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;
583 p->path.dentry = NULL;
584 p->total_link_count = old ? old->total_link_count : 0;
586 current->nameidata = p;
589 static void restore_nameidata(void)
591 struct nameidata *now = current->nameidata, *old = now->saved;
593 current->nameidata = old;
595 old->total_link_count = now->total_link_count;
596 if (now->stack != now->internal)
600 static bool nd_alloc_stack(struct nameidata *nd)
604 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
605 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
608 memcpy(p, nd->internal, sizeof(nd->internal));
614 * path_connected - Verify that a dentry is below mnt.mnt_root
616 * Rename can sometimes move a file or directory outside of a bind
617 * mount, path_connected allows those cases to be detected.
619 static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
621 struct super_block *sb = mnt->mnt_sb;
623 /* Bind mounts can have disconnected paths */
624 if (mnt->mnt_root == sb->s_root)
627 return is_subdir(dentry, mnt->mnt_root);
630 static void drop_links(struct nameidata *nd)
634 struct saved *last = nd->stack + i;
635 do_delayed_call(&last->done);
636 clear_delayed_call(&last->done);
640 static void terminate_walk(struct nameidata *nd)
643 if (!(nd->flags & LOOKUP_RCU)) {
646 for (i = 0; i < nd->depth; i++)
647 path_put(&nd->stack[i].link);
648 if (nd->flags & LOOKUP_ROOT_GRABBED) {
650 nd->flags &= ~LOOKUP_ROOT_GRABBED;
653 nd->flags &= ~LOOKUP_RCU;
658 nd->path.dentry = NULL;
661 /* path_put is needed afterwards regardless of success or failure */
662 static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
664 int res = __legitimize_mnt(path->mnt, mseq);
671 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
675 return !read_seqcount_retry(&path->dentry->d_seq, seq);
678 static inline bool legitimize_path(struct nameidata *nd,
679 struct path *path, unsigned seq)
681 return __legitimize_path(path, seq, nd->m_seq);
684 static bool legitimize_links(struct nameidata *nd)
687 if (unlikely(nd->flags & LOOKUP_CACHED)) {
692 for (i = 0; i < nd->depth; i++) {
693 struct saved *last = nd->stack + i;
694 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
703 static bool legitimize_root(struct nameidata *nd)
706 * For scoped-lookups (where nd->root has been zeroed), we need to
707 * restart the whole lookup from scratch -- because set_root() is wrong
708 * for these lookups (nd->dfd is the root, not the filesystem root).
710 if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED))
712 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
713 if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT))
715 nd->flags |= LOOKUP_ROOT_GRABBED;
716 return legitimize_path(nd, &nd->root, nd->root_seq);
720 * Path walking has 2 modes, rcu-walk and ref-walk (see
721 * Documentation/filesystems/path-lookup.txt). In situations when we can't
722 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
723 * normal reference counts on dentries and vfsmounts to transition to ref-walk
724 * mode. Refcounts are grabbed at the last known good point before rcu-walk
725 * got stuck, so ref-walk may continue from there. If this is not successful
726 * (eg. a seqcount has changed), then failure is returned and it's up to caller
727 * to restart the path walk from the beginning in ref-walk mode.
731 * try_to_unlazy - try to switch to ref-walk mode.
732 * @nd: nameidata pathwalk data
733 * Returns: true on success, false on failure
735 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
737 * Must be called from rcu-walk context.
738 * Nothing should touch nameidata between try_to_unlazy() failure and
741 static bool try_to_unlazy(struct nameidata *nd)
743 struct dentry *parent = nd->path.dentry;
745 BUG_ON(!(nd->flags & LOOKUP_RCU));
747 nd->flags &= ~LOOKUP_RCU;
748 if (unlikely(!legitimize_links(nd)))
750 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
752 if (unlikely(!legitimize_root(nd)))
755 BUG_ON(nd->inode != parent->d_inode);
760 nd->path.dentry = NULL;
767 * try_to_unlazy_next - try to switch to ref-walk mode.
768 * @nd: nameidata pathwalk data
769 * @dentry: next dentry to step into
770 * @seq: seq number to check @dentry against
771 * Returns: true on success, false on failure
773 * Similar to to try_to_unlazy(), but here we have the next dentry already
774 * picked by rcu-walk and want to legitimize that in addition to the current
775 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
776 * Nothing should touch nameidata between try_to_unlazy_next() failure and
779 static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq)
781 BUG_ON(!(nd->flags & LOOKUP_RCU));
783 nd->flags &= ~LOOKUP_RCU;
784 if (unlikely(!legitimize_links(nd)))
786 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
788 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
792 * We need to move both the parent and the dentry from the RCU domain
793 * to be properly refcounted. And the sequence number in the dentry
794 * validates *both* dentry counters, since we checked the sequence
795 * number of the parent after we got the child sequence number. So we
796 * know the parent must still be valid if the child sequence number is
798 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
800 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
803 * Sequence counts matched. Now make sure that the root is
804 * still valid and get it if required.
806 if (unlikely(!legitimize_root(nd)))
814 nd->path.dentry = NULL;
824 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
826 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
827 return dentry->d_op->d_revalidate(dentry, flags);
833 * complete_walk - successful completion of path walk
834 * @nd: pointer nameidata
836 * If we had been in RCU mode, drop out of it and legitimize nd->path.
837 * Revalidate the final result, unless we'd already done that during
838 * the path walk or the filesystem doesn't ask for it. Return 0 on
839 * success, -error on failure. In case of failure caller does not
840 * need to drop nd->path.
842 static int complete_walk(struct nameidata *nd)
844 struct dentry *dentry = nd->path.dentry;
847 if (nd->flags & LOOKUP_RCU) {
849 * We don't want to zero nd->root for scoped-lookups or
850 * externally-managed nd->root.
852 if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED)))
854 nd->flags &= ~LOOKUP_CACHED;
855 if (!try_to_unlazy(nd))
859 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
861 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
862 * ever step outside the root during lookup" and should already
863 * be guaranteed by the rest of namei, we want to avoid a namei
864 * BUG resulting in userspace being given a path that was not
865 * scoped within the root at some point during the lookup.
867 * So, do a final sanity-check to make sure that in the
868 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
869 * we won't silently return an fd completely outside of the
870 * requested root to userspace.
872 * Userspace could move the path outside the root after this
873 * check, but as discussed elsewhere this is not a concern (the
874 * resolved file was inside the root at some point).
876 if (!path_is_under(&nd->path, &nd->root))
880 if (likely(!(nd->flags & LOOKUP_JUMPED)))
883 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
886 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
896 static int set_root(struct nameidata *nd)
898 struct fs_struct *fs = current->fs;
901 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
902 * still have to ensure it doesn't happen because it will cause a breakout
905 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
906 return -ENOTRECOVERABLE;
908 if (nd->flags & LOOKUP_RCU) {
912 seq = read_seqcount_begin(&fs->seq);
914 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
915 } while (read_seqcount_retry(&fs->seq, seq));
917 get_fs_root(fs, &nd->root);
918 nd->flags |= LOOKUP_ROOT_GRABBED;
923 static int nd_jump_root(struct nameidata *nd)
925 if (unlikely(nd->flags & LOOKUP_BENEATH))
927 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
928 /* Absolute path arguments to path_init() are allowed. */
929 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
933 int error = set_root(nd);
937 if (nd->flags & LOOKUP_RCU) {
941 nd->inode = d->d_inode;
942 nd->seq = nd->root_seq;
943 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
949 nd->inode = nd->path.dentry->d_inode;
951 nd->flags |= LOOKUP_JUMPED;
956 * Helper to directly jump to a known parsed path from ->get_link,
957 * caller must have taken a reference to path beforehand.
959 int nd_jump_link(struct path *path)
962 struct nameidata *nd = current->nameidata;
964 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
968 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
969 if (nd->path.mnt != path->mnt)
972 /* Not currently safe for scoped-lookups. */
973 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
978 nd->inode = nd->path.dentry->d_inode;
979 nd->flags |= LOOKUP_JUMPED;
987 static inline void put_link(struct nameidata *nd)
989 struct saved *last = nd->stack + --nd->depth;
990 do_delayed_call(&last->done);
991 if (!(nd->flags & LOOKUP_RCU))
992 path_put(&last->link);
995 int sysctl_protected_symlinks __read_mostly = 0;
996 int sysctl_protected_hardlinks __read_mostly = 0;
997 int sysctl_protected_fifos __read_mostly;
998 int sysctl_protected_regular __read_mostly;
1001 * may_follow_link - Check symlink following for unsafe situations
1002 * @nd: nameidata pathwalk data
1004 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1005 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1006 * in a sticky world-writable directory. This is to protect privileged
1007 * processes from failing races against path names that may change out
1008 * from under them by way of other users creating malicious symlinks.
