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 BUG_ON(name->refcnt <= 0);
252 if (--name->refcnt > 0)
255 if (name->name != name->iname) {
256 __putname(name->name);
263 * check_acl - perform ACL permission checking
264 * @mnt_userns: user namespace of the mount the inode was found from
265 * @inode: inode to check permissions on
266 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
268 * This function performs the ACL permission checking. Since this function
269 * retrieve POSIX acls it needs to know whether it is called from a blocking or
270 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
272 * If the inode has been found through an idmapped mount the user namespace of
273 * the vfsmount must be passed through @mnt_userns. This function will then take
274 * care to map the inode according to @mnt_userns before checking permissions.
275 * On non-idmapped mounts or if permission checking is to be performed on the
276 * raw inode simply passs init_user_ns.
278 static int check_acl(struct user_namespace *mnt_userns,
279 struct inode *inode, int mask)
281 #ifdef CONFIG_FS_POSIX_ACL
282 struct posix_acl *acl;
284 if (mask & MAY_NOT_BLOCK) {
285 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
288 /* no ->get_acl() calls in RCU mode... */
289 if (is_uncached_acl(acl))
291 return posix_acl_permission(mnt_userns, inode, acl, mask);
294 acl = get_acl(inode, ACL_TYPE_ACCESS);
298 int error = posix_acl_permission(mnt_userns, inode, acl, mask);
299 posix_acl_release(acl);
308 * acl_permission_check - perform basic UNIX permission checking
309 * @mnt_userns: user namespace of the mount the inode was found from
310 * @inode: inode to check permissions on
311 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
313 * This function performs the basic UNIX permission checking. Since this
314 * function may retrieve POSIX acls it needs to know whether it is called from a
315 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
317 * If the inode has been found through an idmapped mount the user namespace of
318 * the vfsmount must be passed through @mnt_userns. This function will then take
319 * care to map the inode according to @mnt_userns before checking permissions.
320 * On non-idmapped mounts or if permission checking is to be performed on the
321 * raw inode simply passs init_user_ns.
323 static int acl_permission_check(struct user_namespace *mnt_userns,
324 struct inode *inode, int mask)
326 unsigned int mode = inode->i_mode;
329 /* Are we the owner? If so, ACL's don't matter */
330 i_uid = i_uid_into_mnt(mnt_userns, inode);
331 if (likely(uid_eq(current_fsuid(), i_uid))) {
334 return (mask & ~mode) ? -EACCES : 0;
337 /* Do we have ACL's? */
338 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
339 int error = check_acl(mnt_userns, inode, mask);
340 if (error != -EAGAIN)
344 /* Only RWX matters for group/other mode bits */
348 * Are the group permissions different from
349 * the other permissions in the bits we care
350 * about? Need to check group ownership if so.
352 if (mask & (mode ^ (mode >> 3))) {
353 kgid_t kgid = i_gid_into_mnt(mnt_userns, inode);
354 if (in_group_p(kgid))
358 /* Bits in 'mode' clear that we require? */
359 return (mask & ~mode) ? -EACCES : 0;
363 * generic_permission - check for access rights on a Posix-like filesystem
364 * @mnt_userns: user namespace of the mount the inode was found from
365 * @inode: inode to check access rights for
366 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
367 * %MAY_NOT_BLOCK ...)
369 * Used to check for read/write/execute permissions on a file.
370 * We use "fsuid" for this, letting us set arbitrary permissions
371 * for filesystem access without changing the "normal" uids which
372 * are used for other things.
374 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
375 * request cannot be satisfied (eg. requires blocking or too much complexity).
376 * It would then be called again in ref-walk mode.
378 * If the inode has been found through an idmapped mount the user namespace of
379 * the vfsmount must be passed through @mnt_userns. This function will then take
380 * care to map the inode according to @mnt_userns before checking permissions.
381 * On non-idmapped mounts or if permission checking is to be performed on the
382 * raw inode simply passs init_user_ns.
384 int generic_permission(struct user_namespace *mnt_userns, struct inode *inode,
390 * Do the basic permission checks.
392 ret = acl_permission_check(mnt_userns, inode, mask);
396 if (S_ISDIR(inode->i_mode)) {
397 /* DACs are overridable for directories */
398 if (!(mask & MAY_WRITE))
399 if (capable_wrt_inode_uidgid(mnt_userns, inode,
400 CAP_DAC_READ_SEARCH))
402 if (capable_wrt_inode_uidgid(mnt_userns, inode,
409 * Searching includes executable on directories, else just read.
411 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
412 if (mask == MAY_READ)
413 if (capable_wrt_inode_uidgid(mnt_userns, inode,
414 CAP_DAC_READ_SEARCH))
417 * Read/write DACs are always overridable.
418 * Executable DACs are overridable when there is
419 * at least one exec bit set.
421 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
422 if (capable_wrt_inode_uidgid(mnt_userns, inode,
428 EXPORT_SYMBOL(generic_permission);
431 * do_inode_permission - UNIX permission checking
432 * @mnt_userns: user namespace of the mount the inode was found from
433 * @inode: inode to check permissions on
434 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
436 * We _really_ want to just do "generic_permission()" without
437 * even looking at the inode->i_op values. So we keep a cache
438 * flag in inode->i_opflags, that says "this has not special
439 * permission function, use the fast case".
441 static inline int do_inode_permission(struct user_namespace *mnt_userns,
442 struct inode *inode, int mask)
444 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
445 if (likely(inode->i_op->permission))
446 return inode->i_op->permission(mnt_userns, inode, mask);
448 /* This gets set once for the inode lifetime */
449 spin_lock(&inode->i_lock);
450 inode->i_opflags |= IOP_FASTPERM;
451 spin_unlock(&inode->i_lock);
453 return generic_permission(mnt_userns, inode, mask);
457 * sb_permission - Check superblock-level permissions
458 * @sb: Superblock of inode to check permission on
459 * @inode: Inode to check permission on
460 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
462 * Separate out file-system wide checks from inode-specific permission checks.
464 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
466 if (unlikely(mask & MAY_WRITE)) {
467 umode_t mode = inode->i_mode;
469 /* Nobody gets write access to a read-only fs. */
470 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
477 * inode_permission - Check for access rights to a given inode
478 * @mnt_userns: User namespace of the mount the inode was found from
479 * @inode: Inode to check permission on
480 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
482 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
483 * this, letting us set arbitrary permissions for filesystem access without
484 * changing the "normal" UIDs which are used for other things.
486 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
488 int inode_permission(struct user_namespace *mnt_userns,
489 struct inode *inode, int mask)
493 retval = sb_permission(inode->i_sb, inode, mask);
497 if (unlikely(mask & MAY_WRITE)) {
499 * Nobody gets write access to an immutable file.
501 if (IS_IMMUTABLE(inode))
505 * Updating mtime will likely cause i_uid and i_gid to be
506 * written back improperly if their true value is unknown
509 if (HAS_UNMAPPED_ID(mnt_userns, inode))
513 retval = do_inode_permission(mnt_userns, inode, mask);
517 retval = devcgroup_inode_permission(inode, mask);
521 return security_inode_permission(inode, mask);
523 EXPORT_SYMBOL(inode_permission);
526 * path_get - get a reference to a path
527 * @path: path to get the reference to
529 * Given a path increment the reference count to the dentry and the vfsmount.
531 void path_get(const struct path *path)
536 EXPORT_SYMBOL(path_get);
539 * path_put - put a reference to a path
540 * @path: path to put the reference to
542 * Given a path decrement the reference count to the dentry and the vfsmount.
544 void path_put(const struct path *path)
549 EXPORT_SYMBOL(path_put);
551 #define EMBEDDED_LEVELS 2
556 struct inode *inode; /* path.dentry.d_inode */
558 unsigned seq, m_seq, r_seq;
561 int total_link_count;
564 struct delayed_call done;
567 } *stack, internal[EMBEDDED_LEVELS];
568 struct filename *name;
569 struct nameidata *saved;
574 } __randomize_layout;
576 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
578 struct nameidata *old = current->nameidata;
579 p->stack = p->internal;
582 p->total_link_count = old ? old->total_link_count : 0;
584 current->nameidata = p;
587 static void restore_nameidata(void)
589 struct nameidata *now = current->nameidata, *old = now->saved;
591 current->nameidata = old;
593 old->total_link_count = now->total_link_count;
594 if (now->stack != now->internal)
598 static bool nd_alloc_stack(struct nameidata *nd)
602 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
603 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
606 memcpy(p, nd->internal, sizeof(nd->internal));
612 * path_connected - Verify that a dentry is below mnt.mnt_root
614 * Rename can sometimes move a file or directory outside of a bind
615 * mount, path_connected allows those cases to be detected.
617 static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
619 struct super_block *sb = mnt->mnt_sb;
621 /* Bind mounts can have disconnected paths */
622 if (mnt->mnt_root == sb->s_root)
625 return is_subdir(dentry, mnt->mnt_root);
628 static void drop_links(struct nameidata *nd)
632 struct saved *last = nd->stack + i;
633 do_delayed_call(&last->done);
634 clear_delayed_call(&last->done);
638 static void terminate_walk(struct nameidata *nd)
641 if (!(nd->flags & LOOKUP_RCU)) {
644 for (i = 0; i < nd->depth; i++)
645 path_put(&nd->stack[i].link);
646 if (nd->flags & LOOKUP_ROOT_GRABBED) {
648 nd->flags &= ~LOOKUP_ROOT_GRABBED;
651 nd->flags &= ~LOOKUP_RCU;
657 /* path_put is needed afterwards regardless of success or failure */
658 static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
660 int res = __legitimize_mnt(path->mnt, mseq);
667 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
671 return !read_seqcount_retry(&path->dentry->d_seq, seq);
674 static inline bool legitimize_path(struct nameidata *nd,
675 struct path *path, unsigned seq)
677 return __legitimize_path(path, seq, nd->m_seq);
680 static bool legitimize_links(struct nameidata *nd)
683 for (i = 0; i < nd->depth; i++) {
684 struct saved *last = nd->stack + i;
685 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
694 static bool legitimize_root(struct nameidata *nd)
697 * For scoped-lookups (where nd->root has been zeroed), we need to
698 * restart the whole lookup from scratch -- because set_root() is wrong
699 * for these lookups (nd->dfd is the root, not the filesystem root).
701 if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED))
703 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
704 if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT))
706 nd->flags |= LOOKUP_ROOT_GRABBED;
707 return legitimize_path(nd, &nd->root, nd->root_seq);
711 * Path walking has 2 modes, rcu-walk and ref-walk (see
712 * Documentation/filesystems/path-lookup.txt). In situations when we can't
713 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
714 * normal reference counts on dentries and vfsmounts to transition to ref-walk
715 * mode. Refcounts are grabbed at the last known good point before rcu-walk
716 * got stuck, so ref-walk may continue from there. If this is not successful
717 * (eg. a seqcount has changed), then failure is returned and it's up to caller
718 * to restart the path walk from the beginning in ref-walk mode.
722 * unlazy_walk - try to switch to ref-walk mode.
723 * @nd: nameidata pathwalk data
724 * Returns: 0 on success, -ECHILD on failure
726 * unlazy_walk attempts to legitimize the current nd->path and nd->root
728 * Must be called from rcu-walk context.
729 * Nothing should touch nameidata between unlazy_walk() failure and
732 static int unlazy_walk(struct nameidata *nd)
734 struct dentry *parent = nd->path.dentry;
736 BUG_ON(!(nd->flags & LOOKUP_RCU));
738 nd->flags &= ~LOOKUP_RCU;
739 if (unlikely(!legitimize_links(nd)))
741 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
743 if (unlikely(!legitimize_root(nd)))
746 BUG_ON(nd->inode != parent->d_inode);
751 nd->path.dentry = NULL;
758 * unlazy_child - try to switch to ref-walk mode.
759 * @nd: nameidata pathwalk data
760 * @dentry: child of nd->path.dentry
761 * @seq: seq number to check dentry against
762 * Returns: 0 on success, -ECHILD on failure
764 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
765 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
766 * @nd. Must be called from rcu-walk context.
767 * Nothing should touch nameidata between unlazy_child() failure and
770 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
772 BUG_ON(!(nd->flags & LOOKUP_RCU));
774 nd->flags &= ~LOOKUP_RCU;
775 if (unlikely(!legitimize_links(nd)))
777 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
779 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
783 * We need to move both the parent and the dentry from the RCU domain
784 * to be properly refcounted. And the sequence number in the dentry
785 * validates *both* dentry counters, since we checked the sequence
786 * number of the parent after we got the child sequence number. So we
787 * know the parent must still be valid if the child sequence number is
789 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
791 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
794 * Sequence counts matched. Now make sure that the root is
795 * still valid and get it if required.
797 if (unlikely(!legitimize_root(nd)))
805 nd->path.dentry = NULL;
815 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
817 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
818 return dentry->d_op->d_revalidate(dentry, flags);
824 * complete_walk - successful completion of path walk
825 * @nd: pointer nameidata
827 * If we had been in RCU mode, drop out of it and legitimize nd->path.
828 * Revalidate the final result, unless we'd already done that during
829 * the path walk or the filesystem doesn't ask for it. Return 0 on
830 * success, -error on failure. In case of failure caller does not
831 * need to drop nd->path.
