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/filelock.h>
24 #include <linux/namei.h>
25 #include <linux/pagemap.h>
26 #include <linux/sched/mm.h>
27 #include <linux/fsnotify.h>
28 #include <linux/personality.h>
29 #include <linux/security.h>
30 #include <linux/ima.h>
31 #include <linux/syscalls.h>
32 #include <linux/mount.h>
33 #include <linux/audit.h>
34 #include <linux/capability.h>
35 #include <linux/file.h>
36 #include <linux/fcntl.h>
37 #include <linux/device_cgroup.h>
38 #include <linux/fs_struct.h>
39 #include <linux/posix_acl.h>
40 #include <linux/hash.h>
41 #include <linux/bitops.h>
42 #include <linux/init_task.h>
43 #include <linux/uaccess.h>
48 /* [Feb-1997 T. Schoebel-Theuer]
49 * Fundamental changes in the pathname lookup mechanisms (namei)
50 * were necessary because of omirr. The reason is that omirr needs
51 * to know the _real_ pathname, not the user-supplied one, in case
52 * of symlinks (and also when transname replacements occur).
54 * The new code replaces the old recursive symlink resolution with
55 * an iterative one (in case of non-nested symlink chains). It does
56 * this with calls to <fs>_follow_link().
57 * As a side effect, dir_namei(), _namei() and follow_link() are now
58 * replaced with a single function lookup_dentry() that can handle all
59 * the special cases of the former code.
61 * With the new dcache, the pathname is stored at each inode, at least as
62 * long as the refcount of the inode is positive. As a side effect, the
63 * size of the dcache depends on the inode cache and thus is dynamic.
65 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
66 * resolution to correspond with current state of the code.
68 * Note that the symlink resolution is not *completely* iterative.
69 * There is still a significant amount of tail- and mid- recursion in
70 * the algorithm. Also, note that <fs>_readlink() is not used in
71 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
72 * may return different results than <fs>_follow_link(). Many virtual
73 * filesystems (including /proc) exhibit this behavior.
76 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
77 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
78 * and the name already exists in form of a symlink, try to create the new
79 * name indicated by the symlink. The old code always complained that the
80 * name already exists, due to not following the symlink even if its target
81 * is nonexistent. The new semantics affects also mknod() and link() when
82 * the name is a symlink pointing to a non-existent name.
84 * I don't know which semantics is the right one, since I have no access
85 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
86 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
87 * "old" one. Personally, I think the new semantics is much more logical.
88 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
89 * file does succeed in both HP-UX and SunOs, but not in Solaris
90 * and in the old Linux semantics.
93 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
94 * semantics. See the comments in "open_namei" and "do_link" below.
96 * [10-Sep-98 Alan Modra] Another symlink change.
99 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
100 * inside the path - always follow.
101 * in the last component in creation/removal/renaming - never follow.
102 * if LOOKUP_FOLLOW passed - follow.
103 * if the pathname has trailing slashes - follow.
104 * otherwise - don't follow.
105 * (applied in that order).
107 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
108 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
109 * During the 2.4 we need to fix the userland stuff depending on it -
110 * hopefully we will be able to get rid of that wart in 2.5. So far only
111 * XEmacs seems to be relying on it...
114 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
115 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
116 * any extra contention...
119 /* In order to reduce some races, while at the same time doing additional
120 * checking and hopefully speeding things up, we copy filenames to the
121 * kernel data space before using them..
123 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
124 * PATH_MAX includes the nul terminator --RR.
127 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
130 getname_flags(const char __user *filename, int flags, int *empty)
132 struct filename *result;
136 result = audit_reusename(filename);
140 result = __getname();
141 if (unlikely(!result))
142 return ERR_PTR(-ENOMEM);
145 * First, try to embed the struct filename inside the names_cache
148 kname = (char *)result->iname;
149 result->name = kname;
151 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
152 if (unlikely(len < 0)) {
158 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
159 * separate struct filename so we can dedicate the entire
160 * names_cache allocation for the pathname, and re-do the copy from
163 if (unlikely(len == EMBEDDED_NAME_MAX)) {
164 const size_t size = offsetof(struct filename, iname[1]);
165 kname = (char *)result;
168 * size is chosen that way we to guarantee that
169 * result->iname[0] is within the same object and that
170 * kname can't be equal to result->iname, no matter what.
172 result = kzalloc(size, GFP_KERNEL);
173 if (unlikely(!result)) {
175 return ERR_PTR(-ENOMEM);
177 result->name = kname;
178 len = strncpy_from_user(kname, filename, PATH_MAX);
179 if (unlikely(len < 0)) {
184 if (unlikely(len == PATH_MAX)) {
187 return ERR_PTR(-ENAMETOOLONG);
192 /* The empty path is special. */
193 if (unlikely(!len)) {
196 if (!(flags & LOOKUP_EMPTY)) {
198 return ERR_PTR(-ENOENT);
202 result->uptr = filename;
203 result->aname = NULL;
204 audit_getname(result);
209 getname_uflags(const char __user *filename, int uflags)
211 int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
213 return getname_flags(filename, flags, NULL);
217 getname(const char __user * filename)
219 return getname_flags(filename, 0, NULL);
223 getname_kernel(const char * filename)
225 struct filename *result;
226 int len = strlen(filename) + 1;
228 result = __getname();
229 if (unlikely(!result))
230 return ERR_PTR(-ENOMEM);
232 if (len <= EMBEDDED_NAME_MAX) {
233 result->name = (char *)result->iname;
234 } else if (len <= PATH_MAX) {
235 const size_t size = offsetof(struct filename, iname[1]);
236 struct filename *tmp;
238 tmp = kmalloc(size, GFP_KERNEL);
239 if (unlikely(!tmp)) {
241 return ERR_PTR(-ENOMEM);
243 tmp->name = (char *)result;
247 return ERR_PTR(-ENAMETOOLONG);
249 memcpy((char *)result->name, filename, len);
251 result->aname = NULL;
253 audit_getname(result);
258 void putname(struct filename *name)
263 BUG_ON(name->refcnt <= 0);
265 if (--name->refcnt > 0)
268 if (name->name != name->iname) {
269 __putname(name->name);
276 * check_acl - perform ACL permission checking
277 * @idmap: idmap of the mount the inode was found from
278 * @inode: inode to check permissions on
279 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
281 * This function performs the ACL permission checking. Since this function
282 * retrieve POSIX acls it needs to know whether it is called from a blocking or
283 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
285 * If the inode has been found through an idmapped mount the idmap of
286 * the vfsmount must be passed through @idmap. This function will then take
287 * care to map the inode according to @idmap before checking permissions.
288 * On non-idmapped mounts or if permission checking is to be performed on the
289 * raw inode simply passs @nop_mnt_idmap.
291 static int check_acl(struct mnt_idmap *idmap,
292 struct inode *inode, int mask)
294 #ifdef CONFIG_FS_POSIX_ACL
295 struct posix_acl *acl;
297 if (mask & MAY_NOT_BLOCK) {
298 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
301 /* no ->get_inode_acl() calls in RCU mode... */
302 if (is_uncached_acl(acl))
304 return posix_acl_permission(idmap, inode, acl, mask);
307 acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
311 int error = posix_acl_permission(idmap, inode, acl, mask);
312 posix_acl_release(acl);
321 * acl_permission_check - perform basic UNIX permission checking
322 * @idmap: idmap of the mount the inode was found from
323 * @inode: inode to check permissions on
324 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
326 * This function performs the basic UNIX permission checking. Since this
327 * function may retrieve POSIX acls it needs to know whether it is called from a
328 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
330 * If the inode has been found through an idmapped mount the idmap of
331 * the vfsmount must be passed through @idmap. This function will then take
332 * care to map the inode according to @idmap before checking permissions.
333 * On non-idmapped mounts or if permission checking is to be performed on the
334 * raw inode simply passs @nop_mnt_idmap.
336 static int acl_permission_check(struct mnt_idmap *idmap,
337 struct inode *inode, int mask)
339 unsigned int mode = inode->i_mode;
342 /* Are we the owner? If so, ACL's don't matter */
343 vfsuid = i_uid_into_vfsuid(idmap, inode);
344 if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) {
347 return (mask & ~mode) ? -EACCES : 0;
350 /* Do we have ACL's? */
351 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
352 int error = check_acl(idmap, inode, mask);
353 if (error != -EAGAIN)
357 /* Only RWX matters for group/other mode bits */
361 * Are the group permissions different from
362 * the other permissions in the bits we care
363 * about? Need to check group ownership if so.
365 if (mask & (mode ^ (mode >> 3))) {
366 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
367 if (vfsgid_in_group_p(vfsgid))
371 /* Bits in 'mode' clear that we require? */
372 return (mask & ~mode) ? -EACCES : 0;
376 * generic_permission - check for access rights on a Posix-like filesystem
377 * @idmap: idmap of the mount the inode was found from
378 * @inode: inode to check access rights for
379 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
380 * %MAY_NOT_BLOCK ...)
382 * Used to check for read/write/execute permissions on a file.
383 * We use "fsuid" for this, letting us set arbitrary permissions
384 * for filesystem access without changing the "normal" uids which
385 * are used for other things.
387 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
388 * request cannot be satisfied (eg. requires blocking or too much complexity).
389 * It would then be called again in ref-walk mode.
391 * If the inode has been found through an idmapped mount the idmap of
392 * the vfsmount must be passed through @idmap. This function will then take
393 * care to map the inode according to @idmap before checking permissions.
394 * On non-idmapped mounts or if permission checking is to be performed on the
395 * raw inode simply passs @nop_mnt_idmap.
397 int generic_permission(struct mnt_idmap *idmap, struct inode *inode,
403 * Do the basic permission checks.
405 ret = acl_permission_check(idmap, 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(idmap, inode,
413 CAP_DAC_READ_SEARCH))
415 if (capable_wrt_inode_uidgid(idmap, 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(idmap, 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(idmap, 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 mnt_idmap *idmap;
1101 if (!sysctl_protected_symlinks)
1104 idmap = mnt_idmap(nd->path.mnt);
1105 vfsuid = i_uid_into_vfsuid(idmap, 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 inode *inode = link->dentry->d_inode;
1186 /* Inode writeback is not safe when the uid or gid are invalid. */
1187 if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
1188 !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
1191 if (!sysctl_protected_hardlinks)
1194 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1195 * otherwise, it must be a safe source.
1197 if (safe_hardlink_source(idmap, inode) ||
1198 inode_owner_or_capable(idmap, inode))
1201 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1206 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1207 * should be allowed, or not, on files that already
1209 * @idmap: idmap of the mount the inode was found from
1210 * @nd: nameidata pathwalk data
1211 * @inode: the inode of the file to open
1213 * Block an O_CREAT open of a FIFO (or a regular file) when:
1214 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1215 * - the file already exists
1216 * - we are in a sticky directory
1217 * - we don't own the file
1218 * - the owner of the directory doesn't own the file
1219 * - the directory is world writable
1220 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1221 * the directory doesn't have to be world writable: being group writable will
1224 * If the inode has been found through an idmapped mount the idmap of
1225 * the vfsmount must be passed through @idmap. This function will then take
1226 * care to map the inode according to @idmap before checking permissions.
1227 * On non-idmapped mounts or if permission checking is to be performed on the
1228 * raw inode simply pass @nop_mnt_idmap.
1230 * Returns 0 if the open is allowed, -ve on error.
1232 static int may_create_in_sticky(struct mnt_idmap *idmap,
1233 struct nameidata *nd, struct inode *const inode)
1235 umode_t dir_mode = nd->dir_mode;
1236 vfsuid_t dir_vfsuid = nd->dir_vfsuid;
1238 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1239 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1240 likely(!(dir_mode & S_ISVTX)) ||
1241 vfsuid_eq(i_uid_into_vfsuid(idmap, inode), dir_vfsuid) ||
1242 vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), current_fsuid()))
1245 if (likely(dir_mode & 0002) ||
1247 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1248 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1249 const char *operation = S_ISFIFO(inode->i_mode) ?
1250 "sticky_create_fifo" :
1251 "sticky_create_regular";
1252 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1259 * follow_up - Find the mountpoint of path's vfsmount
1261 * Given a path, find the mountpoint of its source file system.
1262 * Replace @path with the path of the mountpoint in the parent mount.
