1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * Some corrections by tytso.
12 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
15 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
18 #include <linux/init.h>
19 #include <linux/export.h>
20 #include <linux/kernel.h>
21 #include <linux/slab.h>
23 #include <linux/namei.h>
24 #include <linux/pagemap.h>
25 #include <linux/fsnotify.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/ima.h>
29 #include <linux/syscalls.h>
30 #include <linux/mount.h>
31 #include <linux/audit.h>
32 #include <linux/capability.h>
33 #include <linux/file.h>
34 #include <linux/fcntl.h>
35 #include <linux/device_cgroup.h>
36 #include <linux/fs_struct.h>
37 #include <linux/posix_acl.h>
38 #include <linux/hash.h>
39 #include <linux/bitops.h>
40 #include <linux/init_task.h>
41 #include <linux/uaccess.h>
46 /* [Feb-1997 T. Schoebel-Theuer]
47 * Fundamental changes in the pathname lookup mechanisms (namei)
48 * were necessary because of omirr. The reason is that omirr needs
49 * to know the _real_ pathname, not the user-supplied one, in case
50 * of symlinks (and also when transname replacements occur).
52 * The new code replaces the old recursive symlink resolution with
53 * an iterative one (in case of non-nested symlink chains). It does
54 * this with calls to <fs>_follow_link().
55 * As a side effect, dir_namei(), _namei() and follow_link() are now
56 * replaced with a single function lookup_dentry() that can handle all
57 * the special cases of the former code.
59 * With the new dcache, the pathname is stored at each inode, at least as
60 * long as the refcount of the inode is positive. As a side effect, the
61 * size of the dcache depends on the inode cache and thus is dynamic.
63 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 * resolution to correspond with current state of the code.
66 * Note that the symlink resolution is not *completely* iterative.
67 * There is still a significant amount of tail- and mid- recursion in
68 * the algorithm. Also, note that <fs>_readlink() is not used in
69 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 * may return different results than <fs>_follow_link(). Many virtual
71 * filesystems (including /proc) exhibit this behavior.
74 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 * and the name already exists in form of a symlink, try to create the new
77 * name indicated by the symlink. The old code always complained that the
78 * name already exists, due to not following the symlink even if its target
79 * is nonexistent. The new semantics affects also mknod() and link() when
80 * the name is a symlink pointing to a non-existent name.
82 * I don't know which semantics is the right one, since I have no access
83 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 * "old" one. Personally, I think the new semantics is much more logical.
86 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 * file does succeed in both HP-UX and SunOs, but not in Solaris
88 * and in the old Linux semantics.
91 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 * semantics. See the comments in "open_namei" and "do_link" below.
94 * [10-Sep-98 Alan Modra] Another symlink change.
97 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 * inside the path - always follow.
99 * in the last component in creation/removal/renaming - never follow.
100 * if LOOKUP_FOLLOW passed - follow.
101 * if the pathname has trailing slashes - follow.
102 * otherwise - don't follow.
103 * (applied in that order).
105 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 * During the 2.4 we need to fix the userland stuff depending on it -
108 * hopefully we will be able to get rid of that wart in 2.5. So far only
109 * XEmacs seems to be relying on it...
112 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 * any extra contention...
117 /* In order to reduce some races, while at the same time doing additional
118 * checking and hopefully speeding things up, we copy filenames to the
119 * kernel data space before using them..
121 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 * PATH_MAX includes the nul terminator --RR.
125 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
128 getname_flags(const char __user *filename, int flags, int *empty)
130 struct filename *result;
134 result = audit_reusename(filename);
138 result = __getname();
139 if (unlikely(!result))
140 return ERR_PTR(-ENOMEM);
143 * First, try to embed the struct filename inside the names_cache
146 kname = (char *)result->iname;
147 result->name = kname;
149 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
150 if (unlikely(len < 0)) {
156 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 * separate struct filename so we can dedicate the entire
158 * names_cache allocation for the pathname, and re-do the copy from
161 if (unlikely(len == EMBEDDED_NAME_MAX)) {
162 const size_t size = offsetof(struct filename, iname[1]);
163 kname = (char *)result;
166 * size is chosen that way we to guarantee that
167 * result->iname[0] is within the same object and that
168 * kname can't be equal to result->iname, no matter what.
170 result = kzalloc(size, GFP_KERNEL);
171 if (unlikely(!result)) {
173 return ERR_PTR(-ENOMEM);
175 result->name = kname;
176 len = strncpy_from_user(kname, filename, PATH_MAX);
177 if (unlikely(len < 0)) {
182 if (unlikely(len == PATH_MAX)) {
185 return ERR_PTR(-ENAMETOOLONG);
190 /* The empty path is special. */
191 if (unlikely(!len)) {
194 if (!(flags & LOOKUP_EMPTY)) {
196 return ERR_PTR(-ENOENT);
200 result->uptr = filename;
201 result->aname = NULL;
202 audit_getname(result);
207 getname(const char __user * filename)
209 return getname_flags(filename, 0, NULL);
213 getname_kernel(const char * filename)
215 struct filename *result;
216 int len = strlen(filename) + 1;
218 result = __getname();
219 if (unlikely(!result))
220 return ERR_PTR(-ENOMEM);
222 if (len <= EMBEDDED_NAME_MAX) {
223 result->name = (char *)result->iname;
224 } else if (len <= PATH_MAX) {
225 const size_t size = offsetof(struct filename, iname[1]);
226 struct filename *tmp;
228 tmp = kmalloc(size, GFP_KERNEL);
229 if (unlikely(!tmp)) {
231 return ERR_PTR(-ENOMEM);
233 tmp->name = (char *)result;
237 return ERR_PTR(-ENAMETOOLONG);
239 memcpy((char *)result->name, filename, len);
241 result->aname = NULL;
243 audit_getname(result);
248 void putname(struct filename *name)
250 BUG_ON(name->refcnt <= 0);
252 if (--name->refcnt > 0)
255 if (name->name != name->iname) {
256 __putname(name->name);
262 static int check_acl(struct inode *inode, int mask)
264 #ifdef CONFIG_FS_POSIX_ACL
265 struct posix_acl *acl;
267 if (mask & MAY_NOT_BLOCK) {
268 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
271 /* no ->get_acl() calls in RCU mode... */
272 if (is_uncached_acl(acl))
274 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
277 acl = get_acl(inode, ACL_TYPE_ACCESS);
281 int error = posix_acl_permission(inode, acl, mask);
282 posix_acl_release(acl);
291 * This does the basic permission checking
293 static int acl_permission_check(struct inode *inode, int mask)
295 unsigned int mode = inode->i_mode;
297 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
300 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
301 int error = check_acl(inode, mask);
302 if (error != -EAGAIN)
306 if (in_group_p(inode->i_gid))
311 * If the DACs are ok we don't need any capability check.
313 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
319 * generic_permission - check for access rights on a Posix-like filesystem
320 * @inode: inode to check access rights for
321 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
323 * Used to check for read/write/execute permissions on a file.
324 * We use "fsuid" for this, letting us set arbitrary permissions
325 * for filesystem access without changing the "normal" uids which
326 * are used for other things.
328 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
329 * request cannot be satisfied (eg. requires blocking or too much complexity).
330 * It would then be called again in ref-walk mode.
332 int generic_permission(struct inode *inode, int mask)
337 * Do the basic permission checks.
339 ret = acl_permission_check(inode, mask);
343 if (S_ISDIR(inode->i_mode)) {
344 /* DACs are overridable for directories */
345 if (!(mask & MAY_WRITE))
346 if (capable_wrt_inode_uidgid(inode,
347 CAP_DAC_READ_SEARCH))
349 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
355 * Searching includes executable on directories, else just read.
357 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
358 if (mask == MAY_READ)
359 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
362 * Read/write DACs are always overridable.
363 * Executable DACs are overridable when there is
364 * at least one exec bit set.
366 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
367 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
372 EXPORT_SYMBOL(generic_permission);
375 * We _really_ want to just do "generic_permission()" without
376 * even looking at the inode->i_op values. So we keep a cache
377 * flag in inode->i_opflags, that says "this has not special
378 * permission function, use the fast case".
380 static inline int do_inode_permission(struct inode *inode, int mask)
382 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
383 if (likely(inode->i_op->permission))
384 return inode->i_op->permission(inode, mask);
386 /* This gets set once for the inode lifetime */
387 spin_lock(&inode->i_lock);
388 inode->i_opflags |= IOP_FASTPERM;
389 spin_unlock(&inode->i_lock);
391 return generic_permission(inode, mask);
395 * sb_permission - Check superblock-level permissions
396 * @sb: Superblock of inode to check permission on
397 * @inode: Inode to check permission on
398 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
400 * Separate out file-system wide checks from inode-specific permission checks.
402 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
404 if (unlikely(mask & MAY_WRITE)) {
405 umode_t mode = inode->i_mode;
407 /* Nobody gets write access to a read-only fs. */
408 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
415 * inode_permission - Check for access rights to a given inode
416 * @inode: Inode to check permission on
417 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
419 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
420 * this, letting us set arbitrary permissions for filesystem access without
421 * changing the "normal" UIDs which are used for other things.
423 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
425 int inode_permission(struct inode *inode, int mask)
429 retval = sb_permission(inode->i_sb, inode, mask);
433 if (unlikely(mask & MAY_WRITE)) {
435 * Nobody gets write access to an immutable file.
437 if (IS_IMMUTABLE(inode))
441 * Updating mtime will likely cause i_uid and i_gid to be
442 * written back improperly if their true value is unknown
445 if (HAS_UNMAPPED_ID(inode))
449 retval = do_inode_permission(inode, mask);
453 retval = devcgroup_inode_permission(inode, mask);
457 return security_inode_permission(inode, mask);
459 EXPORT_SYMBOL(inode_permission);
462 * path_get - get a reference to a path
463 * @path: path to get the reference to
465 * Given a path increment the reference count to the dentry and the vfsmount.
467 void path_get(const struct path *path)
472 EXPORT_SYMBOL(path_get);
475 * path_put - put a reference to a path
476 * @path: path to put the reference to
478 * Given a path decrement the reference count to the dentry and the vfsmount.
480 void path_put(const struct path *path)
485 EXPORT_SYMBOL(path_put);
487 #define EMBEDDED_LEVELS 2
492 struct inode *inode; /* path.dentry.d_inode */
494 unsigned seq, m_seq, r_seq;
497 int total_link_count;
500 struct delayed_call done;
503 } *stack, internal[EMBEDDED_LEVELS];
504 struct filename *name;
505 struct nameidata *saved;
506 struct inode *link_inode;
509 } __randomize_layout;
511 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
513 struct nameidata *old = current->nameidata;
514 p->stack = p->internal;
517 p->total_link_count = old ? old->total_link_count : 0;
519 current->nameidata = p;
522 static void restore_nameidata(void)
524 struct nameidata *now = current->nameidata, *old = now->saved;
526 current->nameidata = old;
528 old->total_link_count = now->total_link_count;
529 if (now->stack != now->internal)
533 static int __nd_alloc_stack(struct nameidata *nd)
537 if (nd->flags & LOOKUP_RCU) {
538 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
543 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
548 memcpy(p, nd->internal, sizeof(nd->internal));
554 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
555 * @path: nameidate to verify
557 * Rename can sometimes move a file or directory outside of a bind
558 * mount, path_connected allows those cases to be detected.
560 static bool path_connected(const struct path *path)
562 struct vfsmount *mnt = path->mnt;
563 struct super_block *sb = mnt->mnt_sb;
565 /* Bind mounts and multi-root filesystems can have disconnected paths */
566 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
569 return is_subdir(path->dentry, mnt->mnt_root);
572 static inline int nd_alloc_stack(struct nameidata *nd)
574 if (likely(nd->depth != EMBEDDED_LEVELS))
576 if (likely(nd->stack != nd->internal))
578 return __nd_alloc_stack(nd);
581 static void drop_links(struct nameidata *nd)
585 struct saved *last = nd->stack + i;
586 do_delayed_call(&last->done);
587 clear_delayed_call(&last->done);
591 static void terminate_walk(struct nameidata *nd)
594 if (!(nd->flags & LOOKUP_RCU)) {
597 for (i = 0; i < nd->depth; i++)
598 path_put(&nd->stack[i].link);
599 if (nd->flags & LOOKUP_ROOT_GRABBED) {
601 nd->flags &= ~LOOKUP_ROOT_GRABBED;
604 nd->flags &= ~LOOKUP_RCU;
610 /* path_put is needed afterwards regardless of success or failure */
611 static bool legitimize_path(struct nameidata *nd,
612 struct path *path, unsigned seq)
614 int res = __legitimize_mnt(path->mnt, nd->m_seq);
621 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
625 return !read_seqcount_retry(&path->dentry->d_seq, seq);
628 static bool legitimize_links(struct nameidata *nd)
631 for (i = 0; i < nd->depth; i++) {
632 struct saved *last = nd->stack + i;
633 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
642 static bool legitimize_root(struct nameidata *nd)
645 * For scoped-lookups (where nd->root has been zeroed), we need to
646 * restart the whole lookup from scratch -- because set_root() is wrong
647 * for these lookups (nd->dfd is the root, not the filesystem root).
