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 inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1389 struct path *path, struct inode **inode,
1394 path->mnt = nd->path.mnt;
1395 path->dentry = dentry;
1396 if (nd->flags & LOOKUP_RCU) {
1397 unsigned int seq = *seqp;
1398 if (unlikely(!*inode))
1400 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1402 if (unlazy_child(nd, dentry, seq))
1404 // *path might've been clobbered by __follow_mount_rcu()
1405 path->mnt = nd->path.mnt;
1406 path->dentry = dentry;
1408 ret = follow_managed(path, nd);
1409 if (likely(ret >= 0)) {
1410 *inode = d_backing_inode(path->dentry);
1411 *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1416 static int follow_dotdot_rcu(struct nameidata *nd)
1418 struct inode *inode = nd->inode;
1421 if (path_equal(&nd->path, &nd->root)) {
1422 if (unlikely(nd->flags & LOOKUP_BENEATH))
1426 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1427 struct dentry *old = nd->path.dentry;
1428 struct dentry *parent = old->d_parent;
1431 inode = parent->d_inode;
1432 seq = read_seqcount_begin(&parent->d_seq);
1433 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1435 nd->path.dentry = parent;
1437 if (unlikely(!path_connected(&nd->path)))
1441 struct mount *mnt = real_mount(nd->path.mnt);
1442 struct mount *mparent = mnt->mnt_parent;
1443 struct dentry *mountpoint = mnt->mnt_mountpoint;
1444 struct inode *inode2 = mountpoint->d_inode;
1445 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1446 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1448 if (&mparent->mnt == nd->path.mnt)
1450 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1452 /* we know that mountpoint was pinned */
1453 nd->path.dentry = mountpoint;
1454 nd->path.mnt = &mparent->mnt;
1459 while (unlikely(d_mountpoint(nd->path.dentry))) {
1460 struct mount *mounted;
1461 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1462 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1466 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1468 nd->path.mnt = &mounted->mnt;
1469 nd->path.dentry = mounted->mnt.mnt_root;
1470 inode = nd->path.dentry->d_inode;
1471 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1478 * Follow down to the covering mount currently visible to userspace. At each
1479 * point, the filesystem owning that dentry may be queried as to whether the
1480 * caller is permitted to proceed or not.
1482 int follow_down(struct path *path)
1487 while (managed = READ_ONCE(path->dentry->d_flags),
1488 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1489 /* Allow the filesystem to manage the transit without i_mutex
1492 * We indicate to the filesystem if someone is trying to mount
1493 * something here. This gives autofs the chance to deny anyone
1494 * other than its daemon the right to mount on its
1497 * The filesystem may sleep at this point.
1499 if (managed & DCACHE_MANAGE_TRANSIT) {
1500 BUG_ON(!path->dentry->d_op);
1501 BUG_ON(!path->dentry->d_op->d_manage);
1502 ret = path->dentry->d_op->d_manage(path, false);
1504 return ret == -EISDIR ? 0 : ret;
1507 /* Transit to a mounted filesystem. */
1508 if (managed & DCACHE_MOUNTED) {
1509 struct vfsmount *mounted = lookup_mnt(path);
1514 path->mnt = mounted;
1515 path->dentry = dget(mounted->mnt_root);
1519 /* Don't handle automount points here */
1524 EXPORT_SYMBOL(follow_down);
1527 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1529 static void follow_mount(struct path *path)
1531 while (d_mountpoint(path->dentry)) {
1532 struct vfsmount *mounted = lookup_mnt(path);
1537 path->mnt = mounted;
1538 path->dentry = dget(mounted->mnt_root);
1542 static int path_parent_directory(struct path *path)
1544 struct dentry *old = path->dentry;
1545 /* rare case of legitimate dget_parent()... */
1546 path->dentry = dget_parent(path->dentry);
1548 if (unlikely(!path_connected(path)))
1553 static int follow_dotdot(struct nameidata *nd)
1556 if (path_equal(&nd->path, &nd->root)) {
1557 if (unlikely(nd->flags & LOOKUP_BENEATH))
1561 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1562 int ret = path_parent_directory(&nd->path);
1567 if (!follow_up(&nd->path))
1569 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1572 follow_mount(&nd->path);
1573 nd->inode = nd->path.dentry->d_inode;
1578 * This looks up the name in dcache and possibly revalidates the found dentry.
1579 * NULL is returned if the dentry does not exist in the cache.
1581 static struct dentry *lookup_dcache(const struct qstr *name,
1585 struct dentry *dentry = d_lookup(dir, name);
1587 int error = d_revalidate(dentry, flags);
1588 if (unlikely(error <= 0)) {
1590 d_invalidate(dentry);
1592 return ERR_PTR(error);
1599 * Parent directory has inode locked exclusive. This is one
1600 * and only case when ->lookup() gets called on non in-lookup
1601 * dentries - as the matter of fact, this only gets called
1602 * when directory is guaranteed to have no in-lookup children
1605 static struct dentry *__lookup_hash(const struct qstr *name,
1606 struct dentry *base, unsigned int flags)
1608 struct dentry *dentry = lookup_dcache(name, base, flags);
1610 struct inode *dir = base->d_inode;
1615 /* Don't create child dentry for a dead directory. */
1616 if (unlikely(IS_DEADDIR(dir)))
1617 return ERR_PTR(-ENOENT);
1619 dentry = d_alloc(base, name);
1620 if (unlikely(!dentry))
1621 return ERR_PTR(-ENOMEM);
1623 old = dir->i_op->lookup(dir, dentry, flags);
1624 if (unlikely(old)) {
1631 static struct dentry *lookup_fast(struct nameidata *nd,
1632 struct inode **inode,
1635 struct dentry *dentry, *parent = nd->path.dentry;
1639 * Rename seqlock is not required here because in the off chance
1640 * of a false negative due to a concurrent rename, the caller is
1641 * going to fall back to non-racy lookup.
1643 if (nd->flags & LOOKUP_RCU) {
1645 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1646 if (unlikely(!dentry)) {
1647 if (unlazy_walk(nd))
1648 return ERR_PTR(-ECHILD);
1653 * This sequence count validates that the inode matches
1654 * the dentry name information from lookup.
1656 *inode = d_backing_inode(dentry);
1657 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1658 return ERR_PTR(-ECHILD);
1661 * This sequence count validates that the parent had no
1662 * changes while we did the lookup of the dentry above.
1664 * The memory barrier in read_seqcount_begin of child is
1665 * enough, we can use __read_seqcount_retry here.
1667 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1668 return ERR_PTR(-ECHILD);
1671 status = d_revalidate(dentry, nd->flags);
1672 if (likely(status > 0))
1674 if (unlazy_child(nd, dentry, seq))
1675 return ERR_PTR(-ECHILD);
1676 if (unlikely(status == -ECHILD))
1677 /* we'd been told to redo it in non-rcu mode */
1678 status = d_revalidate(dentry, nd->flags);
1680 dentry = __d_lookup(parent, &nd->last);
1681 if (unlikely(!dentry))
1683 status = d_revalidate(dentry, nd->flags);
1685 if (unlikely(status <= 0)) {
1687 d_invalidate(dentry);
1689 return ERR_PTR(status);
1694 /* Fast lookup failed, do it the slow way */
1695 static struct dentry *__lookup_slow(const struct qstr *name,
1699 struct dentry *dentry, *old;
1700 struct inode *inode = dir->d_inode;
1701 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1703 /* Don't go there if it's already dead */
1704 if (unlikely(IS_DEADDIR(inode)))
1705 return ERR_PTR(-ENOENT);
1707 dentry = d_alloc_parallel(dir, name, &wq);
1710 if (unlikely(!d_in_lookup(dentry))) {
1711 int error = d_revalidate(dentry, flags);
1712 if (unlikely(error <= 0)) {
1714 d_invalidate(dentry);
1719 dentry = ERR_PTR(error);
1722 old = inode->i_op->lookup(inode, dentry, flags);
1723 d_lookup_done(dentry);
1724 if (unlikely(old)) {
1732 static struct dentry *lookup_slow(const struct qstr *name,
1736 struct inode *inode = dir->d_inode;
1738 inode_lock_shared(inode);
1739 res = __lookup_slow(name, dir, flags);
1740 inode_unlock_shared(inode);
1744 static inline int may_lookup(struct nameidata *nd)
1746 if (nd->flags & LOOKUP_RCU) {
1747 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1750 if (unlazy_walk(nd))
1753 return inode_permission(nd->inode, MAY_EXEC);
1756 static inline int handle_dots(struct nameidata *nd, int type)
1758 if (type == LAST_DOTDOT) {
1761 if (!nd->root.mnt) {
1762 error = set_root(nd);
1766 if (nd->flags & LOOKUP_RCU)
1767 error = follow_dotdot_rcu(nd);
1769 error = follow_dotdot(nd);
1773 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1775 * If there was a racing rename or mount along our
1776 * path, then we can't be sure that ".." hasn't jumped
1777 * above nd->root (and so userspace should retry or use
1781 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1783 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1790 static int pick_link(struct nameidata *nd, struct path *link,
1791 struct inode *inode, unsigned seq)
1795 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1796 path_to_nameidata(link, nd);
1799 if (!(nd->flags & LOOKUP_RCU)) {
1800 if (link->mnt == nd->path.mnt)
1803 error = nd_alloc_stack(nd);
1804 if (unlikely(error)) {
1805 if (error == -ECHILD) {
1806 if (unlikely(!legitimize_path(nd, link, seq))) {
1809 nd->flags &= ~LOOKUP_RCU;
1810 nd->path.mnt = NULL;
1811 nd->path.dentry = NULL;
1813 } else if (likely(unlazy_walk(nd)) == 0)
1814 error = nd_alloc_stack(nd);
1822 last = nd->stack + nd->depth++;
1824 clear_delayed_call(&last->done);
1825 nd->link_inode = inode;
1830 enum {WALK_FOLLOW = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1833 * Do we need to follow links? We _really_ want to be able
1834 * to do this check without having to look at inode->i_op,
1835 * so we keep a cache of "no, this doesn't need follow_link"
1836 * for the common case.
