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 * @idmap: idmap 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 idmap of
391 * the vfsmount must be passed through @idmap. This function will then take
392 * care to map the inode according to @idmap before checking permissions.
393 * On non-idmapped mounts or if permission checking is to be performed on the
394 * raw inode simply passs @nop_mnt_idmap.
396 int generic_permission(struct mnt_idmap *idmap, struct inode *inode,
400 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
403 * Do the basic permission checks.
405 ret = acl_permission_check(mnt_userns, inode, mask);
409 if (S_ISDIR(inode->i_mode)) {
410 /* DACs are overridable for directories */
411 if (!(mask & MAY_WRITE))
412 if (capable_wrt_inode_uidgid(mnt_userns, inode,
413 CAP_DAC_READ_SEARCH))
415 if (capable_wrt_inode_uidgid(mnt_userns, inode,
422 * Searching includes executable on directories, else just read.
424 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
425 if (mask == MAY_READ)
426 if (capable_wrt_inode_uidgid(mnt_userns, inode,
427 CAP_DAC_READ_SEARCH))
430 * Read/write DACs are always overridable.
431 * Executable DACs are overridable when there is
432 * at least one exec bit set.
434 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
435 if (capable_wrt_inode_uidgid(mnt_userns, inode,
441 EXPORT_SYMBOL(generic_permission);
444 * do_inode_permission - UNIX permission checking
445 * @idmap: idmap of the mount the inode was found from
446 * @inode: inode to check permissions on
447 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
449 * We _really_ want to just do "generic_permission()" without
450 * even looking at the inode->i_op values. So we keep a cache
451 * flag in inode->i_opflags, that says "this has not special
452 * permission function, use the fast case".
454 static inline int do_inode_permission(struct mnt_idmap *idmap,
455 struct inode *inode, int mask)
457 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
458 if (likely(inode->i_op->permission))
459 return inode->i_op->permission(idmap, inode, mask);
461 /* This gets set once for the inode lifetime */
462 spin_lock(&inode->i_lock);
463 inode->i_opflags |= IOP_FASTPERM;
464 spin_unlock(&inode->i_lock);
466 return generic_permission(idmap, inode, mask);
470 * sb_permission - Check superblock-level permissions
471 * @sb: Superblock of inode to check permission on
472 * @inode: Inode to check permission on
473 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
475 * Separate out file-system wide checks from inode-specific permission checks.
477 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
479 if (unlikely(mask & MAY_WRITE)) {
480 umode_t mode = inode->i_mode;
482 /* Nobody gets write access to a read-only fs. */
483 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
490 * inode_permission - Check for access rights to a given inode
491 * @idmap: idmap of the mount the inode was found from
492 * @inode: Inode to check permission on
493 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
495 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
496 * this, letting us set arbitrary permissions for filesystem access without
497 * changing the "normal" UIDs which are used for other things.
499 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
501 int inode_permission(struct mnt_idmap *idmap,
502 struct inode *inode, int mask)
506 retval = sb_permission(inode->i_sb, inode, mask);
510 if (unlikely(mask & MAY_WRITE)) {
512 * Nobody gets write access to an immutable file.
514 if (IS_IMMUTABLE(inode))
518 * Updating mtime will likely cause i_uid and i_gid to be
519 * written back improperly if their true value is unknown
522 if (HAS_UNMAPPED_ID(idmap, inode))
526 retval = do_inode_permission(idmap, inode, mask);
530 retval = devcgroup_inode_permission(inode, mask);
534 return security_inode_permission(inode, mask);
536 EXPORT_SYMBOL(inode_permission);
539 * path_get - get a reference to a path
540 * @path: path to get the reference to
542 * Given a path increment the reference count to the dentry and the vfsmount.
544 void path_get(const struct path *path)
549 EXPORT_SYMBOL(path_get);
552 * path_put - put a reference to a path
553 * @path: path to put the reference to
555 * Given a path decrement the reference count to the dentry and the vfsmount.
557 void path_put(const struct path *path)
562 EXPORT_SYMBOL(path_put);
564 #define EMBEDDED_LEVELS 2
569 struct inode *inode; /* path.dentry.d_inode */
570 unsigned int flags, state;
571 unsigned seq, next_seq, m_seq, r_seq;
574 int total_link_count;
577 struct delayed_call done;
580 } *stack, internal[EMBEDDED_LEVELS];
581 struct filename *name;
582 struct nameidata *saved;
587 } __randomize_layout;
589 #define ND_ROOT_PRESET 1
590 #define ND_ROOT_GRABBED 2
593 static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
595 struct nameidata *old = current->nameidata;
596 p->stack = p->internal;
601 p->path.dentry = NULL;
602 p->total_link_count = old ? old->total_link_count : 0;
604 current->nameidata = p;
607 static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
608 const struct path *root)
610 __set_nameidata(p, dfd, name);
612 if (unlikely(root)) {
613 p->state = ND_ROOT_PRESET;
618 static void restore_nameidata(void)
620 struct nameidata *now = current->nameidata, *old = now->saved;
622 current->nameidata = old;
624 old->total_link_count = now->total_link_count;
625 if (now->stack != now->internal)
629 static bool nd_alloc_stack(struct nameidata *nd)
633 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
634 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
637 memcpy(p, nd->internal, sizeof(nd->internal));
643 * path_connected - Verify that a dentry is below mnt.mnt_root
645 * Rename can sometimes move a file or directory outside of a bind
646 * mount, path_connected allows those cases to be detected.
648 static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
650 struct super_block *sb = mnt->mnt_sb;
652 /* Bind mounts can have disconnected paths */
653 if (mnt->mnt_root == sb->s_root)
656 return is_subdir(dentry, mnt->mnt_root);
659 static void drop_links(struct nameidata *nd)
663 struct saved *last = nd->stack + i;
664 do_delayed_call(&last->done);
665 clear_delayed_call(&last->done);
669 static void leave_rcu(struct nameidata *nd)
671 nd->flags &= ~LOOKUP_RCU;
672 nd->seq = nd->next_seq = 0;
676 static void terminate_walk(struct nameidata *nd)
679 if (!(nd->flags & LOOKUP_RCU)) {
682 for (i = 0; i < nd->depth; i++)
683 path_put(&nd->stack[i].link);
684 if (nd->state & ND_ROOT_GRABBED) {
686 nd->state &= ~ND_ROOT_GRABBED;
693 nd->path.dentry = NULL;
696 /* path_put is needed afterwards regardless of success or failure */
697 static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
699 int res = __legitimize_mnt(path->mnt, mseq);
706 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
710 return !read_seqcount_retry(&path->dentry->d_seq, seq);
713 static inline bool legitimize_path(struct nameidata *nd,
714 struct path *path, unsigned seq)
716 return __legitimize_path(path, seq, nd->m_seq);
719 static bool legitimize_links(struct nameidata *nd)
722 if (unlikely(nd->flags & LOOKUP_CACHED)) {
727 for (i = 0; i < nd->depth; i++) {
728 struct saved *last = nd->stack + i;
729 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
738 static bool legitimize_root(struct nameidata *nd)
740 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
741 if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
743 nd->state |= ND_ROOT_GRABBED;
744 return legitimize_path(nd, &nd->root, nd->root_seq);
748 * Path walking has 2 modes, rcu-walk and ref-walk (see
749 * Documentation/filesystems/path-lookup.txt). In situations when we can't
750 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
751 * normal reference counts on dentries and vfsmounts to transition to ref-walk
752 * mode. Refcounts are grabbed at the last known good point before rcu-walk
753 * got stuck, so ref-walk may continue from there. If this is not successful
754 * (eg. a seqcount has changed), then failure is returned and it's up to caller
755 * to restart the path walk from the beginning in ref-walk mode.
759 * try_to_unlazy - try to switch to ref-walk mode.
760 * @nd: nameidata pathwalk data
761 * Returns: true on success, false on failure
763 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
765 * Must be called from rcu-walk context.
766 * Nothing should touch nameidata between try_to_unlazy() failure and
769 static bool try_to_unlazy(struct nameidata *nd)
771 struct dentry *parent = nd->path.dentry;
773 BUG_ON(!(nd->flags & LOOKUP_RCU));
775 if (unlikely(!legitimize_links(nd)))
777 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
779 if (unlikely(!legitimize_root(nd)))
782 BUG_ON(nd->inode != parent->d_inode);
787 nd->path.dentry = NULL;
794 * try_to_unlazy_next - try to switch to ref-walk mode.
795 * @nd: nameidata pathwalk data
796 * @dentry: next dentry to step into
797 * Returns: true on success, false on failure
799 * Similar to try_to_unlazy(), but here we have the next dentry already
800 * picked by rcu-walk and want to legitimize that in addition to the current
801 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
802 * Nothing should touch nameidata between try_to_unlazy_next() failure and
805 static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry)
808 BUG_ON(!(nd->flags & LOOKUP_RCU));
810 if (unlikely(!legitimize_links(nd)))
812 res = __legitimize_mnt(nd->path.mnt, nd->m_seq);
818 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
822 * We need to move both the parent and the dentry from the RCU domain
823 * to be properly refcounted. And the sequence number in the dentry
824 * validates *both* dentry counters, since we checked the sequence
825 * number of the parent after we got the child sequence number. So we
826 * know the parent must still be valid if the child sequence number is
828 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
830 if (read_seqcount_retry(&dentry->d_seq, nd->next_seq))
833 * Sequence counts matched. Now make sure that the root is
834 * still valid and get it if required.
836 if (unlikely(!legitimize_root(nd)))
844 nd->path.dentry = NULL;
854 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
856 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
857 return dentry->d_op->d_revalidate(dentry, flags);
863 * complete_walk - successful completion of path walk
864 * @nd: pointer nameidata
866 * If we had been in RCU mode, drop out of it and legitimize nd->path.
867 * Revalidate the final result, unless we'd already done that during
868 * the path walk or the filesystem doesn't ask for it. Return 0 on
869 * success, -error on failure. In case of failure caller does not
870 * need to drop nd->path.
872 static int complete_walk(struct nameidata *nd)
874 struct dentry *dentry = nd->path.dentry;
877 if (nd->flags & LOOKUP_RCU) {
879 * We don't want to zero nd->root for scoped-lookups or
880 * externally-managed nd->root.
882 if (!(nd->state & ND_ROOT_PRESET))
883 if (!(nd->flags & LOOKUP_IS_SCOPED))
885 nd->flags &= ~LOOKUP_CACHED;
886 if (!try_to_unlazy(nd))
890 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
892 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
893 * ever step outside the root during lookup" and should already
894 * be guaranteed by the rest of namei, we want to avoid a namei
895 * BUG resulting in userspace being given a path that was not
896 * scoped within the root at some point during the lookup.
898 * So, do a final sanity-check to make sure that in the
899 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
900 * we won't silently return an fd completely outside of the
901 * requested root to userspace.
903 * Userspace could move the path outside the root after this
904 * check, but as discussed elsewhere this is not a concern (the
905 * resolved file was inside the root at some point).
907 if (!path_is_under(&nd->path, &nd->root))
911 if (likely(!(nd->state & ND_JUMPED)))
914 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
917 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
927 static int set_root(struct nameidata *nd)
929 struct fs_struct *fs = current->fs;
932 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
933 * still have to ensure it doesn't happen because it will cause a breakout
936 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
937 return -ENOTRECOVERABLE;
939 if (nd->flags & LOOKUP_RCU) {
943 seq = read_seqcount_begin(&fs->seq);
945 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
946 } while (read_seqcount_retry(&fs->seq, seq));
948 get_fs_root(fs, &nd->root);
949 nd->state |= ND_ROOT_GRABBED;
954 static int nd_jump_root(struct nameidata *nd)
956 if (unlikely(nd->flags & LOOKUP_BENEATH))
958 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
959 /* Absolute path arguments to path_init() are allowed. */
960 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
964 int error = set_root(nd);
968 if (nd->flags & LOOKUP_RCU) {
972 nd->inode = d->d_inode;
973 nd->seq = nd->root_seq;
974 if (read_seqcount_retry(&d->d_seq, nd->seq))
980 nd->inode = nd->path.dentry->d_inode;
982 nd->state |= ND_JUMPED;
987 * Helper to directly jump to a known parsed path from ->get_link,
988 * caller must have taken a reference to path beforehand.