649 if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED))
651 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
652 if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT))
654 nd->flags |= LOOKUP_ROOT_GRABBED;
655 return legitimize_path(nd, &nd->root, nd->root_seq);
659 * Path walking has 2 modes, rcu-walk and ref-walk (see
660 * Documentation/filesystems/path-lookup.txt). In situations when we can't
661 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
662 * normal reference counts on dentries and vfsmounts to transition to ref-walk
663 * mode. Refcounts are grabbed at the last known good point before rcu-walk
664 * got stuck, so ref-walk may continue from there. If this is not successful
665 * (eg. a seqcount has changed), then failure is returned and it's up to caller
666 * to restart the path walk from the beginning in ref-walk mode.
670 * unlazy_walk - try to switch to ref-walk mode.
671 * @nd: nameidata pathwalk data
672 * Returns: 0 on success, -ECHILD on failure
674 * unlazy_walk attempts to legitimize the current nd->path and nd->root
676 * Must be called from rcu-walk context.
677 * Nothing should touch nameidata between unlazy_walk() failure and
680 static int unlazy_walk(struct nameidata *nd)
682 struct dentry *parent = nd->path.dentry;
684 BUG_ON(!(nd->flags & LOOKUP_RCU));
686 nd->flags &= ~LOOKUP_RCU;
687 if (unlikely(!legitimize_links(nd)))
689 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
691 if (unlikely(!legitimize_root(nd)))
694 BUG_ON(nd->inode != parent->d_inode);
699 nd->path.dentry = NULL;
706 * unlazy_child - try to switch to ref-walk mode.
707 * @nd: nameidata pathwalk data
708 * @dentry: child of nd->path.dentry
709 * @seq: seq number to check dentry against
710 * Returns: 0 on success, -ECHILD on failure
712 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
713 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
714 * @nd. Must be called from rcu-walk context.
715 * Nothing should touch nameidata between unlazy_child() failure and
718 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
720 BUG_ON(!(nd->flags & LOOKUP_RCU));
722 nd->flags &= ~LOOKUP_RCU;
723 if (unlikely(!legitimize_links(nd)))
725 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
727 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
731 * We need to move both the parent and the dentry from the RCU domain
732 * to be properly refcounted. And the sequence number in the dentry
733 * validates *both* dentry counters, since we checked the sequence
734 * number of the parent after we got the child sequence number. So we
735 * know the parent must still be valid if the child sequence number is
737 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
739 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
742 * Sequence counts matched. Now make sure that the root is
743 * still valid and get it if required.
745 if (unlikely(!legitimize_root(nd)))
753 nd->path.dentry = NULL;
763 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
765 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
766 return dentry->d_op->d_revalidate(dentry, flags);
772 * complete_walk - successful completion of path walk
773 * @nd: pointer nameidata
775 * If we had been in RCU mode, drop out of it and legitimize nd->path.
776 * Revalidate the final result, unless we'd already done that during
777 * the path walk or the filesystem doesn't ask for it. Return 0 on
778 * success, -error on failure. In case of failure caller does not
779 * need to drop nd->path.
781 static int complete_walk(struct nameidata *nd)
783 struct dentry *dentry = nd->path.dentry;
786 if (nd->flags & LOOKUP_RCU) {
788 * We don't want to zero nd->root for scoped-lookups or
789 * externally-managed nd->root.
791 if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED)))
793 if (unlikely(unlazy_walk(nd)))
797 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
799 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
800 * ever step outside the root during lookup" and should already
801 * be guaranteed by the rest of namei, we want to avoid a namei
802 * BUG resulting in userspace being given a path that was not
803 * scoped within the root at some point during the lookup.
805 * So, do a final sanity-check to make sure that in the
806 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
807 * we won't silently return an fd completely outside of the
808 * requested root to userspace.
810 * Userspace could move the path outside the root after this
811 * check, but as discussed elsewhere this is not a concern (the
812 * resolved file was inside the root at some point).
814 if (!path_is_under(&nd->path, &nd->root))
818 if (likely(!(nd->flags & LOOKUP_JUMPED)))
821 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
824 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
834 static int set_root(struct nameidata *nd)
836 struct fs_struct *fs = current->fs;
839 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
840 * still have to ensure it doesn't happen because it will cause a breakout
843 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
844 return -ENOTRECOVERABLE;
846 if (nd->flags & LOOKUP_RCU) {
850 seq = read_seqcount_begin(&fs->seq);
852 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
853 } while (read_seqcount_retry(&fs->seq, seq));
855 get_fs_root(fs, &nd->root);
856 nd->flags |= LOOKUP_ROOT_GRABBED;
861 static void path_put_conditional(struct path *path, struct nameidata *nd)
864 if (path->mnt != nd->path.mnt)
868 static inline void path_to_nameidata(const struct path *path,
869 struct nameidata *nd)
871 if (!(nd->flags & LOOKUP_RCU)) {
872 dput(nd->path.dentry);
873 if (nd->path.mnt != path->mnt)
874 mntput(nd->path.mnt);
876 nd->path.mnt = path->mnt;
877 nd->path.dentry = path->dentry;
880 static int nd_jump_root(struct nameidata *nd)
882 if (unlikely(nd->flags & LOOKUP_BENEATH))
884 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
885 /* Absolute path arguments to path_init() are allowed. */
886 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
890 int error = set_root(nd);
894 if (nd->flags & LOOKUP_RCU) {
898 nd->inode = d->d_inode;
899 nd->seq = nd->root_seq;
900 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
906 nd->inode = nd->path.dentry->d_inode;
908 nd->flags |= LOOKUP_JUMPED;
913 * Helper to directly jump to a known parsed path from ->get_link,
914 * caller must have taken a reference to path beforehand.
916 int nd_jump_link(struct path *path)
919 struct nameidata *nd = current->nameidata;
921 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
925 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
926 if (nd->path.mnt != path->mnt)
929 /* Not currently safe for scoped-lookups. */
930 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
935 nd->inode = nd->path.dentry->d_inode;
936 nd->flags |= LOOKUP_JUMPED;
944 static inline void put_link(struct nameidata *nd)
946 struct saved *last = nd->stack + --nd->depth;
947 do_delayed_call(&last->done);
948 if (!(nd->flags & LOOKUP_RCU))
949 path_put(&last->link);
952 int sysctl_protected_symlinks __read_mostly = 0;
953 int sysctl_protected_hardlinks __read_mostly = 0;
954 int sysctl_protected_fifos __read_mostly;
955 int sysctl_protected_regular __read_mostly;
958 * may_follow_link - Check symlink following for unsafe situations
959 * @nd: nameidata pathwalk data
961 * In the case of the sysctl_protected_symlinks sysctl being enabled,
962 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
963 * in a sticky world-writable directory. This is to protect privileged
964 * processes from failing races against path names that may change out
965 * from under them by way of other users creating malicious symlinks.
966 * It will permit symlinks to be followed only when outside a sticky
967 * world-writable directory, or when the uid of the symlink and follower
968 * match, or when the directory owner matches the symlink's owner.
970 * Returns 0 if following the symlink is allowed, -ve on error.
972 static inline int may_follow_link(struct nameidata *nd)
974 const struct inode *inode;
975 const struct inode *parent;
978 if (!sysctl_protected_symlinks)
981 /* Allowed if owner and follower match. */
982 inode = nd->link_inode;
983 if (uid_eq(current_cred()->fsuid, inode->i_uid))
986 /* Allowed if parent directory not sticky and world-writable. */
988 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
991 /* Allowed if parent directory and link owner match. */
992 puid = parent->i_uid;
993 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
996 if (nd->flags & LOOKUP_RCU)
999 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1000 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1005 * safe_hardlink_source - Check for safe hardlink conditions
1006 * @inode: the source inode to hardlink from
1008 * Return false if at least one of the following conditions:
1009 * - inode is not a regular file
1011 * - inode is setgid and group-exec
1012 * - access failure for read and write
1014 * Otherwise returns true.
1016 static bool safe_hardlink_source(struct inode *inode)
1018 umode_t mode = inode->i_mode;
1020 /* Special files should not get pinned to the filesystem. */
1024 /* Setuid files should not get pinned to the filesystem. */
1028 /* Executable setgid files should not get pinned to the filesystem. */
1029 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1032 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1033 if (inode_permission(inode, MAY_READ | MAY_WRITE))
1040 * may_linkat - Check permissions for creating a hardlink
1041 * @link: the source to hardlink from
1043 * Block hardlink when all of:
1044 * - sysctl_protected_hardlinks enabled
1045 * - fsuid does not match inode
1046 * - hardlink source is unsafe (see safe_hardlink_source() above)
1047 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1049 * Returns 0 if successful, -ve on error.
1051 static int may_linkat(struct path *link)
1053 struct inode *inode = link->dentry->d_inode;
1055 /* Inode writeback is not safe when the uid or gid are invalid. */
1056 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
1059 if (!sysctl_protected_hardlinks)
1062 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1063 * otherwise, it must be a safe source.
1065 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
1068 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1073 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1074 * should be allowed, or not, on files that already
1076 * @dir_mode: mode bits of directory
1077 * @dir_uid: owner of directory
1078 * @inode: the inode of the file to open
1080 * Block an O_CREAT open of a FIFO (or a regular file) when:
1081 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1082 * - the file already exists
1083 * - we are in a sticky directory
1084 * - we don't own the file
1085 * - the owner of the directory doesn't own the file
1086 * - the directory is world writable
1087 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1088 * the directory doesn't have to be world writable: being group writable will
1091 * Returns 0 if the open is allowed, -ve on error.
1093 static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid,
1094 struct inode * const inode)
1096 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1097 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1098 likely(!(dir_mode & S_ISVTX)) ||
1099 uid_eq(inode->i_uid, dir_uid) ||
1100 uid_eq(current_fsuid(), inode->i_uid))
1103 if (likely(dir_mode & 0002) ||
1105 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1106 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1107 const char *operation = S_ISFIFO(inode->i_mode) ?
1108 "sticky_create_fifo" :
1109 "sticky_create_regular";
1110 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1116 static __always_inline
1117 const char *get_link(struct nameidata *nd)
1119 struct saved *last = nd->stack + nd->depth - 1;
1120 struct dentry *dentry = last->link.dentry;
1121 struct inode *inode = nd->link_inode;
1125 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS))
1126 return ERR_PTR(-ELOOP);
1128 if (!(nd->flags & LOOKUP_RCU)) {
1129 touch_atime(&last->link);
1131 } else if (atime_needs_update(&last->link, inode)) {
1132 if (unlikely(unlazy_walk(nd)))
1133 return ERR_PTR(-ECHILD);
1134 touch_atime(&last->link);
1137 error = security_inode_follow_link(dentry, inode,
1138 nd->flags & LOOKUP_RCU);
1139 if (unlikely(error))
1140 return ERR_PTR(error);
1142 nd->last_type = LAST_BIND;
1143 res = READ_ONCE(inode->i_link);
1145 const char * (*get)(struct dentry *, struct inode *,
1146 struct delayed_call *);
1147 get = inode->i_op->get_link;
1148 if (nd->flags & LOOKUP_RCU) {
1149 res = get(NULL, inode, &last->done);
1150 if (res == ERR_PTR(-ECHILD)) {
1151 if (unlikely(unlazy_walk(nd)))
1152 return ERR_PTR(-ECHILD);
1153 res = get(dentry, inode, &last->done);
1156 res = get(dentry, inode, &last->done);
1158 if (IS_ERR_OR_NULL(res))
1162 error = nd_jump_root(nd);
1163 if (unlikely(error))
1164 return ERR_PTR(error);
1165 while (unlikely(*++res == '/'))
1174 * follow_up - Find the mountpoint of path's vfsmount
1176 * Given a path, find the mountpoint of its source file system.
1177 * Replace @path with the path of the mountpoint in the parent mount.
1180 * Return 1 if we went up a level and 0 if we were already at the
1183 int follow_up(struct path *path)
1185 struct mount *mnt = real_mount(path->mnt);
1186 struct mount *parent;
1187 struct dentry *mountpoint;
1189 read_seqlock_excl(&mount_lock);
1190 parent = mnt->mnt_parent;
1191 if (parent == mnt) {
1192 read_sequnlock_excl(&mount_lock);
1195 mntget(&parent->mnt);
1196 mountpoint = dget(mnt->mnt_mountpoint);
1197 read_sequnlock_excl(&mount_lock);
1199 path->dentry = mountpoint;
1201 path->mnt = &parent->mnt;
1204 EXPORT_SYMBOL(follow_up);
1207 * Perform an automount
1208 * - return -EISDIR to tell follow_managed() to stop and return the path we
1211 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1213 struct dentry *dentry = path->dentry;
1215 /* We don't want to mount if someone's just doing a stat -
1216 * unless they're stat'ing a directory and appended a '/' to
1219 * We do, however, want to mount if someone wants to open or
1220 * create a file of any type under the mountpoint, wants to
1221 * traverse through the mountpoint or wants to open the
1222 * mounted directory. Also, autofs may mark negative dentries
1223 * as being automount points. These will need the attentions
1224 * of the daemon to instantiate them before they can be used.
