->i_mutex on parents, which works but leads to some truly excessive
4464 int vfs_rename(struct renamedata *rd)
4467 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4468 struct dentry *old_dentry = rd->old_dentry;
4469 struct dentry *new_dentry = rd->new_dentry;
4470 struct inode **delegated_inode = rd->delegated_inode;
4471 unsigned int flags = rd->flags;
4472 bool is_dir = d_is_dir(old_dentry);
4473 struct inode *source = old_dentry->d_inode;
4474 struct inode *target = new_dentry->d_inode;
4475 bool new_is_dir = false;
4476 unsigned max_links = new_dir->i_sb->s_max_links;
4477 struct name_snapshot old_name;
4479 if (source == target)
4482 error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir);
4487 error = may_create(rd->new_mnt_userns, new_dir, new_dentry);
4489 new_is_dir = d_is_dir(new_dentry);
4491 if (!(flags & RENAME_EXCHANGE))
4492 error = may_delete(rd->new_mnt_userns, new_dir,
4493 new_dentry, is_dir);
4495 error = may_delete(rd->new_mnt_userns, new_dir,
4496 new_dentry, new_is_dir);
4501 if (!old_dir->i_op->rename)
4505 * If we are going to change the parent - check write permissions,
4506 * we'll need to flip '..'.
4508 if (new_dir != old_dir) {
4510 error = inode_permission(rd->old_mnt_userns, source,
4515 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4516 error = inode_permission(rd->new_mnt_userns, target,
4523 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4528 take_dentry_name_snapshot(&old_name, old_dentry);
4530 if (!is_dir || (flags & RENAME_EXCHANGE))
4531 lock_two_nondirectories(source, target);
4536 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4539 if (max_links && new_dir != old_dir) {
4541 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4543 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4544 old_dir->i_nlink >= max_links)
4548 error = try_break_deleg(source, delegated_inode);
4552 if (target && !new_is_dir) {
4553 error = try_break_deleg(target, delegated_inode);
4557 error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry,
4558 new_dir, new_dentry, flags);
4562 if (!(flags & RENAME_EXCHANGE) && target) {
4564 shrink_dcache_parent(new_dentry);
4565 target->i_flags |= S_DEAD;
4567 dont_mount(new_dentry);
4568 detach_mounts(new_dentry);
4570 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4571 if (!(flags & RENAME_EXCHANGE))
4572 d_move(old_dentry, new_dentry);
4574 d_exchange(old_dentry, new_dentry);
4577 if (!is_dir || (flags & RENAME_EXCHANGE))
4578 unlock_two_nondirectories(source, target);
4580 inode_unlock(target);
4583 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4584 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4585 if (flags & RENAME_EXCHANGE) {
4586 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4587 new_is_dir, NULL, new_dentry);
4590 release_dentry_name_snapshot(&old_name);
4594 EXPORT_SYMBOL(vfs_rename);
4596 int do_renameat2(int olddfd, struct filename *from, int newdfd,
4597 struct filename *to, unsigned int flags)
4599 struct renamedata rd;
4600 struct dentry *old_dentry, *new_dentry;
4601 struct dentry *trap;
4602 struct path old_path, new_path;
4603 struct qstr old_last, new_last;
4604 int old_type, new_type;
4605 struct inode *delegated_inode = NULL;
4606 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4607 bool should_retry = false;
4608 int error = -EINVAL;
4610 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4613 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4614 (flags & RENAME_EXCHANGE))
4617 if (flags & RENAME_EXCHANGE)
4621 error = __filename_parentat(olddfd, from, lookup_flags, &old_path,
4622 &old_last, &old_type);
4626 error = __filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4632 if (old_path.mnt != new_path.mnt)
4636 if (old_type != LAST_NORM)
4639 if (flags & RENAME_NOREPLACE)
4641 if (new_type != LAST_NORM)
4644 error = mnt_want_write(old_path.mnt);
4649 trap = lock_rename(new_path.dentry, old_path.dentry);
4651 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4652 error = PTR_ERR(old_dentry);
4653 if (IS_ERR(old_dentry))
4655 /* source must exist */
4657 if (d_is_negative(old_dentry))
4659 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4660 error = PTR_ERR(new_dentry);
4661 if (IS_ERR(new_dentry))
4664 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4666 if (flags & RENAME_EXCHANGE) {
4668 if (d_is_negative(new_dentry))
4671 if (!d_is_dir(new_dentry)) {
4673 if (new_last.name[new_last.len])
4677 /* unless the source is a directory trailing slashes give -ENOTDIR */
4678 if (!d_is_dir(old_dentry)) {
4680 if (old_last.name[old_last.len])
4682 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4685 /* source should not be ancestor of target */
4687 if (old_dentry == trap)
4689 /* target should not be an ancestor of source */
4690 if (!(flags & RENAME_EXCHANGE))
4692 if (new_dentry == trap)