990 int nd_jump_link(const struct path *path)
993 struct nameidata *nd = current->nameidata;
995 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
999 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
1000 if (nd->path.mnt != path->mnt)
1003 /* Not currently safe for scoped-lookups. */
1004 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1007 path_put(&nd->path);
1009 nd->inode = nd->path.dentry->d_inode;
1010 nd->state |= ND_JUMPED;
1018 static inline void put_link(struct nameidata *nd)
1020 struct saved *last = nd->stack + --nd->depth;
1021 do_delayed_call(&last->done);
1022 if (!(nd->flags & LOOKUP_RCU))
1023 path_put(&last->link);
1026 static int sysctl_protected_symlinks __read_mostly;
1027 static int sysctl_protected_hardlinks __read_mostly;
1028 static int sysctl_protected_fifos __read_mostly;
1029 static int sysctl_protected_regular __read_mostly;
1031 #ifdef CONFIG_SYSCTL
1032 static struct ctl_table namei_sysctls[] = {
1034 .procname = "protected_symlinks",
1035 .data = &sysctl_protected_symlinks,
1036 .maxlen = sizeof(int),
1038 .proc_handler = proc_dointvec_minmax,
1039 .extra1 = SYSCTL_ZERO,
1040 .extra2 = SYSCTL_ONE,
1043 .procname = "protected_hardlinks",
1044 .data = &sysctl_protected_hardlinks,
1045 .maxlen = sizeof(int),
1047 .proc_handler = proc_dointvec_minmax,
1048 .extra1 = SYSCTL_ZERO,
1049 .extra2 = SYSCTL_ONE,
1052 .procname = "protected_fifos",
1053 .data = &sysctl_protected_fifos,
1054 .maxlen = sizeof(int),
1056 .proc_handler = proc_dointvec_minmax,
1057 .extra1 = SYSCTL_ZERO,
1058 .extra2 = SYSCTL_TWO,
1061 .procname = "protected_regular",
1062 .data = &sysctl_protected_regular,
1063 .maxlen = sizeof(int),
1065 .proc_handler = proc_dointvec_minmax,
1066 .extra1 = SYSCTL_ZERO,
1067 .extra2 = SYSCTL_TWO,
1072 static int __init init_fs_namei_sysctls(void)
1074 register_sysctl_init("fs", namei_sysctls);
1077 fs_initcall(init_fs_namei_sysctls);
1079 #endif /* CONFIG_SYSCTL */
1082 * may_follow_link - Check symlink following for unsafe situations
1083 * @nd: nameidata pathwalk data
1085 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1086 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1087 * in a sticky world-writable directory. This is to protect privileged
1088 * processes from failing races against path names that may change out
1089 * from under them by way of other users creating malicious symlinks.
1090 * It will permit symlinks to be followed only when outside a sticky
1091 * world-writable directory, or when the uid of the symlink and follower
1092 * match, or when the directory owner matches the symlink's owner.
1094 * Returns 0 if following the symlink is allowed, -ve on error.
1096 static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1098 struct user_namespace *mnt_userns;
1101 if (!sysctl_protected_symlinks)
1104 mnt_userns = mnt_user_ns(nd->path.mnt);
1105 vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
1106 /* Allowed if owner and follower match. */
1107 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
1110 /* Allowed if parent directory not sticky and world-writable. */
1111 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1114 /* Allowed if parent directory and link owner match. */
1115 if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid))
1118 if (nd->flags & LOOKUP_RCU)
1121 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1122 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1127 * safe_hardlink_source - Check for safe hardlink conditions
1128 * @idmap: idmap of the mount the inode was found from
1129 * @inode: the source inode to hardlink from
1131 * Return false if at least one of the following conditions:
1132 * - inode is not a regular file
1134 * - inode is setgid and group-exec
1135 * - access failure for read and write
1137 * Otherwise returns true.
1139 static bool safe_hardlink_source(struct mnt_idmap *idmap,
1140 struct inode *inode)
1142 umode_t mode = inode->i_mode;
1144 /* Special files should not get pinned to the filesystem. */
1148 /* Setuid files should not get pinned to the filesystem. */
1152 /* Executable setgid files should not get pinned to the filesystem. */
1153 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1156 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1157 if (inode_permission(idmap, inode, MAY_READ | MAY_WRITE))
1164 * may_linkat - Check permissions for creating a hardlink
1165 * @idmap: idmap of the mount the inode was found from
1166 * @link: the source to hardlink from
1168 * Block hardlink when all of:
1169 * - sysctl_protected_hardlinks enabled
1170 * - fsuid does not match inode
1171 * - hardlink source is unsafe (see safe_hardlink_source() above)
1172 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1174 * If the inode has been found through an idmapped mount the idmap of
1175 * the vfsmount must be passed through @idmap. This function will then take
1176 * care to map the inode according to @idmap before checking permissions.
1177 * On non-idmapped mounts or if permission checking is to be performed on the
1178 * raw inode simply pass @nop_mnt_idmap.
1180 * Returns 0 if successful, -ve on error.
1182 int may_linkat(struct mnt_idmap *idmap, const struct path *link)
1184 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
1185 struct inode *inode = link->dentry->d_inode;
1187 /* Inode writeback is not safe when the uid or gid are invalid. */
1188 if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) ||
1189 !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode)))
1192 if (!sysctl_protected_hardlinks)
1195 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1196 * otherwise, it must be a safe source.
1198 if (safe_hardlink_source(idmap, inode) ||
1199 inode_owner_or_capable(mnt_userns, inode))
1202 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1207 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1208 * should be allowed, or not, on files that already
1210 * @mnt_userns: user namespace of the mount the inode was found from
1211 * @nd: nameidata pathwalk data
1212 * @inode: the inode of the file to open
1214 * Block an O_CREAT open of a FIFO (or a regular file) when:
1215 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1216 * - the file already exists
1217 * - we are in a sticky directory
1218 * - we don't own the file
1219 * - the owner of the directory doesn't own the file
1220 * - the directory is world writable
1221 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1222 * the directory doesn't have to be world writable: being group writable will
1225 * If the inode has been found through an idmapped mount the user namespace of
1226 * the vfsmount must be passed through @mnt_userns. This function will then take
1227 * care to map the inode according to @mnt_userns before checking permissions.
1228 * On non-idmapped mounts or if permission checking is to be performed on the
1229 * raw inode simply passs init_user_ns.
1231 * Returns 0 if the open is allowed, -ve on error.
1233 static int may_create_in_sticky(struct user_namespace *mnt_userns,
1234 struct nameidata *nd, struct inode *const inode)
1236 umode_t dir_mode = nd->dir_mode;
1237 vfsuid_t dir_vfsuid = nd->dir_vfsuid;
1239 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1240 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1241 likely(!(dir_mode & S_ISVTX)) ||
1242 vfsuid_eq(i_uid_into_vfsuid(mnt_userns, inode), dir_vfsuid) ||
1243 vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), current_fsuid()))
1246 if (likely(dir_mode & 0002) ||
1248 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1249 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1250 const char *operation = S_ISFIFO(inode->i_mode) ?
1251 "sticky_create_fifo" :
1252 "sticky_create_regular";
1253 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1260 * follow_up - Find the mountpoint of path's vfsmount
1262 * Given a path, find the mountpoint of its source file system.
1263 * Replace @path with the path of the mountpoint in the parent mount.
1266 * Return 1 if we went up a level and 0 if we were already at the
1269 int follow_up(struct path *path)
1271 struct mount *mnt = real_mount(path->mnt);
1272 struct mount *parent;
1273 struct dentry *mountpoint;
1275 read_seqlock_excl(&mount_lock);
1276 parent = mnt->mnt_parent;
1277 if (parent == mnt) {
1278 read_sequnlock_excl(&mount_lock);
1281 mntget(&parent->mnt);
1282 mountpoint = dget(mnt->mnt_mountpoint);
1283 read_sequnlock_excl(&mount_lock);
1285 path->dentry = mountpoint;
1287 path->mnt = &parent->mnt;
1290 EXPORT_SYMBOL(follow_up);
1292 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1293 struct path *path, unsigned *seqp)
1295 while (mnt_has_parent(m)) {
1296 struct dentry *mountpoint = m->mnt_mountpoint;
1299 if (unlikely(root->dentry == mountpoint &&
1300 root->mnt == &m->mnt))
1302 if (mountpoint != m->mnt.mnt_root) {
1303 path->mnt = &m->mnt;
1304 path->dentry = mountpoint;
1305 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1312 static bool choose_mountpoint(struct mount *m, const struct path *root,
1319 unsigned seq, mseq = read_seqbegin(&mount_lock);
1321 found = choose_mountpoint_rcu(m, root, path, &seq);
1322 if (unlikely(!found)) {
1323 if (!read_seqretry(&mount_lock, mseq))
1326 if (likely(__legitimize_path(path, seq, mseq)))
1338 * Perform an automount
1339 * - return -EISDIR to tell follow_managed() to stop and return the path we
1342 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1344 struct dentry *dentry = path->dentry;
1346 /* We don't want to mount if someone's just doing a stat -
1347 * unless they're stat'ing a directory and appended a '/' to
1350 * We do, however, want to mount if someone wants to open or
1351 * create a file of any type under the mountpoint, wants to
1352 * traverse through the mountpoint or wants to open the
1353 * mounted directory. Also, autofs may mark negative dentries
1354 * as being automount points. These will need the attentions
1355 * of the daemon to instantiate them before they can be used.
1357 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1358 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1362 if (count && (*count)++ >= MAXSYMLINKS)
1365 return finish_automount(dentry->d_op->d_automount(path), path);
1369 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1370 * dentries are pinned but not locked here, so negative dentry can go
1371 * positive right under us. Use of smp_load_acquire() provides a barrier
1372 * sufficient for ->d_inode and ->d_flags consistency.
1374 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1375 int *count, unsigned lookup_flags)
1377 struct vfsmount *mnt = path->mnt;
1378 bool need_mntput = false;
1381 while (flags & DCACHE_MANAGED_DENTRY) {
1382 /* Allow the filesystem to manage the transit without i_mutex
1384 if (flags & DCACHE_MANAGE_TRANSIT) {
1385 ret = path->dentry->d_op->d_manage(path, false);
1386 flags = smp_load_acquire(&path->dentry->d_flags);
1391 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1392 struct vfsmount *mounted = lookup_mnt(path);
1393 if (mounted) { // ... in our namespace
1397 path->mnt = mounted;
1398 path->dentry = dget(mounted->mnt_root);
1399 // here we know it's positive
1400 flags = path->dentry->d_flags;
1406 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1409 // uncovered automount point
1410 ret = follow_automount(path, count, lookup_flags);
1411 flags = smp_load_acquire(&path->dentry->d_flags);
1418 // possible if you race with several mount --move
1419 if (need_mntput && path->mnt == mnt)
1421 if (!ret && unlikely(d_flags_negative(flags)))
1423 *jumped = need_mntput;
1427 static inline int traverse_mounts(struct path *path, bool *jumped,
1428 int *count, unsigned lookup_flags)
1430 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1433 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1435 if (unlikely(d_flags_negative(flags)))
1439 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1442 int follow_down_one(struct path *path)
1444 struct vfsmount *mounted;
1446 mounted = lookup_mnt(path);
1450 path->mnt = mounted;
1451 path->dentry = dget(mounted->mnt_root);
1456 EXPORT_SYMBOL(follow_down_one);
1459 * Follow down to the covering mount currently visible to userspace. At each
1460 * point, the filesystem owning that dentry may be queried as to whether the
1461 * caller is permitted to proceed or not.
1463 int follow_down(struct path *path)
1465 struct vfsmount *mnt = path->mnt;
1467 int ret = traverse_mounts(path, &jumped, NULL, 0);
1469 if (path->mnt != mnt)
1473 EXPORT_SYMBOL(follow_down);
1476 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1477 * we meet a managed dentry that would need blocking.
1479 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1481 struct dentry *dentry = path->dentry;
1482 unsigned int flags = dentry->d_flags;
1484 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1487 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1492 * Don't forget we might have a non-mountpoint managed dentry
1493 * that wants to block transit.
1495 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1496 int res = dentry->d_op->d_manage(path, true);
1498 return res == -EISDIR;
1499 flags = dentry->d_flags;
1502 if (flags & DCACHE_MOUNTED) {
1503 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1505 path->mnt = &mounted->mnt;
1506 dentry = path->dentry = mounted->mnt.mnt_root;
1507 nd->state |= ND_JUMPED;
1508 nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1509 flags = dentry->d_flags;
1510 // makes sure that non-RCU pathwalk could reach
1512 if (read_seqretry(&mount_lock, nd->m_seq))
1516 if (read_seqretry(&mount_lock, nd->m_seq))
1519 return !(flags & DCACHE_NEED_AUTOMOUNT);
1523 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1529 path->mnt = nd->path.mnt;
1530 path->dentry = dentry;
1531 if (nd->flags & LOOKUP_RCU) {
1532 unsigned int seq = nd->next_seq;
1533 if (likely(__follow_mount_rcu(nd, path)))
1535 // *path and nd->next_seq might've been clobbered
1536 path->mnt = nd->path.mnt;
1537 path->dentry = dentry;
1539 if (!try_to_unlazy_next(nd, dentry))
1542 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1544 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1547 nd->state |= ND_JUMPED;
1549 if (unlikely(ret)) {
1551 if (path->mnt != nd->path.mnt)
1558 * This looks up the name in dcache and possibly revalidates the found dentry.
1559 * NULL is returned if the dentry does not exist in the cache.