1226 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1227 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1231 if (count && (*count)++ >= MAXSYMLINKS)
1234 return finish_automount(dentry->d_op->d_automount(path), path);
1238 * Handle a dentry that is managed in some way.
1239 * - Flagged for transit management (autofs)
1240 * - Flagged as mountpoint
1241 * - Flagged as automount point
1243 * This may only be called in refwalk mode.
1244 * On success path->dentry is known positive.
1246 * Serialization is taken care of in namespace.c
1248 static int follow_managed(struct path *path, struct nameidata *nd)
1250 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1252 bool need_mntput = false;
1255 /* Given that we're not holding a lock here, we retain the value in a
1256 * local variable for each dentry as we look at it so that we don't see
1257 * the components of that value change under us */
1258 while (flags = smp_load_acquire(&path->dentry->d_flags),
1259 unlikely(flags & DCACHE_MANAGED_DENTRY)) {
1260 /* Allow the filesystem to manage the transit without i_mutex
1262 if (flags & DCACHE_MANAGE_TRANSIT) {
1263 BUG_ON(!path->dentry->d_op);
1264 BUG_ON(!path->dentry->d_op->d_manage);
1265 ret = path->dentry->d_op->d_manage(path, false);
1266 flags = smp_load_acquire(&path->dentry->d_flags);
1271 /* Transit to a mounted filesystem. */
1272 if (flags & DCACHE_MOUNTED) {
1273 struct vfsmount *mounted = lookup_mnt(path);
1278 path->mnt = mounted;
1279 path->dentry = dget(mounted->mnt_root);
1284 /* Something is mounted on this dentry in another
1285 * namespace and/or whatever was mounted there in this
1286 * namespace got unmounted before lookup_mnt() could
1290 /* Handle an automount point */
1291 if (flags & DCACHE_NEED_AUTOMOUNT) {
1292 ret = follow_automount(path, &nd->total_link_count,
1299 /* We didn't change the current path point */
1304 if (path->mnt == mnt)
1306 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1309 nd->flags |= LOOKUP_JUMPED;
1311 if (ret == -EISDIR || !ret)
1313 if (ret > 0 && unlikely(d_flags_negative(flags)))
1315 if (unlikely(ret < 0))
1316 path_put_conditional(path, nd);
1320 int follow_down_one(struct path *path)
1322 struct vfsmount *mounted;
1324 mounted = lookup_mnt(path);
1328 path->mnt = mounted;
1329 path->dentry = dget(mounted->mnt_root);
1334 EXPORT_SYMBOL(follow_down_one);
1336 static inline int managed_dentry_rcu(const struct path *path)
1338 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1339 path->dentry->d_op->d_manage(path, true) : 0;
1343 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1344 * we meet a managed dentry that would need blocking.
1346 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1347 struct inode **inode, unsigned *seqp)
1350 struct mount *mounted;
1352 * Don't forget we might have a non-mountpoint managed dentry
1353 * that wants to block transit.
1355 switch (managed_dentry_rcu(path)) {
1365 if (!d_mountpoint(path->dentry))
1366 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1368 mounted = __lookup_mnt(path->mnt, path->dentry);
1371 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1373 path->mnt = &mounted->mnt;
1374 path->dentry = mounted->mnt.mnt_root;
1375 nd->flags |= LOOKUP_JUMPED;
1376 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1378 * Update the inode too. We don't need to re-check the
1379 * dentry sequence number here after this d_inode read,
1380 * because a mount-point is always pinned.
1382 *inode = path->dentry->d_inode;
1384 return !read_seqretry(&mount_lock, nd->m_seq) &&
1385 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1388 static int follow_dotdot_rcu(struct nameidata *nd)
1390 struct inode *inode = nd->inode;
1393 if (path_equal(&nd->path, &nd->root)) {
1394 if (unlikely(nd->flags & LOOKUP_BENEATH))
1398 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1399 struct dentry *old = nd->path.dentry;
1400 struct dentry *parent = old->d_parent;
1403 inode = parent->d_inode;
1404 seq = read_seqcount_begin(&parent->d_seq);
1405 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1407 nd->path.dentry = parent;
1409 if (unlikely(!path_connected(&nd->path)))
1413 struct mount *mnt = real_mount(nd->path.mnt);
1414 struct mount *mparent = mnt->mnt_parent;
1415 struct dentry *mountpoint = mnt->mnt_mountpoint;
1416 struct inode *inode2 = mountpoint->d_inode;
1417 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1418 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1420 if (&mparent->mnt == nd->path.mnt)
1422 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1424 /* we know that mountpoint was pinned */
1425 nd->path.dentry = mountpoint;
1426 nd->path.mnt = &mparent->mnt;
1431 while (unlikely(d_mountpoint(nd->path.dentry))) {
1432 struct mount *mounted;
1433 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1434 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1438 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1440 nd->path.mnt = &mounted->mnt;
1441 nd->path.dentry = mounted->mnt.mnt_root;
1442 inode = nd->path.dentry->d_inode;
1443 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1450 * Follow down to the covering mount currently visible to userspace. At each
1451 * point, the filesystem owning that dentry may be queried as to whether the
1452 * caller is permitted to proceed or not.
1454 int follow_down(struct path *path)
1459 while (managed = READ_ONCE(path->dentry->d_flags),
1460 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1461 /* Allow the filesystem to manage the transit without i_mutex
1464 * We indicate to the filesystem if someone is trying to mount
1465 * something here. This gives autofs the chance to deny anyone
1466 * other than its daemon the right to mount on its
1469 * The filesystem may sleep at this point.
1471 if (managed & DCACHE_MANAGE_TRANSIT) {
1472 BUG_ON(!path->dentry->d_op);
1473 BUG_ON(!path->dentry->d_op->d_manage);
1474 ret = path->dentry->d_op->d_manage(path, false);
1476 return ret == -EISDIR ? 0 : ret;
1479 /* Transit to a mounted filesystem. */
1480 if (managed & DCACHE_MOUNTED) {
1481 struct vfsmount *mounted = lookup_mnt(path);
1486 path->mnt = mounted;
1487 path->dentry = dget(mounted->mnt_root);
1491 /* Don't handle automount points here */
1496 EXPORT_SYMBOL(follow_down);
1499 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1501 static void follow_mount(struct path *path)
1503 while (d_mountpoint(path->dentry)) {
1504 struct vfsmount *mounted = lookup_mnt(path);
1509 path->mnt = mounted;
1510 path->dentry = dget(mounted->mnt_root);
1514 static int path_parent_directory(struct path *path)
1516 struct dentry *old = path->dentry;
1517 /* rare case of legitimate dget_parent()... */
1518 path->dentry = dget_parent(path->dentry);
1520 if (unlikely(!path_connected(path)))
1525 static int follow_dotdot(struct nameidata *nd)
1528 if (path_equal(&nd->path, &nd->root)) {
1529 if (unlikely(nd->flags & LOOKUP_BENEATH))
1533 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1534 int ret = path_parent_directory(&nd->path);
1539 if (!follow_up(&nd->path))
1541 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1544 follow_mount(&nd->path);
1545 nd->inode = nd->path.dentry->d_inode;
1550 * This looks up the name in dcache and possibly revalidates the found dentry.
1551 * NULL is returned if the dentry does not exist in the cache.
1553 static struct dentry *lookup_dcache(const struct qstr *name,
1557 struct dentry *dentry = d_lookup(dir, name);
1559 int error = d_revalidate(dentry, flags);
1560 if (unlikely(error <= 0)) {
1562 d_invalidate(dentry);
1564 return ERR_PTR(error);
1571 * Parent directory has inode locked exclusive. This is one
1572 * and only case when ->lookup() gets called on non in-lookup
1573 * dentries - as the matter of fact, this only gets called
1574 * when directory is guaranteed to have no in-lookup children
1577 static struct dentry *__lookup_hash(const struct qstr *name,
1578 struct dentry *base, unsigned int flags)
1580 struct dentry *dentry = lookup_dcache(name, base, flags);
1582 struct inode *dir = base->d_inode;
1587 /* Don't create child dentry for a dead directory. */
1588 if (unlikely(IS_DEADDIR(dir)))
1589 return ERR_PTR(-ENOENT);
1591 dentry = d_alloc(base, name);
1592 if (unlikely(!dentry))
1593 return ERR_PTR(-ENOMEM);
1595 old = dir->i_op->lookup(dir, dentry, flags);
1596 if (unlikely(old)) {
1603 static int lookup_fast(struct nameidata *nd,
1604 struct path *path, struct inode **inode,
1607 struct vfsmount *mnt = nd->path.mnt;
1608 struct dentry *dentry, *parent = nd->path.dentry;
1613 * Rename seqlock is not required here because in the off chance
1614 * of a false negative due to a concurrent rename, the caller is
1615 * going to fall back to non-racy lookup.
1617 if (nd->flags & LOOKUP_RCU) {
1620 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1621 if (unlikely(!dentry)) {
1622 if (unlazy_walk(nd))
1628 * This sequence count validates that the inode matches
1629 * the dentry name information from lookup.
1631 *inode = d_backing_inode(dentry);
1632 negative = d_is_negative(dentry);
1633 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1637 * This sequence count validates that the parent had no
1638 * changes while we did the lookup of the dentry above.
1640 * The memory barrier in read_seqcount_begin of child is
1641 * enough, we can use __read_seqcount_retry here.
1643 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1647 status = d_revalidate(dentry, nd->flags);
1648 if (likely(status > 0)) {
1650 * Note: do negative dentry check after revalidation in
1651 * case that drops it.
1653 if (unlikely(negative))
1656 path->dentry = dentry;
1657 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1660 if (unlazy_child(nd, dentry, seq))
1662 if (unlikely(status == -ECHILD))
1663 /* we'd been told to redo it in non-rcu mode */
1664 status = d_revalidate(dentry, nd->flags);
1666 dentry = __d_lookup(parent, &nd->last);
1667 if (unlikely(!dentry))
1669 status = d_revalidate(dentry, nd->flags);
1671 if (unlikely(status <= 0)) {
1673 d_invalidate(dentry);
1679 path->dentry = dentry;
1680 err = follow_managed(path, nd);
1681 if (likely(err > 0))
1682 *inode = d_backing_inode(path->dentry);
1686 /* Fast lookup failed, do it the slow way */
1687 static struct dentry *__lookup_slow(const struct qstr *name,
1691 struct dentry *dentry, *old;
1692 struct inode *inode = dir->d_inode;
1693 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1695 /* Don't go there if it's already dead */
1696 if (unlikely(IS_DEADDIR(inode)))
1697 return ERR_PTR(-ENOENT);
1699 dentry = d_alloc_parallel(dir, name, &wq);
1702 if (unlikely(!d_in_lookup(dentry))) {
1703 int error = d_revalidate(dentry, flags);
1704 if (unlikely(error <= 0)) {
1706 d_invalidate(dentry);
1711 dentry = ERR_PTR(error);
1714 old = inode->i_op->lookup(inode, dentry, flags);
1715 d_lookup_done(dentry);
1716 if (unlikely(old)) {
1724 static struct dentry *lookup_slow(const struct qstr *name,
1728 struct inode *inode = dir->d_inode;
1730 inode_lock_shared(inode);
1731 res = __lookup_slow(name, dir, flags);
1732 inode_unlock_shared(inode);
1736 static inline int may_lookup(struct nameidata *nd)
1738 if (nd->flags & LOOKUP_RCU) {
1739 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1742 if (unlazy_walk(nd))
1745 return inode_permission(nd->inode, MAY_EXEC);
1748 static inline int handle_dots(struct nameidata *nd, int type)
1750 if (type == LAST_DOTDOT) {
1753 if (!nd->root.mnt) {
1754 error = set_root(nd);
1758 if (nd->flags & LOOKUP_RCU)
1759 error = follow_dotdot_rcu(nd);
1761 error = follow_dotdot(nd);
1765 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1767 * If there was a racing rename or mount along our
1768 * path, then we can't be sure that ".." hasn't jumped
1769 * above nd->root (and so userspace should retry or use
1773 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1775 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1782 static int pick_link(struct nameidata *nd, struct path *link,
1783 struct inode *inode, unsigned seq)
1787 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1788 path_to_nameidata(link, nd);
1791 if (!(nd->flags & LOOKUP_RCU)) {
1792 if (link->mnt == nd->path.mnt)
1795 error = nd_alloc_stack(nd);
1796 if (unlikely(error)) {
1797 if (error == -ECHILD) {
1798 if (unlikely(!legitimize_path(nd, link, seq))) {
1801 nd->flags &= ~LOOKUP_RCU;
1802 nd->path.mnt = NULL;
1803 nd->path.dentry = NULL;
1805 } else if (likely(unlazy_walk(nd)) == 0)
1806 error = nd_alloc_stack(nd);
1814 last = nd->stack + nd->depth++;
1816 clear_delayed_call(&last->done);
1817 nd->link_inode = inode;
1822 enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1825 * Do we need to follow links? We _really_ want to be able
1826 * to do this check without having to look at inode->i_op,
1827 * so we keep a cache of "no, this doesn't need follow_link"
1828 * for the common case.
