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/sched/mm.h>
26 #include <linux/fsnotify.h>
27 #include <linux/personality.h>
28 #include <linux/security.h>
29 #include <linux/ima.h>
30 #include <linux/syscalls.h>
31 #include <linux/mount.h>
32 #include <linux/audit.h>
33 #include <linux/capability.h>
34 #include <linux/file.h>
35 #include <linux/fcntl.h>
36 #include <linux/device_cgroup.h>
37 #include <linux/fs_struct.h>
38 #include <linux/posix_acl.h>
39 #include <linux/hash.h>
40 #include <linux/bitops.h>
41 #include <linux/init_task.h>
42 #include <linux/uaccess.h>
47 /* [Feb-1997 T. Schoebel-Theuer]
48 * Fundamental changes in the pathname lookup mechanisms (namei)
49 * were necessary because of omirr. The reason is that omirr needs
50 * to know the _real_ pathname, not the user-supplied one, in case
51 * of symlinks (and also when transname replacements occur).
53 * The new code replaces the old recursive symlink resolution with
54 * an iterative one (in case of non-nested symlink chains). It does
55 * this with calls to <fs>_follow_link().
56 * As a side effect, dir_namei(), _namei() and follow_link() are now
57 * replaced with a single function lookup_dentry() that can handle all
58 * the special cases of the former code.
60 * With the new dcache, the pathname is stored at each inode, at least as
61 * long as the refcount of the inode is positive. As a side effect, the
62 * size of the dcache depends on the inode cache and thus is dynamic.
64 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
65 * resolution to correspond with current state of the code.
67 * Note that the symlink resolution is not *completely* iterative.
68 * There is still a significant amount of tail- and mid- recursion in
69 * the algorithm. Also, note that <fs>_readlink() is not used in
70 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
71 * may return different results than <fs>_follow_link(). Many virtual
72 * filesystems (including /proc) exhibit this behavior.
75 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
76 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
77 * and the name already exists in form of a symlink, try to create the new
78 * name indicated by the symlink. The old code always complained that the
79 * name already exists, due to not following the symlink even if its target
80 * is nonexistent. The new semantics affects also mknod() and link() when
81 * the name is a symlink pointing to a non-existent name.
83 * I don't know which semantics is the right one, since I have no access
84 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
85 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
86 * "old" one. Personally, I think the new semantics is much more logical.
87 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
88 * file does succeed in both HP-UX and SunOs, but not in Solaris
89 * and in the old Linux semantics.
92 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
93 * semantics. See the comments in "open_namei" and "do_link" below.
95 * [10-Sep-98 Alan Modra] Another symlink change.
98 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
99 * inside the path - always follow.
100 * in the last component in creation/removal/renaming - never follow.
101 * if LOOKUP_FOLLOW passed - follow.
102 * if the pathname has trailing slashes - follow.
103 * otherwise - don't follow.
104 * (applied in that order).
106 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
107 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
108 * During the 2.4 we need to fix the userland stuff depending on it -
109 * hopefully we will be able to get rid of that wart in 2.5. So far only
110 * XEmacs seems to be relying on it...
113 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
114 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
115 * any extra contention...
118 /* In order to reduce some races, while at the same time doing additional
119 * checking and hopefully speeding things up, we copy filenames to the
120 * kernel data space before using them..
122 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
123 * PATH_MAX includes the nul terminator --RR.
126 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
129 getname_flags(const char __user *filename, int flags, int *empty)
131 struct filename *result;
135 result = audit_reusename(filename);
139 result = __getname();
140 if (unlikely(!result))
141 return ERR_PTR(-ENOMEM);
144 * First, try to embed the struct filename inside the names_cache
147 kname = (char *)result->iname;
148 result->name = kname;
150 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
151 if (unlikely(len < 0)) {
157 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
158 * separate struct filename so we can dedicate the entire
159 * names_cache allocation for the pathname, and re-do the copy from
162 if (unlikely(len == EMBEDDED_NAME_MAX)) {
163 const size_t size = offsetof(struct filename, iname[1]);
164 kname = (char *)result;
167 * size is chosen that way we to guarantee that
168 * result->iname[0] is within the same object and that
169 * kname can't be equal to result->iname, no matter what.
171 result = kzalloc(size, GFP_KERNEL);
172 if (unlikely(!result)) {
174 return ERR_PTR(-ENOMEM);
176 result->name = kname;
177 len = strncpy_from_user(kname, filename, PATH_MAX);
178 if (unlikely(len < 0)) {
183 if (unlikely(len == PATH_MAX)) {
186 return ERR_PTR(-ENAMETOOLONG);
191 /* The empty path is special. */
192 if (unlikely(!len)) {
195 if (!(flags & LOOKUP_EMPTY)) {
197 return ERR_PTR(-ENOENT);
201 result->uptr = filename;
202 result->aname = NULL;
203 audit_getname(result);
208 getname_uflags(const char __user *filename, int uflags)
210 int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
212 return getname_flags(filename, flags, NULL);
216 getname(const char __user * filename)
218 return getname_flags(filename, 0, NULL);
222 getname_kernel(const char * filename)
224 struct filename *result;
225 int len = strlen(filename) + 1;
227 result = __getname();
228 if (unlikely(!result))
229 return ERR_PTR(-ENOMEM);
231 if (len <= EMBEDDED_NAME_MAX) {
232 result->name = (char *)result->iname;
233 } else if (len <= PATH_MAX) {
234 const size_t size = offsetof(struct filename, iname[1]);
235 struct filename *tmp;
237 tmp = kmalloc(size, GFP_KERNEL);
238 if (unlikely(!tmp)) {
240 return ERR_PTR(-ENOMEM);
242 tmp->name = (char *)result;
246 return ERR_PTR(-ENAMETOOLONG);
248 memcpy((char *)result->name, filename, len);
250 result->aname = NULL;
252 audit_getname(result);
257 void putname(struct filename *name)
262 BUG_ON(name->refcnt <= 0);
264 if (--name->refcnt > 0)
267 if (name->name != name->iname) {
268 __putname(name->name);
275 * check_acl - perform ACL permission checking
276 * @mnt_userns: user namespace of the mount the inode was found from
277 * @inode: inode to check permissions on
278 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
280 * This function performs the ACL permission checking. Since this function
281 * retrieve POSIX acls it needs to know whether it is called from a blocking or
282 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
284 * If the inode has been found through an idmapped mount the user namespace of
285 * the vfsmount must be passed through @mnt_userns. This function will then take
286 * care to map the inode according to @mnt_userns before checking permissions.
287 * On non-idmapped mounts or if permission checking is to be performed on the
288 * raw inode simply passs init_user_ns.
290 static int check_acl(struct user_namespace *mnt_userns,
291 struct inode *inode, int mask)
293 #ifdef CONFIG_FS_POSIX_ACL
294 struct posix_acl *acl;
296 if (mask & MAY_NOT_BLOCK) {
297 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
300 /* no ->get_inode_acl() calls in RCU mode... */
301 if (is_uncached_acl(acl))
303 return posix_acl_permission(mnt_userns, inode, acl, mask);
306 acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
310 int error = posix_acl_permission(mnt_userns, inode, acl, mask);
311 posix_acl_release(acl);
320 * acl_permission_check - perform basic UNIX permission checking
321 * @mnt_userns: user namespace of the mount the inode was found from
322 * @inode: inode to check permissions on
323 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
325 * This function performs the basic UNIX permission checking. Since this
326 * function may retrieve POSIX acls it needs to know whether it is called from a
327 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
329 * If the inode has been found through an idmapped mount the user namespace of
330 * the vfsmount must be passed through @mnt_userns. This function will then take
331 * care to map the inode according to @mnt_userns before checking permissions.
332 * On non-idmapped mounts or if permission checking is to be performed on the
333 * raw inode simply passs init_user_ns.
335 static int acl_permission_check(struct user_namespace *mnt_userns,
336 struct inode *inode, int mask)
338 unsigned int mode = inode->i_mode;
341 /* Are we the owner? If so, ACL's don't matter */
342 vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
343 if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) {
346 return (mask & ~mode) ? -EACCES : 0;
349 /* Do we have ACL's? */
350 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
351 int error = check_acl(mnt_userns, inode, mask);
352 if (error != -EAGAIN)
356 /* Only RWX matters for group/other mode bits */
360 * Are the group permissions different from
361 * the other permissions in the bits we care
362 * about? Need to check group ownership if so.
364 if (mask & (mode ^ (mode >> 3))) {
365 vfsgid_t vfsgid = i_gid_into_vfsgid(mnt_userns, inode);
366 if (vfsgid_in_group_p(vfsgid))
370 /* Bits in 'mode' clear that we require? */
371 return (mask & ~mode) ? -EACCES : 0;
375 * generic_permission - check for access rights on a Posix-like filesystem
376 * @mnt_userns: user namespace of the mount the inode was found from
377 * @inode: inode to check access rights for
378 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
379 * %MAY_NOT_BLOCK ...)
381 * Used to check for read/write/execute permissions on a file.
382 * We use "fsuid" for this, letting us set arbitrary permissions
383 * for filesystem access without changing the "normal" uids which
384 * are used for other things.
386 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
387 * request cannot be satisfied (eg. requires blocking or too much complexity).
388 * It would then be called again in ref-walk mode.
390 * If the inode has been found through an idmapped mount the user namespace of
391 * the vfsmount must be passed through @mnt_userns. This function will then take
392 * care to map the inode according to @mnt_userns before checking permissions.
393 * On non-idmapped mounts or if permission checking is to be performed on the
394 * raw inode simply passs init_user_ns.
396 int generic_permission(struct user_namespace *mnt_userns, struct inode *inode,
402 * Do the basic permission checks.
404 ret = acl_permission_check(mnt_userns, inode, mask);
408 if (S_ISDIR(inode->i_mode)) {
409 /* DACs are overridable for directories */
410 if (!(mask & MAY_WRITE))
411 if (capable_wrt_inode_uidgid(mnt_userns, inode,
412 CAP_DAC_READ_SEARCH))
414 if (capable_wrt_inode_uidgid(mnt_userns, inode,
421 * Searching includes executable on directories, else just read.
423 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
424 if (mask == MAY_READ)
425 if (capable_wrt_inode_uidgid(mnt_userns, inode,
426 CAP_DAC_READ_SEARCH))
429 * Read/write DACs are always overridable.
430 * Executable DACs are overridable when there is
431 * at least one exec bit set.
433 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
434 if (capable_wrt_inode_uidgid(mnt_userns, inode,
440 EXPORT_SYMBOL(generic_permission);
443 * do_inode_permission - UNIX permission checking
444 * @mnt_userns: user namespace of the mount the inode was found from
445 * @inode: inode to check permissions on
446 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
448 * We _really_ want to just do "generic_permission()" without
449 * even looking at the inode->i_op values. So we keep a cache
450 * flag in inode->i_opflags, that says "this has not special
451 * permission function, use the fast case".
453 static inline int do_inode_permission(struct user_namespace *mnt_userns,
454 struct inode *inode, int mask)
456 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
457 if (likely(inode->i_op->permission))
458 return inode->i_op->permission(mnt_userns, inode, mask);
460 /* This gets set once for the inode lifetime */
461 spin_lock(&inode->i_lock);
462 inode->i_opflags |= IOP_FASTPERM;
463 spin_unlock(&inode->i_lock);
465 return generic_permission(mnt_userns, inode, mask);
469 * sb_permission - Check superblock-level permissions
470 * @sb: Superblock of inode to check permission on
471 * @inode: Inode to check permission on
472 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
474 * Separate out file-system wide checks from inode-specific permission checks.
476 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
478 if (unlikely(mask & MAY_WRITE)) {
479 umode_t mode = inode->i_mode;
481 /* Nobody gets write access to a read-only fs. */
482 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
489 * inode_permission - Check for access rights to a given inode
490 * @mnt_userns: User namespace of the mount the inode was found from
491 * @inode: Inode to check permission on
492 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
494 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
495 * this, letting us set arbitrary permissions for filesystem access without
496 * changing the "normal" UIDs which are used for other things.
498 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
500 int inode_permission(struct user_namespace *mnt_userns,
501 struct inode *inode, int mask)
505 retval = sb_permission(inode->i_sb, inode, mask);
509 if (unlikely(mask & MAY_WRITE)) {
511 * Nobody gets write access to an immutable file.
513 if (IS_IMMUTABLE(inode))
517 * Updating mtime will likely cause i_uid and i_gid to be
518 * written back improperly if their true value is unknown
521 if (HAS_UNMAPPED_ID(mnt_userns, inode))
525 retval = do_inode_permission(mnt_userns, inode, mask);
529 retval = devcgroup_inode_permission(inode, mask);
533 return security_inode_permission(inode, mask);
535 EXPORT_SYMBOL(inode_permission);
538 * path_get - get a reference to a path
539 * @path: path to get the reference to
541 * Given a path increment the reference count to the dentry and the vfsmount.
543 void path_get(const struct path *path)
548 EXPORT_SYMBOL(path_get);
551 * path_put - put a reference to a path
552 * @path: path to put the reference to
554 * Given a path decrement the reference count to the dentry and the vfsmount.
556 void path_put(const struct path *path)
561 EXPORT_SYMBOL(path_put);
563 #define EMBEDDED_LEVELS 2
568 struct inode *inode; /* path.dentry.d_inode */
569 unsigned int flags, state;
570 unsigned seq, next_seq, m_seq, r_seq;
573 int total_link_count;
576 struct delayed_call done;
579 } *stack, internal[EMBEDDED_LEVELS];
580 struct filename *name;
581 struct nameidata *saved;
586 } __randomize_layout;
588 #define ND_ROOT_PRESET 1
589 #define ND_ROOT_GRABBED 2
592 static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
594 struct nameidata *old = current->nameidata;
595 p->stack = p->internal;
600 p->path.dentry = NULL;
601 p->total_link_count = old ? old->total_link_count : 0;
603 current->nameidata = p;
606 static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
607 const struct path *root)
609 __set_nameidata(p, dfd, name);
611 if (unlikely(root)) {
612 p->state = ND_ROOT_PRESET;
617 static void restore_nameidata(void)
619 struct nameidata *now = current->nameidata, *old = now->saved;
621 current->nameidata = old;
623 old->total_link_count = now->total_link_count;
624 if (now->stack != now->internal)
628 static bool nd_alloc_stack(struct nameidata *nd)
632 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
633 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
636 memcpy(p, nd->internal, sizeof(nd->internal));
642 * path_connected - Verify that a dentry is below mnt.mnt_root
644 * Rename can sometimes move a file or directory outside of a bind
645 * mount, path_connected allows those cases to be detected.