1561 static struct dentry *lookup_dcache(const struct qstr *name,
1565 struct dentry *dentry = d_lookup(dir, name);
1567 int error = d_revalidate(dentry, flags);
1568 if (unlikely(error <= 0)) {
1570 d_invalidate(dentry);
1572 return ERR_PTR(error);
1579 * Parent directory has inode locked exclusive. This is one
1580 * and only case when ->lookup() gets called on non in-lookup
1581 * dentries - as the matter of fact, this only gets called
1582 * when directory is guaranteed to have no in-lookup children
1585 static struct dentry *__lookup_hash(const struct qstr *name,
1586 struct dentry *base, unsigned int flags)
1588 struct dentry *dentry = lookup_dcache(name, base, flags);
1590 struct inode *dir = base->d_inode;
1595 /* Don't create child dentry for a dead directory. */
1596 if (unlikely(IS_DEADDIR(dir)))
1597 return ERR_PTR(-ENOENT);
1599 dentry = d_alloc(base, name);
1600 if (unlikely(!dentry))
1601 return ERR_PTR(-ENOMEM);
1603 old = dir->i_op->lookup(dir, dentry, flags);
1604 if (unlikely(old)) {
1611 static struct dentry *lookup_fast(struct nameidata *nd)
1613 struct dentry *dentry, *parent = nd->path.dentry;
1617 * Rename seqlock is not required here because in the off chance
1618 * of a false negative due to a concurrent rename, the caller is
1619 * going to fall back to non-racy lookup.
1621 if (nd->flags & LOOKUP_RCU) {
1622 dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq);
1623 if (unlikely(!dentry)) {
1624 if (!try_to_unlazy(nd))
1625 return ERR_PTR(-ECHILD);
1630 * This sequence count validates that the parent had no
1631 * changes while we did the lookup of the dentry above.
1633 if (read_seqcount_retry(&parent->d_seq, nd->seq))
1634 return ERR_PTR(-ECHILD);
1636 status = d_revalidate(dentry, nd->flags);
1637 if (likely(status > 0))
1639 if (!try_to_unlazy_next(nd, dentry))
1640 return ERR_PTR(-ECHILD);
1641 if (status == -ECHILD)
1642 /* we'd been told to redo it in non-rcu mode */
1643 status = d_revalidate(dentry, nd->flags);
1645 dentry = __d_lookup(parent, &nd->last);
1646 if (unlikely(!dentry))
1648 status = d_revalidate(dentry, nd->flags);
1650 if (unlikely(status <= 0)) {
1652 d_invalidate(dentry);
1654 return ERR_PTR(status);
1659 /* Fast lookup failed, do it the slow way */
1660 static struct dentry *__lookup_slow(const struct qstr *name,
1664 struct dentry *dentry, *old;
1665 struct inode *inode = dir->d_inode;
1666 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1668 /* Don't go there if it's already dead */
1669 if (unlikely(IS_DEADDIR(inode)))
1670 return ERR_PTR(-ENOENT);
1672 dentry = d_alloc_parallel(dir, name, &wq);
1675 if (unlikely(!d_in_lookup(dentry))) {
1676 int error = d_revalidate(dentry, flags);
1677 if (unlikely(error <= 0)) {
1679 d_invalidate(dentry);
1684 dentry = ERR_PTR(error);
1687 old = inode->i_op->lookup(inode, dentry, flags);
1688 d_lookup_done(dentry);
1689 if (unlikely(old)) {
1697 static struct dentry *lookup_slow(const struct qstr *name,
1701 struct inode *inode = dir->d_inode;
1703 inode_lock_shared(inode);
1704 res = __lookup_slow(name, dir, flags);
1705 inode_unlock_shared(inode);
1709 static inline int may_lookup(struct mnt_idmap *idmap,
1710 struct nameidata *nd)
1712 if (nd->flags & LOOKUP_RCU) {
1713 int err = inode_permission(idmap, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1714 if (err != -ECHILD || !try_to_unlazy(nd))
1717 return inode_permission(idmap, nd->inode, MAY_EXEC);
1720 static int reserve_stack(struct nameidata *nd, struct path *link)
1722 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1725 if (likely(nd->depth != EMBEDDED_LEVELS))
1727 if (likely(nd->stack != nd->internal))
1729 if (likely(nd_alloc_stack(nd)))
1732 if (nd->flags & LOOKUP_RCU) {
1733 // we need to grab link before we do unlazy. And we can't skip
1734 // unlazy even if we fail to grab the link - cleanup needs it
1735 bool grabbed_link = legitimize_path(nd, link, nd->next_seq);
1737 if (!try_to_unlazy(nd) || !grabbed_link)
1740 if (nd_alloc_stack(nd))
1746 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1748 static const char *pick_link(struct nameidata *nd, struct path *link,
1749 struct inode *inode, int flags)
1753 int error = reserve_stack(nd, link);
1755 if (unlikely(error)) {
1756 if (!(nd->flags & LOOKUP_RCU))
1758 return ERR_PTR(error);
1760 last = nd->stack + nd->depth++;
1762 clear_delayed_call(&last->done);
1763 last->seq = nd->next_seq;
1765 if (flags & WALK_TRAILING) {
1766 error = may_follow_link(nd, inode);
1767 if (unlikely(error))
1768 return ERR_PTR(error);
1771 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1772 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1773 return ERR_PTR(-ELOOP);
1775 if (!(nd->flags & LOOKUP_RCU)) {
1776 touch_atime(&last->link);
1778 } else if (atime_needs_update(&last->link, inode)) {
1779 if (!try_to_unlazy(nd))
1780 return ERR_PTR(-ECHILD);
1781 touch_atime(&last->link);
1784 error = security_inode_follow_link(link->dentry, inode,
1785 nd->flags & LOOKUP_RCU);
1786 if (unlikely(error))
1787 return ERR_PTR(error);
1789 res = READ_ONCE(inode->i_link);
1791 const char * (*get)(struct dentry *, struct inode *,
1792 struct delayed_call *);
1793 get = inode->i_op->get_link;
1794 if (nd->flags & LOOKUP_RCU) {
1795 res = get(NULL, inode, &last->done);
1796 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1797 res = get(link->dentry, inode, &last->done);
1799 res = get(link->dentry, inode, &last->done);
1807 error = nd_jump_root(nd);
1808 if (unlikely(error))
1809 return ERR_PTR(error);
1810 while (unlikely(*++res == '/'))
1815 all_done: // pure jump
1821 * Do we need to follow links? We _really_ want to be able
1822 * to do this check without having to look at inode->i_op,
1823 * so we keep a cache of "no, this doesn't need follow_link"
1824 * for the common case.
1826 * NOTE: dentry must be what nd->next_seq had been sampled from.
1828 static const char *step_into(struct nameidata *nd, int flags,
1829 struct dentry *dentry)
1832 struct inode *inode;
1833 int err = handle_mounts(nd, dentry, &path);
1836 return ERR_PTR(err);
1837 inode = path.dentry->d_inode;
1838 if (likely(!d_is_symlink(path.dentry)) ||
1839 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1840 (flags & WALK_NOFOLLOW)) {
1841 /* not a symlink or should not follow */
1842 if (nd->flags & LOOKUP_RCU) {
1843 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1844 return ERR_PTR(-ECHILD);
1845 if (unlikely(!inode))
1846 return ERR_PTR(-ENOENT);
1848 dput(nd->path.dentry);
1849 if (nd->path.mnt != path.mnt)
1850 mntput(nd->path.mnt);
1854 nd->seq = nd->next_seq;
1857 if (nd->flags & LOOKUP_RCU) {
1858 /* make sure that d_is_symlink above matches inode */
1859 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1860 return ERR_PTR(-ECHILD);
1862 if (path.mnt == nd->path.mnt)
1865 return pick_link(nd, &path, inode, flags);
1868 static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1870 struct dentry *parent, *old;
1872 if (path_equal(&nd->path, &nd->root))
1874 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1877 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1878 &nd->root, &path, &seq))
1880 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1881 return ERR_PTR(-ECHILD);
1883 nd->inode = path.dentry->d_inode;
1885 // makes sure that non-RCU pathwalk could reach this state
1886 if (read_seqretry(&mount_lock, nd->m_seq))
1887 return ERR_PTR(-ECHILD);
1888 /* we know that mountpoint was pinned */
1890 old = nd->path.dentry;
1891 parent = old->d_parent;
1892 nd->next_seq = read_seqcount_begin(&parent->d_seq);
1893 // makes sure that non-RCU pathwalk could reach this state
1894 if (read_seqcount_retry(&old->d_seq, nd->seq))
1895 return ERR_PTR(-ECHILD);
1896 if (unlikely(!path_connected(nd->path.mnt, parent)))
1897 return ERR_PTR(-ECHILD);
1900 if (read_seqretry(&mount_lock, nd->m_seq))
1901 return ERR_PTR(-ECHILD);
1902 if (unlikely(nd->flags & LOOKUP_BENEATH))
1903 return ERR_PTR(-ECHILD);
1904 nd->next_seq = nd->seq;
1905 return nd->path.dentry;
1908 static struct dentry *follow_dotdot(struct nameidata *nd)
1910 struct dentry *parent;
1912 if (path_equal(&nd->path, &nd->root))
1914 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1917 if (!choose_mountpoint(real_mount(nd->path.mnt),
1920 path_put(&nd->path);
1922 nd->inode = path.dentry->d_inode;
1923 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1924 return ERR_PTR(-EXDEV);
1926 /* rare case of legitimate dget_parent()... */
1927 parent = dget_parent(nd->path.dentry);
1928 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1930 return ERR_PTR(-ENOENT);
1935 if (unlikely(nd->flags & LOOKUP_BENEATH))
1936 return ERR_PTR(-EXDEV);
1937 return dget(nd->path.dentry);
1940 static const char *handle_dots(struct nameidata *nd, int type)
1942 if (type == LAST_DOTDOT) {
1943 const char *error = NULL;
1944 struct dentry *parent;
1946 if (!nd->root.mnt) {
1947 error = ERR_PTR(set_root(nd));
1951 if (nd->flags & LOOKUP_RCU)
1952 parent = follow_dotdot_rcu(nd);
1954 parent = follow_dotdot(nd);
1956 return ERR_CAST(parent);
1957 error = step_into(nd, WALK_NOFOLLOW, parent);
1958 if (unlikely(error))
1961 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1963 * If there was a racing rename or mount along our
1964 * path, then we can't be sure that ".." hasn't jumped
1965 * above nd->root (and so userspace should retry or use
1969 if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
1970 return ERR_PTR(-EAGAIN);
1971 if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
1972 return ERR_PTR(-EAGAIN);
1978 static const char *walk_component(struct nameidata *nd, int flags)
1980 struct dentry *dentry;
1982 * "." and ".." are special - ".." especially so because it has
1983 * to be able to know about the current root directory and
1984 * parent relationships.
1986 if (unlikely(nd->last_type != LAST_NORM)) {
1987 if (!(flags & WALK_MORE) && nd->depth)
1989 return handle_dots(nd, nd->last_type);
1991 dentry = lookup_fast(nd);
1993 return ERR_CAST(dentry);
1994 if (unlikely(!dentry)) {
1995 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1997 return ERR_CAST(dentry);
1999 if (!(flags & WALK_MORE) && nd->depth)
2001 return step_into(nd, flags, dentry);
2005 * We can do the critical dentry name comparison and hashing
2006 * operations one word at a time, but we are limited to:
2008 * - Architectures with fast unaligned word accesses. We could
2009 * do a "get_unaligned()" if this helps and is sufficiently
2012 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2013 * do not trap on the (extremely unlikely) case of a page
2014 * crossing operation.
2016 * - Furthermore, we need an efficient 64-bit compile for the
2017 * 64-bit case in order to generate the "number of bytes in
2018 * the final mask". Again, that could be replaced with a
2019 * efficient population count instruction or similar.
2021 #ifdef CONFIG_DCACHE_WORD_ACCESS
2023 #include <asm/word-at-a-time.h>
2027 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2029 #elif defined(CONFIG_64BIT)
2031 * Register pressure in the mixing function is an issue, particularly
2032 * on 32-bit x86, but almost any function requires one state value and
2033 * one temporary. Instead, use a function designed for two state values
2034 * and no temporaries.
2036 * This function cannot create a collision in only two iterations, so
2037 * we have two iterations to achieve avalanche. In those two iterations,
2038 * we have six layers of mixing, which is enough to spread one bit's
2039 * influence out to 2^6 = 64 state bits.
2041 * Rotate constants are scored by considering either 64 one-bit input
2042 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2043 * probability of that delta causing a change to each of the 128 output
2044 * bits, using a sample of random initial states.
2046 * The Shannon entropy of the computed probabilities is then summed
2047 * to produce a score. Ideally, any input change has a 50% chance of
2048 * toggling any given output bit.
2050 * Mixing scores (in bits) for (12,45):
2051 * Input delta: 1-bit 2-bit
2052 * 1 round: 713.3 42542.6
2053 * 2 rounds: 2753.7 140389.8
2054 * 3 rounds: 5954.1 233458.2
2055 * 4 rounds: 7862.6 256672.2
2056 * Perfect: 8192 258048
2057 * (64*128) (64*63/2 * 128)
2059 #define HASH_MIX(x, y, a) \
2061 y ^= x, x = rol64(x,12),\
2062 x += y, y = rol64(y,45),\
2066 * Fold two longs into one 32-bit hash value. This must be fast, but
2067 * latency isn't quite as critical, as there is a fair bit of additional
2068 * work done before the hash value is used.
