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 inline void path_to_nameidata(const struct path *path,
862 struct nameidata *nd)
864 if (!(nd->flags & LOOKUP_RCU)) {
865 dput(nd->path.dentry);
866 if (nd->path.mnt != path->mnt)
867 mntput(nd->path.mnt);
869 nd->path.mnt = path->mnt;
870 nd->path.dentry = path->dentry;
873 static int nd_jump_root(struct nameidata *nd)
875 if (unlikely(nd->flags & LOOKUP_BENEATH))
877 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
878 /* Absolute path arguments to path_init() are allowed. */
879 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
883 int error = set_root(nd);
887 if (nd->flags & LOOKUP_RCU) {
891 nd->inode = d->d_inode;
892 nd->seq = nd->root_seq;
893 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
899 nd->inode = nd->path.dentry->d_inode;
901 nd->flags |= LOOKUP_JUMPED;
906 * Helper to directly jump to a known parsed path from ->get_link,
907 * caller must have taken a reference to path beforehand.
909 int nd_jump_link(struct path *path)
912 struct nameidata *nd = current->nameidata;
914 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
918 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
919 if (nd->path.mnt != path->mnt)
922 /* Not currently safe for scoped-lookups. */
923 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
928 nd->inode = nd->path.dentry->d_inode;
929 nd->flags |= LOOKUP_JUMPED;
937 static inline void put_link(struct nameidata *nd)
939 struct saved *last = nd->stack + --nd->depth;
940 do_delayed_call(&last->done);
941 if (!(nd->flags & LOOKUP_RCU))
942 path_put(&last->link);
945 int sysctl_protected_symlinks __read_mostly = 0;
946 int sysctl_protected_hardlinks __read_mostly = 0;
947 int sysctl_protected_fifos __read_mostly;
948 int sysctl_protected_regular __read_mostly;
951 * may_follow_link - Check symlink following for unsafe situations
952 * @nd: nameidata pathwalk data
954 * In the case of the sysctl_protected_symlinks sysctl being enabled,
955 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
956 * in a sticky world-writable directory. This is to protect privileged
957 * processes from failing races against path names that may change out
958 * from under them by way of other users creating malicious symlinks.
959 * It will permit symlinks to be followed only when outside a sticky
960 * world-writable directory, or when the uid of the symlink and follower
961 * match, or when the directory owner matches the symlink's owner.
963 * Returns 0 if following the symlink is allowed, -ve on error.
965 static inline int may_follow_link(struct nameidata *nd)
967 const struct inode *inode;
968 const struct inode *parent;
971 if (!sysctl_protected_symlinks)
974 /* Allowed if owner and follower match. */
975 inode = nd->link_inode;
976 if (uid_eq(current_cred()->fsuid, inode->i_uid))
979 /* Allowed if parent directory not sticky and world-writable. */
981 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
984 /* Allowed if parent directory and link owner match. */
985 puid = parent->i_uid;
986 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
989 if (nd->flags & LOOKUP_RCU)
992 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
993 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
998 * safe_hardlink_source - Check for safe hardlink conditions
999 * @inode: the source inode to hardlink from
1001 * Return false if at least one of the following conditions:
1002 * - inode is not a regular file
1004 * - inode is setgid and group-exec
1005 * - access failure for read and write
1007 * Otherwise returns true.
1009 static bool safe_hardlink_source(struct inode *inode)
1011 umode_t mode = inode->i_mode;
1013 /* Special files should not get pinned to the filesystem. */
1017 /* Setuid files should not get pinned to the filesystem. */
1021 /* Executable setgid files should not get pinned to the filesystem. */
1022 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1025 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1026 if (inode_permission(inode, MAY_READ | MAY_WRITE))
1033 * may_linkat - Check permissions for creating a hardlink
1034 * @link: the source to hardlink from
1036 * Block hardlink when all of:
1037 * - sysctl_protected_hardlinks enabled
1038 * - fsuid does not match inode
1039 * - hardlink source is unsafe (see safe_hardlink_source() above)
1040 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1042 * Returns 0 if successful, -ve on error.
1044 static int may_linkat(struct path *link)
1046 struct inode *inode = link->dentry->d_inode;
1048 /* Inode writeback is not safe when the uid or gid are invalid. */
1049 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
1052 if (!sysctl_protected_hardlinks)
1055 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1056 * otherwise, it must be a safe source.
1058 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
1061 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1066 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1067 * should be allowed, or not, on files that already
1069 * @dir_mode: mode bits of directory
1070 * @dir_uid: owner of directory
1071 * @inode: the inode of the file to open
1073 * Block an O_CREAT open of a FIFO (or a regular file) when:
1074 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1075 * - the file already exists
1076 * - we are in a sticky directory
1077 * - we don't own the file
1078 * - the owner of the directory doesn't own the file
1079 * - the directory is world writable
1080 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1081 * the directory doesn't have to be world writable: being group writable will
1084 * Returns 0 if the open is allowed, -ve on error.
1086 static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid,
1087 struct inode * const inode)
1089 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1090 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1091 likely(!(dir_mode & S_ISVTX)) ||
1092 uid_eq(inode->i_uid, dir_uid) ||
1093 uid_eq(current_fsuid(), inode->i_uid))
1096 if (likely(dir_mode & 0002) ||
1098 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1099 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1100 const char *operation = S_ISFIFO(inode->i_mode) ?
1101 "sticky_create_fifo" :
1102 "sticky_create_regular";
1103 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1109 static __always_inline
1110 const char *get_link(struct nameidata *nd)
1112 struct saved *last = nd->stack + nd->depth - 1;
1113 struct dentry *dentry = last->link.dentry;
1114 struct inode *inode = nd->link_inode;
1118 if (!(nd->flags & LOOKUP_PARENT)) {
1119 error = may_follow_link(nd);
1120 if (unlikely(error))
1121 return ERR_PTR(error);
1124 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS))
1125 return ERR_PTR(-ELOOP);
1127 if (!(nd->flags & LOOKUP_RCU)) {
1128 touch_atime(&last->link);
1130 } else if (atime_needs_update(&last->link, inode)) {
1131 if (unlikely(unlazy_walk(nd)))
1132 return ERR_PTR(-ECHILD);
1133 touch_atime(&last->link);
1136 error = security_inode_follow_link(dentry, inode,
1137 nd->flags & LOOKUP_RCU);
1138 if (unlikely(error))
1139 return ERR_PTR(error);
1141 nd->last_type = LAST_BIND;
1142 res = READ_ONCE(inode->i_link);
1144 const char * (*get)(struct dentry *, struct inode *,
1145 struct delayed_call *);
1146 get = inode->i_op->get_link;
1147 if (nd->flags & LOOKUP_RCU) {
1148 res = get(NULL, inode, &last->done);
1149 if (res == ERR_PTR(-ECHILD)) {
1150 if (unlikely(unlazy_walk(nd)))
1151 return ERR_PTR(-ECHILD);
1152 res = get(dentry, inode, &last->done);
1155 res = get(dentry, inode, &last->done);
1157 if (IS_ERR_OR_NULL(res))
1161 error = nd_jump_root(nd);
1162 if (unlikely(error))
1163 return ERR_PTR(error);
1164 while (unlikely(*++res == '/'))
1173 * follow_up - Find the mountpoint of path's vfsmount
1175 * Given a path, find the mountpoint of its source file system.
1176 * Replace @path with the path of the mountpoint in the parent mount.