1838 static int step_into(struct nameidata *nd, int flags,
1839 struct dentry *dentry, struct inode *inode, unsigned seq)
1842 int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1846 if (likely(!d_is_symlink(path.dentry)) ||
1847 !((flags & WALK_FOLLOW) || (nd->flags & LOOKUP_FOLLOW)) ||
1848 (flags & WALK_NOFOLLOW)) {
1849 /* not a symlink or should not follow */
1850 path_to_nameidata(&path, nd);
1855 /* make sure that d_is_symlink above matches inode */
1856 if (nd->flags & LOOKUP_RCU) {
1857 if (read_seqcount_retry(&path.dentry->d_seq, seq))
1860 return pick_link(nd, &path, inode, seq);
1863 static int walk_component(struct nameidata *nd, int flags)
1865 struct dentry *dentry;
1866 struct inode *inode;
1870 * "." and ".." are special - ".." especially so because it has
1871 * to be able to know about the current root directory and
1872 * parent relationships.
1874 if (unlikely(nd->last_type != LAST_NORM)) {
1875 if (!(flags & WALK_MORE) && nd->depth)
1877 err = handle_dots(nd, nd->last_type);
1880 dentry = lookup_fast(nd, &inode, &seq);
1882 return PTR_ERR(dentry);
1883 if (unlikely(!dentry)) {
1884 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1886 return PTR_ERR(dentry);
1888 if (!(flags & WALK_MORE) && nd->depth)
1890 return step_into(nd, flags, dentry, inode, seq);
1894 * We can do the critical dentry name comparison and hashing
1895 * operations one word at a time, but we are limited to:
1897 * - Architectures with fast unaligned word accesses. We could
1898 * do a "get_unaligned()" if this helps and is sufficiently
1901 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1902 * do not trap on the (extremely unlikely) case of a page
1903 * crossing operation.
1905 * - Furthermore, we need an efficient 64-bit compile for the
1906 * 64-bit case in order to generate the "number of bytes in
1907 * the final mask". Again, that could be replaced with a
1908 * efficient population count instruction or similar.
1910 #ifdef CONFIG_DCACHE_WORD_ACCESS
1912 #include <asm/word-at-a-time.h>
1916 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1918 #elif defined(CONFIG_64BIT)
1920 * Register pressure in the mixing function is an issue, particularly
1921 * on 32-bit x86, but almost any function requires one state value and
1922 * one temporary. Instead, use a function designed for two state values
1923 * and no temporaries.
1925 * This function cannot create a collision in only two iterations, so
1926 * we have two iterations to achieve avalanche. In those two iterations,
1927 * we have six layers of mixing, which is enough to spread one bit's
1928 * influence out to 2^6 = 64 state bits.
1930 * Rotate constants are scored by considering either 64 one-bit input
1931 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1932 * probability of that delta causing a change to each of the 128 output
1933 * bits, using a sample of random initial states.
1935 * The Shannon entropy of the computed probabilities is then summed
1936 * to produce a score. Ideally, any input change has a 50% chance of
1937 * toggling any given output bit.
1939 * Mixing scores (in bits) for (12,45):
1940 * Input delta: 1-bit 2-bit
1941 * 1 round: 713.3 42542.6
1942 * 2 rounds: 2753.7 140389.8
1943 * 3 rounds: 5954.1 233458.2
1944 * 4 rounds: 7862.6 256672.2
1945 * Perfect: 8192 258048
1946 * (64*128) (64*63/2 * 128)
1948 #define HASH_MIX(x, y, a) \
1950 y ^= x, x = rol64(x,12),\
1951 x += y, y = rol64(y,45),\
1955 * Fold two longs into one 32-bit hash value. This must be fast, but
1956 * latency isn't quite as critical, as there is a fair bit of additional
1957 * work done before the hash value is used.
1959 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1961 y ^= x * GOLDEN_RATIO_64;
1962 y *= GOLDEN_RATIO_64;
1966 #else /* 32-bit case */
1969 * Mixing scores (in bits) for (7,20):
1970 * Input delta: 1-bit 2-bit
1971 * 1 round: 330.3 9201.6
1972 * 2 rounds: 1246.4 25475.4
1973 * 3 rounds: 1907.1 31295.1
1974 * 4 rounds: 2042.3 31718.6
1975 * Perfect: 2048 31744
1976 * (32*64) (32*31/2 * 64)
1978 #define HASH_MIX(x, y, a) \
1980 y ^= x, x = rol32(x, 7),\
1981 x += y, y = rol32(y,20),\
1984 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1986 /* Use arch-optimized multiply if one exists */
1987 return __hash_32(y ^ __hash_32(x));
1993 * Return the hash of a string of known length. This is carfully
1994 * designed to match hash_name(), which is the more critical function.
1995 * In particular, we must end by hashing a final word containing 0..7
1996 * payload bytes, to match the way that hash_name() iterates until it
1997 * finds the delimiter after the name.
1999 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2001 unsigned long a, x = 0, y = (unsigned long)salt;
2006 a = load_unaligned_zeropad(name);
2007 if (len < sizeof(unsigned long))
2010 name += sizeof(unsigned long);
2011 len -= sizeof(unsigned long);
2013 x ^= a & bytemask_from_count(len);
2015 return fold_hash(x, y);
2017 EXPORT_SYMBOL(full_name_hash);
2019 /* Return the "hash_len" (hash and length) of a null-terminated string */
2020 u64 hashlen_string(const void *salt, const char *name)
2022 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2023 unsigned long adata, mask, len;
2024 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2031 len += sizeof(unsigned long);
2033 a = load_unaligned_zeropad(name+len);
2034 } while (!has_zero(a, &adata, &constants));
2036 adata = prep_zero_mask(a, adata, &constants);
2037 mask = create_zero_mask(adata);
2038 x ^= a & zero_bytemask(mask);
2040 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2042 EXPORT_SYMBOL(hashlen_string);
2045 * Calculate the length and hash of the path component, and
2046 * return the "hash_len" as the result.
2048 static inline u64 hash_name(const void *salt, const char *name)
2050 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2051 unsigned long adata, bdata, mask, len;
2052 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2059 len += sizeof(unsigned long);
2061 a = load_unaligned_zeropad(name+len);
2062 b = a ^ REPEAT_BYTE('/');
2063 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2065 adata = prep_zero_mask(a, adata, &constants);
2066 bdata = prep_zero_mask(b, bdata, &constants);
2067 mask = create_zero_mask(adata | bdata);
2068 x ^= a & zero_bytemask(mask);
2070 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2073 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2075 /* Return the hash of a string of known length */
2076 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2078 unsigned long hash = init_name_hash(salt);
2080 hash = partial_name_hash((unsigned char)*name++, hash);
2081 return end_name_hash(hash);
2083 EXPORT_SYMBOL(full_name_hash);
2085 /* Return the "hash_len" (hash and length) of a null-terminated string */
2086 u64 hashlen_string(const void *salt, const char *name)
2088 unsigned long hash = init_name_hash(salt);
2089 unsigned long len = 0, c;
2091 c = (unsigned char)*name;
2094 hash = partial_name_hash(c, hash);
2095 c = (unsigned char)name[len];
2097 return hashlen_create(end_name_hash(hash), len);
2099 EXPORT_SYMBOL(hashlen_string);
2102 * We know there's a real path component here of at least
2105 static inline u64 hash_name(const void *salt, const char *name)
2107 unsigned long hash = init_name_hash(salt);
2108 unsigned long len = 0, c;
2110 c = (unsigned char)*name;
2113 hash = partial_name_hash(c, hash);
2114 c = (unsigned char)name[len];
2115 } while (c && c != '/');
2116 return hashlen_create(end_name_hash(hash), len);
2123 * This is the basic name resolution function, turning a pathname into
2124 * the final dentry. We expect 'base' to be positive and a directory.
2126 * Returns 0 and nd will have valid dentry and mnt on success.
2127 * Returns error and drops reference to input namei data on failure.
2129 static int link_path_walk(const char *name, struct nameidata *nd)
2134 return PTR_ERR(name);
2140 /* At this point we know we have a real path component. */
2145 err = may_lookup(nd);
2149 hash_len = hash_name(nd->path.dentry, name);
2152 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2154 if (name[1] == '.') {
2156 nd->flags |= LOOKUP_JUMPED;
2162 if (likely(type == LAST_NORM)) {
2163 struct dentry *parent = nd->path.dentry;
2164 nd->flags &= ~LOOKUP_JUMPED;
2165 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2166 struct qstr this = { { .hash_len = hash_len }, .name = name };
2167 err = parent->d_op->d_hash(parent, &this);
2170 hash_len = this.hash_len;
2175 nd->last.hash_len = hash_len;
2176 nd->last.name = name;
2177 nd->last_type = type;
2179 name += hashlen_len(hash_len);
2183 * If it wasn't NUL, we know it was '/'. Skip that
2184 * slash, and continue until no more slashes.