1009 * It will permit symlinks to be followed only when outside a sticky
1010 * world-writable directory, or when the uid of the symlink and follower
1011 * match, or when the directory owner matches the symlink's owner.
1013 * Returns 0 if following the symlink is allowed, -ve on error.
1015 static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1017 struct user_namespace *mnt_userns;
1020 if (!sysctl_protected_symlinks)
1023 mnt_userns = mnt_user_ns(nd->path.mnt);
1024 i_uid = i_uid_into_mnt(mnt_userns, inode);
1025 /* Allowed if owner and follower match. */
1026 if (uid_eq(current_cred()->fsuid, i_uid))
1029 /* Allowed if parent directory not sticky and world-writable. */
1030 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1033 /* Allowed if parent directory and link owner match. */
1034 if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid))
1037 if (nd->flags & LOOKUP_RCU)
1040 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1041 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1046 * safe_hardlink_source - Check for safe hardlink conditions
1047 * @mnt_userns: user namespace of the mount the inode was found from
1048 * @inode: the source inode to hardlink from
1050 * Return false if at least one of the following conditions:
1051 * - inode is not a regular file
1053 * - inode is setgid and group-exec
1054 * - access failure for read and write
1056 * Otherwise returns true.
1058 static bool safe_hardlink_source(struct user_namespace *mnt_userns,
1059 struct inode *inode)
1061 umode_t mode = inode->i_mode;
1063 /* Special files should not get pinned to the filesystem. */
1067 /* Setuid files should not get pinned to the filesystem. */
1071 /* Executable setgid files should not get pinned to the filesystem. */
1072 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1075 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1076 if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE))
1083 * may_linkat - Check permissions for creating a hardlink
1084 * @mnt_userns: user namespace of the mount the inode was found from
1085 * @link: the source to hardlink from
1087 * Block hardlink when all of:
1088 * - sysctl_protected_hardlinks enabled
1089 * - fsuid does not match inode
1090 * - hardlink source is unsafe (see safe_hardlink_source() above)
1091 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1093 * If the inode has been found through an idmapped mount the user namespace of
1094 * the vfsmount must be passed through @mnt_userns. This function will then take
1095 * care to map the inode according to @mnt_userns before checking permissions.
1096 * On non-idmapped mounts or if permission checking is to be performed on the
1097 * raw inode simply passs init_user_ns.
1099 * Returns 0 if successful, -ve on error.
1101 int may_linkat(struct user_namespace *mnt_userns, struct path *link)
1103 struct inode *inode = link->dentry->d_inode;
1105 /* Inode writeback is not safe when the uid or gid are invalid. */
1106 if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
1107 !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
1110 if (!sysctl_protected_hardlinks)
1113 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1114 * otherwise, it must be a safe source.
1116 if (safe_hardlink_source(mnt_userns, inode) ||
1117 inode_owner_or_capable(mnt_userns, inode))
1120 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1125 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1126 * should be allowed, or not, on files that already
1128 * @mnt_userns: user namespace of the mount the inode was found from
1129 * @dir_mode: mode bits of directory
1130 * @dir_uid: owner of directory
1131 * @inode: the inode of the file to open
1133 * Block an O_CREAT open of a FIFO (or a regular file) when:
1134 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1135 * - the file already exists
1136 * - we are in a sticky directory
1137 * - we don't own the file
1138 * - the owner of the directory doesn't own the file
1139 * - the directory is world writable
1140 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1141 * the directory doesn't have to be world writable: being group writable will
1144 * If the inode has been found through an idmapped mount the user namespace of
1145 * the vfsmount must be passed through @mnt_userns. This function will then take
1146 * care to map the inode according to @mnt_userns before checking permissions.
1147 * On non-idmapped mounts or if permission checking is to be performed on the
1148 * raw inode simply passs init_user_ns.
1150 * Returns 0 if the open is allowed, -ve on error.
1152 static int may_create_in_sticky(struct user_namespace *mnt_userns,
1153 struct nameidata *nd, struct inode *const inode)
1155 umode_t dir_mode = nd->dir_mode;
1156 kuid_t dir_uid = nd->dir_uid;
1158 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1159 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1160 likely(!(dir_mode & S_ISVTX)) ||
1161 uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) ||
1162 uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode)))
1165 if (likely(dir_mode & 0002) ||
1167 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1168 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1169 const char *operation = S_ISFIFO(inode->i_mode) ?
1170 "sticky_create_fifo" :
1171 "sticky_create_regular";
1172 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1179 * follow_up - Find the mountpoint of path's vfsmount
1181 * Given a path, find the mountpoint of its source file system.
1182 * Replace @path with the path of the mountpoint in the parent mount.
1185 * Return 1 if we went up a level and 0 if we were already at the
1188 int follow_up(struct path *path)
1190 struct mount *mnt = real_mount(path->mnt);
1191 struct mount *parent;
1192 struct dentry *mountpoint;
1194 read_seqlock_excl(&mount_lock);
1195 parent = mnt->mnt_parent;
1196 if (parent == mnt) {
1197 read_sequnlock_excl(&mount_lock);
1200 mntget(&parent->mnt);
1201 mountpoint = dget(mnt->mnt_mountpoint);
1202 read_sequnlock_excl(&mount_lock);
1204 path->dentry = mountpoint;
1206 path->mnt = &parent->mnt;
1209 EXPORT_SYMBOL(follow_up);
1211 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1212 struct path *path, unsigned *seqp)
1214 while (mnt_has_parent(m)) {
1215 struct dentry *mountpoint = m->mnt_mountpoint;
1218 if (unlikely(root->dentry == mountpoint &&
1219 root->mnt == &m->mnt))
1221 if (mountpoint != m->mnt.mnt_root) {
1222 path->mnt = &m->mnt;
1223 path->dentry = mountpoint;
1224 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1231 static bool choose_mountpoint(struct mount *m, const struct path *root,
1238 unsigned seq, mseq = read_seqbegin(&mount_lock);
1240 found = choose_mountpoint_rcu(m, root, path, &seq);
1241 if (unlikely(!found)) {
1242 if (!read_seqretry(&mount_lock, mseq))
1245 if (likely(__legitimize_path(path, seq, mseq)))
1257 * Perform an automount
1258 * - return -EISDIR to tell follow_managed() to stop and return the path we
1261 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1263 struct dentry *dentry = path->dentry;
1265 /* We don't want to mount if someone's just doing a stat -
1266 * unless they're stat'ing a directory and appended a '/' to
1269 * We do, however, want to mount if someone wants to open or
1270 * create a file of any type under the mountpoint, wants to
1271 * traverse through the mountpoint or wants to open the
1272 * mounted directory. Also, autofs may mark negative dentries
1273 * as being automount points. These will need the attentions
1274 * of the daemon to instantiate them before they can be used.
1276 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1277 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1281 if (count && (*count)++ >= MAXSYMLINKS)
1284 return finish_automount(dentry->d_op->d_automount(path), path);
1288 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1289 * dentries are pinned but not locked here, so negative dentry can go
1290 * positive right under us. Use of smp_load_acquire() provides a barrier
1291 * sufficient for ->d_inode and ->d_flags consistency.