1265 * Return 1 if we went up a level and 0 if we were already at the
1268 int follow_up(struct path *path)
1270 struct mount *mnt = real_mount(path->mnt);
1271 struct mount *parent;
1272 struct dentry *mountpoint;
1274 read_seqlock_excl(&mount_lock);
1275 parent = mnt->mnt_parent;
1276 if (parent == mnt) {
1277 read_sequnlock_excl(&mount_lock);
1280 mntget(&parent->mnt);
1281 mountpoint = dget(mnt->mnt_mountpoint);
1282 read_sequnlock_excl(&mount_lock);
1284 path->dentry = mountpoint;
1286 path->mnt = &parent->mnt;
1289 EXPORT_SYMBOL(follow_up);
1291 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1292 struct path *path, unsigned *seqp)
1294 while (mnt_has_parent(m)) {
1295 struct dentry *mountpoint = m->mnt_mountpoint;
1298 if (unlikely(root->dentry == mountpoint &&
1299 root->mnt == &m->mnt))
1301 if (mountpoint != m->mnt.mnt_root) {
1302 path->mnt = &m->mnt;
1303 path->dentry = mountpoint;
1304 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1311 static bool choose_mountpoint(struct mount *m, const struct path *root,
1318 unsigned seq, mseq = read_seqbegin(&mount_lock);
1320 found = choose_mountpoint_rcu(m, root, path, &seq);
1321 if (unlikely(!found)) {
1322 if (!read_seqretry(&mount_lock, mseq))
1325 if (likely(__legitimize_path(path, seq, mseq)))
1337 * Perform an automount
1338 * - return -EISDIR to tell follow_managed() to stop and return the path we
1341 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1343 struct dentry *dentry = path->dentry;
1345 /* We don't want to mount if someone's just doing a stat -
1346 * unless they're stat'ing a directory and appended a '/' to
1349 * We do, however, want to mount if someone wants to open or
1350 * create a file of any type under the mountpoint, wants to
1351 * traverse through the mountpoint or wants to open the
1352 * mounted directory. Also, autofs may mark negative dentries
1353 * as being automount points. These will need the attentions
1354 * of the daemon to instantiate them before they can be used.
1356 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1357 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1361 if (count && (*count)++ >= MAXSYMLINKS)
1364 return finish_automount(dentry->d_op->d_automount(path), path);
1368 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1369 * dentries are pinned but not locked here, so negative dentry can go
1370 * positive right under us. Use of smp_load_acquire() provides a barrier
1371 * sufficient for ->d_inode and ->d_flags consistency.
1373 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1374 int *count, unsigned lookup_flags)
1376 struct vfsmount *mnt = path->mnt;
1377 bool need_mntput = false;
1380 while (flags & DCACHE_MANAGED_DENTRY) {
1381 /* Allow the filesystem to manage the transit without i_mutex
1383 if (flags & DCACHE_MANAGE_TRANSIT) {
1384 ret = path->dentry->d_op->d_manage(path, false);
1385 flags = smp_load_acquire(&path->dentry->d_flags);
1390 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1391 struct vfsmount *mounted = lookup_mnt(path);
1392 if (mounted) { // ... in our namespace
1396 path->mnt = mounted;
1397 path->dentry = dget(mounted->mnt_root);
1398 // here we know it's positive
1399 flags = path->dentry->d_flags;
1405 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1408 // uncovered automount point
1409 ret = follow_automount(path, count, lookup_flags);
1410 flags = smp_load_acquire(&path->dentry->d_flags);
1417 // possible if you race with several mount --move
1418 if (need_mntput && path->mnt == mnt)
1420 if (!ret && unlikely(d_flags_negative(flags)))
1422 *jumped = need_mntput;
1426 static inline int traverse_mounts(struct path *path, bool *jumped,
1427 int *count, unsigned lookup_flags)
1429 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1432 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1434 if (unlikely(d_flags_negative(flags)))
1438 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1441 int follow_down_one(struct path *path)
1443 struct vfsmount *mounted;
1445 mounted = lookup_mnt(path);
1449 path->mnt = mounted;
1450 path->dentry = dget(mounted->mnt_root);
1455 EXPORT_SYMBOL(follow_down_one);
1458 * Follow down to the covering mount currently visible to userspace. At each
1459 * point, the filesystem owning that dentry may be queried as to whether the
1460 * caller is permitted to proceed or not.
1462 int follow_down(struct path *path, unsigned int flags)
1464 struct vfsmount *mnt = path->mnt;
1466 int ret = traverse_mounts(path, &jumped, NULL, flags);
1468 if (path->mnt != mnt)
1472 EXPORT_SYMBOL(follow_down);
1475 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1476 * we meet a managed dentry that would need blocking.
1478 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1480 struct dentry *dentry = path->dentry;
1481 unsigned int flags = dentry->d_flags;
1483 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1486 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1491 * Don't forget we might have a non-mountpoint managed dentry
1492 * that wants to block transit.
1494 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1495 int res = dentry->d_op->d_manage(path, true);
1497 return res == -EISDIR;
1498 flags = dentry->d_flags;
1501 if (flags & DCACHE_MOUNTED) {
1502 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1504 path->mnt = &mounted->mnt;
1505 dentry = path->dentry = mounted->mnt.mnt_root;
1506 nd->state |= ND_JUMPED;
1507 nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1508 flags = dentry->d_flags;
1509 // makes sure that non-RCU pathwalk could reach
1511 if (read_seqretry(&mount_lock, nd->m_seq))
1515 if (read_seqretry(&mount_lock, nd->m_seq))
1518 return !(flags & DCACHE_NEED_AUTOMOUNT);
1522 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1528 path->mnt = nd->path.mnt;
1529 path->dentry = dentry;
1530 if (nd->flags & LOOKUP_RCU) {
1531 unsigned int seq = nd->next_seq;
1532 if (likely(__follow_mount_rcu(nd, path)))
1534 // *path and nd->next_seq might've been clobbered
1535 path->mnt = nd->path.mnt;
1536 path->dentry = dentry;
1538 if (!try_to_unlazy_next(nd, dentry))
1541 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1543 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1546 nd->state |= ND_JUMPED;
1548 if (unlikely(ret)) {
1550 if (path->mnt != nd->path.mnt)
1557 * This looks up the name in dcache and possibly revalidates the found dentry.
1558 * NULL is returned if the dentry does not exist in the cache.
1560 static struct dentry *lookup_dcache(const struct qstr *name,
1564 struct dentry *dentry = d_lookup(dir, name);
1566 int error = d_revalidate(dentry, flags);
1567 if (unlikely(error <= 0)) {
1569 d_invalidate(dentry);
1571 return ERR_PTR(error);
1578 * Parent directory has inode locked exclusive. This is one
1579 * and only case when ->lookup() gets called on non in-lookup
1580 * dentries - as the matter of fact, this only gets called
1581 * when directory is guaranteed to have no in-lookup children
1584 static struct dentry *__lookup_hash(const struct qstr *name,
1585 struct dentry *base, unsigned int flags)
1587 struct dentry *dentry = lookup_dcache(name, base, flags);
1589 struct inode *dir = base->d_inode;
1594 /* Don't create child dentry for a dead directory. */
1595 if (unlikely(IS_DEADDIR(dir)))
1596 return ERR_PTR(-ENOENT);
1598 dentry = d_alloc(base, name);
1599 if (unlikely(!dentry))
1600 return ERR_PTR(-ENOMEM);
1602 old = dir->i_op->lookup(dir, dentry, flags);
1603 if (unlikely(old)) {
1610 static struct dentry *lookup_fast(struct nameidata *nd)
1612 struct dentry *dentry, *parent = nd->path.dentry;
1616 * Rename seqlock is not required here because in the off chance
1617 * of a false negative due to a concurrent rename, the caller is
1618 * going to fall back to non-racy lookup.
1620 if (nd->flags & LOOKUP_RCU) {
1621 dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq);
1622 if (unlikely(!dentry)) {
1623 if (!try_to_unlazy(nd))
1624 return ERR_PTR(-ECHILD);
1629 * This sequence count validates that the parent had no
1630 * changes while we did the lookup of the dentry above.
1632 if (read_seqcount_retry(&parent->d_seq, nd->seq))
1633 return ERR_PTR(-ECHILD);
1635 status = d_revalidate(dentry, nd->flags);
1636 if (likely(status > 0))
1638 if (!try_to_unlazy_next(nd, dentry))
1639 return ERR_PTR(-ECHILD);
1640 if (status == -ECHILD)
1641 /* we'd been told to redo it in non-rcu mode */
1642 status = d_revalidate(dentry, nd->flags);
1644 dentry = __d_lookup(parent, &nd->last);
1645 if (unlikely(!dentry))
1647 status = d_revalidate(dentry, nd->flags);
1649 if (unlikely(status <= 0)) {
1651 d_invalidate(dentry);
1653 return ERR_PTR(status);
1658 /* Fast lookup failed, do it the slow way */
1659 static struct dentry *__lookup_slow(const struct qstr *name,
1663 struct dentry *dentry, *old;
1664 struct inode *inode = dir->d_inode;
1665 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1667 /* Don't go there if it's already dead */
1668 if (unlikely(IS_DEADDIR(inode)))
1669 return ERR_PTR(-ENOENT);
1671 dentry = d_alloc_parallel(dir, name, &wq);
1674 if (unlikely(!d_in_lookup(dentry))) {
1675 int error = d_revalidate(dentry, flags);
1676 if (unlikely(error <= 0)) {
1678 d_invalidate(dentry);
1683 dentry = ERR_PTR(error);
1686 old = inode->i_op->lookup(inode, dentry, flags);
1687 d_lookup_done(dentry);
1688 if (unlikely(old)) {
1696 static struct dentry *lookup_slow(const struct qstr *name,
1700 struct inode *inode = dir->d_inode;
1702 inode_lock_shared(inode);
1703 res = __lookup_slow(name, dir, flags);
1704 inode_unlock_shared(inode);
1708 static inline int may_lookup(struct mnt_idmap *idmap,
1709 struct nameidata *nd)
1711 if (nd->flags & LOOKUP_RCU) {
1712 int err = inode_permission(idmap, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1713 if (err != -ECHILD || !try_to_unlazy(nd))
1716 return inode_permission(idmap, nd->inode, MAY_EXEC);
1719 static int reserve_stack(struct nameidata *nd, struct path *link)
1721 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1724 if (likely(nd->depth != EMBEDDED_LEVELS))
1726 if (likely(nd->stack != nd->internal))
1728 if (likely(nd_alloc_stack(nd)))
1731 if (nd->flags & LOOKUP_RCU) {
1732 // we need to grab link before we do unlazy. And we can't skip
1733 // unlazy even if we fail to grab the link - cleanup needs it
1734 bool grabbed_link = legitimize_path(nd, link, nd->next_seq);
1736 if (!try_to_unlazy(nd) || !grabbed_link)
1739 if (nd_alloc_stack(nd))
1745 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1747 static const char *pick_link(struct nameidata *nd, struct path *link,
1748 struct inode *inode, int flags)
1752 int error = reserve_stack(nd, link);
1754 if (unlikely(error)) {
1755 if (!(nd->flags & LOOKUP_RCU))
1757 return ERR_PTR(error);
1759 last = nd->stack + nd->depth++;
1761 clear_delayed_call(&last->done);
1762 last->seq = nd->next_seq;
1764 if (flags & WALK_TRAILING) {
1765 error = may_follow_link(nd, inode);
1766 if (unlikely(error))
1767 return ERR_PTR(error);
1770 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1771 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1772 return ERR_PTR(-ELOOP);
1774 if (!(nd->flags & LOOKUP_RCU)) {
1775 touch_atime(&last->link);
1777 } else if (atime_needs_update(&last->link, inode)) {
1778 if (!try_to_unlazy(nd))
1779 return ERR_PTR(-ECHILD);
1780 touch_atime(&last->link);
1783 error = security_inode_follow_link(link->dentry, inode,
1784 nd->flags & LOOKUP_RCU);
1785 if (unlikely(error))
1786 return ERR_PTR(error);
1788 res = READ_ONCE(inode->i_link);
1790 const char * (*get)(struct dentry *, struct inode *,
1791 struct delayed_call *);
1792 get = inode->i_op->get_link;
1793 if (nd->flags & LOOKUP_RCU) {
1794 res = get(NULL, inode, &last->done);
1795 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1796 res = get(link->dentry, inode, &last->done);
1798 res = get(link->dentry, inode, &last->done);
1806 error = nd_jump_root(nd);
1807 if (unlikely(error))
1808 return ERR_PTR(error);
1809 while (unlikely(*++res == '/'))
1814 all_done: // pure jump
1820 * Do we need to follow links? We _really_ want to be able
1821 * to do this check without having to look at inode->i_op,
1822 * so we keep a cache of "no, this doesn't need follow_link"
1823 * for the common case.
1825 * NOTE: dentry must be what nd->next_seq had been sampled from.