1830 static inline int step_into(struct nameidata *nd, struct path *path,
1831 int flags, struct inode *inode, unsigned seq)
1833 if (!(flags & WALK_MORE) && nd->depth)
1835 if (likely(!d_is_symlink(path->dentry)) ||
1836 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1837 /* not a symlink or should not follow */
1838 path_to_nameidata(path, nd);
1843 /* make sure that d_is_symlink above matches inode */
1844 if (nd->flags & LOOKUP_RCU) {
1845 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1848 return pick_link(nd, path, inode, seq);
1851 static int walk_component(struct nameidata *nd, int flags)
1854 struct inode *inode;
1858 * "." and ".." are special - ".." especially so because it has
1859 * to be able to know about the current root directory and
1860 * parent relationships.
1862 if (unlikely(nd->last_type != LAST_NORM)) {
1863 err = handle_dots(nd, nd->last_type);
1864 if (!(flags & WALK_MORE) && nd->depth)
1868 err = lookup_fast(nd, &path, &inode, &seq);
1869 if (unlikely(err <= 0)) {
1872 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1874 if (IS_ERR(path.dentry))
1875 return PTR_ERR(path.dentry);
1877 path.mnt = nd->path.mnt;
1878 err = follow_managed(&path, nd);
1879 if (unlikely(err < 0))
1882 seq = 0; /* we are already out of RCU mode */
1883 inode = d_backing_inode(path.dentry);
1886 return step_into(nd, &path, flags, inode, seq);
1890 * We can do the critical dentry name comparison and hashing
1891 * operations one word at a time, but we are limited to:
1893 * - Architectures with fast unaligned word accesses. We could
1894 * do a "get_unaligned()" if this helps and is sufficiently
1897 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1898 * do not trap on the (extremely unlikely) case of a page
1899 * crossing operation.
1901 * - Furthermore, we need an efficient 64-bit compile for the
1902 * 64-bit case in order to generate the "number of bytes in
1903 * the final mask". Again, that could be replaced with a
1904 * efficient population count instruction or similar.
1906 #ifdef CONFIG_DCACHE_WORD_ACCESS
1908 #include <asm/word-at-a-time.h>
1912 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1914 #elif defined(CONFIG_64BIT)
1916 * Register pressure in the mixing function is an issue, particularly
1917 * on 32-bit x86, but almost any function requires one state value and
1918 * one temporary. Instead, use a function designed for two state values
1919 * and no temporaries.
1921 * This function cannot create a collision in only two iterations, so
1922 * we have two iterations to achieve avalanche. In those two iterations,
1923 * we have six layers of mixing, which is enough to spread one bit's
1924 * influence out to 2^6 = 64 state bits.
1926 * Rotate constants are scored by considering either 64 one-bit input
1927 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1928 * probability of that delta causing a change to each of the 128 output
1929 * bits, using a sample of random initial states.
1931 * The Shannon entropy of the computed probabilities is then summed
1932 * to produce a score. Ideally, any input change has a 50% chance of
1933 * toggling any given output bit.
1935 * Mixing scores (in bits) for (12,45):
1936 * Input delta: 1-bit 2-bit
1937 * 1 round: 713.3 42542.6
1938 * 2 rounds: 2753.7 140389.8
1939 * 3 rounds: 5954.1 233458.2
1940 * 4 rounds: 7862.6 256672.2
1941 * Perfect: 8192 258048
1942 * (64*128) (64*63/2 * 128)
1944 #define HASH_MIX(x, y, a) \
1946 y ^= x, x = rol64(x,12),\
1947 x += y, y = rol64(y,45),\
1951 * Fold two longs into one 32-bit hash value. This must be fast, but
1952 * latency isn't quite as critical, as there is a fair bit of additional
1953 * work done before the hash value is used.
1955 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1957 y ^= x * GOLDEN_RATIO_64;
1958 y *= GOLDEN_RATIO_64;
1962 #else /* 32-bit case */
1965 * Mixing scores (in bits) for (7,20):
1966 * Input delta: 1-bit 2-bit
1967 * 1 round: 330.3 9201.6
1968 * 2 rounds: 1246.4 25475.4
1969 * 3 rounds: 1907.1 31295.1
1970 * 4 rounds: 2042.3 31718.6
1971 * Perfect: 2048 31744
1972 * (32*64) (32*31/2 * 64)
1974 #define HASH_MIX(x, y, a) \
1976 y ^= x, x = rol32(x, 7),\
1977 x += y, y = rol32(y,20),\
1980 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1982 /* Use arch-optimized multiply if one exists */
1983 return __hash_32(y ^ __hash_32(x));
1989 * Return the hash of a string of known length. This is carfully
1990 * designed to match hash_name(), which is the more critical function.
1991 * In particular, we must end by hashing a final word containing 0..7
1992 * payload bytes, to match the way that hash_name() iterates until it
1993 * finds the delimiter after the name.
1995 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1997 unsigned long a, x = 0, y = (unsigned long)salt;
2002 a = load_unaligned_zeropad(name);
2003 if (len < sizeof(unsigned long))
2006 name += sizeof(unsigned long);
2007 len -= sizeof(unsigned long);
2009 x ^= a & bytemask_from_count(len);
2011 return fold_hash(x, y);
2013 EXPORT_SYMBOL(full_name_hash);
2015 /* Return the "hash_len" (hash and length) of a null-terminated string */
2016 u64 hashlen_string(const void *salt, const char *name)
2018 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2019 unsigned long adata, mask, len;
2020 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2027 len += sizeof(unsigned long);
2029 a = load_unaligned_zeropad(name+len);
2030 } while (!has_zero(a, &adata, &constants));
2032 adata = prep_zero_mask(a, adata, &constants);
2033 mask = create_zero_mask(adata);
2034 x ^= a & zero_bytemask(mask);
2036 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2038 EXPORT_SYMBOL(hashlen_string);
2041 * Calculate the length and hash of the path component, and
2042 * return the "hash_len" as the result.
2044 static inline u64 hash_name(const void *salt, const char *name)
2046 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2047 unsigned long adata, bdata, mask, len;
2048 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2055 len += sizeof(unsigned long);
2057 a = load_unaligned_zeropad(name+len);
2058 b = a ^ REPEAT_BYTE('/');
2059 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2061 adata = prep_zero_mask(a, adata, &constants);
2062 bdata = prep_zero_mask(b, bdata, &constants);
2063 mask = create_zero_mask(adata | bdata);
2064 x ^= a & zero_bytemask(mask);
2066 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2069 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2071 /* Return the hash of a string of known length */
2072 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2074 unsigned long hash = init_name_hash(salt);
2076 hash = partial_name_hash((unsigned char)*name++, hash);
2077 return end_name_hash(hash);
2079 EXPORT_SYMBOL(full_name_hash);
2081 /* Return the "hash_len" (hash and length) of a null-terminated string */
2082 u64 hashlen_string(const void *salt, const char *name)
2084 unsigned long hash = init_name_hash(salt);
2085 unsigned long len = 0, c;
2087 c = (unsigned char)*name;
2090 hash = partial_name_hash(c, hash);
2091 c = (unsigned char)name[len];
2093 return hashlen_create(end_name_hash(hash), len);
2095 EXPORT_SYMBOL(hashlen_string);
2098 * We know there's a real path component here of at least
2101 static inline u64 hash_name(const void *salt, const char *name)
2103 unsigned long hash = init_name_hash(salt);
2104 unsigned long len = 0, c;
2106 c = (unsigned char)*name;
2109 hash = partial_name_hash(c, hash);
2110 c = (unsigned char)name[len];
2111 } while (c && c != '/');
2112 return hashlen_create(end_name_hash(hash), len);
2119 * This is the basic name resolution function, turning a pathname into
2120 * the final dentry. We expect 'base' to be positive and a directory.
2122 * Returns 0 and nd will have valid dentry and mnt on success.
2123 * Returns error and drops reference to input namei data on failure.
2125 static int link_path_walk(const char *name, struct nameidata *nd)
2130 return PTR_ERR(name);
2136 /* At this point we know we have a real path component. */
2141 err = may_lookup(nd);
2145 hash_len = hash_name(nd->path.dentry, name);
2148 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2150 if (name[1] == '.') {
2152 nd->flags |= LOOKUP_JUMPED;
2158 if (likely(type == LAST_NORM)) {
2159 struct dentry *parent = nd->path.dentry;
2160 nd->flags &= ~LOOKUP_JUMPED;
2161 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2162 struct qstr this = { { .hash_len = hash_len }, .name = name };
2163 err = parent->d_op->d_hash(parent, &this);
2166 hash_len = this.hash_len;
2171 nd->last.hash_len = hash_len;
2172 nd->last.name = name;
2173 nd->last_type = type;
2175 name += hashlen_len(hash_len);
2179 * If it wasn't NUL, we know it was '/'. Skip that
2180 * slash, and continue until no more slashes.