833 static int complete_walk(struct nameidata *nd)
835 struct dentry *dentry = nd->path.dentry;
838 if (nd->flags & LOOKUP_RCU) {
840 * We don't want to zero nd->root for scoped-lookups or
841 * externally-managed nd->root.
843 if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED)))
845 if (unlikely(unlazy_walk(nd)))
849 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
851 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
852 * ever step outside the root during lookup" and should already
853 * be guaranteed by the rest of namei, we want to avoid a namei
854 * BUG resulting in userspace being given a path that was not
855 * scoped within the root at some point during the lookup.
857 * So, do a final sanity-check to make sure that in the
858 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
859 * we won't silently return an fd completely outside of the
860 * requested root to userspace.
862 * Userspace could move the path outside the root after this
863 * check, but as discussed elsewhere this is not a concern (the
864 * resolved file was inside the root at some point).
866 if (!path_is_under(&nd->path, &nd->root))
870 if (likely(!(nd->flags & LOOKUP_JUMPED)))
873 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
876 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
886 static int set_root(struct nameidata *nd)
888 struct fs_struct *fs = current->fs;
891 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
892 * still have to ensure it doesn't happen because it will cause a breakout
895 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
896 return -ENOTRECOVERABLE;
898 if (nd->flags & LOOKUP_RCU) {
902 seq = read_seqcount_begin(&fs->seq);
904 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
905 } while (read_seqcount_retry(&fs->seq, seq));
907 get_fs_root(fs, &nd->root);
908 nd->flags |= LOOKUP_ROOT_GRABBED;
913 static int nd_jump_root(struct nameidata *nd)
915 if (unlikely(nd->flags & LOOKUP_BENEATH))
917 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
918 /* Absolute path arguments to path_init() are allowed. */
919 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
923 int error = set_root(nd);
927 if (nd->flags & LOOKUP_RCU) {
931 nd->inode = d->d_inode;
932 nd->seq = nd->root_seq;
933 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
939 nd->inode = nd->path.dentry->d_inode;
941 nd->flags |= LOOKUP_JUMPED;
946 * Helper to directly jump to a known parsed path from ->get_link,
947 * caller must have taken a reference to path beforehand.
949 int nd_jump_link(struct path *path)
952 struct nameidata *nd = current->nameidata;
954 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
958 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
959 if (nd->path.mnt != path->mnt)
962 /* Not currently safe for scoped-lookups. */
963 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
968 nd->inode = nd->path.dentry->d_inode;
969 nd->flags |= LOOKUP_JUMPED;
977 static inline void put_link(struct nameidata *nd)
979 struct saved *last = nd->stack + --nd->depth;
980 do_delayed_call(&last->done);
981 if (!(nd->flags & LOOKUP_RCU))
982 path_put(&last->link);
985 int sysctl_protected_symlinks __read_mostly = 0;
986 int sysctl_protected_hardlinks __read_mostly = 0;
987 int sysctl_protected_fifos __read_mostly;
988 int sysctl_protected_regular __read_mostly;
991 * may_follow_link - Check symlink following for unsafe situations
992 * @nd: nameidata pathwalk data
994 * In the case of the sysctl_protected_symlinks sysctl being enabled,
995 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
996 * in a sticky world-writable directory. This is to protect privileged
997 * processes from failing races against path names that may change out
998 * from under them by way of other users creating malicious symlinks.
999 * It will permit symlinks to be followed only when outside a sticky
1000 * world-writable directory, or when the uid of the symlink and follower
1001 * match, or when the directory owner matches the symlink's owner.
1003 * Returns 0 if following the symlink is allowed, -ve on error.
1005 static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1007 struct user_namespace *mnt_userns;
1010 if (!sysctl_protected_symlinks)
1013 mnt_userns = mnt_user_ns(nd->path.mnt);
1014 i_uid = i_uid_into_mnt(mnt_userns, inode);
1015 /* Allowed if owner and follower match. */
1016 if (uid_eq(current_cred()->fsuid, i_uid))
1019 /* Allowed if parent directory not sticky and world-writable. */
1020 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1023 /* Allowed if parent directory and link owner match. */
1024 if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid))
1027 if (nd->flags & LOOKUP_RCU)
1030 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1031 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1036 * safe_hardlink_source - Check for safe hardlink conditions
1037 * @mnt_userns: user namespace of the mount the inode was found from
1038 * @inode: the source inode to hardlink from
1040 * Return false if at least one of the following conditions:
1041 * - inode is not a regular file
1043 * - inode is setgid and group-exec
1044 * - access failure for read and write
1046 * Otherwise returns true.
1048 static bool safe_hardlink_source(struct user_namespace *mnt_userns,
1049 struct inode *inode)
1051 umode_t mode = inode->i_mode;
1053 /* Special files should not get pinned to the filesystem. */
1057 /* Setuid files should not get pinned to the filesystem. */
1061 /* Executable setgid files should not get pinned to the filesystem. */
1062 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1065 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1066 if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE))
1073 * may_linkat - Check permissions for creating a hardlink
1074 * @mnt_userns: user namespace of the mount the inode was found from
1075 * @link: the source to hardlink from
1077 * Block hardlink when all of:
1078 * - sysctl_protected_hardlinks enabled
1079 * - fsuid does not match inode
1080 * - hardlink source is unsafe (see safe_hardlink_source() above)
1081 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1083 * If the inode has been found through an idmapped mount the user namespace of
1084 * the vfsmount must be passed through @mnt_userns. This function will then take
1085 * care to map the inode according to @mnt_userns before checking permissions.
1086 * On non-idmapped mounts or if permission checking is to be performed on the
1087 * raw inode simply passs init_user_ns.
1089 * Returns 0 if successful, -ve on error.
1091 int may_linkat(struct user_namespace *mnt_userns, struct path *link)
1093 struct inode *inode = link->dentry->d_inode;
1095 /* Inode writeback is not safe when the uid or gid are invalid. */
1096 if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
1097 !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
1100 if (!sysctl_protected_hardlinks)
1103 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1104 * otherwise, it must be a safe source.
1106 if (safe_hardlink_source(mnt_userns, inode) ||
1107 inode_owner_or_capable(mnt_userns, inode))
1110 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1115 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1116 * should be allowed, or not, on files that already
1118 * @mnt_userns: user namespace of the mount the inode was found from
1119 * @dir_mode: mode bits of directory
1120 * @dir_uid: owner of directory
1121 * @inode: the inode of the file to open
1123 * Block an O_CREAT open of a FIFO (or a regular file) when:
1124 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1125 * - the file already exists
1126 * - we are in a sticky directory
1127 * - we don't own the file
1128 * - the owner of the directory doesn't own the file
1129 * - the directory is world writable
1130 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1131 * the directory doesn't have to be world writable: being group writable will
1134 * If the inode has been found through an idmapped mount the user namespace of
1135 * the vfsmount must be passed through @mnt_userns. This function will then take
1136 * care to map the inode according to @mnt_userns before checking permissions.
1137 * On non-idmapped mounts or if permission checking is to be performed on the
1138 * raw inode simply passs init_user_ns.
1140 * Returns 0 if the open is allowed, -ve on error.
1142 static int may_create_in_sticky(struct user_namespace *mnt_userns,
1143 struct nameidata *nd, struct inode *const inode)
1145 umode_t dir_mode = nd->dir_mode;
1146 kuid_t dir_uid = nd->dir_uid;
1148 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1149 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1150 likely(!(dir_mode & S_ISVTX)) ||
1151 uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) ||
1152 uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode)))
1155 if (likely(dir_mode & 0002) ||
1157 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1158 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1159 const char *operation = S_ISFIFO(inode->i_mode) ?
1160 "sticky_create_fifo" :
1161 "sticky_create_regular";
1162 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1169 * follow_up - Find the mountpoint of path's vfsmount
1171 * Given a path, find the mountpoint of its source file system.
1172 * Replace @path with the path of the mountpoint in the parent mount.
1175 * Return 1 if we went up a level and 0 if we were already at the
1178 int follow_up(struct path *path)
1180 struct mount *mnt = real_mount(path->mnt);
1181 struct mount *parent;
1182 struct dentry *mountpoint;
1184 read_seqlock_excl(&mount_lock);
1185 parent = mnt->mnt_parent;
1186 if (parent == mnt) {
1187 read_sequnlock_excl(&mount_lock);
1190 mntget(&parent->mnt);
1191 mountpoint = dget(mnt->mnt_mountpoint);
1192 read_sequnlock_excl(&mount_lock);
1194 path->dentry = mountpoint;
1196 path->mnt = &parent->mnt;
1199 EXPORT_SYMBOL(follow_up);
1201 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1202 struct path *path, unsigned *seqp)
1204 while (mnt_has_parent(m)) {
1205 struct dentry *mountpoint = m->mnt_mountpoint;
1208 if (unlikely(root->dentry == mountpoint &&
1209 root->mnt == &m->mnt))
1211 if (mountpoint != m->mnt.mnt_root) {
1212 path->mnt = &m->mnt;
1213 path->dentry = mountpoint;
1214 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1221 static bool choose_mountpoint(struct mount *m, const struct path *root,
1228 unsigned seq, mseq = read_seqbegin(&mount_lock);
1230 found = choose_mountpoint_rcu(m, root, path, &seq);
1231 if (unlikely(!found)) {
1232 if (!read_seqretry(&mount_lock, mseq))
1235 if (likely(__legitimize_path(path, seq, mseq)))
1247 * Perform an automount
1248 * - return -EISDIR to tell follow_managed() to stop and return the path we
1251 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1253 struct dentry *dentry = path->dentry;
1255 /* We don't want to mount if someone's just doing a stat -
1256 * unless they're stat'ing a directory and appended a '/' to
1259 * We do, however, want to mount if someone wants to open or
1260 * create a file of any type under the mountpoint, wants to
1261 * traverse through the mountpoint or wants to open the
1262 * mounted directory. Also, autofs may mark negative dentries
1263 * as being automount points. These will need the attentions
1264 * of the daemon to instantiate them before they can be used.
1266 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1267 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1271 if (count && (*count)++ >= MAXSYMLINKS)
1274 return finish_automount(dentry->d_op->d_automount(path), path);
1278 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1279 * dentries are pinned but not locked here, so negative dentry can go
1280 * positive right under us. Use of smp_load_acquire() provides a barrier
1281 * sufficient for ->d_inode and ->d_flags consistency.
1283 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1284 int *count, unsigned lookup_flags)
1286 struct vfsmount *mnt = path->mnt;
1287 bool need_mntput = false;
1290 while (flags & DCACHE_MANAGED_DENTRY) {
1291 /* Allow the filesystem to manage the transit without i_mutex
1293 if (flags & DCACHE_MANAGE_TRANSIT) {
1294 ret = path->dentry->d_op->d_manage(path, false);
1295 flags = smp_load_acquire(&path->dentry->d_flags);
1300 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1301 struct vfsmount *mounted = lookup_mnt(path);
1302 if (mounted) { // ... in our namespace
1306 path->mnt = mounted;
1307 path->dentry = dget(mounted->mnt_root);
1308 // here we know it's positive
1309 flags = path->dentry->d_flags;
1315 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1318 // uncovered automount point
1319 ret = follow_automount(path, count, lookup_flags);
1320 flags = smp_load_acquire(&path->dentry->d_flags);
1327 // possible if you race with several mount --move
1328 if (need_mntput && path->mnt == mnt)
1330 if (!ret && unlikely(d_flags_negative(flags)))
1332 *jumped = need_mntput;
1336 static inline int traverse_mounts(struct path *path, bool *jumped,
1337 int *count, unsigned lookup_flags)
1339 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1342 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1344 if (unlikely(d_flags_negative(flags)))
1348 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1351 int follow_down_one(struct path *path)
1353 struct vfsmount *mounted;
1355 mounted = lookup_mnt(path);
1359 path->mnt = mounted;
1360 path->dentry = dget(mounted->mnt_root);
1365 EXPORT_SYMBOL(follow_down_one);
1368 * Follow down to the covering mount currently visible to userspace. At each
1369 * point, the filesystem owning that dentry may be queried as to whether the
1370 * caller is permitted to proceed or not.
1372 int follow_down(struct path *path)
1374 struct vfsmount *mnt = path->mnt;
1376 int ret = traverse_mounts(path, &jumped, NULL, 0);
1378 if (path->mnt != mnt)
1382 EXPORT_SYMBOL(follow_down);
1385 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1386 * we meet a managed dentry that would need blocking.
1388 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1389 struct inode **inode, unsigned *seqp)
1391 struct dentry *dentry = path->dentry;
1392 unsigned int flags = dentry->d_flags;
1394 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1397 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1402 * Don't forget we might have a non-mountpoint managed dentry
1403 * that wants to block transit.
1405 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1406 int res = dentry->d_op->d_manage(path, true);
1408 return res == -EISDIR;
1409 flags = dentry->d_flags;
1412 if (flags & DCACHE_MOUNTED) {
1413 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1415 path->mnt = &mounted->mnt;
1416 dentry = path->dentry = mounted->mnt.mnt_root;
1417 nd->flags |= LOOKUP_JUMPED;
1418 *seqp = read_seqcount_begin(&dentry->d_seq);
1419 *inode = dentry->d_inode;
1421 * We don't need to re-check ->d_seq after this
1422 * ->d_inode read - there will be an RCU delay
1423 * between mount hash removal and ->mnt_root
1424 * becoming unpinned.