1179 * Return 1 if we went up a level and 0 if we were already at the
1182 int follow_up(struct path *path)
1184 struct mount *mnt = real_mount(path->mnt);
1185 struct mount *parent;
1186 struct dentry *mountpoint;
1188 read_seqlock_excl(&mount_lock);
1189 parent = mnt->mnt_parent;
1190 if (parent == mnt) {
1191 read_sequnlock_excl(&mount_lock);
1194 mntget(&parent->mnt);
1195 mountpoint = dget(mnt->mnt_mountpoint);
1196 read_sequnlock_excl(&mount_lock);
1198 path->dentry = mountpoint;
1200 path->mnt = &parent->mnt;
1203 EXPORT_SYMBOL(follow_up);
1206 * Perform an automount
1207 * - return -EISDIR to tell follow_managed() to stop and return the path we
1210 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1212 struct dentry *dentry = path->dentry;
1214 /* We don't want to mount if someone's just doing a stat -
1215 * unless they're stat'ing a directory and appended a '/' to
1218 * We do, however, want to mount if someone wants to open or
1219 * create a file of any type under the mountpoint, wants to
1220 * traverse through the mountpoint or wants to open the
1221 * mounted directory. Also, autofs may mark negative dentries
1222 * as being automount points. These will need the attentions
1223 * of the daemon to instantiate them before they can be used.
1225 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1226 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1230 if (count && (*count)++ >= MAXSYMLINKS)
1233 return finish_automount(dentry->d_op->d_automount(path), path);
1237 * Handle a dentry that is managed in some way.
1238 * - Flagged for transit management (autofs)
1239 * - Flagged as mountpoint
1240 * - Flagged as automount point
1242 * This may only be called in refwalk mode.
1243 * On success path->dentry is known positive.
1245 * Serialization is taken care of in namespace.c
1247 static int follow_managed(struct path *path, struct nameidata *nd)
1249 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1251 bool need_mntput = false;
1254 /* Given that we're not holding a lock here, we retain the value in a
1255 * local variable for each dentry as we look at it so that we don't see
1256 * the components of that value change under us */
1257 while (flags = smp_load_acquire(&path->dentry->d_flags),
1258 unlikely(flags & DCACHE_MANAGED_DENTRY)) {
1259 /* Allow the filesystem to manage the transit without i_mutex
1261 if (flags & DCACHE_MANAGE_TRANSIT) {
1262 BUG_ON(!path->dentry->d_op);
1263 BUG_ON(!path->dentry->d_op->d_manage);
1264 ret = path->dentry->d_op->d_manage(path, false);
1265 flags = smp_load_acquire(&path->dentry->d_flags);
1270 /* Transit to a mounted filesystem. */
1271 if (flags & DCACHE_MOUNTED) {
1272 struct vfsmount *mounted = lookup_mnt(path);
1277 path->mnt = mounted;
1278 path->dentry = dget(mounted->mnt_root);
1283 /* Something is mounted on this dentry in another
1284 * namespace and/or whatever was mounted there in this
1285 * namespace got unmounted before lookup_mnt() could
1289 /* Handle an automount point */
1290 if (flags & DCACHE_NEED_AUTOMOUNT) {
1291 ret = follow_automount(path, &nd->total_link_count,
1298 /* We didn't change the current path point */
1303 if (path->mnt == mnt)
1305 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1308 nd->flags |= LOOKUP_JUMPED;
1310 if (ret == -EISDIR || !ret)
1312 if (ret > 0 && unlikely(d_flags_negative(flags)))
1314 if (unlikely(ret < 0)) {
1316 if (path->mnt != nd->path.mnt)
1322 int follow_down_one(struct path *path)
1324 struct vfsmount *mounted;
1326 mounted = lookup_mnt(path);
1330 path->mnt = mounted;
1331 path->dentry = dget(mounted->mnt_root);
1336 EXPORT_SYMBOL(follow_down_one);
1338 static inline int managed_dentry_rcu(const struct path *path)
1340 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1341 path->dentry->d_op->d_manage(path, true) : 0;
1345 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1346 * we meet a managed dentry that would need blocking.
1348 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1349 struct inode **inode, unsigned *seqp)
1352 struct mount *mounted;
1354 * Don't forget we might have a non-mountpoint managed dentry
1355 * that wants to block transit.
1357 switch (managed_dentry_rcu(path)) {
1367 if (!d_mountpoint(path->dentry))
1368 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1370 mounted = __lookup_mnt(path->mnt, path->dentry);
1373 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1375 path->mnt = &mounted->mnt;
1376 path->dentry = mounted->mnt.mnt_root;
1377 nd->flags |= LOOKUP_JUMPED;
1378 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1380 * Update the inode too. We don't need to re-check the
1381 * dentry sequence number here after this d_inode read,
1382 * because a mount-point is always pinned.
1384 *inode = path->dentry->d_inode;
1386 return !read_seqretry(&mount_lock, nd->m_seq) &&
1387 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1390 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1391 struct path *path, struct inode **inode,
1396 path->mnt = nd->path.mnt;
1397 path->dentry = dentry;
1398 if (nd->flags & LOOKUP_RCU) {
1399 unsigned int seq = *seqp;
1400 if (unlikely(!*inode))
1402 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1404 if (unlazy_child(nd, dentry, seq))
1406 // *path might've been clobbered by __follow_mount_rcu()
1407 path->mnt = nd->path.mnt;
1408 path->dentry = dentry;
1410 ret = follow_managed(path, nd);
1411 if (likely(ret >= 0)) {
1412 *inode = d_backing_inode(path->dentry);
1413 *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1418 static int follow_dotdot_rcu(struct nameidata *nd)
1420 struct inode *inode = nd->inode;
1423 if (path_equal(&nd->path, &nd->root)) {
1424 if (unlikely(nd->flags & LOOKUP_BENEATH))
1428 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1429 struct dentry *old = nd->path.dentry;
1430 struct dentry *parent = old->d_parent;
1433 inode = parent->d_inode;
1434 seq = read_seqcount_begin(&parent->d_seq);
1435 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1437 nd->path.dentry = parent;
1439 if (unlikely(!path_connected(&nd->path)))
1443 struct mount *mnt = real_mount(nd->path.mnt);
1444 struct mount *mparent = mnt->mnt_parent;
1445 struct dentry *mountpoint = mnt->mnt_mountpoint;
1446 struct inode *inode2 = mountpoint->d_inode;
1447 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1448 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1450 if (&mparent->mnt == nd->path.mnt)
1452 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1454 /* we know that mountpoint was pinned */
1455 nd->path.dentry = mountpoint;
1456 nd->path.mnt = &mparent->mnt;
1461 while (unlikely(d_mountpoint(nd->path.dentry))) {
1462 struct mount *mounted;
1463 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1464 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1468 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1470 nd->path.mnt = &mounted->mnt;
1471 nd->path.dentry = mounted->mnt.mnt_root;
1472 inode = nd->path.dentry->d_inode;
1473 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1480 * Follow down to the covering mount currently visible to userspace. At each
1481 * point, the filesystem owning that dentry may be queried as to whether the
1482 * caller is permitted to proceed or not.
1484 int follow_down(struct path *path)
1489 while (managed = READ_ONCE(path->dentry->d_flags),
1490 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1491 /* Allow the filesystem to manage the transit without i_mutex
1494 * We indicate to the filesystem if someone is trying to mount
1495 * something here. This gives autofs the chance to deny anyone
1496 * other than its daemon the right to mount on its
1499 * The filesystem may sleep at this point.
1501 if (managed & DCACHE_MANAGE_TRANSIT) {
1502 BUG_ON(!path->dentry->d_op);
1503 BUG_ON(!path->dentry->d_op->d_manage);
1504 ret = path->dentry->d_op->d_manage(path, false);
1506 return ret == -EISDIR ? 0 : ret;
1509 /* Transit to a mounted filesystem. */
1510 if (managed & DCACHE_MOUNTED) {
1511 struct vfsmount *mounted = lookup_mnt(path);
1516 path->mnt = mounted;
1517 path->dentry = dget(mounted->mnt_root);
1521 /* Don't handle automount points here */
1526 EXPORT_SYMBOL(follow_down);
1529 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1531 static void follow_mount(struct path *path)
1533 while (d_mountpoint(path->dentry)) {
1534 struct vfsmount *mounted = lookup_mnt(path);
1539 path->mnt = mounted;
1540 path->dentry = dget(mounted->mnt_root);
1544 static int path_parent_directory(struct path *path)
1546 struct dentry *old = path->dentry;
1547 /* rare case of legitimate dget_parent()... */
1548 path->dentry = dget_parent(path->dentry);
1550 if (unlikely(!path_connected(path)))
1555 static int follow_dotdot(struct nameidata *nd)
1558 if (path_equal(&nd->path, &nd->root)) {
1559 if (unlikely(nd->flags & LOOKUP_BENEATH))
1563 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1564 int ret = path_parent_directory(&nd->path);
1569 if (!follow_up(&nd->path))
1571 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1574 follow_mount(&nd->path);
1575 nd->inode = nd->path.dentry->d_inode;
1580 * This looks up the name in dcache and possibly revalidates the found dentry.