647 static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
649 struct super_block *sb = mnt->mnt_sb;
651 /* Bind mounts can have disconnected paths */
652 if (mnt->mnt_root == sb->s_root)
655 return is_subdir(dentry, mnt->mnt_root);
658 static void drop_links(struct nameidata *nd)
662 struct saved *last = nd->stack + i;
663 do_delayed_call(&last->done);
664 clear_delayed_call(&last->done);
668 static void leave_rcu(struct nameidata *nd)
670 nd->flags &= ~LOOKUP_RCU;
671 nd->seq = nd->next_seq = 0;
675 static void terminate_walk(struct nameidata *nd)
678 if (!(nd->flags & LOOKUP_RCU)) {
681 for (i = 0; i < nd->depth; i++)
682 path_put(&nd->stack[i].link);
683 if (nd->state & ND_ROOT_GRABBED) {
685 nd->state &= ~ND_ROOT_GRABBED;
692 nd->path.dentry = NULL;
695 /* path_put is needed afterwards regardless of success or failure */
696 static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
698 int res = __legitimize_mnt(path->mnt, mseq);
705 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
709 return !read_seqcount_retry(&path->dentry->d_seq, seq);
712 static inline bool legitimize_path(struct nameidata *nd,
713 struct path *path, unsigned seq)
715 return __legitimize_path(path, seq, nd->m_seq);
718 static bool legitimize_links(struct nameidata *nd)
721 if (unlikely(nd->flags & LOOKUP_CACHED)) {
726 for (i = 0; i < nd->depth; i++) {
727 struct saved *last = nd->stack + i;
728 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
737 static bool legitimize_root(struct nameidata *nd)
739 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
740 if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
742 nd->state |= ND_ROOT_GRABBED;
743 return legitimize_path(nd, &nd->root, nd->root_seq);
747 * Path walking has 2 modes, rcu-walk and ref-walk (see
748 * Documentation/filesystems/path-lookup.txt). In situations when we can't
749 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
750 * normal reference counts on dentries and vfsmounts to transition to ref-walk
751 * mode. Refcounts are grabbed at the last known good point before rcu-walk
752 * got stuck, so ref-walk may continue from there. If this is not successful
753 * (eg. a seqcount has changed), then failure is returned and it's up to caller
754 * to restart the path walk from the beginning in ref-walk mode.
758 * try_to_unlazy - try to switch to ref-walk mode.
759 * @nd: nameidata pathwalk data
760 * Returns: true on success, false on failure
762 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
764 * Must be called from rcu-walk context.
765 * Nothing should touch nameidata between try_to_unlazy() failure and
768 static bool try_to_unlazy(struct nameidata *nd)
770 struct dentry *parent = nd->path.dentry;
772 BUG_ON(!(nd->flags & LOOKUP_RCU));
774 if (unlikely(!legitimize_links(nd)))
776 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
778 if (unlikely(!legitimize_root(nd)))
781 BUG_ON(nd->inode != parent->d_inode);
786 nd->path.dentry = NULL;
793 * try_to_unlazy_next - try to switch to ref-walk mode.
794 * @nd: nameidata pathwalk data
795 * @dentry: next dentry to step into
796 * Returns: true on success, false on failure
798 * Similar to try_to_unlazy(), but here we have the next dentry already
799 * picked by rcu-walk and want to legitimize that in addition to the current
800 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
801 * Nothing should touch nameidata between try_to_unlazy_next() failure and
804 static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry)
807 BUG_ON(!(nd->flags & LOOKUP_RCU));
809 if (unlikely(!legitimize_links(nd)))
811 res = __legitimize_mnt(nd->path.mnt, nd->m_seq);
817 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
821 * We need to move both the parent and the dentry from the RCU domain
822 * to be properly refcounted. And the sequence number in the dentry
823 * validates *both* dentry counters, since we checked the sequence
824 * number of the parent after we got the child sequence number. So we
825 * know the parent must still be valid if the child sequence number is
827 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
829 if (read_seqcount_retry(&dentry->d_seq, nd->next_seq))
832 * Sequence counts matched. Now make sure that the root is
833 * still valid and get it if required.
835 if (unlikely(!legitimize_root(nd)))
843 nd->path.dentry = NULL;
853 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
855 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
856 return dentry->d_op->d_revalidate(dentry, flags);
862 * complete_walk - successful completion of path walk
863 * @nd: pointer nameidata
865 * If we had been in RCU mode, drop out of it and legitimize nd->path.
866 * Revalidate the final result, unless we'd already done that during
867 * the path walk or the filesystem doesn't ask for it. Return 0 on
868 * success, -error on failure. In case of failure caller does not
869 * need to drop nd->path.
871 static int complete_walk(struct nameidata *nd)
873 struct dentry *dentry = nd->path.dentry;
876 if (nd->flags & LOOKUP_RCU) {
878 * We don't want to zero nd->root for scoped-lookups or
879 * externally-managed nd->root.
881 if (!(nd->state & ND_ROOT_PRESET))
882 if (!(nd->flags & LOOKUP_IS_SCOPED))
884 nd->flags &= ~LOOKUP_CACHED;
885 if (!try_to_unlazy(nd))
889 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
891 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
892 * ever step outside the root during lookup" and should already
893 * be guaranteed by the rest of namei, we want to avoid a namei
894 * BUG resulting in userspace being given a path that was not
895 * scoped within the root at some point during the lookup.
897 * So, do a final sanity-check to make sure that in the
898 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
899 * we won't silently return an fd completely outside of the
900 * requested root to userspace.
902 * Userspace could move the path outside the root after this
903 * check, but as discussed elsewhere this is not a concern (the
904 * resolved file was inside the root at some point).
906 if (!path_is_under(&nd->path, &nd->root))
910 if (likely(!(nd->state & ND_JUMPED)))
913 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
916 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
926 static int set_root(struct nameidata *nd)
928 struct fs_struct *fs = current->fs;
931 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
932 * still have to ensure it doesn't happen because it will cause a breakout
935 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
936 return -ENOTRECOVERABLE;
938 if (nd->flags & LOOKUP_RCU) {
942 seq = read_seqcount_begin(&fs->seq);
944 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
945 } while (read_seqcount_retry(&fs->seq, seq));
947 get_fs_root(fs, &nd->root);
948 nd->state |= ND_ROOT_GRABBED;
953 static int nd_jump_root(struct nameidata *nd)
955 if (unlikely(nd->flags & LOOKUP_BENEATH))
957 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
958 /* Absolute path arguments to path_init() are allowed. */
959 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
963 int error = set_root(nd);
967 if (nd->flags & LOOKUP_RCU) {
971 nd->inode = d->d_inode;
972 nd->seq = nd->root_seq;
973 if (read_seqcount_retry(&d->d_seq, nd->seq))
979 nd->inode = nd->path.dentry->d_inode;
981 nd->state |= ND_JUMPED;
986 * Helper to directly jump to a known parsed path from ->get_link,
987 * caller must have taken a reference to path beforehand.
989 int nd_jump_link(const struct path *path)
992 struct nameidata *nd = current->nameidata;
994 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
998 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
999 if (nd->path.mnt != path->mnt)
1002 /* Not currently safe for scoped-lookups. */
1003 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1006 path_put(&nd->path);
1008 nd->inode = nd->path.dentry->d_inode;
1009 nd->state |= ND_JUMPED;
1017 static inline void put_link(struct nameidata *nd)
1019 struct saved *last = nd->stack + --nd->depth;
1020 do_delayed_call(&last->done);
1021 if (!(nd->flags & LOOKUP_RCU))
1022 path_put(&last->link);
1025 static int sysctl_protected_symlinks __read_mostly;
1026 static int sysctl_protected_hardlinks __read_mostly;
1027 static int sysctl_protected_fifos __read_mostly;
1028 static int sysctl_protected_regular __read_mostly;
1030 #ifdef CONFIG_SYSCTL
1031 static struct ctl_table namei_sysctls[] = {
1033 .procname = "protected_symlinks",
1034 .data = &sysctl_protected_symlinks,
1035 .maxlen = sizeof(int),
1037 .proc_handler = proc_dointvec_minmax,
1038 .extra1 = SYSCTL_ZERO,
1039 .extra2 = SYSCTL_ONE,
1042 .procname = "protected_hardlinks",
1043 .data = &sysctl_protected_hardlinks,
1044 .maxlen = sizeof(int),
1046 .proc_handler = proc_dointvec_minmax,
1047 .extra1 = SYSCTL_ZERO,
1048 .extra2 = SYSCTL_ONE,
1051 .procname = "protected_fifos",
1052 .data = &sysctl_protected_fifos,
1053 .maxlen = sizeof(int),
1055 .proc_handler = proc_dointvec_minmax,
1056 .extra1 = SYSCTL_ZERO,
1057 .extra2 = SYSCTL_TWO,
1060 .procname = "protected_regular",
1061 .data = &sysctl_protected_regular,
1062 .maxlen = sizeof(int),
1064 .proc_handler = proc_dointvec_minmax,
1065 .extra1 = SYSCTL_ZERO,
1066 .extra2 = SYSCTL_TWO,
1071 static int __init init_fs_namei_sysctls(void)
1073 register_sysctl_init("fs", namei_sysctls);
1076 fs_initcall(init_fs_namei_sysctls);
1078 #endif /* CONFIG_SYSCTL */
1081 * may_follow_link - Check symlink following for unsafe situations
1082 * @nd: nameidata pathwalk data
1084 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1085 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1086 * in a sticky world-writable directory. This is to protect privileged
1087 * processes from failing races against path names that may change out
1088 * from under them by way of other users creating malicious symlinks.
1089 * It will permit symlinks to be followed only when outside a sticky
1090 * world-writable directory, or when the uid of the symlink and follower
1091 * match, or when the directory owner matches the symlink's owner.
1093 * Returns 0 if following the symlink is allowed, -ve on error.
1095 static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1097 struct user_namespace *mnt_userns;
1100 if (!sysctl_protected_symlinks)
1103 mnt_userns = mnt_user_ns(nd->path.mnt);
1104 vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
1105 /* Allowed if owner and follower match. */
1106 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
1109 /* Allowed if parent directory not sticky and world-writable. */
1110 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1113 /* Allowed if parent directory and link owner match. */
1114 if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid))
1117 if (nd->flags & LOOKUP_RCU)
1120 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1121 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1126 * safe_hardlink_source - Check for safe hardlink conditions
1127 * @mnt_userns: user namespace of the mount the inode was found from
1128 * @inode: the source inode to hardlink from
1130 * Return false if at least one of the following conditions:
1131 * - inode is not a regular file
1133 * - inode is setgid and group-exec
1134 * - access failure for read and write
1136 * Otherwise returns true.
1138 static bool safe_hardlink_source(struct user_namespace *mnt_userns,
1139 struct inode *inode)
1141 umode_t mode = inode->i_mode;
1143 /* Special files should not get pinned to the filesystem. */
1147 /* Setuid files should not get pinned to the filesystem. */
1151 /* Executable setgid files should not get pinned to the filesystem. */
1152 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1155 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1156 if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE))
1163 * may_linkat - Check permissions for creating a hardlink
1164 * @mnt_userns: user namespace of the mount the inode was found from
1165 * @link: the source to hardlink from
1167 * Block hardlink when all of:
1168 * - sysctl_protected_hardlinks enabled
1169 * - fsuid does not match inode
1170 * - hardlink source is unsafe (see safe_hardlink_source() above)
1171 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1173 * If the inode has been found through an idmapped mount the user namespace of
1174 * the vfsmount must be passed through @mnt_userns. This function will then take
1175 * care to map the inode according to @mnt_userns before checking permissions.
1176 * On non-idmapped mounts or if permission checking is to be performed on the
1177 * raw inode simply passs init_user_ns.
1179 * Returns 0 if successful, -ve on error.
1181 int may_linkat(struct user_namespace *mnt_userns, const struct path *link)
1183 struct inode *inode = link->dentry->d_inode;
1185 /* Inode writeback is not safe when the uid or gid are invalid. */
1186 if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) ||
1187 !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode)))
1190 if (!sysctl_protected_hardlinks)
1193 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1194 * otherwise, it must be a safe source.
1196 if (safe_hardlink_source(mnt_userns, inode) ||
1197 inode_owner_or_capable(mnt_userns, inode))
1200 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1205 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1206 * should be allowed, or not, on files that already
1208 * @mnt_userns: user namespace of the mount the inode was found from
1209 * @nd: nameidata pathwalk data
1210 * @inode: the inode of the file to open
1212 * Block an O_CREAT open of a FIFO (or a regular file) when:
1213 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1214 * - the file already exists
1215 * - we are in a sticky directory
1216 * - we don't own the file
1217 * - the owner of the directory doesn't own the file
1218 * - the directory is world writable
1219 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1220 * the directory doesn't have to be world writable: being group writable will
1223 * If the inode has been found through an idmapped mount the user namespace of
1224 * the vfsmount must be passed through @mnt_userns. This function will then take
1225 * care to map the inode according to @mnt_userns before checking permissions.
1226 * On non-idmapped mounts or if permission checking is to be performed on the
1227 * raw inode simply passs init_user_ns.
1229 * Returns 0 if the open is allowed, -ve on error.
1231 static int may_create_in_sticky(struct user_namespace *mnt_userns,
1232 struct nameidata *nd, struct inode *const inode)
1234 umode_t dir_mode = nd->dir_mode;
1235 vfsuid_t dir_vfsuid = nd->dir_vfsuid;
1237 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1238 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1239 likely(!(dir_mode & S_ISVTX)) ||
1240 vfsuid_eq(i_uid_into_vfsuid(mnt_userns, inode), dir_vfsuid) ||
1241 vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), current_fsuid()))
1244 if (likely(dir_mode & 0002) ||
1246 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1247 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1248 const char *operation = S_ISFIFO(inode->i_mode) ?
1249 "sticky_create_fifo" :
1250 "sticky_create_regular";
1251 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1258 * follow_up - Find the mountpoint of path's vfsmount
1260 * Given a path, find the mountpoint of its source file system.
1261 * Replace @path with the path of the mountpoint in the parent mount.