1426 flags = dentry->d_flags;
1429 if (read_seqretry(&mount_lock, nd->m_seq))
1432 return !(flags & DCACHE_NEED_AUTOMOUNT);
1436 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1437 struct path *path, struct inode **inode,
1443 path->mnt = nd->path.mnt;
1444 path->dentry = dentry;
1445 if (nd->flags & LOOKUP_RCU) {
1446 unsigned int seq = *seqp;
1447 if (unlikely(!*inode))
1449 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1451 if (unlazy_child(nd, dentry, seq))
1453 // *path might've been clobbered by __follow_mount_rcu()
1454 path->mnt = nd->path.mnt;
1455 path->dentry = dentry;
1457 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1459 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1462 nd->flags |= LOOKUP_JUMPED;
1464 if (unlikely(ret)) {
1466 if (path->mnt != nd->path.mnt)
1469 *inode = d_backing_inode(path->dentry);
1470 *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1476 * This looks up the name in dcache and possibly revalidates the found dentry.
1477 * NULL is returned if the dentry does not exist in the cache.
1479 static struct dentry *lookup_dcache(const struct qstr *name,
1483 struct dentry *dentry = d_lookup(dir, name);
1485 int error = d_revalidate(dentry, flags);
1486 if (unlikely(error <= 0)) {
1488 d_invalidate(dentry);
1490 return ERR_PTR(error);
1497 * Parent directory has inode locked exclusive. This is one
1498 * and only case when ->lookup() gets called on non in-lookup
1499 * dentries - as the matter of fact, this only gets called
1500 * when directory is guaranteed to have no in-lookup children
1503 static struct dentry *__lookup_hash(const struct qstr *name,
1504 struct dentry *base, unsigned int flags)
1506 struct dentry *dentry = lookup_dcache(name, base, flags);
1508 struct inode *dir = base->d_inode;
1513 /* Don't create child dentry for a dead directory. */
1514 if (unlikely(IS_DEADDIR(dir)))
1515 return ERR_PTR(-ENOENT);
1517 dentry = d_alloc(base, name);
1518 if (unlikely(!dentry))
1519 return ERR_PTR(-ENOMEM);
1521 old = dir->i_op->lookup(dir, dentry, flags);
1522 if (unlikely(old)) {
1529 static struct dentry *lookup_fast(struct nameidata *nd,
1530 struct inode **inode,
1533 struct dentry *dentry, *parent = nd->path.dentry;
1537 * Rename seqlock is not required here because in the off chance
1538 * of a false negative due to a concurrent rename, the caller is
1539 * going to fall back to non-racy lookup.
1541 if (nd->flags & LOOKUP_RCU) {
1543 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1544 if (unlikely(!dentry)) {
1545 if (unlazy_walk(nd))
1546 return ERR_PTR(-ECHILD);
1551 * This sequence count validates that the inode matches
1552 * the dentry name information from lookup.
1554 *inode = d_backing_inode(dentry);
1555 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1556 return ERR_PTR(-ECHILD);
1559 * This sequence count validates that the parent had no
1560 * changes while we did the lookup of the dentry above.
1562 * The memory barrier in read_seqcount_begin of child is
1563 * enough, we can use __read_seqcount_retry here.
1565 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1566 return ERR_PTR(-ECHILD);
1569 status = d_revalidate(dentry, nd->flags);
1570 if (likely(status > 0))
1572 if (unlazy_child(nd, dentry, seq))
1573 return ERR_PTR(-ECHILD);
1574 if (unlikely(status == -ECHILD))
1575 /* we'd been told to redo it in non-rcu mode */
1576 status = d_revalidate(dentry, nd->flags);
1578 dentry = __d_lookup(parent, &nd->last);
1579 if (unlikely(!dentry))
1581 status = d_revalidate(dentry, nd->flags);
1583 if (unlikely(status <= 0)) {
1585 d_invalidate(dentry);
1587 return ERR_PTR(status);
1592 /* Fast lookup failed, do it the slow way */
1593 static struct dentry *__lookup_slow(const struct qstr *name,
1597 struct dentry *dentry, *old;
1598 struct inode *inode = dir->d_inode;
1599 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1601 /* Don't go there if it's already dead */
1602 if (unlikely(IS_DEADDIR(inode)))
1603 return ERR_PTR(-ENOENT);
1605 dentry = d_alloc_parallel(dir, name, &wq);
1608 if (unlikely(!d_in_lookup(dentry))) {
1609 int error = d_revalidate(dentry, flags);
1610 if (unlikely(error <= 0)) {
1612 d_invalidate(dentry);
1617 dentry = ERR_PTR(error);
1620 old = inode->i_op->lookup(inode, dentry, flags);
1621 d_lookup_done(dentry);
1622 if (unlikely(old)) {
1630 static struct dentry *lookup_slow(const struct qstr *name,
1634 struct inode *inode = dir->d_inode;
1636 inode_lock_shared(inode);
1637 res = __lookup_slow(name, dir, flags);
1638 inode_unlock_shared(inode);
1642 static inline int may_lookup(struct user_namespace *mnt_userns,
1643 struct nameidata *nd)
1645 if (nd->flags & LOOKUP_RCU) {
1646 int err = inode_permission(mnt_userns, nd->inode,
1647 MAY_EXEC | MAY_NOT_BLOCK);
1650 if (unlazy_walk(nd))
1653 return inode_permission(mnt_userns, nd->inode, MAY_EXEC);
1656 static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq)
1658 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1661 if (likely(nd->depth != EMBEDDED_LEVELS))
1663 if (likely(nd->stack != nd->internal))
1665 if (likely(nd_alloc_stack(nd)))
1668 if (nd->flags & LOOKUP_RCU) {
1669 // we need to grab link before we do unlazy. And we can't skip
1670 // unlazy even if we fail to grab the link - cleanup needs it
1671 bool grabbed_link = legitimize_path(nd, link, seq);
1673 if (unlazy_walk(nd) != 0 || !grabbed_link)
1676 if (nd_alloc_stack(nd))
1682 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1684 static const char *pick_link(struct nameidata *nd, struct path *link,
1685 struct inode *inode, unsigned seq, int flags)
1689 int error = reserve_stack(nd, link, seq);
1691 if (unlikely(error)) {
1692 if (!(nd->flags & LOOKUP_RCU))
1694 return ERR_PTR(error);
1696 last = nd->stack + nd->depth++;
1698 clear_delayed_call(&last->done);
1701 if (flags & WALK_TRAILING) {
1702 error = may_follow_link(nd, inode);
1703 if (unlikely(error))
1704 return ERR_PTR(error);
1707 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1708 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1709 return ERR_PTR(-ELOOP);
1711 if (!(nd->flags & LOOKUP_RCU)) {
1712 touch_atime(&last->link);
1714 } else if (atime_needs_update(&last->link, inode)) {
1715 if (unlikely(unlazy_walk(nd)))
1716 return ERR_PTR(-ECHILD);
1717 touch_atime(&last->link);
1720 error = security_inode_follow_link(link->dentry, inode,
1721 nd->flags & LOOKUP_RCU);
1722 if (unlikely(error))
1723 return ERR_PTR(error);
1725 res = READ_ONCE(inode->i_link);
1727 const char * (*get)(struct dentry *, struct inode *,
1728 struct delayed_call *);
1729 get = inode->i_op->get_link;
1730 if (nd->flags & LOOKUP_RCU) {
1731 res = get(NULL, inode, &last->done);
1732 if (res == ERR_PTR(-ECHILD)) {
1733 if (unlikely(unlazy_walk(nd)))
1734 return ERR_PTR(-ECHILD);
1735 res = get(link->dentry, inode, &last->done);
1738 res = get(link->dentry, inode, &last->done);
1746 error = nd_jump_root(nd);
1747 if (unlikely(error))
1748 return ERR_PTR(error);
1749 while (unlikely(*++res == '/'))
1754 all_done: // pure jump
1760 * Do we need to follow links? We _really_ want to be able
1761 * to do this check without having to look at inode->i_op,
1762 * so we keep a cache of "no, this doesn't need follow_link"
1763 * for the common case.
1765 static const char *step_into(struct nameidata *nd, int flags,
1766 struct dentry *dentry, struct inode *inode, unsigned seq)
1769 int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1772 return ERR_PTR(err);
1773 if (likely(!d_is_symlink(path.dentry)) ||
1774 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1775 (flags & WALK_NOFOLLOW)) {
1776 /* not a symlink or should not follow */
1777 if (!(nd->flags & LOOKUP_RCU)) {
1778 dput(nd->path.dentry);
1779 if (nd->path.mnt != path.mnt)
1780 mntput(nd->path.mnt);
1787 if (nd->flags & LOOKUP_RCU) {
1788 /* make sure that d_is_symlink above matches inode */
1789 if (read_seqcount_retry(&path.dentry->d_seq, seq))
1790 return ERR_PTR(-ECHILD);
1792 if (path.mnt == nd->path.mnt)
1795 return pick_link(nd, &path, inode, seq, flags);
1798 static struct dentry *follow_dotdot_rcu(struct nameidata *nd,
1799 struct inode **inodep,
1802 struct dentry *parent, *old;
1804 if (path_equal(&nd->path, &nd->root))
1806 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1809 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1810 &nd->root, &path, &seq))
1812 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1813 return ERR_PTR(-ECHILD);
1815 nd->inode = path.dentry->d_inode;
1817 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1818 return ERR_PTR(-ECHILD);
1819 /* we know that mountpoint was pinned */
1821 old = nd->path.dentry;
1822 parent = old->d_parent;
1823 *inodep = parent->d_inode;
1824 *seqp = read_seqcount_begin(&parent->d_seq);
1825 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1826 return ERR_PTR(-ECHILD);
1827 if (unlikely(!path_connected(nd->path.mnt, parent)))
1828 return ERR_PTR(-ECHILD);
1831 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1832 return ERR_PTR(-ECHILD);
1833 if (unlikely(nd->flags & LOOKUP_BENEATH))
1834 return ERR_PTR(-ECHILD);
1838 static struct dentry *follow_dotdot(struct nameidata *nd,
1839 struct inode **inodep,
1842 struct dentry *parent;
1844 if (path_equal(&nd->path, &nd->root))
1846 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1849 if (!choose_mountpoint(real_mount(nd->path.mnt),
1852 path_put(&nd->path);
1854 nd->inode = path.dentry->d_inode;
1855 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1856 return ERR_PTR(-EXDEV);
1858 /* rare case of legitimate dget_parent()... */
1859 parent = dget_parent(nd->path.dentry);
1860 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1862 return ERR_PTR(-ENOENT);
1865 *inodep = parent->d_inode;
1869 if (unlikely(nd->flags & LOOKUP_BENEATH))
1870 return ERR_PTR(-EXDEV);
1871 dget(nd->path.dentry);
1875 static const char *handle_dots(struct nameidata *nd, int type)
1877 if (type == LAST_DOTDOT) {
1878 const char *error = NULL;
1879 struct dentry *parent;
1880 struct inode *inode;
1883 if (!nd->root.mnt) {
1884 error = ERR_PTR(set_root(nd));
1888 if (nd->flags & LOOKUP_RCU)
1889 parent = follow_dotdot_rcu(nd, &inode, &seq);
1891 parent = follow_dotdot(nd, &inode, &seq);
1893 return ERR_CAST(parent);
1894 if (unlikely(!parent))
1895 error = step_into(nd, WALK_NOFOLLOW,
1896 nd->path.dentry, nd->inode, nd->seq);
1898 error = step_into(nd, WALK_NOFOLLOW,
1899 parent, inode, seq);
1900 if (unlikely(error))
1903 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1905 * If there was a racing rename or mount along our
1906 * path, then we can't be sure that ".." hasn't jumped
1907 * above nd->root (and so userspace should retry or use
1911 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1912 return ERR_PTR(-EAGAIN);
1913 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1914 return ERR_PTR(-EAGAIN);
1920 static const char *walk_component(struct nameidata *nd, int flags)
1922 struct dentry *dentry;
1923 struct inode *inode;
1926 * "." and ".." are special - ".." especially so because it has
1927 * to be able to know about the current root directory and
1928 * parent relationships.
1930 if (unlikely(nd->last_type != LAST_NORM)) {
1931 if (!(flags & WALK_MORE) && nd->depth)
1933 return handle_dots(nd, nd->last_type);
1935 dentry = lookup_fast(nd, &inode, &seq);
1937 return ERR_CAST(dentry);
1938 if (unlikely(!dentry)) {
1939 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1941 return ERR_CAST(dentry);
1943 if (!(flags & WALK_MORE) && nd->depth)
1945 return step_into(nd, flags, dentry, inode, seq);
1949 * We can do the critical dentry name comparison and hashing
1950 * operations one word at a time, but we are limited to:
1952 * - Architectures with fast unaligned word accesses. We could
1953 * do a "get_unaligned()" if this helps and is sufficiently
1956 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1957 * do not trap on the (extremely unlikely) case of a page
1958 * crossing operation.
1960 * - Furthermore, we need an efficient 64-bit compile for the
1961 * 64-bit case in order to generate the "number of bytes in
1962 * the final mask". Again, that could be replaced with a
1963 * efficient population count instruction or similar.
1965 #ifdef CONFIG_DCACHE_WORD_ACCESS
1967 #include <asm/word-at-a-time.h>
1971 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1973 #elif defined(CONFIG_64BIT)
1975 * Register pressure in the mixing function is an issue, particularly
1976 * on 32-bit x86, but almost any function requires one state value and
1977 * one temporary. Instead, use a function designed for two state values
1978 * and no temporaries.