1581 * NULL is returned if the dentry does not exist in the cache.
1583 static struct dentry *lookup_dcache(const struct qstr *name,
1587 struct dentry *dentry = d_lookup(dir, name);
1589 int error = d_revalidate(dentry, flags);
1590 if (unlikely(error <= 0)) {
1592 d_invalidate(dentry);
1594 return ERR_PTR(error);
1601 * Parent directory has inode locked exclusive. This is one
1602 * and only case when ->lookup() gets called on non in-lookup
1603 * dentries - as the matter of fact, this only gets called
1604 * when directory is guaranteed to have no in-lookup children
1607 static struct dentry *__lookup_hash(const struct qstr *name,
1608 struct dentry *base, unsigned int flags)
1610 struct dentry *dentry = lookup_dcache(name, base, flags);
1612 struct inode *dir = base->d_inode;
1617 /* Don't create child dentry for a dead directory. */
1618 if (unlikely(IS_DEADDIR(dir)))
1619 return ERR_PTR(-ENOENT);
1621 dentry = d_alloc(base, name);
1622 if (unlikely(!dentry))
1623 return ERR_PTR(-ENOMEM);
1625 old = dir->i_op->lookup(dir, dentry, flags);
1626 if (unlikely(old)) {
1633 static struct dentry *lookup_fast(struct nameidata *nd,
1634 struct inode **inode,
1637 struct dentry *dentry, *parent = nd->path.dentry;
1641 * Rename seqlock is not required here because in the off chance
1642 * of a false negative due to a concurrent rename, the caller is
1643 * going to fall back to non-racy lookup.
1645 if (nd->flags & LOOKUP_RCU) {
1647 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1648 if (unlikely(!dentry)) {
1649 if (unlazy_walk(nd))
1650 return ERR_PTR(-ECHILD);
1655 * This sequence count validates that the inode matches
1656 * the dentry name information from lookup.
1658 *inode = d_backing_inode(dentry);
1659 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1660 return ERR_PTR(-ECHILD);
1663 * This sequence count validates that the parent had no
1664 * changes while we did the lookup of the dentry above.
1666 * The memory barrier in read_seqcount_begin of child is
1667 * enough, we can use __read_seqcount_retry here.
1669 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1670 return ERR_PTR(-ECHILD);
1673 status = d_revalidate(dentry, nd->flags);
1674 if (likely(status > 0))
1676 if (unlazy_child(nd, dentry, seq))
1677 return ERR_PTR(-ECHILD);
1678 if (unlikely(status == -ECHILD))
1679 /* we'd been told to redo it in non-rcu mode */
1680 status = d_revalidate(dentry, nd->flags);
1682 dentry = __d_lookup(parent, &nd->last);
1683 if (unlikely(!dentry))
1685 status = d_revalidate(dentry, nd->flags);
1687 if (unlikely(status <= 0)) {
1689 d_invalidate(dentry);
1691 return ERR_PTR(status);
1696 /* Fast lookup failed, do it the slow way */
1697 static struct dentry *__lookup_slow(const struct qstr *name,
1701 struct dentry *dentry, *old;
1702 struct inode *inode = dir->d_inode;
1703 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1705 /* Don't go there if it's already dead */
1706 if (unlikely(IS_DEADDIR(inode)))
1707 return ERR_PTR(-ENOENT);
1709 dentry = d_alloc_parallel(dir, name, &wq);
1712 if (unlikely(!d_in_lookup(dentry))) {
1713 int error = d_revalidate(dentry, flags);
1714 if (unlikely(error <= 0)) {
1716 d_invalidate(dentry);
1721 dentry = ERR_PTR(error);
1724 old = inode->i_op->lookup(inode, dentry, flags);
1725 d_lookup_done(dentry);
1726 if (unlikely(old)) {
1734 static struct dentry *lookup_slow(const struct qstr *name,
1738 struct inode *inode = dir->d_inode;
1740 inode_lock_shared(inode);
1741 res = __lookup_slow(name, dir, flags);
1742 inode_unlock_shared(inode);
1746 static inline int may_lookup(struct nameidata *nd)
1748 if (nd->flags & LOOKUP_RCU) {
1749 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1752 if (unlazy_walk(nd))
1755 return inode_permission(nd->inode, MAY_EXEC);
1758 static inline int handle_dots(struct nameidata *nd, int type)
1760 if (type == LAST_DOTDOT) {
1763 if (!nd->root.mnt) {
1764 error = set_root(nd);
1768 if (nd->flags & LOOKUP_RCU)
1769 error = follow_dotdot_rcu(nd);
1771 error = follow_dotdot(nd);
1775 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1777 * If there was a racing rename or mount along our
1778 * path, then we can't be sure that ".." hasn't jumped
1779 * above nd->root (and so userspace should retry or use
1783 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1785 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1792 static int pick_link(struct nameidata *nd, struct path *link,
1793 struct inode *inode, unsigned seq)
1797 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1798 path_to_nameidata(link, nd);
1801 if (!(nd->flags & LOOKUP_RCU)) {
1802 if (link->mnt == nd->path.mnt)
1805 error = nd_alloc_stack(nd);
1806 if (unlikely(error)) {
1807 if (error == -ECHILD) {
1808 if (unlikely(!legitimize_path(nd, link, seq))) {
1811 nd->flags &= ~LOOKUP_RCU;
1812 nd->path.mnt = NULL;
1813 nd->path.dentry = NULL;
1815 } else if (likely(unlazy_walk(nd)) == 0)
1816 error = nd_alloc_stack(nd);
1824 last = nd->stack + nd->depth++;
1826 clear_delayed_call(&last->done);
1827 nd->link_inode = inode;
1832 enum {WALK_FOLLOW = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1835 * Do we need to follow links? We _really_ want to be able
1836 * to do this check without having to look at inode->i_op,
1837 * so we keep a cache of "no, this doesn't need follow_link"
1838 * for the common case.
1840 static int step_into(struct nameidata *nd, int flags,
1841 struct dentry *dentry, struct inode *inode, unsigned seq)
1844 int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1848 if (likely(!d_is_symlink(path.dentry)) ||
1849 !((flags & WALK_FOLLOW) || (nd->flags & LOOKUP_FOLLOW)) ||
1850 (flags & WALK_NOFOLLOW)) {
1851 /* not a symlink or should not follow */
1852 path_to_nameidata(&path, nd);
1857 /* make sure that d_is_symlink above matches inode */
1858 if (nd->flags & LOOKUP_RCU) {
1859 if (read_seqcount_retry(&path.dentry->d_seq, seq))
1862 return pick_link(nd, &path, inode, seq);
1865 static int walk_component(struct nameidata *nd, int flags)
1867 struct dentry *dentry;
1868 struct inode *inode;
1872 * "." and ".." are special - ".." especially so because it has
1873 * to be able to know about the current root directory and
1874 * parent relationships.
1876 if (unlikely(nd->last_type != LAST_NORM)) {
1877 if (!(flags & WALK_MORE) && nd->depth)
1879 err = handle_dots(nd, nd->last_type);
1882 dentry = lookup_fast(nd, &inode, &seq);
1884 return PTR_ERR(dentry);
1885 if (unlikely(!dentry)) {
1886 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1888 return PTR_ERR(dentry);
1890 if (!(flags & WALK_MORE) && nd->depth)
1892 return step_into(nd, flags, dentry, inode, seq);
1896 * We can do the critical dentry name comparison and hashing
1897 * operations one word at a time, but we are limited to:
1899 * - Architectures with fast unaligned word accesses. We could
1900 * do a "get_unaligned()" if this helps and is sufficiently
1903 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1904 * do not trap on the (extremely unlikely) case of a page
1905 * crossing operation.