2070 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2072 y ^= x * GOLDEN_RATIO_64;
2073 y *= GOLDEN_RATIO_64;
2077 #else /* 32-bit case */
2080 * Mixing scores (in bits) for (7,20):
2081 * Input delta: 1-bit 2-bit
2082 * 1 round: 330.3 9201.6
2083 * 2 rounds: 1246.4 25475.4
2084 * 3 rounds: 1907.1 31295.1
2085 * 4 rounds: 2042.3 31718.6
2086 * Perfect: 2048 31744
2087 * (32*64) (32*31/2 * 64)
2089 #define HASH_MIX(x, y, a) \
2091 y ^= x, x = rol32(x, 7),\
2092 x += y, y = rol32(y,20),\
2095 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2097 /* Use arch-optimized multiply if one exists */
2098 return __hash_32(y ^ __hash_32(x));
2104 * Return the hash of a string of known length. This is carfully
2105 * designed to match hash_name(), which is the more critical function.
2106 * In particular, we must end by hashing a final word containing 0..7
2107 * payload bytes, to match the way that hash_name() iterates until it
2108 * finds the delimiter after the name.
2110 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2112 unsigned long a, x = 0, y = (unsigned long)salt;
2117 a = load_unaligned_zeropad(name);
2118 if (len < sizeof(unsigned long))
2121 name += sizeof(unsigned long);
2122 len -= sizeof(unsigned long);
2124 x ^= a & bytemask_from_count(len);
2126 return fold_hash(x, y);
2128 EXPORT_SYMBOL(full_name_hash);
2130 /* Return the "hash_len" (hash and length) of a null-terminated string */
2131 u64 hashlen_string(const void *salt, const char *name)
2133 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2134 unsigned long adata, mask, len;
2135 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2142 len += sizeof(unsigned long);
2144 a = load_unaligned_zeropad(name+len);
2145 } while (!has_zero(a, &adata, &constants));
2147 adata = prep_zero_mask(a, adata, &constants);
2148 mask = create_zero_mask(adata);
2149 x ^= a & zero_bytemask(mask);
2151 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2153 EXPORT_SYMBOL(hashlen_string);
2156 * Calculate the length and hash of the path component, and
2157 * return the "hash_len" as the result.
2159 static inline u64 hash_name(const void *salt, const char *name)
2161 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2162 unsigned long adata, bdata, mask, len;
2163 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2170 len += sizeof(unsigned long);
2172 a = load_unaligned_zeropad(name+len);
2173 b = a ^ REPEAT_BYTE('/');
2174 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2176 adata = prep_zero_mask(a, adata, &constants);
2177 bdata = prep_zero_mask(b, bdata, &constants);
2178 mask = create_zero_mask(adata | bdata);
2179 x ^= a & zero_bytemask(mask);
2181 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2184 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2186 /* Return the hash of a string of known length */
2187 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2189 unsigned long hash = init_name_hash(salt);
2191 hash = partial_name_hash((unsigned char)*name++, hash);
2192 return end_name_hash(hash);
2194 EXPORT_SYMBOL(full_name_hash);
2196 /* Return the "hash_len" (hash and length) of a null-terminated string */
2197 u64 hashlen_string(const void *salt, const char *name)
2199 unsigned long hash = init_name_hash(salt);
2200 unsigned long len = 0, c;
2202 c = (unsigned char)*name;
2205 hash = partial_name_hash(c, hash);
2206 c = (unsigned char)name[len];
2208 return hashlen_create(end_name_hash(hash), len);
2210 EXPORT_SYMBOL(hashlen_string);
2213 * We know there's a real path component here of at least
2216 static inline u64 hash_name(const void *salt, const char *name)
2218 unsigned long hash = init_name_hash(salt);
2219 unsigned long len = 0, c;
2221 c = (unsigned char)*name;
2224 hash = partial_name_hash(c, hash);
2225 c = (unsigned char)name[len];
2226 } while (c && c != '/');
2227 return hashlen_create(end_name_hash(hash), len);
2234 * This is the basic name resolution function, turning a pathname into
2235 * the final dentry. We expect 'base' to be positive and a directory.
2237 * Returns 0 and nd will have valid dentry and mnt on success.
2238 * Returns error and drops reference to input namei data on failure.
2240 static int link_path_walk(const char *name, struct nameidata *nd)
2242 int depth = 0; // depth <= nd->depth
2245 nd->last_type = LAST_ROOT;
2246 nd->flags |= LOOKUP_PARENT;
2248 return PTR_ERR(name);
2252 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2256 /* At this point we know we have a real path component. */
2258 struct mnt_idmap *idmap;
2259 struct user_namespace *mnt_userns;
2264 idmap = mnt_idmap(nd->path.mnt);
2265 mnt_userns = mnt_idmap_owner(idmap);
2266 err = may_lookup(idmap, nd);
2270 hash_len = hash_name(nd->path.dentry, name);
2273 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2275 if (name[1] == '.') {
2277 nd->state |= ND_JUMPED;
2283 if (likely(type == LAST_NORM)) {
2284 struct dentry *parent = nd->path.dentry;
2285 nd->state &= ~ND_JUMPED;
2286 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2287 struct qstr this = { { .hash_len = hash_len }, .name = name };
2288 err = parent->d_op->d_hash(parent, &this);
2291 hash_len = this.hash_len;
2296 nd->last.hash_len = hash_len;
2297 nd->last.name = name;
2298 nd->last_type = type;
2300 name += hashlen_len(hash_len);
2304 * If it wasn't NUL, we know it was '/'. Skip that
2305 * slash, and continue until no more slashes.
2309 } while (unlikely(*name == '/'));
2310 if (unlikely(!*name)) {
2312 /* pathname or trailing symlink, done */
2314 nd->dir_vfsuid = i_uid_into_vfsuid(mnt_userns, nd->inode);
2315 nd->dir_mode = nd->inode->i_mode;
2316 nd->flags &= ~LOOKUP_PARENT;
2319 /* last component of nested symlink */
2320 name = nd->stack[--depth].name;
2321 link = walk_component(nd, 0);
2323 /* not the last component */
2324 link = walk_component(nd, WALK_MORE);
2326 if (unlikely(link)) {
2328 return PTR_ERR(link);
2329 /* a symlink to follow */
2330 nd->stack[depth++].name = name;
2334 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2335 if (nd->flags & LOOKUP_RCU) {
2336 if (!try_to_unlazy(nd))
2344 /* must be paired with terminate_walk() */
2345 static const char *path_init(struct nameidata *nd, unsigned flags)
2348 const char *s = nd->name->name;
2350 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2351 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2352 return ERR_PTR(-EAGAIN);
2355 flags &= ~LOOKUP_RCU;
2356 if (flags & LOOKUP_RCU)
2359 nd->seq = nd->next_seq = 0;
2362 nd->state |= ND_JUMPED;
2364 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2365 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2368 if (nd->state & ND_ROOT_PRESET) {
2369 struct dentry *root = nd->root.dentry;
2370 struct inode *inode = root->d_inode;
2371 if (*s && unlikely(!d_can_lookup(root)))
2372 return ERR_PTR(-ENOTDIR);
2373 nd->path = nd->root;
2375 if (flags & LOOKUP_RCU) {
2376 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2377 nd->root_seq = nd->seq;
2379 path_get(&nd->path);
2384 nd->root.mnt = NULL;
2386 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2387 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2388 error = nd_jump_root(nd);
2389 if (unlikely(error))
2390 return ERR_PTR(error);
2394 /* Relative pathname -- get the starting-point it is relative to. */
2395 if (nd->dfd == AT_FDCWD) {
2396 if (flags & LOOKUP_RCU) {
2397 struct fs_struct *fs = current->fs;
2401 seq = read_seqcount_begin(&fs->seq);
2403 nd->inode = nd->path.dentry->d_inode;
2404 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2405 } while (read_seqcount_retry(&fs->seq, seq));
2407 get_fs_pwd(current->fs, &nd->path);
2408 nd->inode = nd->path.dentry->d_inode;
2411 /* Caller must check execute permissions on the starting path component */
2412 struct fd f = fdget_raw(nd->dfd);
2413 struct dentry *dentry;
2416 return ERR_PTR(-EBADF);
2418 dentry = f.file->f_path.dentry;
2420 if (*s && unlikely(!d_can_lookup(dentry))) {
2422 return ERR_PTR(-ENOTDIR);
2425 nd->path = f.file->f_path;
2426 if (flags & LOOKUP_RCU) {
2427 nd->inode = nd->path.dentry->d_inode;
2428 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2430 path_get(&nd->path);
2431 nd->inode = nd->path.dentry->d_inode;
2436 /* For scoped-lookups we need to set the root to the dirfd as well. */
2437 if (flags & LOOKUP_IS_SCOPED) {
2438 nd->root = nd->path;
2439 if (flags & LOOKUP_RCU) {
2440 nd->root_seq = nd->seq;
2442 path_get(&nd->root);
2443 nd->state |= ND_ROOT_GRABBED;
2449 static inline const char *lookup_last(struct nameidata *nd)
2451 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2452 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2454 return walk_component(nd, WALK_TRAILING);
2457 static int handle_lookup_down(struct nameidata *nd)
2459 if (!(nd->flags & LOOKUP_RCU))
2460 dget(nd->path.dentry);
2461 nd->next_seq = nd->seq;
2462 return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry));
2465 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2466 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2468 const char *s = path_init(nd, flags);
2471 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2472 err = handle_lookup_down(nd);
2473 if (unlikely(err < 0))
2477 while (!(err = link_path_walk(s, nd)) &&
2478 (s = lookup_last(nd)) != NULL)
2480 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2481 err = handle_lookup_down(nd);
2482 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2485 err = complete_walk(nd);
2487 if (!err && nd->flags & LOOKUP_DIRECTORY)
2488 if (!d_can_lookup(nd->path.dentry))
2492 nd->path.mnt = NULL;
2493 nd->path.dentry = NULL;
2499 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2500 struct path *path, struct path *root)
2503 struct nameidata nd;
2505 return PTR_ERR(name);
2506 set_nameidata(&nd, dfd, name, root);
2507 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2508 if (unlikely(retval == -ECHILD))
2509 retval = path_lookupat(&nd, flags, path);
2510 if (unlikely(retval == -ESTALE))
2511 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2513 if (likely(!retval))
2514 audit_inode(name, path->dentry,
2515 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2516 restore_nameidata();
2520 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2521 static int path_parentat(struct nameidata *nd, unsigned flags,
2522 struct path *parent)
2524 const char *s = path_init(nd, flags);
2525 int err = link_path_walk(s, nd);
2527 err = complete_walk(nd);
2530 nd->path.mnt = NULL;
2531 nd->path.dentry = NULL;
2537 /* Note: this does not consume "name" */
2538 static int filename_parentat(int dfd, struct filename *name,
2539 unsigned int flags, struct path *parent,
2540 struct qstr *last, int *type)
2543 struct nameidata nd;
2546 return PTR_ERR(name);
2547 set_nameidata(&nd, dfd, name, NULL);
2548 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2549 if (unlikely(retval == -ECHILD))
2550 retval = path_parentat(&nd, flags, parent);
2551 if (unlikely(retval == -ESTALE))
2552 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2553 if (likely(!retval)) {
2555 *type = nd.last_type;
2556 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2558 restore_nameidata();
2562 /* does lookup, returns the object with parent locked */
2563 static struct dentry *__kern_path_locked(struct filename *name, struct path *path)
2569 error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type);
2571 return ERR_PTR(error);
2572 if (unlikely(type != LAST_NORM)) {
2574 return ERR_PTR(-EINVAL);
2576 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2577 d = __lookup_hash(&last, path->dentry, 0);
2579 inode_unlock(path->dentry->d_inode);
2585 struct dentry *kern_path_locked(const char *name, struct path *path)
2587 struct filename *filename = getname_kernel(name);
2588 struct dentry *res = __kern_path_locked(filename, path);
2594 int kern_path(const char *name, unsigned int flags, struct path *path)
2596 struct filename *filename = getname_kernel(name);
2597 int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2603 EXPORT_SYMBOL(kern_path);
2606 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2607 * @dentry: pointer to dentry of the base directory
2608 * @mnt: pointer to vfs mount of the base directory
2609 * @name: pointer to file name
2610 * @flags: lookup flags
2611 * @path: pointer to struct path to fill
2613 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2614 const char *name, unsigned int flags,
2617 struct filename *filename;
2618 struct path root = {.mnt = mnt, .dentry = dentry};
2621 filename = getname_kernel(name);
2622 /* the first argument of filename_lookup() is ignored with root */
2623 ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2627 EXPORT_SYMBOL(vfs_path_lookup);
2629 static int lookup_one_common(struct mnt_idmap *idmap,
2630 const char *name, struct dentry *base, int len,
2635 this->hash = full_name_hash(base, name, len);
2639 if (unlikely(name[0] == '.')) {
2640 if (len < 2 || (len == 2 && name[1] == '.'))
2645 unsigned int c = *(const unsigned char *)name++;
2646 if (c == '/' || c == '\0')
2650 * See if the low-level filesystem might want
2651 * to use its own hash..
2653 if (base->d_flags & DCACHE_OP_HASH) {
2654 int err = base->d_op->d_hash(base, this);
2659 return inode_permission(idmap, base->d_inode, MAY_EXEC);
2663 * try_lookup_one_len - filesystem helper to lookup single pathname component
2664 * @name: pathname component to lookup
2665 * @base: base directory to lookup from
2666 * @len: maximum length @len should be interpreted to
2668 * Look up a dentry by name in the dcache, returning NULL if it does not
2669 * currently exist. The function does not try to create a dentry.