1980 * This function cannot create a collision in only two iterations, so
1981 * we have two iterations to achieve avalanche. In those two iterations,
1982 * we have six layers of mixing, which is enough to spread one bit's
1983 * influence out to 2^6 = 64 state bits.
1985 * Rotate constants are scored by considering either 64 one-bit input
1986 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1987 * probability of that delta causing a change to each of the 128 output
1988 * bits, using a sample of random initial states.
1990 * The Shannon entropy of the computed probabilities is then summed
1991 * to produce a score. Ideally, any input change has a 50% chance of
1992 * toggling any given output bit.
1994 * Mixing scores (in bits) for (12,45):
1995 * Input delta: 1-bit 2-bit
1996 * 1 round: 713.3 42542.6
1997 * 2 rounds: 2753.7 140389.8
1998 * 3 rounds: 5954.1 233458.2
1999 * 4 rounds: 7862.6 256672.2
2000 * Perfect: 8192 258048
2001 * (64*128) (64*63/2 * 128)
2003 #define HASH_MIX(x, y, a) \
2005 y ^= x, x = rol64(x,12),\
2006 x += y, y = rol64(y,45),\
2010 * Fold two longs into one 32-bit hash value. This must be fast, but
2011 * latency isn't quite as critical, as there is a fair bit of additional
2012 * work done before the hash value is used.
2014 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2016 y ^= x * GOLDEN_RATIO_64;
2017 y *= GOLDEN_RATIO_64;
2021 #else /* 32-bit case */
2024 * Mixing scores (in bits) for (7,20):
2025 * Input delta: 1-bit 2-bit
2026 * 1 round: 330.3 9201.6
2027 * 2 rounds: 1246.4 25475.4
2028 * 3 rounds: 1907.1 31295.1
2029 * 4 rounds: 2042.3 31718.6
2030 * Perfect: 2048 31744
2031 * (32*64) (32*31/2 * 64)
2033 #define HASH_MIX(x, y, a) \
2035 y ^= x, x = rol32(x, 7),\
2036 x += y, y = rol32(y,20),\
2039 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2041 /* Use arch-optimized multiply if one exists */
2042 return __hash_32(y ^ __hash_32(x));
2048 * Return the hash of a string of known length. This is carfully
2049 * designed to match hash_name(), which is the more critical function.
2050 * In particular, we must end by hashing a final word containing 0..7
2051 * payload bytes, to match the way that hash_name() iterates until it
2052 * finds the delimiter after the name.
2054 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2056 unsigned long a, x = 0, y = (unsigned long)salt;
2061 a = load_unaligned_zeropad(name);
2062 if (len < sizeof(unsigned long))
2065 name += sizeof(unsigned long);
2066 len -= sizeof(unsigned long);
2068 x ^= a & bytemask_from_count(len);
2070 return fold_hash(x, y);
2072 EXPORT_SYMBOL(full_name_hash);
2074 /* Return the "hash_len" (hash and length) of a null-terminated string */
2075 u64 hashlen_string(const void *salt, const char *name)
2077 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2078 unsigned long adata, mask, len;
2079 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2086 len += sizeof(unsigned long);
2088 a = load_unaligned_zeropad(name+len);
2089 } while (!has_zero(a, &adata, &constants));
2091 adata = prep_zero_mask(a, adata, &constants);
2092 mask = create_zero_mask(adata);
2093 x ^= a & zero_bytemask(mask);
2095 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2097 EXPORT_SYMBOL(hashlen_string);
2100 * Calculate the length and hash of the path component, and
2101 * return the "hash_len" as the result.
2103 static inline u64 hash_name(const void *salt, const char *name)
2105 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2106 unsigned long adata, bdata, mask, len;
2107 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2114 len += sizeof(unsigned long);
2116 a = load_unaligned_zeropad(name+len);
2117 b = a ^ REPEAT_BYTE('/');
2118 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2120 adata = prep_zero_mask(a, adata, &constants);
2121 bdata = prep_zero_mask(b, bdata, &constants);
2122 mask = create_zero_mask(adata | bdata);
2123 x ^= a & zero_bytemask(mask);
2125 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2128 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2130 /* Return the hash of a string of known length */
2131 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2133 unsigned long hash = init_name_hash(salt);
2135 hash = partial_name_hash((unsigned char)*name++, hash);
2136 return end_name_hash(hash);
2138 EXPORT_SYMBOL(full_name_hash);
2140 /* Return the "hash_len" (hash and length) of a null-terminated string */
2141 u64 hashlen_string(const void *salt, const char *name)
2143 unsigned long hash = init_name_hash(salt);
2144 unsigned long len = 0, c;
2146 c = (unsigned char)*name;
2149 hash = partial_name_hash(c, hash);
2150 c = (unsigned char)name[len];
2152 return hashlen_create(end_name_hash(hash), len);
2154 EXPORT_SYMBOL(hashlen_string);
2157 * We know there's a real path component here of at least
2160 static inline u64 hash_name(const void *salt, const char *name)
2162 unsigned long hash = init_name_hash(salt);
2163 unsigned long len = 0, c;
2165 c = (unsigned char)*name;
2168 hash = partial_name_hash(c, hash);
2169 c = (unsigned char)name[len];
2170 } while (c && c != '/');
2171 return hashlen_create(end_name_hash(hash), len);
2178 * This is the basic name resolution function, turning a pathname into
2179 * the final dentry. We expect 'base' to be positive and a directory.
2181 * Returns 0 and nd will have valid dentry and mnt on success.
2182 * Returns error and drops reference to input namei data on failure.
2184 static int link_path_walk(const char *name, struct nameidata *nd)
2186 int depth = 0; // depth <= nd->depth
2189 nd->last_type = LAST_ROOT;
2190 nd->flags |= LOOKUP_PARENT;
2192 return PTR_ERR(name);
2196 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2200 /* At this point we know we have a real path component. */
2202 struct user_namespace *mnt_userns;
2207 mnt_userns = mnt_user_ns(nd->path.mnt);
2208 err = may_lookup(mnt_userns, nd);
2212 hash_len = hash_name(nd->path.dentry, name);
2215 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2217 if (name[1] == '.') {
2219 nd->flags |= LOOKUP_JUMPED;
2225 if (likely(type == LAST_NORM)) {
2226 struct dentry *parent = nd->path.dentry;
2227 nd->flags &= ~LOOKUP_JUMPED;
2228 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2229 struct qstr this = { { .hash_len = hash_len }, .name = name };
2230 err = parent->d_op->d_hash(parent, &this);
2233 hash_len = this.hash_len;
2238 nd->last.hash_len = hash_len;
2239 nd->last.name = name;
2240 nd->last_type = type;
2242 name += hashlen_len(hash_len);
2246 * If it wasn't NUL, we know it was '/'. Skip that
2247 * slash, and continue until no more slashes.
2251 } while (unlikely(*name == '/'));
2252 if (unlikely(!*name)) {
2254 /* pathname or trailing symlink, done */
2256 nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode);
2257 nd->dir_mode = nd->inode->i_mode;
2258 nd->flags &= ~LOOKUP_PARENT;
2261 /* last component of nested symlink */
2262 name = nd->stack[--depth].name;
2263 link = walk_component(nd, 0);
2265 /* not the last component */
2266 link = walk_component(nd, WALK_MORE);
2268 if (unlikely(link)) {
2270 return PTR_ERR(link);
2271 /* a symlink to follow */
2272 nd->stack[depth++].name = name;
2276 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2277 if (nd->flags & LOOKUP_RCU) {
2278 if (unlazy_walk(nd))
2286 /* must be paired with terminate_walk() */
2287 static const char *path_init(struct nameidata *nd, unsigned flags)
2290 const char *s = nd->name->name;
2293 flags &= ~LOOKUP_RCU;
2294 if (flags & LOOKUP_RCU)
2297 nd->flags = flags | LOOKUP_JUMPED;
2300 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2301 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2304 if (flags & LOOKUP_ROOT) {
2305 struct dentry *root = nd->root.dentry;
2306 struct inode *inode = root->d_inode;
2307 if (*s && unlikely(!d_can_lookup(root)))
2308 return ERR_PTR(-ENOTDIR);
2309 nd->path = nd->root;
2311 if (flags & LOOKUP_RCU) {
2312 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2313 nd->root_seq = nd->seq;
2315 path_get(&nd->path);
2320 nd->root.mnt = NULL;
2321 nd->path.mnt = NULL;
2322 nd->path.dentry = NULL;
2324 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2325 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2326 error = nd_jump_root(nd);
2327 if (unlikely(error))
2328 return ERR_PTR(error);
2332 /* Relative pathname -- get the starting-point it is relative to. */
2333 if (nd->dfd == AT_FDCWD) {
2334 if (flags & LOOKUP_RCU) {
2335 struct fs_struct *fs = current->fs;
2339 seq = read_seqcount_begin(&fs->seq);
2341 nd->inode = nd->path.dentry->d_inode;
2342 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2343 } while (read_seqcount_retry(&fs->seq, seq));
2345 get_fs_pwd(current->fs, &nd->path);
2346 nd->inode = nd->path.dentry->d_inode;
2349 /* Caller must check execute permissions on the starting path component */
2350 struct fd f = fdget_raw(nd->dfd);
2351 struct dentry *dentry;
2354 return ERR_PTR(-EBADF);
2356 dentry = f.file->f_path.dentry;
2358 if (*s && unlikely(!d_can_lookup(dentry))) {
2360 return ERR_PTR(-ENOTDIR);
2363 nd->path = f.file->f_path;
2364 if (flags & LOOKUP_RCU) {
2365 nd->inode = nd->path.dentry->d_inode;
2366 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2368 path_get(&nd->path);
2369 nd->inode = nd->path.dentry->d_inode;
2374 /* For scoped-lookups we need to set the root to the dirfd as well. */
2375 if (flags & LOOKUP_IS_SCOPED) {
2376 nd->root = nd->path;
2377 if (flags & LOOKUP_RCU) {
2378 nd->root_seq = nd->seq;
2380 path_get(&nd->root);
2381 nd->flags |= LOOKUP_ROOT_GRABBED;
2387 static inline const char *lookup_last(struct nameidata *nd)
2389 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2390 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2392 return walk_component(nd, WALK_TRAILING);
2395 static int handle_lookup_down(struct nameidata *nd)
2397 if (!(nd->flags & LOOKUP_RCU))
2398 dget(nd->path.dentry);
2399 return PTR_ERR(step_into(nd, WALK_NOFOLLOW,
2400 nd->path.dentry, nd->inode, nd->seq));
2403 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2404 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2406 const char *s = path_init(nd, flags);
2409 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2410 err = handle_lookup_down(nd);
2411 if (unlikely(err < 0))
2415 while (!(err = link_path_walk(s, nd)) &&
2416 (s = lookup_last(nd)) != NULL)
2419 err = complete_walk(nd);
2421 if (!err && nd->flags & LOOKUP_DIRECTORY)
2422 if (!d_can_lookup(nd->path.dentry))
2424 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2425 err = handle_lookup_down(nd);
2426 nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please...
2430 nd->path.mnt = NULL;
2431 nd->path.dentry = NULL;
2437 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2438 struct path *path, struct path *root)
2441 struct nameidata nd;
2443 return PTR_ERR(name);
2444 if (unlikely(root)) {
2446 flags |= LOOKUP_ROOT;
2448 set_nameidata(&nd, dfd, name);
2449 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2450 if (unlikely(retval == -ECHILD))
2451 retval = path_lookupat(&nd, flags, path);
2452 if (unlikely(retval == -ESTALE))
2453 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2455 if (likely(!retval))
2456 audit_inode(name, path->dentry,
2457 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2458 restore_nameidata();
2463 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2464 static int path_parentat(struct nameidata *nd, unsigned flags,
2465 struct path *parent)
2467 const char *s = path_init(nd, flags);
2468 int err = link_path_walk(s, nd);
2470 err = complete_walk(nd);
2473 nd->path.mnt = NULL;
2474 nd->path.dentry = NULL;
2480 static struct filename *filename_parentat(int dfd, struct filename *name,
2481 unsigned int flags, struct path *parent,
2482 struct qstr *last, int *type)
2485 struct nameidata nd;
2489 set_nameidata(&nd, dfd, name);
2490 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2491 if (unlikely(retval == -ECHILD))
2492 retval = path_parentat(&nd, flags, parent);
2493 if (unlikely(retval == -ESTALE))
2494 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2495 if (likely(!retval)) {
2497 *type = nd.last_type;
2498 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2501 name = ERR_PTR(retval);
2503 restore_nameidata();
2507 /* does lookup, returns the object with parent locked */
2508 struct dentry *kern_path_locked(const char *name, struct path *path)
2510 struct filename *filename;
2515 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2517 if (IS_ERR(filename))
2518 return ERR_CAST(filename);
2519 if (unlikely(type != LAST_NORM)) {
2522 return ERR_PTR(-EINVAL);
2524 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2525 d = __lookup_hash(&last, path->dentry, 0);
2527 inode_unlock(path->dentry->d_inode);
2534 int kern_path(const char *name, unsigned int flags, struct path *path)
2536 return filename_lookup(AT_FDCWD, getname_kernel(name),
2539 EXPORT_SYMBOL(kern_path);
2542 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2543 * @dentry: pointer to dentry of the base directory
2544 * @mnt: pointer to vfs mount of the base directory
2545 * @name: pointer to file name
2546 * @flags: lookup flags
2547 * @path: pointer to struct path to fill
2549 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2550 const char *name, unsigned int flags,
2553 struct path root = {.mnt = mnt, .dentry = dentry};
2554 /* the first argument of filename_lookup() is ignored with root */
2555 return filename_lookup(AT_FDCWD, getname_kernel(name),
2556 flags , path, &root);
2558 EXPORT_SYMBOL(vfs_path_lookup);
2560 static int lookup_one_len_common(const char *name, struct dentry *base,
2561 int len, struct qstr *this)
2565 this->hash = full_name_hash(base, name, len);
2569 if (unlikely(name[0] == '.')) {
2570 if (len < 2 || (len == 2 && name[1] == '.'))