1827 static const char *step_into(struct nameidata *nd, int flags,
1828 struct dentry *dentry)
1831 struct inode *inode;
1832 int err = handle_mounts(nd, dentry, &path);
1835 return ERR_PTR(err);
1836 inode = path.dentry->d_inode;
1837 if (likely(!d_is_symlink(path.dentry)) ||
1838 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1839 (flags & WALK_NOFOLLOW)) {
1840 /* not a symlink or should not follow */
1841 if (nd->flags & LOOKUP_RCU) {
1842 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1843 return ERR_PTR(-ECHILD);
1844 if (unlikely(!inode))
1845 return ERR_PTR(-ENOENT);
1847 dput(nd->path.dentry);
1848 if (nd->path.mnt != path.mnt)
1849 mntput(nd->path.mnt);
1853 nd->seq = nd->next_seq;
1856 if (nd->flags & LOOKUP_RCU) {
1857 /* make sure that d_is_symlink above matches inode */
1858 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1859 return ERR_PTR(-ECHILD);
1861 if (path.mnt == nd->path.mnt)
1864 return pick_link(nd, &path, inode, flags);
1867 static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1869 struct dentry *parent, *old;
1871 if (path_equal(&nd->path, &nd->root))
1873 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1876 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1877 &nd->root, &path, &seq))
1879 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1880 return ERR_PTR(-ECHILD);
1882 nd->inode = path.dentry->d_inode;
1884 // makes sure that non-RCU pathwalk could reach this state
1885 if (read_seqretry(&mount_lock, nd->m_seq))
1886 return ERR_PTR(-ECHILD);
1887 /* we know that mountpoint was pinned */
1889 old = nd->path.dentry;
1890 parent = old->d_parent;
1891 nd->next_seq = read_seqcount_begin(&parent->d_seq);
1892 // makes sure that non-RCU pathwalk could reach this state
1893 if (read_seqcount_retry(&old->d_seq, nd->seq))
1894 return ERR_PTR(-ECHILD);
1895 if (unlikely(!path_connected(nd->path.mnt, parent)))
1896 return ERR_PTR(-ECHILD);
1899 if (read_seqretry(&mount_lock, nd->m_seq))
1900 return ERR_PTR(-ECHILD);
1901 if (unlikely(nd->flags & LOOKUP_BENEATH))
1902 return ERR_PTR(-ECHILD);
1903 nd->next_seq = nd->seq;
1904 return nd->path.dentry;
1907 static struct dentry *follow_dotdot(struct nameidata *nd)
1909 struct dentry *parent;
1911 if (path_equal(&nd->path, &nd->root))
1913 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1916 if (!choose_mountpoint(real_mount(nd->path.mnt),
1919 path_put(&nd->path);
1921 nd->inode = path.dentry->d_inode;
1922 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1923 return ERR_PTR(-EXDEV);
1925 /* rare case of legitimate dget_parent()... */
1926 parent = dget_parent(nd->path.dentry);
1927 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1929 return ERR_PTR(-ENOENT);
1934 if (unlikely(nd->flags & LOOKUP_BENEATH))
1935 return ERR_PTR(-EXDEV);
1936 return dget(nd->path.dentry);
1939 static const char *handle_dots(struct nameidata *nd, int type)
1941 if (type == LAST_DOTDOT) {
1942 const char *error = NULL;
1943 struct dentry *parent;
1945 if (!nd->root.mnt) {
1946 error = ERR_PTR(set_root(nd));
1950 if (nd->flags & LOOKUP_RCU)
1951 parent = follow_dotdot_rcu(nd);
1953 parent = follow_dotdot(nd);
1955 return ERR_CAST(parent);
1956 error = step_into(nd, WALK_NOFOLLOW, parent);
1957 if (unlikely(error))
1960 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1962 * If there was a racing rename or mount along our
1963 * path, then we can't be sure that ".." hasn't jumped
1964 * above nd->root (and so userspace should retry or use
1968 if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
1969 return ERR_PTR(-EAGAIN);
1970 if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
1971 return ERR_PTR(-EAGAIN);
1977 static const char *walk_component(struct nameidata *nd, int flags)
1979 struct dentry *dentry;
1981 * "." and ".." are special - ".." especially so because it has
1982 * to be able to know about the current root directory and
1983 * parent relationships.
1985 if (unlikely(nd->last_type != LAST_NORM)) {
1986 if (!(flags & WALK_MORE) && nd->depth)
1988 return handle_dots(nd, nd->last_type);
1990 dentry = lookup_fast(nd);
1992 return ERR_CAST(dentry);
1993 if (unlikely(!dentry)) {
1994 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1996 return ERR_CAST(dentry);
1998 if (!(flags & WALK_MORE) && nd->depth)
2000 return step_into(nd, flags, dentry);
2004 * We can do the critical dentry name comparison and hashing
2005 * operations one word at a time, but we are limited to:
2007 * - Architectures with fast unaligned word accesses. We could
2008 * do a "get_unaligned()" if this helps and is sufficiently
2011 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2012 * do not trap on the (extremely unlikely) case of a page
2013 * crossing operation.
2015 * - Furthermore, we need an efficient 64-bit compile for the
2016 * 64-bit case in order to generate the "number of bytes in
2017 * the final mask". Again, that could be replaced with a
2018 * efficient population count instruction or similar.
2020 #ifdef CONFIG_DCACHE_WORD_ACCESS
2022 #include <asm/word-at-a-time.h>
2026 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2028 #elif defined(CONFIG_64BIT)
2030 * Register pressure in the mixing function is an issue, particularly
2031 * on 32-bit x86, but almost any function requires one state value and
2032 * one temporary. Instead, use a function designed for two state values
2033 * and no temporaries.
2035 * This function cannot create a collision in only two iterations, so
2036 * we have two iterations to achieve avalanche. In those two iterations,
2037 * we have six layers of mixing, which is enough to spread one bit's
2038 * influence out to 2^6 = 64 state bits.
2040 * Rotate constants are scored by considering either 64 one-bit input
2041 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2042 * probability of that delta causing a change to each of the 128 output
2043 * bits, using a sample of random initial states.
2045 * The Shannon entropy of the computed probabilities is then summed
2046 * to produce a score. Ideally, any input change has a 50% chance of
2047 * toggling any given output bit.
2049 * Mixing scores (in bits) for (12,45):
2050 * Input delta: 1-bit 2-bit
2051 * 1 round: 713.3 42542.6
2052 * 2 rounds: 2753.7 140389.8
2053 * 3 rounds: 5954.1 233458.2
2054 * 4 rounds: 7862.6 256672.2
2055 * Perfect: 8192 258048
2056 * (64*128) (64*63/2 * 128)
2058 #define HASH_MIX(x, y, a) \
2060 y ^= x, x = rol64(x,12),\
2061 x += y, y = rol64(y,45),\
2065 * Fold two longs into one 32-bit hash value. This must be fast, but
2066 * latency isn't quite as critical, as there is a fair bit of additional
2067 * work done before the hash value is used.
2069 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2071 y ^= x * GOLDEN_RATIO_64;
2072 y *= GOLDEN_RATIO_64;
2076 #else /* 32-bit case */
2079 * Mixing scores (in bits) for (7,20):
2080 * Input delta: 1-bit 2-bit
2081 * 1 round: 330.3 9201.6
2082 * 2 rounds: 1246.4 25475.4
2083 * 3 rounds: 1907.1 31295.1
2084 * 4 rounds: 2042.3 31718.6
2085 * Perfect: 2048 31744
2086 * (32*64) (32*31/2 * 64)
2088 #define HASH_MIX(x, y, a) \
2090 y ^= x, x = rol32(x, 7),\
2091 x += y, y = rol32(y,20),\
2094 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2096 /* Use arch-optimized multiply if one exists */
2097 return __hash_32(y ^ __hash_32(x));
2103 * Return the hash of a string of known length. This is carfully
2104 * designed to match hash_name(), which is the more critical function.
2105 * In particular, we must end by hashing a final word containing 0..7
2106 * payload bytes, to match the way that hash_name() iterates until it
2107 * finds the delimiter after the name.
2109 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2111 unsigned long a, x = 0, y = (unsigned long)salt;
2116 a = load_unaligned_zeropad(name);
2117 if (len < sizeof(unsigned long))
2120 name += sizeof(unsigned long);
2121 len -= sizeof(unsigned long);
2123 x ^= a & bytemask_from_count(len);
2125 return fold_hash(x, y);
2127 EXPORT_SYMBOL(full_name_hash);
2129 /* Return the "hash_len" (hash and length) of a null-terminated string */
2130 u64 hashlen_string(const void *salt, const char *name)
2132 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2133 unsigned long adata, mask, len;
2134 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2141 len += sizeof(unsigned long);
2143 a = load_unaligned_zeropad(name+len);
2144 } while (!has_zero(a, &adata, &constants));
2146 adata = prep_zero_mask(a, adata, &constants);
2147 mask = create_zero_mask(adata);
2148 x ^= a & zero_bytemask(mask);
2150 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2152 EXPORT_SYMBOL(hashlen_string);
2155 * Calculate the length and hash of the path component, and
2156 * return the "hash_len" as the result.
2158 static inline u64 hash_name(const void *salt, const char *name)
2160 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2161 unsigned long adata, bdata, mask, len;
2162 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2169 len += sizeof(unsigned long);
2171 a = load_unaligned_zeropad(name+len);
2172 b = a ^ REPEAT_BYTE('/');
2173 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2175 adata = prep_zero_mask(a, adata, &constants);
2176 bdata = prep_zero_mask(b, bdata, &constants);
2177 mask = create_zero_mask(adata | bdata);
2178 x ^= a & zero_bytemask(mask);
2180 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2183 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2185 /* Return the hash of a string of known length */
2186 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2188 unsigned long hash = init_name_hash(salt);
2190 hash = partial_name_hash((unsigned char)*name++, hash);
2191 return end_name_hash(hash);
2193 EXPORT_SYMBOL(full_name_hash);
2195 /* Return the "hash_len" (hash and length) of a null-terminated string */
2196 u64 hashlen_string(const void *salt, const char *name)
2198 unsigned long hash = init_name_hash(salt);
2199 unsigned long len = 0, c;
2201 c = (unsigned char)*name;
2204 hash = partial_name_hash(c, hash);
2205 c = (unsigned char)name[len];
2207 return hashlen_create(end_name_hash(hash), len);
2209 EXPORT_SYMBOL(hashlen_string);
2212 * We know there's a real path component here of at least
2215 static inline u64 hash_name(const void *salt, const char *name)
2217 unsigned long hash = init_name_hash(salt);
2218 unsigned long len = 0, c;
2220 c = (unsigned char)*name;
2223 hash = partial_name_hash(c, hash);
2224 c = (unsigned char)name[len];
2225 } while (c && c != '/');
2226 return hashlen_create(end_name_hash(hash), len);
2233 * This is the basic name resolution function, turning a pathname into
2234 * the final dentry. We expect 'base' to be positive and a directory.
2236 * Returns 0 and nd will have valid dentry and mnt on success.
2237 * Returns error and drops reference to input namei data on failure.
2239 static int link_path_walk(const char *name, struct nameidata *nd)
2241 int depth = 0; // depth <= nd->depth
2244 nd->last_type = LAST_ROOT;
2245 nd->flags |= LOOKUP_PARENT;
2247 return PTR_ERR(name);
2251 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2255 /* At this point we know we have a real path component. */
2257 struct mnt_idmap *idmap;
2262 idmap = mnt_idmap(nd->path.mnt);
2263 err = may_lookup(idmap, nd);
2267 hash_len = hash_name(nd->path.dentry, name);
2270 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2272 if (name[1] == '.') {
2274 nd->state |= ND_JUMPED;
2280 if (likely(type == LAST_NORM)) {
2281 struct dentry *parent = nd->path.dentry;
2282 nd->state &= ~ND_JUMPED;
2283 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2284 struct qstr this = { { .hash_len = hash_len }, .name = name };
2285 err = parent->d_op->d_hash(parent, &this);
2288 hash_len = this.hash_len;
2293 nd->last.hash_len = hash_len;
2294 nd->last.name = name;
2295 nd->last_type = type;
2297 name += hashlen_len(hash_len);
2301 * If it wasn't NUL, we know it was '/'. Skip that
2302 * slash, and continue until no more slashes.