4695 error = security_path_rename(&old_path, old_dentry,
4696 &new_path, new_dentry, flags);
4700 rd.old_dir = old_path.dentry->d_inode;
4701 rd.old_dentry = old_dentry;
4702 rd.old_mnt_userns = mnt_user_ns(old_path.mnt);
4703 rd.new_dir = new_path.dentry->d_inode;
4704 rd.new_dentry = new_dentry;
4705 rd.new_mnt_userns = mnt_user_ns(new_path.mnt);
4706 rd.delegated_inode = &delegated_inode;
4708 error = vfs_rename(&rd);
4714 unlock_rename(new_path.dentry, old_path.dentry);
4715 if (delegated_inode) {
4716 error = break_deleg_wait(&delegated_inode);
4720 mnt_drop_write(old_path.mnt);
4722 if (retry_estale(error, lookup_flags))
4723 should_retry = true;
4724 path_put(&new_path);
4726 path_put(&old_path);
4728 should_retry = false;
4729 lookup_flags |= LOOKUP_REVAL;
4738 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4739 int, newdfd, const char __user *, newname, unsigned int, flags)
4741 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4745 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4746 int, newdfd, const char __user *, newname)
4748 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4752 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4754 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4755 getname(newname), 0);
4758 int readlink_copy(char __user *buffer, int buflen, const char *link)
4760 int len = PTR_ERR(link);
4765 if (len > (unsigned) buflen)
4767 if (copy_to_user(buffer, link, len))
4774 * vfs_readlink - copy symlink body into userspace buffer
4775 * @dentry: dentry on which to get symbolic link
4776 * @buffer: user memory pointer
4777 * @buflen: size of buffer
4779 * Does not touch atime. That's up to the caller if necessary
4781 * Does not call security hook.
4783 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4785 struct inode *inode = d_inode(dentry);
4786 DEFINE_DELAYED_CALL(done);
4790 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4791 if (unlikely(inode->i_op->readlink))
4792 return inode->i_op->readlink(dentry, buffer, buflen);
4794 if (!d_is_symlink(dentry))
4797 spin_lock(&inode->i_lock);
4798 inode->i_opflags |= IOP_DEFAULT_READLINK;
4799 spin_unlock(&inode->i_lock);
4802 link = READ_ONCE(inode->i_link);
4804 link = inode->i_op->get_link(dentry, inode, &done);
4806 return PTR_ERR(link);
4808 res = readlink_copy(buffer, buflen, link);
4809 do_delayed_call(&done);
4812 EXPORT_SYMBOL(vfs_readlink);
4815 * vfs_get_link - get symlink body
4816 * @dentry: dentry on which to get symbolic link
4817 * @done: caller needs to free returned data with this
4819 * Calls security hook and i_op->get_link() on the supplied inode.
4821 * It does not touch atime. That's up to the caller if necessary.
4823 * Does not work on "special" symlinks like /proc/$$/fd/N
4825 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4827 const char *res = ERR_PTR(-EINVAL);
4828 struct inode *inode = d_inode(dentry);
4830 if (d_is_symlink(dentry)) {
4831 res = ERR_PTR(security_inode_readlink(dentry));
4833 res = inode->i_op->get_link(dentry, inode, done);
4837 EXPORT_SYMBOL(vfs_get_link);
4839 /* get the link contents into pagecache */
4840 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4841 struct delayed_call *callback)
4845 struct address_space *mapping = inode->i_mapping;
4848 page = find_get_page(mapping, 0);
4850 return ERR_PTR(-ECHILD);
4851 if (!PageUptodate(page)) {
4853 return ERR_PTR(-ECHILD);
4856 page = read_mapping_page(mapping, 0, NULL);
4860 set_delayed_call(callback, page_put_link, page);
4861 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4862 kaddr = page_address(page);
4863 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4867 EXPORT_SYMBOL(page_get_link);
4869 void page_put_link(void *arg)
4873 EXPORT_SYMBOL(page_put_link);
4875 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4877 DEFINE_DELAYED_CALL(done);
4878 int res = readlink_copy(buffer, buflen,
4879 page_get_link(dentry, d_inode(dentry),
4881 do_delayed_call(&done);
4884 EXPORT_SYMBOL(page_readlink);
4887 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4889 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4891 struct address_space *mapping = inode->i_mapping;
4895 unsigned int flags = 0;
4897 flags |= AOP_FLAG_NOFS;
4900 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4901 flags, &page, &fsdata);
4905 memcpy(page_address(page), symname, len-1);
4907 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4914 mark_inode_dirty(inode);
4919 EXPORT_SYMBOL(__page_symlink);
4921 int page_symlink(struct inode *inode, const char *symname, int len)
4923 return __page_symlink(inode, symname, len,
4924 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4926 EXPORT_SYMBOL(page_symlink);
4928 const struct inode_operations page_symlink_inode_operations = {
4929 .get_link = page_get_link,
4931 EXPORT_SYMBOL(page_symlink_inode_operations);