1907 * - Furthermore, we need an efficient 64-bit compile for the
1908 * 64-bit case in order to generate the "number of bytes in
1909 * the final mask". Again, that could be replaced with a
1910 * efficient population count instruction or similar.
1912 #ifdef CONFIG_DCACHE_WORD_ACCESS
1914 #include <asm/word-at-a-time.h>
1918 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1920 #elif defined(CONFIG_64BIT)
1922 * Register pressure in the mixing function is an issue, particularly
1923 * on 32-bit x86, but almost any function requires one state value and
1924 * one temporary. Instead, use a function designed for two state values
1925 * and no temporaries.
1927 * This function cannot create a collision in only two iterations, so
1928 * we have two iterations to achieve avalanche. In those two iterations,
1929 * we have six layers of mixing, which is enough to spread one bit's
1930 * influence out to 2^6 = 64 state bits.
1932 * Rotate constants are scored by considering either 64 one-bit input
1933 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1934 * probability of that delta causing a change to each of the 128 output
1935 * bits, using a sample of random initial states.
1937 * The Shannon entropy of the computed probabilities is then summed
1938 * to produce a score. Ideally, any input change has a 50% chance of
1939 * toggling any given output bit.
1941 * Mixing scores (in bits) for (12,45):
1942 * Input delta: 1-bit 2-bit
1943 * 1 round: 713.3 42542.6
1944 * 2 rounds: 2753.7 140389.8
1945 * 3 rounds: 5954.1 233458.2
1946 * 4 rounds: 7862.6 256672.2
1947 * Perfect: 8192 258048
1948 * (64*128) (64*63/2 * 128)
1950 #define HASH_MIX(x, y, a) \
1952 y ^= x, x = rol64(x,12),\
1953 x += y, y = rol64(y,45),\
1957 * Fold two longs into one 32-bit hash value. This must be fast, but
1958 * latency isn't quite as critical, as there is a fair bit of additional
1959 * work done before the hash value is used.
1961 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1963 y ^= x * GOLDEN_RATIO_64;
1964 y *= GOLDEN_RATIO_64;
1968 #else /* 32-bit case */
1971 * Mixing scores (in bits) for (7,20):
1972 * Input delta: 1-bit 2-bit
1973 * 1 round: 330.3 9201.6
1974 * 2 rounds: 1246.4 25475.4
1975 * 3 rounds: 1907.1 31295.1
1976 * 4 rounds: 2042.3 31718.6
1977 * Perfect: 2048 31744
1978 * (32*64) (32*31/2 * 64)
1980 #define HASH_MIX(x, y, a) \
1982 y ^= x, x = rol32(x, 7),\
1983 x += y, y = rol32(y,20),\
1986 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1988 /* Use arch-optimized multiply if one exists */
1989 return __hash_32(y ^ __hash_32(x));
1995 * Return the hash of a string of known length. This is carfully
1996 * designed to match hash_name(), which is the more critical function.
1997 * In particular, we must end by hashing a final word containing 0..7
1998 * payload bytes, to match the way that hash_name() iterates until it
1999 * finds the delimiter after the name.
2001 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2003 unsigned long a, x = 0, y = (unsigned long)salt;
2008 a = load_unaligned_zeropad(name);
2009 if (len < sizeof(unsigned long))
2012 name += sizeof(unsigned long);
2013 len -= sizeof(unsigned long);
2015 x ^= a & bytemask_from_count(len);
2017 return fold_hash(x, y);
2019 EXPORT_SYMBOL(full_name_hash);
2021 /* Return the "hash_len" (hash and length) of a null-terminated string */
2022 u64 hashlen_string(const void *salt, const char *name)
2024 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2025 unsigned long adata, mask, len;
2026 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2033 len += sizeof(unsigned long);
2035 a = load_unaligned_zeropad(name+len);
2036 } while (!has_zero(a, &adata, &constants));
2038 adata = prep_zero_mask(a, adata, &constants);
2039 mask = create_zero_mask(adata);
2040 x ^= a & zero_bytemask(mask);
2042 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2044 EXPORT_SYMBOL(hashlen_string);
2047 * Calculate the length and hash of the path component, and
2048 * return the "hash_len" as the result.
2050 static inline u64 hash_name(const void *salt, const char *name)
2052 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2053 unsigned long adata, bdata, mask, len;
2054 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2061 len += sizeof(unsigned long);
2063 a = load_unaligned_zeropad(name+len);
2064 b = a ^ REPEAT_BYTE('/');
2065 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2067 adata = prep_zero_mask(a, adata, &constants);
2068 bdata = prep_zero_mask(b, bdata, &constants);
2069 mask = create_zero_mask(adata | bdata);
2070 x ^= a & zero_bytemask(mask);
2072 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2075 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2077 /* Return the hash of a string of known length */
2078 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2080 unsigned long hash = init_name_hash(salt);
2082 hash = partial_name_hash((unsigned char)*name++, hash);
2083 return end_name_hash(hash);
2085 EXPORT_SYMBOL(full_name_hash);
2087 /* Return the "hash_len" (hash and length) of a null-terminated string */
2088 u64 hashlen_string(const void *salt, const char *name)
2090 unsigned long hash = init_name_hash(salt);
2091 unsigned long len = 0, c;
2093 c = (unsigned char)*name;
2096 hash = partial_name_hash(c, hash);
2097 c = (unsigned char)name[len];
2099 return hashlen_create(end_name_hash(hash), len);
2101 EXPORT_SYMBOL(hashlen_string);
2104 * We know there's a real path component here of at least
2107 static inline u64 hash_name(const void *salt, const char *name)
2109 unsigned long hash = init_name_hash(salt);
2110 unsigned long len = 0, c;
2112 c = (unsigned char)*name;
2115 hash = partial_name_hash(c, hash);
2116 c = (unsigned char)name[len];
2117 } while (c && c != '/');
2118 return hashlen_create(end_name_hash(hash), len);
2125 * This is the basic name resolution function, turning a pathname into
2126 * the final dentry. We expect 'base' to be positive and a directory.
2128 * Returns 0 and nd will have valid dentry and mnt on success.
2129 * Returns error and drops reference to input namei data on failure.
2131 static int link_path_walk(const char *name, struct nameidata *nd)
2136 return PTR_ERR(name);
2142 /* At this point we know we have a real path component. */
2147 err = may_lookup(nd);
2151 hash_len = hash_name(nd->path.dentry, name);
2154 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2156 if (name[1] == '.') {
2158 nd->flags |= LOOKUP_JUMPED;
2164 if (likely(type == LAST_NORM)) {
2165 struct dentry *parent = nd->path.dentry;
2166 nd->flags &= ~LOOKUP_JUMPED;
2167 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2168 struct qstr this = { { .hash_len = hash_len }, .name = name };
2169 err = parent->d_op->d_hash(parent, &this);
2172 hash_len = this.hash_len;
2177 nd->last.hash_len = hash_len;
2178 nd->last.name = name;
2179 nd->last_type = type;
2181 name += hashlen_len(hash_len);
2185 * If it wasn't NUL, we know it was '/'. Skip that
2186 * slash, and continue until no more slashes.