1293 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1294 int *count, unsigned lookup_flags)
1296 struct vfsmount *mnt = path->mnt;
1297 bool need_mntput = false;
1300 while (flags & DCACHE_MANAGED_DENTRY) {
1301 /* Allow the filesystem to manage the transit without i_mutex
1303 if (flags & DCACHE_MANAGE_TRANSIT) {
1304 ret = path->dentry->d_op->d_manage(path, false);
1305 flags = smp_load_acquire(&path->dentry->d_flags);
1310 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1311 struct vfsmount *mounted = lookup_mnt(path);
1312 if (mounted) { // ... in our namespace
1316 path->mnt = mounted;
1317 path->dentry = dget(mounted->mnt_root);
1318 // here we know it's positive
1319 flags = path->dentry->d_flags;
1325 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1328 // uncovered automount point
1329 ret = follow_automount(path, count, lookup_flags);
1330 flags = smp_load_acquire(&path->dentry->d_flags);
1337 // possible if you race with several mount --move
1338 if (need_mntput && path->mnt == mnt)
1340 if (!ret && unlikely(d_flags_negative(flags)))
1342 *jumped = need_mntput;
1346 static inline int traverse_mounts(struct path *path, bool *jumped,
1347 int *count, unsigned lookup_flags)
1349 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1352 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1354 if (unlikely(d_flags_negative(flags)))
1358 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1361 int follow_down_one(struct path *path)
1363 struct vfsmount *mounted;
1365 mounted = lookup_mnt(path);
1369 path->mnt = mounted;
1370 path->dentry = dget(mounted->mnt_root);
1375 EXPORT_SYMBOL(follow_down_one);
1378 * Follow down to the covering mount currently visible to userspace. At each
1379 * point, the filesystem owning that dentry may be queried as to whether the
1380 * caller is permitted to proceed or not.
1382 int follow_down(struct path *path)
1384 struct vfsmount *mnt = path->mnt;
1386 int ret = traverse_mounts(path, &jumped, NULL, 0);
1388 if (path->mnt != mnt)
1392 EXPORT_SYMBOL(follow_down);
1395 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1396 * we meet a managed dentry that would need blocking.
1398 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1399 struct inode **inode, unsigned *seqp)
1401 struct dentry *dentry = path->dentry;
1402 unsigned int flags = dentry->d_flags;
1404 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1407 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1412 * Don't forget we might have a non-mountpoint managed dentry
1413 * that wants to block transit.
1415 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1416 int res = dentry->d_op->d_manage(path, true);
1418 return res == -EISDIR;
1419 flags = dentry->d_flags;
1422 if (flags & DCACHE_MOUNTED) {
1423 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1425 path->mnt = &mounted->mnt;
1426 dentry = path->dentry = mounted->mnt.mnt_root;
1427 nd->flags |= LOOKUP_JUMPED;
1428 *seqp = read_seqcount_begin(&dentry->d_seq);
1429 *inode = dentry->d_inode;
1431 * We don't need to re-check ->d_seq after this
1432 * ->d_inode read - there will be an RCU delay
1433 * between mount hash removal and ->mnt_root
1434 * becoming unpinned.
1436 flags = dentry->d_flags;
1439 if (read_seqretry(&mount_lock, nd->m_seq))
1442 return !(flags & DCACHE_NEED_AUTOMOUNT);
1446 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1447 struct path *path, struct inode **inode,
1453 path->mnt = nd->path.mnt;
1454 path->dentry = dentry;
1455 if (nd->flags & LOOKUP_RCU) {
1456 unsigned int seq = *seqp;
1457 if (unlikely(!*inode))
1459 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1461 if (!try_to_unlazy_next(nd, dentry, seq))
1463 // *path might've been clobbered by __follow_mount_rcu()
1464 path->mnt = nd->path.mnt;
1465 path->dentry = dentry;
1467 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1469 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1472 nd->flags |= LOOKUP_JUMPED;
1474 if (unlikely(ret)) {
1476 if (path->mnt != nd->path.mnt)
1479 *inode = d_backing_inode(path->dentry);
1480 *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1486 * This looks up the name in dcache and possibly revalidates the found dentry.
1487 * NULL is returned if the dentry does not exist in the cache.
1489 static struct dentry *lookup_dcache(const struct qstr *name,
1493 struct dentry *dentry = d_lookup(dir, name);
1495 int error = d_revalidate(dentry, flags);
1496 if (unlikely(error <= 0)) {
1498 d_invalidate(dentry);
1500 return ERR_PTR(error);
1507 * Parent directory has inode locked exclusive. This is one
1508 * and only case when ->lookup() gets called on non in-lookup
1509 * dentries - as the matter of fact, this only gets called
1510 * when directory is guaranteed to have no in-lookup children
1513 static struct dentry *__lookup_hash(const struct qstr *name,
1514 struct dentry *base, unsigned int flags)
1516 struct dentry *dentry = lookup_dcache(name, base, flags);
1518 struct inode *dir = base->d_inode;
1523 /* Don't create child dentry for a dead directory. */
1524 if (unlikely(IS_DEADDIR(dir)))
1525 return ERR_PTR(-ENOENT);
1527 dentry = d_alloc(base, name);
1528 if (unlikely(!dentry))
1529 return ERR_PTR(-ENOMEM);
1531 old = dir->i_op->lookup(dir, dentry, flags);
1532 if (unlikely(old)) {
1539 static struct dentry *lookup_fast(struct nameidata *nd,
1540 struct inode **inode,
1543 struct dentry *dentry, *parent = nd->path.dentry;
1547 * Rename seqlock is not required here because in the off chance
1548 * of a false negative due to a concurrent rename, the caller is
1549 * going to fall back to non-racy lookup.
1551 if (nd->flags & LOOKUP_RCU) {
1553 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1554 if (unlikely(!dentry)) {
1555 if (!try_to_unlazy(nd))
1556 return ERR_PTR(-ECHILD);
1561 * This sequence count validates that the inode matches
1562 * the dentry name information from lookup.
1564 *inode = d_backing_inode(dentry);
1565 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1566 return ERR_PTR(-ECHILD);
1569 * This sequence count validates that the parent had no
1570 * changes while we did the lookup of the dentry above.
1572 * The memory barrier in read_seqcount_begin of child is
1573 * enough, we can use __read_seqcount_retry here.
1575 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1576 return ERR_PTR(-ECHILD);
1579 status = d_revalidate(dentry, nd->flags);
1580 if (likely(status > 0))
1582 if (!try_to_unlazy_next(nd, dentry, seq))
1583 return ERR_PTR(-ECHILD);
1584 if (status == -ECHILD)
1585 /* we'd been told to redo it in non-rcu mode */
1586 status = d_revalidate(dentry, nd->flags);
1588 dentry = __d_lookup(parent, &nd->last);
1589 if (unlikely(!dentry))
1591 status = d_revalidate(dentry, nd->flags);
1593 if (unlikely(status <= 0)) {
1595 d_invalidate(dentry);
1597 return ERR_PTR(status);
1602 /* Fast lookup failed, do it the slow way */
1603 static struct dentry *__lookup_slow(const struct qstr *name,
1607 struct dentry *dentry, *old;
1608 struct inode *inode = dir->d_inode;
1609 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1611 /* Don't go there if it's already dead */
1612 if (unlikely(IS_DEADDIR(inode)))
1613 return ERR_PTR(-ENOENT);
1615 dentry = d_alloc_parallel(dir, name, &wq);
1618 if (unlikely(!d_in_lookup(dentry))) {
1619 int error = d_revalidate(dentry, flags);
1620 if (unlikely(error <= 0)) {
1622 d_invalidate(dentry);
1627 dentry = ERR_PTR(error);
1630 old = inode->i_op->lookup(inode, dentry, flags);
1631 d_lookup_done(dentry);
1632 if (unlikely(old)) {
1640 static struct dentry *lookup_slow(const struct qstr *name,
1644 struct inode *inode = dir->d_inode;
1646 inode_lock_shared(inode);
1647 res = __lookup_slow(name, dir, flags);
1648 inode_unlock_shared(inode);
1652 static inline int may_lookup(struct user_namespace *mnt_userns,
1653 struct nameidata *nd)
1655 if (nd->flags & LOOKUP_RCU) {
1656 int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1657 if (err != -ECHILD || !try_to_unlazy(nd))
1660 return inode_permission(mnt_userns, nd->inode, MAY_EXEC);
1663 static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq)
1665 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1668 if (likely(nd->depth != EMBEDDED_LEVELS))
1670 if (likely(nd->stack != nd->internal))
1672 if (likely(nd_alloc_stack(nd)))
1675 if (nd->flags & LOOKUP_RCU) {
1676 // we need to grab link before we do unlazy. And we can't skip
1677 // unlazy even if we fail to grab the link - cleanup needs it
1678 bool grabbed_link = legitimize_path(nd, link, seq);
1680 if (!try_to_unlazy(nd) != 0 || !grabbed_link)
1683 if (nd_alloc_stack(nd))
1689 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1691 static const char *pick_link(struct nameidata *nd, struct path *link,
1692 struct inode *inode, unsigned seq, int flags)
1696 int error = reserve_stack(nd, link, seq);
1698 if (unlikely(error)) {
1699 if (!(nd->flags & LOOKUP_RCU))
1701 return ERR_PTR(error);
1703 last = nd->stack + nd->depth++;
1705 clear_delayed_call(&last->done);
1708 if (flags & WALK_TRAILING) {
1709 error = may_follow_link(nd, inode);
1710 if (unlikely(error))
1711 return ERR_PTR(error);
1714 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1715 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1716 return ERR_PTR(-ELOOP);
1718 if (!(nd->flags & LOOKUP_RCU)) {
1719 touch_atime(&last->link);
1721 } else if (atime_needs_update(&last->link, inode)) {
1722 if (!try_to_unlazy(nd))
1723 return ERR_PTR(-ECHILD);
1724 touch_atime(&last->link);
1727 error = security_inode_follow_link(link->dentry, inode,
1728 nd->flags & LOOKUP_RCU);
1729 if (unlikely(error))
1730 return ERR_PTR(error);
1732 res = READ_ONCE(inode->i_link);
1734 const char * (*get)(struct dentry *, struct inode *,
1735 struct delayed_call *);
1736 get = inode->i_op->get_link;
1737 if (nd->flags & LOOKUP_RCU) {
1738 res = get(NULL, inode, &last->done);
1739 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1740 res = get(link->dentry, inode, &last->done);
1742 res = get(link->dentry, inode, &last->done);
1750 error = nd_jump_root(nd);
1751 if (unlikely(error))
1752 return ERR_PTR(error);
1753 while (unlikely(*++res == '/'))
1758 all_done: // pure jump
1764 * Do we need to follow links? We _really_ want to be able
1765 * to do this check without having to look at inode->i_op,
1766 * so we keep a cache of "no, this doesn't need follow_link"
1767 * for the common case.