2306 } while (unlikely(*name == '/'));
2307 if (unlikely(!*name)) {
2309 /* pathname or trailing symlink, done */
2311 nd->dir_vfsuid = i_uid_into_vfsuid(idmap, nd->inode);
2312 nd->dir_mode = nd->inode->i_mode;
2313 nd->flags &= ~LOOKUP_PARENT;
2316 /* last component of nested symlink */
2317 name = nd->stack[--depth].name;
2318 link = walk_component(nd, 0);
2320 /* not the last component */
2321 link = walk_component(nd, WALK_MORE);
2323 if (unlikely(link)) {
2325 return PTR_ERR(link);
2326 /* a symlink to follow */
2327 nd->stack[depth++].name = name;
2331 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2332 if (nd->flags & LOOKUP_RCU) {
2333 if (!try_to_unlazy(nd))
2341 /* must be paired with terminate_walk() */
2342 static const char *path_init(struct nameidata *nd, unsigned flags)
2345 const char *s = nd->name->name;
2347 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2348 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2349 return ERR_PTR(-EAGAIN);
2352 flags &= ~LOOKUP_RCU;
2353 if (flags & LOOKUP_RCU)
2356 nd->seq = nd->next_seq = 0;
2359 nd->state |= ND_JUMPED;
2361 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2362 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2365 if (nd->state & ND_ROOT_PRESET) {
2366 struct dentry *root = nd->root.dentry;
2367 struct inode *inode = root->d_inode;
2368 if (*s && unlikely(!d_can_lookup(root)))
2369 return ERR_PTR(-ENOTDIR);
2370 nd->path = nd->root;
2372 if (flags & LOOKUP_RCU) {
2373 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2374 nd->root_seq = nd->seq;
2376 path_get(&nd->path);
2381 nd->root.mnt = NULL;
2383 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2384 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2385 error = nd_jump_root(nd);
2386 if (unlikely(error))
2387 return ERR_PTR(error);
2391 /* Relative pathname -- get the starting-point it is relative to. */
2392 if (nd->dfd == AT_FDCWD) {
2393 if (flags & LOOKUP_RCU) {
2394 struct fs_struct *fs = current->fs;
2398 seq = read_seqcount_begin(&fs->seq);
2400 nd->inode = nd->path.dentry->d_inode;
2401 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2402 } while (read_seqcount_retry(&fs->seq, seq));
2404 get_fs_pwd(current->fs, &nd->path);
2405 nd->inode = nd->path.dentry->d_inode;
2408 /* Caller must check execute permissions on the starting path component */
2409 struct fd f = fdget_raw(nd->dfd);
2410 struct dentry *dentry;
2413 return ERR_PTR(-EBADF);
2415 dentry = f.file->f_path.dentry;
2417 if (*s && unlikely(!d_can_lookup(dentry))) {
2419 return ERR_PTR(-ENOTDIR);
2422 nd->path = f.file->f_path;
2423 if (flags & LOOKUP_RCU) {
2424 nd->inode = nd->path.dentry->d_inode;
2425 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2427 path_get(&nd->path);
2428 nd->inode = nd->path.dentry->d_inode;
2433 /* For scoped-lookups we need to set the root to the dirfd as well. */
2434 if (flags & LOOKUP_IS_SCOPED) {
2435 nd->root = nd->path;
2436 if (flags & LOOKUP_RCU) {
2437 nd->root_seq = nd->seq;
2439 path_get(&nd->root);
2440 nd->state |= ND_ROOT_GRABBED;
2446 static inline const char *lookup_last(struct nameidata *nd)
2448 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2449 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2451 return walk_component(nd, WALK_TRAILING);
2454 static int handle_lookup_down(struct nameidata *nd)
2456 if (!(nd->flags & LOOKUP_RCU))
2457 dget(nd->path.dentry);
2458 nd->next_seq = nd->seq;
2459 return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry));
2462 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2463 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2465 const char *s = path_init(nd, flags);
2468 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2469 err = handle_lookup_down(nd);
2470 if (unlikely(err < 0))
2474 while (!(err = link_path_walk(s, nd)) &&
2475 (s = lookup_last(nd)) != NULL)
2477 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2478 err = handle_lookup_down(nd);
2479 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2482 err = complete_walk(nd);
2484 if (!err && nd->flags & LOOKUP_DIRECTORY)
2485 if (!d_can_lookup(nd->path.dentry))
2489 nd->path.mnt = NULL;
2490 nd->path.dentry = NULL;
2496 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2497 struct path *path, struct path *root)
2500 struct nameidata nd;
2502 return PTR_ERR(name);
2503 set_nameidata(&nd, dfd, name, root);
2504 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2505 if (unlikely(retval == -ECHILD))
2506 retval = path_lookupat(&nd, flags, path);
2507 if (unlikely(retval == -ESTALE))
2508 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2510 if (likely(!retval))
2511 audit_inode(name, path->dentry,
2512 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2513 restore_nameidata();
2517 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2518 static int path_parentat(struct nameidata *nd, unsigned flags,
2519 struct path *parent)
2521 const char *s = path_init(nd, flags);
2522 int err = link_path_walk(s, nd);
2524 err = complete_walk(nd);
2527 nd->path.mnt = NULL;
2528 nd->path.dentry = NULL;
2534 /* Note: this does not consume "name" */
2535 static int filename_parentat(int dfd, struct filename *name,
2536 unsigned int flags, struct path *parent,
2537 struct qstr *last, int *type)
2540 struct nameidata nd;
2543 return PTR_ERR(name);
2544 set_nameidata(&nd, dfd, name, NULL);
2545 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2546 if (unlikely(retval == -ECHILD))
2547 retval = path_parentat(&nd, flags, parent);
2548 if (unlikely(retval == -ESTALE))
2549 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2550 if (likely(!retval)) {
2552 *type = nd.last_type;
2553 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2555 restore_nameidata();
2559 /* does lookup, returns the object with parent locked */
2560 static struct dentry *__kern_path_locked(struct filename *name, struct path *path)
2566 error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type);
2568 return ERR_PTR(error);
2569 if (unlikely(type != LAST_NORM)) {
2571 return ERR_PTR(-EINVAL);
2573 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2574 d = __lookup_hash(&last, path->dentry, 0);
2576 inode_unlock(path->dentry->d_inode);
2582 struct dentry *kern_path_locked(const char *name, struct path *path)
2584 struct filename *filename = getname_kernel(name);
2585 struct dentry *res = __kern_path_locked(filename, path);
2591 int kern_path(const char *name, unsigned int flags, struct path *path)
2593 struct filename *filename = getname_kernel(name);
2594 int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2600 EXPORT_SYMBOL(kern_path);
2603 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2604 * @dentry: pointer to dentry of the base directory
2605 * @mnt: pointer to vfs mount of the base directory
2606 * @name: pointer to file name
2607 * @flags: lookup flags
2608 * @path: pointer to struct path to fill
2610 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2611 const char *name, unsigned int flags,
2614 struct filename *filename;
2615 struct path root = {.mnt = mnt, .dentry = dentry};
2618 filename = getname_kernel(name);
2619 /* the first argument of filename_lookup() is ignored with root */
2620 ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2624 EXPORT_SYMBOL(vfs_path_lookup);
2626 static int lookup_one_common(struct mnt_idmap *idmap,
2627 const char *name, struct dentry *base, int len,
2632 this->hash = full_name_hash(base, name, len);
2636 if (unlikely(name[0] == '.')) {
2637 if (len < 2 || (len == 2 && name[1] == '.'))
2642 unsigned int c = *(const unsigned char *)name++;
2643 if (c == '/' || c == '\0')
2647 * See if the low-level filesystem might want
2648 * to use its own hash..
2650 if (base->d_flags & DCACHE_OP_HASH) {
2651 int err = base->d_op->d_hash(base, this);
2656 return inode_permission(idmap, base->d_inode, MAY_EXEC);
2660 * try_lookup_one_len - filesystem helper to lookup single pathname component
2661 * @name: pathname component to lookup
2662 * @base: base directory to lookup from
2663 * @len: maximum length @len should be interpreted to
2665 * Look up a dentry by name in the dcache, returning NULL if it does not
2666 * currently exist. The function does not try to create a dentry.
2668 * Note that this routine is purely a helper for filesystem usage and should
2669 * not be called by generic code.
2671 * The caller must hold base->i_mutex.
2673 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2678 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2680 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2682 return ERR_PTR(err);
2684 return lookup_dcache(&this, base, 0);
2686 EXPORT_SYMBOL(try_lookup_one_len);
2689 * lookup_one_len - filesystem helper to lookup single pathname component
2690 * @name: pathname component to lookup
2691 * @base: base directory to lookup from
2692 * @len: maximum length @len should be interpreted to
2694 * Note that this routine is purely a helper for filesystem usage and should
2695 * not be called by generic code.
2697 * The caller must hold base->i_mutex.
2699 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2701 struct dentry *dentry;
2705 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2707 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2709 return ERR_PTR(err);
2711 dentry = lookup_dcache(&this, base, 0);
2712 return dentry ? dentry : __lookup_slow(&this, base, 0);
2714 EXPORT_SYMBOL(lookup_one_len);
2717 * lookup_one - filesystem helper to lookup single pathname component
2718 * @idmap: idmap of the mount the lookup is performed from
2719 * @name: pathname component to lookup
2720 * @base: base directory to lookup from
2721 * @len: maximum length @len should be interpreted to
2723 * Note that this routine is purely a helper for filesystem usage and should
2724 * not be called by generic code.
2726 * The caller must hold base->i_mutex.
2728 struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name,
2729 struct dentry *base, int len)
2731 struct dentry *dentry;
2735 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2737 err = lookup_one_common(idmap, name, base, len, &this);
2739 return ERR_PTR(err);
2741 dentry = lookup_dcache(&this, base, 0);
2742 return dentry ? dentry : __lookup_slow(&this, base, 0);
2744 EXPORT_SYMBOL(lookup_one);
2747 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2748 * @idmap: idmap of the mount the lookup is performed from
2749 * @name: pathname component to lookup
2750 * @base: base directory to lookup from
2751 * @len: maximum length @len should be interpreted to
2753 * Note that this routine is purely a helper for filesystem usage and should
2754 * not be called by generic code.
2756 * Unlike lookup_one_len, it should be called without the parent
2757 * i_mutex held, and will take the i_mutex itself if necessary.
2759 struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap,
2760 const char *name, struct dentry *base,
2767 err = lookup_one_common(idmap, name, base, len, &this);
2769 return ERR_PTR(err);
2771 ret = lookup_dcache(&this, base, 0);
2773 ret = lookup_slow(&this, base, 0);
2776 EXPORT_SYMBOL(lookup_one_unlocked);
2779 * lookup_one_positive_unlocked - filesystem helper to lookup single
2780 * pathname component
2781 * @idmap: idmap of the mount the lookup is performed from
2782 * @name: pathname component to lookup
2783 * @base: base directory to lookup from
2784 * @len: maximum length @len should be interpreted to
2786 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2787 * known positive or ERR_PTR(). This is what most of the users want.
2789 * Note that pinned negative with unlocked parent _can_ become positive at any
2790 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2791 * positives have >d_inode stable, so this one avoids such problems.
2793 * Note that this routine is purely a helper for filesystem usage and should
2794 * not be called by generic code.
2796 * The helper should be called without i_mutex held.
2798 struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap,
2800 struct dentry *base, int len)
2802 struct dentry *ret = lookup_one_unlocked(idmap, name, base, len);
2804 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2806 ret = ERR_PTR(-ENOENT);
2810 EXPORT_SYMBOL(lookup_one_positive_unlocked);
2813 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2814 * @name: pathname component to lookup
2815 * @base: base directory to lookup from
2816 * @len: maximum length @len should be interpreted to
2818 * Note that this routine is purely a helper for filesystem usage and should
2819 * not be called by generic code.
2821 * Unlike lookup_one_len, it should be called without the parent
2822 * i_mutex held, and will take the i_mutex itself if necessary.
2824 struct dentry *lookup_one_len_unlocked(const char *name,
2825 struct dentry *base, int len)
2827 return lookup_one_unlocked(&nop_mnt_idmap, name, base, len);
2829 EXPORT_SYMBOL(lookup_one_len_unlocked);
2832 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2833 * on negatives. Returns known positive or ERR_PTR(); that's what
2834 * most of the users want. Note that pinned negative with unlocked parent
2835 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2836 * need to be very careful; pinned positives have ->d_inode stable, so
2837 * this one avoids such problems.
2839 struct dentry *lookup_positive_unlocked(const char *name,
2840 struct dentry *base, int len)
2842 return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len);
2844 EXPORT_SYMBOL(lookup_positive_unlocked);
2846 #ifdef CONFIG_UNIX98_PTYS
2847 int path_pts(struct path *path)
2849 /* Find something mounted on "pts" in the same directory as
2852 struct dentry *parent = dget_parent(path->dentry);
2853 struct dentry *child;
2854 struct qstr this = QSTR_INIT("pts", 3);
2856 if (unlikely(!path_connected(path->mnt, parent))) {
2861 path->dentry = parent;
2862 child = d_hash_and_lookup(parent, &this);
2866 path->dentry = child;
2868 follow_down(path, 0);
2873 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2874 struct path *path, int *empty)
2876 struct filename *filename = getname_flags(name, flags, empty);
2877 int ret = filename_lookup(dfd, filename, flags, path, NULL);
2882 EXPORT_SYMBOL(user_path_at_empty);
2884 int __check_sticky(struct mnt_idmap *idmap, struct inode *dir,
2885 struct inode *inode)
2887 kuid_t fsuid = current_fsuid();
2889 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), fsuid))
2891 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, dir), fsuid))
2893 return !capable_wrt_inode_uidgid(idmap, inode, CAP_FOWNER);
2895 EXPORT_SYMBOL(__check_sticky);
2898 * Check whether we can remove a link victim from directory dir, check
2899 * whether the type of victim is right.