2184 } while (unlikely(*name == '/'));
2185 if (unlikely(!*name)) {
2187 /* pathname body, done */
2190 name = nd->stack[nd->depth - 1].name;
2191 /* trailing symlink, done */
2194 /* last component of nested symlink */
2195 err = walk_component(nd, WALK_FOLLOW);
2197 /* not the last component */
2198 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2204 const char *s = get_link(nd);
2213 nd->stack[nd->depth - 1].name = name;
2218 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2219 if (nd->flags & LOOKUP_RCU) {
2220 if (unlazy_walk(nd))
2228 /* must be paired with terminate_walk() */
2229 static const char *path_init(struct nameidata *nd, unsigned flags)
2232 const char *s = nd->name->name;
2235 flags &= ~LOOKUP_RCU;
2236 if (flags & LOOKUP_RCU)
2239 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2240 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2243 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2244 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2247 if (flags & LOOKUP_ROOT) {
2248 struct dentry *root = nd->root.dentry;
2249 struct inode *inode = root->d_inode;
2250 if (*s && unlikely(!d_can_lookup(root)))
2251 return ERR_PTR(-ENOTDIR);
2252 nd->path = nd->root;
2254 if (flags & LOOKUP_RCU) {
2255 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2256 nd->root_seq = nd->seq;
2258 path_get(&nd->path);
2263 nd->root.mnt = NULL;
2264 nd->path.mnt = NULL;
2265 nd->path.dentry = NULL;
2267 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2268 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2269 error = nd_jump_root(nd);
2270 if (unlikely(error))
2271 return ERR_PTR(error);
2275 /* Relative pathname -- get the starting-point it is relative to. */
2276 if (nd->dfd == AT_FDCWD) {
2277 if (flags & LOOKUP_RCU) {
2278 struct fs_struct *fs = current->fs;
2282 seq = read_seqcount_begin(&fs->seq);
2284 nd->inode = nd->path.dentry->d_inode;
2285 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2286 } while (read_seqcount_retry(&fs->seq, seq));
2288 get_fs_pwd(current->fs, &nd->path);
2289 nd->inode = nd->path.dentry->d_inode;
2292 /* Caller must check execute permissions on the starting path component */
2293 struct fd f = fdget_raw(nd->dfd);
2294 struct dentry *dentry;
2297 return ERR_PTR(-EBADF);
2299 dentry = f.file->f_path.dentry;
2301 if (*s && unlikely(!d_can_lookup(dentry))) {
2303 return ERR_PTR(-ENOTDIR);
2306 nd->path = f.file->f_path;
2307 if (flags & LOOKUP_RCU) {
2308 nd->inode = nd->path.dentry->d_inode;
2309 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2311 path_get(&nd->path);
2312 nd->inode = nd->path.dentry->d_inode;
2317 /* For scoped-lookups we need to set the root to the dirfd as well. */
2318 if (flags & LOOKUP_IS_SCOPED) {
2319 nd->root = nd->path;
2320 if (flags & LOOKUP_RCU) {
2321 nd->root_seq = nd->seq;
2323 path_get(&nd->root);
2324 nd->flags |= LOOKUP_ROOT_GRABBED;
2330 static const char *trailing_symlink(struct nameidata *nd)
2333 int error = may_follow_link(nd);
2334 if (unlikely(error))
2335 return ERR_PTR(error);
2336 nd->flags |= LOOKUP_PARENT;
2337 nd->stack[0].name = NULL;
2342 static inline int lookup_last(struct nameidata *nd)
2344 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2345 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2347 nd->flags &= ~LOOKUP_PARENT;
2348 return walk_component(nd, 0);
2351 static int handle_lookup_down(struct nameidata *nd)
2353 struct path path = nd->path;
2354 struct inode *inode = nd->inode;
2355 unsigned seq = nd->seq;
2358 if (nd->flags & LOOKUP_RCU) {
2360 * don't bother with unlazy_walk on failure - we are
2361 * at the very beginning of walk, so we lose nothing
2362 * if we simply redo everything in non-RCU mode
2364 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2368 err = follow_managed(&path, nd);
2369 if (unlikely(err < 0))
2371 inode = d_backing_inode(path.dentry);
2374 path_to_nameidata(&path, nd);
2380 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2381 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2383 const char *s = path_init(nd, flags);
2386 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2387 err = handle_lookup_down(nd);
2388 if (unlikely(err < 0))
2392 while (!(err = link_path_walk(s, nd))
2393 && ((err = lookup_last(nd)) > 0)) {
2394 s = trailing_symlink(nd);
2397 err = complete_walk(nd);
2399 if (!err && nd->flags & LOOKUP_DIRECTORY)
2400 if (!d_can_lookup(nd->path.dentry))
2404 nd->path.mnt = NULL;
2405 nd->path.dentry = NULL;
2411 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2412 struct path *path, struct path *root)
2415 struct nameidata nd;
2417 return PTR_ERR(name);
2418 if (unlikely(root)) {
2420 flags |= LOOKUP_ROOT;
2422 set_nameidata(&nd, dfd, name);
2423 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2424 if (unlikely(retval == -ECHILD))
2425 retval = path_lookupat(&nd, flags, path);
2426 if (unlikely(retval == -ESTALE))
2427 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2429 if (likely(!retval))
2430 audit_inode(name, path->dentry, 0);
2431 restore_nameidata();
2436 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2437 static int path_parentat(struct nameidata *nd, unsigned flags,
2438 struct path *parent)
2440 const char *s = path_init(nd, flags);
2441 int err = link_path_walk(s, nd);
2443 err = complete_walk(nd);
2446 nd->path.mnt = NULL;
2447 nd->path.dentry = NULL;
2453 static struct filename *filename_parentat(int dfd, struct filename *name,
2454 unsigned int flags, struct path *parent,
2455 struct qstr *last, int *type)
2458 struct nameidata nd;
2462 set_nameidata(&nd, dfd, name);
2463 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2464 if (unlikely(retval == -ECHILD))
2465 retval = path_parentat(&nd, flags, parent);
2466 if (unlikely(retval == -ESTALE))
2467 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2468 if (likely(!retval)) {
2470 *type = nd.last_type;
2471 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2474 name = ERR_PTR(retval);
2476 restore_nameidata();
2480 /* does lookup, returns the object with parent locked */
2481 struct dentry *kern_path_locked(const char *name, struct path *path)
2483 struct filename *filename;
2488 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2490 if (IS_ERR(filename))
2491 return ERR_CAST(filename);
2492 if (unlikely(type != LAST_NORM)) {
2495 return ERR_PTR(-EINVAL);
2497 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2498 d = __lookup_hash(&last, path->dentry, 0);
2500 inode_unlock(path->dentry->d_inode);
2507 int kern_path(const char *name, unsigned int flags, struct path *path)
2509 return filename_lookup(AT_FDCWD, getname_kernel(name),
2512 EXPORT_SYMBOL(kern_path);
2515 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2516 * @dentry: pointer to dentry of the base directory
2517 * @mnt: pointer to vfs mount of the base directory
2518 * @name: pointer to file name
2519 * @flags: lookup flags
2520 * @path: pointer to struct path to fill
2522 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2523 const char *name, unsigned int flags,
2526 struct path root = {.mnt = mnt, .dentry = dentry};
2527 /* the first argument of filename_lookup() is ignored with root */
2528 return filename_lookup(AT_FDCWD, getname_kernel(name),
2529 flags , path, &root);
2531 EXPORT_SYMBOL(vfs_path_lookup);
2533 static int lookup_one_len_common(const char *name, struct dentry *base,
2534 int len, struct qstr *this)
2538 this->hash = full_name_hash(base, name, len);
2542 if (unlikely(name[0] == '.')) {
2543 if (len < 2 || (len == 2 && name[1] == '.'))
2548 unsigned int c = *(const unsigned char *)name++;
2549 if (c == '/' || c == '\0')
2553 * See if the low-level filesystem might want
2554 * to use its own hash..
2556 if (base->d_flags & DCACHE_OP_HASH) {
2557 int err = base->d_op->d_hash(base, this);
2562 return inode_permission(base->d_inode, MAY_EXEC);
2566 * try_lookup_one_len - filesystem helper to lookup single pathname component
2567 * @name: pathname component to lookup
2568 * @base: base directory to lookup from
2569 * @len: maximum length @len should be interpreted to
2571 * Look up a dentry by name in the dcache, returning NULL if it does not
2572 * currently exist. The function does not try to create a dentry.
2574 * Note that this routine is purely a helper for filesystem usage and should
2575 * not be called by generic code.
2577 * The caller must hold base->i_mutex.
2579 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2584 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2586 err = lookup_one_len_common(name, base, len, &this);
2588 return ERR_PTR(err);
2590 return lookup_dcache(&this, base, 0);
2592 EXPORT_SYMBOL(try_lookup_one_len);
2595 * lookup_one_len - filesystem helper to lookup single pathname component
2596 * @name: pathname component to lookup
2597 * @base: base directory to lookup from
2598 * @len: maximum length @len should be interpreted to
2600 * Note that this routine is purely a helper for filesystem usage and should
2601 * not be called by generic code.
2603 * The caller must hold base->i_mutex.
2605 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2607 struct dentry *dentry;
2611 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2613 err = lookup_one_len_common(name, base, len, &this);
2615 return ERR_PTR(err);
2617 dentry = lookup_dcache(&this, base, 0);
2618 return dentry ? dentry : __lookup_slow(&this, base, 0);
2620 EXPORT_SYMBOL(lookup_one_len);
2623 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2624 * @name: pathname component to lookup
2625 * @base: base directory to lookup from
2626 * @len: maximum length @len should be interpreted to
2628 * Note that this routine is purely a helper for filesystem usage and should
2629 * not be called by generic code.
2631 * Unlike lookup_one_len, it should be called without the parent
2632 * i_mutex held, and will take the i_mutex itself if necessary.
2634 struct dentry *lookup_one_len_unlocked(const char *name,
2635 struct dentry *base, int len)
2641 err = lookup_one_len_common(name, base, len, &this);
2643 return ERR_PTR(err);
2645 ret = lookup_dcache(&this, base, 0);
2647 ret = lookup_slow(&this, base, 0);
2650 EXPORT_SYMBOL(lookup_one_len_unlocked);
2653 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2654 * on negatives. Returns known positive or ERR_PTR(); that's what
2655 * most of the users want. Note that pinned negative with unlocked parent
2656 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2657 * need to be very careful; pinned positives have ->d_inode stable, so
2658 * this one avoids such problems.
2660 struct dentry *lookup_positive_unlocked(const char *name,
2661 struct dentry *base, int len)
2663 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2664 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2666 ret = ERR_PTR(-ENOENT);
2670 EXPORT_SYMBOL(lookup_positive_unlocked);
2672 #ifdef CONFIG_UNIX98_PTYS
2673 int path_pts(struct path *path)
2675 /* Find something mounted on "pts" in the same directory as
2678 struct dentry *child, *parent;
2682 ret = path_parent_directory(path);
2686 parent = path->dentry;
2689 child = d_hash_and_lookup(parent, &this);
2693 path->dentry = child;
2700 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2701 struct path *path, int *empty)
2703 return filename_lookup(dfd, getname_flags(name, flags, empty),
2706 EXPORT_SYMBOL(user_path_at_empty);
2709 * path_mountpoint - look up a path to be umounted
2710 * @nd: lookup context
2711 * @flags: lookup flags
2712 * @path: pointer to container for result
2714 * Look up the given name, but don't attempt to revalidate the last component.
2715 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2718 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2720 const char *s = path_init(nd, flags);
2723 while (!(err = link_path_walk(s, nd)) &&
2724 (err = lookup_last(nd)) > 0) {
2725 s = trailing_symlink(nd);
2727 if (!err && (nd->flags & LOOKUP_RCU))
2728 err = unlazy_walk(nd);
2730 err = handle_lookup_down(nd);
2733 nd->path.mnt = NULL;
2734 nd->path.dentry = NULL;
2741 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2744 struct nameidata nd;
2747 return PTR_ERR(name);
2748 set_nameidata(&nd, dfd, name);
2749 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2750 if (unlikely(error == -ECHILD))
2751 error = path_mountpoint(&nd, flags, path);
2752 if (unlikely(error == -ESTALE))
2753 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2755 audit_inode(name, path->dentry, AUDIT_INODE_NOEVAL);
2756 restore_nameidata();
2762 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2763 * @dfd: directory file descriptor
2764 * @name: pathname from userland
2765 * @flags: lookup flags
2766 * @path: pointer to container to hold result
2768 * A umount is a special case for path walking. We're not actually interested
2769 * in the inode in this situation, and ESTALE errors can be a problem. We
2770 * simply want track down the dentry and vfsmount attached at the mountpoint
2771 * and avoid revalidating the last component.
2773 * Returns 0 and populates "path" on success.
2776 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2779 return filename_mountpoint(dfd, getname(name), path, flags);
2783 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2786 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2788 EXPORT_SYMBOL(kern_path_mountpoint);
2790 int __check_sticky(struct inode *dir, struct inode *inode)
2792 kuid_t fsuid = current_fsuid();
2794 if (uid_eq(inode->i_uid, fsuid))
2796 if (uid_eq(dir->i_uid, fsuid))
2798 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2800 EXPORT_SYMBOL(__check_sticky);
2803 * Check whether we can remove a link victim from directory dir, check
2804 * whether the type of victim is right.
2805 * 1. We can't do it if dir is read-only (done in permission())
2806 * 2. We should have write and exec permissions on dir
2807 * 3. We can't remove anything from append-only dir
2808 * 4. We can't do anything with immutable dir (done in permission())
2809 * 5. If the sticky bit on dir is set we should either
2810 * a. be owner of dir, or
2811 * b. be owner of victim, or
2812 * c. have CAP_FOWNER capability
2813 * 6. If the victim is append-only or immutable we can't do antyhing with
2814 * links pointing to it.
2815 * 7. If the victim has an unknown uid or gid we can't change the inode.
2816 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2817 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2818 * 10. We can't remove a root or mountpoint.
2819 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2820 * nfs_async_unlink().
2822 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2824 struct inode *inode = d_backing_inode(victim);
2827 if (d_is_negative(victim))
2831 BUG_ON(victim->d_parent->d_inode != dir);
2833 /* Inode writeback is not safe when the uid or gid are invalid. */
2834 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2837 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2839 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2845 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2846 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2849 if (!d_is_dir(victim))
2851 if (IS_ROOT(victim))
2853 } else if (d_is_dir(victim))
2855 if (IS_DEADDIR(dir))
2857 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2862 /* Check whether we can create an object with dentry child in directory
2864 * 1. We can't do it if child already exists (open has special treatment for
2865 * this case, but since we are inlined it's OK)
2866 * 2. We can't do it if dir is read-only (done in permission())
2867 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2868 * 4. We should have write and exec permissions on dir
2869 * 5. We can't do it if dir is immutable (done in permission())
2871 static inline int may_create(struct inode *dir, struct dentry *child)
2873 struct user_namespace *s_user_ns;
2874 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2877 if (IS_DEADDIR(dir))
2879 s_user_ns = dir->i_sb->s_user_ns;
2880 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2881 !kgid_has_mapping(s_user_ns, current_fsgid()))
2883 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2887 * p1 and p2 should be directories on the same fs.
2889 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2894 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2898 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2900 p = d_ancestor(p2, p1);
2902 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2903 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2907 p = d_ancestor(p1, p2);
2909 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2910 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2914 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2915 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2918 EXPORT_SYMBOL(lock_rename);
2920 void unlock_rename(struct dentry *p1, struct dentry *p2)
2922 inode_unlock(p1->d_inode);
2924 inode_unlock(p2->d_inode);
2925 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2928 EXPORT_SYMBOL(unlock_rename);
2930 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2933 int error = may_create(dir, dentry);
2937 if (!dir->i_op->create)
2938 return -EACCES; /* shouldn't it be ENOSYS? */
2941 error = security_inode_create(dir, dentry, mode);
2944 error = dir->i_op->create(dir, dentry, mode, want_excl);
2946 fsnotify_create(dir, dentry);
2949 EXPORT_SYMBOL(vfs_create);
2951 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2952 int (*f)(struct dentry *, umode_t, void *),
2955 struct inode *dir = dentry->d_parent->d_inode;
2956 int error = may_create(dir, dentry);
2962 error = security_inode_create(dir, dentry, mode);
2965 error = f(dentry, mode, arg);
2967 fsnotify_create(dir, dentry);
2970 EXPORT_SYMBOL(vfs_mkobj);
2972 bool may_open_dev(const struct path *path)
2974 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2975 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2978 static int may_open(const struct path *path, int acc_mode, int flag)
2980 struct dentry *dentry = path->dentry;
2981 struct inode *inode = dentry->d_inode;
2987 switch (inode->i_mode & S_IFMT) {
2991 if (acc_mode & MAY_WRITE)
2996 if (!may_open_dev(path))
3005 error = inode_permission(inode, MAY_OPEN | acc_mode);
3010 * An append-only file must be opened in append mode for writing.