2671 * Note that this routine is purely a helper for filesystem usage and should
2672 * not be called by generic code.
2674 * The caller must hold base->i_mutex.
2676 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2681 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2683 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2685 return ERR_PTR(err);
2687 return lookup_dcache(&this, base, 0);
2689 EXPORT_SYMBOL(try_lookup_one_len);
2692 * lookup_one_len - filesystem helper to lookup single pathname component
2693 * @name: pathname component to lookup
2694 * @base: base directory to lookup from
2695 * @len: maximum length @len should be interpreted to
2697 * Note that this routine is purely a helper for filesystem usage and should
2698 * not be called by generic code.
2700 * The caller must hold base->i_mutex.
2702 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2704 struct dentry *dentry;
2708 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2710 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2712 return ERR_PTR(err);
2714 dentry = lookup_dcache(&this, base, 0);
2715 return dentry ? dentry : __lookup_slow(&this, base, 0);
2717 EXPORT_SYMBOL(lookup_one_len);
2720 * lookup_one - filesystem helper to lookup single pathname component
2721 * @idmap: idmap of the mount the lookup is performed from
2722 * @name: pathname component to lookup
2723 * @base: base directory to lookup from
2724 * @len: maximum length @len should be interpreted to
2726 * Note that this routine is purely a helper for filesystem usage and should
2727 * not be called by generic code.
2729 * The caller must hold base->i_mutex.
2731 struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name,
2732 struct dentry *base, int len)
2734 struct dentry *dentry;
2738 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2740 err = lookup_one_common(idmap, name, base, len, &this);
2742 return ERR_PTR(err);
2744 dentry = lookup_dcache(&this, base, 0);
2745 return dentry ? dentry : __lookup_slow(&this, base, 0);
2747 EXPORT_SYMBOL(lookup_one);
2750 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2751 * @idmap: idmap of the mount the lookup is performed from
2752 * @name: pathname component to lookup
2753 * @base: base directory to lookup from
2754 * @len: maximum length @len should be interpreted to
2756 * Note that this routine is purely a helper for filesystem usage and should
2757 * not be called by generic code.
2759 * Unlike lookup_one_len, it should be called without the parent
2760 * i_mutex held, and will take the i_mutex itself if necessary.
2762 struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap,
2763 const char *name, struct dentry *base,
2770 err = lookup_one_common(idmap, name, base, len, &this);
2772 return ERR_PTR(err);
2774 ret = lookup_dcache(&this, base, 0);
2776 ret = lookup_slow(&this, base, 0);
2779 EXPORT_SYMBOL(lookup_one_unlocked);
2782 * lookup_one_positive_unlocked - filesystem helper to lookup single
2783 * pathname component
2784 * @idmap: idmap of the mount the lookup is performed from
2785 * @name: pathname component to lookup
2786 * @base: base directory to lookup from
2787 * @len: maximum length @len should be interpreted to
2789 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2790 * known positive or ERR_PTR(). This is what most of the users want.
2792 * Note that pinned negative with unlocked parent _can_ become positive at any
2793 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2794 * positives have >d_inode stable, so this one avoids such problems.
2796 * Note that this routine is purely a helper for filesystem usage and should
2797 * not be called by generic code.
2799 * The helper should be called without i_mutex held.
2801 struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap,
2803 struct dentry *base, int len)
2805 struct dentry *ret = lookup_one_unlocked(idmap, name, base, len);
2807 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2809 ret = ERR_PTR(-ENOENT);
2813 EXPORT_SYMBOL(lookup_one_positive_unlocked);
2816 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2817 * @name: pathname component to lookup
2818 * @base: base directory to lookup from
2819 * @len: maximum length @len should be interpreted to
2821 * Note that this routine is purely a helper for filesystem usage and should
2822 * not be called by generic code.
2824 * Unlike lookup_one_len, it should be called without the parent
2825 * i_mutex held, and will take the i_mutex itself if necessary.
2827 struct dentry *lookup_one_len_unlocked(const char *name,
2828 struct dentry *base, int len)
2830 return lookup_one_unlocked(&nop_mnt_idmap, name, base, len);
2832 EXPORT_SYMBOL(lookup_one_len_unlocked);
2835 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2836 * on negatives. Returns known positive or ERR_PTR(); that's what
2837 * most of the users want. Note that pinned negative with unlocked parent
2838 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2839 * need to be very careful; pinned positives have ->d_inode stable, so
2840 * this one avoids such problems.
2842 struct dentry *lookup_positive_unlocked(const char *name,
2843 struct dentry *base, int len)
2845 return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len);
2847 EXPORT_SYMBOL(lookup_positive_unlocked);
2849 #ifdef CONFIG_UNIX98_PTYS
2850 int path_pts(struct path *path)
2852 /* Find something mounted on "pts" in the same directory as
2855 struct dentry *parent = dget_parent(path->dentry);
2856 struct dentry *child;
2857 struct qstr this = QSTR_INIT("pts", 3);
2859 if (unlikely(!path_connected(path->mnt, parent))) {
2864 path->dentry = parent;
2865 child = d_hash_and_lookup(parent, &this);
2869 path->dentry = child;
2876 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2877 struct path *path, int *empty)
2879 struct filename *filename = getname_flags(name, flags, empty);
2880 int ret = filename_lookup(dfd, filename, flags, path, NULL);
2885 EXPORT_SYMBOL(user_path_at_empty);
2887 int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir,
2888 struct inode *inode)
2890 kuid_t fsuid = current_fsuid();
2892 if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), fsuid))
2894 if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, dir), fsuid))
2896 return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER);
2898 EXPORT_SYMBOL(__check_sticky);
2901 * Check whether we can remove a link victim from directory dir, check
2902 * whether the type of victim is right.
2903 * 1. We can't do it if dir is read-only (done in permission())
2904 * 2. We should have write and exec permissions on dir
2905 * 3. We can't remove anything from append-only dir
2906 * 4. We can't do anything with immutable dir (done in permission())
2907 * 5. If the sticky bit on dir is set we should either
2908 * a. be owner of dir, or
2909 * b. be owner of victim, or
2910 * c. have CAP_FOWNER capability
2911 * 6. If the victim is append-only or immutable we can't do antyhing with
2912 * links pointing to it.
2913 * 7. If the victim has an unknown uid or gid we can't change the inode.
2914 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2915 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2916 * 10. We can't remove a root or mountpoint.
2917 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2918 * nfs_async_unlink().
2920 static int may_delete(struct mnt_idmap *idmap, struct inode *dir,
2921 struct dentry *victim, bool isdir)
2923 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
2924 struct inode *inode = d_backing_inode(victim);
2927 if (d_is_negative(victim))
2931 BUG_ON(victim->d_parent->d_inode != dir);
2933 /* Inode writeback is not safe when the uid or gid are invalid. */
2934 if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) ||
2935 !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode)))
2938 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2940 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
2946 if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) ||
2947 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2948 HAS_UNMAPPED_ID(idmap, inode))
2951 if (!d_is_dir(victim))
2953 if (IS_ROOT(victim))
2955 } else if (d_is_dir(victim))
2957 if (IS_DEADDIR(dir))
2959 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2964 /* Check whether we can create an object with dentry child in directory
2966 * 1. We can't do it if child already exists (open has special treatment for
2967 * this case, but since we are inlined it's OK)
2968 * 2. We can't do it if dir is read-only (done in permission())
2969 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2970 * 4. We should have write and exec permissions on dir
2971 * 5. We can't do it if dir is immutable (done in permission())
2973 static inline int may_create(struct mnt_idmap *idmap,
2974 struct inode *dir, struct dentry *child)
2976 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2979 if (IS_DEADDIR(dir))
2981 if (!fsuidgid_has_mapping(dir->i_sb, idmap))
2984 return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
2988 * p1 and p2 should be directories on the same fs.
2990 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2995 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2999 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
3001 p = d_ancestor(p2, p1);
3003 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3004 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
3008 p = d_ancestor(p1, p2);
3010 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3011 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
3015 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3016 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3019 EXPORT_SYMBOL(lock_rename);
3021 void unlock_rename(struct dentry *p1, struct dentry *p2)
3023 inode_unlock(p1->d_inode);
3025 inode_unlock(p2->d_inode);
3026 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3029 EXPORT_SYMBOL(unlock_rename);
3032 * mode_strip_umask - handle vfs umask stripping
3033 * @dir: parent directory of the new inode
3034 * @mode: mode of the new inode to be created in @dir
3036 * Umask stripping depends on whether or not the filesystem supports POSIX
3037 * ACLs. If the filesystem doesn't support it umask stripping is done directly
3038 * in here. If the filesystem does support POSIX ACLs umask stripping is
3039 * deferred until the filesystem calls posix_acl_create().
3043 static inline umode_t mode_strip_umask(const struct inode *dir, umode_t mode)
3045 if (!IS_POSIXACL(dir))
3046 mode &= ~current_umask();
3051 * vfs_prepare_mode - prepare the mode to be used for a new inode
3052 * @mnt_userns: user namespace of the mount the inode was found from
3053 * @dir: parent directory of the new inode
3054 * @mode: mode of the new inode
3055 * @mask_perms: allowed permission by the vfs
3056 * @type: type of file to be created
3058 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3059 * object to be created.
3061 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3062 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3063 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3064 * POSIX ACL supporting filesystems.
3066 * Note that it's currently valid for @type to be 0 if a directory is created.
3067 * Filesystems raise that flag individually and we need to check whether each
3068 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3071 * Returns: mode to be passed to the filesystem
3073 static inline umode_t vfs_prepare_mode(struct user_namespace *mnt_userns,
3074 const struct inode *dir, umode_t mode,
3075 umode_t mask_perms, umode_t type)
3077 mode = mode_strip_sgid(mnt_userns, dir, mode);
3078 mode = mode_strip_umask(dir, mode);
3081 * Apply the vfs mandated allowed permission mask and set the type of
3082 * file to be created before we call into the filesystem.
3084 mode &= (mask_perms & ~S_IFMT);
3085 mode |= (type & S_IFMT);
3091 * vfs_create - create new file
3092 * @idmap: idmap of the mount the inode was found from
3093 * @dir: inode of @dentry
3094 * @dentry: pointer to dentry of the base directory
3095 * @mode: mode of the new file
3096 * @want_excl: whether the file must not yet exist
3098 * Create a new file.
3100 * If the inode has been found through an idmapped mount the idmap of
3101 * the vfsmount must be passed through @idmap. This function will then take
3102 * care to map the inode according to @idmap before checking permissions.
3103 * On non-idmapped mounts or if permission checking is to be performed on the
3104 * raw inode simply passs @nop_mnt_idmap.
3106 int vfs_create(struct mnt_idmap *idmap, struct inode *dir,
3107 struct dentry *dentry, umode_t mode, bool want_excl)
3109 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
3112 error = may_create(idmap, dir, dentry);
3116 if (!dir->i_op->create)
3117 return -EACCES; /* shouldn't it be ENOSYS? */
3119 mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IALLUGO, S_IFREG);
3120 error = security_inode_create(dir, dentry, mode);
3123 error = dir->i_op->create(idmap, dir, dentry, mode, want_excl);
3125 fsnotify_create(dir, dentry);
3128 EXPORT_SYMBOL(vfs_create);
3130 int vfs_mkobj(struct dentry *dentry, umode_t mode,
3131 int (*f)(struct dentry *, umode_t, void *),
3134 struct inode *dir = dentry->d_parent->d_inode;
3135 int error = may_create(&nop_mnt_idmap, dir, dentry);
3141 error = security_inode_create(dir, dentry, mode);
3144 error = f(dentry, mode, arg);
3146 fsnotify_create(dir, dentry);
3149 EXPORT_SYMBOL(vfs_mkobj);
3151 bool may_open_dev(const struct path *path)
3153 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3154 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3157 static int may_open(struct mnt_idmap *idmap, const struct path *path,
3158 int acc_mode, int flag)
3160 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
3161 struct dentry *dentry = path->dentry;
3162 struct inode *inode = dentry->d_inode;
3168 switch (inode->i_mode & S_IFMT) {
3172 if (acc_mode & MAY_WRITE)
3174 if (acc_mode & MAY_EXEC)
3179 if (!may_open_dev(path))
3184 if (acc_mode & MAY_EXEC)
3189 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3194 error = inode_permission(idmap, inode, MAY_OPEN | acc_mode);
3199 * An append-only file must be opened in append mode for writing.
3201 if (IS_APPEND(inode)) {
3202 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3208 /* O_NOATIME can only be set by the owner or superuser */
3209 if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode))
3215 static int handle_truncate(struct mnt_idmap *idmap, struct file *filp)
3217 const struct path *path = &filp->f_path;
3218 struct inode *inode = path->dentry->d_inode;
3219 int error = get_write_access(inode);
3223 error = security_file_truncate(filp);
3225 error = do_truncate(idmap, path->dentry, 0,
3226 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3229 put_write_access(inode);
3233 static inline int open_to_namei_flags(int flag)
3235 if ((flag & O_ACCMODE) == 3)
3240 static int may_o_create(struct mnt_idmap *idmap,
3241 const struct path *dir, struct dentry *dentry,
3244 int error = security_path_mknod(dir, dentry, mode, 0);
3248 if (!fsuidgid_has_mapping(dir->dentry->d_sb, idmap))
3251 error = inode_permission(idmap, dir->dentry->d_inode,
3252 MAY_WRITE | MAY_EXEC);
3256 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3260 * Attempt to atomically look up, create and open a file from a negative
3263 * Returns 0 if successful. The file will have been created and attached to
3264 * @file by the filesystem calling finish_open().