2190 } while (unlikely(*name == '/'));
2191 if (unlikely(!*name)) {
2193 /* pathname body, done */
2196 name = nd->stack[nd->depth - 1].name;
2197 /* trailing symlink, done */
2200 /* last component of nested symlink */
2201 err = walk_component(nd, WALK_FOLLOW);
2203 /* not the last component */
2204 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2210 const char *s = get_link(nd);
2219 nd->stack[nd->depth - 1].name = name;
2224 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2225 if (nd->flags & LOOKUP_RCU) {
2226 if (unlazy_walk(nd))
2234 /* must be paired with terminate_walk() */
2235 static const char *path_init(struct nameidata *nd, unsigned flags)
2238 const char *s = nd->name->name;
2241 flags &= ~LOOKUP_RCU;
2242 if (flags & LOOKUP_RCU)
2245 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2246 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2249 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2250 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2253 if (flags & LOOKUP_ROOT) {
2254 struct dentry *root = nd->root.dentry;
2255 struct inode *inode = root->d_inode;
2256 if (*s && unlikely(!d_can_lookup(root)))
2257 return ERR_PTR(-ENOTDIR);
2258 nd->path = nd->root;
2260 if (flags & LOOKUP_RCU) {
2261 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2262 nd->root_seq = nd->seq;
2264 path_get(&nd->path);
2269 nd->root.mnt = NULL;
2270 nd->path.mnt = NULL;
2271 nd->path.dentry = NULL;
2273 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2274 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2275 error = nd_jump_root(nd);
2276 if (unlikely(error))
2277 return ERR_PTR(error);
2281 /* Relative pathname -- get the starting-point it is relative to. */
2282 if (nd->dfd == AT_FDCWD) {
2283 if (flags & LOOKUP_RCU) {
2284 struct fs_struct *fs = current->fs;
2288 seq = read_seqcount_begin(&fs->seq);
2290 nd->inode = nd->path.dentry->d_inode;
2291 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2292 } while (read_seqcount_retry(&fs->seq, seq));
2294 get_fs_pwd(current->fs, &nd->path);
2295 nd->inode = nd->path.dentry->d_inode;
2298 /* Caller must check execute permissions on the starting path component */
2299 struct fd f = fdget_raw(nd->dfd);
2300 struct dentry *dentry;
2303 return ERR_PTR(-EBADF);
2305 dentry = f.file->f_path.dentry;
2307 if (*s && unlikely(!d_can_lookup(dentry))) {
2309 return ERR_PTR(-ENOTDIR);
2312 nd->path = f.file->f_path;
2313 if (flags & LOOKUP_RCU) {
2314 nd->inode = nd->path.dentry->d_inode;
2315 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2317 path_get(&nd->path);
2318 nd->inode = nd->path.dentry->d_inode;
2323 /* For scoped-lookups we need to set the root to the dirfd as well. */
2324 if (flags & LOOKUP_IS_SCOPED) {
2325 nd->root = nd->path;
2326 if (flags & LOOKUP_RCU) {
2327 nd->root_seq = nd->seq;
2329 path_get(&nd->root);
2330 nd->flags |= LOOKUP_ROOT_GRABBED;
2336 static inline const char *lookup_last(struct nameidata *nd)
2339 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2340 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2342 nd->flags &= ~LOOKUP_PARENT;
2343 err = walk_component(nd, 0);
2344 if (unlikely(err)) {
2347 return PTR_ERR(err);
2350 nd->flags |= LOOKUP_PARENT;
2351 nd->stack[0].name = NULL;
2358 static int handle_lookup_down(struct nameidata *nd)
2360 if (!(nd->flags & LOOKUP_RCU))
2361 dget(nd->path.dentry);
2362 return step_into(nd, WALK_NOFOLLOW,
2363 nd->path.dentry, nd->inode, nd->seq);
2366 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2367 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2369 const char *s = path_init(nd, flags);
2372 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2373 err = handle_lookup_down(nd);
2374 if (unlikely(err < 0))
2378 while (!(err = link_path_walk(s, nd)) &&
2379 (s = lookup_last(nd)) != NULL)
2382 err = complete_walk(nd);
2384 if (!err && nd->flags & LOOKUP_DIRECTORY)
2385 if (!d_can_lookup(nd->path.dentry))
2387 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2388 err = handle_lookup_down(nd);
2389 nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please...
2393 nd->path.mnt = NULL;
2394 nd->path.dentry = NULL;
2400 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2401 struct path *path, struct path *root)
2404 struct nameidata nd;
2406 return PTR_ERR(name);
2407 if (unlikely(root)) {
2409 flags |= LOOKUP_ROOT;
2411 set_nameidata(&nd, dfd, name);
2412 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2413 if (unlikely(retval == -ECHILD))
2414 retval = path_lookupat(&nd, flags, path);
2415 if (unlikely(retval == -ESTALE))
2416 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2418 if (likely(!retval))
2419 audit_inode(name, path->dentry,
2420 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2421 restore_nameidata();
2426 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2427 static int path_parentat(struct nameidata *nd, unsigned flags,
2428 struct path *parent)
2430 const char *s = path_init(nd, flags);
2431 int err = link_path_walk(s, nd);
2433 err = complete_walk(nd);
2436 nd->path.mnt = NULL;
2437 nd->path.dentry = NULL;
2443 static struct filename *filename_parentat(int dfd, struct filename *name,
2444 unsigned int flags, struct path *parent,
2445 struct qstr *last, int *type)
2448 struct nameidata nd;
2452 set_nameidata(&nd, dfd, name);
2453 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2454 if (unlikely(retval == -ECHILD))
2455 retval = path_parentat(&nd, flags, parent);
2456 if (unlikely(retval == -ESTALE))
2457 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2458 if (likely(!retval)) {
2460 *type = nd.last_type;
2461 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2464 name = ERR_PTR(retval);
2466 restore_nameidata();
2470 /* does lookup, returns the object with parent locked */
2471 struct dentry *kern_path_locked(const char *name, struct path *path)
2473 struct filename *filename;
2478 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2480 if (IS_ERR(filename))
2481 return ERR_CAST(filename);
2482 if (unlikely(type != LAST_NORM)) {
2485 return ERR_PTR(-EINVAL);
2487 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2488 d = __lookup_hash(&last, path->dentry, 0);
2490 inode_unlock(path->dentry->d_inode);
2497 int kern_path(const char *name, unsigned int flags, struct path *path)
2499 return filename_lookup(AT_FDCWD, getname_kernel(name),
2502 EXPORT_SYMBOL(kern_path);
2505 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2506 * @dentry: pointer to dentry of the base directory
2507 * @mnt: pointer to vfs mount of the base directory
2508 * @name: pointer to file name
2509 * @flags: lookup flags
2510 * @path: pointer to struct path to fill
2512 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2513 const char *name, unsigned int flags,
2516 struct path root = {.mnt = mnt, .dentry = dentry};
2517 /* the first argument of filename_lookup() is ignored with root */
2518 return filename_lookup(AT_FDCWD, getname_kernel(name),
2519 flags , path, &root);
2521 EXPORT_SYMBOL(vfs_path_lookup);
2523 static int lookup_one_len_common(const char *name, struct dentry *base,
2524 int len, struct qstr *this)
2528 this->hash = full_name_hash(base, name, len);
2532 if (unlikely(name[0] == '.')) {
2533 if (len < 2 || (len == 2 && name[1] == '.'))
2538 unsigned int c = *(const unsigned char *)name++;
2539 if (c == '/' || c == '\0')
2543 * See if the low-level filesystem might want
2544 * to use its own hash..
2546 if (base->d_flags & DCACHE_OP_HASH) {
2547 int err = base->d_op->d_hash(base, this);
2552 return inode_permission(base->d_inode, MAY_EXEC);
2556 * try_lookup_one_len - filesystem helper to lookup single pathname component
2557 * @name: pathname component to lookup
2558 * @base: base directory to lookup from
2559 * @len: maximum length @len should be interpreted to
2561 * Look up a dentry by name in the dcache, returning NULL if it does not
2562 * currently exist. The function does not try to create a dentry.
2564 * Note that this routine is purely a helper for filesystem usage and should
2565 * not be called by generic code.
2567 * The caller must hold base->i_mutex.
2569 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2574 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2576 err = lookup_one_len_common(name, base, len, &this);
2578 return ERR_PTR(err);
2580 return lookup_dcache(&this, base, 0);
2582 EXPORT_SYMBOL(try_lookup_one_len);
2585 * lookup_one_len - filesystem helper to lookup single pathname component
2586 * @name: pathname component to lookup
2587 * @base: base directory to lookup from
2588 * @len: maximum length @len should be interpreted to
2590 * Note that this routine is purely a helper for filesystem usage and should
2591 * not be called by generic code.
2593 * The caller must hold base->i_mutex.