1264 * Return 1 if we went up a level and 0 if we were already at the
1267 int follow_up(struct path *path)
1269 struct mount *mnt = real_mount(path->mnt);
1270 struct mount *parent;
1271 struct dentry *mountpoint;
1273 read_seqlock_excl(&mount_lock);
1274 parent = mnt->mnt_parent;
1275 if (parent == mnt) {
1276 read_sequnlock_excl(&mount_lock);
1279 mntget(&parent->mnt);
1280 mountpoint = dget(mnt->mnt_mountpoint);
1281 read_sequnlock_excl(&mount_lock);
1283 path->dentry = mountpoint;
1285 path->mnt = &parent->mnt;
1288 EXPORT_SYMBOL(follow_up);
1290 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1291 struct path *path, unsigned *seqp)
1293 while (mnt_has_parent(m)) {
1294 struct dentry *mountpoint = m->mnt_mountpoint;
1297 if (unlikely(root->dentry == mountpoint &&
1298 root->mnt == &m->mnt))
1300 if (mountpoint != m->mnt.mnt_root) {
1301 path->mnt = &m->mnt;
1302 path->dentry = mountpoint;
1303 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1310 static bool choose_mountpoint(struct mount *m, const struct path *root,
1317 unsigned seq, mseq = read_seqbegin(&mount_lock);
1319 found = choose_mountpoint_rcu(m, root, path, &seq);
1320 if (unlikely(!found)) {
1321 if (!read_seqretry(&mount_lock, mseq))
1324 if (likely(__legitimize_path(path, seq, mseq)))
1336 * Perform an automount
1337 * - return -EISDIR to tell follow_managed() to stop and return the path we
1340 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1342 struct dentry *dentry = path->dentry;
1344 /* We don't want to mount if someone's just doing a stat -
1345 * unless they're stat'ing a directory and appended a '/' to
1348 * We do, however, want to mount if someone wants to open or
1349 * create a file of any type under the mountpoint, wants to
1350 * traverse through the mountpoint or wants to open the
1351 * mounted directory. Also, autofs may mark negative dentries
1352 * as being automount points. These will need the attentions
1353 * of the daemon to instantiate them before they can be used.
1355 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1356 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1360 if (count && (*count)++ >= MAXSYMLINKS)
1363 return finish_automount(dentry->d_op->d_automount(path), path);
1367 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1368 * dentries are pinned but not locked here, so negative dentry can go
1369 * positive right under us. Use of smp_load_acquire() provides a barrier
1370 * sufficient for ->d_inode and ->d_flags consistency.
1372 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1373 int *count, unsigned lookup_flags)
1375 struct vfsmount *mnt = path->mnt;
1376 bool need_mntput = false;
1379 while (flags & DCACHE_MANAGED_DENTRY) {
1380 /* Allow the filesystem to manage the transit without i_mutex
1382 if (flags & DCACHE_MANAGE_TRANSIT) {
1383 ret = path->dentry->d_op->d_manage(path, false);
1384 flags = smp_load_acquire(&path->dentry->d_flags);
1389 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1390 struct vfsmount *mounted = lookup_mnt(path);
1391 if (mounted) { // ... in our namespace
1395 path->mnt = mounted;
1396 path->dentry = dget(mounted->mnt_root);
1397 // here we know it's positive
1398 flags = path->dentry->d_flags;
1404 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1407 // uncovered automount point
1408 ret = follow_automount(path, count, lookup_flags);
1409 flags = smp_load_acquire(&path->dentry->d_flags);
1416 // possible if you race with several mount --move
1417 if (need_mntput && path->mnt == mnt)
1419 if (!ret && unlikely(d_flags_negative(flags)))
1421 *jumped = need_mntput;
1425 static inline int traverse_mounts(struct path *path, bool *jumped,
1426 int *count, unsigned lookup_flags)
1428 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1431 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1433 if (unlikely(d_flags_negative(flags)))
1437 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1440 int follow_down_one(struct path *path)
1442 struct vfsmount *mounted;
1444 mounted = lookup_mnt(path);
1448 path->mnt = mounted;
1449 path->dentry = dget(mounted->mnt_root);
1454 EXPORT_SYMBOL(follow_down_one);
1457 * Follow down to the covering mount currently visible to userspace. At each
1458 * point, the filesystem owning that dentry may be queried as to whether the
1459 * caller is permitted to proceed or not.
1461 int follow_down(struct path *path)
1463 struct vfsmount *mnt = path->mnt;
1465 int ret = traverse_mounts(path, &jumped, NULL, 0);
1467 if (path->mnt != mnt)
1471 EXPORT_SYMBOL(follow_down);
1474 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1475 * we meet a managed dentry that would need blocking.
1477 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1479 struct dentry *dentry = path->dentry;
1480 unsigned int flags = dentry->d_flags;
1482 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1485 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1490 * Don't forget we might have a non-mountpoint managed dentry
1491 * that wants to block transit.
1493 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1494 int res = dentry->d_op->d_manage(path, true);
1496 return res == -EISDIR;
1497 flags = dentry->d_flags;
1500 if (flags & DCACHE_MOUNTED) {
1501 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1503 path->mnt = &mounted->mnt;
1504 dentry = path->dentry = mounted->mnt.mnt_root;
1505 nd->state |= ND_JUMPED;
1506 nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1507 flags = dentry->d_flags;
1508 // makes sure that non-RCU pathwalk could reach
1510 if (read_seqretry(&mount_lock, nd->m_seq))
1514 if (read_seqretry(&mount_lock, nd->m_seq))
1517 return !(flags & DCACHE_NEED_AUTOMOUNT);
1521 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1527 path->mnt = nd->path.mnt;
1528 path->dentry = dentry;
1529 if (nd->flags & LOOKUP_RCU) {
1530 unsigned int seq = nd->next_seq;
1531 if (likely(__follow_mount_rcu(nd, path)))
1533 // *path and nd->next_seq might've been clobbered
1534 path->mnt = nd->path.mnt;
1535 path->dentry = dentry;
1537 if (!try_to_unlazy_next(nd, dentry))
1540 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1542 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1545 nd->state |= ND_JUMPED;
1547 if (unlikely(ret)) {
1549 if (path->mnt != nd->path.mnt)
1556 * This looks up the name in dcache and possibly revalidates the found dentry.
1557 * NULL is returned if the dentry does not exist in the cache.
1559 static struct dentry *lookup_dcache(const struct qstr *name,
1563 struct dentry *dentry = d_lookup(dir, name);
1565 int error = d_revalidate(dentry, flags);
1566 if (unlikely(error <= 0)) {
1568 d_invalidate(dentry);
1570 return ERR_PTR(error);
1577 * Parent directory has inode locked exclusive. This is one
1578 * and only case when ->lookup() gets called on non in-lookup
1579 * dentries - as the matter of fact, this only gets called
1580 * when directory is guaranteed to have no in-lookup children
1583 static struct dentry *__lookup_hash(const struct qstr *name,
1584 struct dentry *base, unsigned int flags)
1586 struct dentry *dentry = lookup_dcache(name, base, flags);
1588 struct inode *dir = base->d_inode;
1593 /* Don't create child dentry for a dead directory. */
1594 if (unlikely(IS_DEADDIR(dir)))
1595 return ERR_PTR(-ENOENT);
1597 dentry = d_alloc(base, name);
1598 if (unlikely(!dentry))
1599 return ERR_PTR(-ENOMEM);
1601 old = dir->i_op->lookup(dir, dentry, flags);
1602 if (unlikely(old)) {
1609 static struct dentry *lookup_fast(struct nameidata *nd)
1611 struct dentry *dentry, *parent = nd->path.dentry;
1615 * Rename seqlock is not required here because in the off chance
1616 * of a false negative due to a concurrent rename, the caller is
1617 * going to fall back to non-racy lookup.
1619 if (nd->flags & LOOKUP_RCU) {
1620 dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq);
1621 if (unlikely(!dentry)) {
1622 if (!try_to_unlazy(nd))
1623 return ERR_PTR(-ECHILD);
1628 * This sequence count validates that the parent had no
1629 * changes while we did the lookup of the dentry above.
1631 if (read_seqcount_retry(&parent->d_seq, nd->seq))
1632 return ERR_PTR(-ECHILD);
1634 status = d_revalidate(dentry, nd->flags);
1635 if (likely(status > 0))
1637 if (!try_to_unlazy_next(nd, dentry))
1638 return ERR_PTR(-ECHILD);
1639 if (status == -ECHILD)
1640 /* we'd been told to redo it in non-rcu mode */
1641 status = d_revalidate(dentry, nd->flags);
1643 dentry = __d_lookup(parent, &nd->last);
1644 if (unlikely(!dentry))
1646 status = d_revalidate(dentry, nd->flags);
1648 if (unlikely(status <= 0)) {
1650 d_invalidate(dentry);
1652 return ERR_PTR(status);
1657 /* Fast lookup failed, do it the slow way */
1658 static struct dentry *__lookup_slow(const struct qstr *name,
1662 struct dentry *dentry, *old;
1663 struct inode *inode = dir->d_inode;
1664 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1666 /* Don't go there if it's already dead */
1667 if (unlikely(IS_DEADDIR(inode)))
1668 return ERR_PTR(-ENOENT);
1670 dentry = d_alloc_parallel(dir, name, &wq);
1673 if (unlikely(!d_in_lookup(dentry))) {
1674 int error = d_revalidate(dentry, flags);
1675 if (unlikely(error <= 0)) {
1677 d_invalidate(dentry);
1682 dentry = ERR_PTR(error);
1685 old = inode->i_op->lookup(inode, dentry, flags);
1686 d_lookup_done(dentry);
1687 if (unlikely(old)) {
1695 static struct dentry *lookup_slow(const struct qstr *name,
1699 struct inode *inode = dir->d_inode;
1701 inode_lock_shared(inode);
1702 res = __lookup_slow(name, dir, flags);
1703 inode_unlock_shared(inode);
1707 static inline int may_lookup(struct user_namespace *mnt_userns,
1708 struct nameidata *nd)
1710 if (nd->flags & LOOKUP_RCU) {
1711 int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1712 if (err != -ECHILD || !try_to_unlazy(nd))
1715 return inode_permission(mnt_userns, nd->inode, MAY_EXEC);
1718 static int reserve_stack(struct nameidata *nd, struct path *link)
1720 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1723 if (likely(nd->depth != EMBEDDED_LEVELS))
1725 if (likely(nd->stack != nd->internal))
1727 if (likely(nd_alloc_stack(nd)))
1730 if (nd->flags & LOOKUP_RCU) {
1731 // we need to grab link before we do unlazy. And we can't skip
1732 // unlazy even if we fail to grab the link - cleanup needs it
1733 bool grabbed_link = legitimize_path(nd, link, nd->next_seq);
1735 if (!try_to_unlazy(nd) || !grabbed_link)
1738 if (nd_alloc_stack(nd))
1744 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1746 static const char *pick_link(struct nameidata *nd, struct path *link,
1747 struct inode *inode, int flags)
1751 int error = reserve_stack(nd, link);
1753 if (unlikely(error)) {
1754 if (!(nd->flags & LOOKUP_RCU))
1756 return ERR_PTR(error);
1758 last = nd->stack + nd->depth++;
1760 clear_delayed_call(&last->done);
1761 last->seq = nd->next_seq;
1763 if (flags & WALK_TRAILING) {
1764 error = may_follow_link(nd, inode);
1765 if (unlikely(error))
1766 return ERR_PTR(error);
1769 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1770 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1771 return ERR_PTR(-ELOOP);
1773 if (!(nd->flags & LOOKUP_RCU)) {
1774 touch_atime(&last->link);
1776 } else if (atime_needs_update(&last->link, inode)) {
1777 if (!try_to_unlazy(nd))
1778 return ERR_PTR(-ECHILD);
1779 touch_atime(&last->link);
1782 error = security_inode_follow_link(link->dentry, inode,
1783 nd->flags & LOOKUP_RCU);
1784 if (unlikely(error))
1785 return ERR_PTR(error);
1787 res = READ_ONCE(inode->i_link);
1789 const char * (*get)(struct dentry *, struct inode *,
1790 struct delayed_call *);
1791 get = inode->i_op->get_link;
1792 if (nd->flags & LOOKUP_RCU) {
1793 res = get(NULL, inode, &last->done);
1794 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1795 res = get(link->dentry, inode, &last->done);
1797 res = get(link->dentry, inode, &last->done);
1805 error = nd_jump_root(nd);
1806 if (unlikely(error))
1807 return ERR_PTR(error);
1808 while (unlikely(*++res == '/'))
1813 all_done: // pure jump
1819 * Do we need to follow links? We _really_ want to be able
1820 * to do this check without having to look at inode->i_op,
1821 * so we keep a cache of "no, this doesn't need follow_link"
1822 * for the common case.
1824 * NOTE: dentry must be what nd->next_seq had been sampled from.
1826 static const char *step_into(struct nameidata *nd, int flags,
1827 struct dentry *dentry)
1830 struct inode *inode;
1831 int err = handle_mounts(nd, dentry, &path);
1834 return ERR_PTR(err);
1835 inode = path.dentry->d_inode;
1836 if (likely(!d_is_symlink(path.dentry)) ||
1837 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1838 (flags & WALK_NOFOLLOW)) {
1839 /* not a symlink or should not follow */
1840 if (nd->flags & LOOKUP_RCU) {
1841 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1842 return ERR_PTR(-ECHILD);
1843 if (unlikely(!inode))
1844 return ERR_PTR(-ENOENT);
1846 dput(nd->path.dentry);
1847 if (nd->path.mnt != path.mnt)
1848 mntput(nd->path.mnt);
1852 nd->seq = nd->next_seq;
1855 if (nd->flags & LOOKUP_RCU) {
1856 /* make sure that d_is_symlink above matches inode */
1857 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1858 return ERR_PTR(-ECHILD);
1860 if (path.mnt == nd->path.mnt)
1863 return pick_link(nd, &path, inode, flags);
1866 static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1868 struct dentry *parent, *old;
1870 if (path_equal(&nd->path, &nd->root))
1872 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1875 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1876 &nd->root, &path, &seq))
1878 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1879 return ERR_PTR(-ECHILD);
1881 nd->inode = path.dentry->d_inode;
1883 // makes sure that non-RCU pathwalk could reach this state
1884 if (read_seqretry(&mount_lock, nd->m_seq))
1885 return ERR_PTR(-ECHILD);
1886 /* we know that mountpoint was pinned */
1888 old = nd->path.dentry;
1889 parent = old->d_parent;
1890 nd->next_seq = read_seqcount_begin(&parent->d_seq);
1891 // makes sure that non-RCU pathwalk could reach this state
1892 if (read_seqcount_retry(&old->d_seq, nd->seq))
1893 return ERR_PTR(-ECHILD);
1894 if (unlikely(!path_connected(nd->path.mnt, parent)))
1895 return ERR_PTR(-ECHILD);
1898 if (read_seqretry(&mount_lock, nd->m_seq))
1899 return ERR_PTR(-ECHILD);
1900 if (unlikely(nd->flags & LOOKUP_BENEATH))
1901 return ERR_PTR(-ECHILD);
1902 nd->next_seq = nd->seq;
1903 return nd->path.dentry;
1906 static struct dentry *follow_dotdot(struct nameidata *nd)
1908 struct dentry *parent;
1910 if (path_equal(&nd->path, &nd->root))
1912 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1915 if (!choose_mountpoint(real_mount(nd->path.mnt),
1918 path_put(&nd->path);
1920 nd->inode = path.dentry->d_inode;
1921 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1922 return ERR_PTR(-EXDEV);
1924 /* rare case of legitimate dget_parent()... */
1925 parent = dget_parent(nd->path.dentry);
1926 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1928 return ERR_PTR(-ENOENT);
1933 if (unlikely(nd->flags & LOOKUP_BENEATH))
1934 return ERR_PTR(-EXDEV);
1935 return dget(nd->path.dentry);
1938 static const char *handle_dots(struct nameidata *nd, int type)
1940 if (type == LAST_DOTDOT) {
1941 const char *error = NULL;
1942 struct dentry *parent;
1944 if (!nd->root.mnt) {
1945 error = ERR_PTR(set_root(nd));
1949 if (nd->flags & LOOKUP_RCU)
1950 parent = follow_dotdot_rcu(nd);
1952 parent = follow_dotdot(nd);
1954 return ERR_CAST(parent);
1955 error = step_into(nd, WALK_NOFOLLOW, parent);
1956 if (unlikely(error))
1959 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1961 * If there was a racing rename or mount along our
1962 * path, then we can't be sure that ".." hasn't jumped
1963 * above nd->root (and so userspace should retry or use
1967 if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
1968 return ERR_PTR(-EAGAIN);
1969 if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
1970 return ERR_PTR(-EAGAIN);
1976 static const char *walk_component(struct nameidata *nd, int flags)
1978 struct dentry *dentry;
1980 * "." and ".." are special - ".." especially so because it has
1981 * to be able to know about the current root directory and
1982 * parent relationships.