3266 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3267 * be set. The caller will need to perform the open themselves. @path will
3268 * have been updated to point to the new dentry. This may be negative.
3270 * Returns an error code otherwise.
3272 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3274 int open_flag, umode_t mode)
3276 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3277 struct inode *dir = nd->path.dentry->d_inode;
3280 if (nd->flags & LOOKUP_DIRECTORY)
3281 open_flag |= O_DIRECTORY;
3283 file->f_path.dentry = DENTRY_NOT_SET;
3284 file->f_path.mnt = nd->path.mnt;
3285 error = dir->i_op->atomic_open(dir, dentry, file,
3286 open_to_namei_flags(open_flag), mode);
3287 d_lookup_done(dentry);
3289 if (file->f_mode & FMODE_OPENED) {
3290 if (unlikely(dentry != file->f_path.dentry)) {
3292 dentry = dget(file->f_path.dentry);
3294 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3297 if (file->f_path.dentry) {
3299 dentry = file->f_path.dentry;
3301 if (unlikely(d_is_negative(dentry)))
3307 dentry = ERR_PTR(error);
3313 * Look up and maybe create and open the last component.
3315 * Must be called with parent locked (exclusive in O_CREAT case).
3317 * Returns 0 on success, that is, if
3318 * the file was successfully atomically created (if necessary) and opened, or
3319 * the file was not completely opened at this time, though lookups and
3320 * creations were performed.
3321 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3322 * In the latter case dentry returned in @path might be negative if O_CREAT
3323 * hadn't been specified.
3325 * An error code is returned on failure.
3327 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3328 const struct open_flags *op,
3331 struct mnt_idmap *idmap;
3332 struct user_namespace *mnt_userns;
3333 struct dentry *dir = nd->path.dentry;
3334 struct inode *dir_inode = dir->d_inode;
3335 int open_flag = op->open_flag;
3336 struct dentry *dentry;
3337 int error, create_error = 0;
3338 umode_t mode = op->mode;
3339 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3341 if (unlikely(IS_DEADDIR(dir_inode)))
3342 return ERR_PTR(-ENOENT);
3344 file->f_mode &= ~FMODE_CREATED;
3345 dentry = d_lookup(dir, &nd->last);
3348 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3352 if (d_in_lookup(dentry))
3355 error = d_revalidate(dentry, nd->flags);
3356 if (likely(error > 0))
3360 d_invalidate(dentry);
3364 if (dentry->d_inode) {
3365 /* Cached positive dentry: will open in f_op->open */
3370 * Checking write permission is tricky, bacuse we don't know if we are
3371 * going to actually need it: O_CREAT opens should work as long as the
3372 * file exists. But checking existence breaks atomicity. The trick is
3373 * to check access and if not granted clear O_CREAT from the flags.
3375 * Another problem is returing the "right" error value (e.g. for an
3376 * O_EXCL open we want to return EEXIST not EROFS).
3378 if (unlikely(!got_write))
3379 open_flag &= ~O_TRUNC;
3380 idmap = mnt_idmap(nd->path.mnt);
3381 mnt_userns = mnt_idmap_owner(idmap);
3382 if (open_flag & O_CREAT) {
3383 if (open_flag & O_EXCL)
3384 open_flag &= ~O_TRUNC;
3385 mode = vfs_prepare_mode(mnt_userns, dir->d_inode, mode, mode, mode);
3386 if (likely(got_write))
3387 create_error = may_o_create(idmap, &nd->path,
3390 create_error = -EROFS;
3393 open_flag &= ~O_CREAT;
3394 if (dir_inode->i_op->atomic_open) {
3395 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3396 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3397 dentry = ERR_PTR(create_error);
3401 if (d_in_lookup(dentry)) {
3402 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3404 d_lookup_done(dentry);
3405 if (unlikely(res)) {
3407 error = PTR_ERR(res);
3415 /* Negative dentry, just create the file */
3416 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3417 file->f_mode |= FMODE_CREATED;
3418 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3419 if (!dir_inode->i_op->create) {
3424 error = dir_inode->i_op->create(idmap, dir_inode, dentry,
3425 mode, open_flag & O_EXCL);
3429 if (unlikely(create_error) && !dentry->d_inode) {
3430 error = create_error;
3437 return ERR_PTR(error);
3440 static const char *open_last_lookups(struct nameidata *nd,
3441 struct file *file, const struct open_flags *op)
3443 struct dentry *dir = nd->path.dentry;
3444 int open_flag = op->open_flag;
3445 bool got_write = false;
3446 struct dentry *dentry;
3449 nd->flags |= op->intent;
3451 if (nd->last_type != LAST_NORM) {
3454 return handle_dots(nd, nd->last_type);
3457 if (!(open_flag & O_CREAT)) {
3458 if (nd->last.name[nd->last.len])
3459 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3460 /* we _can_ be in RCU mode here */
3461 dentry = lookup_fast(nd);
3463 return ERR_CAST(dentry);
3467 BUG_ON(nd->flags & LOOKUP_RCU);
3469 /* create side of things */
3470 if (nd->flags & LOOKUP_RCU) {
3471 if (!try_to_unlazy(nd))
3472 return ERR_PTR(-ECHILD);
3474 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3475 /* trailing slashes? */
3476 if (unlikely(nd->last.name[nd->last.len]))
3477 return ERR_PTR(-EISDIR);
3480 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3481 got_write = !mnt_want_write(nd->path.mnt);
3483 * do _not_ fail yet - we might not need that or fail with
3484 * a different error; let lookup_open() decide; we'll be
3485 * dropping this one anyway.
3488 if (open_flag & O_CREAT)
3489 inode_lock(dir->d_inode);
3491 inode_lock_shared(dir->d_inode);
3492 dentry = lookup_open(nd, file, op, got_write);
3493 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3494 fsnotify_create(dir->d_inode, dentry);
3495 if (open_flag & O_CREAT)
3496 inode_unlock(dir->d_inode);
3498 inode_unlock_shared(dir->d_inode);
3501 mnt_drop_write(nd->path.mnt);
3504 return ERR_CAST(dentry);
3506 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3507 dput(nd->path.dentry);
3508 nd->path.dentry = dentry;
3515 res = step_into(nd, WALK_TRAILING, dentry);
3517 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3522 * Handle the last step of open()
3524 static int do_open(struct nameidata *nd,
3525 struct file *file, const struct open_flags *op)
3527 struct mnt_idmap *idmap;
3528 struct user_namespace *mnt_userns;
3529 int open_flag = op->open_flag;
3534 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3535 error = complete_walk(nd);
3539 if (!(file->f_mode & FMODE_CREATED))
3540 audit_inode(nd->name, nd->path.dentry, 0);
3541 idmap = mnt_idmap(nd->path.mnt);
3542 mnt_userns = mnt_idmap_owner(idmap);
3543 if (open_flag & O_CREAT) {
3544 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3546 if (d_is_dir(nd->path.dentry))
3548 error = may_create_in_sticky(mnt_userns, nd,
3549 d_backing_inode(nd->path.dentry));
3550 if (unlikely(error))
3553 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3556 do_truncate = false;
3557 acc_mode = op->acc_mode;
3558 if (file->f_mode & FMODE_CREATED) {
3559 /* Don't check for write permission, don't truncate */
3560 open_flag &= ~O_TRUNC;
3562 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3563 error = mnt_want_write(nd->path.mnt);
3568 error = may_open(idmap, &nd->path, acc_mode, open_flag);
3569 if (!error && !(file->f_mode & FMODE_OPENED))
3570 error = vfs_open(&nd->path, file);
3572 error = ima_file_check(file, op->acc_mode);
3573 if (!error && do_truncate)
3574 error = handle_truncate(idmap, file);
3575 if (unlikely(error > 0)) {
3580 mnt_drop_write(nd->path.mnt);
3585 * vfs_tmpfile - create tmpfile
3586 * @idmap: idmap of the mount the inode was found from
3587 * @dentry: pointer to dentry of the base directory
3588 * @mode: mode of the new tmpfile
3591 * Create a temporary file.
3593 * If the inode has been found through an idmapped mount the idmap of
3594 * the vfsmount must be passed through @idmap. This function will then take
3595 * care to map the inode according to @idmap before checking permissions.
3596 * On non-idmapped mounts or if permission checking is to be performed on the
3597 * raw inode simply passs @nop_mnt_idmap.
3599 static int vfs_tmpfile(struct mnt_idmap *idmap,
3600 const struct path *parentpath,
3601 struct file *file, umode_t mode)
3603 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
3604 struct dentry *child;
3605 struct inode *dir = d_inode(parentpath->dentry);
3606 struct inode *inode;
3608 int open_flag = file->f_flags;
3610 /* we want directory to be writable */
3611 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3614 if (!dir->i_op->tmpfile)
3616 child = d_alloc(parentpath->dentry, &slash_name);
3617 if (unlikely(!child))
3619 file->f_path.mnt = parentpath->mnt;
3620 file->f_path.dentry = child;
3621 mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode);
3622 error = dir->i_op->tmpfile(idmap, dir, file, mode);
3626 /* Don't check for other permissions, the inode was just created */
3627 error = may_open(idmap, &file->f_path, 0, file->f_flags);
3630 inode = file_inode(file);
3631 if (!(open_flag & O_EXCL)) {
3632 spin_lock(&inode->i_lock);
3633 inode->i_state |= I_LINKABLE;
3634 spin_unlock(&inode->i_lock);
3636 ima_post_create_tmpfile(mnt_userns, inode);
3641 * vfs_tmpfile_open - open a tmpfile for kernel internal use
3642 * @idmap: idmap of the mount the inode was found from
3643 * @parentpath: path of the base directory
3644 * @mode: mode of the new tmpfile
3646 * @cred: credentials for open
3648 * Create and open a temporary file. The file is not accounted in nr_files,
3649 * hence this is only for kernel internal use, and must not be installed into
3650 * file tables or such.
3652 struct file *vfs_tmpfile_open(struct mnt_idmap *idmap,
3653 const struct path *parentpath,
3654 umode_t mode, int open_flag, const struct cred *cred)
3659 file = alloc_empty_file_noaccount(open_flag, cred);
3660 if (!IS_ERR(file)) {
3661 error = vfs_tmpfile(idmap, parentpath, file, mode);
3664 file = ERR_PTR(error);
3669 EXPORT_SYMBOL(vfs_tmpfile_open);
3671 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3672 const struct open_flags *op,
3676 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3678 if (unlikely(error))
3680 error = mnt_want_write(path.mnt);
3681 if (unlikely(error))
3683 error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode);
3686 audit_inode(nd->name, file->f_path.dentry, 0);
3688 mnt_drop_write(path.mnt);
3694 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3697 int error = path_lookupat(nd, flags, &path);
3699 audit_inode(nd->name, path.dentry, 0);
3700 error = vfs_open(&path, file);
3706 static struct file *path_openat(struct nameidata *nd,
3707 const struct open_flags *op, unsigned flags)
3712 file = alloc_empty_file(op->open_flag, current_cred());
3716 if (unlikely(file->f_flags & __O_TMPFILE)) {
3717 error = do_tmpfile(nd, flags, op, file);
3718 } else if (unlikely(file->f_flags & O_PATH)) {
3719 error = do_o_path(nd, flags, file);
3721 const char *s = path_init(nd, flags);
3722 while (!(error = link_path_walk(s, nd)) &&
3723 (s = open_last_lookups(nd, file, op)) != NULL)
3726 error = do_open(nd, file, op);
3729 if (likely(!error)) {
3730 if (likely(file->f_mode & FMODE_OPENED))
3736 if (error == -EOPENSTALE) {
3737 if (flags & LOOKUP_RCU)
3742 return ERR_PTR(error);
3745 struct file *do_filp_open(int dfd, struct filename *pathname,
3746 const struct open_flags *op)
3748 struct nameidata nd;
3749 int flags = op->lookup_flags;
3752 set_nameidata(&nd, dfd, pathname, NULL);
3753 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3754 if (unlikely(filp == ERR_PTR(-ECHILD)))
3755 filp = path_openat(&nd, op, flags);
3756 if (unlikely(filp == ERR_PTR(-ESTALE)))
3757 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3758 restore_nameidata();
3762 struct file *do_file_open_root(const struct path *root,
3763 const char *name, const struct open_flags *op)
3765 struct nameidata nd;
3767 struct filename *filename;
3768 int flags = op->lookup_flags;
3770 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3771 return ERR_PTR(-ELOOP);
3773 filename = getname_kernel(name);
3774 if (IS_ERR(filename))
3775 return ERR_CAST(filename);
3777 set_nameidata(&nd, -1, filename, root);
3778 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3779 if (unlikely(file == ERR_PTR(-ECHILD)))
3780 file = path_openat(&nd, op, flags);
3781 if (unlikely(file == ERR_PTR(-ESTALE)))
3782 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3783 restore_nameidata();
3788 static struct dentry *filename_create(int dfd, struct filename *name,
3789 struct path *path, unsigned int lookup_flags)
3791 struct dentry *dentry = ERR_PTR(-EEXIST);
3793 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3794 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3795 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3800 error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3802 return ERR_PTR(error);
3805 * Yucky last component or no last component at all?