2575 unsigned int c = *(const unsigned char *)name++;
2576 if (c == '/' || c == '\0')
2580 * See if the low-level filesystem might want
2581 * to use its own hash..
2583 if (base->d_flags & DCACHE_OP_HASH) {
2584 int err = base->d_op->d_hash(base, this);
2589 return inode_permission(&init_user_ns, base->d_inode, MAY_EXEC);
2593 * try_lookup_one_len - filesystem helper to lookup single pathname component
2594 * @name: pathname component to lookup
2595 * @base: base directory to lookup from
2596 * @len: maximum length @len should be interpreted to
2598 * Look up a dentry by name in the dcache, returning NULL if it does not
2599 * currently exist. The function does not try to create a dentry.
2601 * Note that this routine is purely a helper for filesystem usage and should
2602 * not be called by generic code.
2604 * The caller must hold base->i_mutex.
2606 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2611 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2613 err = lookup_one_len_common(name, base, len, &this);
2615 return ERR_PTR(err);
2617 return lookup_dcache(&this, base, 0);
2619 EXPORT_SYMBOL(try_lookup_one_len);
2622 * lookup_one_len - filesystem helper to lookup single pathname component
2623 * @name: pathname component to lookup
2624 * @base: base directory to lookup from
2625 * @len: maximum length @len should be interpreted to
2627 * Note that this routine is purely a helper for filesystem usage and should
2628 * not be called by generic code.
2630 * The caller must hold base->i_mutex.
2632 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2634 struct dentry *dentry;
2638 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2640 err = lookup_one_len_common(name, base, len, &this);
2642 return ERR_PTR(err);
2644 dentry = lookup_dcache(&this, base, 0);
2645 return dentry ? dentry : __lookup_slow(&this, base, 0);
2647 EXPORT_SYMBOL(lookup_one_len);
2650 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2651 * @name: pathname component to lookup
2652 * @base: base directory to lookup from
2653 * @len: maximum length @len should be interpreted to
2655 * Note that this routine is purely a helper for filesystem usage and should
2656 * not be called by generic code.
2658 * Unlike lookup_one_len, it should be called without the parent
2659 * i_mutex held, and will take the i_mutex itself if necessary.
2661 struct dentry *lookup_one_len_unlocked(const char *name,
2662 struct dentry *base, int len)
2668 err = lookup_one_len_common(name, base, len, &this);
2670 return ERR_PTR(err);
2672 ret = lookup_dcache(&this, base, 0);
2674 ret = lookup_slow(&this, base, 0);
2677 EXPORT_SYMBOL(lookup_one_len_unlocked);
2680 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2681 * on negatives. Returns known positive or ERR_PTR(); that's what
2682 * most of the users want. Note that pinned negative with unlocked parent
2683 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2684 * need to be very careful; pinned positives have ->d_inode stable, so
2685 * this one avoids such problems.
2687 struct dentry *lookup_positive_unlocked(const char *name,
2688 struct dentry *base, int len)
2690 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2691 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2693 ret = ERR_PTR(-ENOENT);
2697 EXPORT_SYMBOL(lookup_positive_unlocked);
2699 #ifdef CONFIG_UNIX98_PTYS
2700 int path_pts(struct path *path)
2702 /* Find something mounted on "pts" in the same directory as
2705 struct dentry *parent = dget_parent(path->dentry);
2706 struct dentry *child;
2707 struct qstr this = QSTR_INIT("pts", 3);
2709 if (unlikely(!path_connected(path->mnt, parent))) {
2714 path->dentry = parent;
2715 child = d_hash_and_lookup(parent, &this);
2719 path->dentry = child;
2726 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2727 struct path *path, int *empty)
2729 return filename_lookup(dfd, getname_flags(name, flags, empty),
2732 EXPORT_SYMBOL(user_path_at_empty);
2734 int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir,
2735 struct inode *inode)
2737 kuid_t fsuid = current_fsuid();
2739 if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid))
2741 if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid))
2743 return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER);
2745 EXPORT_SYMBOL(__check_sticky);
2748 * Check whether we can remove a link victim from directory dir, check
2749 * whether the type of victim is right.
2750 * 1. We can't do it if dir is read-only (done in permission())
2751 * 2. We should have write and exec permissions on dir
2752 * 3. We can't remove anything from append-only dir
2753 * 4. We can't do anything with immutable dir (done in permission())
2754 * 5. If the sticky bit on dir is set we should either
2755 * a. be owner of dir, or
2756 * b. be owner of victim, or
2757 * c. have CAP_FOWNER capability
2758 * 6. If the victim is append-only or immutable we can't do antyhing with
2759 * links pointing to it.
2760 * 7. If the victim has an unknown uid or gid we can't change the inode.
2761 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2762 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2763 * 10. We can't remove a root or mountpoint.
2764 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2765 * nfs_async_unlink().
2767 static int may_delete(struct user_namespace *mnt_userns, struct inode *dir,
2768 struct dentry *victim, bool isdir)
2770 struct inode *inode = d_backing_inode(victim);
2773 if (d_is_negative(victim))
2777 BUG_ON(victim->d_parent->d_inode != dir);
2779 /* Inode writeback is not safe when the uid or gid are invalid. */
2780 if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
2781 !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
2784 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2786 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2792 if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) ||
2793 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2794 HAS_UNMAPPED_ID(mnt_userns, inode))
2797 if (!d_is_dir(victim))
2799 if (IS_ROOT(victim))
2801 } else if (d_is_dir(victim))
2803 if (IS_DEADDIR(dir))
2805 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2810 /* Check whether we can create an object with dentry child in directory
2812 * 1. We can't do it if child already exists (open has special treatment for
2813 * this case, but since we are inlined it's OK)
2814 * 2. We can't do it if dir is read-only (done in permission())
2815 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2816 * 4. We should have write and exec permissions on dir
2817 * 5. We can't do it if dir is immutable (done in permission())
2819 static inline int may_create(struct user_namespace *mnt_userns,
2820 struct inode *dir, struct dentry *child)
2822 struct user_namespace *s_user_ns;
2823 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2826 if (IS_DEADDIR(dir))
2828 s_user_ns = dir->i_sb->s_user_ns;
2829 if (!kuid_has_mapping(s_user_ns, fsuid_into_mnt(mnt_userns)) ||
2830 !kgid_has_mapping(s_user_ns, fsgid_into_mnt(mnt_userns)))
2832 return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2836 * p1 and p2 should be directories on the same fs.
2838 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2843 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2847 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2849 p = d_ancestor(p2, p1);
2851 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2852 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2856 p = d_ancestor(p1, p2);
2858 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2859 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2863 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2864 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2867 EXPORT_SYMBOL(lock_rename);
2869 void unlock_rename(struct dentry *p1, struct dentry *p2)
2871 inode_unlock(p1->d_inode);
2873 inode_unlock(p2->d_inode);
2874 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2877 EXPORT_SYMBOL(unlock_rename);
2880 * vfs_create - create new file
2881 * @mnt_userns: user namespace of the mount the inode was found from
2882 * @dir: inode of @dentry
2883 * @dentry: pointer to dentry of the base directory
2884 * @mode: mode of the new file
2885 * @want_excl: whether the file must not yet exist
2887 * Create a new file.
2889 * If the inode has been found through an idmapped mount the user namespace of
2890 * the vfsmount must be passed through @mnt_userns. This function will then take
2891 * care to map the inode according to @mnt_userns before checking permissions.
2892 * On non-idmapped mounts or if permission checking is to be performed on the
2893 * raw inode simply passs init_user_ns.
2895 int vfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2896 struct dentry *dentry, umode_t mode, bool want_excl)
2898 int error = may_create(mnt_userns, dir, dentry);
2902 if (!dir->i_op->create)
2903 return -EACCES; /* shouldn't it be ENOSYS? */
2906 error = security_inode_create(dir, dentry, mode);
2909 error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl);
2911 fsnotify_create(dir, dentry);
2914 EXPORT_SYMBOL(vfs_create);
2916 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2917 int (*f)(struct dentry *, umode_t, void *),
2920 struct inode *dir = dentry->d_parent->d_inode;
2921 int error = may_create(&init_user_ns, dir, dentry);
2927 error = security_inode_create(dir, dentry, mode);
2930 error = f(dentry, mode, arg);
2932 fsnotify_create(dir, dentry);
2935 EXPORT_SYMBOL(vfs_mkobj);
2937 bool may_open_dev(const struct path *path)
2939 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2940 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2943 static int may_open(struct user_namespace *mnt_userns, const struct path *path,
2944 int acc_mode, int flag)
2946 struct dentry *dentry = path->dentry;
2947 struct inode *inode = dentry->d_inode;
2953 switch (inode->i_mode & S_IFMT) {
2957 if (acc_mode & MAY_WRITE)
2959 if (acc_mode & MAY_EXEC)
2964 if (!may_open_dev(path))
2969 if (acc_mode & MAY_EXEC)
2974 if ((acc_mode & MAY_EXEC) && path_noexec(path))
2979 error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode);
2984 * An append-only file must be opened in append mode for writing.
2986 if (IS_APPEND(inode)) {
2987 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2993 /* O_NOATIME can only be set by the owner or superuser */
2994 if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode))
3000 static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp)
3002 const struct path *path = &filp->f_path;
3003 struct inode *inode = path->dentry->d_inode;
3004 int error = get_write_access(inode);
3008 * Refuse to truncate files with mandatory locks held on them.
3010 error = locks_verify_locked(filp);
3012 error = security_path_truncate(path);
3014 error = do_truncate(mnt_userns, path->dentry, 0,
3015 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3018 put_write_access(inode);
3022 static inline int open_to_namei_flags(int flag)
3024 if ((flag & O_ACCMODE) == 3)
3029 static int may_o_create(struct user_namespace *mnt_userns,
3030 const struct path *dir, struct dentry *dentry,
3033 struct user_namespace *s_user_ns;
3034 int error = security_path_mknod(dir, dentry, mode, 0);
3038 s_user_ns = dir->dentry->d_sb->s_user_ns;
3039 if (!kuid_has_mapping(s_user_ns, fsuid_into_mnt(mnt_userns)) ||
3040 !kgid_has_mapping(s_user_ns, fsgid_into_mnt(mnt_userns)))
3043 error = inode_permission(mnt_userns, dir->dentry->d_inode,
3044 MAY_WRITE | MAY_EXEC);
3048 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3052 * Attempt to atomically look up, create and open a file from a negative
3055 * Returns 0 if successful. The file will have been created and attached to
3056 * @file by the filesystem calling finish_open().
3058 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3059 * be set. The caller will need to perform the open themselves. @path will
3060 * have been updated to point to the new dentry. This may be negative.
3062 * Returns an error code otherwise.
3064 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3066 int open_flag, umode_t mode)
3068 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3069 struct inode *dir = nd->path.dentry->d_inode;
3072 if (nd->flags & LOOKUP_DIRECTORY)
3073 open_flag |= O_DIRECTORY;
3075 file->f_path.dentry = DENTRY_NOT_SET;
3076 file->f_path.mnt = nd->path.mnt;
3077 error = dir->i_op->atomic_open(dir, dentry, file,
3078 open_to_namei_flags(open_flag), mode);
3079 d_lookup_done(dentry);
3081 if (file->f_mode & FMODE_OPENED) {
3082 if (unlikely(dentry != file->f_path.dentry)) {
3084 dentry = dget(file->f_path.dentry);
3086 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3089 if (file->f_path.dentry) {
3091 dentry = file->f_path.dentry;
3093 if (unlikely(d_is_negative(dentry)))
3099 dentry = ERR_PTR(error);
3105 * Look up and maybe create and open the last component.
3107 * Must be called with parent locked (exclusive in O_CREAT case).
3109 * Returns 0 on success, that is, if
3110 * the file was successfully atomically created (if necessary) and opened, or
3111 * the file was not completely opened at this time, though lookups and
3112 * creations were performed.
3113 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3114 * In the latter case dentry returned in @path might be negative if O_CREAT
3115 * hadn't been specified.
3117 * An error code is returned on failure.