1769 static const char *step_into(struct nameidata *nd, int flags,
1770 struct dentry *dentry, struct inode *inode, unsigned seq)
1773 int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1776 return ERR_PTR(err);
1777 if (likely(!d_is_symlink(path.dentry)) ||
1778 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1779 (flags & WALK_NOFOLLOW)) {
1780 /* not a symlink or should not follow */
1781 if (!(nd->flags & LOOKUP_RCU)) {
1782 dput(nd->path.dentry);
1783 if (nd->path.mnt != path.mnt)
1784 mntput(nd->path.mnt);
1791 if (nd->flags & LOOKUP_RCU) {
1792 /* make sure that d_is_symlink above matches inode */
1793 if (read_seqcount_retry(&path.dentry->d_seq, seq))
1794 return ERR_PTR(-ECHILD);
1796 if (path.mnt == nd->path.mnt)
1799 return pick_link(nd, &path, inode, seq, flags);
1802 static struct dentry *follow_dotdot_rcu(struct nameidata *nd,
1803 struct inode **inodep,
1806 struct dentry *parent, *old;
1808 if (path_equal(&nd->path, &nd->root))
1810 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1813 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1814 &nd->root, &path, &seq))
1816 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1817 return ERR_PTR(-ECHILD);
1819 nd->inode = path.dentry->d_inode;
1821 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1822 return ERR_PTR(-ECHILD);
1823 /* we know that mountpoint was pinned */
1825 old = nd->path.dentry;
1826 parent = old->d_parent;
1827 *inodep = parent->d_inode;
1828 *seqp = read_seqcount_begin(&parent->d_seq);
1829 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1830 return ERR_PTR(-ECHILD);
1831 if (unlikely(!path_connected(nd->path.mnt, parent)))
1832 return ERR_PTR(-ECHILD);
1835 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1836 return ERR_PTR(-ECHILD);
1837 if (unlikely(nd->flags & LOOKUP_BENEATH))
1838 return ERR_PTR(-ECHILD);
1842 static struct dentry *follow_dotdot(struct nameidata *nd,
1843 struct inode **inodep,
1846 struct dentry *parent;
1848 if (path_equal(&nd->path, &nd->root))
1850 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1853 if (!choose_mountpoint(real_mount(nd->path.mnt),
1856 path_put(&nd->path);
1858 nd->inode = path.dentry->d_inode;
1859 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1860 return ERR_PTR(-EXDEV);
1862 /* rare case of legitimate dget_parent()... */
1863 parent = dget_parent(nd->path.dentry);
1864 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1866 return ERR_PTR(-ENOENT);
1869 *inodep = parent->d_inode;
1873 if (unlikely(nd->flags & LOOKUP_BENEATH))
1874 return ERR_PTR(-EXDEV);
1875 dget(nd->path.dentry);
1879 static const char *handle_dots(struct nameidata *nd, int type)
1881 if (type == LAST_DOTDOT) {
1882 const char *error = NULL;
1883 struct dentry *parent;
1884 struct inode *inode;
1887 if (!nd->root.mnt) {
1888 error = ERR_PTR(set_root(nd));
1892 if (nd->flags & LOOKUP_RCU)
1893 parent = follow_dotdot_rcu(nd, &inode, &seq);
1895 parent = follow_dotdot(nd, &inode, &seq);
1897 return ERR_CAST(parent);
1898 if (unlikely(!parent))
1899 error = step_into(nd, WALK_NOFOLLOW,
1900 nd->path.dentry, nd->inode, nd->seq);
1902 error = step_into(nd, WALK_NOFOLLOW,
1903 parent, inode, seq);
1904 if (unlikely(error))
1907 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1909 * If there was a racing rename or mount along our
1910 * path, then we can't be sure that ".." hasn't jumped
1911 * above nd->root (and so userspace should retry or use
1915 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1916 return ERR_PTR(-EAGAIN);
1917 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1918 return ERR_PTR(-EAGAIN);
1924 static const char *walk_component(struct nameidata *nd, int flags)
1926 struct dentry *dentry;
1927 struct inode *inode;
1930 * "." and ".." are special - ".." especially so because it has
1931 * to be able to know about the current root directory and
1932 * parent relationships.
1934 if (unlikely(nd->last_type != LAST_NORM)) {
1935 if (!(flags & WALK_MORE) && nd->depth)
1937 return handle_dots(nd, nd->last_type);
1939 dentry = lookup_fast(nd, &inode, &seq);
1941 return ERR_CAST(dentry);
1942 if (unlikely(!dentry)) {
1943 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1945 return ERR_CAST(dentry);
1947 if (!(flags & WALK_MORE) && nd->depth)
1949 return step_into(nd, flags, dentry, inode, seq);
1953 * We can do the critical dentry name comparison and hashing
1954 * operations one word at a time, but we are limited to:
1956 * - Architectures with fast unaligned word accesses. We could
1957 * do a "get_unaligned()" if this helps and is sufficiently
1960 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1961 * do not trap on the (extremely unlikely) case of a page
1962 * crossing operation.
1964 * - Furthermore, we need an efficient 64-bit compile for the
1965 * 64-bit case in order to generate the "number of bytes in
1966 * the final mask". Again, that could be replaced with a
1967 * efficient population count instruction or similar.
1969 #ifdef CONFIG_DCACHE_WORD_ACCESS
1971 #include <asm/word-at-a-time.h>
1975 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1977 #elif defined(CONFIG_64BIT)
1979 * Register pressure in the mixing function is an issue, particularly
1980 * on 32-bit x86, but almost any function requires one state value and
1981 * one temporary. Instead, use a function designed for two state values
1982 * and no temporaries.
1984 * This function cannot create a collision in only two iterations, so
1985 * we have two iterations to achieve avalanche. In those two iterations,
1986 * we have six layers of mixing, which is enough to spread one bit's
1987 * influence out to 2^6 = 64 state bits.
1989 * Rotate constants are scored by considering either 64 one-bit input
1990 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1991 * probability of that delta causing a change to each of the 128 output
1992 * bits, using a sample of random initial states.
1994 * The Shannon entropy of the computed probabilities is then summed
1995 * to produce a score. Ideally, any input change has a 50% chance of
1996 * toggling any given output bit.
1998 * Mixing scores (in bits) for (12,45):
1999 * Input delta: 1-bit 2-bit
2000 * 1 round: 713.3 42542.6
2001 * 2 rounds: 2753.7 140389.8
2002 * 3 rounds: 5954.1 233458.2
2003 * 4 rounds: 7862.6 256672.2
2004 * Perfect: 8192 258048
2005 * (64*128) (64*63/2 * 128)
2007 #define HASH_MIX(x, y, a) \
2009 y ^= x, x = rol64(x,12),\
2010 x += y, y = rol64(y,45),\
2014 * Fold two longs into one 32-bit hash value. This must be fast, but
2015 * latency isn't quite as critical, as there is a fair bit of additional
2016 * work done before the hash value is used.