2188 } while (unlikely(*name == '/'));
2189 if (unlikely(!*name)) {
2191 /* pathname body, done */
2194 name = nd->stack[nd->depth - 1].name;
2195 /* trailing symlink, done */
2198 /* last component of nested symlink */
2199 err = walk_component(nd, WALK_FOLLOW);
2201 /* not the last component */
2202 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2208 const char *s = get_link(nd);
2217 nd->stack[nd->depth - 1].name = name;
2222 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2223 if (nd->flags & LOOKUP_RCU) {
2224 if (unlazy_walk(nd))
2232 /* must be paired with terminate_walk() */
2233 static const char *path_init(struct nameidata *nd, unsigned flags)
2236 const char *s = nd->name->name;
2239 flags &= ~LOOKUP_RCU;
2240 if (flags & LOOKUP_RCU)
2243 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2244 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2247 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2248 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2251 if (flags & LOOKUP_ROOT) {
2252 struct dentry *root = nd->root.dentry;
2253 struct inode *inode = root->d_inode;
2254 if (*s && unlikely(!d_can_lookup(root)))
2255 return ERR_PTR(-ENOTDIR);
2256 nd->path = nd->root;
2258 if (flags & LOOKUP_RCU) {
2259 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2260 nd->root_seq = nd->seq;
2262 path_get(&nd->path);
2267 nd->root.mnt = NULL;
2268 nd->path.mnt = NULL;
2269 nd->path.dentry = NULL;
2271 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2272 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2273 error = nd_jump_root(nd);
2274 if (unlikely(error))
2275 return ERR_PTR(error);
2279 /* Relative pathname -- get the starting-point it is relative to. */
2280 if (nd->dfd == AT_FDCWD) {
2281 if (flags & LOOKUP_RCU) {
2282 struct fs_struct *fs = current->fs;
2286 seq = read_seqcount_begin(&fs->seq);
2288 nd->inode = nd->path.dentry->d_inode;
2289 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2290 } while (read_seqcount_retry(&fs->seq, seq));
2292 get_fs_pwd(current->fs, &nd->path);
2293 nd->inode = nd->path.dentry->d_inode;
2296 /* Caller must check execute permissions on the starting path component */
2297 struct fd f = fdget_raw(nd->dfd);
2298 struct dentry *dentry;
2301 return ERR_PTR(-EBADF);
2303 dentry = f.file->f_path.dentry;
2305 if (*s && unlikely(!d_can_lookup(dentry))) {
2307 return ERR_PTR(-ENOTDIR);
2310 nd->path = f.file->f_path;
2311 if (flags & LOOKUP_RCU) {
2312 nd->inode = nd->path.dentry->d_inode;
2313 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2315 path_get(&nd->path);
2316 nd->inode = nd->path.dentry->d_inode;
2321 /* For scoped-lookups we need to set the root to the dirfd as well. */
2322 if (flags & LOOKUP_IS_SCOPED) {
2323 nd->root = nd->path;
2324 if (flags & LOOKUP_RCU) {
2325 nd->root_seq = nd->seq;
2327 path_get(&nd->root);
2328 nd->flags |= LOOKUP_ROOT_GRABBED;
2334 static const char *trailing_symlink(struct nameidata *nd)
2337 int error = may_follow_link(nd);
2338 if (unlikely(error))
2339 return ERR_PTR(error);
2340 nd->flags |= LOOKUP_PARENT;
2341 nd->stack[0].name = NULL;
2346 static inline int lookup_last(struct nameidata *nd)
2348 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2349 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2351 nd->flags &= ~LOOKUP_PARENT;
2352 return walk_component(nd, 0);
2355 static int handle_lookup_down(struct nameidata *nd)
2357 if (!(nd->flags & LOOKUP_RCU))
2358 dget(nd->path.dentry);
2359 return step_into(nd, WALK_NOFOLLOW,
2360 nd->path.dentry, nd->inode, nd->seq);
2363 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2364 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2366 const char *s = path_init(nd, flags);
2369 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2370 err = handle_lookup_down(nd);
2371 if (unlikely(err < 0))
2375 while (!(err = link_path_walk(s, nd))
2376 && ((err = lookup_last(nd)) > 0)) {
2377 s = trailing_symlink(nd);
2380 err = complete_walk(nd);
2382 if (!err && nd->flags & LOOKUP_DIRECTORY)
2383 if (!d_can_lookup(nd->path.dentry))
2387 nd->path.mnt = NULL;
2388 nd->path.dentry = NULL;
2394 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2395 struct path *path, struct path *root)
2398 struct nameidata nd;
2400 return PTR_ERR(name);
2401 if (unlikely(root)) {
2403 flags |= LOOKUP_ROOT;
2405 set_nameidata(&nd, dfd, name);
2406 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2407 if (unlikely(retval == -ECHILD))
2408 retval = path_lookupat(&nd, flags, path);
2409 if (unlikely(retval == -ESTALE))
2410 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2412 if (likely(!retval))
2413 audit_inode(name, path->dentry, 0);
2414 restore_nameidata();
2419 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2420 static int path_parentat(struct nameidata *nd, unsigned flags,
2421 struct path *parent)
2423 const char *s = path_init(nd, flags);
2424 int err = link_path_walk(s, nd);
2426 err = complete_walk(nd);
2429 nd->path.mnt = NULL;
2430 nd->path.dentry = NULL;
2436 static struct filename *filename_parentat(int dfd, struct filename *name,
2437 unsigned int flags, struct path *parent,
2438 struct qstr *last, int *type)
2441 struct nameidata nd;
2445 set_nameidata(&nd, dfd, name);
2446 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2447 if (unlikely(retval == -ECHILD))
2448 retval = path_parentat(&nd, flags, parent);
2449 if (unlikely(retval == -ESTALE))
2450 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2451 if (likely(!retval)) {
2453 *type = nd.last_type;
2454 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2457 name = ERR_PTR(retval);
2459 restore_nameidata();
2463 /* does lookup, returns the object with parent locked */
2464 struct dentry *kern_path_locked(const char *name, struct path *path)
2466 struct filename *filename;
2471 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2473 if (IS_ERR(filename))
2474 return ERR_CAST(filename);
2475 if (unlikely(type != LAST_NORM)) {
2478 return ERR_PTR(-EINVAL);
2480 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2481 d = __lookup_hash(&last, path->dentry, 0);
2483 inode_unlock(path->dentry->d_inode);
2490 int kern_path(const char *name, unsigned int flags, struct path *path)
2492 return filename_lookup(AT_FDCWD, getname_kernel(name),
2495 EXPORT_SYMBOL(kern_path);
2498 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2499 * @dentry: pointer to dentry of the base directory
2500 * @mnt: pointer to vfs mount of the base directory
2501 * @name: pointer to file name
2502 * @flags: lookup flags
2503 * @path: pointer to struct path to fill
2505 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2506 const char *name, unsigned int flags,
2509 struct path root = {.mnt = mnt, .dentry = dentry};
2510 /* the first argument of filename_lookup() is ignored with root */
2511 return filename_lookup(AT_FDCWD, getname_kernel(name),
2512 flags , path, &root);
2514 EXPORT_SYMBOL(vfs_path_lookup);
2516 static int lookup_one_len_common(const char *name, struct dentry *base,
2517 int len, struct qstr *this)
2521 this->hash = full_name_hash(base, name, len);
2525 if (unlikely(name[0] == '.')) {
2526 if (len < 2 || (len == 2 && name[1] == '.'))
2531 unsigned int c = *(const unsigned char *)name++;
2532 if (c == '/' || c == '\0')
2536 * See if the low-level filesystem might want
2537 * to use its own hash..
2539 if (base->d_flags & DCACHE_OP_HASH) {
2540 int err = base->d_op->d_hash(base, this);
2545 return inode_permission(base->d_inode, MAY_EXEC);
2549 * try_lookup_one_len - filesystem helper to lookup single pathname component
2550 * @name: pathname component to lookup
2551 * @base: base directory to lookup from
2552 * @len: maximum length @len should be interpreted to
2554 * Look up a dentry by name in the dcache, returning NULL if it does not
2555 * currently exist. The function does not try to create a dentry.
2557 * Note that this routine is purely a helper for filesystem usage and should
2558 * not be called by generic code.
2560 * The caller must hold base->i_mutex.
2562 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2567 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2569 err = lookup_one_len_common(name, base, len, &this);
2571 return ERR_PTR(err);
2573 return lookup_dcache(&this, base, 0);
2575 EXPORT_SYMBOL(try_lookup_one_len);
2578 * lookup_one_len - filesystem helper to lookup single pathname component
2579 * @name: pathname component to lookup
2580 * @base: base directory to lookup from
2581 * @len: maximum length @len should be interpreted to
2583 * Note that this routine is purely a helper for filesystem usage and should
2584 * not be called by generic code.
2586 * The caller must hold base->i_mutex.