1984 if (unlikely(nd->last_type != LAST_NORM)) {
1985 if (!(flags & WALK_MORE) && nd->depth)
1987 return handle_dots(nd, nd->last_type);
1989 dentry = lookup_fast(nd);
1991 return ERR_CAST(dentry);
1992 if (unlikely(!dentry)) {
1993 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1995 return ERR_CAST(dentry);
1997 if (!(flags & WALK_MORE) && nd->depth)
1999 return step_into(nd, flags, dentry);
2003 * We can do the critical dentry name comparison and hashing
2004 * operations one word at a time, but we are limited to:
2006 * - Architectures with fast unaligned word accesses. We could
2007 * do a "get_unaligned()" if this helps and is sufficiently
2010 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2011 * do not trap on the (extremely unlikely) case of a page
2012 * crossing operation.
2014 * - Furthermore, we need an efficient 64-bit compile for the
2015 * 64-bit case in order to generate the "number of bytes in
2016 * the final mask". Again, that could be replaced with a
2017 * efficient population count instruction or similar.
2019 #ifdef CONFIG_DCACHE_WORD_ACCESS
2021 #include <asm/word-at-a-time.h>
2025 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2027 #elif defined(CONFIG_64BIT)
2029 * Register pressure in the mixing function is an issue, particularly
2030 * on 32-bit x86, but almost any function requires one state value and
2031 * one temporary. Instead, use a function designed for two state values
2032 * and no temporaries.
2034 * This function cannot create a collision in only two iterations, so
2035 * we have two iterations to achieve avalanche. In those two iterations,
2036 * we have six layers of mixing, which is enough to spread one bit's
2037 * influence out to 2^6 = 64 state bits.
2039 * Rotate constants are scored by considering either 64 one-bit input
2040 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2041 * probability of that delta causing a change to each of the 128 output
2042 * bits, using a sample of random initial states.
2044 * The Shannon entropy of the computed probabilities is then summed
2045 * to produce a score. Ideally, any input change has a 50% chance of
2046 * toggling any given output bit.
2048 * Mixing scores (in bits) for (12,45):
2049 * Input delta: 1-bit 2-bit
2050 * 1 round: 713.3 42542.6
2051 * 2 rounds: 2753.7 140389.8
2052 * 3 rounds: 5954.1 233458.2
2053 * 4 rounds: 7862.6 256672.2
2054 * Perfect: 8192 258048
2055 * (64*128) (64*63/2 * 128)
2057 #define HASH_MIX(x, y, a) \
2059 y ^= x, x = rol64(x,12),\
2060 x += y, y = rol64(y,45),\
2064 * Fold two longs into one 32-bit hash value. This must be fast, but
2065 * latency isn't quite as critical, as there is a fair bit of additional
2066 * work done before the hash value is used.
2068 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2070 y ^= x * GOLDEN_RATIO_64;
2071 y *= GOLDEN_RATIO_64;
2075 #else /* 32-bit case */
2078 * Mixing scores (in bits) for (7,20):
2079 * Input delta: 1-bit 2-bit
2080 * 1 round: 330.3 9201.6
2081 * 2 rounds: 1246.4 25475.4
2082 * 3 rounds: 1907.1 31295.1
2083 * 4 rounds: 2042.3 31718.6
2084 * Perfect: 2048 31744
2085 * (32*64) (32*31/2 * 64)
2087 #define HASH_MIX(x, y, a) \
2089 y ^= x, x = rol32(x, 7),\
2090 x += y, y = rol32(y,20),\
2093 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2095 /* Use arch-optimized multiply if one exists */
2096 return __hash_32(y ^ __hash_32(x));
2102 * Return the hash of a string of known length. This is carfully
2103 * designed to match hash_name(), which is the more critical function.
2104 * In particular, we must end by hashing a final word containing 0..7
2105 * payload bytes, to match the way that hash_name() iterates until it
2106 * finds the delimiter after the name.
2108 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2110 unsigned long a, x = 0, y = (unsigned long)salt;
2115 a = load_unaligned_zeropad(name);
2116 if (len < sizeof(unsigned long))
2119 name += sizeof(unsigned long);
2120 len -= sizeof(unsigned long);
2122 x ^= a & bytemask_from_count(len);
2124 return fold_hash(x, y);
2126 EXPORT_SYMBOL(full_name_hash);
2128 /* Return the "hash_len" (hash and length) of a null-terminated string */
2129 u64 hashlen_string(const void *salt, const char *name)
2131 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2132 unsigned long adata, mask, len;
2133 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2140 len += sizeof(unsigned long);
2142 a = load_unaligned_zeropad(name+len);
2143 } while (!has_zero(a, &adata, &constants));
2145 adata = prep_zero_mask(a, adata, &constants);
2146 mask = create_zero_mask(adata);
2147 x ^= a & zero_bytemask(mask);
2149 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2151 EXPORT_SYMBOL(hashlen_string);
2154 * Calculate the length and hash of the path component, and
2155 * return the "hash_len" as the result.
2157 static inline u64 hash_name(const void *salt, const char *name)
2159 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2160 unsigned long adata, bdata, mask, len;
2161 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2168 len += sizeof(unsigned long);
2170 a = load_unaligned_zeropad(name+len);
2171 b = a ^ REPEAT_BYTE('/');
2172 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2174 adata = prep_zero_mask(a, adata, &constants);
2175 bdata = prep_zero_mask(b, bdata, &constants);
2176 mask = create_zero_mask(adata | bdata);
2177 x ^= a & zero_bytemask(mask);
2179 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2182 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2184 /* Return the hash of a string of known length */
2185 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2187 unsigned long hash = init_name_hash(salt);
2189 hash = partial_name_hash((unsigned char)*name++, hash);
2190 return end_name_hash(hash);
2192 EXPORT_SYMBOL(full_name_hash);
2194 /* Return the "hash_len" (hash and length) of a null-terminated string */
2195 u64 hashlen_string(const void *salt, const char *name)
2197 unsigned long hash = init_name_hash(salt);
2198 unsigned long len = 0, c;
2200 c = (unsigned char)*name;
2203 hash = partial_name_hash(c, hash);
2204 c = (unsigned char)name[len];
2206 return hashlen_create(end_name_hash(hash), len);
2208 EXPORT_SYMBOL(hashlen_string);
2211 * We know there's a real path component here of at least
2214 static inline u64 hash_name(const void *salt, const char *name)
2216 unsigned long hash = init_name_hash(salt);
2217 unsigned long len = 0, c;
2219 c = (unsigned char)*name;
2222 hash = partial_name_hash(c, hash);
2223 c = (unsigned char)name[len];
2224 } while (c && c != '/');
2225 return hashlen_create(end_name_hash(hash), len);
2232 * This is the basic name resolution function, turning a pathname into
2233 * the final dentry. We expect 'base' to be positive and a directory.
2235 * Returns 0 and nd will have valid dentry and mnt on success.
2236 * Returns error and drops reference to input namei data on failure.
2238 static int link_path_walk(const char *name, struct nameidata *nd)
2240 int depth = 0; // depth <= nd->depth
2243 nd->last_type = LAST_ROOT;
2244 nd->flags |= LOOKUP_PARENT;
2246 return PTR_ERR(name);
2250 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2254 /* At this point we know we have a real path component. */
2256 struct user_namespace *mnt_userns;
2261 mnt_userns = mnt_user_ns(nd->path.mnt);
2262 err = may_lookup(mnt_userns, nd);
2266 hash_len = hash_name(nd->path.dentry, name);
2269 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2271 if (name[1] == '.') {
2273 nd->state |= ND_JUMPED;
2279 if (likely(type == LAST_NORM)) {
2280 struct dentry *parent = nd->path.dentry;
2281 nd->state &= ~ND_JUMPED;
2282 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2283 struct qstr this = { { .hash_len = hash_len }, .name = name };
2284 err = parent->d_op->d_hash(parent, &this);
2287 hash_len = this.hash_len;
2292 nd->last.hash_len = hash_len;
2293 nd->last.name = name;
2294 nd->last_type = type;
2296 name += hashlen_len(hash_len);
2300 * If it wasn't NUL, we know it was '/'. Skip that
2301 * slash, and continue until no more slashes.
2305 } while (unlikely(*name == '/'));
2306 if (unlikely(!*name)) {
2308 /* pathname or trailing symlink, done */
2310 nd->dir_vfsuid = i_uid_into_vfsuid(mnt_userns, nd->inode);
2311 nd->dir_mode = nd->inode->i_mode;
2312 nd->flags &= ~LOOKUP_PARENT;
2315 /* last component of nested symlink */
2316 name = nd->stack[--depth].name;
2317 link = walk_component(nd, 0);
2319 /* not the last component */
2320 link = walk_component(nd, WALK_MORE);
2322 if (unlikely(link)) {
2324 return PTR_ERR(link);
2325 /* a symlink to follow */
2326 nd->stack[depth++].name = name;
2330 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2331 if (nd->flags & LOOKUP_RCU) {
2332 if (!try_to_unlazy(nd))
2340 /* must be paired with terminate_walk() */
2341 static const char *path_init(struct nameidata *nd, unsigned flags)
2344 const char *s = nd->name->name;
2346 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2347 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2348 return ERR_PTR(-EAGAIN);
2351 flags &= ~LOOKUP_RCU;
2352 if (flags & LOOKUP_RCU)
2355 nd->seq = nd->next_seq = 0;
2358 nd->state |= ND_JUMPED;
2360 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2361 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2364 if (nd->state & ND_ROOT_PRESET) {
2365 struct dentry *root = nd->root.dentry;
2366 struct inode *inode = root->d_inode;
2367 if (*s && unlikely(!d_can_lookup(root)))
2368 return ERR_PTR(-ENOTDIR);
2369 nd->path = nd->root;
2371 if (flags & LOOKUP_RCU) {
2372 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2373 nd->root_seq = nd->seq;
2375 path_get(&nd->path);
2380 nd->root.mnt = NULL;
2382 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2383 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2384 error = nd_jump_root(nd);
2385 if (unlikely(error))
2386 return ERR_PTR(error);
2390 /* Relative pathname -- get the starting-point it is relative to. */
2391 if (nd->dfd == AT_FDCWD) {
2392 if (flags & LOOKUP_RCU) {
2393 struct fs_struct *fs = current->fs;
2397 seq = read_seqcount_begin(&fs->seq);
2399 nd->inode = nd->path.dentry->d_inode;
2400 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2401 } while (read_seqcount_retry(&fs->seq, seq));
2403 get_fs_pwd(current->fs, &nd->path);
2404 nd->inode = nd->path.dentry->d_inode;
2407 /* Caller must check execute permissions on the starting path component */
2408 struct fd f = fdget_raw(nd->dfd);
2409 struct dentry *dentry;
2412 return ERR_PTR(-EBADF);
2414 dentry = f.file->f_path.dentry;
2416 if (*s && unlikely(!d_can_lookup(dentry))) {
2418 return ERR_PTR(-ENOTDIR);
2421 nd->path = f.file->f_path;
2422 if (flags & LOOKUP_RCU) {
2423 nd->inode = nd->path.dentry->d_inode;
2424 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2426 path_get(&nd->path);
2427 nd->inode = nd->path.dentry->d_inode;
2432 /* For scoped-lookups we need to set the root to the dirfd as well. */
2433 if (flags & LOOKUP_IS_SCOPED) {
2434 nd->root = nd->path;
2435 if (flags & LOOKUP_RCU) {
2436 nd->root_seq = nd->seq;
2438 path_get(&nd->root);
2439 nd->state |= ND_ROOT_GRABBED;
2445 static inline const char *lookup_last(struct nameidata *nd)
2447 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2448 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2450 return walk_component(nd, WALK_TRAILING);
2453 static int handle_lookup_down(struct nameidata *nd)
2455 if (!(nd->flags & LOOKUP_RCU))
2456 dget(nd->path.dentry);
2457 nd->next_seq = nd->seq;
2458 return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry));
2461 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2462 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2464 const char *s = path_init(nd, flags);
2467 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2468 err = handle_lookup_down(nd);
2469 if (unlikely(err < 0))
2473 while (!(err = link_path_walk(s, nd)) &&
2474 (s = lookup_last(nd)) != NULL)
2476 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2477 err = handle_lookup_down(nd);
2478 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2481 err = complete_walk(nd);
2483 if (!err && nd->flags & LOOKUP_DIRECTORY)
2484 if (!d_can_lookup(nd->path.dentry))
2488 nd->path.mnt = NULL;
2489 nd->path.dentry = NULL;
2495 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2496 struct path *path, struct path *root)
2499 struct nameidata nd;
2501 return PTR_ERR(name);
2502 set_nameidata(&nd, dfd, name, root);
2503 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2504 if (unlikely(retval == -ECHILD))
2505 retval = path_lookupat(&nd, flags, path);
2506 if (unlikely(retval == -ESTALE))
2507 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2509 if (likely(!retval))
2510 audit_inode(name, path->dentry,
2511 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2512 restore_nameidata();
2516 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2517 static int path_parentat(struct nameidata *nd, unsigned flags,
2518 struct path *parent)
2520 const char *s = path_init(nd, flags);
2521 int err = link_path_walk(s, nd);
2523 err = complete_walk(nd);
2526 nd->path.mnt = NULL;
2527 nd->path.dentry = NULL;
2533 /* Note: this does not consume "name" */
2534 static int filename_parentat(int dfd, struct filename *name,
2535 unsigned int flags, struct path *parent,
2536 struct qstr *last, int *type)
2539 struct nameidata nd;
2542 return PTR_ERR(name);
2543 set_nameidata(&nd, dfd, name, NULL);
2544 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2545 if (unlikely(retval == -ECHILD))
2546 retval = path_parentat(&nd, flags, parent);
2547 if (unlikely(retval == -ESTALE))
2548 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2549 if (likely(!retval)) {
2551 *type = nd.last_type;
2552 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2554 restore_nameidata();
2558 /* does lookup, returns the object with parent locked */
2559 static struct dentry *__kern_path_locked(struct filename *name, struct path *path)
2565 error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type);
2567 return ERR_PTR(error);
2568 if (unlikely(type != LAST_NORM)) {
2570 return ERR_PTR(-EINVAL);
2572 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2573 d = __lookup_hash(&last, path->dentry, 0);
2575 inode_unlock(path->dentry->d_inode);
2581 struct dentry *kern_path_locked(const char *name, struct path *path)
2583 struct filename *filename = getname_kernel(name);
2584 struct dentry *res = __kern_path_locked(filename, path);
2590 int kern_path(const char *name, unsigned int flags, struct path *path)
2592 struct filename *filename = getname_kernel(name);
2593 int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2599 EXPORT_SYMBOL(kern_path);
2602 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2603 * @dentry: pointer to dentry of the base directory
2604 * @mnt: pointer to vfs mount of the base directory
2605 * @name: pointer to file name
2606 * @flags: lookup flags
2607 * @path: pointer to struct path to fill
2609 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2610 const char *name, unsigned int flags,
2613 struct filename *filename;
2614 struct path root = {.mnt = mnt, .dentry = dentry};
2617 filename = getname_kernel(name);
2618 /* the first argument of filename_lookup() is ignored with root */
2619 ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2623 EXPORT_SYMBOL(vfs_path_lookup);
2625 static int lookup_one_common(struct user_namespace *mnt_userns,
2626 const char *name, struct dentry *base, int len,
2631 this->hash = full_name_hash(base, name, len);
2635 if (unlikely(name[0] == '.')) {
2636 if (len < 2 || (len == 2 && name[1] == '.'))