2900 * 1. We can't do it if dir is read-only (done in permission())
2901 * 2. We should have write and exec permissions on dir
2902 * 3. We can't remove anything from append-only dir
2903 * 4. We can't do anything with immutable dir (done in permission())
2904 * 5. If the sticky bit on dir is set we should either
2905 * a. be owner of dir, or
2906 * b. be owner of victim, or
2907 * c. have CAP_FOWNER capability
2908 * 6. If the victim is append-only or immutable we can't do antyhing with
2909 * links pointing to it.
2910 * 7. If the victim has an unknown uid or gid we can't change the inode.
2911 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2912 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2913 * 10. We can't remove a root or mountpoint.
2914 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2915 * nfs_async_unlink().
2917 static int may_delete(struct mnt_idmap *idmap, struct inode *dir,
2918 struct dentry *victim, bool isdir)
2920 struct inode *inode = d_backing_inode(victim);
2923 if (d_is_negative(victim))
2927 BUG_ON(victim->d_parent->d_inode != dir);
2929 /* Inode writeback is not safe when the uid or gid are invalid. */
2930 if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
2931 !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
2934 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2936 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
2942 if (check_sticky(idmap, dir, inode) || IS_APPEND(inode) ||
2943 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2944 HAS_UNMAPPED_ID(idmap, inode))
2947 if (!d_is_dir(victim))
2949 if (IS_ROOT(victim))
2951 } else if (d_is_dir(victim))
2953 if (IS_DEADDIR(dir))
2955 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2960 /* Check whether we can create an object with dentry child in directory
2962 * 1. We can't do it if child already exists (open has special treatment for
2963 * this case, but since we are inlined it's OK)
2964 * 2. We can't do it if dir is read-only (done in permission())
2965 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2966 * 4. We should have write and exec permissions on dir
2967 * 5. We can't do it if dir is immutable (done in permission())
2969 static inline int may_create(struct mnt_idmap *idmap,
2970 struct inode *dir, struct dentry *child)
2972 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2975 if (IS_DEADDIR(dir))
2977 if (!fsuidgid_has_mapping(dir->i_sb, idmap))
2980 return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
2983 static struct dentry *lock_two_directories(struct dentry *p1, struct dentry *p2)
2987 p = d_ancestor(p2, p1);
2989 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2990 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2994 p = d_ancestor(p1, p2);
2996 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2997 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
3001 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3002 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3007 * p1 and p2 should be directories on the same fs.
3009 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
3012 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3016 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
3017 return lock_two_directories(p1, p2);
3019 EXPORT_SYMBOL(lock_rename);
3022 * c1 and p2 should be on the same fs.
3024 struct dentry *lock_rename_child(struct dentry *c1, struct dentry *p2)
3026 if (READ_ONCE(c1->d_parent) == p2) {
3028 * hopefully won't need to touch ->s_vfs_rename_mutex at all.
3030 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3032 * now that p2 is locked, nobody can move in or out of it,
3033 * so the test below is safe.
3035 if (likely(c1->d_parent == p2))
3039 * c1 got moved out of p2 while we'd been taking locks;
3040 * unlock and fall back to slow case.
3042 inode_unlock(p2->d_inode);
3045 mutex_lock(&c1->d_sb->s_vfs_rename_mutex);
3047 * nobody can move out of any directories on this fs.
3049 if (likely(c1->d_parent != p2))
3050 return lock_two_directories(c1->d_parent, p2);
3053 * c1 got moved into p2 while we were taking locks;
3054 * we need p2 locked and ->s_vfs_rename_mutex unlocked,
3055 * for consistency with lock_rename().
3057 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3058 mutex_unlock(&c1->d_sb->s_vfs_rename_mutex);
3061 EXPORT_SYMBOL(lock_rename_child);
3063 void unlock_rename(struct dentry *p1, struct dentry *p2)
3065 inode_unlock(p1->d_inode);
3067 inode_unlock(p2->d_inode);
3068 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3071 EXPORT_SYMBOL(unlock_rename);
3074 * mode_strip_umask - handle vfs umask stripping
3075 * @dir: parent directory of the new inode
3076 * @mode: mode of the new inode to be created in @dir
3078 * Umask stripping depends on whether or not the filesystem supports POSIX
3079 * ACLs. If the filesystem doesn't support it umask stripping is done directly
3080 * in here. If the filesystem does support POSIX ACLs umask stripping is
3081 * deferred until the filesystem calls posix_acl_create().
3085 static inline umode_t mode_strip_umask(const struct inode *dir, umode_t mode)
3087 if (!IS_POSIXACL(dir))
3088 mode &= ~current_umask();
3093 * vfs_prepare_mode - prepare the mode to be used for a new inode
3094 * @idmap: idmap of the mount the inode was found from
3095 * @dir: parent directory of the new inode
3096 * @mode: mode of the new inode
3097 * @mask_perms: allowed permission by the vfs
3098 * @type: type of file to be created
3100 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3101 * object to be created.
3103 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3104 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3105 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3106 * POSIX ACL supporting filesystems.
3108 * Note that it's currently valid for @type to be 0 if a directory is created.
3109 * Filesystems raise that flag individually and we need to check whether each
3110 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3113 * Returns: mode to be passed to the filesystem
3115 static inline umode_t vfs_prepare_mode(struct mnt_idmap *idmap,
3116 const struct inode *dir, umode_t mode,
3117 umode_t mask_perms, umode_t type)
3119 mode = mode_strip_sgid(idmap, dir, mode);
3120 mode = mode_strip_umask(dir, mode);
3123 * Apply the vfs mandated allowed permission mask and set the type of
3124 * file to be created before we call into the filesystem.
3126 mode &= (mask_perms & ~S_IFMT);
3127 mode |= (type & S_IFMT);
3133 * vfs_create - create new file
3134 * @idmap: idmap of the mount the inode was found from
3135 * @dir: inode of @dentry
3136 * @dentry: pointer to dentry of the base directory
3137 * @mode: mode of the new file
3138 * @want_excl: whether the file must not yet exist
3140 * Create a new file.
3142 * If the inode has been found through an idmapped mount the idmap of
3143 * the vfsmount must be passed through @idmap. This function will then take
3144 * care to map the inode according to @idmap before checking permissions.
3145 * On non-idmapped mounts or if permission checking is to be performed on the
3146 * raw inode simply passs @nop_mnt_idmap.
3148 int vfs_create(struct mnt_idmap *idmap, struct inode *dir,
3149 struct dentry *dentry, umode_t mode, bool want_excl)
3153 error = may_create(idmap, dir, dentry);
3157 if (!dir->i_op->create)
3158 return -EACCES; /* shouldn't it be ENOSYS? */
3160 mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG);
3161 error = security_inode_create(dir, dentry, mode);
3164 error = dir->i_op->create(idmap, dir, dentry, mode, want_excl);
3166 fsnotify_create(dir, dentry);
3169 EXPORT_SYMBOL(vfs_create);
3171 int vfs_mkobj(struct dentry *dentry, umode_t mode,
3172 int (*f)(struct dentry *, umode_t, void *),
3175 struct inode *dir = dentry->d_parent->d_inode;
3176 int error = may_create(&nop_mnt_idmap, dir, dentry);
3182 error = security_inode_create(dir, dentry, mode);
3185 error = f(dentry, mode, arg);
3187 fsnotify_create(dir, dentry);
3190 EXPORT_SYMBOL(vfs_mkobj);
3192 bool may_open_dev(const struct path *path)
3194 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3195 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3198 static int may_open(struct mnt_idmap *idmap, const struct path *path,
3199 int acc_mode, int flag)
3201 struct dentry *dentry = path->dentry;
3202 struct inode *inode = dentry->d_inode;
3208 switch (inode->i_mode & S_IFMT) {
3212 if (acc_mode & MAY_WRITE)
3214 if (acc_mode & MAY_EXEC)
3219 if (!may_open_dev(path))
3224 if (acc_mode & MAY_EXEC)
3229 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3234 error = inode_permission(idmap, inode, MAY_OPEN | acc_mode);
3239 * An append-only file must be opened in append mode for writing.
3241 if (IS_APPEND(inode)) {
3242 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3248 /* O_NOATIME can only be set by the owner or superuser */
3249 if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode))
3255 static int handle_truncate(struct mnt_idmap *idmap, struct file *filp)
3257 const struct path *path = &filp->f_path;
3258 struct inode *inode = path->dentry->d_inode;
3259 int error = get_write_access(inode);
3263 error = security_file_truncate(filp);
3265 error = do_truncate(idmap, path->dentry, 0,
3266 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3269 put_write_access(inode);
3273 static inline int open_to_namei_flags(int flag)
3275 if ((flag & O_ACCMODE) == 3)
3280 static int may_o_create(struct mnt_idmap *idmap,
3281 const struct path *dir, struct dentry *dentry,
3284 int error = security_path_mknod(dir, dentry, mode, 0);
3288 if (!fsuidgid_has_mapping(dir->dentry->d_sb, idmap))
3291 error = inode_permission(idmap, dir->dentry->d_inode,
3292 MAY_WRITE | MAY_EXEC);
3296 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3300 * Attempt to atomically look up, create and open a file from a negative
3303 * Returns 0 if successful. The file will have been created and attached to
3304 * @file by the filesystem calling finish_open().
3306 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3307 * be set. The caller will need to perform the open themselves. @path will
3308 * have been updated to point to the new dentry. This may be negative.
3310 * Returns an error code otherwise.
3312 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3314 int open_flag, umode_t mode)
3316 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3317 struct inode *dir = nd->path.dentry->d_inode;
3320 if (nd->flags & LOOKUP_DIRECTORY)
3321 open_flag |= O_DIRECTORY;
3323 file->f_path.dentry = DENTRY_NOT_SET;
3324 file->f_path.mnt = nd->path.mnt;
3325 error = dir->i_op->atomic_open(dir, dentry, file,
3326 open_to_namei_flags(open_flag), mode);
3327 d_lookup_done(dentry);
3329 if (file->f_mode & FMODE_OPENED) {
3330 if (unlikely(dentry != file->f_path.dentry)) {
3332 dentry = dget(file->f_path.dentry);
3334 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3337 if (file->f_path.dentry) {
3339 dentry = file->f_path.dentry;
3341 if (unlikely(d_is_negative(dentry)))
3347 dentry = ERR_PTR(error);
3353 * Look up and maybe create and open the last component.
3355 * Must be called with parent locked (exclusive in O_CREAT case).
3357 * Returns 0 on success, that is, if
3358 * the file was successfully atomically created (if necessary) and opened, or
3359 * the file was not completely opened at this time, though lookups and
3360 * creations were performed.
3361 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3362 * In the latter case dentry returned in @path might be negative if O_CREAT
3363 * hadn't been specified.
3365 * An error code is returned on failure.
3367 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3368 const struct open_flags *op,
3371 struct mnt_idmap *idmap;
3372 struct dentry *dir = nd->path.dentry;
3373 struct inode *dir_inode = dir->d_inode;
3374 int open_flag = op->open_flag;
3375 struct dentry *dentry;
3376 int error, create_error = 0;
3377 umode_t mode = op->mode;
3378 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3380 if (unlikely(IS_DEADDIR(dir_inode)))
3381 return ERR_PTR(-ENOENT);
3383 file->f_mode &= ~FMODE_CREATED;
3384 dentry = d_lookup(dir, &nd->last);
3387 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3391 if (d_in_lookup(dentry))
3394 error = d_revalidate(dentry, nd->flags);
3395 if (likely(error > 0))
3399 d_invalidate(dentry);
3403 if (dentry->d_inode) {
3404 /* Cached positive dentry: will open in f_op->open */
3409 * Checking write permission is tricky, bacuse we don't know if we are
3410 * going to actually need it: O_CREAT opens should work as long as the
3411 * file exists. But checking existence breaks atomicity. The trick is
3412 * to check access and if not granted clear O_CREAT from the flags.
3414 * Another problem is returing the "right" error value (e.g. for an
3415 * O_EXCL open we want to return EEXIST not EROFS).