3119 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3120 const struct open_flags *op,
3123 struct user_namespace *mnt_userns;
3124 struct dentry *dir = nd->path.dentry;
3125 struct inode *dir_inode = dir->d_inode;
3126 int open_flag = op->open_flag;
3127 struct dentry *dentry;
3128 int error, create_error = 0;
3129 umode_t mode = op->mode;
3130 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3132 if (unlikely(IS_DEADDIR(dir_inode)))
3133 return ERR_PTR(-ENOENT);
3135 file->f_mode &= ~FMODE_CREATED;
3136 dentry = d_lookup(dir, &nd->last);
3139 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3143 if (d_in_lookup(dentry))
3146 error = d_revalidate(dentry, nd->flags);
3147 if (likely(error > 0))
3151 d_invalidate(dentry);
3155 if (dentry->d_inode) {
3156 /* Cached positive dentry: will open in f_op->open */
3161 * Checking write permission is tricky, bacuse we don't know if we are
3162 * going to actually need it: O_CREAT opens should work as long as the
3163 * file exists. But checking existence breaks atomicity. The trick is
3164 * to check access and if not granted clear O_CREAT from the flags.
3166 * Another problem is returing the "right" error value (e.g. for an
3167 * O_EXCL open we want to return EEXIST not EROFS).
3169 if (unlikely(!got_write))
3170 open_flag &= ~O_TRUNC;
3171 mnt_userns = mnt_user_ns(nd->path.mnt);
3172 if (open_flag & O_CREAT) {
3173 if (open_flag & O_EXCL)
3174 open_flag &= ~O_TRUNC;
3175 if (!IS_POSIXACL(dir->d_inode))
3176 mode &= ~current_umask();
3177 if (likely(got_write))
3178 create_error = may_o_create(mnt_userns, &nd->path,
3181 create_error = -EROFS;
3184 open_flag &= ~O_CREAT;
3185 if (dir_inode->i_op->atomic_open) {
3186 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3187 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3188 dentry = ERR_PTR(create_error);
3192 if (d_in_lookup(dentry)) {
3193 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3195 d_lookup_done(dentry);
3196 if (unlikely(res)) {
3198 error = PTR_ERR(res);
3206 /* Negative dentry, just create the file */
3207 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3208 file->f_mode |= FMODE_CREATED;
3209 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3210 if (!dir_inode->i_op->create) {
3215 error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry,
3216 mode, open_flag & O_EXCL);
3220 if (unlikely(create_error) && !dentry->d_inode) {
3221 error = create_error;
3228 return ERR_PTR(error);
3231 static const char *open_last_lookups(struct nameidata *nd,
3232 struct file *file, const struct open_flags *op)
3234 struct dentry *dir = nd->path.dentry;
3235 int open_flag = op->open_flag;
3236 bool got_write = false;
3238 struct inode *inode;
3239 struct dentry *dentry;
3243 nd->flags |= op->intent;
3245 if (nd->last_type != LAST_NORM) {
3248 return handle_dots(nd, nd->last_type);
3251 if (!(open_flag & O_CREAT)) {
3252 if (nd->last.name[nd->last.len])
3253 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3254 /* we _can_ be in RCU mode here */
3255 dentry = lookup_fast(nd, &inode, &seq);
3257 return ERR_CAST(dentry);
3261 BUG_ON(nd->flags & LOOKUP_RCU);
3263 /* create side of things */
3264 if (nd->flags & LOOKUP_RCU) {
3265 error = unlazy_walk(nd);
3266 if (unlikely(error))
3267 return ERR_PTR(error);
3269 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3270 /* trailing slashes? */
3271 if (unlikely(nd->last.name[nd->last.len]))
3272 return ERR_PTR(-EISDIR);
3275 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3276 error = mnt_want_write(nd->path.mnt);
3280 * do _not_ fail yet - we might not need that or fail with
3281 * a different error; let lookup_open() decide; we'll be
3282 * dropping this one anyway.
3285 if (open_flag & O_CREAT)
3286 inode_lock(dir->d_inode);
3288 inode_lock_shared(dir->d_inode);
3289 dentry = lookup_open(nd, file, op, got_write);
3290 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3291 fsnotify_create(dir->d_inode, dentry);
3292 if (open_flag & O_CREAT)
3293 inode_unlock(dir->d_inode);
3295 inode_unlock_shared(dir->d_inode);
3298 mnt_drop_write(nd->path.mnt);
3301 return ERR_CAST(dentry);
3303 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3304 dput(nd->path.dentry);
3305 nd->path.dentry = dentry;
3312 res = step_into(nd, WALK_TRAILING, dentry, inode, seq);
3314 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3319 * Handle the last step of open()
3321 static int do_open(struct nameidata *nd,
3322 struct file *file, const struct open_flags *op)
3324 struct user_namespace *mnt_userns;
3325 int open_flag = op->open_flag;
3330 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3331 error = complete_walk(nd);
3335 if (!(file->f_mode & FMODE_CREATED))
3336 audit_inode(nd->name, nd->path.dentry, 0);
3337 mnt_userns = mnt_user_ns(nd->path.mnt);
3338 if (open_flag & O_CREAT) {
3339 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3341 if (d_is_dir(nd->path.dentry))
3343 error = may_create_in_sticky(mnt_userns, nd,
3344 d_backing_inode(nd->path.dentry));
3345 if (unlikely(error))
3348 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3351 do_truncate = false;
3352 acc_mode = op->acc_mode;
3353 if (file->f_mode & FMODE_CREATED) {
3354 /* Don't check for write permission, don't truncate */
3355 open_flag &= ~O_TRUNC;
3357 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3358 error = mnt_want_write(nd->path.mnt);
3363 error = may_open(mnt_userns, &nd->path, acc_mode, open_flag);
3364 if (!error && !(file->f_mode & FMODE_OPENED))
3365 error = vfs_open(&nd->path, file);
3367 error = ima_file_check(file, op->acc_mode);
3368 if (!error && do_truncate)
3369 error = handle_truncate(mnt_userns, file);
3370 if (unlikely(error > 0)) {
3375 mnt_drop_write(nd->path.mnt);
3380 * vfs_tmpfile - create tmpfile
3381 * @mnt_userns: user namespace of the mount the inode was found from
3382 * @dentry: pointer to dentry of the base directory
3383 * @mode: mode of the new tmpfile
3384 * @open_flags: flags
3386 * Create a temporary file.
3388 * If the inode has been found through an idmapped mount the user namespace of
3389 * the vfsmount must be passed through @mnt_userns. This function will then take
3390 * care to map the inode according to @mnt_userns before checking permissions.
3391 * On non-idmapped mounts or if permission checking is to be performed on the
3392 * raw inode simply passs init_user_ns.
3394 struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns,
3395 struct dentry *dentry, umode_t mode, int open_flag)
3397 struct dentry *child = NULL;
3398 struct inode *dir = dentry->d_inode;
3399 struct inode *inode;
3402 /* we want directory to be writable */
3403 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
3406 error = -EOPNOTSUPP;
3407 if (!dir->i_op->tmpfile)
3410 child = d_alloc(dentry, &slash_name);
3411 if (unlikely(!child))
3413 error = dir->i_op->tmpfile(mnt_userns, dir, child, mode);
3417 inode = child->d_inode;
3418 if (unlikely(!inode))
3420 if (!(open_flag & O_EXCL)) {
3421 spin_lock(&inode->i_lock);
3422 inode->i_state |= I_LINKABLE;
3423 spin_unlock(&inode->i_lock);
3425 ima_post_create_tmpfile(mnt_userns, inode);
3430 return ERR_PTR(error);
3432 EXPORT_SYMBOL(vfs_tmpfile);
3434 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3435 const struct open_flags *op,
3438 struct user_namespace *mnt_userns;
3439 struct dentry *child;
3441 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3442 if (unlikely(error))
3444 error = mnt_want_write(path.mnt);
3445 if (unlikely(error))
3447 mnt_userns = mnt_user_ns(path.mnt);
3448 child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag);
3449 error = PTR_ERR(child);
3453 path.dentry = child;
3454 audit_inode(nd->name, child, 0);
3455 /* Don't check for other permissions, the inode was just created */
3456 error = may_open(mnt_userns, &path, 0, op->open_flag);
3459 file->f_path.mnt = path.mnt;
3460 error = finish_open(file, child, NULL);
3462 mnt_drop_write(path.mnt);
3468 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3471 int error = path_lookupat(nd, flags, &path);
3473 audit_inode(nd->name, path.dentry, 0);
3474 error = vfs_open(&path, file);
3480 static struct file *path_openat(struct nameidata *nd,
3481 const struct open_flags *op, unsigned flags)
3486 file = alloc_empty_file(op->open_flag, current_cred());
3490 if (unlikely(file->f_flags & __O_TMPFILE)) {
3491 error = do_tmpfile(nd, flags, op, file);
3492 } else if (unlikely(file->f_flags & O_PATH)) {
3493 error = do_o_path(nd, flags, file);
3495 const char *s = path_init(nd, flags);
3496 while (!(error = link_path_walk(s, nd)) &&
3497 (s = open_last_lookups(nd, file, op)) != NULL)
3500 error = do_open(nd, file, op);
3503 if (likely(!error)) {
3504 if (likely(file->f_mode & FMODE_OPENED))
3510 if (error == -EOPENSTALE) {
3511 if (flags & LOOKUP_RCU)
3516 return ERR_PTR(error);
3519 struct file *do_filp_open(int dfd, struct filename *pathname,
3520 const struct open_flags *op)
3522 struct nameidata nd;
3523 int flags = op->lookup_flags;
3526 set_nameidata(&nd, dfd, pathname);
3527 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3528 if (unlikely(filp == ERR_PTR(-ECHILD)))
3529 filp = path_openat(&nd, op, flags);
3530 if (unlikely(filp == ERR_PTR(-ESTALE)))
3531 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3532 restore_nameidata();
3536 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3537 const char *name, const struct open_flags *op)
3539 struct nameidata nd;
3541 struct filename *filename;
3542 int flags = op->lookup_flags | LOOKUP_ROOT;
3545 nd.root.dentry = dentry;
3547 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3548 return ERR_PTR(-ELOOP);
3550 filename = getname_kernel(name);
3551 if (IS_ERR(filename))
3552 return ERR_CAST(filename);
3554 set_nameidata(&nd, -1, filename);
3555 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3556 if (unlikely(file == ERR_PTR(-ECHILD)))
3557 file = path_openat(&nd, op, flags);
3558 if (unlikely(file == ERR_PTR(-ESTALE)))
3559 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3560 restore_nameidata();
3565 static struct dentry *filename_create(int dfd, struct filename *name,
3566 struct path *path, unsigned int lookup_flags)
3568 struct dentry *dentry = ERR_PTR(-EEXIST);
3573 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3576 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3577 * other flags passed in are ignored!
3579 lookup_flags &= LOOKUP_REVAL;
3581 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3583 return ERR_CAST(name);
3586 * Yucky last component or no last component at all?
3587 * (foo/., foo/.., /////)
3589 if (unlikely(type != LAST_NORM))
3592 /* don't fail immediately if it's r/o, at least try to report other errors */
3593 err2 = mnt_want_write(path->mnt);
3595 * Do the final lookup.
3597 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3598 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3599 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3604 if (d_is_positive(dentry))
3608 * Special case - lookup gave negative, but... we had foo/bar/
3609 * From the vfs_mknod() POV we just have a negative dentry -
3610 * all is fine. Let's be bastards - you had / on the end, you've
3611 * been asking for (non-existent) directory. -ENOENT for you.
3613 if (unlikely(!is_dir && last.name[last.len])) {
3617 if (unlikely(err2)) {
3625 dentry = ERR_PTR(error);
3627 inode_unlock(path->dentry->d_inode);
3629 mnt_drop_write(path->mnt);
3636 struct dentry *kern_path_create(int dfd, const char *pathname,
3637 struct path *path, unsigned int lookup_flags)
3639 return filename_create(dfd, getname_kernel(pathname),
3640 path, lookup_flags);
3642 EXPORT_SYMBOL(kern_path_create);
3644 void done_path_create(struct path *path, struct dentry *dentry)
3647 inode_unlock(path->dentry->d_inode);
3648 mnt_drop_write(path->mnt);
3651 EXPORT_SYMBOL(done_path_create);
3653 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3654 struct path *path, unsigned int lookup_flags)
3656 return filename_create(dfd, getname(pathname), path, lookup_flags);
3658 EXPORT_SYMBOL(user_path_create);
3661 * vfs_mknod - create device node or file
3662 * @mnt_userns: user namespace of the mount the inode was found from
3663 * @dir: inode of @dentry
3664 * @dentry: pointer to dentry of the base directory
3665 * @mode: mode of the new device node or file
3666 * @dev: device number of device to create
3668 * Create a device node or file.
3670 * If the inode has been found through an idmapped mount the user namespace of
3671 * the vfsmount must be passed through @mnt_userns. This function will then take
3672 * care to map the inode according to @mnt_userns before checking permissions.
3673 * On non-idmapped mounts or if permission checking is to be performed on the
3674 * raw inode simply passs init_user_ns.