2588 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2590 struct dentry *dentry;
2594 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2596 err = lookup_one_len_common(name, base, len, &this);
2598 return ERR_PTR(err);
2600 dentry = lookup_dcache(&this, base, 0);
2601 return dentry ? dentry : __lookup_slow(&this, base, 0);
2603 EXPORT_SYMBOL(lookup_one_len);
2606 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2607 * @name: pathname component to lookup
2608 * @base: base directory to lookup from
2609 * @len: maximum length @len should be interpreted to
2611 * Note that this routine is purely a helper for filesystem usage and should
2612 * not be called by generic code.
2614 * Unlike lookup_one_len, it should be called without the parent
2615 * i_mutex held, and will take the i_mutex itself if necessary.
2617 struct dentry *lookup_one_len_unlocked(const char *name,
2618 struct dentry *base, int len)
2624 err = lookup_one_len_common(name, base, len, &this);
2626 return ERR_PTR(err);
2628 ret = lookup_dcache(&this, base, 0);
2630 ret = lookup_slow(&this, base, 0);
2633 EXPORT_SYMBOL(lookup_one_len_unlocked);
2636 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2637 * on negatives. Returns known positive or ERR_PTR(); that's what
2638 * most of the users want. Note that pinned negative with unlocked parent
2639 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2640 * need to be very careful; pinned positives have ->d_inode stable, so
2641 * this one avoids such problems.
2643 struct dentry *lookup_positive_unlocked(const char *name,
2644 struct dentry *base, int len)
2646 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2647 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2649 ret = ERR_PTR(-ENOENT);
2653 EXPORT_SYMBOL(lookup_positive_unlocked);
2655 #ifdef CONFIG_UNIX98_PTYS
2656 int path_pts(struct path *path)
2658 /* Find something mounted on "pts" in the same directory as
2661 struct dentry *child, *parent;
2665 ret = path_parent_directory(path);
2669 parent = path->dentry;
2672 child = d_hash_and_lookup(parent, &this);
2676 path->dentry = child;
2683 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2684 struct path *path, int *empty)
2686 return filename_lookup(dfd, getname_flags(name, flags, empty),
2689 EXPORT_SYMBOL(user_path_at_empty);
2692 * path_mountpoint - look up a path to be umounted
2693 * @nd: lookup context
2694 * @flags: lookup flags
2695 * @path: pointer to container for result
2697 * Look up the given name, but don't attempt to revalidate the last component.
2698 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2701 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2703 const char *s = path_init(nd, flags);
2706 while (!(err = link_path_walk(s, nd)) &&
2707 (err = lookup_last(nd)) > 0) {
2708 s = trailing_symlink(nd);
2710 if (!err && (nd->flags & LOOKUP_RCU))
2711 err = unlazy_walk(nd);
2713 err = handle_lookup_down(nd);
2716 nd->path.mnt = NULL;
2717 nd->path.dentry = NULL;
2724 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2727 struct nameidata nd;
2730 return PTR_ERR(name);
2731 set_nameidata(&nd, dfd, name);
2732 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2733 if (unlikely(error == -ECHILD))
2734 error = path_mountpoint(&nd, flags, path);
2735 if (unlikely(error == -ESTALE))
2736 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2738 audit_inode(name, path->dentry, AUDIT_INODE_NOEVAL);
2739 restore_nameidata();
2745 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2746 * @dfd: directory file descriptor
2747 * @name: pathname from userland
2748 * @flags: lookup flags
2749 * @path: pointer to container to hold result
2751 * A umount is a special case for path walking. We're not actually interested
2752 * in the inode in this situation, and ESTALE errors can be a problem. We
2753 * simply want track down the dentry and vfsmount attached at the mountpoint
2754 * and avoid revalidating the last component.
2756 * Returns 0 and populates "path" on success.
2759 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2762 return filename_mountpoint(dfd, getname(name), path, flags);
2766 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2769 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2771 EXPORT_SYMBOL(kern_path_mountpoint);
2773 int __check_sticky(struct inode *dir, struct inode *inode)
2775 kuid_t fsuid = current_fsuid();
2777 if (uid_eq(inode->i_uid, fsuid))
2779 if (uid_eq(dir->i_uid, fsuid))
2781 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2783 EXPORT_SYMBOL(__check_sticky);
2786 * Check whether we can remove a link victim from directory dir, check
2787 * whether the type of victim is right.
2788 * 1. We can't do it if dir is read-only (done in permission())
2789 * 2. We should have write and exec permissions on dir
2790 * 3. We can't remove anything from append-only dir
2791 * 4. We can't do anything with immutable dir (done in permission())
2792 * 5. If the sticky bit on dir is set we should either
2793 * a. be owner of dir, or
2794 * b. be owner of victim, or
2795 * c. have CAP_FOWNER capability
2796 * 6. If the victim is append-only or immutable we can't do antyhing with
2797 * links pointing to it.
2798 * 7. If the victim has an unknown uid or gid we can't change the inode.
2799 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2800 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2801 * 10. We can't remove a root or mountpoint.
2802 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2803 * nfs_async_unlink().
2805 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2807 struct inode *inode = d_backing_inode(victim);
2810 if (d_is_negative(victim))
2814 BUG_ON(victim->d_parent->d_inode != dir);
2816 /* Inode writeback is not safe when the uid or gid are invalid. */
2817 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2820 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2822 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2828 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2829 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2832 if (!d_is_dir(victim))
2834 if (IS_ROOT(victim))
2836 } else if (d_is_dir(victim))
2838 if (IS_DEADDIR(dir))
2840 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2845 /* Check whether we can create an object with dentry child in directory
2847 * 1. We can't do it if child already exists (open has special treatment for
2848 * this case, but since we are inlined it's OK)
2849 * 2. We can't do it if dir is read-only (done in permission())
2850 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2851 * 4. We should have write and exec permissions on dir
2852 * 5. We can't do it if dir is immutable (done in permission())
2854 static inline int may_create(struct inode *dir, struct dentry *child)
2856 struct user_namespace *s_user_ns;
2857 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2860 if (IS_DEADDIR(dir))
2862 s_user_ns = dir->i_sb->s_user_ns;
2863 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2864 !kgid_has_mapping(s_user_ns, current_fsgid()))
2866 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2870 * p1 and p2 should be directories on the same fs.
2872 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2877 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2881 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2883 p = d_ancestor(p2, p1);
2885 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2886 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2890 p = d_ancestor(p1, p2);
2892 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2893 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2897 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2898 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2901 EXPORT_SYMBOL(lock_rename);
2903 void unlock_rename(struct dentry *p1, struct dentry *p2)
2905 inode_unlock(p1->d_inode);
2907 inode_unlock(p2->d_inode);
2908 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2911 EXPORT_SYMBOL(unlock_rename);
2913 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2916 int error = may_create(dir, dentry);
2920 if (!dir->i_op->create)
2921 return -EACCES; /* shouldn't it be ENOSYS? */
2924 error = security_inode_create(dir, dentry, mode);
2927 error = dir->i_op->create(dir, dentry, mode, want_excl);
2929 fsnotify_create(dir, dentry);
2932 EXPORT_SYMBOL(vfs_create);
2934 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2935 int (*f)(struct dentry *, umode_t, void *),
2938 struct inode *dir = dentry->d_parent->d_inode;
2939 int error = may_create(dir, dentry);
2945 error = security_inode_create(dir, dentry, mode);
2948 error = f(dentry, mode, arg);
2950 fsnotify_create(dir, dentry);
2953 EXPORT_SYMBOL(vfs_mkobj);
2955 bool may_open_dev(const struct path *path)
2957 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2958 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2961 static int may_open(const struct path *path, int acc_mode, int flag)
2963 struct dentry *dentry = path->dentry;
2964 struct inode *inode = dentry->d_inode;
2970 switch (inode->i_mode & S_IFMT) {
2974 if (acc_mode & MAY_WRITE)
2979 if (!may_open_dev(path))
2988 error = inode_permission(inode, MAY_OPEN | acc_mode);
2993 * An append-only file must be opened in append mode for writing.
2995 if (IS_APPEND(inode)) {
2996 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3002 /* O_NOATIME can only be set by the owner or superuser */
3003 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
3009 static int handle_truncate(struct file *filp)
3011 const struct path *path = &filp->f_path;
3012 struct inode *inode = path->dentry->d_inode;
3013 int error = get_write_access(inode);
3017 * Refuse to truncate files with mandatory locks held on them.
3019 error = locks_verify_locked(filp);
3021 error = security_path_truncate(path);
3023 error = do_truncate(path->dentry, 0,
3024 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3027 put_write_access(inode);
3031 static inline int open_to_namei_flags(int flag)
3033 if ((flag & O_ACCMODE) == 3)
3038 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
3040 struct user_namespace *s_user_ns;
3041 int error = security_path_mknod(dir, dentry, mode, 0);
3045 s_user_ns = dir->dentry->d_sb->s_user_ns;
3046 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
3047 !kgid_has_mapping(s_user_ns, current_fsgid()))
3050 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
3054 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3058 * Attempt to atomically look up, create and open a file from a negative
3061 * Returns 0 if successful. The file will have been created and attached to
3062 * @file by the filesystem calling finish_open().
3064 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3065 * be set. The caller will need to perform the open themselves. @path will
3066 * have been updated to point to the new dentry. This may be negative.
3068 * Returns an error code otherwise.