3012 if (IS_APPEND(inode)) {
3013 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3019 /* O_NOATIME can only be set by the owner or superuser */
3020 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
3026 static int handle_truncate(struct file *filp)
3028 const struct path *path = &filp->f_path;
3029 struct inode *inode = path->dentry->d_inode;
3030 int error = get_write_access(inode);
3034 * Refuse to truncate files with mandatory locks held on them.
3036 error = locks_verify_locked(filp);
3038 error = security_path_truncate(path);
3040 error = do_truncate(path->dentry, 0,
3041 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3044 put_write_access(inode);
3048 static inline int open_to_namei_flags(int flag)
3050 if ((flag & O_ACCMODE) == 3)
3055 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
3057 struct user_namespace *s_user_ns;
3058 int error = security_path_mknod(dir, dentry, mode, 0);
3062 s_user_ns = dir->dentry->d_sb->s_user_ns;
3063 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
3064 !kgid_has_mapping(s_user_ns, current_fsgid()))
3067 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
3071 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3075 * Attempt to atomically look up, create and open a file from a negative
3078 * Returns 0 if successful. The file will have been created and attached to
3079 * @file by the filesystem calling finish_open().
3081 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3082 * be set. The caller will need to perform the open themselves. @path will
3083 * have been updated to point to the new dentry. This may be negative.
3085 * Returns an error code otherwise.
3087 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3088 struct path *path, struct file *file,
3089 const struct open_flags *op,
3090 int open_flag, umode_t mode)
3092 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3093 struct inode *dir = nd->path.dentry->d_inode;
3096 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3097 open_flag &= ~O_TRUNC;
3099 if (nd->flags & LOOKUP_DIRECTORY)
3100 open_flag |= O_DIRECTORY;
3102 file->f_path.dentry = DENTRY_NOT_SET;
3103 file->f_path.mnt = nd->path.mnt;
3104 error = dir->i_op->atomic_open(dir, dentry, file,
3105 open_to_namei_flags(open_flag), mode);
3106 d_lookup_done(dentry);
3108 if (file->f_mode & FMODE_OPENED) {
3110 * We didn't have the inode before the open, so check open
3113 int acc_mode = op->acc_mode;
3114 if (file->f_mode & FMODE_CREATED) {
3115 WARN_ON(!(open_flag & O_CREAT));
3116 fsnotify_create(dir, dentry);
3119 error = may_open(&file->f_path, acc_mode, open_flag);
3120 if (WARN_ON(error > 0))
3122 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3125 if (file->f_path.dentry) {
3127 dentry = file->f_path.dentry;
3129 if (file->f_mode & FMODE_CREATED)
3130 fsnotify_create(dir, dentry);
3131 if (unlikely(d_is_negative(dentry))) {
3134 path->dentry = dentry;
3135 path->mnt = nd->path.mnt;
3145 * Look up and maybe create and open the last component.
3147 * Must be called with parent locked (exclusive in O_CREAT case).
3149 * Returns 0 on success, that is, if
3150 * the file was successfully atomically created (if necessary) and opened, or
3151 * the file was not completely opened at this time, though lookups and
3152 * creations were performed.
3153 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3154 * In the latter case dentry returned in @path might be negative if O_CREAT
3155 * hadn't been specified.
3157 * An error code is returned on failure.
3159 static int lookup_open(struct nameidata *nd, struct path *path,
3161 const struct open_flags *op,
3164 struct dentry *dir = nd->path.dentry;
3165 struct inode *dir_inode = dir->d_inode;
3166 int open_flag = op->open_flag;
3167 struct dentry *dentry;
3168 int error, create_error = 0;
3169 umode_t mode = op->mode;
3170 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3172 if (unlikely(IS_DEADDIR(dir_inode)))
3175 file->f_mode &= ~FMODE_CREATED;
3176 dentry = d_lookup(dir, &nd->last);
3179 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3181 return PTR_ERR(dentry);
3183 if (d_in_lookup(dentry))
3186 error = d_revalidate(dentry, nd->flags);
3187 if (likely(error > 0))
3191 d_invalidate(dentry);
3195 if (dentry->d_inode) {
3196 /* Cached positive dentry: will open in f_op->open */
3201 * Checking write permission is tricky, bacuse we don't know if we are
3202 * going to actually need it: O_CREAT opens should work as long as the
3203 * file exists. But checking existence breaks atomicity. The trick is
3204 * to check access and if not granted clear O_CREAT from the flags.
3206 * Another problem is returing the "right" error value (e.g. for an
3207 * O_EXCL open we want to return EEXIST not EROFS).
3209 if (open_flag & O_CREAT) {
3210 if (!IS_POSIXACL(dir->d_inode))
3211 mode &= ~current_umask();
3212 if (unlikely(!got_write)) {
3213 create_error = -EROFS;
3214 open_flag &= ~O_CREAT;
3215 if (open_flag & (O_EXCL | O_TRUNC))
3217 /* No side effects, safe to clear O_CREAT */
3219 create_error = may_o_create(&nd->path, dentry, mode);
3221 open_flag &= ~O_CREAT;
3222 if (open_flag & O_EXCL)
3226 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3227 unlikely(!got_write)) {
3229 * No O_CREATE -> atomicity not a requirement -> fall
3230 * back to lookup + open
3235 if (dir_inode->i_op->atomic_open) {
3236 error = atomic_open(nd, dentry, path, file, op, open_flag,
3238 if (unlikely(error == -ENOENT) && create_error)
3239 error = create_error;
3244 if (d_in_lookup(dentry)) {
3245 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3247 d_lookup_done(dentry);
3248 if (unlikely(res)) {
3250 error = PTR_ERR(res);
3258 /* Negative dentry, just create the file */
3259 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3260 file->f_mode |= FMODE_CREATED;
3261 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3262 if (!dir_inode->i_op->create) {
3266 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3267 open_flag & O_EXCL);
3270 fsnotify_create(dir_inode, dentry);
3272 if (unlikely(create_error) && !dentry->d_inode) {
3273 error = create_error;
3277 path->dentry = dentry;
3278 path->mnt = nd->path.mnt;
3287 * Handle the last step of open()
3289 static int do_last(struct nameidata *nd,
3290 struct file *file, const struct open_flags *op)
3292 struct dentry *dir = nd->path.dentry;
3293 kuid_t dir_uid = nd->inode->i_uid;
3294 umode_t dir_mode = nd->inode->i_mode;
3295 int open_flag = op->open_flag;
3296 bool will_truncate = (open_flag & O_TRUNC) != 0;
3297 bool got_write = false;
3298 int acc_mode = op->acc_mode;
3300 struct inode *inode;
3304 nd->flags &= ~LOOKUP_PARENT;
3305 nd->flags |= op->intent;
3307 if (nd->last_type != LAST_NORM) {
3308 error = handle_dots(nd, nd->last_type);
3309 if (unlikely(error))
3314 if (!(open_flag & O_CREAT)) {
3315 if (nd->last.name[nd->last.len])
3316 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3317 /* we _can_ be in RCU mode here */
3318 error = lookup_fast(nd, &path, &inode, &seq);
3319 if (likely(error > 0))
3325 BUG_ON(nd->inode != dir->d_inode);
3326 BUG_ON(nd->flags & LOOKUP_RCU);
3328 /* create side of things */
3330 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3331 * has been cleared when we got to the last component we are
3334 error = complete_walk(nd);
3338 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3339 /* trailing slashes? */
3340 if (unlikely(nd->last.name[nd->last.len]))
3344 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3345 error = mnt_want_write(nd->path.mnt);
3349 * do _not_ fail yet - we might not need that or fail with
3350 * a different error; let lookup_open() decide; we'll be
3351 * dropping this one anyway.
3354 if (open_flag & O_CREAT)
3355 inode_lock(dir->d_inode);
3357 inode_lock_shared(dir->d_inode);
3358 error = lookup_open(nd, &path, file, op, got_write);
3359 if (open_flag & O_CREAT)
3360 inode_unlock(dir->d_inode);
3362 inode_unlock_shared(dir->d_inode);
3367 if (file->f_mode & FMODE_OPENED) {
3368 if ((file->f_mode & FMODE_CREATED) ||
3369 !S_ISREG(file_inode(file)->i_mode))
3370 will_truncate = false;
3372 audit_inode(nd->name, file->f_path.dentry, 0);
3376 if (file->f_mode & FMODE_CREATED) {
3377 /* Don't check for write permission, don't truncate */
3378 open_flag &= ~O_TRUNC;
3379 will_truncate = false;
3381 path_to_nameidata(&path, nd);
3382 goto finish_open_created;
3386 * If atomic_open() acquired write access it is dropped now due to
3387 * possible mount and symlink following (this might be optimized away if
3391 mnt_drop_write(nd->path.mnt);
3395 error = follow_managed(&path, nd);
3396 if (unlikely(error < 0))
3399 seq = 0; /* out of RCU mode, so the value doesn't matter */
3400 inode = d_backing_inode(path.dentry);
3402 error = step_into(nd, &path, 0, inode, seq);
3403 if (unlikely(error))
3406 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3407 audit_inode(nd->name, nd->path.dentry, 0);
3411 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3412 error = complete_walk(nd);
3415 audit_inode(nd->name, nd->path.dentry, 0);
3416 if (open_flag & O_CREAT) {
3418 if (d_is_dir(nd->path.dentry))
3420 error = may_create_in_sticky(dir_mode, dir_uid,
3421 d_backing_inode(nd->path.dentry));
3422 if (unlikely(error))
3426 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3428 if (!d_is_reg(nd->path.dentry))
3429 will_truncate = false;
3431 if (will_truncate) {
3432 error = mnt_want_write(nd->path.mnt);
3437 finish_open_created:
3438 error = may_open(&nd->path, acc_mode, open_flag);
3441 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
3442 error = vfs_open(&nd->path, file);
3446 error = ima_file_check(file, op->acc_mode);
3447 if (!error && will_truncate)
3448 error = handle_truncate(file);
3450 if (unlikely(error > 0)) {
3455 mnt_drop_write(nd->path.mnt);
3459 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3461 struct dentry *child = NULL;
3462 struct inode *dir = dentry->d_inode;
3463 struct inode *inode;
3466 /* we want directory to be writable */
3467 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3470 error = -EOPNOTSUPP;
3471 if (!dir->i_op->tmpfile)
3474 child = d_alloc(dentry, &slash_name);
3475 if (unlikely(!child))
3477 error = dir->i_op->tmpfile(dir, child, mode);
3481 inode = child->d_inode;
3482 if (unlikely(!inode))
3484 if (!(open_flag & O_EXCL)) {
3485 spin_lock(&inode->i_lock);
3486 inode->i_state |= I_LINKABLE;
3487 spin_unlock(&inode->i_lock);
3489 ima_post_create_tmpfile(inode);
3494 return ERR_PTR(error);
3496 EXPORT_SYMBOL(vfs_tmpfile);
3498 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3499 const struct open_flags *op,
3502 struct dentry *child;
3504 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3505 if (unlikely(error))
3507 error = mnt_want_write(path.mnt);
3508 if (unlikely(error))
3510 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3511 error = PTR_ERR(child);
3515 path.dentry = child;
3516 audit_inode(nd->name, child, 0);
3517 /* Don't check for other permissions, the inode was just created */
3518 error = may_open(&path, 0, op->open_flag);
3521 file->f_path.mnt = path.mnt;
3522 error = finish_open(file, child, NULL);
3524 mnt_drop_write(path.mnt);
3530 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3533 int error = path_lookupat(nd, flags, &path);
3535 audit_inode(nd->name, path.dentry, 0);
3536 error = vfs_open(&path, file);
3542 static struct file *path_openat(struct nameidata *nd,
3543 const struct open_flags *op, unsigned flags)
3548 file = alloc_empty_file(op->open_flag, current_cred());
3552 if (unlikely(file->f_flags & __O_TMPFILE)) {
3553 error = do_tmpfile(nd, flags, op, file);
3554 } else if (unlikely(file->f_flags & O_PATH)) {
3555 error = do_o_path(nd, flags, file);
3557 const char *s = path_init(nd, flags);
3558 while (!(error = link_path_walk(s, nd)) &&
3559 (error = do_last(nd, file, op)) > 0) {
3560 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3561 s = trailing_symlink(nd);
3565 if (likely(!error)) {
3566 if (likely(file->f_mode & FMODE_OPENED))
3572 if (error == -EOPENSTALE) {
3573 if (flags & LOOKUP_RCU)
3578 return ERR_PTR(error);
3581 struct file *do_filp_open(int dfd, struct filename *pathname,
3582 const struct open_flags *op)
3584 struct nameidata nd;
3585 int flags = op->lookup_flags;
3588 set_nameidata(&nd, dfd, pathname);
3589 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3590 if (unlikely(filp == ERR_PTR(-ECHILD)))
3591 filp = path_openat(&nd, op, flags);
3592 if (unlikely(filp == ERR_PTR(-ESTALE)))
3593 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3594 restore_nameidata();
3598 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3599 const char *name, const struct open_flags *op)
3601 struct nameidata nd;
3603 struct filename *filename;
3604 int flags = op->lookup_flags | LOOKUP_ROOT;
3607 nd.root.dentry = dentry;
3609 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3610 return ERR_PTR(-ELOOP);
3612 filename = getname_kernel(name);
3613 if (IS_ERR(filename))
3614 return ERR_CAST(filename);
3616 set_nameidata(&nd, -1, filename);
3617 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3618 if (unlikely(file == ERR_PTR(-ECHILD)))
3619 file = path_openat(&nd, op, flags);
3620 if (unlikely(file == ERR_PTR(-ESTALE)))
3621 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3622 restore_nameidata();
3627 static struct dentry *filename_create(int dfd, struct filename *name,
3628 struct path *path, unsigned int lookup_flags)
3630 struct dentry *dentry = ERR_PTR(-EEXIST);
3635 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3638 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3639 * other flags passed in are ignored!