2641 unsigned int c = *(const unsigned char *)name++;
2642 if (c == '/' || c == '\0')
2646 * See if the low-level filesystem might want
2647 * to use its own hash..
2649 if (base->d_flags & DCACHE_OP_HASH) {
2650 int err = base->d_op->d_hash(base, this);
2655 return inode_permission(mnt_userns, base->d_inode, MAY_EXEC);
2659 * try_lookup_one_len - filesystem helper to lookup single pathname component
2660 * @name: pathname component to lookup
2661 * @base: base directory to lookup from
2662 * @len: maximum length @len should be interpreted to
2664 * Look up a dentry by name in the dcache, returning NULL if it does not
2665 * currently exist. The function does not try to create a dentry.
2667 * Note that this routine is purely a helper for filesystem usage and should
2668 * not be called by generic code.
2670 * The caller must hold base->i_mutex.
2672 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2677 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2679 err = lookup_one_common(&init_user_ns, name, base, len, &this);
2681 return ERR_PTR(err);
2683 return lookup_dcache(&this, base, 0);
2685 EXPORT_SYMBOL(try_lookup_one_len);
2688 * lookup_one_len - filesystem helper to lookup single pathname component
2689 * @name: pathname component to lookup
2690 * @base: base directory to lookup from
2691 * @len: maximum length @len should be interpreted to
2693 * Note that this routine is purely a helper for filesystem usage and should
2694 * not be called by generic code.
2696 * The caller must hold base->i_mutex.
2698 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2700 struct dentry *dentry;
2704 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2706 err = lookup_one_common(&init_user_ns, name, base, len, &this);
2708 return ERR_PTR(err);
2710 dentry = lookup_dcache(&this, base, 0);
2711 return dentry ? dentry : __lookup_slow(&this, base, 0);
2713 EXPORT_SYMBOL(lookup_one_len);
2716 * lookup_one - filesystem helper to lookup single pathname component
2717 * @mnt_userns: user namespace of the mount the lookup is performed from
2718 * @name: pathname component to lookup
2719 * @base: base directory to lookup from
2720 * @len: maximum length @len should be interpreted to
2722 * Note that this routine is purely a helper for filesystem usage and should
2723 * not be called by generic code.
2725 * The caller must hold base->i_mutex.
2727 struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name,
2728 struct dentry *base, int len)
2730 struct dentry *dentry;
2734 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2736 err = lookup_one_common(mnt_userns, name, base, len, &this);
2738 return ERR_PTR(err);
2740 dentry = lookup_dcache(&this, base, 0);
2741 return dentry ? dentry : __lookup_slow(&this, base, 0);
2743 EXPORT_SYMBOL(lookup_one);
2746 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2747 * @mnt_userns: idmapping of the mount the lookup is performed from
2748 * @name: pathname component to lookup
2749 * @base: base directory to lookup from
2750 * @len: maximum length @len should be interpreted to
2752 * Note that this routine is purely a helper for filesystem usage and should
2753 * not be called by generic code.
2755 * Unlike lookup_one_len, it should be called without the parent
2756 * i_mutex held, and will take the i_mutex itself if necessary.
2758 struct dentry *lookup_one_unlocked(struct user_namespace *mnt_userns,
2759 const char *name, struct dentry *base,
2766 err = lookup_one_common(mnt_userns, name, base, len, &this);
2768 return ERR_PTR(err);
2770 ret = lookup_dcache(&this, base, 0);
2772 ret = lookup_slow(&this, base, 0);
2775 EXPORT_SYMBOL(lookup_one_unlocked);
2778 * lookup_one_positive_unlocked - filesystem helper to lookup single
2779 * pathname component
2780 * @mnt_userns: idmapping of the mount the lookup is performed from
2781 * @name: pathname component to lookup
2782 * @base: base directory to lookup from
2783 * @len: maximum length @len should be interpreted to
2785 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2786 * known positive or ERR_PTR(). This is what most of the users want.
2788 * Note that pinned negative with unlocked parent _can_ become positive at any
2789 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2790 * positives have >d_inode stable, so this one avoids such problems.
2792 * Note that this routine is purely a helper for filesystem usage and should
2793 * not be called by generic code.
2795 * The helper should be called without i_mutex held.
2797 struct dentry *lookup_one_positive_unlocked(struct user_namespace *mnt_userns,
2799 struct dentry *base, int len)
2801 struct dentry *ret = lookup_one_unlocked(mnt_userns, name, base, len);
2803 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2805 ret = ERR_PTR(-ENOENT);
2809 EXPORT_SYMBOL(lookup_one_positive_unlocked);
2812 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2813 * @name: pathname component to lookup
2814 * @base: base directory to lookup from
2815 * @len: maximum length @len should be interpreted to
2817 * Note that this routine is purely a helper for filesystem usage and should
2818 * not be called by generic code.
2820 * Unlike lookup_one_len, it should be called without the parent
2821 * i_mutex held, and will take the i_mutex itself if necessary.
2823 struct dentry *lookup_one_len_unlocked(const char *name,
2824 struct dentry *base, int len)
2826 return lookup_one_unlocked(&init_user_ns, name, base, len);
2828 EXPORT_SYMBOL(lookup_one_len_unlocked);
2831 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2832 * on negatives. Returns known positive or ERR_PTR(); that's what
2833 * most of the users want. Note that pinned negative with unlocked parent
2834 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2835 * need to be very careful; pinned positives have ->d_inode stable, so
2836 * this one avoids such problems.
2838 struct dentry *lookup_positive_unlocked(const char *name,
2839 struct dentry *base, int len)
2841 return lookup_one_positive_unlocked(&init_user_ns, name, base, len);
2843 EXPORT_SYMBOL(lookup_positive_unlocked);
2845 #ifdef CONFIG_UNIX98_PTYS
2846 int path_pts(struct path *path)
2848 /* Find something mounted on "pts" in the same directory as
2851 struct dentry *parent = dget_parent(path->dentry);
2852 struct dentry *child;
2853 struct qstr this = QSTR_INIT("pts", 3);
2855 if (unlikely(!path_connected(path->mnt, parent))) {
2860 path->dentry = parent;
2861 child = d_hash_and_lookup(parent, &this);
2865 path->dentry = child;
2872 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2873 struct path *path, int *empty)
2875 struct filename *filename = getname_flags(name, flags, empty);
2876 int ret = filename_lookup(dfd, filename, flags, path, NULL);
2881 EXPORT_SYMBOL(user_path_at_empty);
2883 int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir,
2884 struct inode *inode)
2886 kuid_t fsuid = current_fsuid();
2888 if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), fsuid))
2890 if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, dir), fsuid))
2892 return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER);
2894 EXPORT_SYMBOL(__check_sticky);
2897 * Check whether we can remove a link victim from directory dir, check
2898 * whether the type of victim is right.
2899 * 1. We can't do it if dir is read-only (done in permission())
2900 * 2. We should have write and exec permissions on dir
2901 * 3. We can't remove anything from append-only dir
2902 * 4. We can't do anything with immutable dir (done in permission())
2903 * 5. If the sticky bit on dir is set we should either
2904 * a. be owner of dir, or
2905 * b. be owner of victim, or
2906 * c. have CAP_FOWNER capability
2907 * 6. If the victim is append-only or immutable we can't do antyhing with
2908 * links pointing to it.
2909 * 7. If the victim has an unknown uid or gid we can't change the inode.
2910 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2911 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2912 * 10. We can't remove a root or mountpoint.
2913 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2914 * nfs_async_unlink().
2916 static int may_delete(struct user_namespace *mnt_userns, struct inode *dir,
2917 struct dentry *victim, bool isdir)
2919 struct inode *inode = d_backing_inode(victim);
2922 if (d_is_negative(victim))
2926 BUG_ON(victim->d_parent->d_inode != dir);
2928 /* Inode writeback is not safe when the uid or gid are invalid. */
2929 if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) ||
2930 !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode)))
2933 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2935 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2941 if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) ||
2942 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2943 HAS_UNMAPPED_ID(mnt_userns, inode))
2946 if (!d_is_dir(victim))
2948 if (IS_ROOT(victim))
2950 } else if (d_is_dir(victim))
2952 if (IS_DEADDIR(dir))
2954 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2959 /* Check whether we can create an object with dentry child in directory
2961 * 1. We can't do it if child already exists (open has special treatment for
2962 * this case, but since we are inlined it's OK)
2963 * 2. We can't do it if dir is read-only (done in permission())
2964 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2965 * 4. We should have write and exec permissions on dir
2966 * 5. We can't do it if dir is immutable (done in permission())
2968 static inline int may_create(struct user_namespace *mnt_userns,
2969 struct inode *dir, struct dentry *child)
2971 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2974 if (IS_DEADDIR(dir))
2976 if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
2979 return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2983 * p1 and p2 should be directories on the same fs.
2985 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2990 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2994 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2996 p = d_ancestor(p2, p1);
2998 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2999 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
3003 p = d_ancestor(p1, p2);
3005 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3006 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
3010 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3011 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3014 EXPORT_SYMBOL(lock_rename);
3016 void unlock_rename(struct dentry *p1, struct dentry *p2)
3018 inode_unlock(p1->d_inode);
3020 inode_unlock(p2->d_inode);
3021 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3024 EXPORT_SYMBOL(unlock_rename);
3027 * mode_strip_umask - handle vfs umask stripping
3028 * @dir: parent directory of the new inode
3029 * @mode: mode of the new inode to be created in @dir
3031 * Umask stripping depends on whether or not the filesystem supports POSIX
3032 * ACLs. If the filesystem doesn't support it umask stripping is done directly
3033 * in here. If the filesystem does support POSIX ACLs umask stripping is
3034 * deferred until the filesystem calls posix_acl_create().
3038 static inline umode_t mode_strip_umask(const struct inode *dir, umode_t mode)
3040 if (!IS_POSIXACL(dir))
3041 mode &= ~current_umask();
3046 * vfs_prepare_mode - prepare the mode to be used for a new inode
3047 * @mnt_userns: user namespace of the mount the inode was found from
3048 * @dir: parent directory of the new inode
3049 * @mode: mode of the new inode
3050 * @mask_perms: allowed permission by the vfs
3051 * @type: type of file to be created
3053 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3054 * object to be created.
3056 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3057 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3058 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3059 * POSIX ACL supporting filesystems.
3061 * Note that it's currently valid for @type to be 0 if a directory is created.
3062 * Filesystems raise that flag individually and we need to check whether each
3063 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3066 * Returns: mode to be passed to the filesystem
3068 static inline umode_t vfs_prepare_mode(struct user_namespace *mnt_userns,
3069 const struct inode *dir, umode_t mode,
3070 umode_t mask_perms, umode_t type)
3072 mode = mode_strip_sgid(mnt_userns, dir, mode);
3073 mode = mode_strip_umask(dir, mode);
3076 * Apply the vfs mandated allowed permission mask and set the type of
3077 * file to be created before we call into the filesystem.
3079 mode &= (mask_perms & ~S_IFMT);
3080 mode |= (type & S_IFMT);
3086 * vfs_create - create new file
3087 * @idmap: idmap of the mount the inode was found from
3088 * @dir: inode of @dentry
3089 * @dentry: pointer to dentry of the base directory
3090 * @mode: mode of the new file
3091 * @want_excl: whether the file must not yet exist
3093 * Create a new file.
3095 * If the inode has been found through an idmapped mount the idmap of
3096 * the vfsmount must be passed through @idmap. This function will then take
3097 * care to map the inode according to @idmap before checking permissions.
3098 * On non-idmapped mounts or if permission checking is to be performed on the
3099 * raw inode simply passs @nop_mnt_idmap.
3101 int vfs_create(struct mnt_idmap *idmap, struct inode *dir,
3102 struct dentry *dentry, umode_t mode, bool want_excl)
3104 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
3107 error = may_create(mnt_userns, dir, dentry);
3111 if (!dir->i_op->create)
3112 return -EACCES; /* shouldn't it be ENOSYS? */
3114 mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IALLUGO, S_IFREG);
3115 error = security_inode_create(dir, dentry, mode);
3118 error = dir->i_op->create(idmap, dir, dentry, mode, want_excl);
3120 fsnotify_create(dir, dentry);
3123 EXPORT_SYMBOL(vfs_create);
3125 int vfs_mkobj(struct dentry *dentry, umode_t mode,
3126 int (*f)(struct dentry *, umode_t, void *),
3129 struct inode *dir = dentry->d_parent->d_inode;
3130 int error = may_create(&init_user_ns, dir, dentry);
3136 error = security_inode_create(dir, dentry, mode);
3139 error = f(dentry, mode, arg);
3141 fsnotify_create(dir, dentry);
3144 EXPORT_SYMBOL(vfs_mkobj);
3146 bool may_open_dev(const struct path *path)
3148 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3149 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3152 static int may_open(struct user_namespace *mnt_userns, const struct path *path,
3153 int acc_mode, int flag)
3155 struct dentry *dentry = path->dentry;
3156 struct inode *inode = dentry->d_inode;
3162 switch (inode->i_mode & S_IFMT) {
3166 if (acc_mode & MAY_WRITE)
3168 if (acc_mode & MAY_EXEC)
3173 if (!may_open_dev(path))
3178 if (acc_mode & MAY_EXEC)
3183 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3188 error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode);
3193 * An append-only file must be opened in append mode for writing.