2018 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2020 y ^= x * GOLDEN_RATIO_64;
2021 y *= GOLDEN_RATIO_64;
2025 #else /* 32-bit case */
2028 * Mixing scores (in bits) for (7,20):
2029 * Input delta: 1-bit 2-bit
2030 * 1 round: 330.3 9201.6
2031 * 2 rounds: 1246.4 25475.4
2032 * 3 rounds: 1907.1 31295.1
2033 * 4 rounds: 2042.3 31718.6
2034 * Perfect: 2048 31744
2035 * (32*64) (32*31/2 * 64)
2037 #define HASH_MIX(x, y, a) \
2039 y ^= x, x = rol32(x, 7),\
2040 x += y, y = rol32(y,20),\
2043 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2045 /* Use arch-optimized multiply if one exists */
2046 return __hash_32(y ^ __hash_32(x));
2052 * Return the hash of a string of known length. This is carfully
2053 * designed to match hash_name(), which is the more critical function.
2054 * In particular, we must end by hashing a final word containing 0..7
2055 * payload bytes, to match the way that hash_name() iterates until it
2056 * finds the delimiter after the name.
2058 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2060 unsigned long a, x = 0, y = (unsigned long)salt;
2065 a = load_unaligned_zeropad(name);
2066 if (len < sizeof(unsigned long))
2069 name += sizeof(unsigned long);
2070 len -= sizeof(unsigned long);
2072 x ^= a & bytemask_from_count(len);
2074 return fold_hash(x, y);
2076 EXPORT_SYMBOL(full_name_hash);
2078 /* Return the "hash_len" (hash and length) of a null-terminated string */
2079 u64 hashlen_string(const void *salt, const char *name)
2081 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2082 unsigned long adata, mask, len;
2083 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2090 len += sizeof(unsigned long);
2092 a = load_unaligned_zeropad(name+len);
2093 } while (!has_zero(a, &adata, &constants));
2095 adata = prep_zero_mask(a, adata, &constants);
2096 mask = create_zero_mask(adata);
2097 x ^= a & zero_bytemask(mask);
2099 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2101 EXPORT_SYMBOL(hashlen_string);
2104 * Calculate the length and hash of the path component, and
2105 * return the "hash_len" as the result.
2107 static inline u64 hash_name(const void *salt, const char *name)
2109 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2110 unsigned long adata, bdata, mask, len;
2111 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2118 len += sizeof(unsigned long);
2120 a = load_unaligned_zeropad(name+len);
2121 b = a ^ REPEAT_BYTE('/');
2122 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2124 adata = prep_zero_mask(a, adata, &constants);
2125 bdata = prep_zero_mask(b, bdata, &constants);
2126 mask = create_zero_mask(adata | bdata);
2127 x ^= a & zero_bytemask(mask);
2129 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2132 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2134 /* Return the hash of a string of known length */
2135 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2137 unsigned long hash = init_name_hash(salt);
2139 hash = partial_name_hash((unsigned char)*name++, hash);
2140 return end_name_hash(hash);
2142 EXPORT_SYMBOL(full_name_hash);
2144 /* Return the "hash_len" (hash and length) of a null-terminated string */
2145 u64 hashlen_string(const void *salt, const char *name)
2147 unsigned long hash = init_name_hash(salt);
2148 unsigned long len = 0, c;
2150 c = (unsigned char)*name;
2153 hash = partial_name_hash(c, hash);
2154 c = (unsigned char)name[len];
2156 return hashlen_create(end_name_hash(hash), len);
2158 EXPORT_SYMBOL(hashlen_string);
2161 * We know there's a real path component here of at least
2164 static inline u64 hash_name(const void *salt, const char *name)
2166 unsigned long hash = init_name_hash(salt);
2167 unsigned long len = 0, c;
2169 c = (unsigned char)*name;
2172 hash = partial_name_hash(c, hash);
2173 c = (unsigned char)name[len];
2174 } while (c && c != '/');
2175 return hashlen_create(end_name_hash(hash), len);
2182 * This is the basic name resolution function, turning a pathname into
2183 * the final dentry. We expect 'base' to be positive and a directory.
2185 * Returns 0 and nd will have valid dentry and mnt on success.
2186 * Returns error and drops reference to input namei data on failure.
2188 static int link_path_walk(const char *name, struct nameidata *nd)
2190 int depth = 0; // depth <= nd->depth
2193 nd->last_type = LAST_ROOT;
2194 nd->flags |= LOOKUP_PARENT;
2196 return PTR_ERR(name);
2200 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2204 /* At this point we know we have a real path component. */
2206 struct user_namespace *mnt_userns;
2211 mnt_userns = mnt_user_ns(nd->path.mnt);
2212 err = may_lookup(mnt_userns, nd);
2216 hash_len = hash_name(nd->path.dentry, name);
2219 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2221 if (name[1] == '.') {
2223 nd->flags |= LOOKUP_JUMPED;
2229 if (likely(type == LAST_NORM)) {
2230 struct dentry *parent = nd->path.dentry;
2231 nd->flags &= ~LOOKUP_JUMPED;
2232 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2233 struct qstr this = { { .hash_len = hash_len }, .name = name };
2234 err = parent->d_op->d_hash(parent, &this);
2237 hash_len = this.hash_len;
2242 nd->last.hash_len = hash_len;
2243 nd->last.name = name;
2244 nd->last_type = type;
2246 name += hashlen_len(hash_len);
2250 * If it wasn't NUL, we know it was '/'. Skip that
2251 * slash, and continue until no more slashes.
2255 } while (unlikely(*name == '/'));
2256 if (unlikely(!*name)) {
2258 /* pathname or trailing symlink, done */
2260 nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode);
2261 nd->dir_mode = nd->inode->i_mode;
2262 nd->flags &= ~LOOKUP_PARENT;
2265 /* last component of nested symlink */
2266 name = nd->stack[--depth].name;
2267 link = walk_component(nd, 0);
2269 /* not the last component */
2270 link = walk_component(nd, WALK_MORE);
2272 if (unlikely(link)) {
2274 return PTR_ERR(link);
2275 /* a symlink to follow */
2276 nd->stack[depth++].name = name;
2280 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2281 if (nd->flags & LOOKUP_RCU) {
2282 if (!try_to_unlazy(nd))
2290 /* must be paired with terminate_walk() */
2291 static const char *path_init(struct nameidata *nd, unsigned flags)
2294 const char *s = nd->name->name;
2296 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2297 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2298 return ERR_PTR(-EAGAIN);
2301 flags &= ~LOOKUP_RCU;
2302 if (flags & LOOKUP_RCU)
2305 nd->flags = flags | LOOKUP_JUMPED;
2308 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2309 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2312 if (flags & LOOKUP_ROOT) {
2313 struct dentry *root = nd->root.dentry;
2314 struct inode *inode = root->d_inode;
2315 if (*s && unlikely(!d_can_lookup(root)))
2316 return ERR_PTR(-ENOTDIR);
2317 nd->path = nd->root;
2319 if (flags & LOOKUP_RCU) {
2320 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2321 nd->root_seq = nd->seq;
2323 path_get(&nd->path);
2328 nd->root.mnt = NULL;
2330 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2331 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2332 error = nd_jump_root(nd);
2333 if (unlikely(error))
2334 return ERR_PTR(error);
2338 /* Relative pathname -- get the starting-point it is relative to. */
2339 if (nd->dfd == AT_FDCWD) {
2340 if (flags & LOOKUP_RCU) {
2341 struct fs_struct *fs = current->fs;
2345 seq = read_seqcount_begin(&fs->seq);
2347 nd->inode = nd->path.dentry->d_inode;
2348 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2349 } while (read_seqcount_retry(&fs->seq, seq));
2351 get_fs_pwd(current->fs, &nd->path);
2352 nd->inode = nd->path.dentry->d_inode;
2355 /* Caller must check execute permissions on the starting path component */
2356 struct fd f = fdget_raw(nd->dfd);
2357 struct dentry *dentry;
2360 return ERR_PTR(-EBADF);
2362 dentry = f.file->f_path.dentry;
2364 if (*s && unlikely(!d_can_lookup(dentry))) {
2366 return ERR_PTR(-ENOTDIR);
2369 nd->path = f.file->f_path;
2370 if (flags & LOOKUP_RCU) {
2371 nd->inode = nd->path.dentry->d_inode;
2372 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2374 path_get(&nd->path);
2375 nd->inode = nd->path.dentry->d_inode;
2380 /* For scoped-lookups we need to set the root to the dirfd as well. */
2381 if (flags & LOOKUP_IS_SCOPED) {
2382 nd->root = nd->path;
2383 if (flags & LOOKUP_RCU) {
2384 nd->root_seq = nd->seq;
2386 path_get(&nd->root);
2387 nd->flags |= LOOKUP_ROOT_GRABBED;
2393 static inline const char *lookup_last(struct nameidata *nd)
2395 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2396 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2398 return walk_component(nd, WALK_TRAILING);
2401 static int handle_lookup_down(struct nameidata *nd)
2403 if (!(nd->flags & LOOKUP_RCU))
2404 dget(nd->path.dentry);
2405 return PTR_ERR(step_into(nd, WALK_NOFOLLOW,
2406 nd->path.dentry, nd->inode, nd->seq));
2409 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2410 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2412 const char *s = path_init(nd, flags);
2415 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2416 err = handle_lookup_down(nd);
2417 if (unlikely(err < 0))
2421 while (!(err = link_path_walk(s, nd)) &&
2422 (s = lookup_last(nd)) != NULL)
2424 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2425 err = handle_lookup_down(nd);
2426 nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please...