2595 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2597 struct dentry *dentry;
2601 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2603 err = lookup_one_len_common(name, base, len, &this);
2605 return ERR_PTR(err);
2607 dentry = lookup_dcache(&this, base, 0);
2608 return dentry ? dentry : __lookup_slow(&this, base, 0);
2610 EXPORT_SYMBOL(lookup_one_len);
2613 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2614 * @name: pathname component to lookup
2615 * @base: base directory to lookup from
2616 * @len: maximum length @len should be interpreted to
2618 * Note that this routine is purely a helper for filesystem usage and should
2619 * not be called by generic code.
2621 * Unlike lookup_one_len, it should be called without the parent
2622 * i_mutex held, and will take the i_mutex itself if necessary.
2624 struct dentry *lookup_one_len_unlocked(const char *name,
2625 struct dentry *base, int len)
2631 err = lookup_one_len_common(name, base, len, &this);
2633 return ERR_PTR(err);
2635 ret = lookup_dcache(&this, base, 0);
2637 ret = lookup_slow(&this, base, 0);
2640 EXPORT_SYMBOL(lookup_one_len_unlocked);
2643 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2644 * on negatives. Returns known positive or ERR_PTR(); that's what
2645 * most of the users want. Note that pinned negative with unlocked parent
2646 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2647 * need to be very careful; pinned positives have ->d_inode stable, so
2648 * this one avoids such problems.
2650 struct dentry *lookup_positive_unlocked(const char *name,
2651 struct dentry *base, int len)
2653 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2654 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2656 ret = ERR_PTR(-ENOENT);
2660 EXPORT_SYMBOL(lookup_positive_unlocked);
2662 #ifdef CONFIG_UNIX98_PTYS
2663 int path_pts(struct path *path)
2665 /* Find something mounted on "pts" in the same directory as
2668 struct dentry *child, *parent;
2672 ret = path_parent_directory(path);
2676 parent = path->dentry;
2679 child = d_hash_and_lookup(parent, &this);
2683 path->dentry = child;
2690 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2691 struct path *path, int *empty)
2693 return filename_lookup(dfd, getname_flags(name, flags, empty),
2696 EXPORT_SYMBOL(user_path_at_empty);
2698 int __check_sticky(struct inode *dir, struct inode *inode)
2700 kuid_t fsuid = current_fsuid();
2702 if (uid_eq(inode->i_uid, fsuid))
2704 if (uid_eq(dir->i_uid, fsuid))
2706 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2708 EXPORT_SYMBOL(__check_sticky);
2711 * Check whether we can remove a link victim from directory dir, check
2712 * whether the type of victim is right.
2713 * 1. We can't do it if dir is read-only (done in permission())
2714 * 2. We should have write and exec permissions on dir
2715 * 3. We can't remove anything from append-only dir
2716 * 4. We can't do anything with immutable dir (done in permission())
2717 * 5. If the sticky bit on dir is set we should either
2718 * a. be owner of dir, or
2719 * b. be owner of victim, or
2720 * c. have CAP_FOWNER capability
2721 * 6. If the victim is append-only or immutable we can't do antyhing with
2722 * links pointing to it.
2723 * 7. If the victim has an unknown uid or gid we can't change the inode.
2724 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2725 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2726 * 10. We can't remove a root or mountpoint.
2727 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2728 * nfs_async_unlink().
2730 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2732 struct inode *inode = d_backing_inode(victim);
2735 if (d_is_negative(victim))
2739 BUG_ON(victim->d_parent->d_inode != dir);
2741 /* Inode writeback is not safe when the uid or gid are invalid. */
2742 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2745 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2747 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2753 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2754 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2757 if (!d_is_dir(victim))
2759 if (IS_ROOT(victim))
2761 } else if (d_is_dir(victim))
2763 if (IS_DEADDIR(dir))
2765 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2770 /* Check whether we can create an object with dentry child in directory
2772 * 1. We can't do it if child already exists (open has special treatment for
2773 * this case, but since we are inlined it's OK)
2774 * 2. We can't do it if dir is read-only (done in permission())
2775 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2776 * 4. We should have write and exec permissions on dir
2777 * 5. We can't do it if dir is immutable (done in permission())
2779 static inline int may_create(struct inode *dir, struct dentry *child)
2781 struct user_namespace *s_user_ns;
2782 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2785 if (IS_DEADDIR(dir))
2787 s_user_ns = dir->i_sb->s_user_ns;
2788 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2789 !kgid_has_mapping(s_user_ns, current_fsgid()))
2791 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2795 * p1 and p2 should be directories on the same fs.
2797 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2802 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2806 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2808 p = d_ancestor(p2, p1);
2810 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2811 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2815 p = d_ancestor(p1, p2);
2817 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2818 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2822 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2823 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2826 EXPORT_SYMBOL(lock_rename);
2828 void unlock_rename(struct dentry *p1, struct dentry *p2)
2830 inode_unlock(p1->d_inode);
2832 inode_unlock(p2->d_inode);
2833 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2836 EXPORT_SYMBOL(unlock_rename);
2838 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2841 int error = may_create(dir, dentry);
2845 if (!dir->i_op->create)
2846 return -EACCES; /* shouldn't it be ENOSYS? */
2849 error = security_inode_create(dir, dentry, mode);
2852 error = dir->i_op->create(dir, dentry, mode, want_excl);
2854 fsnotify_create(dir, dentry);
2857 EXPORT_SYMBOL(vfs_create);
2859 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2860 int (*f)(struct dentry *, umode_t, void *),
2863 struct inode *dir = dentry->d_parent->d_inode;
2864 int error = may_create(dir, dentry);
2870 error = security_inode_create(dir, dentry, mode);
2873 error = f(dentry, mode, arg);
2875 fsnotify_create(dir, dentry);
2878 EXPORT_SYMBOL(vfs_mkobj);
2880 bool may_open_dev(const struct path *path)
2882 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2883 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2886 static int may_open(const struct path *path, int acc_mode, int flag)
2888 struct dentry *dentry = path->dentry;
2889 struct inode *inode = dentry->d_inode;
2895 switch (inode->i_mode & S_IFMT) {
2899 if (acc_mode & MAY_WRITE)
2904 if (!may_open_dev(path))
2913 error = inode_permission(inode, MAY_OPEN | acc_mode);
2918 * An append-only file must be opened in append mode for writing.
2920 if (IS_APPEND(inode)) {
2921 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2927 /* O_NOATIME can only be set by the owner or superuser */
2928 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2934 static int handle_truncate(struct file *filp)
2936 const struct path *path = &filp->f_path;
2937 struct inode *inode = path->dentry->d_inode;
2938 int error = get_write_access(inode);
2942 * Refuse to truncate files with mandatory locks held on them.
2944 error = locks_verify_locked(filp);
2946 error = security_path_truncate(path);
2948 error = do_truncate(path->dentry, 0,
2949 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2952 put_write_access(inode);
2956 static inline int open_to_namei_flags(int flag)
2958 if ((flag & O_ACCMODE) == 3)
2963 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2965 struct user_namespace *s_user_ns;
2966 int error = security_path_mknod(dir, dentry, mode, 0);
2970 s_user_ns = dir->dentry->d_sb->s_user_ns;
2971 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2972 !kgid_has_mapping(s_user_ns, current_fsgid()))
2975 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2979 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2983 * Attempt to atomically look up, create and open a file from a negative
2986 * Returns 0 if successful. The file will have been created and attached to
2987 * @file by the filesystem calling finish_open().
2989 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
2990 * be set. The caller will need to perform the open themselves. @path will
2991 * have been updated to point to the new dentry. This may be negative.
2993 * Returns an error code otherwise.