3195 if (IS_APPEND(inode)) {
3196 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3202 /* O_NOATIME can only be set by the owner or superuser */
3203 if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode))
3209 static int handle_truncate(struct mnt_idmap *idmap, struct file *filp)
3211 const struct path *path = &filp->f_path;
3212 struct inode *inode = path->dentry->d_inode;
3213 int error = get_write_access(inode);
3217 error = security_file_truncate(filp);
3219 error = do_truncate(idmap, path->dentry, 0,
3220 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3223 put_write_access(inode);
3227 static inline int open_to_namei_flags(int flag)
3229 if ((flag & O_ACCMODE) == 3)
3234 static int may_o_create(struct user_namespace *mnt_userns,
3235 const struct path *dir, struct dentry *dentry,
3238 int error = security_path_mknod(dir, dentry, mode, 0);
3242 if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns))
3245 error = inode_permission(mnt_userns, dir->dentry->d_inode,
3246 MAY_WRITE | MAY_EXEC);
3250 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3254 * Attempt to atomically look up, create and open a file from a negative
3257 * Returns 0 if successful. The file will have been created and attached to
3258 * @file by the filesystem calling finish_open().
3260 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3261 * be set. The caller will need to perform the open themselves. @path will
3262 * have been updated to point to the new dentry. This may be negative.
3264 * Returns an error code otherwise.
3266 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3268 int open_flag, umode_t mode)
3270 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3271 struct inode *dir = nd->path.dentry->d_inode;
3274 if (nd->flags & LOOKUP_DIRECTORY)
3275 open_flag |= O_DIRECTORY;
3277 file->f_path.dentry = DENTRY_NOT_SET;
3278 file->f_path.mnt = nd->path.mnt;
3279 error = dir->i_op->atomic_open(dir, dentry, file,
3280 open_to_namei_flags(open_flag), mode);
3281 d_lookup_done(dentry);
3283 if (file->f_mode & FMODE_OPENED) {
3284 if (unlikely(dentry != file->f_path.dentry)) {
3286 dentry = dget(file->f_path.dentry);
3288 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3291 if (file->f_path.dentry) {
3293 dentry = file->f_path.dentry;
3295 if (unlikely(d_is_negative(dentry)))
3301 dentry = ERR_PTR(error);
3307 * Look up and maybe create and open the last component.
3309 * Must be called with parent locked (exclusive in O_CREAT case).
3311 * Returns 0 on success, that is, if
3312 * the file was successfully atomically created (if necessary) and opened, or
3313 * the file was not completely opened at this time, though lookups and
3314 * creations were performed.
3315 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3316 * In the latter case dentry returned in @path might be negative if O_CREAT
3317 * hadn't been specified.
3319 * An error code is returned on failure.
3321 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3322 const struct open_flags *op,
3325 struct mnt_idmap *idmap;
3326 struct user_namespace *mnt_userns;
3327 struct dentry *dir = nd->path.dentry;
3328 struct inode *dir_inode = dir->d_inode;
3329 int open_flag = op->open_flag;
3330 struct dentry *dentry;
3331 int error, create_error = 0;
3332 umode_t mode = op->mode;
3333 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3335 if (unlikely(IS_DEADDIR(dir_inode)))
3336 return ERR_PTR(-ENOENT);
3338 file->f_mode &= ~FMODE_CREATED;
3339 dentry = d_lookup(dir, &nd->last);
3342 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3346 if (d_in_lookup(dentry))
3349 error = d_revalidate(dentry, nd->flags);
3350 if (likely(error > 0))
3354 d_invalidate(dentry);
3358 if (dentry->d_inode) {
3359 /* Cached positive dentry: will open in f_op->open */
3364 * Checking write permission is tricky, bacuse we don't know if we are
3365 * going to actually need it: O_CREAT opens should work as long as the
3366 * file exists. But checking existence breaks atomicity. The trick is
3367 * to check access and if not granted clear O_CREAT from the flags.
3369 * Another problem is returing the "right" error value (e.g. for an
3370 * O_EXCL open we want to return EEXIST not EROFS).
3372 if (unlikely(!got_write))
3373 open_flag &= ~O_TRUNC;
3374 idmap = mnt_idmap(nd->path.mnt);
3375 mnt_userns = mnt_idmap_owner(idmap);
3376 if (open_flag & O_CREAT) {
3377 if (open_flag & O_EXCL)
3378 open_flag &= ~O_TRUNC;
3379 mode = vfs_prepare_mode(mnt_userns, dir->d_inode, mode, mode, mode);
3380 if (likely(got_write))
3381 create_error = may_o_create(mnt_userns, &nd->path,
3384 create_error = -EROFS;
3387 open_flag &= ~O_CREAT;
3388 if (dir_inode->i_op->atomic_open) {
3389 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3390 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3391 dentry = ERR_PTR(create_error);
3395 if (d_in_lookup(dentry)) {
3396 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3398 d_lookup_done(dentry);
3399 if (unlikely(res)) {
3401 error = PTR_ERR(res);
3409 /* Negative dentry, just create the file */
3410 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3411 file->f_mode |= FMODE_CREATED;
3412 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3413 if (!dir_inode->i_op->create) {
3418 error = dir_inode->i_op->create(idmap, dir_inode, dentry,
3419 mode, open_flag & O_EXCL);
3423 if (unlikely(create_error) && !dentry->d_inode) {
3424 error = create_error;
3431 return ERR_PTR(error);
3434 static const char *open_last_lookups(struct nameidata *nd,
3435 struct file *file, const struct open_flags *op)
3437 struct dentry *dir = nd->path.dentry;
3438 int open_flag = op->open_flag;
3439 bool got_write = false;
3440 struct dentry *dentry;
3443 nd->flags |= op->intent;
3445 if (nd->last_type != LAST_NORM) {
3448 return handle_dots(nd, nd->last_type);
3451 if (!(open_flag & O_CREAT)) {
3452 if (nd->last.name[nd->last.len])
3453 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3454 /* we _can_ be in RCU mode here */
3455 dentry = lookup_fast(nd);
3457 return ERR_CAST(dentry);
3461 BUG_ON(nd->flags & LOOKUP_RCU);
3463 /* create side of things */
3464 if (nd->flags & LOOKUP_RCU) {
3465 if (!try_to_unlazy(nd))
3466 return ERR_PTR(-ECHILD);
3468 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3469 /* trailing slashes? */
3470 if (unlikely(nd->last.name[nd->last.len]))
3471 return ERR_PTR(-EISDIR);
3474 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3475 got_write = !mnt_want_write(nd->path.mnt);
3477 * do _not_ fail yet - we might not need that or fail with
3478 * a different error; let lookup_open() decide; we'll be
3479 * dropping this one anyway.
3482 if (open_flag & O_CREAT)
3483 inode_lock(dir->d_inode);
3485 inode_lock_shared(dir->d_inode);
3486 dentry = lookup_open(nd, file, op, got_write);
3487 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3488 fsnotify_create(dir->d_inode, dentry);
3489 if (open_flag & O_CREAT)
3490 inode_unlock(dir->d_inode);
3492 inode_unlock_shared(dir->d_inode);
3495 mnt_drop_write(nd->path.mnt);
3498 return ERR_CAST(dentry);
3500 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3501 dput(nd->path.dentry);
3502 nd->path.dentry = dentry;
3509 res = step_into(nd, WALK_TRAILING, dentry);
3511 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3516 * Handle the last step of open()
3518 static int do_open(struct nameidata *nd,
3519 struct file *file, const struct open_flags *op)
3521 struct mnt_idmap *idmap;
3522 struct user_namespace *mnt_userns;
3523 int open_flag = op->open_flag;
3528 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3529 error = complete_walk(nd);
3533 if (!(file->f_mode & FMODE_CREATED))
3534 audit_inode(nd->name, nd->path.dentry, 0);
3535 idmap = mnt_idmap(nd->path.mnt);
3536 mnt_userns = mnt_idmap_owner(idmap);
3537 if (open_flag & O_CREAT) {
3538 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3540 if (d_is_dir(nd->path.dentry))
3542 error = may_create_in_sticky(mnt_userns, nd,
3543 d_backing_inode(nd->path.dentry));
3544 if (unlikely(error))
3547 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3550 do_truncate = false;
3551 acc_mode = op->acc_mode;
3552 if (file->f_mode & FMODE_CREATED) {
3553 /* Don't check for write permission, don't truncate */
3554 open_flag &= ~O_TRUNC;
3556 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3557 error = mnt_want_write(nd->path.mnt);
3562 error = may_open(mnt_userns, &nd->path, acc_mode, open_flag);
3563 if (!error && !(file->f_mode & FMODE_OPENED))
3564 error = vfs_open(&nd->path, file);
3566 error = ima_file_check(file, op->acc_mode);
3567 if (!error && do_truncate)
3568 error = handle_truncate(idmap, file);
3569 if (unlikely(error > 0)) {
3574 mnt_drop_write(nd->path.mnt);
3579 * vfs_tmpfile - create tmpfile
3580 * @idmap: idmap of the mount the inode was found from
3581 * @dentry: pointer to dentry of the base directory
3582 * @mode: mode of the new tmpfile
3585 * Create a temporary file.
3587 * If the inode has been found through an idmapped mount the idmap of
3588 * the vfsmount must be passed through @idmap. This function will then take
3589 * care to map the inode according to @idmap before checking permissions.
3590 * On non-idmapped mounts or if permission checking is to be performed on the
3591 * raw inode simply passs @nop_mnt_idmap.
3593 static int vfs_tmpfile(struct mnt_idmap *idmap,
3594 const struct path *parentpath,
3595 struct file *file, umode_t mode)
3597 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
3598 struct dentry *child;
3599 struct inode *dir = d_inode(parentpath->dentry);
3600 struct inode *inode;
3602 int open_flag = file->f_flags;
3604 /* we want directory to be writable */
3605 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
3608 if (!dir->i_op->tmpfile)
3610 child = d_alloc(parentpath->dentry, &slash_name);
3611 if (unlikely(!child))
3613 file->f_path.mnt = parentpath->mnt;
3614 file->f_path.dentry = child;
3615 mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode);
3616 error = dir->i_op->tmpfile(idmap, dir, file, mode);
3620 /* Don't check for other permissions, the inode was just created */
3621 error = may_open(mnt_userns, &file->f_path, 0, file->f_flags);
3624 inode = file_inode(file);
3625 if (!(open_flag & O_EXCL)) {
3626 spin_lock(&inode->i_lock);
3627 inode->i_state |= I_LINKABLE;
3628 spin_unlock(&inode->i_lock);
3630 ima_post_create_tmpfile(mnt_userns, inode);
3635 * vfs_tmpfile_open - open a tmpfile for kernel internal use
3636 * @idmap: idmap of the mount the inode was found from
3637 * @parentpath: path of the base directory
3638 * @mode: mode of the new tmpfile
3640 * @cred: credentials for open
3642 * Create and open a temporary file. The file is not accounted in nr_files,
3643 * hence this is only for kernel internal use, and must not be installed into
3644 * file tables or such.
3646 struct file *vfs_tmpfile_open(struct mnt_idmap *idmap,
3647 const struct path *parentpath,
3648 umode_t mode, int open_flag, const struct cred *cred)
3653 file = alloc_empty_file_noaccount(open_flag, cred);
3654 if (!IS_ERR(file)) {
3655 error = vfs_tmpfile(idmap, parentpath, file, mode);
3658 file = ERR_PTR(error);
3663 EXPORT_SYMBOL(vfs_tmpfile_open);
3665 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3666 const struct open_flags *op,
3670 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3672 if (unlikely(error))
3674 error = mnt_want_write(path.mnt);
3675 if (unlikely(error))
3677 error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode);
3680 audit_inode(nd->name, file->f_path.dentry, 0);
3682 mnt_drop_write(path.mnt);
3688 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3691 int error = path_lookupat(nd, flags, &path);
3693 audit_inode(nd->name, path.dentry, 0);
3694 error = vfs_open(&path, file);
3700 static struct file *path_openat(struct nameidata *nd,
3701 const struct open_flags *op, unsigned flags)
3706 file = alloc_empty_file(op->open_flag, current_cred());
3710 if (unlikely(file->f_flags & __O_TMPFILE)) {
3711 error = do_tmpfile(nd, flags, op, file);
3712 } else if (unlikely(file->f_flags & O_PATH)) {
3713 error = do_o_path(nd, flags, file);
3715 const char *s = path_init(nd, flags);
3716 while (!(error = link_path_walk(s, nd)) &&
3717 (s = open_last_lookups(nd, file, op)) != NULL)
3720 error = do_open(nd, file, op);
3723 if (likely(!error)) {
3724 if (likely(file->f_mode & FMODE_OPENED))
3730 if (error == -EOPENSTALE) {
3731 if (flags & LOOKUP_RCU)
3736 return ERR_PTR(error);
3739 struct file *do_filp_open(int dfd, struct filename *pathname,
3740 const struct open_flags *op)
3742 struct nameidata nd;
3743 int flags = op->lookup_flags;
3746 set_nameidata(&nd, dfd, pathname, NULL);
3747 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3748 if (unlikely(filp == ERR_PTR(-ECHILD)))
3749 filp = path_openat(&nd, op, flags);
3750 if (unlikely(filp == ERR_PTR(-ESTALE)))
3751 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3752 restore_nameidata();
3756 struct file *do_file_open_root(const struct path *root,
3757 const char *name, const struct open_flags *op)
3759 struct nameidata nd;
3761 struct filename *filename;
3762 int flags = op->lookup_flags;
3764 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3765 return ERR_PTR(-ELOOP);
3767 filename = getname_kernel(name);
3768 if (IS_ERR(filename))
3769 return ERR_CAST(filename);
3771 set_nameidata(&nd, -1, filename, root);
3772 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3773 if (unlikely(file == ERR_PTR(-ECHILD)))
3774 file = path_openat(&nd, op, flags);
3775 if (unlikely(file == ERR_PTR(-ESTALE)))
3776 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3777 restore_nameidata();
3782 static struct dentry *filename_create(int dfd, struct filename *name,
3783 struct path *path, unsigned int lookup_flags)
3785 struct dentry *dentry = ERR_PTR(-EEXIST);
3787 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3788 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3789 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3794 error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3796 return ERR_PTR(error);
3799 * Yucky last component or no last component at all?