3070 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3072 const struct open_flags *op,
3073 int open_flag, umode_t mode)
3075 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3076 struct inode *dir = nd->path.dentry->d_inode;
3079 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3080 open_flag &= ~O_TRUNC;
3082 if (nd->flags & LOOKUP_DIRECTORY)
3083 open_flag |= O_DIRECTORY;
3085 file->f_path.dentry = DENTRY_NOT_SET;
3086 file->f_path.mnt = nd->path.mnt;
3087 error = dir->i_op->atomic_open(dir, dentry, file,
3088 open_to_namei_flags(open_flag), mode);
3089 d_lookup_done(dentry);
3091 if (file->f_mode & FMODE_OPENED) {
3093 * We didn't have the inode before the open, so check open
3096 int acc_mode = op->acc_mode;
3097 if (file->f_mode & FMODE_CREATED) {
3098 WARN_ON(!(open_flag & O_CREAT));
3099 fsnotify_create(dir, dentry);
3102 error = may_open(&file->f_path, acc_mode, open_flag);
3103 if (WARN_ON(error > 0))
3105 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3108 if (file->f_path.dentry) {
3110 dentry = file->f_path.dentry;
3112 if (file->f_mode & FMODE_CREATED)
3113 fsnotify_create(dir, dentry);
3114 if (unlikely(d_is_negative(dentry)))
3120 dentry = ERR_PTR(error);
3126 * Look up and maybe create and open the last component.
3128 * Must be called with parent locked (exclusive in O_CREAT case).
3130 * Returns 0 on success, that is, if
3131 * the file was successfully atomically created (if necessary) and opened, or
3132 * the file was not completely opened at this time, though lookups and
3133 * creations were performed.
3134 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3135 * In the latter case dentry returned in @path might be negative if O_CREAT
3136 * hadn't been specified.
3138 * An error code is returned on failure.
3140 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3141 const struct open_flags *op,
3144 struct dentry *dir = nd->path.dentry;
3145 struct inode *dir_inode = dir->d_inode;
3146 int open_flag = op->open_flag;
3147 struct dentry *dentry;
3148 int error, create_error = 0;
3149 umode_t mode = op->mode;
3150 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3152 if (unlikely(IS_DEADDIR(dir_inode)))
3153 return ERR_PTR(-ENOENT);
3155 file->f_mode &= ~FMODE_CREATED;
3156 dentry = d_lookup(dir, &nd->last);
3159 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3163 if (d_in_lookup(dentry))
3166 error = d_revalidate(dentry, nd->flags);
3167 if (likely(error > 0))
3171 d_invalidate(dentry);
3175 if (dentry->d_inode) {
3176 /* Cached positive dentry: will open in f_op->open */
3181 * Checking write permission is tricky, bacuse we don't know if we are
3182 * going to actually need it: O_CREAT opens should work as long as the
3183 * file exists. But checking existence breaks atomicity. The trick is
3184 * to check access and if not granted clear O_CREAT from the flags.
3186 * Another problem is returing the "right" error value (e.g. for an
3187 * O_EXCL open we want to return EEXIST not EROFS).
3189 if (open_flag & O_CREAT) {
3190 if (!IS_POSIXACL(dir->d_inode))
3191 mode &= ~current_umask();
3192 if (unlikely(!got_write)) {
3193 create_error = -EROFS;
3194 open_flag &= ~O_CREAT;
3195 if (open_flag & (O_EXCL | O_TRUNC))
3197 /* No side effects, safe to clear O_CREAT */
3199 create_error = may_o_create(&nd->path, dentry, mode);
3201 open_flag &= ~O_CREAT;
3202 if (open_flag & O_EXCL)
3206 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3207 unlikely(!got_write)) {
3209 * No O_CREATE -> atomicity not a requirement -> fall
3210 * back to lookup + open
3215 if (dir_inode->i_op->atomic_open) {
3216 dentry = atomic_open(nd, dentry, file, op, open_flag, mode);
3217 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3218 dentry = ERR_PTR(create_error);
3223 if (d_in_lookup(dentry)) {
3224 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3226 d_lookup_done(dentry);
3227 if (unlikely(res)) {
3229 error = PTR_ERR(res);
3237 /* Negative dentry, just create the file */
3238 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3239 file->f_mode |= FMODE_CREATED;
3240 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3241 if (!dir_inode->i_op->create) {
3245 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3246 open_flag & O_EXCL);
3249 fsnotify_create(dir_inode, dentry);
3251 if (unlikely(create_error) && !dentry->d_inode) {
3252 error = create_error;
3259 return ERR_PTR(error);
3263 * Handle the last step of open()
3265 static int do_last(struct nameidata *nd,
3266 struct file *file, const struct open_flags *op)
3268 struct dentry *dir = nd->path.dentry;
3269 kuid_t dir_uid = nd->inode->i_uid;
3270 umode_t dir_mode = nd->inode->i_mode;
3271 int open_flag = op->open_flag;
3272 bool will_truncate = (open_flag & O_TRUNC) != 0;
3273 bool got_write = false;
3274 int acc_mode = op->acc_mode;
3276 struct inode *inode;
3277 struct dentry *dentry;
3280 nd->flags &= ~LOOKUP_PARENT;
3281 nd->flags |= op->intent;
3283 if (nd->last_type != LAST_NORM) {
3286 error = handle_dots(nd, nd->last_type);
3287 if (unlikely(error))
3292 if (!(open_flag & O_CREAT)) {
3293 if (nd->last.name[nd->last.len])
3294 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3295 /* we _can_ be in RCU mode here */
3296 dentry = lookup_fast(nd, &inode, &seq);
3298 return PTR_ERR(dentry);
3302 BUG_ON(nd->inode != dir->d_inode);
3303 BUG_ON(nd->flags & LOOKUP_RCU);
3305 /* create side of things */
3307 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3308 * has been cleared when we got to the last component we are
3311 error = complete_walk(nd);
3315 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3316 /* trailing slashes? */
3317 if (unlikely(nd->last.name[nd->last.len]))
3321 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3322 error = mnt_want_write(nd->path.mnt);
3326 * do _not_ fail yet - we might not need that or fail with
3327 * a different error; let lookup_open() decide; we'll be
3328 * dropping this one anyway.
3331 if (open_flag & O_CREAT)
3332 inode_lock(dir->d_inode);
3334 inode_lock_shared(dir->d_inode);
3335 dentry = lookup_open(nd, file, op, got_write);
3336 if (open_flag & O_CREAT)
3337 inode_unlock(dir->d_inode);
3339 inode_unlock_shared(dir->d_inode);
3341 if (IS_ERR(dentry)) {
3342 error = PTR_ERR(dentry);
3346 if (file->f_mode & FMODE_OPENED) {
3347 if ((file->f_mode & FMODE_CREATED) ||
3348 !S_ISREG(file_inode(file)->i_mode))
3349 will_truncate = false;
3351 audit_inode(nd->name, file->f_path.dentry, 0);
3356 if (file->f_mode & FMODE_CREATED) {
3357 /* Don't check for write permission, don't truncate */
3358 open_flag &= ~O_TRUNC;
3359 will_truncate = false;
3361 dput(nd->path.dentry);
3362 nd->path.dentry = dentry;
3363 goto finish_open_created;
3367 * If atomic_open() acquired write access it is dropped now due to
3368 * possible mount and symlink following (this might be optimized away if
3372 mnt_drop_write(nd->path.mnt);
3379 error = step_into(nd, 0, dentry, inode, seq);
3380 if (unlikely(error))
3383 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3384 audit_inode(nd->name, nd->path.dentry, 0);
3388 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3389 error = complete_walk(nd);
3392 audit_inode(nd->name, nd->path.dentry, 0);
3393 if (open_flag & O_CREAT) {
3395 if (d_is_dir(nd->path.dentry))
3397 error = may_create_in_sticky(dir_mode, dir_uid,
3398 d_backing_inode(nd->path.dentry));
3399 if (unlikely(error))
3403 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3405 if (!d_is_reg(nd->path.dentry))
3406 will_truncate = false;
3408 if (will_truncate) {
3409 error = mnt_want_write(nd->path.mnt);
3414 finish_open_created:
3415 error = may_open(&nd->path, acc_mode, open_flag);
3418 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
3419 error = vfs_open(&nd->path, file);
3423 error = ima_file_check(file, op->acc_mode);
3424 if (!error && will_truncate)
3425 error = handle_truncate(file);
3427 if (unlikely(error > 0)) {
3432 mnt_drop_write(nd->path.mnt);
3436 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3438 struct dentry *child = NULL;
3439 struct inode *dir = dentry->d_inode;
3440 struct inode *inode;
3443 /* we want directory to be writable */
3444 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3447 error = -EOPNOTSUPP;
3448 if (!dir->i_op->tmpfile)
3451 child = d_alloc(dentry, &slash_name);
3452 if (unlikely(!child))
3454 error = dir->i_op->tmpfile(dir, child, mode);
3458 inode = child->d_inode;
3459 if (unlikely(!inode))
3461 if (!(open_flag & O_EXCL)) {
3462 spin_lock(&inode->i_lock);
3463 inode->i_state |= I_LINKABLE;
3464 spin_unlock(&inode->i_lock);
3466 ima_post_create_tmpfile(inode);
3471 return ERR_PTR(error);
3473 EXPORT_SYMBOL(vfs_tmpfile);
3475 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3476 const struct open_flags *op,
3479 struct dentry *child;
3481 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3482 if (unlikely(error))
3484 error = mnt_want_write(path.mnt);
3485 if (unlikely(error))
3487 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3488 error = PTR_ERR(child);
3492 path.dentry = child;