3806 * (foo/., foo/.., /////)
3808 if (unlikely(type != LAST_NORM))
3811 /* don't fail immediately if it's r/o, at least try to report other errors */
3812 err2 = mnt_want_write(path->mnt);
3814 * Do the final lookup. Suppress 'create' if there is a trailing
3815 * '/', and a directory wasn't requested.
3817 if (last.name[last.len] && !want_dir)
3819 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3820 dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags);
3825 if (d_is_positive(dentry))
3829 * Special case - lookup gave negative, but... we had foo/bar/
3830 * From the vfs_mknod() POV we just have a negative dentry -
3831 * all is fine. Let's be bastards - you had / on the end, you've
3832 * been asking for (non-existent) directory. -ENOENT for you.
3834 if (unlikely(!create_flags)) {
3838 if (unlikely(err2)) {
3845 dentry = ERR_PTR(error);
3847 inode_unlock(path->dentry->d_inode);
3849 mnt_drop_write(path->mnt);
3855 struct dentry *kern_path_create(int dfd, const char *pathname,
3856 struct path *path, unsigned int lookup_flags)
3858 struct filename *filename = getname_kernel(pathname);
3859 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3864 EXPORT_SYMBOL(kern_path_create);
3866 void done_path_create(struct path *path, struct dentry *dentry)
3869 inode_unlock(path->dentry->d_inode);
3870 mnt_drop_write(path->mnt);
3873 EXPORT_SYMBOL(done_path_create);
3875 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3876 struct path *path, unsigned int lookup_flags)
3878 struct filename *filename = getname(pathname);
3879 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3884 EXPORT_SYMBOL(user_path_create);
3887 * vfs_mknod - create device node or file
3888 * @idmap: idmap of the mount the inode was found from
3889 * @dir: inode of @dentry
3890 * @dentry: pointer to dentry of the base directory
3891 * @mode: mode of the new device node or file
3892 * @dev: device number of device to create
3894 * Create a device node or file.
3896 * If the inode has been found through an idmapped mount the idmap of
3897 * the vfsmount must be passed through @idmap. This function will then take
3898 * care to map the inode according to @idmap before checking permissions.
3899 * On non-idmapped mounts or if permission checking is to be performed on the
3900 * raw inode simply passs @nop_mnt_idmap.
3902 int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
3903 struct dentry *dentry, umode_t mode, dev_t dev)
3905 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
3906 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3907 int error = may_create(idmap, dir, dentry);
3912 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3913 !capable(CAP_MKNOD))
3916 if (!dir->i_op->mknod)
3919 mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode);
3920 error = devcgroup_inode_mknod(mode, dev);
3924 error = security_inode_mknod(dir, dentry, mode, dev);
3928 error = dir->i_op->mknod(idmap, dir, dentry, mode, dev);
3930 fsnotify_create(dir, dentry);
3933 EXPORT_SYMBOL(vfs_mknod);
3935 static int may_mknod(umode_t mode)
3937 switch (mode & S_IFMT) {
3943 case 0: /* zero mode translates to S_IFREG */
3952 static int do_mknodat(int dfd, struct filename *name, umode_t mode,
3955 struct mnt_idmap *idmap;
3956 struct user_namespace *mnt_userns;
3957 struct dentry *dentry;
3960 unsigned int lookup_flags = 0;
3962 error = may_mknod(mode);
3966 dentry = filename_create(dfd, name, &path, lookup_flags);
3967 error = PTR_ERR(dentry);
3971 error = security_path_mknod(&path, dentry,
3972 mode_strip_umask(path.dentry->d_inode, mode), dev);
3976 idmap = mnt_idmap(path.mnt);
3977 mnt_userns = mnt_idmap_owner(idmap);
3978 switch (mode & S_IFMT) {
3979 case 0: case S_IFREG:
3980 error = vfs_create(idmap, path.dentry->d_inode,
3981 dentry, mode, true);
3983 ima_post_path_mknod(mnt_userns, dentry);
3985 case S_IFCHR: case S_IFBLK:
3986 error = vfs_mknod(idmap, path.dentry->d_inode,
3987 dentry, mode, new_decode_dev(dev));
3989 case S_IFIFO: case S_IFSOCK:
3990 error = vfs_mknod(idmap, path.dentry->d_inode,
3995 done_path_create(&path, dentry);
3996 if (retry_estale(error, lookup_flags)) {
3997 lookup_flags |= LOOKUP_REVAL;
4005 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
4008 return do_mknodat(dfd, getname(filename), mode, dev);
4011 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
4013 return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
4017 * vfs_mkdir - create directory
4018 * @idmap: idmap of the mount the inode was found from
4019 * @dir: inode of @dentry
4020 * @dentry: pointer to dentry of the base directory
4021 * @mode: mode of the new directory
4023 * Create a directory.
4025 * If the inode has been found through an idmapped mount the idmap of
4026 * the vfsmount must be passed through @idmap. This function will then take
4027 * care to map the inode according to @idmap before checking permissions.
4028 * On non-idmapped mounts or if permission checking is to be performed on the
4029 * raw inode simply passs @nop_mnt_idmap.
4031 int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
4032 struct dentry *dentry, umode_t mode)
4034 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
4036 unsigned max_links = dir->i_sb->s_max_links;
4038 error = may_create(idmap, dir, dentry);
4042 if (!dir->i_op->mkdir)
4045 mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IRWXUGO | S_ISVTX, 0);
4046 error = security_inode_mkdir(dir, dentry, mode);
4050 if (max_links && dir->i_nlink >= max_links)
4053 error = dir->i_op->mkdir(idmap, dir, dentry, mode);
4055 fsnotify_mkdir(dir, dentry);
4058 EXPORT_SYMBOL(vfs_mkdir);
4060 int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4062 struct dentry *dentry;
4065 unsigned int lookup_flags = LOOKUP_DIRECTORY;
4068 dentry = filename_create(dfd, name, &path, lookup_flags);
4069 error = PTR_ERR(dentry);
4073 error = security_path_mkdir(&path, dentry,
4074 mode_strip_umask(path.dentry->d_inode, mode));
4076 error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode,
4079 done_path_create(&path, dentry);
4080 if (retry_estale(error, lookup_flags)) {
4081 lookup_flags |= LOOKUP_REVAL;
4089 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4091 return do_mkdirat(dfd, getname(pathname), mode);
4094 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4096 return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4100 * vfs_rmdir - remove directory
4101 * @idmap: idmap of the mount the inode was found from
4102 * @dir: inode of @dentry
4103 * @dentry: pointer to dentry of the base directory
4105 * Remove a directory.
4107 * If the inode has been found through an idmapped mount the idmap of
4108 * the vfsmount must be passed through @idmap. This function will then take
4109 * care to map the inode according to @idmap before checking permissions.
4110 * On non-idmapped mounts or if permission checking is to be performed on the
4111 * raw inode simply passs @nop_mnt_idmap.
4113 int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir,
4114 struct dentry *dentry)
4116 int error = may_delete(idmap, dir, dentry, 1);
4121 if (!dir->i_op->rmdir)
4125 inode_lock(dentry->d_inode);
4128 if (is_local_mountpoint(dentry) ||
4129 (dentry->d_inode->i_flags & S_KERNEL_FILE))
4132 error = security_inode_rmdir(dir, dentry);
4136 error = dir->i_op->rmdir(dir, dentry);
4140 shrink_dcache_parent(dentry);
4141 dentry->d_inode->i_flags |= S_DEAD;
4143 detach_mounts(dentry);
4146 inode_unlock(dentry->d_inode);
4149 d_delete_notify(dir, dentry);
4152 EXPORT_SYMBOL(vfs_rmdir);
4154 int do_rmdir(int dfd, struct filename *name)
4157 struct dentry *dentry;
4161 unsigned int lookup_flags = 0;
4163 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4179 error = mnt_want_write(path.mnt);
4183 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4184 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4185 error = PTR_ERR(dentry);
4188 if (!dentry->d_inode) {
4192 error = security_path_rmdir(&path, dentry);
4195 error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry);
4199 inode_unlock(path.dentry->d_inode);
4200 mnt_drop_write(path.mnt);
4203 if (retry_estale(error, lookup_flags)) {
4204 lookup_flags |= LOOKUP_REVAL;
4212 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4214 return do_rmdir(AT_FDCWD, getname(pathname));
4218 * vfs_unlink - unlink a filesystem object
4219 * @idmap: idmap of the mount the inode was found from
4220 * @dir: parent directory
4222 * @delegated_inode: returns victim inode, if the inode is delegated.
4224 * The caller must hold dir->i_mutex.
4226 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4227 * return a reference to the inode in delegated_inode. The caller
4228 * should then break the delegation on that inode and retry. Because
4229 * breaking a delegation may take a long time, the caller should drop
4230 * dir->i_mutex before doing so.
4232 * Alternatively, a caller may pass NULL for delegated_inode. This may
4233 * be appropriate for callers that expect the underlying filesystem not
4234 * to be NFS exported.
4236 * If the inode has been found through an idmapped mount the idmap of
4237 * the vfsmount must be passed through @idmap. This function will then take
4238 * care to map the inode according to @idmap before checking permissions.
4239 * On non-idmapped mounts or if permission checking is to be performed on the
4240 * raw inode simply passs @nop_mnt_idmap.
4242 int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir,
4243 struct dentry *dentry, struct inode **delegated_inode)
4245 struct inode *target = dentry->d_inode;
4246 int error = may_delete(idmap, dir, dentry, 0);
4251 if (!dir->i_op->unlink)
4255 if (IS_SWAPFILE(target))
4257 else if (is_local_mountpoint(dentry))
4260 error = security_inode_unlink(dir, dentry);
4262 error = try_break_deleg(target, delegated_inode);
4265 error = dir->i_op->unlink(dir, dentry);
4268 detach_mounts(dentry);
4273 inode_unlock(target);
4275 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4276 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4277 fsnotify_unlink(dir, dentry);
4278 } else if (!error) {
4279 fsnotify_link_count(target);
4280 d_delete_notify(dir, dentry);
4285 EXPORT_SYMBOL(vfs_unlink);
4288 * Make sure that the actual truncation of the file will occur outside its
4289 * directory's i_mutex. Truncate can take a long time if there is a lot of
4290 * writeout happening, and we don't want to prevent access to the directory
4291 * while waiting on the I/O.
4293 int do_unlinkat(int dfd, struct filename *name)
4296 struct dentry *dentry;
4300 struct inode *inode = NULL;
4301 struct inode *delegated_inode = NULL;
4302 unsigned int lookup_flags = 0;
4304 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4309 if (type != LAST_NORM)
4312 error = mnt_want_write(path.mnt);
4316 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4317 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4318 error = PTR_ERR(dentry);
4319 if (!IS_ERR(dentry)) {
4321 /* Why not before? Because we want correct error value */
4322 if (last.name[last.len])
4324 inode = dentry->d_inode;
4325 if (d_is_negative(dentry))
4328 error = security_path_unlink(&path, dentry);
4331 error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4332 dentry, &delegated_inode);
4336 inode_unlock(path.dentry->d_inode);
4338 iput(inode); /* truncate the inode here */
4340 if (delegated_inode) {
4341 error = break_deleg_wait(&delegated_inode);
4345 mnt_drop_write(path.mnt);
4348 if (retry_estale(error, lookup_flags)) {
4349 lookup_flags |= LOOKUP_REVAL;
4358 if (d_is_negative(dentry))
4360 else if (d_is_dir(dentry))
4367 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4369 if ((flag & ~AT_REMOVEDIR) != 0)
4372 if (flag & AT_REMOVEDIR)
4373 return do_rmdir(dfd, getname(pathname));
4374 return do_unlinkat(dfd, getname(pathname));
4377 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4379 return do_unlinkat(AT_FDCWD, getname(pathname));
4383 * vfs_symlink - create symlink
4384 * @idmap: idmap of the mount the inode was found from
4385 * @dir: inode of @dentry
4386 * @dentry: pointer to dentry of the base directory
4387 * @oldname: name of the file to link to
4391 * If the inode has been found through an idmapped mount the idmap of
4392 * the vfsmount must be passed through @idmap. This function will then take
4393 * care to map the inode according to @idmap before checking permissions.
4394 * On non-idmapped mounts or if permission checking is to be performed on the
4395 * raw inode simply passs @nop_mnt_idmap.