3800 * (foo/., foo/.., /////)
3802 if (unlikely(type != LAST_NORM))
3805 /* don't fail immediately if it's r/o, at least try to report other errors */
3806 err2 = mnt_want_write(path->mnt);
3808 * Do the final lookup. Suppress 'create' if there is a trailing
3809 * '/', and a directory wasn't requested.
3811 if (last.name[last.len] && !want_dir)
3813 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3814 dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags);
3819 if (d_is_positive(dentry))
3823 * Special case - lookup gave negative, but... we had foo/bar/
3824 * From the vfs_mknod() POV we just have a negative dentry -
3825 * all is fine. Let's be bastards - you had / on the end, you've
3826 * been asking for (non-existent) directory. -ENOENT for you.
3828 if (unlikely(!create_flags)) {
3832 if (unlikely(err2)) {
3839 dentry = ERR_PTR(error);
3841 inode_unlock(path->dentry->d_inode);
3843 mnt_drop_write(path->mnt);
3849 struct dentry *kern_path_create(int dfd, const char *pathname,
3850 struct path *path, unsigned int lookup_flags)
3852 struct filename *filename = getname_kernel(pathname);
3853 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3858 EXPORT_SYMBOL(kern_path_create);
3860 void done_path_create(struct path *path, struct dentry *dentry)
3863 inode_unlock(path->dentry->d_inode);
3864 mnt_drop_write(path->mnt);
3867 EXPORT_SYMBOL(done_path_create);
3869 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3870 struct path *path, unsigned int lookup_flags)
3872 struct filename *filename = getname(pathname);
3873 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3878 EXPORT_SYMBOL(user_path_create);
3881 * vfs_mknod - create device node or file
3882 * @idmap: idmap of the mount the inode was found from
3883 * @dir: inode of @dentry
3884 * @dentry: pointer to dentry of the base directory
3885 * @mode: mode of the new device node or file
3886 * @dev: device number of device to create
3888 * Create a device node or file.
3890 * If the inode has been found through an idmapped mount the idmap of
3891 * the vfsmount must be passed through @idmap. This function will then take
3892 * care to map the inode according to @idmap before checking permissions.
3893 * On non-idmapped mounts or if permission checking is to be performed on the
3894 * raw inode simply passs @nop_mnt_idmap.
3896 int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
3897 struct dentry *dentry, umode_t mode, dev_t dev)
3899 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
3900 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3901 int error = may_create(mnt_userns, dir, dentry);
3906 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3907 !capable(CAP_MKNOD))
3910 if (!dir->i_op->mknod)
3913 mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode);
3914 error = devcgroup_inode_mknod(mode, dev);
3918 error = security_inode_mknod(dir, dentry, mode, dev);
3922 error = dir->i_op->mknod(idmap, dir, dentry, mode, dev);
3924 fsnotify_create(dir, dentry);
3927 EXPORT_SYMBOL(vfs_mknod);
3929 static int may_mknod(umode_t mode)
3931 switch (mode & S_IFMT) {
3937 case 0: /* zero mode translates to S_IFREG */
3946 static int do_mknodat(int dfd, struct filename *name, umode_t mode,
3949 struct mnt_idmap *idmap;
3950 struct user_namespace *mnt_userns;
3951 struct dentry *dentry;
3954 unsigned int lookup_flags = 0;
3956 error = may_mknod(mode);
3960 dentry = filename_create(dfd, name, &path, lookup_flags);
3961 error = PTR_ERR(dentry);
3965 error = security_path_mknod(&path, dentry,
3966 mode_strip_umask(path.dentry->d_inode, mode), dev);
3970 idmap = mnt_idmap(path.mnt);
3971 mnt_userns = mnt_idmap_owner(idmap);
3972 switch (mode & S_IFMT) {
3973 case 0: case S_IFREG:
3974 error = vfs_create(idmap, path.dentry->d_inode,
3975 dentry, mode, true);
3977 ima_post_path_mknod(mnt_userns, dentry);
3979 case S_IFCHR: case S_IFBLK:
3980 error = vfs_mknod(idmap, path.dentry->d_inode,
3981 dentry, mode, new_decode_dev(dev));
3983 case S_IFIFO: case S_IFSOCK:
3984 error = vfs_mknod(idmap, path.dentry->d_inode,
3989 done_path_create(&path, dentry);
3990 if (retry_estale(error, lookup_flags)) {
3991 lookup_flags |= LOOKUP_REVAL;
3999 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
4002 return do_mknodat(dfd, getname(filename), mode, dev);
4005 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
4007 return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
4011 * vfs_mkdir - create directory
4012 * @idmap: idmap of the mount the inode was found from
4013 * @dir: inode of @dentry
4014 * @dentry: pointer to dentry of the base directory
4015 * @mode: mode of the new directory
4017 * Create a directory.
4019 * If the inode has been found through an idmapped mount the idmap of
4020 * the vfsmount must be passed through @idmap. This function will then take
4021 * care to map the inode according to @idmap before checking permissions.
4022 * On non-idmapped mounts or if permission checking is to be performed on the
4023 * raw inode simply passs @nop_mnt_idmap.
4025 int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
4026 struct dentry *dentry, umode_t mode)
4028 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
4030 unsigned max_links = dir->i_sb->s_max_links;
4032 error = may_create(mnt_userns, dir, dentry);
4036 if (!dir->i_op->mkdir)
4039 mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IRWXUGO | S_ISVTX, 0);
4040 error = security_inode_mkdir(dir, dentry, mode);
4044 if (max_links && dir->i_nlink >= max_links)
4047 error = dir->i_op->mkdir(idmap, dir, dentry, mode);
4049 fsnotify_mkdir(dir, dentry);
4052 EXPORT_SYMBOL(vfs_mkdir);
4054 int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4056 struct dentry *dentry;
4059 unsigned int lookup_flags = LOOKUP_DIRECTORY;
4062 dentry = filename_create(dfd, name, &path, lookup_flags);
4063 error = PTR_ERR(dentry);
4067 error = security_path_mkdir(&path, dentry,
4068 mode_strip_umask(path.dentry->d_inode, mode));
4070 error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode,
4073 done_path_create(&path, dentry);
4074 if (retry_estale(error, lookup_flags)) {
4075 lookup_flags |= LOOKUP_REVAL;
4083 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4085 return do_mkdirat(dfd, getname(pathname), mode);
4088 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4090 return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4094 * vfs_rmdir - remove directory
4095 * @idmap: idmap of the mount the inode was found from
4096 * @dir: inode of @dentry
4097 * @dentry: pointer to dentry of the base directory
4099 * Remove a directory.
4101 * If the inode has been found through an idmapped mount the idmap of
4102 * the vfsmount must be passed through @idmap. This function will then take
4103 * care to map the inode according to @idmap before checking permissions.
4104 * On non-idmapped mounts or if permission checking is to be performed on the
4105 * raw inode simply passs @nop_mnt_idmap.
4107 int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir,
4108 struct dentry *dentry)
4110 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
4111 int error = may_delete(mnt_userns, dir, dentry, 1);
4116 if (!dir->i_op->rmdir)
4120 inode_lock(dentry->d_inode);
4123 if (is_local_mountpoint(dentry) ||
4124 (dentry->d_inode->i_flags & S_KERNEL_FILE))
4127 error = security_inode_rmdir(dir, dentry);
4131 error = dir->i_op->rmdir(dir, dentry);
4135 shrink_dcache_parent(dentry);
4136 dentry->d_inode->i_flags |= S_DEAD;
4138 detach_mounts(dentry);
4141 inode_unlock(dentry->d_inode);
4144 d_delete_notify(dir, dentry);
4147 EXPORT_SYMBOL(vfs_rmdir);
4149 int do_rmdir(int dfd, struct filename *name)
4152 struct dentry *dentry;
4156 unsigned int lookup_flags = 0;
4158 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4174 error = mnt_want_write(path.mnt);
4178 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4179 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4180 error = PTR_ERR(dentry);
4183 if (!dentry->d_inode) {
4187 error = security_path_rmdir(&path, dentry);
4190 error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry);
4194 inode_unlock(path.dentry->d_inode);
4195 mnt_drop_write(path.mnt);
4198 if (retry_estale(error, lookup_flags)) {
4199 lookup_flags |= LOOKUP_REVAL;
4207 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4209 return do_rmdir(AT_FDCWD, getname(pathname));
4213 * vfs_unlink - unlink a filesystem object
4214 * @idmap: idmap of the mount the inode was found from
4215 * @dir: parent directory
4217 * @delegated_inode: returns victim inode, if the inode is delegated.
4219 * The caller must hold dir->i_mutex.
4221 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4222 * return a reference to the inode in delegated_inode. The caller
4223 * should then break the delegation on that inode and retry. Because
4224 * breaking a delegation may take a long time, the caller should drop
4225 * dir->i_mutex before doing so.
4227 * Alternatively, a caller may pass NULL for delegated_inode. This may
4228 * be appropriate for callers that expect the underlying filesystem not
4229 * to be NFS exported.
4231 * If the inode has been found through an idmapped mount the idmap of
4232 * the vfsmount must be passed through @idmap. This function will then take
4233 * care to map the inode according to @idmap before checking permissions.
4234 * On non-idmapped mounts or if permission checking is to be performed on the
4235 * raw inode simply passs @nop_mnt_idmap.
4237 int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir,
4238 struct dentry *dentry, struct inode **delegated_inode)
4240 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
4241 struct inode *target = dentry->d_inode;
4242 int error = may_delete(mnt_userns, dir, dentry, 0);
4247 if (!dir->i_op->unlink)
4251 if (IS_SWAPFILE(target))
4253 else if (is_local_mountpoint(dentry))
4256 error = security_inode_unlink(dir, dentry);
4258 error = try_break_deleg(target, delegated_inode);
4261 error = dir->i_op->unlink(dir, dentry);
4264 detach_mounts(dentry);
4269 inode_unlock(target);
4271 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4272 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4273 fsnotify_unlink(dir, dentry);
4274 } else if (!error) {
4275 fsnotify_link_count(target);
4276 d_delete_notify(dir, dentry);
4281 EXPORT_SYMBOL(vfs_unlink);
4284 * Make sure that the actual truncation of the file will occur outside its
4285 * directory's i_mutex. Truncate can take a long time if there is a lot of
4286 * writeout happening, and we don't want to prevent access to the directory
4287 * while waiting on the I/O.
4289 int do_unlinkat(int dfd, struct filename *name)
4292 struct dentry *dentry;
4296 struct inode *inode = NULL;
4297 struct inode *delegated_inode = NULL;
4298 unsigned int lookup_flags = 0;
4300 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4305 if (type != LAST_NORM)
4308 error = mnt_want_write(path.mnt);
4312 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4313 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4314 error = PTR_ERR(dentry);
4315 if (!IS_ERR(dentry)) {
4317 /* Why not before? Because we want correct error value */
4318 if (last.name[last.len])
4320 inode = dentry->d_inode;
4321 if (d_is_negative(dentry))
4324 error = security_path_unlink(&path, dentry);
4327 error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4328 dentry, &delegated_inode);
4332 inode_unlock(path.dentry->d_inode);
4334 iput(inode); /* truncate the inode here */
4336 if (delegated_inode) {
4337 error = break_deleg_wait(&delegated_inode);
4341 mnt_drop_write(path.mnt);
4344 if (retry_estale(error, lookup_flags)) {
4345 lookup_flags |= LOOKUP_REVAL;
4354 if (d_is_negative(dentry))
4356 else if (d_is_dir(dentry))
4363 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4365 if ((flag & ~AT_REMOVEDIR) != 0)
4368 if (flag & AT_REMOVEDIR)
4369 return do_rmdir(dfd, getname(pathname));
4370 return do_unlinkat(dfd, getname(pathname));
4373 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4375 return do_unlinkat(AT_FDCWD, getname(pathname));
4379 * vfs_symlink - create symlink
4380 * @idmap: idmap of the mount the inode was found from
4381 * @dir: inode of @dentry
4382 * @dentry: pointer to dentry of the base directory
4383 * @oldname: name of the file to link to
4387 * If the inode has been found through an idmapped mount the idmap of
4388 * the vfsmount must be passed through @idmap. This function will then take
4389 * care to map the inode according to @idmap before checking permissions.
4390 * On non-idmapped mounts or if permission checking is to be performed on the
4391 * raw inode simply passs @nop_mnt_idmap.
4393 int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
4394 struct dentry *dentry, const char *oldname)
4396 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
4399 error = may_create(mnt_userns, dir, dentry);
4403 if (!dir->i_op->symlink)
4406 error = security_inode_symlink(dir, dentry, oldname);
4410 error = dir->i_op->symlink(idmap, dir, dentry, oldname);
4412 fsnotify_create(dir, dentry);
4415 EXPORT_SYMBOL(vfs_symlink);
4417 int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4420 struct dentry *dentry;
4422 unsigned int lookup_flags = 0;
4425 error = PTR_ERR(from);
4429 dentry = filename_create(newdfd, to, &path, lookup_flags);
4430 error = PTR_ERR(dentry);
4434 error = security_path_symlink(&path, dentry, from->name);
4436 error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4437 dentry, from->name);
4438 done_path_create(&path, dentry);
4439 if (retry_estale(error, lookup_flags)) {
4440 lookup_flags |= LOOKUP_REVAL;
4449 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4450 int, newdfd, const char __user *, newname)
4452 return do_symlinkat(getname(oldname), newdfd, getname(newname));
4455 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4457 return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4461 * vfs_link - create a new link
4462 * @old_dentry: object to be linked
4463 * @idmap: idmap of the mount
4465 * @new_dentry: where to create the new link
4466 * @delegated_inode: returns inode needing a delegation break
4468 * The caller must hold dir->i_mutex
4470 * If vfs_link discovers a delegation on the to-be-linked file in need
4471 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4472 * inode in delegated_inode. The caller should then break the delegation
4473 * and retry. Because breaking a delegation may take a long time, the
4474 * caller should drop the i_mutex before doing so.
4476 * Alternatively, a caller may pass NULL for delegated_inode. This may
4477 * be appropriate for callers that expect the underlying filesystem not
4478 * to be NFS exported.
4480 * If the inode has been found through an idmapped mount the idmap of
4481 * the vfsmount must be passed through @idmap. This function will then take
4482 * care to map the inode according to @idmap before checking permissions.
4483 * On non-idmapped mounts or if permission checking is to be performed on the
4484 * raw inode simply passs @nop_mnt_idmap.
4486 int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap,
4487 struct inode *dir, struct dentry *new_dentry,
4488 struct inode **delegated_inode)
4490 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
4491 struct inode *inode = old_dentry->d_inode;
4492 unsigned max_links = dir->i_sb->s_max_links;
4498 error = may_create(mnt_userns, dir, new_dentry);
4502 if (dir->i_sb != inode->i_sb)
4506 * A link to an append-only or immutable file cannot be created.