3493 audit_inode(nd->name, child, 0);
3494 /* Don't check for other permissions, the inode was just created */
3495 error = may_open(&path, 0, op->open_flag);
3498 file->f_path.mnt = path.mnt;
3499 error = finish_open(file, child, NULL);
3501 mnt_drop_write(path.mnt);
3507 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3510 int error = path_lookupat(nd, flags, &path);
3512 audit_inode(nd->name, path.dentry, 0);
3513 error = vfs_open(&path, file);
3519 static struct file *path_openat(struct nameidata *nd,
3520 const struct open_flags *op, unsigned flags)
3525 file = alloc_empty_file(op->open_flag, current_cred());
3529 if (unlikely(file->f_flags & __O_TMPFILE)) {
3530 error = do_tmpfile(nd, flags, op, file);
3531 } else if (unlikely(file->f_flags & O_PATH)) {
3532 error = do_o_path(nd, flags, file);
3534 const char *s = path_init(nd, flags);
3535 while (!(error = link_path_walk(s, nd)) &&
3536 (error = do_last(nd, file, op)) > 0) {
3537 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3538 s = trailing_symlink(nd);
3542 if (likely(!error)) {
3543 if (likely(file->f_mode & FMODE_OPENED))
3549 if (error == -EOPENSTALE) {
3550 if (flags & LOOKUP_RCU)
3555 return ERR_PTR(error);
3558 struct file *do_filp_open(int dfd, struct filename *pathname,
3559 const struct open_flags *op)
3561 struct nameidata nd;
3562 int flags = op->lookup_flags;
3565 set_nameidata(&nd, dfd, pathname);
3566 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3567 if (unlikely(filp == ERR_PTR(-ECHILD)))
3568 filp = path_openat(&nd, op, flags);
3569 if (unlikely(filp == ERR_PTR(-ESTALE)))
3570 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3571 restore_nameidata();
3575 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3576 const char *name, const struct open_flags *op)
3578 struct nameidata nd;
3580 struct filename *filename;
3581 int flags = op->lookup_flags | LOOKUP_ROOT;
3584 nd.root.dentry = dentry;
3586 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3587 return ERR_PTR(-ELOOP);
3589 filename = getname_kernel(name);
3590 if (IS_ERR(filename))
3591 return ERR_CAST(filename);
3593 set_nameidata(&nd, -1, filename);
3594 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3595 if (unlikely(file == ERR_PTR(-ECHILD)))
3596 file = path_openat(&nd, op, flags);
3597 if (unlikely(file == ERR_PTR(-ESTALE)))
3598 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3599 restore_nameidata();
3604 static struct dentry *filename_create(int dfd, struct filename *name,
3605 struct path *path, unsigned int lookup_flags)
3607 struct dentry *dentry = ERR_PTR(-EEXIST);
3612 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3615 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3616 * other flags passed in are ignored!
3618 lookup_flags &= LOOKUP_REVAL;
3620 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3622 return ERR_CAST(name);
3625 * Yucky last component or no last component at all?
3626 * (foo/., foo/.., /////)
3628 if (unlikely(type != LAST_NORM))
3631 /* don't fail immediately if it's r/o, at least try to report other errors */
3632 err2 = mnt_want_write(path->mnt);
3634 * Do the final lookup.
3636 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3637 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3638 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3643 if (d_is_positive(dentry))
3647 * Special case - lookup gave negative, but... we had foo/bar/
3648 * From the vfs_mknod() POV we just have a negative dentry -
3649 * all is fine. Let's be bastards - you had / on the end, you've
3650 * been asking for (non-existent) directory. -ENOENT for you.
3652 if (unlikely(!is_dir && last.name[last.len])) {
3656 if (unlikely(err2)) {
3664 dentry = ERR_PTR(error);
3666 inode_unlock(path->dentry->d_inode);
3668 mnt_drop_write(path->mnt);
3675 struct dentry *kern_path_create(int dfd, const char *pathname,
3676 struct path *path, unsigned int lookup_flags)
3678 return filename_create(dfd, getname_kernel(pathname),
3679 path, lookup_flags);
3681 EXPORT_SYMBOL(kern_path_create);
3683 void done_path_create(struct path *path, struct dentry *dentry)
3686 inode_unlock(path->dentry->d_inode);
3687 mnt_drop_write(path->mnt);
3690 EXPORT_SYMBOL(done_path_create);
3692 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3693 struct path *path, unsigned int lookup_flags)
3695 return filename_create(dfd, getname(pathname), path, lookup_flags);
3697 EXPORT_SYMBOL(user_path_create);
3699 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3701 int error = may_create(dir, dentry);
3706 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3709 if (!dir->i_op->mknod)
3712 error = devcgroup_inode_mknod(mode, dev);
3716 error = security_inode_mknod(dir, dentry, mode, dev);
3720 error = dir->i_op->mknod(dir, dentry, mode, dev);
3722 fsnotify_create(dir, dentry);
3725 EXPORT_SYMBOL(vfs_mknod);
3727 static int may_mknod(umode_t mode)
3729 switch (mode & S_IFMT) {
3735 case 0: /* zero mode translates to S_IFREG */
3744 long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3747 struct dentry *dentry;
3750 unsigned int lookup_flags = 0;
3752 error = may_mknod(mode);
3756 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3758 return PTR_ERR(dentry);
3760 if (!IS_POSIXACL(path.dentry->d_inode))
3761 mode &= ~current_umask();
3762 error = security_path_mknod(&path, dentry, mode, dev);
3765 switch (mode & S_IFMT) {
3766 case 0: case S_IFREG:
3767 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3769 ima_post_path_mknod(dentry);
3771 case S_IFCHR: case S_IFBLK:
3772 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3773 new_decode_dev(dev));
3775 case S_IFIFO: case S_IFSOCK:
3776 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3780 done_path_create(&path, dentry);
3781 if (retry_estale(error, lookup_flags)) {
3782 lookup_flags |= LOOKUP_REVAL;
3788 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3791 return do_mknodat(dfd, filename, mode, dev);
3794 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3796 return do_mknodat(AT_FDCWD, filename, mode, dev);
3799 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3801 int error = may_create(dir, dentry);
3802 unsigned max_links = dir->i_sb->s_max_links;
3807 if (!dir->i_op->mkdir)
3810 mode &= (S_IRWXUGO|S_ISVTX);
3811 error = security_inode_mkdir(dir, dentry, mode);
3815 if (max_links && dir->i_nlink >= max_links)
3818 error = dir->i_op->mkdir(dir, dentry, mode);
3820 fsnotify_mkdir(dir, dentry);
3823 EXPORT_SYMBOL(vfs_mkdir);
3825 long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3827 struct dentry *dentry;
3830 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3833 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3835 return PTR_ERR(dentry);
3837 if (!IS_POSIXACL(path.dentry->d_inode))
3838 mode &= ~current_umask();
3839 error = security_path_mkdir(&path, dentry, mode);
3841 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3842 done_path_create(&path, dentry);
3843 if (retry_estale(error, lookup_flags)) {
3844 lookup_flags |= LOOKUP_REVAL;
3850 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3852 return do_mkdirat(dfd, pathname, mode);
3855 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3857 return do_mkdirat(AT_FDCWD, pathname, mode);
3860 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3862 int error = may_delete(dir, dentry, 1);
3867 if (!dir->i_op->rmdir)
3871 inode_lock(dentry->d_inode);
3874 if (is_local_mountpoint(dentry))
3877 error = security_inode_rmdir(dir, dentry);
3881 error = dir->i_op->rmdir(dir, dentry);
3885 shrink_dcache_parent(dentry);
3886 dentry->d_inode->i_flags |= S_DEAD;
3888 detach_mounts(dentry);
3889 fsnotify_rmdir(dir, dentry);
3892 inode_unlock(dentry->d_inode);
3898 EXPORT_SYMBOL(vfs_rmdir);
3900 long do_rmdir(int dfd, const char __user *pathname)
3903 struct filename *name;
3904 struct dentry *dentry;
3908 unsigned int lookup_flags = 0;
3910 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3911 &path, &last, &type);
3913 return PTR_ERR(name);
3927 error = mnt_want_write(path.mnt);
3931 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3932 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3933 error = PTR_ERR(dentry);
3936 if (!dentry->d_inode) {
3940 error = security_path_rmdir(&path, dentry);
3943 error = vfs_rmdir(path.dentry->d_inode, dentry);
3947 inode_unlock(path.dentry->d_inode);
3948 mnt_drop_write(path.mnt);
3952 if (retry_estale(error, lookup_flags)) {
3953 lookup_flags |= LOOKUP_REVAL;
3959 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3961 return do_rmdir(AT_FDCWD, pathname);
3965 * vfs_unlink - unlink a filesystem object
3966 * @dir: parent directory
3968 * @delegated_inode: returns victim inode, if the inode is delegated.
3970 * The caller must hold dir->i_mutex.
3972 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3973 * return a reference to the inode in delegated_inode. The caller
3974 * should then break the delegation on that inode and retry. Because
3975 * breaking a delegation may take a long time, the caller should drop
3976 * dir->i_mutex before doing so.