3641 lookup_flags &= LOOKUP_REVAL;
3643 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3645 return ERR_CAST(name);
3648 * Yucky last component or no last component at all?
3649 * (foo/., foo/.., /////)
3651 if (unlikely(type != LAST_NORM))
3654 /* don't fail immediately if it's r/o, at least try to report other errors */
3655 err2 = mnt_want_write(path->mnt);
3657 * Do the final lookup.
3659 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3660 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3661 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3666 if (d_is_positive(dentry))
3670 * Special case - lookup gave negative, but... we had foo/bar/
3671 * From the vfs_mknod() POV we just have a negative dentry -
3672 * all is fine. Let's be bastards - you had / on the end, you've
3673 * been asking for (non-existent) directory. -ENOENT for you.
3675 if (unlikely(!is_dir && last.name[last.len])) {
3679 if (unlikely(err2)) {
3687 dentry = ERR_PTR(error);
3689 inode_unlock(path->dentry->d_inode);
3691 mnt_drop_write(path->mnt);
3698 struct dentry *kern_path_create(int dfd, const char *pathname,
3699 struct path *path, unsigned int lookup_flags)
3701 return filename_create(dfd, getname_kernel(pathname),
3702 path, lookup_flags);
3704 EXPORT_SYMBOL(kern_path_create);
3706 void done_path_create(struct path *path, struct dentry *dentry)
3709 inode_unlock(path->dentry->d_inode);
3710 mnt_drop_write(path->mnt);
3713 EXPORT_SYMBOL(done_path_create);
3715 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3716 struct path *path, unsigned int lookup_flags)
3718 return filename_create(dfd, getname(pathname), path, lookup_flags);
3720 EXPORT_SYMBOL(user_path_create);
3722 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3724 int error = may_create(dir, dentry);
3729 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3732 if (!dir->i_op->mknod)
3735 error = devcgroup_inode_mknod(mode, dev);
3739 error = security_inode_mknod(dir, dentry, mode, dev);
3743 error = dir->i_op->mknod(dir, dentry, mode, dev);
3745 fsnotify_create(dir, dentry);
3748 EXPORT_SYMBOL(vfs_mknod);
3750 static int may_mknod(umode_t mode)
3752 switch (mode & S_IFMT) {
3758 case 0: /* zero mode translates to S_IFREG */
3767 long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3770 struct dentry *dentry;
3773 unsigned int lookup_flags = 0;
3775 error = may_mknod(mode);
3779 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3781 return PTR_ERR(dentry);
3783 if (!IS_POSIXACL(path.dentry->d_inode))
3784 mode &= ~current_umask();
3785 error = security_path_mknod(&path, dentry, mode, dev);
3788 switch (mode & S_IFMT) {
3789 case 0: case S_IFREG:
3790 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3792 ima_post_path_mknod(dentry);
3794 case S_IFCHR: case S_IFBLK:
3795 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3796 new_decode_dev(dev));
3798 case S_IFIFO: case S_IFSOCK:
3799 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3803 done_path_create(&path, dentry);
3804 if (retry_estale(error, lookup_flags)) {
3805 lookup_flags |= LOOKUP_REVAL;
3811 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3814 return do_mknodat(dfd, filename, mode, dev);
3817 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3819 return do_mknodat(AT_FDCWD, filename, mode, dev);
3822 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3824 int error = may_create(dir, dentry);
3825 unsigned max_links = dir->i_sb->s_max_links;
3830 if (!dir->i_op->mkdir)
3833 mode &= (S_IRWXUGO|S_ISVTX);
3834 error = security_inode_mkdir(dir, dentry, mode);
3838 if (max_links && dir->i_nlink >= max_links)
3841 error = dir->i_op->mkdir(dir, dentry, mode);
3843 fsnotify_mkdir(dir, dentry);
3846 EXPORT_SYMBOL(vfs_mkdir);
3848 long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3850 struct dentry *dentry;
3853 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3856 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3858 return PTR_ERR(dentry);
3860 if (!IS_POSIXACL(path.dentry->d_inode))
3861 mode &= ~current_umask();
3862 error = security_path_mkdir(&path, dentry, mode);
3864 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3865 done_path_create(&path, dentry);
3866 if (retry_estale(error, lookup_flags)) {
3867 lookup_flags |= LOOKUP_REVAL;
3873 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3875 return do_mkdirat(dfd, pathname, mode);
3878 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3880 return do_mkdirat(AT_FDCWD, pathname, mode);
3883 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3885 int error = may_delete(dir, dentry, 1);
3890 if (!dir->i_op->rmdir)
3894 inode_lock(dentry->d_inode);
3897 if (is_local_mountpoint(dentry))
3900 error = security_inode_rmdir(dir, dentry);
3904 error = dir->i_op->rmdir(dir, dentry);
3908 shrink_dcache_parent(dentry);
3909 dentry->d_inode->i_flags |= S_DEAD;
3911 detach_mounts(dentry);
3912 fsnotify_rmdir(dir, dentry);
3915 inode_unlock(dentry->d_inode);
3921 EXPORT_SYMBOL(vfs_rmdir);
3923 long do_rmdir(int dfd, const char __user *pathname)
3926 struct filename *name;
3927 struct dentry *dentry;
3931 unsigned int lookup_flags = 0;
3933 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3934 &path, &last, &type);
3936 return PTR_ERR(name);
3950 error = mnt_want_write(path.mnt);
3954 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3955 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3956 error = PTR_ERR(dentry);
3959 if (!dentry->d_inode) {
3963 error = security_path_rmdir(&path, dentry);
3966 error = vfs_rmdir(path.dentry->d_inode, dentry);
3970 inode_unlock(path.dentry->d_inode);
3971 mnt_drop_write(path.mnt);
3975 if (retry_estale(error, lookup_flags)) {
3976 lookup_flags |= LOOKUP_REVAL;
3982 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3984 return do_rmdir(AT_FDCWD, pathname);
3988 * vfs_unlink - unlink a filesystem object
3989 * @dir: parent directory
3991 * @delegated_inode: returns victim inode, if the inode is delegated.
3993 * The caller must hold dir->i_mutex.
3995 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3996 * return a reference to the inode in delegated_inode. The caller
3997 * should then break the delegation on that inode and retry. Because
3998 * breaking a delegation may take a long time, the caller should drop
3999 * dir->i_mutex before doing so.
4001 * Alternatively, a caller may pass NULL for delegated_inode. This may
4002 * be appropriate for callers that expect the underlying filesystem not
4003 * to be NFS exported.
4005 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
4007 struct inode *target = dentry->d_inode;
4008 int error = may_delete(dir, dentry, 0);
4013 if (!dir->i_op->unlink)
4017 if (is_local_mountpoint(dentry))
4020 error = security_inode_unlink(dir, dentry);
4022 error = try_break_deleg(target, delegated_inode);
4025 error = dir->i_op->unlink(dir, dentry);
4028 detach_mounts(dentry);
4029 fsnotify_unlink(dir, dentry);
4034 inode_unlock(target);
4036 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4037 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
4038 fsnotify_link_count(target);
4044 EXPORT_SYMBOL(vfs_unlink);
4047 * Make sure that the actual truncation of the file will occur outside its
4048 * directory's i_mutex. Truncate can take a long time if there is a lot of
4049 * writeout happening, and we don't want to prevent access to the directory
4050 * while waiting on the I/O.
4052 long do_unlinkat(int dfd, struct filename *name)
4055 struct dentry *dentry;
4059 struct inode *inode = NULL;
4060 struct inode *delegated_inode = NULL;
4061 unsigned int lookup_flags = 0;
4063 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4065 return PTR_ERR(name);
4068 if (type != LAST_NORM)
4071 error = mnt_want_write(path.mnt);
4075 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4076 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4077 error = PTR_ERR(dentry);
4078 if (!IS_ERR(dentry)) {
4079 /* Why not before? Because we want correct error value */
4080 if (last.name[last.len])
4082 inode = dentry->d_inode;
4083 if (d_is_negative(dentry))
4086 error = security_path_unlink(&path, dentry);
4089 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4093 inode_unlock(path.dentry->d_inode);
4095 iput(inode); /* truncate the inode here */
4097 if (delegated_inode) {
4098 error = break_deleg_wait(&delegated_inode);
4102 mnt_drop_write(path.mnt);
4105 if (retry_estale(error, lookup_flags)) {
4106 lookup_flags |= LOOKUP_REVAL;
4114 if (d_is_negative(dentry))
4116 else if (d_is_dir(dentry))
4123 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4125 if ((flag & ~AT_REMOVEDIR) != 0)
4128 if (flag & AT_REMOVEDIR)
4129 return do_rmdir(dfd, pathname);
4131 return do_unlinkat(dfd, getname(pathname));
4134 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4136 return do_unlinkat(AT_FDCWD, getname(pathname));
4139 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4141 int error = may_create(dir, dentry);
4146 if (!dir->i_op->symlink)
4149 error = security_inode_symlink(dir, dentry, oldname);
4153 error = dir->i_op->symlink(dir, dentry, oldname);
4155 fsnotify_create(dir, dentry);
4158 EXPORT_SYMBOL(vfs_symlink);
4160 long do_symlinkat(const char __user *oldname, int newdfd,
4161 const char __user *newname)
4164 struct filename *from;
4165 struct dentry *dentry;
4167 unsigned int lookup_flags = 0;
4169 from = getname(oldname);
4171 return PTR_ERR(from);
4173 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4174 error = PTR_ERR(dentry);
4178 error = security_path_symlink(&path, dentry, from->name);
4180 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4181 done_path_create(&path, dentry);
4182 if (retry_estale(error, lookup_flags)) {
4183 lookup_flags |= LOOKUP_REVAL;
4191 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4192 int, newdfd, const char __user *, newname)
4194 return do_symlinkat(oldname, newdfd, newname);
4197 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4199 return do_symlinkat(oldname, AT_FDCWD, newname);
4203 * vfs_link - create a new link
4204 * @old_dentry: object to be linked
4206 * @new_dentry: where to create the new link
4207 * @delegated_inode: returns inode needing a delegation break
4209 * The caller must hold dir->i_mutex
4211 * If vfs_link discovers a delegation on the to-be-linked file in need
4212 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4213 * inode in delegated_inode. The caller should then break the delegation
4214 * and retry. Because breaking a delegation may take a long time, the
4215 * caller should drop the i_mutex before doing so.
4217 * Alternatively, a caller may pass NULL for delegated_inode. This may
4218 * be appropriate for callers that expect the underlying filesystem not
4219 * to be NFS exported.
4221 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4223 struct inode *inode = old_dentry->d_inode;
4224 unsigned max_links = dir->i_sb->s_max_links;
4230 error = may_create(dir, new_dentry);
4234 if (dir->i_sb != inode->i_sb)
4238 * A link to an append-only or immutable file cannot be created.
4240 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4243 * Updating the link count will likely cause i_uid and i_gid to
4244 * be writen back improperly if their true value is unknown to
4247 if (HAS_UNMAPPED_ID(inode))
4249 if (!dir->i_op->link)
4251 if (S_ISDIR(inode->i_mode))
4254 error = security_inode_link(old_dentry, dir, new_dentry);
4259 /* Make sure we don't allow creating hardlink to an unlinked file */
4260 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4262 else if (max_links && inode->i_nlink >= max_links)
4265 error = try_break_deleg(inode, delegated_inode);
4267 error = dir->i_op->link(old_dentry, dir, new_dentry);
4270 if (!error && (inode->i_state & I_LINKABLE)) {
4271 spin_lock(&inode->i_lock);
4272 inode->i_state &= ~I_LINKABLE;
4273 spin_unlock(&inode->i_lock);
4275 inode_unlock(inode);
4277 fsnotify_link(dir, inode, new_dentry);
4280 EXPORT_SYMBOL(vfs_link);
4283 * Hardlinks are often used in delicate situations. We avoid
4284 * security-related surprises by not following symlinks on the
4287 * We don't follow them on the oldname either to be compatible
4288 * with linux 2.0, and to avoid hard-linking to directories
4289 * and other special files. --ADM
4291 int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4292 const char __user *newname, int flags)
4294 struct dentry *new_dentry;
4295 struct path old_path, new_path;
4296 struct inode *delegated_inode = NULL;
4300 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4303 * To use null names we require CAP_DAC_READ_SEARCH
4304 * This ensures that not everyone will be able to create
4305 * handlink using the passed filedescriptor.