2429 err = complete_walk(nd);
2431 if (!err && nd->flags & LOOKUP_DIRECTORY)
2432 if (!d_can_lookup(nd->path.dentry))
2436 nd->path.mnt = NULL;
2437 nd->path.dentry = NULL;
2443 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2444 struct path *path, struct path *root)
2447 struct nameidata nd;
2449 return PTR_ERR(name);
2450 if (unlikely(root)) {
2452 flags |= LOOKUP_ROOT;
2454 set_nameidata(&nd, dfd, name);
2455 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2456 if (unlikely(retval == -ECHILD))
2457 retval = path_lookupat(&nd, flags, path);
2458 if (unlikely(retval == -ESTALE))
2459 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2461 if (likely(!retval))
2462 audit_inode(name, path->dentry,
2463 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2464 restore_nameidata();
2469 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2470 static int path_parentat(struct nameidata *nd, unsigned flags,
2471 struct path *parent)
2473 const char *s = path_init(nd, flags);
2474 int err = link_path_walk(s, nd);
2476 err = complete_walk(nd);
2479 nd->path.mnt = NULL;
2480 nd->path.dentry = NULL;
2486 static struct filename *filename_parentat(int dfd, struct filename *name,
2487 unsigned int flags, struct path *parent,
2488 struct qstr *last, int *type)
2491 struct nameidata nd;
2495 set_nameidata(&nd, dfd, name);
2496 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2497 if (unlikely(retval == -ECHILD))
2498 retval = path_parentat(&nd, flags, parent);
2499 if (unlikely(retval == -ESTALE))
2500 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2501 if (likely(!retval)) {
2503 *type = nd.last_type;
2504 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2507 name = ERR_PTR(retval);
2509 restore_nameidata();
2513 /* does lookup, returns the object with parent locked */
2514 struct dentry *kern_path_locked(const char *name, struct path *path)
2516 struct filename *filename;
2521 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2523 if (IS_ERR(filename))
2524 return ERR_CAST(filename);
2525 if (unlikely(type != LAST_NORM)) {
2528 return ERR_PTR(-EINVAL);
2530 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2531 d = __lookup_hash(&last, path->dentry, 0);
2533 inode_unlock(path->dentry->d_inode);
2540 int kern_path(const char *name, unsigned int flags, struct path *path)
2542 return filename_lookup(AT_FDCWD, getname_kernel(name),
2545 EXPORT_SYMBOL(kern_path);
2548 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2549 * @dentry: pointer to dentry of the base directory
2550 * @mnt: pointer to vfs mount of the base directory
2551 * @name: pointer to file name
2552 * @flags: lookup flags
2553 * @path: pointer to struct path to fill
2555 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2556 const char *name, unsigned int flags,
2559 struct path root = {.mnt = mnt, .dentry = dentry};
2560 /* the first argument of filename_lookup() is ignored with root */
2561 return filename_lookup(AT_FDCWD, getname_kernel(name),
2562 flags , path, &root);
2564 EXPORT_SYMBOL(vfs_path_lookup);
2566 static int lookup_one_len_common(const char *name, struct dentry *base,
2567 int len, struct qstr *this)
2571 this->hash = full_name_hash(base, name, len);
2575 if (unlikely(name[0] == '.')) {
2576 if (len < 2 || (len == 2 && name[1] == '.'))
2581 unsigned int c = *(const unsigned char *)name++;
2582 if (c == '/' || c == '\0')
2586 * See if the low-level filesystem might want
2587 * to use its own hash..
2589 if (base->d_flags & DCACHE_OP_HASH) {
2590 int err = base->d_op->d_hash(base, this);
2595 return inode_permission(&init_user_ns, base->d_inode, MAY_EXEC);
2599 * try_lookup_one_len - filesystem helper to lookup single pathname component
2600 * @name: pathname component to lookup
2601 * @base: base directory to lookup from
2602 * @len: maximum length @len should be interpreted to
2604 * Look up a dentry by name in the dcache, returning NULL if it does not
2605 * currently exist. The function does not try to create a dentry.
2607 * Note that this routine is purely a helper for filesystem usage and should
2608 * not be called by generic code.
2610 * The caller must hold base->i_mutex.
2612 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2617 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2619 err = lookup_one_len_common(name, base, len, &this);
2621 return ERR_PTR(err);
2623 return lookup_dcache(&this, base, 0);
2625 EXPORT_SYMBOL(try_lookup_one_len);
2628 * lookup_one_len - filesystem helper to lookup single pathname component
2629 * @name: pathname component to lookup
2630 * @base: base directory to lookup from
2631 * @len: maximum length @len should be interpreted to
2633 * Note that this routine is purely a helper for filesystem usage and should
2634 * not be called by generic code.
2636 * The caller must hold base->i_mutex.
2638 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2640 struct dentry *dentry;
2644 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2646 err = lookup_one_len_common(name, base, len, &this);
2648 return ERR_PTR(err);
2650 dentry = lookup_dcache(&this, base, 0);
2651 return dentry ? dentry : __lookup_slow(&this, base, 0);
2653 EXPORT_SYMBOL(lookup_one_len);
2656 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2657 * @name: pathname component to lookup
2658 * @base: base directory to lookup from
2659 * @len: maximum length @len should be interpreted to
2661 * Note that this routine is purely a helper for filesystem usage and should
2662 * not be called by generic code.
2664 * Unlike lookup_one_len, it should be called without the parent
2665 * i_mutex held, and will take the i_mutex itself if necessary.
2667 struct dentry *lookup_one_len_unlocked(const char *name,
2668 struct dentry *base, int len)
2674 err = lookup_one_len_common(name, base, len, &this);
2676 return ERR_PTR(err);
2678 ret = lookup_dcache(&this, base, 0);
2680 ret = lookup_slow(&this, base, 0);
2683 EXPORT_SYMBOL(lookup_one_len_unlocked);
2686 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2687 * on negatives. Returns known positive or ERR_PTR(); that's what
2688 * most of the users want. Note that pinned negative with unlocked parent
2689 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2690 * need to be very careful; pinned positives have ->d_inode stable, so
2691 * this one avoids such problems.
2693 struct dentry *lookup_positive_unlocked(const char *name,
2694 struct dentry *base, int len)
2696 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2697 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2699 ret = ERR_PTR(-ENOENT);
2703 EXPORT_SYMBOL(lookup_positive_unlocked);
2705 #ifdef CONFIG_UNIX98_PTYS
2706 int path_pts(struct path *path)
2708 /* Find something mounted on "pts" in the same directory as
2711 struct dentry *parent = dget_parent(path->dentry);
2712 struct dentry *child;
2713 struct qstr this = QSTR_INIT("pts", 3);
2715 if (unlikely(!path_connected(path->mnt, parent))) {
2720 path->dentry = parent;
2721 child = d_hash_and_lookup(parent, &this);
2725 path->dentry = child;
2732 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2733 struct path *path, int *empty)
2735 return filename_lookup(dfd, getname_flags(name, flags, empty),
2738 EXPORT_SYMBOL(user_path_at_empty);
2740 int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir,
2741 struct inode *inode)
2743 kuid_t fsuid = current_fsuid();
2745 if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid))
2747 if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid))
2749 return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER);
2751 EXPORT_SYMBOL(__check_sticky);
2754 * Check whether we can remove a link victim from directory dir, check
2755 * whether the type of victim is right.