3978 * Alternatively, a caller may pass NULL for delegated_inode. This may
3979 * be appropriate for callers that expect the underlying filesystem not
3980 * to be NFS exported.
3982 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3984 struct inode *target = dentry->d_inode;
3985 int error = may_delete(dir, dentry, 0);
3990 if (!dir->i_op->unlink)
3994 if (is_local_mountpoint(dentry))
3997 error = security_inode_unlink(dir, dentry);
3999 error = try_break_deleg(target, delegated_inode);
4002 error = dir->i_op->unlink(dir, dentry);
4005 detach_mounts(dentry);
4006 fsnotify_unlink(dir, dentry);
4011 inode_unlock(target);
4013 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4014 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
4015 fsnotify_link_count(target);
4021 EXPORT_SYMBOL(vfs_unlink);
4024 * Make sure that the actual truncation of the file will occur outside its
4025 * directory's i_mutex. Truncate can take a long time if there is a lot of
4026 * writeout happening, and we don't want to prevent access to the directory
4027 * while waiting on the I/O.
4029 long do_unlinkat(int dfd, struct filename *name)
4032 struct dentry *dentry;
4036 struct inode *inode = NULL;
4037 struct inode *delegated_inode = NULL;
4038 unsigned int lookup_flags = 0;
4040 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4042 return PTR_ERR(name);
4045 if (type != LAST_NORM)
4048 error = mnt_want_write(path.mnt);
4052 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4053 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4054 error = PTR_ERR(dentry);
4055 if (!IS_ERR(dentry)) {
4056 /* Why not before? Because we want correct error value */
4057 if (last.name[last.len])
4059 inode = dentry->d_inode;
4060 if (d_is_negative(dentry))
4063 error = security_path_unlink(&path, dentry);
4066 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4070 inode_unlock(path.dentry->d_inode);
4072 iput(inode); /* truncate the inode here */
4074 if (delegated_inode) {
4075 error = break_deleg_wait(&delegated_inode);
4079 mnt_drop_write(path.mnt);
4082 if (retry_estale(error, lookup_flags)) {
4083 lookup_flags |= LOOKUP_REVAL;
4091 if (d_is_negative(dentry))
4093 else if (d_is_dir(dentry))
4100 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4102 if ((flag & ~AT_REMOVEDIR) != 0)
4105 if (flag & AT_REMOVEDIR)
4106 return do_rmdir(dfd, pathname);
4108 return do_unlinkat(dfd, getname(pathname));
4111 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4113 return do_unlinkat(AT_FDCWD, getname(pathname));
4116 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4118 int error = may_create(dir, dentry);
4123 if (!dir->i_op->symlink)
4126 error = security_inode_symlink(dir, dentry, oldname);
4130 error = dir->i_op->symlink(dir, dentry, oldname);
4132 fsnotify_create(dir, dentry);
4135 EXPORT_SYMBOL(vfs_symlink);
4137 long do_symlinkat(const char __user *oldname, int newdfd,
4138 const char __user *newname)
4141 struct filename *from;
4142 struct dentry *dentry;
4144 unsigned int lookup_flags = 0;
4146 from = getname(oldname);
4148 return PTR_ERR(from);
4150 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4151 error = PTR_ERR(dentry);
4155 error = security_path_symlink(&path, dentry, from->name);
4157 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4158 done_path_create(&path, dentry);
4159 if (retry_estale(error, lookup_flags)) {
4160 lookup_flags |= LOOKUP_REVAL;
4168 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4169 int, newdfd, const char __user *, newname)
4171 return do_symlinkat(oldname, newdfd, newname);
4174 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4176 return do_symlinkat(oldname, AT_FDCWD, newname);
4180 * vfs_link - create a new link
4181 * @old_dentry: object to be linked
4183 * @new_dentry: where to create the new link
4184 * @delegated_inode: returns inode needing a delegation break
4186 * The caller must hold dir->i_mutex
4188 * If vfs_link discovers a delegation on the to-be-linked file in need
4189 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4190 * inode in delegated_inode. The caller should then break the delegation
4191 * and retry. Because breaking a delegation may take a long time, the
4192 * caller should drop the i_mutex before doing so.
4194 * Alternatively, a caller may pass NULL for delegated_inode. This may
4195 * be appropriate for callers that expect the underlying filesystem not
4196 * to be NFS exported.
4198 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4200 struct inode *inode = old_dentry->d_inode;
4201 unsigned max_links = dir->i_sb->s_max_links;
4207 error = may_create(dir, new_dentry);
4211 if (dir->i_sb != inode->i_sb)
4215 * A link to an append-only or immutable file cannot be created.
4217 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4220 * Updating the link count will likely cause i_uid and i_gid to
4221 * be writen back improperly if their true value is unknown to
4224 if (HAS_UNMAPPED_ID(inode))
4226 if (!dir->i_op->link)
4228 if (S_ISDIR(inode->i_mode))
4231 error = security_inode_link(old_dentry, dir, new_dentry);
4236 /* Make sure we don't allow creating hardlink to an unlinked file */
4237 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4239 else if (max_links && inode->i_nlink >= max_links)
4242 error = try_break_deleg(inode, delegated_inode);
4244 error = dir->i_op->link(old_dentry, dir, new_dentry);
4247 if (!error && (inode->i_state & I_LINKABLE)) {
4248 spin_lock(&inode->i_lock);
4249 inode->i_state &= ~I_LINKABLE;
4250 spin_unlock(&inode->i_lock);
4252 inode_unlock(inode);
4254 fsnotify_link(dir, inode, new_dentry);
4257 EXPORT_SYMBOL(vfs_link);
4260 * Hardlinks are often used in delicate situations. We avoid
4261 * security-related surprises by not following symlinks on the
4264 * We don't follow them on the oldname either to be compatible
4265 * with linux 2.0, and to avoid hard-linking to directories
4266 * and other special files. --ADM
4268 int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4269 const char __user *newname, int flags)
4271 struct dentry *new_dentry;
4272 struct path old_path, new_path;
4273 struct inode *delegated_inode = NULL;
4277 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4280 * To use null names we require CAP_DAC_READ_SEARCH
4281 * This ensures that not everyone will be able to create
4282 * handlink using the passed filedescriptor.
4284 if (flags & AT_EMPTY_PATH) {
4285 if (!capable(CAP_DAC_READ_SEARCH))
4290 if (flags & AT_SYMLINK_FOLLOW)
4291 how |= LOOKUP_FOLLOW;
4293 error = user_path_at(olddfd, oldname, how, &old_path);
4297 new_dentry = user_path_create(newdfd, newname, &new_path,
4298 (how & LOOKUP_REVAL));
4299 error = PTR_ERR(new_dentry);
4300 if (IS_ERR(new_dentry))
4304 if (old_path.mnt != new_path.mnt)
4306 error = may_linkat(&old_path);
4307 if (unlikely(error))
4309 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4312 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4314 done_path_create(&new_path, new_dentry);
4315 if (delegated_inode) {
4316 error = break_deleg_wait(&delegated_inode);
4318 path_put(&old_path);
4322 if (retry_estale(error, how)) {
4323 path_put(&old_path);
4324 how |= LOOKUP_REVAL;
4328 path_put(&old_path);
4333 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4334 int, newdfd, const char __user *, newname, int, flags)
4336 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4339 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4341 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4345 * vfs_rename - rename a filesystem object
4346 * @old_dir: parent of source
4347 * @old_dentry: source
4348 * @new_dir: parent of destination
4349 * @new_dentry: destination
4350 * @delegated_inode: returns an inode needing a delegation break
4351 * @flags: rename flags
4353 * The caller must hold multiple mutexes--see lock_rename()).
4355 * If vfs_rename discovers a delegation in need of breaking at either
4356 * the source or destination, it will return -EWOULDBLOCK and return a
4357 * reference to the inode in delegated_inode. The caller should then
4358 * break the delegation and retry. Because breaking a delegation may
4359 * take a long time, the caller should drop all locks before doing
4362 * Alternatively, a caller may pass NULL for delegated_inode. This may
4363 * be appropriate for callers that expect the underlying filesystem not
4364 * to be NFS exported.
4366 * The worst of all namespace operations - renaming directory. "Perverted"
4367 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4370 * a) we can get into loop creation.
4371 * b) race potential - two innocent renames can create a loop together.
4372 * That's where 4.4 screws up. Current fix: serialization on
4373 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4375 * c) we have to lock _four_ objects - parents and victim (if it exists),
4376 * and source (if it is not a directory).
4377 * And that - after we got ->i_mutex on parents (until then we don't know
4378 * whether the target exists). Solution: try to be smart with locking
4379 * order for inodes. We rely on the fact that tree topology may change
4380 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4381 * move will be locked. Thus we can rank directories by the tree
4382 * (ancestors first) and rank all non-directories after them.
4383 * That works since everybody except rename does "lock parent, lookup,
4384 * lock child" and rename is under ->s_vfs_rename_mutex.
4385 * HOWEVER, it relies on the assumption that any object with ->lookup()
4386 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4387 * we'd better make sure that there's no link(2) for them.