4307 if (flags & AT_EMPTY_PATH) {
4308 if (!capable(CAP_DAC_READ_SEARCH))
4313 if (flags & AT_SYMLINK_FOLLOW)
4314 how |= LOOKUP_FOLLOW;
4316 error = user_path_at(olddfd, oldname, how, &old_path);
4320 new_dentry = user_path_create(newdfd, newname, &new_path,
4321 (how & LOOKUP_REVAL));
4322 error = PTR_ERR(new_dentry);
4323 if (IS_ERR(new_dentry))
4327 if (old_path.mnt != new_path.mnt)
4329 error = may_linkat(&old_path);
4330 if (unlikely(error))
4332 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4335 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4337 done_path_create(&new_path, new_dentry);
4338 if (delegated_inode) {
4339 error = break_deleg_wait(&delegated_inode);
4341 path_put(&old_path);
4345 if (retry_estale(error, how)) {
4346 path_put(&old_path);
4347 how |= LOOKUP_REVAL;
4351 path_put(&old_path);
4356 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4357 int, newdfd, const char __user *, newname, int, flags)
4359 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4362 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4364 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4368 * vfs_rename - rename a filesystem object
4369 * @old_dir: parent of source
4370 * @old_dentry: source
4371 * @new_dir: parent of destination
4372 * @new_dentry: destination
4373 * @delegated_inode: returns an inode needing a delegation break
4374 * @flags: rename flags
4376 * The caller must hold multiple mutexes--see lock_rename()).
4378 * If vfs_rename discovers a delegation in need of breaking at either
4379 * the source or destination, it will return -EWOULDBLOCK and return a
4380 * reference to the inode in delegated_inode. The caller should then
4381 * break the delegation and retry. Because breaking a delegation may
4382 * take a long time, the caller should drop all locks before doing
4385 * Alternatively, a caller may pass NULL for delegated_inode. This may
4386 * be appropriate for callers that expect the underlying filesystem not
4387 * to be NFS exported.
4389 * The worst of all namespace operations - renaming directory. "Perverted"
4390 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4393 * a) we can get into loop creation.
4394 * b) race potential - two innocent renames can create a loop together.
4395 * That's where 4.4 screws up. Current fix: serialization on
4396 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4398 * c) we have to lock _four_ objects - parents and victim (if it exists),
4399 * and source (if it is not a directory).
4400 * And that - after we got ->i_mutex on parents (until then we don't know
4401 * whether the target exists). Solution: try to be smart with locking
4402 * order for inodes. We rely on the fact that tree topology may change
4403 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4404 * move will be locked. Thus we can rank directories by the tree
4405 * (ancestors first) and rank all non-directories after them.
4406 * That works since everybody except rename does "lock parent, lookup,
4407 * lock child" and rename is under ->s_vfs_rename_mutex.
4408 * HOWEVER, it relies on the assumption that any object with ->lookup()
4409 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4410 * we'd better make sure that there's no link(2) for them.
4411 * d) conversion from fhandle to dentry may come in the wrong moment - when
4412 * we are removing the target. Solution: we will have to grab ->i_mutex
4413 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4414 * ->i_mutex on parents, which works but leads to some truly excessive
4417 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4418 struct inode *new_dir, struct dentry *new_dentry,
4419 struct inode **delegated_inode, unsigned int flags)
4422 bool is_dir = d_is_dir(old_dentry);
4423 struct inode *source = old_dentry->d_inode;
4424 struct inode *target = new_dentry->d_inode;
4425 bool new_is_dir = false;
4426 unsigned max_links = new_dir->i_sb->s_max_links;
4427 struct name_snapshot old_name;
4429 if (source == target)
4432 error = may_delete(old_dir, old_dentry, is_dir);
4437 error = may_create(new_dir, new_dentry);
4439 new_is_dir = d_is_dir(new_dentry);
4441 if (!(flags & RENAME_EXCHANGE))
4442 error = may_delete(new_dir, new_dentry, is_dir);
4444 error = may_delete(new_dir, new_dentry, new_is_dir);
4449 if (!old_dir->i_op->rename)
4453 * If we are going to change the parent - check write permissions,
4454 * we'll need to flip '..'.
4456 if (new_dir != old_dir) {
4458 error = inode_permission(source, MAY_WRITE);
4462 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4463 error = inode_permission(target, MAY_WRITE);
4469 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4474 take_dentry_name_snapshot(&old_name, old_dentry);
4476 if (!is_dir || (flags & RENAME_EXCHANGE))
4477 lock_two_nondirectories(source, target);
4482 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4485 if (max_links && new_dir != old_dir) {
4487 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4489 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4490 old_dir->i_nlink >= max_links)
4494 error = try_break_deleg(source, delegated_inode);
4498 if (target && !new_is_dir) {
4499 error = try_break_deleg(target, delegated_inode);
4503 error = old_dir->i_op->rename(old_dir, old_dentry,
4504 new_dir, new_dentry, flags);
4508 if (!(flags & RENAME_EXCHANGE) && target) {
4510 shrink_dcache_parent(new_dentry);
4511 target->i_flags |= S_DEAD;
4513 dont_mount(new_dentry);
4514 detach_mounts(new_dentry);
4516 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4517 if (!(flags & RENAME_EXCHANGE))
4518 d_move(old_dentry, new_dentry);
4520 d_exchange(old_dentry, new_dentry);
4523 if (!is_dir || (flags & RENAME_EXCHANGE))
4524 unlock_two_nondirectories(source, target);
4526 inode_unlock(target);
4529 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4530 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4531 if (flags & RENAME_EXCHANGE) {
4532 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4533 new_is_dir, NULL, new_dentry);
4536 release_dentry_name_snapshot(&old_name);
4540 EXPORT_SYMBOL(vfs_rename);
4542 static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4543 const char __user *newname, unsigned int flags)
4545 struct dentry *old_dentry, *new_dentry;
4546 struct dentry *trap;
4547 struct path old_path, new_path;
4548 struct qstr old_last, new_last;
4549 int old_type, new_type;
4550 struct inode *delegated_inode = NULL;
4551 struct filename *from;
4552 struct filename *to;
4553 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4554 bool should_retry = false;
4557 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4560 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4561 (flags & RENAME_EXCHANGE))
4564 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4567 if (flags & RENAME_EXCHANGE)
4571 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4572 &old_path, &old_last, &old_type);
4574 error = PTR_ERR(from);
4578 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4579 &new_path, &new_last, &new_type);
4581 error = PTR_ERR(to);
4586 if (old_path.mnt != new_path.mnt)
4590 if (old_type != LAST_NORM)
4593 if (flags & RENAME_NOREPLACE)
4595 if (new_type != LAST_NORM)
4598 error = mnt_want_write(old_path.mnt);
4603 trap = lock_rename(new_path.dentry, old_path.dentry);
4605 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4606 error = PTR_ERR(old_dentry);
4607 if (IS_ERR(old_dentry))
4609 /* source must exist */
4611 if (d_is_negative(old_dentry))
4613 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4614 error = PTR_ERR(new_dentry);
4615 if (IS_ERR(new_dentry))
4618 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4620 if (flags & RENAME_EXCHANGE) {
4622 if (d_is_negative(new_dentry))
4625 if (!d_is_dir(new_dentry)) {
4627 if (new_last.name[new_last.len])
4631 /* unless the source is a directory trailing slashes give -ENOTDIR */
4632 if (!d_is_dir(old_dentry)) {
4634 if (old_last.name[old_last.len])
4636 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4639 /* source should not be ancestor of target */
4641 if (old_dentry == trap)
4643 /* target should not be an ancestor of source */
4644 if (!(flags & RENAME_EXCHANGE))
4646 if (new_dentry == trap)
4649 error = security_path_rename(&old_path, old_dentry,
4650 &new_path, new_dentry, flags);
4653 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4654 new_path.dentry->d_inode, new_dentry,
4655 &delegated_inode, flags);
4661 unlock_rename(new_path.dentry, old_path.dentry);
4662 if (delegated_inode) {
4663 error = break_deleg_wait(&delegated_inode);
4667 mnt_drop_write(old_path.mnt);
4669 if (retry_estale(error, lookup_flags))
4670 should_retry = true;
4671 path_put(&new_path);
4674 path_put(&old_path);
4677 should_retry = false;
4678 lookup_flags |= LOOKUP_REVAL;
4685 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4686 int, newdfd, const char __user *, newname, unsigned int, flags)
4688 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4691 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4692 int, newdfd, const char __user *, newname)
4694 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
4697 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4699 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4702 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4704 int error = may_create(dir, dentry);
4708 if (!dir->i_op->mknod)
4711 return dir->i_op->mknod(dir, dentry,
4712 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4714 EXPORT_SYMBOL(vfs_whiteout);
4716 int readlink_copy(char __user *buffer, int buflen, const char *link)
4718 int len = PTR_ERR(link);
4723 if (len > (unsigned) buflen)
4725 if (copy_to_user(buffer, link, len))
4732 * vfs_readlink - copy symlink body into userspace buffer
4733 * @dentry: dentry on which to get symbolic link
4734 * @buffer: user memory pointer
4735 * @buflen: size of buffer
4737 * Does not touch atime. That's up to the caller if necessary
4739 * Does not call security hook.
4741 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4743 struct inode *inode = d_inode(dentry);
4744 DEFINE_DELAYED_CALL(done);
4748 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4749 if (unlikely(inode->i_op->readlink))
4750 return inode->i_op->readlink(dentry, buffer, buflen);
4752 if (!d_is_symlink(dentry))
4755 spin_lock(&inode->i_lock);
4756 inode->i_opflags |= IOP_DEFAULT_READLINK;
4757 spin_unlock(&inode->i_lock);
4760 link = READ_ONCE(inode->i_link);
4762 link = inode->i_op->get_link(dentry, inode, &done);
4764 return PTR_ERR(link);
4766 res = readlink_copy(buffer, buflen, link);
4767 do_delayed_call(&done);
4770 EXPORT_SYMBOL(vfs_readlink);
4773 * vfs_get_link - get symlink body
4774 * @dentry: dentry on which to get symbolic link
4775 * @done: caller needs to free returned data with this
4777 * Calls security hook and i_op->get_link() on the supplied inode.
4779 * It does not touch atime. That's up to the caller if necessary.
4781 * Does not work on "special" symlinks like /proc/$$/fd/N
4783 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4785 const char *res = ERR_PTR(-EINVAL);
4786 struct inode *inode = d_inode(dentry);
4788 if (d_is_symlink(dentry)) {
4789 res = ERR_PTR(security_inode_readlink(dentry));
4791 res = inode->i_op->get_link(dentry, inode, done);
4795 EXPORT_SYMBOL(vfs_get_link);
4797 /* get the link contents into pagecache */
4798 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4799 struct delayed_call *callback)
4803 struct address_space *mapping = inode->i_mapping;
4806 page = find_get_page(mapping, 0);
4808 return ERR_PTR(-ECHILD);
4809 if (!PageUptodate(page)) {
4811 return ERR_PTR(-ECHILD);
4814 page = read_mapping_page(mapping, 0, NULL);
4818 set_delayed_call(callback, page_put_link, page);
4819 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4820 kaddr = page_address(page);
4821 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4825 EXPORT_SYMBOL(page_get_link);
4827 void page_put_link(void *arg)
4831 EXPORT_SYMBOL(page_put_link);
4833 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4835 DEFINE_DELAYED_CALL(done);
4836 int res = readlink_copy(buffer, buflen,
4837 page_get_link(dentry, d_inode(dentry),
4839 do_delayed_call(&done);
4842 EXPORT_SYMBOL(page_readlink);
4845 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4847 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4849 struct address_space *mapping = inode->i_mapping;
4853 unsigned int flags = 0;
4855 flags |= AOP_FLAG_NOFS;
4858 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4859 flags, &page, &fsdata);
4863 memcpy(page_address(page), symname, len-1);
4865 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4872 mark_inode_dirty(inode);
4877 EXPORT_SYMBOL(__page_symlink);
4879 int page_symlink(struct inode *inode, const char *symname, int len)
4881 return __page_symlink(inode, symname, len,
4882 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4884 EXPORT_SYMBOL(page_symlink);
4886 const struct inode_operations page_symlink_inode_operations = {
4887 .get_link = page_get_link,
4889 EXPORT_SYMBOL(page_symlink_inode_operations);