4397 int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
4398 struct dentry *dentry, const char *oldname)
4402 error = may_create(idmap, dir, dentry);
4406 if (!dir->i_op->symlink)
4409 error = security_inode_symlink(dir, dentry, oldname);
4413 error = dir->i_op->symlink(idmap, dir, dentry, oldname);
4415 fsnotify_create(dir, dentry);
4418 EXPORT_SYMBOL(vfs_symlink);
4420 int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4423 struct dentry *dentry;
4425 unsigned int lookup_flags = 0;
4428 error = PTR_ERR(from);
4432 dentry = filename_create(newdfd, to, &path, lookup_flags);
4433 error = PTR_ERR(dentry);
4437 error = security_path_symlink(&path, dentry, from->name);
4439 error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4440 dentry, from->name);
4441 done_path_create(&path, dentry);
4442 if (retry_estale(error, lookup_flags)) {
4443 lookup_flags |= LOOKUP_REVAL;
4452 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4453 int, newdfd, const char __user *, newname)
4455 return do_symlinkat(getname(oldname), newdfd, getname(newname));
4458 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4460 return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4464 * vfs_link - create a new link
4465 * @old_dentry: object to be linked
4466 * @idmap: idmap of the mount
4468 * @new_dentry: where to create the new link
4469 * @delegated_inode: returns inode needing a delegation break
4471 * The caller must hold dir->i_mutex
4473 * If vfs_link discovers a delegation on the to-be-linked file in need
4474 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4475 * inode in delegated_inode. The caller should then break the delegation
4476 * and retry. Because breaking a delegation may take a long time, the
4477 * caller should drop the i_mutex before doing so.
4479 * Alternatively, a caller may pass NULL for delegated_inode. This may
4480 * be appropriate for callers that expect the underlying filesystem not
4481 * to be NFS exported.
4483 * If the inode has been found through an idmapped mount the idmap of
4484 * the vfsmount must be passed through @idmap. This function will then take
4485 * care to map the inode according to @idmap before checking permissions.
4486 * On non-idmapped mounts or if permission checking is to be performed on the
4487 * raw inode simply passs @nop_mnt_idmap.
4489 int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap,
4490 struct inode *dir, struct dentry *new_dentry,
4491 struct inode **delegated_inode)
4493 struct inode *inode = old_dentry->d_inode;
4494 unsigned max_links = dir->i_sb->s_max_links;
4500 error = may_create(idmap, dir, new_dentry);
4504 if (dir->i_sb != inode->i_sb)
4508 * A link to an append-only or immutable file cannot be created.
4510 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4513 * Updating the link count will likely cause i_uid and i_gid to
4514 * be writen back improperly if their true value is unknown to
4517 if (HAS_UNMAPPED_ID(idmap, inode))
4519 if (!dir->i_op->link)
4521 if (S_ISDIR(inode->i_mode))
4524 error = security_inode_link(old_dentry, dir, new_dentry);
4529 /* Make sure we don't allow creating hardlink to an unlinked file */
4530 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4532 else if (max_links && inode->i_nlink >= max_links)
4535 error = try_break_deleg(inode, delegated_inode);
4537 error = dir->i_op->link(old_dentry, dir, new_dentry);
4540 if (!error && (inode->i_state & I_LINKABLE)) {
4541 spin_lock(&inode->i_lock);
4542 inode->i_state &= ~I_LINKABLE;
4543 spin_unlock(&inode->i_lock);
4545 inode_unlock(inode);
4547 fsnotify_link(dir, inode, new_dentry);
4550 EXPORT_SYMBOL(vfs_link);
4553 * Hardlinks are often used in delicate situations. We avoid
4554 * security-related surprises by not following symlinks on the
4557 * We don't follow them on the oldname either to be compatible
4558 * with linux 2.0, and to avoid hard-linking to directories
4559 * and other special files. --ADM
4561 int do_linkat(int olddfd, struct filename *old, int newdfd,
4562 struct filename *new, int flags)
4564 struct mnt_idmap *idmap;
4565 struct dentry *new_dentry;
4566 struct path old_path, new_path;
4567 struct inode *delegated_inode = NULL;
4571 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4576 * To use null names we require CAP_DAC_READ_SEARCH
4577 * This ensures that not everyone will be able to create
4578 * handlink using the passed filedescriptor.
4580 if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4585 if (flags & AT_SYMLINK_FOLLOW)
4586 how |= LOOKUP_FOLLOW;
4588 error = filename_lookup(olddfd, old, how, &old_path, NULL);
4592 new_dentry = filename_create(newdfd, new, &new_path,
4593 (how & LOOKUP_REVAL));
4594 error = PTR_ERR(new_dentry);
4595 if (IS_ERR(new_dentry))
4599 if (old_path.mnt != new_path.mnt)
4601 idmap = mnt_idmap(new_path.mnt);
4602 error = may_linkat(idmap, &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;
4705 if (source == target)
4708 error = may_delete(rd->old_mnt_idmap, old_dir, old_dentry, is_dir);
4713 error = may_create(rd->new_mnt_idmap, new_dir, new_dentry);
4715 new_is_dir = d_is_dir(new_dentry);
4717 if (!(flags & RENAME_EXCHANGE))
4718 error = may_delete(rd->new_mnt_idmap, new_dir,
4719 new_dentry, is_dir);
4721 error = may_delete(rd->new_mnt_idmap, new_dir,
4722 new_dentry, new_is_dir);
4727 if (!old_dir->i_op->rename)
4731 * If we are going to change the parent - check write permissions,
4732 * we'll need to flip '..'.
4734 if (new_dir != old_dir) {
4736 error = inode_permission(rd->old_mnt_idmap, source,
4741 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4742 error = inode_permission(rd->new_mnt_idmap, target,
4749 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4754 take_dentry_name_snapshot(&old_name, old_dentry);
4756 if (!is_dir || (flags & RENAME_EXCHANGE))
4757 lock_two_nondirectories(source, target);
4762 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4766 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4769 if (max_links && new_dir != old_dir) {
4771 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4773 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4774 old_dir->i_nlink >= max_links)
4778 error = try_break_deleg(source, delegated_inode);
4782 if (target && !new_is_dir) {
4783 error = try_break_deleg(target, delegated_inode);
4787 error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry,
4788 new_dir, new_dentry, flags);
4792 if (!(flags & RENAME_EXCHANGE) && target) {
4794 shrink_dcache_parent(new_dentry);
4795 target->i_flags |= S_DEAD;
4797 dont_mount(new_dentry);
4798 detach_mounts(new_dentry);
4800 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4801 if (!(flags & RENAME_EXCHANGE))
4802 d_move(old_dentry, new_dentry);
4804 d_exchange(old_dentry, new_dentry);
4807 if (!is_dir || (flags & RENAME_EXCHANGE))
4808 unlock_two_nondirectories(source, target);
4810 inode_unlock(target);
4813 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4814 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4815 if (flags & RENAME_EXCHANGE) {
4816 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4817 new_is_dir, NULL, new_dentry);
4820 release_dentry_name_snapshot(&old_name);
4824 EXPORT_SYMBOL(vfs_rename);
4826 int do_renameat2(int olddfd, struct filename *from, int newdfd,
4827 struct filename *to, unsigned int flags)
4829 struct renamedata rd;
4830 struct dentry *old_dentry, *new_dentry;
4831 struct dentry *trap;
4832 struct path old_path, new_path;
4833 struct qstr old_last, new_last;
4834 int old_type, new_type;
4835 struct inode *delegated_inode = NULL;
4836 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4837 bool should_retry = false;
4838 int error = -EINVAL;
4840 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4843 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4844 (flags & RENAME_EXCHANGE))
4847 if (flags & RENAME_EXCHANGE)
4851 error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4852 &old_last, &old_type);
4856 error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4862 if (old_path.mnt != new_path.mnt)
4866 if (old_type != LAST_NORM)
4869 if (flags & RENAME_NOREPLACE)
4871 if (new_type != LAST_NORM)
4874 error = mnt_want_write(old_path.mnt);
4879 trap = lock_rename(new_path.dentry, old_path.dentry);
4881 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4882 error = PTR_ERR(old_dentry);
4883 if (IS_ERR(old_dentry))
4885 /* source must exist */
4887 if (d_is_negative(old_dentry))
4889 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4890 error = PTR_ERR(new_dentry);
4891 if (IS_ERR(new_dentry))
4894 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4896 if (flags & RENAME_EXCHANGE) {
4898 if (d_is_negative(new_dentry))
4901 if (!d_is_dir(new_dentry)) {
4903 if (new_last.name[new_last.len])
4907 /* unless the source is a directory trailing slashes give -ENOTDIR */
4908 if (!d_is_dir(old_dentry)) {
4910 if (old_last.name[old_last.len])
4912 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4915 /* source should not be ancestor of target */
4917 if (old_dentry == trap)
4919 /* target should not be an ancestor of source */
4920 if (!(flags & RENAME_EXCHANGE))
4922 if (new_dentry == trap)
4925 error = security_path_rename(&old_path, old_dentry,
4926 &new_path, new_dentry, flags);
4930 rd.old_dir = old_path.dentry->d_inode;
4931 rd.old_dentry = old_dentry;
4932 rd.old_mnt_idmap = mnt_idmap(old_path.mnt);
4933 rd.new_dir = new_path.dentry->d_inode;
4934 rd.new_dentry = new_dentry;
4935 rd.new_mnt_idmap = mnt_idmap(new_path.mnt);
4936 rd.delegated_inode = &delegated_inode;
4938 error = vfs_rename(&rd);
4944 unlock_rename(new_path.dentry, old_path.dentry);
4945 if (delegated_inode) {
4946 error = break_deleg_wait(&delegated_inode);
4950 mnt_drop_write(old_path.mnt);
4952 if (retry_estale(error, lookup_flags))
4953 should_retry = true;
4954 path_put(&new_path);
4956 path_put(&old_path);
4958 should_retry = false;
4959 lookup_flags |= LOOKUP_REVAL;
4968 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4969 int, newdfd, const char __user *, newname, unsigned int, flags)
4971 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4975 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4976 int, newdfd, const char __user *, newname)
4978 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4982 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4984 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4985 getname(newname), 0);
4988 int readlink_copy(char __user *buffer, int buflen, const char *link)
4990 int len = PTR_ERR(link);
4995 if (len > (unsigned) buflen)
4997 if (copy_to_user(buffer, link, len))
5004 * vfs_readlink - copy symlink body into userspace buffer
5005 * @dentry: dentry on which to get symbolic link
5006 * @buffer: user memory pointer
5007 * @buflen: size of buffer
5009 * Does not touch atime. That's up to the caller if necessary
5011 * Does not call security hook.
5013 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5015 struct inode *inode = d_inode(dentry);
5016 DEFINE_DELAYED_CALL(done);
5020 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5021 if (unlikely(inode->i_op->readlink))
5022 return inode->i_op->readlink(dentry, buffer, buflen);
5024 if (!d_is_symlink(dentry))
5027 spin_lock(&inode->i_lock);
5028 inode->i_opflags |= IOP_DEFAULT_READLINK;
5029 spin_unlock(&inode->i_lock);
5032 link = READ_ONCE(inode->i_link);
5034 link = inode->i_op->get_link(dentry, inode, &done);
5036 return PTR_ERR(link);
5038 res = readlink_copy(buffer, buflen, link);
5039 do_delayed_call(&done);
5042 EXPORT_SYMBOL(vfs_readlink);
5045 * vfs_get_link - get symlink body
5046 * @dentry: dentry on which to get symbolic link
5047 * @done: caller needs to free returned data with this
5049 * Calls security hook and i_op->get_link() on the supplied inode.
5051 * It does not touch atime. That's up to the caller if necessary.
5053 * Does not work on "special" symlinks like /proc/$$/fd/N
5055 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5057 const char *res = ERR_PTR(-EINVAL);
5058 struct inode *inode = d_inode(dentry);
5060 if (d_is_symlink(dentry)) {
5061 res = ERR_PTR(security_inode_readlink(dentry));
5063 res = inode->i_op->get_link(dentry, inode, done);
5067 EXPORT_SYMBOL(vfs_get_link);
5069 /* get the link contents into pagecache */
5070 const char *page_get_link(struct dentry *dentry, struct inode *inode,
5071 struct delayed_call *callback)
5075 struct address_space *mapping = inode->i_mapping;
5078 page = find_get_page(mapping, 0);
5080 return ERR_PTR(-ECHILD);
5081 if (!PageUptodate(page)) {
5083 return ERR_PTR(-ECHILD);
5086 page = read_mapping_page(mapping, 0, NULL);
5090 set_delayed_call(callback, page_put_link, page);
5091 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5092 kaddr = page_address(page);
5093 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5097 EXPORT_SYMBOL(page_get_link);
5099 void page_put_link(void *arg)
5103 EXPORT_SYMBOL(page_put_link);
5105 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5107 DEFINE_DELAYED_CALL(done);
5108 int res = readlink_copy(buffer, buflen,
5109 page_get_link(dentry, d_inode(dentry),
5111 do_delayed_call(&done);
5114 EXPORT_SYMBOL(page_readlink);
5116 int page_symlink(struct inode *inode, const char *symname, int len)
5118 struct address_space *mapping = inode->i_mapping;
5119 const struct address_space_operations *aops = mapping->a_ops;
5120 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5122 void *fsdata = NULL;
5128 flags = memalloc_nofs_save();
5129 err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5131 memalloc_nofs_restore(flags);
5135 memcpy(page_address(page), symname, len-1);
5137 err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5144 mark_inode_dirty(inode);
5149 EXPORT_SYMBOL(page_symlink);
5151 const struct inode_operations page_symlink_inode_operations = {
5152 .get_link = page_get_link,
5154 EXPORT_SYMBOL(page_symlink_inode_operations);