4388 * d) conversion from fhandle to dentry may come in the wrong moment - when
4389 * we are removing the target. Solution: we will have to grab ->i_mutex
4390 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4391 * ->i_mutex on parents, which works but leads to some truly excessive
4394 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4395 struct inode *new_dir, struct dentry *new_dentry,
4396 struct inode **delegated_inode, unsigned int flags)
4399 bool is_dir = d_is_dir(old_dentry);
4400 struct inode *source = old_dentry->d_inode;
4401 struct inode *target = new_dentry->d_inode;
4402 bool new_is_dir = false;
4403 unsigned max_links = new_dir->i_sb->s_max_links;
4404 struct name_snapshot old_name;
4406 if (source == target)
4409 error = may_delete(old_dir, old_dentry, is_dir);
4414 error = may_create(new_dir, new_dentry);
4416 new_is_dir = d_is_dir(new_dentry);
4418 if (!(flags & RENAME_EXCHANGE))
4419 error = may_delete(new_dir, new_dentry, is_dir);
4421 error = may_delete(new_dir, new_dentry, new_is_dir);
4426 if (!old_dir->i_op->rename)
4430 * If we are going to change the parent - check write permissions,
4431 * we'll need to flip '..'.
4433 if (new_dir != old_dir) {
4435 error = inode_permission(source, MAY_WRITE);
4439 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4440 error = inode_permission(target, MAY_WRITE);
4446 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4451 take_dentry_name_snapshot(&old_name, old_dentry);
4453 if (!is_dir || (flags & RENAME_EXCHANGE))
4454 lock_two_nondirectories(source, target);
4459 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4462 if (max_links && new_dir != old_dir) {
4464 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4466 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4467 old_dir->i_nlink >= max_links)
4471 error = try_break_deleg(source, delegated_inode);
4475 if (target && !new_is_dir) {
4476 error = try_break_deleg(target, delegated_inode);
4480 error = old_dir->i_op->rename(old_dir, old_dentry,
4481 new_dir, new_dentry, flags);
4485 if (!(flags & RENAME_EXCHANGE) && target) {
4487 shrink_dcache_parent(new_dentry);
4488 target->i_flags |= S_DEAD;
4490 dont_mount(new_dentry);
4491 detach_mounts(new_dentry);
4493 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4494 if (!(flags & RENAME_EXCHANGE))
4495 d_move(old_dentry, new_dentry);
4497 d_exchange(old_dentry, new_dentry);
4500 if (!is_dir || (flags & RENAME_EXCHANGE))
4501 unlock_two_nondirectories(source, target);
4503 inode_unlock(target);
4506 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4507 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4508 if (flags & RENAME_EXCHANGE) {
4509 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4510 new_is_dir, NULL, new_dentry);
4513 release_dentry_name_snapshot(&old_name);
4517 EXPORT_SYMBOL(vfs_rename);
4519 static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4520 const char __user *newname, unsigned int flags)
4522 struct dentry *old_dentry, *new_dentry;
4523 struct dentry *trap;
4524 struct path old_path, new_path;
4525 struct qstr old_last, new_last;
4526 int old_type, new_type;
4527 struct inode *delegated_inode = NULL;
4528 struct filename *from;
4529 struct filename *to;
4530 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4531 bool should_retry = false;
4534 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4537 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4538 (flags & RENAME_EXCHANGE))
4541 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4544 if (flags & RENAME_EXCHANGE)
4548 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4549 &old_path, &old_last, &old_type);
4551 error = PTR_ERR(from);
4555 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4556 &new_path, &new_last, &new_type);
4558 error = PTR_ERR(to);
4563 if (old_path.mnt != new_path.mnt)
4567 if (old_type != LAST_NORM)
4570 if (flags & RENAME_NOREPLACE)
4572 if (new_type != LAST_NORM)
4575 error = mnt_want_write(old_path.mnt);
4580 trap = lock_rename(new_path.dentry, old_path.dentry);
4582 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4583 error = PTR_ERR(old_dentry);
4584 if (IS_ERR(old_dentry))
4586 /* source must exist */
4588 if (d_is_negative(old_dentry))
4590 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4591 error = PTR_ERR(new_dentry);
4592 if (IS_ERR(new_dentry))
4595 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4597 if (flags & RENAME_EXCHANGE) {
4599 if (d_is_negative(new_dentry))
4602 if (!d_is_dir(new_dentry)) {
4604 if (new_last.name[new_last.len])
4608 /* unless the source is a directory trailing slashes give -ENOTDIR */
4609 if (!d_is_dir(old_dentry)) {
4611 if (old_last.name[old_last.len])
4613 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4616 /* source should not be ancestor of target */
4618 if (old_dentry == trap)
4620 /* target should not be an ancestor of source */
4621 if (!(flags & RENAME_EXCHANGE))
4623 if (new_dentry == trap)
4626 error = security_path_rename(&old_path, old_dentry,
4627 &new_path, new_dentry, flags);
4630 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4631 new_path.dentry->d_inode, new_dentry,
4632 &delegated_inode, flags);
4638 unlock_rename(new_path.dentry, old_path.dentry);
4639 if (delegated_inode) {
4640 error = break_deleg_wait(&delegated_inode);
4644 mnt_drop_write(old_path.mnt);
4646 if (retry_estale(error, lookup_flags))
4647 should_retry = true;
4648 path_put(&new_path);
4651 path_put(&old_path);
4654 should_retry = false;
4655 lookup_flags |= LOOKUP_REVAL;
4662 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4663 int, newdfd, const char __user *, newname, unsigned int, flags)
4665 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4668 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4669 int, newdfd, const char __user *, newname)
4671 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
4674 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4676 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4679 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4681 int error = may_create(dir, dentry);
4685 if (!dir->i_op->mknod)
4688 return dir->i_op->mknod(dir, dentry,
4689 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4691 EXPORT_SYMBOL(vfs_whiteout);
4693 int readlink_copy(char __user *buffer, int buflen, const char *link)
4695 int len = PTR_ERR(link);
4700 if (len > (unsigned) buflen)
4702 if (copy_to_user(buffer, link, len))
4709 * vfs_readlink - copy symlink body into userspace buffer
4710 * @dentry: dentry on which to get symbolic link
4711 * @buffer: user memory pointer
4712 * @buflen: size of buffer
4714 * Does not touch atime. That's up to the caller if necessary
4716 * Does not call security hook.
4718 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4720 struct inode *inode = d_inode(dentry);
4721 DEFINE_DELAYED_CALL(done);
4725 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4726 if (unlikely(inode->i_op->readlink))
4727 return inode->i_op->readlink(dentry, buffer, buflen);
4729 if (!d_is_symlink(dentry))
4732 spin_lock(&inode->i_lock);
4733 inode->i_opflags |= IOP_DEFAULT_READLINK;
4734 spin_unlock(&inode->i_lock);
4737 link = READ_ONCE(inode->i_link);
4739 link = inode->i_op->get_link(dentry, inode, &done);
4741 return PTR_ERR(link);
4743 res = readlink_copy(buffer, buflen, link);
4744 do_delayed_call(&done);
4747 EXPORT_SYMBOL(vfs_readlink);
4750 * vfs_get_link - get symlink body
4751 * @dentry: dentry on which to get symbolic link
4752 * @done: caller needs to free returned data with this
4754 * Calls security hook and i_op->get_link() on the supplied inode.
4756 * It does not touch atime. That's up to the caller if necessary.
4758 * Does not work on "special" symlinks like /proc/$$/fd/N
4760 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4762 const char *res = ERR_PTR(-EINVAL);
4763 struct inode *inode = d_inode(dentry);
4765 if (d_is_symlink(dentry)) {
4766 res = ERR_PTR(security_inode_readlink(dentry));
4768 res = inode->i_op->get_link(dentry, inode, done);
4772 EXPORT_SYMBOL(vfs_get_link);
4774 /* get the link contents into pagecache */
4775 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4776 struct delayed_call *callback)
4780 struct address_space *mapping = inode->i_mapping;
4783 page = find_get_page(mapping, 0);
4785 return ERR_PTR(-ECHILD);
4786 if (!PageUptodate(page)) {
4788 return ERR_PTR(-ECHILD);
4791 page = read_mapping_page(mapping, 0, NULL);
4795 set_delayed_call(callback, page_put_link, page);
4796 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4797 kaddr = page_address(page);
4798 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4802 EXPORT_SYMBOL(page_get_link);
4804 void page_put_link(void *arg)
4808 EXPORT_SYMBOL(page_put_link);
4810 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4812 DEFINE_DELAYED_CALL(done);
4813 int res = readlink_copy(buffer, buflen,
4814 page_get_link(dentry, d_inode(dentry),
4816 do_delayed_call(&done);
4819 EXPORT_SYMBOL(page_readlink);
4822 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4824 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4826 struct address_space *mapping = inode->i_mapping;
4830 unsigned int flags = 0;
4832 flags |= AOP_FLAG_NOFS;
4835 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4836 flags, &page, &fsdata);
4840 memcpy(page_address(page), symname, len-1);
4842 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4849 mark_inode_dirty(inode);
4854 EXPORT_SYMBOL(__page_symlink);
4856 int page_symlink(struct inode *inode, const char *symname, int len)
4858 return __page_symlink(inode, symname, len,
4859 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4861 EXPORT_SYMBOL(page_symlink);
4863 const struct inode_operations page_symlink_inode_operations = {
4864 .get_link = page_get_link,
4866 EXPORT_SYMBOL(page_symlink_inode_operations);