3417 if (unlikely(!got_write))
3418 open_flag &= ~O_TRUNC;
3419 idmap = mnt_idmap(nd->path.mnt);
3420 if (open_flag & O_CREAT) {
3421 if (open_flag & O_EXCL)
3422 open_flag &= ~O_TRUNC;
3423 mode = vfs_prepare_mode(idmap, dir->d_inode, mode, mode, mode);
3424 if (likely(got_write))
3425 create_error = may_o_create(idmap, &nd->path,
3428 create_error = -EROFS;
3431 open_flag &= ~O_CREAT;
3432 if (dir_inode->i_op->atomic_open) {
3433 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3434 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3435 dentry = ERR_PTR(create_error);
3439 if (d_in_lookup(dentry)) {
3440 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3442 d_lookup_done(dentry);
3443 if (unlikely(res)) {
3445 error = PTR_ERR(res);
3453 /* Negative dentry, just create the file */
3454 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3455 file->f_mode |= FMODE_CREATED;
3456 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3457 if (!dir_inode->i_op->create) {
3462 error = dir_inode->i_op->create(idmap, dir_inode, dentry,
3463 mode, open_flag & O_EXCL);
3467 if (unlikely(create_error) && !dentry->d_inode) {
3468 error = create_error;
3475 return ERR_PTR(error);
3478 static const char *open_last_lookups(struct nameidata *nd,
3479 struct file *file, const struct open_flags *op)
3481 struct dentry *dir = nd->path.dentry;
3482 int open_flag = op->open_flag;
3483 bool got_write = false;
3484 struct dentry *dentry;
3487 nd->flags |= op->intent;
3489 if (nd->last_type != LAST_NORM) {
3492 return handle_dots(nd, nd->last_type);
3495 if (!(open_flag & O_CREAT)) {
3496 if (nd->last.name[nd->last.len])
3497 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3498 /* we _can_ be in RCU mode here */
3499 dentry = lookup_fast(nd);
3501 return ERR_CAST(dentry);
3505 BUG_ON(nd->flags & LOOKUP_RCU);
3507 /* create side of things */
3508 if (nd->flags & LOOKUP_RCU) {
3509 if (!try_to_unlazy(nd))
3510 return ERR_PTR(-ECHILD);
3512 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3513 /* trailing slashes? */
3514 if (unlikely(nd->last.name[nd->last.len]))
3515 return ERR_PTR(-EISDIR);
3518 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3519 got_write = !mnt_want_write(nd->path.mnt);
3521 * do _not_ fail yet - we might not need that or fail with
3522 * a different error; let lookup_open() decide; we'll be
3523 * dropping this one anyway.
3526 if (open_flag & O_CREAT)
3527 inode_lock(dir->d_inode);
3529 inode_lock_shared(dir->d_inode);
3530 dentry = lookup_open(nd, file, op, got_write);
3531 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3532 fsnotify_create(dir->d_inode, dentry);
3533 if (open_flag & O_CREAT)
3534 inode_unlock(dir->d_inode);
3536 inode_unlock_shared(dir->d_inode);
3539 mnt_drop_write(nd->path.mnt);
3542 return ERR_CAST(dentry);
3544 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3545 dput(nd->path.dentry);
3546 nd->path.dentry = dentry;
3553 res = step_into(nd, WALK_TRAILING, dentry);
3555 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3560 * Handle the last step of open()
3562 static int do_open(struct nameidata *nd,
3563 struct file *file, const struct open_flags *op)
3565 struct mnt_idmap *idmap;
3566 int open_flag = op->open_flag;
3571 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3572 error = complete_walk(nd);
3576 if (!(file->f_mode & FMODE_CREATED))
3577 audit_inode(nd->name, nd->path.dentry, 0);
3578 idmap = mnt_idmap(nd->path.mnt);
3579 if (open_flag & O_CREAT) {
3580 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3582 if (d_is_dir(nd->path.dentry))
3584 error = may_create_in_sticky(idmap, nd,
3585 d_backing_inode(nd->path.dentry));
3586 if (unlikely(error))
3589 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3592 do_truncate = false;
3593 acc_mode = op->acc_mode;
3594 if (file->f_mode & FMODE_CREATED) {
3595 /* Don't check for write permission, don't truncate */
3596 open_flag &= ~O_TRUNC;
3598 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3599 error = mnt_want_write(nd->path.mnt);
3604 error = may_open(idmap, &nd->path, acc_mode, open_flag);
3605 if (!error && !(file->f_mode & FMODE_OPENED))
3606 error = vfs_open(&nd->path, file);
3608 error = ima_file_check(file, op->acc_mode);
3609 if (!error && do_truncate)
3610 error = handle_truncate(idmap, file);
3611 if (unlikely(error > 0)) {
3616 mnt_drop_write(nd->path.mnt);
3621 * vfs_tmpfile - create tmpfile
3622 * @idmap: idmap of the mount the inode was found from
3623 * @dentry: pointer to dentry of the base directory
3624 * @mode: mode of the new tmpfile
3627 * Create a temporary file.
3629 * If the inode has been found through an idmapped mount the idmap of
3630 * the vfsmount must be passed through @idmap. This function will then take
3631 * care to map the inode according to @idmap before checking permissions.
3632 * On non-idmapped mounts or if permission checking is to be performed on the
3633 * raw inode simply passs @nop_mnt_idmap.
3635 static int vfs_tmpfile(struct mnt_idmap *idmap,
3636 const struct path *parentpath,
3637 struct file *file, umode_t mode)
3639 struct dentry *child;
3640 struct inode *dir = d_inode(parentpath->dentry);
3641 struct inode *inode;
3643 int open_flag = file->f_flags;
3645 /* we want directory to be writable */
3646 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3649 if (!dir->i_op->tmpfile)
3651 child = d_alloc(parentpath->dentry, &slash_name);
3652 if (unlikely(!child))
3654 file->f_path.mnt = parentpath->mnt;
3655 file->f_path.dentry = child;
3656 mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3657 error = dir->i_op->tmpfile(idmap, dir, file, mode);
3661 /* Don't check for other permissions, the inode was just created */
3662 error = may_open(idmap, &file->f_path, 0, file->f_flags);
3665 inode = file_inode(file);
3666 if (!(open_flag & O_EXCL)) {
3667 spin_lock(&inode->i_lock);
3668 inode->i_state |= I_LINKABLE;
3669 spin_unlock(&inode->i_lock);
3671 ima_post_create_tmpfile(idmap, inode);
3676 * vfs_tmpfile_open - open a tmpfile for kernel internal use
3677 * @idmap: idmap of the mount the inode was found from
3678 * @parentpath: path of the base directory
3679 * @mode: mode of the new tmpfile
3681 * @cred: credentials for open
3683 * Create and open a temporary file. The file is not accounted in nr_files,
3684 * hence this is only for kernel internal use, and must not be installed into
3685 * file tables or such.
3687 struct file *vfs_tmpfile_open(struct mnt_idmap *idmap,
3688 const struct path *parentpath,
3689 umode_t mode, int open_flag, const struct cred *cred)
3694 file = alloc_empty_file_noaccount(open_flag, cred);
3695 if (!IS_ERR(file)) {
3696 error = vfs_tmpfile(idmap, parentpath, file, mode);
3699 file = ERR_PTR(error);
3704 EXPORT_SYMBOL(vfs_tmpfile_open);
3706 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3707 const struct open_flags *op,
3711 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3713 if (unlikely(error))
3715 error = mnt_want_write(path.mnt);
3716 if (unlikely(error))
3718 error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode);
3721 audit_inode(nd->name, file->f_path.dentry, 0);
3723 mnt_drop_write(path.mnt);
3729 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3732 int error = path_lookupat(nd, flags, &path);
3734 audit_inode(nd->name, path.dentry, 0);
3735 error = vfs_open(&path, file);
3741 static struct file *path_openat(struct nameidata *nd,
3742 const struct open_flags *op, unsigned flags)
3747 file = alloc_empty_file(op->open_flag, current_cred());
3751 if (unlikely(file->f_flags & __O_TMPFILE)) {
3752 error = do_tmpfile(nd, flags, op, file);
3753 } else if (unlikely(file->f_flags & O_PATH)) {
3754 error = do_o_path(nd, flags, file);
3756 const char *s = path_init(nd, flags);
3757 while (!(error = link_path_walk(s, nd)) &&
3758 (s = open_last_lookups(nd, file, op)) != NULL)
3761 error = do_open(nd, file, op);
3764 if (likely(!error)) {
3765 if (likely(file->f_mode & FMODE_OPENED))
3771 if (error == -EOPENSTALE) {
3772 if (flags & LOOKUP_RCU)
3777 return ERR_PTR(error);
3780 struct file *do_filp_open(int dfd, struct filename *pathname,
3781 const struct open_flags *op)
3783 struct nameidata nd;
3784 int flags = op->lookup_flags;
3787 set_nameidata(&nd, dfd, pathname, NULL);
3788 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3789 if (unlikely(filp == ERR_PTR(-ECHILD)))
3790 filp = path_openat(&nd, op, flags);
3791 if (unlikely(filp == ERR_PTR(-ESTALE)))
3792 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3793 restore_nameidata();
3797 struct file *do_file_open_root(const struct path *root,
3798 const char *name, const struct open_flags *op)
3800 struct nameidata nd;
3802 struct filename *filename;
3803 int flags = op->lookup_flags;
3805 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3806 return ERR_PTR(-ELOOP);
3808 filename = getname_kernel(name);
3809 if (IS_ERR(filename))
3810 return ERR_CAST(filename);
3812 set_nameidata(&nd, -1, filename, root);
3813 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3814 if (unlikely(file == ERR_PTR(-ECHILD)))
3815 file = path_openat(&nd, op, flags);
3816 if (unlikely(file == ERR_PTR(-ESTALE)))
3817 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3818 restore_nameidata();
3823 static struct dentry *filename_create(int dfd, struct filename *name,
3824 struct path *path, unsigned int lookup_flags)
3826 struct dentry *dentry = ERR_PTR(-EEXIST);
3828 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3829 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3830 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3835 error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3837 return ERR_PTR(error);
3840 * Yucky last component or no last component at all?
3841 * (foo/., foo/.., /////)
3843 if (unlikely(type != LAST_NORM))
3846 /* don't fail immediately if it's r/o, at least try to report other errors */
3847 err2 = mnt_want_write(path->mnt);
3849 * Do the final lookup. Suppress 'create' if there is a trailing
3850 * '/', and a directory wasn't requested.
3852 if (last.name[last.len] && !want_dir)
3854 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3855 dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags);
3860 if (d_is_positive(dentry))
3864 * Special case - lookup gave negative, but... we had foo/bar/
3865 * From the vfs_mknod() POV we just have a negative dentry -
3866 * all is fine. Let's be bastards - you had / on the end, you've
3867 * been asking for (non-existent) directory. -ENOENT for you.
3869 if (unlikely(!create_flags)) {
3873 if (unlikely(err2)) {
3880 dentry = ERR_PTR(error);
3882 inode_unlock(path->dentry->d_inode);
3884 mnt_drop_write(path->mnt);
3890 struct dentry *kern_path_create(int dfd, const char *pathname,
3891 struct path *path, unsigned int lookup_flags)
3893 struct filename *filename = getname_kernel(pathname);
3894 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3899 EXPORT_SYMBOL(kern_path_create);
3901 void done_path_create(struct path *path, struct dentry *dentry)
3904 inode_unlock(path->dentry->d_inode);
3905 mnt_drop_write(path->mnt);
3908 EXPORT_SYMBOL(done_path_create);
3910 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3911 struct path *path, unsigned int lookup_flags)
3913 struct filename *filename = getname(pathname);
3914 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3919 EXPORT_SYMBOL(user_path_create);
3922 * vfs_mknod - create device node or file
3923 * @idmap: idmap of the mount the inode was found from
3924 * @dir: inode of @dentry
3925 * @dentry: pointer to dentry of the base directory
3926 * @mode: mode of the new device node or file
3927 * @dev: device number of device to create
3929 * Create a device node or file.
3931 * If the inode has been found through an idmapped mount the idmap of
3932 * the vfsmount must be passed through @idmap. This function will then take
3933 * care to map the inode according to @idmap before checking permissions.
3934 * On non-idmapped mounts or if permission checking is to be performed on the
3935 * raw inode simply passs @nop_mnt_idmap.
3937 int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
3938 struct dentry *dentry, umode_t mode, dev_t dev)
3940 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3941 int error = may_create(idmap, dir, dentry);
3946 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3947 !capable(CAP_MKNOD))
3950 if (!dir->i_op->mknod)
3953 mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3954 error = devcgroup_inode_mknod(mode, dev);
3958 error = security_inode_mknod(dir, dentry, mode, dev);
3962 error = dir->i_op->mknod(idmap, dir, dentry, mode, dev);
3964 fsnotify_create(dir, dentry);
3967 EXPORT_SYMBOL(vfs_mknod);
3969 static int may_mknod(umode_t mode)
3971 switch (mode & S_IFMT) {
3977 case 0: /* zero mode translates to S_IFREG */
3986 static int do_mknodat(int dfd, struct filename *name, umode_t mode,
3989 struct mnt_idmap *idmap;
3990 struct dentry *dentry;
3993 unsigned int lookup_flags = 0;
3995 error = may_mknod(mode);
3999 dentry = filename_create(dfd, name, &path, lookup_flags);
4000 error = PTR_ERR(dentry);
4004 error = security_path_mknod(&path, dentry,
4005 mode_strip_umask(path.dentry->d_inode, mode), dev);
4009 idmap = mnt_idmap(path.mnt);
4010 switch (mode & S_IFMT) {
4011 case 0: case S_IFREG:
4012 error = vfs_create(idmap, path.dentry->d_inode,
4013 dentry, mode, true);
4015 ima_post_path_mknod(idmap, dentry);
4017 case S_IFCHR: case S_IFBLK:
4018 error = vfs_mknod(idmap, path.dentry->d_inode,
4019 dentry, mode, new_decode_dev(dev));
4021 case S_IFIFO: case S_IFSOCK:
4022 error = vfs_mknod(idmap, path.dentry->d_inode,
4027 done_path_create(&path, dentry);
4028 if (retry_estale(error, lookup_flags)) {
4029 lookup_flags |= LOOKUP_REVAL;
4037 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
4040 return do_mknodat(dfd, getname(filename), mode, dev);
4043 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
4045 return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
4049 * vfs_mkdir - create directory
4050 * @idmap: idmap of the mount the inode was found from
4051 * @dir: inode of @dentry
4052 * @dentry: pointer to dentry of the base directory
4053 * @mode: mode of the new directory
4055 * Create a directory.