3676 int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
3677 struct dentry *dentry, umode_t mode, dev_t dev)
3679 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3680 int error = may_create(mnt_userns, dir, dentry);
3685 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3686 !capable(CAP_MKNOD))
3689 if (!dir->i_op->mknod)
3692 error = devcgroup_inode_mknod(mode, dev);
3696 error = security_inode_mknod(dir, dentry, mode, dev);
3700 error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev);
3702 fsnotify_create(dir, dentry);
3705 EXPORT_SYMBOL(vfs_mknod);
3707 static int may_mknod(umode_t mode)
3709 switch (mode & S_IFMT) {
3715 case 0: /* zero mode translates to S_IFREG */
3724 static long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3727 struct user_namespace *mnt_userns;
3728 struct dentry *dentry;
3731 unsigned int lookup_flags = 0;
3733 error = may_mknod(mode);
3737 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3739 return PTR_ERR(dentry);
3741 if (!IS_POSIXACL(path.dentry->d_inode))
3742 mode &= ~current_umask();
3743 error = security_path_mknod(&path, dentry, mode, dev);
3747 mnt_userns = mnt_user_ns(path.mnt);
3748 switch (mode & S_IFMT) {
3749 case 0: case S_IFREG:
3750 error = vfs_create(mnt_userns, path.dentry->d_inode,
3751 dentry, mode, true);
3753 ima_post_path_mknod(mnt_userns, dentry);
3755 case S_IFCHR: case S_IFBLK:
3756 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3757 dentry, mode, new_decode_dev(dev));
3759 case S_IFIFO: case S_IFSOCK:
3760 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3765 done_path_create(&path, dentry);
3766 if (retry_estale(error, lookup_flags)) {
3767 lookup_flags |= LOOKUP_REVAL;
3773 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3776 return do_mknodat(dfd, filename, mode, dev);
3779 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3781 return do_mknodat(AT_FDCWD, filename, mode, dev);
3785 * vfs_mkdir - create directory
3786 * @mnt_userns: user namespace of the mount the inode was found from
3787 * @dir: inode of @dentry
3788 * @dentry: pointer to dentry of the base directory
3789 * @mode: mode of the new directory
3791 * Create a directory.
3793 * If the inode has been found through an idmapped mount the user namespace of
3794 * the vfsmount must be passed through @mnt_userns. This function will then take
3795 * care to map the inode according to @mnt_userns before checking permissions.
3796 * On non-idmapped mounts or if permission checking is to be performed on the
3797 * raw inode simply passs init_user_ns.
3799 int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
3800 struct dentry *dentry, umode_t mode)
3802 int error = may_create(mnt_userns, dir, dentry);
3803 unsigned max_links = dir->i_sb->s_max_links;
3808 if (!dir->i_op->mkdir)
3811 mode &= (S_IRWXUGO|S_ISVTX);
3812 error = security_inode_mkdir(dir, dentry, mode);
3816 if (max_links && dir->i_nlink >= max_links)
3819 error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode);
3821 fsnotify_mkdir(dir, dentry);
3824 EXPORT_SYMBOL(vfs_mkdir);
3826 static long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3828 struct dentry *dentry;
3831 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3834 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3836 return PTR_ERR(dentry);
3838 if (!IS_POSIXACL(path.dentry->d_inode))
3839 mode &= ~current_umask();
3840 error = security_path_mkdir(&path, dentry, mode);
3842 struct user_namespace *mnt_userns;
3843 mnt_userns = mnt_user_ns(path.mnt);
3844 error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry,
3847 done_path_create(&path, dentry);
3848 if (retry_estale(error, lookup_flags)) {
3849 lookup_flags |= LOOKUP_REVAL;
3855 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3857 return do_mkdirat(dfd, pathname, mode);
3860 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3862 return do_mkdirat(AT_FDCWD, pathname, mode);
3866 * vfs_rmdir - remove directory
3867 * @mnt_userns: user namespace of the mount the inode was found from
3868 * @dir: inode of @dentry
3869 * @dentry: pointer to dentry of the base directory
3871 * Remove a directory.
3873 * If the inode has been found through an idmapped mount the user namespace of
3874 * the vfsmount must be passed through @mnt_userns. This function will then take
3875 * care to map the inode according to @mnt_userns before checking permissions.
3876 * On non-idmapped mounts or if permission checking is to be performed on the
3877 * raw inode simply passs init_user_ns.
3879 int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir,
3880 struct dentry *dentry)
3882 int error = may_delete(mnt_userns, dir, dentry, 1);
3887 if (!dir->i_op->rmdir)
3891 inode_lock(dentry->d_inode);
3894 if (is_local_mountpoint(dentry))
3897 error = security_inode_rmdir(dir, dentry);
3901 error = dir->i_op->rmdir(dir, dentry);
3905 shrink_dcache_parent(dentry);
3906 dentry->d_inode->i_flags |= S_DEAD;
3908 detach_mounts(dentry);
3909 fsnotify_rmdir(dir, dentry);
3912 inode_unlock(dentry->d_inode);
3918 EXPORT_SYMBOL(vfs_rmdir);
3920 long do_rmdir(int dfd, struct filename *name)
3922 struct user_namespace *mnt_userns;
3924 struct dentry *dentry;
3928 unsigned int lookup_flags = 0;
3930 name = filename_parentat(dfd, name, lookup_flags,
3931 &path, &last, &type);
3933 return PTR_ERR(name);
3947 error = mnt_want_write(path.mnt);
3951 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3952 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3953 error = PTR_ERR(dentry);
3956 if (!dentry->d_inode) {
3960 error = security_path_rmdir(&path, dentry);
3963 mnt_userns = mnt_user_ns(path.mnt);
3964 error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry);
3968 inode_unlock(path.dentry->d_inode);
3969 mnt_drop_write(path.mnt);
3972 if (retry_estale(error, lookup_flags)) {
3973 lookup_flags |= LOOKUP_REVAL;
3980 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3982 return do_rmdir(AT_FDCWD, getname(pathname));
3986 * vfs_unlink - unlink a filesystem object
3987 * @mnt_userns: user namespace of the mount the inode was found from
3988 * @dir: parent directory
3990 * @delegated_inode: returns victim inode, if the inode is delegated.
3992 * The caller must hold dir->i_mutex.
3994 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3995 * return a reference to the inode in delegated_inode. The caller
3996 * should then break the delegation on that inode and retry. Because
3997 * breaking a delegation may take a long time, the caller should drop
3998 * dir->i_mutex before doing so.
4000 * Alternatively, a caller may pass NULL for delegated_inode. This may
4001 * be appropriate for callers that expect the underlying filesystem not
4002 * to be NFS exported.
4004 * If the inode has been found through an idmapped mount the user namespace of
4005 * the vfsmount must be passed through @mnt_userns. This function will then take
4006 * care to map the inode according to @mnt_userns before checking permissions.
4007 * On non-idmapped mounts or if permission checking is to be performed on the
4008 * raw inode simply passs init_user_ns.
4010 int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir,
4011 struct dentry *dentry, struct inode **delegated_inode)
4013 struct inode *target = dentry->d_inode;
4014 int error = may_delete(mnt_userns, dir, dentry, 0);
4019 if (!dir->i_op->unlink)
4023 if (is_local_mountpoint(dentry))
4026 error = security_inode_unlink(dir, dentry);
4028 error = try_break_deleg(target, delegated_inode);
4031 error = dir->i_op->unlink(dir, dentry);
4034 detach_mounts(dentry);
4035 fsnotify_unlink(dir, dentry);
4040 inode_unlock(target);
4042 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4043 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
4044 fsnotify_link_count(target);
4050 EXPORT_SYMBOL(vfs_unlink);
4053 * Make sure that the actual truncation of the file will occur outside its
4054 * directory's i_mutex. Truncate can take a long time if there is a lot of
4055 * writeout happening, and we don't want to prevent access to the directory
4056 * while waiting on the I/O.
4058 long do_unlinkat(int dfd, struct filename *name)
4061 struct dentry *dentry;
4065 struct inode *inode = NULL;
4066 struct inode *delegated_inode = NULL;
4067 unsigned int lookup_flags = 0;
4069 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4071 return PTR_ERR(name);
4074 if (type != LAST_NORM)
4077 error = mnt_want_write(path.mnt);
4081 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4082 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4083 error = PTR_ERR(dentry);
4084 if (!IS_ERR(dentry)) {
4085 struct user_namespace *mnt_userns;
4087 /* Why not before? Because we want correct error value */
4088 if (last.name[last.len])
4090 inode = dentry->d_inode;
4091 if (d_is_negative(dentry))
4094 error = security_path_unlink(&path, dentry);
4097 mnt_userns = mnt_user_ns(path.mnt);
4098 error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry,
4103 inode_unlock(path.dentry->d_inode);
4105 iput(inode); /* truncate the inode here */
4107 if (delegated_inode) {
4108 error = break_deleg_wait(&delegated_inode);
4112 mnt_drop_write(path.mnt);
4115 if (retry_estale(error, lookup_flags)) {
4116 lookup_flags |= LOOKUP_REVAL;
4124 if (d_is_negative(dentry))
4126 else if (d_is_dir(dentry))
4133 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4135 if ((flag & ~AT_REMOVEDIR) != 0)
4138 if (flag & AT_REMOVEDIR)
4139 return do_rmdir(dfd, getname(pathname));
4140 return do_unlinkat(dfd, getname(pathname));
4143 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4145 return do_unlinkat(AT_FDCWD, getname(pathname));
4149 * vfs_symlink - create symlink
4150 * @mnt_userns: user namespace of the mount the inode was found from
4151 * @dir: inode of @dentry
4152 * @dentry: pointer to dentry of the base directory
4153 * @oldname: name of the file to link to
4157 * If the inode has been found through an idmapped mount the user namespace of
4158 * the vfsmount must be passed through @mnt_userns. This function will then take
4159 * care to map the inode according to @mnt_userns before checking permissions.
4160 * On non-idmapped mounts or if permission checking is to be performed on the
4161 * raw inode simply passs init_user_ns.
4163 int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
4164 struct dentry *dentry, const char *oldname)
4166 int error = may_create(mnt_userns, dir, dentry);
4171 if (!dir->i_op->symlink)
4174 error = security_inode_symlink(dir, dentry, oldname);
4178 error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname);
4180 fsnotify_create(dir, dentry);
4183 EXPORT_SYMBOL(vfs_symlink);
4185 static long do_symlinkat(const char __user *oldname, int newdfd,
4186 const char __user *newname)
4189 struct filename *from;
4190 struct dentry *dentry;
4192 unsigned int lookup_flags = 0;
4194 from = getname(oldname);
4196 return PTR_ERR(from);
4198 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4199 error = PTR_ERR(dentry);
4203 error = security_path_symlink(&path, dentry, from->name);
4205 struct user_namespace *mnt_userns;
4207 mnt_userns = mnt_user_ns(path.mnt);
4208 error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry,
4211 done_path_create(&path, dentry);
4212 if (retry_estale(error, lookup_flags)) {
4213 lookup_flags |= LOOKUP_REVAL;
4221 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4222 int, newdfd, const char __user *, newname)
4224 return do_symlinkat(oldname, newdfd, newname);
4227 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4229 return do_symlinkat(oldname, AT_FDCWD, newname);
4233 * vfs_link - create a new link
4234 * @old_dentry: object to be linked
4235 * @mnt_userns: the user namespace of the mount
4237 * @new_dentry: where to create the new link
4238 * @delegated_inode: returns inode needing a delegation break
4240 * The caller must hold dir->i_mutex
4242 * If vfs_link discovers a delegation on the to-be-linked file in need
4243 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4244 * inode in delegated_inode. The caller should then break the delegation
4245 * and retry. Because breaking a delegation may take a long time, the
4246 * caller should drop the i_mutex before doing so.
4248 * Alternatively, a caller may pass NULL for delegated_inode. This may
4249 * be appropriate for callers that expect the underlying filesystem not
4250 * to be NFS exported.
4252 * If the inode has been found through an idmapped mount the user namespace of
4253 * the vfsmount must be passed through @mnt_userns. This function will then take
4254 * care to map the inode according to @mnt_userns before checking permissions.
4255 * On non-idmapped mounts or if permission checking is to be performed on the
4256 * raw inode simply passs init_user_ns.
4258 int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns,
4259 struct inode *dir, struct dentry *new_dentry,
4260 struct inode **delegated_inode)
4262 struct inode *inode = old_dentry->d_inode;
4263 unsigned max_links = dir->i_sb->s_max_links;
4269 error = may_create(mnt_userns, dir, new_dentry);
4273 if (dir->i_sb != inode->i_sb)
4277 * A link to an append-only or immutable file cannot be created.
4279 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4282 * Updating the link count will likely cause i_uid and i_gid to
4283 * be writen back improperly if their true value is unknown to
4286 if (HAS_UNMAPPED_ID(mnt_userns, inode))
4288 if (!dir->i_op->link)
4290 if (S_ISDIR(inode->i_mode))
4293 error = security_inode_link(old_dentry, dir, new_dentry);
4298 /* Make sure we don't allow creating hardlink to an unlinked file */
4299 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4301 else if (max_links && inode->i_nlink >= max_links)
4304 error = try_break_deleg(inode, delegated_inode);
4306 error = dir->i_op->link(old_dentry, dir, new_dentry);
4309 if (!error && (inode->i_state & I_LINKABLE)) {
4310 spin_lock(&inode->i_lock);
4311 inode->i_state &= ~I_LINKABLE;
4312 spin_unlock(&inode->i_lock);
4314 inode_unlock(inode);
4316 fsnotify_link(dir, inode, new_dentry);
4319 EXPORT_SYMBOL(vfs_link);
4322 * Hardlinks are often used in delicate situations. We avoid
4323 * security-related surprises by not following symlinks on the
4326 * We don't follow them on the oldname either to be compatible
4327 * with linux 2.0, and to avoid hard-linking to directories
4328 * and other special files. --ADM
4330 static int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4331 const char __user *newname, int flags)
4333 struct user_namespace *mnt_userns;
4334 struct dentry *new_dentry;
4335 struct path old_path, new_path;
4336 struct inode *delegated_inode = NULL;
4340 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4343 * To use null names we require CAP_DAC_READ_SEARCH
4344 * This ensures that not everyone will be able to create
4345 * handlink using the passed filedescriptor.