4508 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4511 * Updating the link count will likely cause i_uid and i_gid to
4512 * be writen back improperly if their true value is unknown to
4515 if (HAS_UNMAPPED_ID(mnt_userns, inode))
4517 if (!dir->i_op->link)
4519 if (S_ISDIR(inode->i_mode))
4522 error = security_inode_link(old_dentry, dir, new_dentry);
4527 /* Make sure we don't allow creating hardlink to an unlinked file */
4528 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4530 else if (max_links && inode->i_nlink >= max_links)
4533 error = try_break_deleg(inode, delegated_inode);
4535 error = dir->i_op->link(old_dentry, dir, new_dentry);
4538 if (!error && (inode->i_state & I_LINKABLE)) {
4539 spin_lock(&inode->i_lock);
4540 inode->i_state &= ~I_LINKABLE;
4541 spin_unlock(&inode->i_lock);
4543 inode_unlock(inode);
4545 fsnotify_link(dir, inode, new_dentry);
4548 EXPORT_SYMBOL(vfs_link);
4551 * Hardlinks are often used in delicate situations. We avoid
4552 * security-related surprises by not following symlinks on the
4555 * We don't follow them on the oldname either to be compatible
4556 * with linux 2.0, and to avoid hard-linking to directories
4557 * and other special files. --ADM
4559 int do_linkat(int olddfd, struct filename *old, int newdfd,
4560 struct filename *new, int flags)
4562 struct mnt_idmap *idmap;
4563 struct user_namespace *mnt_userns;
4564 struct dentry *new_dentry;
4565 struct path old_path, new_path;
4566 struct inode *delegated_inode = NULL;
4570 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4575 * To use null names we require CAP_DAC_READ_SEARCH
4576 * This ensures that not everyone will be able to create
4577 * handlink using the passed filedescriptor.
4579 if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4584 if (flags & AT_SYMLINK_FOLLOW)
4585 how |= LOOKUP_FOLLOW;
4587 error = filename_lookup(olddfd, old, how, &old_path, NULL);
4591 new_dentry = filename_create(newdfd, new, &new_path,
4592 (how & LOOKUP_REVAL));
4593 error = PTR_ERR(new_dentry);
4594 if (IS_ERR(new_dentry))
4598 if (old_path.mnt != new_path.mnt)
4600 idmap = mnt_idmap(new_path.mnt);
4601 mnt_userns = mnt_idmap_owner(idmap);
4602 error = may_linkat(mnt_userns, &old_path);
4603 if (unlikely(error))
4605 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4608 error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode,
4609 new_dentry, &delegated_inode);
4611 done_path_create(&new_path, new_dentry);
4612 if (delegated_inode) {
4613 error = break_deleg_wait(&delegated_inode);
4615 path_put(&old_path);
4619 if (retry_estale(error, how)) {
4620 path_put(&old_path);
4621 how |= LOOKUP_REVAL;
4625 path_put(&old_path);
4633 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4634 int, newdfd, const char __user *, newname, int, flags)
4636 return do_linkat(olddfd, getname_uflags(oldname, flags),
4637 newdfd, getname(newname), flags);
4640 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4642 return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4646 * vfs_rename - rename a filesystem object
4647 * @rd: pointer to &struct renamedata info
4649 * The caller must hold multiple mutexes--see lock_rename()).
4651 * If vfs_rename discovers a delegation in need of breaking at either
4652 * the source or destination, it will return -EWOULDBLOCK and return a
4653 * reference to the inode in delegated_inode. The caller should then
4654 * break the delegation and retry. Because breaking a delegation may
4655 * take a long time, the caller should drop all locks before doing
4658 * Alternatively, a caller may pass NULL for delegated_inode. This may
4659 * be appropriate for callers that expect the underlying filesystem not
4660 * to be NFS exported.
4662 * The worst of all namespace operations - renaming directory. "Perverted"
4663 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4666 * a) we can get into loop creation.
4667 * b) race potential - two innocent renames can create a loop together.
4668 * That's where 4.4 screws up. Current fix: serialization on
4669 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4671 * c) we have to lock _four_ objects - parents and victim (if it exists),
4672 * and source (if it is not a directory).
4673 * And that - after we got ->i_mutex on parents (until then we don't know
4674 * whether the target exists). Solution: try to be smart with locking
4675 * order for inodes. We rely on the fact that tree topology may change
4676 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4677 * move will be locked. Thus we can rank directories by the tree
4678 * (ancestors first) and rank all non-directories after them.
4679 * That works since everybody except rename does "lock parent, lookup,
4680 * lock child" and rename is under ->s_vfs_rename_mutex.
4681 * HOWEVER, it relies on the assumption that any object with ->lookup()
4682 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4683 * we'd better make sure that there's no link(2) for them.
4684 * d) conversion from fhandle to dentry may come in the wrong moment - when
4685 * we are removing the target. Solution: we will have to grab ->i_mutex
4686 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4687 * ->i_mutex on parents, which works but leads to some truly excessive
4690 int vfs_rename(struct renamedata *rd)
4693 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4694 struct dentry *old_dentry = rd->old_dentry;
4695 struct dentry *new_dentry = rd->new_dentry;
4696 struct inode **delegated_inode = rd->delegated_inode;
4697 unsigned int flags = rd->flags;
4698 bool is_dir = d_is_dir(old_dentry);
4699 struct inode *source = old_dentry->d_inode;
4700 struct inode *target = new_dentry->d_inode;
4701 bool new_is_dir = false;
4702 unsigned max_links = new_dir->i_sb->s_max_links;
4703 struct name_snapshot old_name;
4704 struct user_namespace *old_mnt_userns = mnt_idmap_owner(rd->old_mnt_idmap),
4705 *new_mnt_userns = mnt_idmap_owner(rd->new_mnt_idmap);
4707 if (source == target)
4710 error = may_delete(old_mnt_userns, old_dir, old_dentry, is_dir);
4715 error = may_create(new_mnt_userns, new_dir, new_dentry);
4717 new_is_dir = d_is_dir(new_dentry);
4719 if (!(flags & RENAME_EXCHANGE))
4720 error = may_delete(new_mnt_userns, new_dir,
4721 new_dentry, is_dir);
4723 error = may_delete(new_mnt_userns, new_dir,
4724 new_dentry, new_is_dir);
4729 if (!old_dir->i_op->rename)
4733 * If we are going to change the parent - check write permissions,
4734 * we'll need to flip '..'.
4736 if (new_dir != old_dir) {
4738 error = inode_permission(old_mnt_userns, source,
4743 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4744 error = inode_permission(new_mnt_userns, target,
4751 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4756 take_dentry_name_snapshot(&old_name, old_dentry);
4758 if (!is_dir || (flags & RENAME_EXCHANGE))
4759 lock_two_nondirectories(source, target);
4764 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4768 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4771 if (max_links && new_dir != old_dir) {
4773 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4775 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4776 old_dir->i_nlink >= max_links)
4780 error = try_break_deleg(source, delegated_inode);
4784 if (target && !new_is_dir) {
4785 error = try_break_deleg(target, delegated_inode);
4789 error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry,
4790 new_dir, new_dentry, flags);
4794 if (!(flags & RENAME_EXCHANGE) && target) {
4796 shrink_dcache_parent(new_dentry);
4797 target->i_flags |= S_DEAD;
4799 dont_mount(new_dentry);
4800 detach_mounts(new_dentry);
4802 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4803 if (!(flags & RENAME_EXCHANGE))
4804 d_move(old_dentry, new_dentry);
4806 d_exchange(old_dentry, new_dentry);
4809 if (!is_dir || (flags & RENAME_EXCHANGE))
4810 unlock_two_nondirectories(source, target);
4812 inode_unlock(target);
4815 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4816 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4817 if (flags & RENAME_EXCHANGE) {
4818 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4819 new_is_dir, NULL, new_dentry);
4822 release_dentry_name_snapshot(&old_name);
4826 EXPORT_SYMBOL(vfs_rename);
4828 int do_renameat2(int olddfd, struct filename *from, int newdfd,
4829 struct filename *to, unsigned int flags)
4831 struct renamedata rd;
4832 struct dentry *old_dentry, *new_dentry;
4833 struct dentry *trap;
4834 struct path old_path, new_path;
4835 struct qstr old_last, new_last;
4836 int old_type, new_type;
4837 struct inode *delegated_inode = NULL;
4838 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4839 bool should_retry = false;
4840 int error = -EINVAL;
4842 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4845 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4846 (flags & RENAME_EXCHANGE))
4849 if (flags & RENAME_EXCHANGE)
4853 error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4854 &old_last, &old_type);
4858 error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4864 if (old_path.mnt != new_path.mnt)
4868 if (old_type != LAST_NORM)
4871 if (flags & RENAME_NOREPLACE)
4873 if (new_type != LAST_NORM)
4876 error = mnt_want_write(old_path.mnt);
4881 trap = lock_rename(new_path.dentry, old_path.dentry);
4883 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4884 error = PTR_ERR(old_dentry);
4885 if (IS_ERR(old_dentry))
4887 /* source must exist */
4889 if (d_is_negative(old_dentry))
4891 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4892 error = PTR_ERR(new_dentry);
4893 if (IS_ERR(new_dentry))
4896 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4898 if (flags & RENAME_EXCHANGE) {
4900 if (d_is_negative(new_dentry))
4903 if (!d_is_dir(new_dentry)) {
4905 if (new_last.name[new_last.len])
4909 /* unless the source is a directory trailing slashes give -ENOTDIR */
4910 if (!d_is_dir(old_dentry)) {
4912 if (old_last.name[old_last.len])
4914 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4917 /* source should not be ancestor of target */
4919 if (old_dentry == trap)
4921 /* target should not be an ancestor of source */
4922 if (!(flags & RENAME_EXCHANGE))
4924 if (new_dentry == trap)
4927 error = security_path_rename(&old_path, old_dentry,
4928 &new_path, new_dentry, flags);
4932 rd.old_dir = old_path.dentry->d_inode;
4933 rd.old_dentry = old_dentry;
4934 rd.old_mnt_idmap = mnt_idmap(old_path.mnt);
4935 rd.new_dir = new_path.dentry->d_inode;
4936 rd.new_dentry = new_dentry;
4937 rd.new_mnt_idmap = mnt_idmap(new_path.mnt);
4938 rd.delegated_inode = &delegated_inode;
4940 error = vfs_rename(&rd);
4946 unlock_rename(new_path.dentry, old_path.dentry);
4947 if (delegated_inode) {
4948 error = break_deleg_wait(&delegated_inode);
4952 mnt_drop_write(old_path.mnt);
4954 if (retry_estale(error, lookup_flags))
4955 should_retry = true;
4956 path_put(&new_path);
4958 path_put(&old_path);
4960 should_retry = false;
4961 lookup_flags |= LOOKUP_REVAL;
4970 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4971 int, newdfd, const char __user *, newname, unsigned int, flags)
4973 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4977 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4978 int, newdfd, const char __user *, newname)
4980 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4984 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4986 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4987 getname(newname), 0);
4990 int readlink_copy(char __user *buffer, int buflen, const char *link)
4992 int len = PTR_ERR(link);
4997 if (len > (unsigned) buflen)
4999 if (copy_to_user(buffer, link, len))
5006 * vfs_readlink - copy symlink body into userspace buffer
5007 * @dentry: dentry on which to get symbolic link
5008 * @buffer: user memory pointer
5009 * @buflen: size of buffer
5011 * Does not touch atime. That's up to the caller if necessary
5013 * Does not call security hook.
5015 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5017 struct inode *inode = d_inode(dentry);
5018 DEFINE_DELAYED_CALL(done);
5022 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5023 if (unlikely(inode->i_op->readlink))
5024 return inode->i_op->readlink(dentry, buffer, buflen);
5026 if (!d_is_symlink(dentry))
5029 spin_lock(&inode->i_lock);
5030 inode->i_opflags |= IOP_DEFAULT_READLINK;
5031 spin_unlock(&inode->i_lock);
5034 link = READ_ONCE(inode->i_link);
5036 link = inode->i_op->get_link(dentry, inode, &done);
5038 return PTR_ERR(link);
5040 res = readlink_copy(buffer, buflen, link);
5041 do_delayed_call(&done);
5044 EXPORT_SYMBOL(vfs_readlink);
5047 * vfs_get_link - get symlink body
5048 * @dentry: dentry on which to get symbolic link
5049 * @done: caller needs to free returned data with this
5051 * Calls security hook and i_op->get_link() on the supplied inode.
5053 * It does not touch atime. That's up to the caller if necessary.
5055 * Does not work on "special" symlinks like /proc/$$/fd/N
5057 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5059 const char *res = ERR_PTR(-EINVAL);
5060 struct inode *inode = d_inode(dentry);
5062 if (d_is_symlink(dentry)) {
5063 res = ERR_PTR(security_inode_readlink(dentry));
5065 res = inode->i_op->get_link(dentry, inode, done);
5069 EXPORT_SYMBOL(vfs_get_link);
5071 /* get the link contents into pagecache */
5072 const char *page_get_link(struct dentry *dentry, struct inode *inode,
5073 struct delayed_call *callback)
5077 struct address_space *mapping = inode->i_mapping;
5080 page = find_get_page(mapping, 0);
5082 return ERR_PTR(-ECHILD);
5083 if (!PageUptodate(page)) {
5085 return ERR_PTR(-ECHILD);
5088 page = read_mapping_page(mapping, 0, NULL);
5092 set_delayed_call(callback, page_put_link, page);
5093 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5094 kaddr = page_address(page);
5095 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5099 EXPORT_SYMBOL(page_get_link);
5101 void page_put_link(void *arg)
5105 EXPORT_SYMBOL(page_put_link);
5107 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5109 DEFINE_DELAYED_CALL(done);
5110 int res = readlink_copy(buffer, buflen,
5111 page_get_link(dentry, d_inode(dentry),
5113 do_delayed_call(&done);
5116 EXPORT_SYMBOL(page_readlink);
5118 int page_symlink(struct inode *inode, const char *symname, int len)
5120 struct address_space *mapping = inode->i_mapping;
5121 const struct address_space_operations *aops = mapping->a_ops;
5122 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5124 void *fsdata = NULL;
5130 flags = memalloc_nofs_save();
5131 err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5133 memalloc_nofs_restore(flags);
5137 memcpy(page_address(page), symname, len-1);
5139 err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5146 mark_inode_dirty(inode);
5151 EXPORT_SYMBOL(page_symlink);
5153 const struct inode_operations page_symlink_inode_operations = {
5154 .get_link = page_get_link,
5156 EXPORT_SYMBOL(page_symlink_inode_operations);