4057 * If the inode has been found through an idmapped mount the idmap of
4058 * the vfsmount must be passed through @idmap. This function will then take
4059 * care to map the inode according to @idmap before checking permissions.
4060 * On non-idmapped mounts or if permission checking is to be performed on the
4061 * raw inode simply passs @nop_mnt_idmap.
4063 int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
4064 struct dentry *dentry, umode_t mode)
4067 unsigned max_links = dir->i_sb->s_max_links;
4069 error = may_create(idmap, dir, dentry);
4073 if (!dir->i_op->mkdir)
4076 mode = vfs_prepare_mode(idmap, dir, mode, S_IRWXUGO | S_ISVTX, 0);
4077 error = security_inode_mkdir(dir, dentry, mode);
4081 if (max_links && dir->i_nlink >= max_links)
4084 error = dir->i_op->mkdir(idmap, dir, dentry, mode);
4086 fsnotify_mkdir(dir, dentry);
4089 EXPORT_SYMBOL(vfs_mkdir);
4091 int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4093 struct dentry *dentry;
4096 unsigned int lookup_flags = LOOKUP_DIRECTORY;
4099 dentry = filename_create(dfd, name, &path, lookup_flags);
4100 error = PTR_ERR(dentry);
4104 error = security_path_mkdir(&path, dentry,
4105 mode_strip_umask(path.dentry->d_inode, mode));
4107 error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode,
4110 done_path_create(&path, dentry);
4111 if (retry_estale(error, lookup_flags)) {
4112 lookup_flags |= LOOKUP_REVAL;
4120 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4122 return do_mkdirat(dfd, getname(pathname), mode);
4125 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4127 return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4131 * vfs_rmdir - remove directory
4132 * @idmap: idmap of the mount the inode was found from
4133 * @dir: inode of @dentry
4134 * @dentry: pointer to dentry of the base directory
4136 * Remove a directory.
4138 * If the inode has been found through an idmapped mount the idmap of
4139 * the vfsmount must be passed through @idmap. This function will then take
4140 * care to map the inode according to @idmap before checking permissions.
4141 * On non-idmapped mounts or if permission checking is to be performed on the
4142 * raw inode simply passs @nop_mnt_idmap.
4144 int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir,
4145 struct dentry *dentry)
4147 int error = may_delete(idmap, dir, dentry, 1);
4152 if (!dir->i_op->rmdir)
4156 inode_lock(dentry->d_inode);
4159 if (is_local_mountpoint(dentry) ||
4160 (dentry->d_inode->i_flags & S_KERNEL_FILE))
4163 error = security_inode_rmdir(dir, dentry);
4167 error = dir->i_op->rmdir(dir, dentry);
4171 shrink_dcache_parent(dentry);
4172 dentry->d_inode->i_flags |= S_DEAD;
4174 detach_mounts(dentry);
4177 inode_unlock(dentry->d_inode);
4180 d_delete_notify(dir, dentry);
4183 EXPORT_SYMBOL(vfs_rmdir);
4185 int do_rmdir(int dfd, struct filename *name)
4188 struct dentry *dentry;
4192 unsigned int lookup_flags = 0;
4194 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4210 error = mnt_want_write(path.mnt);
4214 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4215 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4216 error = PTR_ERR(dentry);
4219 if (!dentry->d_inode) {
4223 error = security_path_rmdir(&path, dentry);
4226 error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry);
4230 inode_unlock(path.dentry->d_inode);
4231 mnt_drop_write(path.mnt);
4234 if (retry_estale(error, lookup_flags)) {
4235 lookup_flags |= LOOKUP_REVAL;
4243 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4245 return do_rmdir(AT_FDCWD, getname(pathname));
4249 * vfs_unlink - unlink a filesystem object
4250 * @idmap: idmap of the mount the inode was found from
4251 * @dir: parent directory
4253 * @delegated_inode: returns victim inode, if the inode is delegated.
4255 * The caller must hold dir->i_mutex.
4257 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4258 * return a reference to the inode in delegated_inode. The caller
4259 * should then break the delegation on that inode and retry. Because
4260 * breaking a delegation may take a long time, the caller should drop
4261 * dir->i_mutex before doing so.
4263 * Alternatively, a caller may pass NULL for delegated_inode. This may
4264 * be appropriate for callers that expect the underlying filesystem not
4265 * to be NFS exported.
4267 * If the inode has been found through an idmapped mount the idmap of
4268 * the vfsmount must be passed through @idmap. This function will then take
4269 * care to map the inode according to @idmap before checking permissions.
4270 * On non-idmapped mounts or if permission checking is to be performed on the
4271 * raw inode simply passs @nop_mnt_idmap.
4273 int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir,
4274 struct dentry *dentry, struct inode **delegated_inode)
4276 struct inode *target = dentry->d_inode;
4277 int error = may_delete(idmap, dir, dentry, 0);
4282 if (!dir->i_op->unlink)
4286 if (IS_SWAPFILE(target))
4288 else if (is_local_mountpoint(dentry))
4291 error = security_inode_unlink(dir, dentry);
4293 error = try_break_deleg(target, delegated_inode);
4296 error = dir->i_op->unlink(dir, dentry);
4299 detach_mounts(dentry);
4304 inode_unlock(target);
4306 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4307 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4308 fsnotify_unlink(dir, dentry);
4309 } else if (!error) {
4310 fsnotify_link_count(target);
4311 d_delete_notify(dir, dentry);
4316 EXPORT_SYMBOL(vfs_unlink);
4319 * Make sure that the actual truncation of the file will occur outside its
4320 * directory's i_mutex. Truncate can take a long time if there is a lot of
4321 * writeout happening, and we don't want to prevent access to the directory
4322 * while waiting on the I/O.
4324 int do_unlinkat(int dfd, struct filename *name)
4327 struct dentry *dentry;
4331 struct inode *inode = NULL;
4332 struct inode *delegated_inode = NULL;
4333 unsigned int lookup_flags = 0;
4335 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4340 if (type != LAST_NORM)
4343 error = mnt_want_write(path.mnt);
4347 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4348 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4349 error = PTR_ERR(dentry);
4350 if (!IS_ERR(dentry)) {
4352 /* Why not before? Because we want correct error value */
4353 if (last.name[last.len])
4355 inode = dentry->d_inode;
4356 if (d_is_negative(dentry))
4359 error = security_path_unlink(&path, dentry);
4362 error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4363 dentry, &delegated_inode);
4367 inode_unlock(path.dentry->d_inode);
4369 iput(inode); /* truncate the inode here */
4371 if (delegated_inode) {
4372 error = break_deleg_wait(&delegated_inode);
4376 mnt_drop_write(path.mnt);
4379 if (retry_estale(error, lookup_flags)) {
4380 lookup_flags |= LOOKUP_REVAL;
4389 if (d_is_negative(dentry))
4391 else if (d_is_dir(dentry))
4398 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4400 if ((flag & ~AT_REMOVEDIR) != 0)
4403 if (flag & AT_REMOVEDIR)
4404 return do_rmdir(dfd, getname(pathname));
4405 return do_unlinkat(dfd, getname(pathname));
4408 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4410 return do_unlinkat(AT_FDCWD, getname(pathname));
4414 * vfs_symlink - create symlink
4415 * @idmap: idmap of the mount the inode was found from
4416 * @dir: inode of @dentry
4417 * @dentry: pointer to dentry of the base directory
4418 * @oldname: name of the file to link to
4422 * If the inode has been found through an idmapped mount the idmap of
4423 * the vfsmount must be passed through @idmap. This function will then take
4424 * care to map the inode according to @idmap before checking permissions.
4425 * On non-idmapped mounts or if permission checking is to be performed on the
4426 * raw inode simply passs @nop_mnt_idmap.
4428 int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
4429 struct dentry *dentry, const char *oldname)
4433 error = may_create(idmap, dir, dentry);
4437 if (!dir->i_op->symlink)
4440 error = security_inode_symlink(dir, dentry, oldname);
4444 error = dir->i_op->symlink(idmap, dir, dentry, oldname);
4446 fsnotify_create(dir, dentry);
4449 EXPORT_SYMBOL(vfs_symlink);
4451 int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4454 struct dentry *dentry;
4456 unsigned int lookup_flags = 0;
4459 error = PTR_ERR(from);
4463 dentry = filename_create(newdfd, to, &path, lookup_flags);
4464 error = PTR_ERR(dentry);
4468 error = security_path_symlink(&path, dentry, from->name);
4470 error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4471 dentry, from->name);
4472 done_path_create(&path, dentry);
4473 if (retry_estale(error, lookup_flags)) {
4474 lookup_flags |= LOOKUP_REVAL;
4483 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4484 int, newdfd, const char __user *, newname)
4486 return do_symlinkat(getname(oldname), newdfd, getname(newname));
4489 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4491 return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4495 * vfs_link - create a new link
4496 * @old_dentry: object to be linked
4497 * @idmap: idmap of the mount
4499 * @new_dentry: where to create the new link
4500 * @delegated_inode: returns inode needing a delegation break
4502 * The caller must hold dir->i_mutex
4504 * If vfs_link discovers a delegation on the to-be-linked file in need
4505 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4506 * inode in delegated_inode. The caller should then break the delegation
4507 * and retry. Because breaking a delegation may take a long time, the
4508 * caller should drop the i_mutex before doing so.
4510 * Alternatively, a caller may pass NULL for delegated_inode. This may
4511 * be appropriate for callers that expect the underlying filesystem not
4512 * to be NFS exported.
4514 * If the inode has been found through an idmapped mount the idmap of
4515 * the vfsmount must be passed through @idmap. This function will then take
4516 * care to map the inode according to @idmap before checking permissions.
4517 * On non-idmapped mounts or if permission checking is to be performed on the
4518 * raw inode simply passs @nop_mnt_idmap.
4520 int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap,
4521 struct inode *dir, struct dentry *new_dentry,
4522 struct inode **delegated_inode)
4524 struct inode *inode = old_dentry->d_inode;
4525 unsigned max_links = dir->i_sb->s_max_links;
4531 error = may_create(idmap, dir, new_dentry);
4535 if (dir->i_sb != inode->i_sb)
4539 * A link to an append-only or immutable file cannot be created.
4541 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4544 * Updating the link count will likely cause i_uid and i_gid to
4545 * be writen back improperly if their true value is unknown to
4548 if (HAS_UNMAPPED_ID(idmap, inode))
4550 if (!dir->i_op->link)
4552 if (S_ISDIR(inode->i_mode))
4555 error = security_inode_link(old_dentry, dir, new_dentry);
4560 /* Make sure we don't allow creating hardlink to an unlinked file */
4561 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4563 else if (max_links && inode->i_nlink >= max_links)
4566 error = try_break_deleg(inode, delegated_inode);
4568 error = dir->i_op->link(old_dentry, dir, new_dentry);
4571 if (!error && (inode->i_state & I_LINKABLE)) {
4572 spin_lock(&inode->i_lock);
4573 inode->i_state &= ~I_LINKABLE;
4574 spin_unlock(&inode->i_lock);
4576 inode_unlock(inode);
4578 fsnotify_link(dir, inode, new_dentry);
4581 EXPORT_SYMBOL(vfs_link);
4584 * Hardlinks are often used in delicate situations. We avoid
4585 * security-related surprises by not following symlinks on the
4588 * We don't follow them on the oldname either to be compatible
4589 * with linux 2.0, and to avoid hard-linking to directories
4590 * and other special files. --ADM
4592 int do_linkat(int olddfd, struct filename *old, int newdfd,
4593 struct filename *new, int flags)
4595 struct mnt_idmap *idmap;
4596 struct dentry *new_dentry;
4597 struct path old_path, new_path;
4598 struct inode *delegated_inode = NULL;
4602 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4607 * To use null names we require CAP_DAC_READ_SEARCH
4608 * This ensures that not everyone will be able to create
4609 * handlink using the passed filedescriptor.