4347 if (flags & AT_EMPTY_PATH) {
4348 if (!capable(CAP_DAC_READ_SEARCH))
4353 if (flags & AT_SYMLINK_FOLLOW)
4354 how |= LOOKUP_FOLLOW;
4356 error = user_path_at(olddfd, oldname, how, &old_path);
4360 new_dentry = user_path_create(newdfd, newname, &new_path,
4361 (how & LOOKUP_REVAL));
4362 error = PTR_ERR(new_dentry);
4363 if (IS_ERR(new_dentry))
4367 if (old_path.mnt != new_path.mnt)
4369 mnt_userns = mnt_user_ns(new_path.mnt);
4370 error = may_linkat(mnt_userns, &old_path);
4371 if (unlikely(error))
4373 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4376 error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode,
4377 new_dentry, &delegated_inode);
4379 done_path_create(&new_path, new_dentry);
4380 if (delegated_inode) {
4381 error = break_deleg_wait(&delegated_inode);
4383 path_put(&old_path);
4387 if (retry_estale(error, how)) {
4388 path_put(&old_path);
4389 how |= LOOKUP_REVAL;
4393 path_put(&old_path);
4398 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4399 int, newdfd, const char __user *, newname, int, flags)
4401 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4404 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4406 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4410 * vfs_rename - rename a filesystem object
4411 * @old_mnt_userns: old user namespace of the mount the inode was found from
4412 * @old_dir: parent of source
4413 * @old_dentry: source
4414 * @new_mnt_userns: new user namespace of the mount the inode was found from
4415 * @new_dir: parent of destination
4416 * @new_dentry: destination
4417 * @delegated_inode: returns an inode needing a delegation break
4418 * @flags: rename flags
4420 * The caller must hold multiple mutexes--see lock_rename()).
4422 * If vfs_rename discovers a delegation in need of breaking at either
4423 * the source or destination, it will return -EWOULDBLOCK and return a
4424 * reference to the inode in delegated_inode. The caller should then
4425 * break the delegation and retry. Because breaking a delegation may
4426 * take a long time, the caller should drop all locks before doing
4429 * Alternatively, a caller may pass NULL for delegated_inode. This may
4430 * be appropriate for callers that expect the underlying filesystem not
4431 * to be NFS exported.
4433 * The worst of all namespace operations - renaming directory. "Perverted"
4434 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4437 * a) we can get into loop creation.
4438 * b) race potential - two innocent renames can create a loop together.
4439 * That's where 4.4 screws up. Current fix: serialization on
4440 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4442 * c) we have to lock _four_ objects - parents and victim (if it exists),
4443 * and source (if it is not a directory).
4444 * And that - after we got ->i_mutex on parents (until then we don't know
4445 * whether the target exists). Solution: try to be smart with locking
4446 * order for inodes. We rely on the fact that tree topology may change
4447 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4448 * move will be locked. Thus we can rank directories by the tree
4449 * (ancestors first) and rank all non-directories after them.
4450 * That works since everybody except rename does "lock parent, lookup,
4451 * lock child" and rename is under ->s_vfs_rename_mutex.
4452 * HOWEVER, it relies on the assumption that any object with ->lookup()
4453 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4454 * we'd better make sure that there's no link(2) for them.
4455 * d) conversion from fhandle to dentry may come in the wrong moment - when
4456 * we are removing the target. Solution: we will have to grab ->i_mutex
4457 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4458 * ->i_mutex on parents, which works but leads to some truly excessive
4461 int vfs_rename(struct renamedata *rd)
4464 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4465 struct dentry *old_dentry = rd->old_dentry;
4466 struct dentry *new_dentry = rd->new_dentry;
4467 struct inode **delegated_inode = rd->delegated_inode;
4468 unsigned int flags = rd->flags;
4469 bool is_dir = d_is_dir(old_dentry);
4470 struct inode *source = old_dentry->d_inode;
4471 struct inode *target = new_dentry->d_inode;
4472 bool new_is_dir = false;
4473 unsigned max_links = new_dir->i_sb->s_max_links;
4474 struct name_snapshot old_name;
4476 if (source == target)
4479 error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir);
4484 error = may_create(rd->new_mnt_userns, new_dir, new_dentry);
4486 new_is_dir = d_is_dir(new_dentry);
4488 if (!(flags & RENAME_EXCHANGE))
4489 error = may_delete(rd->new_mnt_userns, new_dir,
4490 new_dentry, is_dir);
4492 error = may_delete(rd->new_mnt_userns, new_dir,
4493 new_dentry, new_is_dir);
4498 if (!old_dir->i_op->rename)
4502 * If we are going to change the parent - check write permissions,
4503 * we'll need to flip '..'.
4505 if (new_dir != old_dir) {
4507 error = inode_permission(rd->old_mnt_userns, source,
4512 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4513 error = inode_permission(rd->new_mnt_userns, target,
4520 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4525 take_dentry_name_snapshot(&old_name, old_dentry);
4527 if (!is_dir || (flags & RENAME_EXCHANGE))
4528 lock_two_nondirectories(source, target);
4533 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4536 if (max_links && new_dir != old_dir) {
4538 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4540 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4541 old_dir->i_nlink >= max_links)
4545 error = try_break_deleg(source, delegated_inode);
4549 if (target && !new_is_dir) {
4550 error = try_break_deleg(target, delegated_inode);
4554 error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry,
4555 new_dir, new_dentry, flags);
4559 if (!(flags & RENAME_EXCHANGE) && target) {
4561 shrink_dcache_parent(new_dentry);
4562 target->i_flags |= S_DEAD;
4564 dont_mount(new_dentry);
4565 detach_mounts(new_dentry);
4567 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4568 if (!(flags & RENAME_EXCHANGE))
4569 d_move(old_dentry, new_dentry);
4571 d_exchange(old_dentry, new_dentry);
4574 if (!is_dir || (flags & RENAME_EXCHANGE))
4575 unlock_two_nondirectories(source, target);
4577 inode_unlock(target);
4580 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4581 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4582 if (flags & RENAME_EXCHANGE) {
4583 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4584 new_is_dir, NULL, new_dentry);
4587 release_dentry_name_snapshot(&old_name);
4591 EXPORT_SYMBOL(vfs_rename);
4593 int do_renameat2(int olddfd, struct filename *from, int newdfd,
4594 struct filename *to, unsigned int flags)
4596 struct renamedata rd;
4597 struct dentry *old_dentry, *new_dentry;
4598 struct dentry *trap;
4599 struct path old_path, new_path;
4600 struct qstr old_last, new_last;
4601 int old_type, new_type;
4602 struct inode *delegated_inode = NULL;
4603 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4604 bool should_retry = false;
4605 int error = -EINVAL;
4607 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4610 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4611 (flags & RENAME_EXCHANGE))
4614 if (flags & RENAME_EXCHANGE)
4618 from = filename_parentat(olddfd, from, lookup_flags, &old_path,
4619 &old_last, &old_type);
4621 error = PTR_ERR(from);
4625 to = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4628 error = PTR_ERR(to);
4633 if (old_path.mnt != new_path.mnt)
4637 if (old_type != LAST_NORM)
4640 if (flags & RENAME_NOREPLACE)
4642 if (new_type != LAST_NORM)
4645 error = mnt_want_write(old_path.mnt);
4650 trap = lock_rename(new_path.dentry, old_path.dentry);
4652 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4653 error = PTR_ERR(old_dentry);
4654 if (IS_ERR(old_dentry))
4656 /* source must exist */
4658 if (d_is_negative(old_dentry))
4660 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4661 error = PTR_ERR(new_dentry);
4662 if (IS_ERR(new_dentry))
4665 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4667 if (flags & RENAME_EXCHANGE) {
4669 if (d_is_negative(new_dentry))
4672 if (!d_is_dir(new_dentry)) {
4674 if (new_last.name[new_last.len])
4678 /* unless the source is a directory trailing slashes give -ENOTDIR */
4679 if (!d_is_dir(old_dentry)) {
4681 if (old_last.name[old_last.len])
4683 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4686 /* source should not be ancestor of target */
4688 if (old_dentry == trap)
4690 /* target should not be an ancestor of source */
4691 if (!(flags & RENAME_EXCHANGE))
4693 if (new_dentry == trap)
4696 error = security_path_rename(&old_path, old_dentry,
4697 &new_path, new_dentry, flags);
4701 rd.old_dir = old_path.dentry->d_inode;
4702 rd.old_dentry = old_dentry;
4703 rd.old_mnt_userns = mnt_user_ns(old_path.mnt);
4704 rd.new_dir = new_path.dentry->d_inode;
4705 rd.new_dentry = new_dentry;
4706 rd.new_mnt_userns = mnt_user_ns(new_path.mnt);
4707 rd.delegated_inode = &delegated_inode;
4709 error = vfs_rename(&rd);
4715 unlock_rename(new_path.dentry, old_path.dentry);
4716 if (delegated_inode) {
4717 error = break_deleg_wait(&delegated_inode);
4721 mnt_drop_write(old_path.mnt);
4723 if (retry_estale(error, lookup_flags))
4724 should_retry = true;
4725 path_put(&new_path);
4727 path_put(&old_path);
4729 should_retry = false;
4730 lookup_flags |= LOOKUP_REVAL;
4742 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4743 int, newdfd, const char __user *, newname, unsigned int, flags)
4745 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4749 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4750 int, newdfd, const char __user *, newname)
4752 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4756 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4758 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4759 getname(newname), 0);
4762 int readlink_copy(char __user *buffer, int buflen, const char *link)
4764 int len = PTR_ERR(link);
4769 if (len > (unsigned) buflen)
4771 if (copy_to_user(buffer, link, len))
4778 * vfs_readlink - copy symlink body into userspace buffer
4779 * @dentry: dentry on which to get symbolic link
4780 * @buffer: user memory pointer
4781 * @buflen: size of buffer
4783 * Does not touch atime. That's up to the caller if necessary
4785 * Does not call security hook.
4787 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4789 struct inode *inode = d_inode(dentry);
4790 DEFINE_DELAYED_CALL(done);
4794 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4795 if (unlikely(inode->i_op->readlink))
4796 return inode->i_op->readlink(dentry, buffer, buflen);
4798 if (!d_is_symlink(dentry))
4801 spin_lock(&inode->i_lock);
4802 inode->i_opflags |= IOP_DEFAULT_READLINK;
4803 spin_unlock(&inode->i_lock);
4806 link = READ_ONCE(inode->i_link);
4808 link = inode->i_op->get_link(dentry, inode, &done);
4810 return PTR_ERR(link);
4812 res = readlink_copy(buffer, buflen, link);
4813 do_delayed_call(&done);
4816 EXPORT_SYMBOL(vfs_readlink);
4819 * vfs_get_link - get symlink body
4820 * @dentry: dentry on which to get symbolic link
4821 * @done: caller needs to free returned data with this
4823 * Calls security hook and i_op->get_link() on the supplied inode.
4825 * It does not touch atime. That's up to the caller if necessary.
4827 * Does not work on "special" symlinks like /proc/$$/fd/N
4829 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4831 const char *res = ERR_PTR(-EINVAL);
4832 struct inode *inode = d_inode(dentry);
4834 if (d_is_symlink(dentry)) {
4835 res = ERR_PTR(security_inode_readlink(dentry));
4837 res = inode->i_op->get_link(dentry, inode, done);
4841 EXPORT_SYMBOL(vfs_get_link);
4843 /* get the link contents into pagecache */
4844 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4845 struct delayed_call *callback)
4849 struct address_space *mapping = inode->i_mapping;
4852 page = find_get_page(mapping, 0);
4854 return ERR_PTR(-ECHILD);
4855 if (!PageUptodate(page)) {
4857 return ERR_PTR(-ECHILD);
4860 page = read_mapping_page(mapping, 0, NULL);
4864 set_delayed_call(callback, page_put_link, page);
4865 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4866 kaddr = page_address(page);
4867 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4871 EXPORT_SYMBOL(page_get_link);
4873 void page_put_link(void *arg)
4877 EXPORT_SYMBOL(page_put_link);
4879 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4881 DEFINE_DELAYED_CALL(done);
4882 int res = readlink_copy(buffer, buflen,
4883 page_get_link(dentry, d_inode(dentry),
4885 do_delayed_call(&done);
4888 EXPORT_SYMBOL(page_readlink);
4891 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4893 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4895 struct address_space *mapping = inode->i_mapping;
4899 unsigned int flags = 0;
4901 flags |= AOP_FLAG_NOFS;
4904 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4905 flags, &page, &fsdata);
4909 memcpy(page_address(page), symname, len-1);
4911 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4918 mark_inode_dirty(inode);
4923 EXPORT_SYMBOL(__page_symlink);
4925 int page_symlink(struct inode *inode, const char *symname, int len)
4927 return __page_symlink(inode, symname, len,
4928 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4930 EXPORT_SYMBOL(page_symlink);
4932 const struct inode_operations page_symlink_inode_operations = {
4933 .get_link = page_get_link,
4935 EXPORT_SYMBOL(page_symlink_inode_operations);