2756 * 1. We can't do it if dir is read-only (done in permission())
2757 * 2. We should have write and exec permissions on dir
2758 * 3. We can't remove anything from append-only dir
2759 * 4. We can't do anything with immutable dir (done in permission())
2760 * 5. If the sticky bit on dir is set we should either
2761 * a. be owner of dir, or
2762 * b. be owner of victim, or
2763 * c. have CAP_FOWNER capability
2764 * 6. If the victim is append-only or immutable we can't do antyhing with
2765 * links pointing to it.
2766 * 7. If the victim has an unknown uid or gid we can't change the inode.
2767 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2768 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2769 * 10. We can't remove a root or mountpoint.
2770 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2771 * nfs_async_unlink().
2773 static int may_delete(struct user_namespace *mnt_userns, struct inode *dir,
2774 struct dentry *victim, bool isdir)
2776 struct inode *inode = d_backing_inode(victim);
2779 if (d_is_negative(victim))
2783 BUG_ON(victim->d_parent->d_inode != dir);
2785 /* Inode writeback is not safe when the uid or gid are invalid. */
2786 if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
2787 !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
2790 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2792 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2798 if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) ||
2799 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2800 HAS_UNMAPPED_ID(mnt_userns, inode))
2803 if (!d_is_dir(victim))
2805 if (IS_ROOT(victim))
2807 } else if (d_is_dir(victim))
2809 if (IS_DEADDIR(dir))
2811 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2816 /* Check whether we can create an object with dentry child in directory
2818 * 1. We can't do it if child already exists (open has special treatment for
2819 * this case, but since we are inlined it's OK)
2820 * 2. We can't do it if dir is read-only (done in permission())
2821 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2822 * 4. We should have write and exec permissions on dir
2823 * 5. We can't do it if dir is immutable (done in permission())
2825 static inline int may_create(struct user_namespace *mnt_userns,
2826 struct inode *dir, struct dentry *child)
2828 struct user_namespace *s_user_ns;
2829 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2832 if (IS_DEADDIR(dir))
2834 s_user_ns = dir->i_sb->s_user_ns;
2835 if (!kuid_has_mapping(s_user_ns, fsuid_into_mnt(mnt_userns)) ||
2836 !kgid_has_mapping(s_user_ns, fsgid_into_mnt(mnt_userns)))
2838 return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2842 * p1 and p2 should be directories on the same fs.
2844 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2849 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2853 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2855 p = d_ancestor(p2, p1);
2857 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2858 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2862 p = d_ancestor(p1, p2);
2864 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2865 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2869 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2870 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2873 EXPORT_SYMBOL(lock_rename);
2875 void unlock_rename(struct dentry *p1, struct dentry *p2)
2877 inode_unlock(p1->d_inode);
2879 inode_unlock(p2->d_inode);
2880 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2883 EXPORT_SYMBOL(unlock_rename);
2886 * vfs_create - create new file
2887 * @mnt_userns: user namespace of the mount the inode was found from
2888 * @dir: inode of @dentry
2889 * @dentry: pointer to dentry of the base directory
2890 * @mode: mode of the new file
2891 * @want_excl: whether the file must not yet exist
2893 * Create a new file.
2895 * If the inode has been found through an idmapped mount the user namespace of
2896 * the vfsmount must be passed through @mnt_userns. This function will then take
2897 * care to map the inode according to @mnt_userns before checking permissions.
2898 * On non-idmapped mounts or if permission checking is to be performed on the
2899 * raw inode simply passs init_user_ns.
2901 int vfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2902 struct dentry *dentry, umode_t mode, bool want_excl)
2904 int error = may_create(mnt_userns, dir, dentry);
2908 if (!dir->i_op->create)
2909 return -EACCES; /* shouldn't it be ENOSYS? */
2912 error = security_inode_create(dir, dentry, mode);
2915 error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl);
2917 fsnotify_create(dir, dentry);
2920 EXPORT_SYMBOL(vfs_create);
2922 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2923 int (*f)(struct dentry *, umode_t, void *),
2926 struct inode *dir = dentry->d_parent->d_inode;
2927 int error = may_create(&init_user_ns, dir, dentry);
2933 error = security_inode_create(dir, dentry, mode);
2936 error = f(dentry, mode, arg);
2938 fsnotify_create(dir, dentry);
2941 EXPORT_SYMBOL(vfs_mkobj);
2943 bool may_open_dev(const struct path *path)
2945 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2946 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2949 static int may_open(struct user_namespace *mnt_userns, const struct path *path,
2950 int acc_mode, int flag)
2952 struct dentry *dentry = path->dentry;
2953 struct inode *inode = dentry->d_inode;
2959 switch (inode->i_mode & S_IFMT) {
2963 if (acc_mode & MAY_WRITE)
2965 if (acc_mode & MAY_EXEC)
2970 if (!may_open_dev(path))
2975 if (acc_mode & MAY_EXEC)
2980 if ((acc_mode & MAY_EXEC) && path_noexec(path))
2985 error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode);
2990 * An append-only file must be opened in append mode for writing.
2992 if (IS_APPEND(inode)) {
2993 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2999 /* O_NOATIME can only be set by the owner or superuser */
3000 if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode))
3006 static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp)
3008 const struct path *path = &filp->f_path;
3009 struct inode *inode = path->dentry->d_inode;
3010 int error = get_write_access(inode);
3014 * Refuse to truncate files with mandatory locks held on them.
3016 error = locks_verify_locked(filp);
3018 error = security_path_truncate(path);
3020 error = do_truncate(mnt_userns, path->dentry, 0,
3021 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3024 put_write_access(inode);
3028 static inline int open_to_namei_flags(int flag)
3030 if ((flag & O_ACCMODE) == 3)
3035 static int may_o_create(struct user_namespace *mnt_userns,
3036 const struct path *dir, struct dentry *dentry,
3039 struct user_namespace *s_user_ns;
3040 int error = security_path_mknod(dir, dentry, mode, 0);
3044 s_user_ns = dir->dentry->d_sb->s_user_ns;
3045 if (!kuid_has_mapping(s_user_ns, fsuid_into_mnt(mnt_userns)) ||
3046 !kgid_has_mapping(s_user_ns, fsgid_into_mnt(mnt_userns)))
3049 error = inode_permission(mnt_userns, dir->dentry->d_inode,
3050 MAY_WRITE | MAY_EXEC);
3054 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3058 * Attempt to atomically look up, create and open a file from a negative
3061 * Returns 0 if successful. The file will have been created and attached to
3062 * @file by the filesystem calling finish_open().
3064 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3065 * be set. The caller will need to perform the open themselves. @path will
3066 * have been updated to point to the new dentry. This may be negative.
3068 * Returns an error code otherwise.
3070 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3072 int open_flag, umode_t mode)
3074 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3075 struct inode *dir = nd->path.dentry->d_inode;
3078 if (nd->flags & LOOKUP_DIRECTORY)
3079 open_flag |= O_DIRECTORY;
3081 file->f_path.dentry = DENTRY_NOT_SET;
3082 file->f_path.mnt = nd->path.mnt;
3083 error = dir->i_op->atomic_open(dir, dentry, file,
3084 open_to_namei_flags(open_flag), mode);
3085 d_lookup_done(dentry);
3087 if (file->f_mode & FMODE_OPENED) {
3088 if (unlikely(dentry != file->f_path.dentry)) {
3090 dentry = dget(file->f_path.dentry);
3092 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3095 if (file->f_path.dentry) {
3097 dentry = file->f_path.dentry;
3099 if (unlikely(d_is_negative(dentry)))
3105 dentry = ERR_PTR(error);
3111 * Look up and maybe create and open the last component.
3113 * Must be called with parent locked (exclusive in O_CREAT case).
3115 * Returns 0 on success, that is, if
3116 * the file was successfully atomically created (if necessary) and opened, or
3117 * the file was not completely opened at this time, though lookups and
3118 * creations were performed.