2995 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
2997 const struct open_flags *op,
2998 int open_flag, umode_t mode)
3000 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3001 struct inode *dir = nd->path.dentry->d_inode;
3004 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3005 open_flag &= ~O_TRUNC;
3007 if (nd->flags & LOOKUP_DIRECTORY)
3008 open_flag |= O_DIRECTORY;
3010 file->f_path.dentry = DENTRY_NOT_SET;
3011 file->f_path.mnt = nd->path.mnt;
3012 error = dir->i_op->atomic_open(dir, dentry, file,
3013 open_to_namei_flags(open_flag), mode);
3014 d_lookup_done(dentry);
3016 if (file->f_mode & FMODE_OPENED) {
3018 * We didn't have the inode before the open, so check open
3021 int acc_mode = op->acc_mode;
3022 if (file->f_mode & FMODE_CREATED) {
3023 WARN_ON(!(open_flag & O_CREAT));
3024 fsnotify_create(dir, dentry);
3027 error = may_open(&file->f_path, acc_mode, open_flag);
3028 if (WARN_ON(error > 0))
3030 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3033 if (file->f_path.dentry) {
3035 dentry = file->f_path.dentry;
3037 if (file->f_mode & FMODE_CREATED)
3038 fsnotify_create(dir, dentry);
3039 if (unlikely(d_is_negative(dentry)))
3045 dentry = ERR_PTR(error);
3051 * Look up and maybe create and open the last component.
3053 * Must be called with parent locked (exclusive in O_CREAT case).
3055 * Returns 0 on success, that is, if
3056 * the file was successfully atomically created (if necessary) and opened, or
3057 * the file was not completely opened at this time, though lookups and
3058 * creations were performed.
3059 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3060 * In the latter case dentry returned in @path might be negative if O_CREAT
3061 * hadn't been specified.
3063 * An error code is returned on failure.
3065 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3066 const struct open_flags *op,
3069 struct dentry *dir = nd->path.dentry;
3070 struct inode *dir_inode = dir->d_inode;
3071 int open_flag = op->open_flag;
3072 struct dentry *dentry;
3073 int error, create_error = 0;
3074 umode_t mode = op->mode;
3075 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3077 if (unlikely(IS_DEADDIR(dir_inode)))
3078 return ERR_PTR(-ENOENT);
3080 file->f_mode &= ~FMODE_CREATED;
3081 dentry = d_lookup(dir, &nd->last);
3084 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3088 if (d_in_lookup(dentry))
3091 error = d_revalidate(dentry, nd->flags);
3092 if (likely(error > 0))
3096 d_invalidate(dentry);
3100 if (dentry->d_inode) {
3101 /* Cached positive dentry: will open in f_op->open */
3106 * Checking write permission is tricky, bacuse we don't know if we are
3107 * going to actually need it: O_CREAT opens should work as long as the
3108 * file exists. But checking existence breaks atomicity. The trick is
3109 * to check access and if not granted clear O_CREAT from the flags.
3111 * Another problem is returing the "right" error value (e.g. for an
3112 * O_EXCL open we want to return EEXIST not EROFS).
3114 if (open_flag & O_CREAT) {
3115 if (!IS_POSIXACL(dir->d_inode))
3116 mode &= ~current_umask();
3117 if (unlikely(!got_write)) {
3118 create_error = -EROFS;
3119 open_flag &= ~O_CREAT;
3120 if (open_flag & (O_EXCL | O_TRUNC))
3122 /* No side effects, safe to clear O_CREAT */
3124 create_error = may_o_create(&nd->path, dentry, mode);
3126 open_flag &= ~O_CREAT;
3127 if (open_flag & O_EXCL)
3131 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3132 unlikely(!got_write)) {
3134 * No O_CREATE -> atomicity not a requirement -> fall
3135 * back to lookup + open
3140 if (dir_inode->i_op->atomic_open) {
3141 dentry = atomic_open(nd, dentry, file, op, open_flag, mode);
3142 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3143 dentry = ERR_PTR(create_error);
3148 if (d_in_lookup(dentry)) {
3149 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3151 d_lookup_done(dentry);
3152 if (unlikely(res)) {
3154 error = PTR_ERR(res);
3162 /* Negative dentry, just create the file */
3163 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3164 file->f_mode |= FMODE_CREATED;
3165 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3166 if (!dir_inode->i_op->create) {
3170 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3171 open_flag & O_EXCL);
3174 fsnotify_create(dir_inode, dentry);
3176 if (unlikely(create_error) && !dentry->d_inode) {
3177 error = create_error;
3184 return ERR_PTR(error);
3188 * Handle the last step of open()
3190 static const char *do_last(struct nameidata *nd,
3191 struct file *file, const struct open_flags *op)
3193 struct dentry *dir = nd->path.dentry;
3194 kuid_t dir_uid = nd->inode->i_uid;
3195 umode_t dir_mode = nd->inode->i_mode;
3196 int open_flag = op->open_flag;
3197 bool will_truncate = (open_flag & O_TRUNC) != 0;
3198 bool got_write = false;
3199 int acc_mode = op->acc_mode;
3201 struct inode *inode;
3202 struct dentry *dentry;
3205 nd->flags &= ~LOOKUP_PARENT;
3206 nd->flags |= op->intent;
3208 if (nd->last_type != LAST_NORM) {
3211 error = handle_dots(nd, nd->last_type);
3212 if (unlikely(error))
3213 return ERR_PTR(error);
3217 if (!(open_flag & O_CREAT)) {
3218 if (nd->last.name[nd->last.len])
3219 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3220 /* we _can_ be in RCU mode here */
3221 dentry = lookup_fast(nd, &inode, &seq);
3223 return ERR_CAST(dentry);
3227 BUG_ON(nd->inode != dir->d_inode);
3228 BUG_ON(nd->flags & LOOKUP_RCU);
3230 /* create side of things */
3232 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3233 * has been cleared when we got to the last component we are
3236 error = complete_walk(nd);
3238 return ERR_PTR(error);
3240 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3241 /* trailing slashes? */
3242 if (unlikely(nd->last.name[nd->last.len]))
3243 return ERR_PTR(-EISDIR);
3246 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3247 error = mnt_want_write(nd->path.mnt);
3251 * do _not_ fail yet - we might not need that or fail with
3252 * a different error; let lookup_open() decide; we'll be
3253 * dropping this one anyway.