4611 if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4616 if (flags & AT_SYMLINK_FOLLOW)
4617 how |= LOOKUP_FOLLOW;
4619 error = filename_lookup(olddfd, old, how, &old_path, NULL);
4623 new_dentry = filename_create(newdfd, new, &new_path,
4624 (how & LOOKUP_REVAL));
4625 error = PTR_ERR(new_dentry);
4626 if (IS_ERR(new_dentry))
4630 if (old_path.mnt != new_path.mnt)
4632 idmap = mnt_idmap(new_path.mnt);
4633 error = may_linkat(idmap, &old_path);
4634 if (unlikely(error))
4636 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4639 error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode,
4640 new_dentry, &delegated_inode);
4642 done_path_create(&new_path, new_dentry);
4643 if (delegated_inode) {
4644 error = break_deleg_wait(&delegated_inode);
4646 path_put(&old_path);
4650 if (retry_estale(error, how)) {
4651 path_put(&old_path);
4652 how |= LOOKUP_REVAL;
4656 path_put(&old_path);
4664 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4665 int, newdfd, const char __user *, newname, int, flags)
4667 return do_linkat(olddfd, getname_uflags(oldname, flags),
4668 newdfd, getname(newname), flags);
4671 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4673 return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4677 * vfs_rename - rename a filesystem object
4678 * @rd: pointer to &struct renamedata info
4680 * The caller must hold multiple mutexes--see lock_rename()).
4682 * If vfs_rename discovers a delegation in need of breaking at either
4683 * the source or destination, it will return -EWOULDBLOCK and return a
4684 * reference to the inode in delegated_inode. The caller should then
4685 * break the delegation and retry. Because breaking a delegation may
4686 * take a long time, the caller should drop all locks before doing
4689 * Alternatively, a caller may pass NULL for delegated_inode. This may
4690 * be appropriate for callers that expect the underlying filesystem not
4691 * to be NFS exported.
4693 * The worst of all namespace operations - renaming directory. "Perverted"
4694 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4697 * a) we can get into loop creation.
4698 * b) race potential - two innocent renames can create a loop together.
4699 * That's where 4.4 screws up. Current fix: serialization on
4700 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4702 * c) we have to lock _four_ objects - parents and victim (if it exists),
4703 * and source (if it is not a directory).
4704 * And that - after we got ->i_mutex on parents (until then we don't know
4705 * whether the target exists). Solution: try to be smart with locking
4706 * order for inodes. We rely on the fact that tree topology may change
4707 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4708 * move will be locked. Thus we can rank directories by the tree
4709 * (ancestors first) and rank all non-directories after them.
4710 * That works since everybody except rename does "lock parent, lookup,
4711 * lock child" and rename is under ->s_vfs_rename_mutex.
4712 * HOWEVER, it relies on the assumption that any object with ->lookup()
4713 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4714 * we'd better make sure that there's no link(2) for them.
4715 * d) conversion from fhandle to dentry may come in the wrong moment - when
4716 * we are removing the target. Solution: we will have to grab ->i_mutex
4717 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4718 * ->i_mutex on parents, which works but leads to some truly excessive
4721 int vfs_rename(struct renamedata *rd)
4724 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4725 struct dentry *old_dentry = rd->old_dentry;
4726 struct dentry *new_dentry = rd->new_dentry;
4727 struct inode **delegated_inode = rd->delegated_inode;
4728 unsigned int flags = rd->flags;
4729 bool is_dir = d_is_dir(old_dentry);
4730 struct inode *source = old_dentry->d_inode;
4731 struct inode *target = new_dentry->d_inode;
4732 bool new_is_dir = false;
4733 unsigned max_links = new_dir->i_sb->s_max_links;
4734 struct name_snapshot old_name;
4736 if (source == target)
4739 error = may_delete(rd->old_mnt_idmap, old_dir, old_dentry, is_dir);
4744 error = may_create(rd->new_mnt_idmap, new_dir, new_dentry);
4746 new_is_dir = d_is_dir(new_dentry);
4748 if (!(flags & RENAME_EXCHANGE))
4749 error = may_delete(rd->new_mnt_idmap, new_dir,
4750 new_dentry, is_dir);
4752 error = may_delete(rd->new_mnt_idmap, new_dir,
4753 new_dentry, new_is_dir);
4758 if (!old_dir->i_op->rename)
4762 * If we are going to change the parent - check write permissions,
4763 * we'll need to flip '..'.
4765 if (new_dir != old_dir) {
4767 error = inode_permission(rd->old_mnt_idmap, source,
4772 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4773 error = inode_permission(rd->new_mnt_idmap, target,
4780 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4785 take_dentry_name_snapshot(&old_name, old_dentry);
4787 if (!is_dir || (flags & RENAME_EXCHANGE))
4788 lock_two_nondirectories(source, target);
4793 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4797 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4800 if (max_links && new_dir != old_dir) {
4802 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4804 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4805 old_dir->i_nlink >= max_links)
4809 error = try_break_deleg(source, delegated_inode);
4813 if (target && !new_is_dir) {
4814 error = try_break_deleg(target, delegated_inode);
4818 error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry,
4819 new_dir, new_dentry, flags);
4823 if (!(flags & RENAME_EXCHANGE) && target) {
4825 shrink_dcache_parent(new_dentry);
4826 target->i_flags |= S_DEAD;
4828 dont_mount(new_dentry);
4829 detach_mounts(new_dentry);
4831 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4832 if (!(flags & RENAME_EXCHANGE))
4833 d_move(old_dentry, new_dentry);
4835 d_exchange(old_dentry, new_dentry);
4838 if (!is_dir || (flags & RENAME_EXCHANGE))
4839 unlock_two_nondirectories(source, target);
4841 inode_unlock(target);
4844 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4845 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4846 if (flags & RENAME_EXCHANGE) {
4847 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4848 new_is_dir, NULL, new_dentry);
4851 release_dentry_name_snapshot(&old_name);
4855 EXPORT_SYMBOL(vfs_rename);
4857 int do_renameat2(int olddfd, struct filename *from, int newdfd,
4858 struct filename *to, unsigned int flags)
4860 struct renamedata rd;
4861 struct dentry *old_dentry, *new_dentry;
4862 struct dentry *trap;
4863 struct path old_path, new_path;
4864 struct qstr old_last, new_last;
4865 int old_type, new_type;
4866 struct inode *delegated_inode = NULL;
4867 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4868 bool should_retry = false;
4869 int error = -EINVAL;
4871 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4874 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4875 (flags & RENAME_EXCHANGE))
4878 if (flags & RENAME_EXCHANGE)
4882 error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4883 &old_last, &old_type);
4887 error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4893 if (old_path.mnt != new_path.mnt)
4897 if (old_type != LAST_NORM)
4900 if (flags & RENAME_NOREPLACE)
4902 if (new_type != LAST_NORM)
4905 error = mnt_want_write(old_path.mnt);
4910 trap = lock_rename(new_path.dentry, old_path.dentry);
4912 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4913 error = PTR_ERR(old_dentry);
4914 if (IS_ERR(old_dentry))
4916 /* source must exist */
4918 if (d_is_negative(old_dentry))
4920 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4921 error = PTR_ERR(new_dentry);
4922 if (IS_ERR(new_dentry))
4925 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4927 if (flags & RENAME_EXCHANGE) {
4929 if (d_is_negative(new_dentry))
4932 if (!d_is_dir(new_dentry)) {
4934 if (new_last.name[new_last.len])
4938 /* unless the source is a directory trailing slashes give -ENOTDIR */
4939 if (!d_is_dir(old_dentry)) {
4941 if (old_last.name[old_last.len])
4943 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4946 /* source should not be ancestor of target */
4948 if (old_dentry == trap)
4950 /* target should not be an ancestor of source */
4951 if (!(flags & RENAME_EXCHANGE))
4953 if (new_dentry == trap)
4956 error = security_path_rename(&old_path, old_dentry,
4957 &new_path, new_dentry, flags);
4961 rd.old_dir = old_path.dentry->d_inode;
4962 rd.old_dentry = old_dentry;
4963 rd.old_mnt_idmap = mnt_idmap(old_path.mnt);
4964 rd.new_dir = new_path.dentry->d_inode;
4965 rd.new_dentry = new_dentry;
4966 rd.new_mnt_idmap = mnt_idmap(new_path.mnt);
4967 rd.delegated_inode = &delegated_inode;
4969 error = vfs_rename(&rd);
4975 unlock_rename(new_path.dentry, old_path.dentry);
4976 if (delegated_inode) {
4977 error = break_deleg_wait(&delegated_inode);
4981 mnt_drop_write(old_path.mnt);
4983 if (retry_estale(error, lookup_flags))
4984 should_retry = true;
4985 path_put(&new_path);
4987 path_put(&old_path);
4989 should_retry = false;
4990 lookup_flags |= LOOKUP_REVAL;
4999 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
5000 int, newdfd, const char __user *, newname, unsigned int, flags)
5002 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5006 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
5007 int, newdfd, const char __user *, newname)
5009 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5013 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5015 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
5016 getname(newname), 0);
5019 int readlink_copy(char __user *buffer, int buflen, const char *link)
5021 int len = PTR_ERR(link);
5026 if (len > (unsigned) buflen)
5028 if (copy_to_user(buffer, link, len))
5035 * vfs_readlink - copy symlink body into userspace buffer
5036 * @dentry: dentry on which to get symbolic link
5037 * @buffer: user memory pointer
5038 * @buflen: size of buffer
5040 * Does not touch atime. That's up to the caller if necessary
5042 * Does not call security hook.
5044 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5046 struct inode *inode = d_inode(dentry);
5047 DEFINE_DELAYED_CALL(done);
5051 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5052 if (unlikely(inode->i_op->readlink))
5053 return inode->i_op->readlink(dentry, buffer, buflen);
5055 if (!d_is_symlink(dentry))
5058 spin_lock(&inode->i_lock);
5059 inode->i_opflags |= IOP_DEFAULT_READLINK;
5060 spin_unlock(&inode->i_lock);
5063 link = READ_ONCE(inode->i_link);
5065 link = inode->i_op->get_link(dentry, inode, &done);
5067 return PTR_ERR(link);
5069 res = readlink_copy(buffer, buflen, link);
5070 do_delayed_call(&done);
5073 EXPORT_SYMBOL(vfs_readlink);
5076 * vfs_get_link - get symlink body
5077 * @dentry: dentry on which to get symbolic link
5078 * @done: caller needs to free returned data with this
5080 * Calls security hook and i_op->get_link() on the supplied inode.
5082 * It does not touch atime. That's up to the caller if necessary.
5084 * Does not work on "special" symlinks like /proc/$$/fd/N
5086 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5088 const char *res = ERR_PTR(-EINVAL);
5089 struct inode *inode = d_inode(dentry);
5091 if (d_is_symlink(dentry)) {
5092 res = ERR_PTR(security_inode_readlink(dentry));
5094 res = inode->i_op->get_link(dentry, inode, done);
5098 EXPORT_SYMBOL(vfs_get_link);
5100 /* get the link contents into pagecache */
5101 const char *page_get_link(struct dentry *dentry, struct inode *inode,
5102 struct delayed_call *callback)
5106 struct address_space *mapping = inode->i_mapping;
5109 page = find_get_page(mapping, 0);
5111 return ERR_PTR(-ECHILD);
5112 if (!PageUptodate(page)) {
5114 return ERR_PTR(-ECHILD);
5117 page = read_mapping_page(mapping, 0, NULL);
5121 set_delayed_call(callback, page_put_link, page);
5122 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5123 kaddr = page_address(page);
5124 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5128 EXPORT_SYMBOL(page_get_link);
5130 void page_put_link(void *arg)
5134 EXPORT_SYMBOL(page_put_link);
5136 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5138 DEFINE_DELAYED_CALL(done);
5139 int res = readlink_copy(buffer, buflen,
5140 page_get_link(dentry, d_inode(dentry),
5142 do_delayed_call(&done);
5145 EXPORT_SYMBOL(page_readlink);
5147 int page_symlink(struct inode *inode, const char *symname, int len)
5149 struct address_space *mapping = inode->i_mapping;
5150 const struct address_space_operations *aops = mapping->a_ops;
5151 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5153 void *fsdata = NULL;
5159 flags = memalloc_nofs_save();
5160 err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5162 memalloc_nofs_restore(flags);
5166 memcpy(page_address(page), symname, len-1);
5168 err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5175 mark_inode_dirty(inode);
5180 EXPORT_SYMBOL(page_symlink);
5182 const struct inode_operations page_symlink_inode_operations = {
5183 .get_link = page_get_link,
5185 EXPORT_SYMBOL(page_symlink_inode_operations);