3119 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3120 * In the latter case dentry returned in @path might be negative if O_CREAT
3121 * hadn't been specified.
3123 * An error code is returned on failure.
3125 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3126 const struct open_flags *op,
3129 struct user_namespace *mnt_userns;
3130 struct dentry *dir = nd->path.dentry;
3131 struct inode *dir_inode = dir->d_inode;
3132 int open_flag = op->open_flag;
3133 struct dentry *dentry;
3134 int error, create_error = 0;
3135 umode_t mode = op->mode;
3136 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3138 if (unlikely(IS_DEADDIR(dir_inode)))
3139 return ERR_PTR(-ENOENT);
3141 file->f_mode &= ~FMODE_CREATED;
3142 dentry = d_lookup(dir, &nd->last);
3145 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3149 if (d_in_lookup(dentry))
3152 error = d_revalidate(dentry, nd->flags);
3153 if (likely(error > 0))
3157 d_invalidate(dentry);
3161 if (dentry->d_inode) {
3162 /* Cached positive dentry: will open in f_op->open */
3167 * Checking write permission is tricky, bacuse we don't know if we are
3168 * going to actually need it: O_CREAT opens should work as long as the
3169 * file exists. But checking existence breaks atomicity. The trick is
3170 * to check access and if not granted clear O_CREAT from the flags.
3172 * Another problem is returing the "right" error value (e.g. for an
3173 * O_EXCL open we want to return EEXIST not EROFS).
3175 if (unlikely(!got_write))
3176 open_flag &= ~O_TRUNC;
3177 mnt_userns = mnt_user_ns(nd->path.mnt);
3178 if (open_flag & O_CREAT) {
3179 if (open_flag & O_EXCL)
3180 open_flag &= ~O_TRUNC;
3181 if (!IS_POSIXACL(dir->d_inode))
3182 mode &= ~current_umask();
3183 if (likely(got_write))
3184 create_error = may_o_create(mnt_userns, &nd->path,
3187 create_error = -EROFS;
3190 open_flag &= ~O_CREAT;
3191 if (dir_inode->i_op->atomic_open) {
3192 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3193 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3194 dentry = ERR_PTR(create_error);
3198 if (d_in_lookup(dentry)) {
3199 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3201 d_lookup_done(dentry);
3202 if (unlikely(res)) {
3204 error = PTR_ERR(res);
3212 /* Negative dentry, just create the file */
3213 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3214 file->f_mode |= FMODE_CREATED;
3215 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3216 if (!dir_inode->i_op->create) {
3221 error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry,
3222 mode, open_flag & O_EXCL);
3226 if (unlikely(create_error) && !dentry->d_inode) {
3227 error = create_error;
3234 return ERR_PTR(error);
3237 static const char *open_last_lookups(struct nameidata *nd,
3238 struct file *file, const struct open_flags *op)
3240 struct dentry *dir = nd->path.dentry;
3241 int open_flag = op->open_flag;
3242 bool got_write = false;
3244 struct inode *inode;
3245 struct dentry *dentry;
3248 nd->flags |= op->intent;
3250 if (nd->last_type != LAST_NORM) {
3253 return handle_dots(nd, nd->last_type);
3256 if (!(open_flag & O_CREAT)) {
3257 if (nd->last.name[nd->last.len])
3258 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3259 /* we _can_ be in RCU mode here */
3260 dentry = lookup_fast(nd, &inode, &seq);
3262 return ERR_CAST(dentry);
3266 BUG_ON(nd->flags & LOOKUP_RCU);
3268 /* create side of things */
3269 if (nd->flags & LOOKUP_RCU) {
3270 if (!try_to_unlazy(nd))
3271 return ERR_PTR(-ECHILD);
3273 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3274 /* trailing slashes? */
3275 if (unlikely(nd->last.name[nd->last.len]))
3276 return ERR_PTR(-EISDIR);
3279 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3280 got_write = !mnt_want_write(nd->path.mnt);
3282 * do _not_ fail yet - we might not need that or fail with
3283 * a different error; let lookup_open() decide; we'll be
3284 * dropping this one anyway.
3287 if (open_flag & O_CREAT)
3288 inode_lock(dir->d_inode);
3290 inode_lock_shared(dir->d_inode);
3291 dentry = lookup_open(nd, file, op, got_write);
3292 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3293 fsnotify_create(dir->d_inode, dentry);
3294 if (open_flag & O_CREAT)
3295 inode_unlock(dir->d_inode);
3297 inode_unlock_shared(dir->d_inode);
3300 mnt_drop_write(nd->path.mnt);
3303 return ERR_CAST(dentry);
3305 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3306 dput(nd->path.dentry);
3307 nd->path.dentry = dentry;
3314 res = step_into(nd, WALK_TRAILING, dentry, inode, seq);
3316 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3321 * Handle the last step of open()
3323 static int do_open(struct nameidata *nd,
3324 struct file *file, const struct open_flags *op)
3326 struct user_namespace *mnt_userns;
3327 int open_flag = op->open_flag;
3332 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3333 error = complete_walk(nd);
3337 if (!(file->f_mode & FMODE_CREATED))
3338 audit_inode(nd->name, nd->path.dentry, 0);
3339 mnt_userns = mnt_user_ns(nd->path.mnt);
3340 if (open_flag & O_CREAT) {
3341 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3343 if (d_is_dir(nd->path.dentry))
3345 error = may_create_in_sticky(mnt_userns, nd,
3346 d_backing_inode(nd->path.dentry));
3347 if (unlikely(error))
3350 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3353 do_truncate = false;
3354 acc_mode = op->acc_mode;
3355 if (file->f_mode & FMODE_CREATED) {
3356 /* Don't check for write permission, don't truncate */
3357 open_flag &= ~O_TRUNC;
3359 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3360 error = mnt_want_write(nd->path.mnt);
3365 error = may_open(mnt_userns, &nd->path, acc_mode, open_flag);
3366 if (!error && !(file->f_mode & FMODE_OPENED))
3367 error = vfs_open(&nd->path, file);
3369 error = ima_file_check(file, op->acc_mode);
3370 if (!error && do_truncate)
3371 error = handle_truncate(mnt_userns, file);
3372 if (unlikely(error > 0)) {
3377 mnt_drop_write(nd->path.mnt);
3382 * vfs_tmpfile - create tmpfile
3383 * @mnt_userns: user namespace of the mount the inode was found from
3384 * @dentry: pointer to dentry of the base directory
3385 * @mode: mode of the new tmpfile
3386 * @open_flags: flags
3388 * Create a temporary file.
3390 * If the inode has been found through an idmapped mount the user namespace of
3391 * the vfsmount must be passed through @mnt_userns. This function will then take
3392 * care to map the inode according to @mnt_userns before checking permissions.
3393 * On non-idmapped mounts or if permission checking is to be performed on the
3394 * raw inode simply passs init_user_ns.
3396 struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns,
3397 struct dentry *dentry, umode_t mode, int open_flag)
3399 struct dentry *child = NULL;
3400 struct inode *dir = dentry->d_inode;
3401 struct inode *inode;
3404 /* we want directory to be writable */
3405 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
3408 error = -EOPNOTSUPP;
3409 if (!dir->i_op->tmpfile)
3412 child = d_alloc(dentry, &slash_name);
3413 if (unlikely(!child))
3415 error = dir->i_op->tmpfile(mnt_userns, dir, child, mode);
3419 inode = child->d_inode;
3420 if (unlikely(!inode))
3422 if (!(open_flag & O_EXCL)) {
3423 spin_lock(&inode->i_lock);
3424 inode->i_state |= I_LINKABLE;
3425 spin_unlock(&inode->i_lock);
3427 ima_post_create_tmpfile(mnt_userns, inode);
3432 return ERR_PTR(error);
3434 EXPORT_SYMBOL(vfs_tmpfile);
3436 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3437 const struct open_flags *op,
3440 struct user_namespace *mnt_userns;
3441 struct dentry *child;
3443 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3444 if (unlikely(error))
3446 error = mnt_want_write(path.mnt);
3447 if (unlikely(error))
3449 mnt_userns = mnt_user_ns(path.mnt);
3450 child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag);
3451 error = PTR_ERR(child);
3455 path.dentry = child;
3456 audit_inode(nd->name, child, 0);
3457 /* Don't check for other permissions, the inode was just created */
3458 error = may_open(mnt_userns, &path, 0, op->open_flag);
3460 error = vfs_open(&path, file);
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);