3256 if (open_flag & O_CREAT)
3257 inode_lock(dir->d_inode);
3259 inode_lock_shared(dir->d_inode);
3260 dentry = lookup_open(nd, file, op, got_write);
3261 if (open_flag & O_CREAT)
3262 inode_unlock(dir->d_inode);
3264 inode_unlock_shared(dir->d_inode);
3266 if (IS_ERR(dentry)) {
3267 error = PTR_ERR(dentry);
3271 if (file->f_mode & FMODE_OPENED) {
3272 if ((file->f_mode & FMODE_CREATED) ||
3273 !S_ISREG(file_inode(file)->i_mode))
3274 will_truncate = false;
3276 audit_inode(nd->name, file->f_path.dentry, 0);
3281 if (file->f_mode & FMODE_CREATED) {
3282 /* Don't check for write permission, don't truncate */
3283 open_flag &= ~O_TRUNC;
3284 will_truncate = false;
3286 dput(nd->path.dentry);
3287 nd->path.dentry = dentry;
3288 goto finish_open_created;
3292 * If atomic_open() acquired write access it is dropped now due to
3293 * possible mount and symlink following (this might be optimized away if
3297 mnt_drop_write(nd->path.mnt);
3304 error = step_into(nd, 0, dentry, inode, seq);
3305 if (unlikely(error)) {
3308 return ERR_PTR(error);
3311 nd->flags |= LOOKUP_PARENT;
3312 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3313 nd->stack[0].name = NULL;
3318 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3319 audit_inode(nd->name, nd->path.dentry, 0);
3320 return ERR_PTR(-EEXIST);
3323 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3324 error = complete_walk(nd);
3326 return ERR_PTR(error);
3327 audit_inode(nd->name, nd->path.dentry, 0);
3328 if (open_flag & O_CREAT) {
3330 if (d_is_dir(nd->path.dentry))
3332 error = may_create_in_sticky(dir_mode, dir_uid,
3333 d_backing_inode(nd->path.dentry));
3334 if (unlikely(error))
3338 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3340 if (!d_is_reg(nd->path.dentry))
3341 will_truncate = false;
3343 if (will_truncate) {
3344 error = mnt_want_write(nd->path.mnt);
3349 finish_open_created:
3350 error = may_open(&nd->path, acc_mode, open_flag);
3353 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
3354 error = vfs_open(&nd->path, file);
3358 error = ima_file_check(file, op->acc_mode);
3359 if (!error && will_truncate)
3360 error = handle_truncate(file);
3362 if (unlikely(error > 0)) {
3367 mnt_drop_write(nd->path.mnt);
3368 return ERR_PTR(error);
3371 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3373 struct dentry *child = NULL;
3374 struct inode *dir = dentry->d_inode;
3375 struct inode *inode;
3378 /* we want directory to be writable */
3379 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3382 error = -EOPNOTSUPP;
3383 if (!dir->i_op->tmpfile)
3386 child = d_alloc(dentry, &slash_name);
3387 if (unlikely(!child))
3389 error = dir->i_op->tmpfile(dir, child, mode);
3393 inode = child->d_inode;
3394 if (unlikely(!inode))
3396 if (!(open_flag & O_EXCL)) {
3397 spin_lock(&inode->i_lock);
3398 inode->i_state |= I_LINKABLE;
3399 spin_unlock(&inode->i_lock);
3401 ima_post_create_tmpfile(inode);
3406 return ERR_PTR(error);
3408 EXPORT_SYMBOL(vfs_tmpfile);
3410 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3411 const struct open_flags *op,
3414 struct dentry *child;
3416 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3417 if (unlikely(error))
3419 error = mnt_want_write(path.mnt);
3420 if (unlikely(error))
3422 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3423 error = PTR_ERR(child);
3427 path.dentry = child;
3428 audit_inode(nd->name, child, 0);
3429 /* Don't check for other permissions, the inode was just created */
3430 error = may_open(&path, 0, op->open_flag);
3433 file->f_path.mnt = path.mnt;
3434 error = finish_open(file, child, NULL);
3436 mnt_drop_write(path.mnt);
3442 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3445 int error = path_lookupat(nd, flags, &path);
3447 audit_inode(nd->name, path.dentry, 0);
3448 error = vfs_open(&path, file);
3454 static struct file *path_openat(struct nameidata *nd,
3455 const struct open_flags *op, unsigned flags)
3460 file = alloc_empty_file(op->open_flag, current_cred());
3464 if (unlikely(file->f_flags & __O_TMPFILE)) {
3465 error = do_tmpfile(nd, flags, op, file);
3466 } else if (unlikely(file->f_flags & O_PATH)) {
3467 error = do_o_path(nd, flags, file);
3469 const char *s = path_init(nd, flags);
3470 while (!(error = link_path_walk(s, nd)) &&
3471 (s = do_last(nd, file, op)) != NULL)
3475 if (likely(!error)) {
3476 if (likely(file->f_mode & FMODE_OPENED))
3482 if (error == -EOPENSTALE) {
3483 if (flags & LOOKUP_RCU)
3488 return ERR_PTR(error);
3491 struct file *do_filp_open(int dfd, struct filename *pathname,
3492 const struct open_flags *op)
3494 struct nameidata nd;
3495 int flags = op->lookup_flags;
3498 set_nameidata(&nd, dfd, pathname);
3499 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3500 if (unlikely(filp == ERR_PTR(-ECHILD)))
3501 filp = path_openat(&nd, op, flags);
3502 if (unlikely(filp == ERR_PTR(-ESTALE)))
3503 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3504 restore_nameidata();
3508 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3509 const char *name, const struct open_flags *op)
3511 struct nameidata nd;
3513 struct filename *filename;
3514 int flags = op->lookup_flags | LOOKUP_ROOT;
3517 nd.root.dentry = dentry;
3519 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3520 return ERR_PTR(-ELOOP);
3522 filename = getname_kernel(name);
3523 if (IS_ERR(filename))
3524 return ERR_CAST(filename);
3526 set_nameidata(&nd, -1, filename);
3527 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3528 if (unlikely(file == ERR_PTR(-ECHILD)))
3529 file = path_openat(&nd, op, flags);
3530 if (unlikely(file == ERR_PTR(-ESTALE)))
3531 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3532 restore_nameidata();
3537 static struct dentry *filename_create(int dfd, struct filename *name,
3538 struct path *path, unsigned int lookup_flags)
3540 struct dentry *dentry = ERR_PTR(-EEXIST);
3545 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3548 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3549 * other flags passed in are ignored!
3551 lookup_flags &= LOOKUP_REVAL;
3553 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3555 return ERR_CAST(name);
3558 * Yucky last component or no last component at all?
3559 * (foo/., foo/.., /////)
3561 if (unlikely(type != LAST_NORM))
3564 /* don't fail immediately if it's r/o, at least try to report other errors */
3565 err2 = mnt_want_write(path->mnt);
3567 * Do the final lookup.
3569 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3570 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3571 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3576 if (d_is_positive(dentry))
3580 * Special case - lookup gave negative, but... we had foo/bar/
3581 * From the vfs_mknod() POV we just have a negative dentry -
3582 * all is fine. Let's be bastards - you had / on the end, you've
3583 * been asking for (non-existent) directory. -ENOENT for you.
3585 if (unlikely(!is_dir && last.name[last.len])) {
3589 if (unlikely(err2)) {
3597 dentry = ERR_PTR(error);
3599 inode_unlock(path->dentry->d_inode);
3601 mnt_drop_write(path->mnt);
3608 struct dentry *kern_path_create(int dfd, const char *pathname,
3609 struct path *path, unsigned int lookup_flags)
3611 return filename_create(dfd, getname_kernel(pathname),
3612 path, lookup_flags);
3614 EXPORT_SYMBOL(kern_path_create);
3616 void done_path_create(struct path *path, struct dentry *dentry)
3619 inode_unlock(path->dentry->d_inode);
3620 mnt_drop_write(path->mnt);
3623 EXPORT_SYMBOL(done_path_create);
3625 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3626 struct path *path, unsigned int lookup_flags)
3628 return filename_create(dfd, getname(pathname), path, lookup_flags);
3630 EXPORT_SYMBOL(user_path_create);
3632 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3634 int error = may_create(dir, dentry);
3639 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3642 if (!dir->i_op->mknod)
3645 error = devcgroup_inode_mknod(mode, dev);
3649 error = security_inode_mknod(dir, dentry, mode, dev);
3653 error = dir->i_op->mknod(dir, dentry, mode, dev);
3655 fsnotify_create(dir, dentry);
3658 EXPORT_SYMBOL(vfs_mknod);
3660 static int may_mknod(umode_t mode)
3662 switch (mode & S_IFMT) {
3668 case 0: /* zero mode translates to S_IFREG */
3677 long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3680 struct dentry *dentry;
3683 unsigned int lookup_flags = 0;
3685 error = may_mknod(mode);
3689 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3691 return PTR_ERR(dentry);
3693 if (!IS_POSIXACL(path.dentry->d_inode))
3694 mode &= ~current_umask();
3695 error = security_path_mknod(&path, dentry, mode, dev);
3698 switch (mode & S_IFMT) {
3699 case 0: case S_IFREG:
3700 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3702 ima_post_path_mknod(dentry);
3704 case S_IFCHR: case S_IFBLK:
3705 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3706 new_decode_dev(dev));
3708 case S_IFIFO: case S_IFSOCK:
3709 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3713 done_path_create(&path, dentry);
3714 if (retry_estale(error, lookup_flags)) {
3715 lookup_flags |= LOOKUP_REVAL;
3721 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3724 return do_mknodat(dfd, filename, mode, dev);
3727 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3729 return do_mknodat(AT_FDCWD, filename, mode, dev);
3732 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3734 int error = may_create(dir, dentry);
3735 unsigned max_links = dir->i_sb->s_max_links;
3740 if (!dir->i_op->mkdir)
3743 mode &= (S_IRWXUGO|S_ISVTX);
3744 error = security_inode_mkdir(dir, dentry, mode);
3748 if (max_links && dir->i_nlink >= max_links)
3751 error = dir->i_op->mkdir(dir, dentry, mode);
3753 fsnotify_mkdir(dir, dentry);
3756 EXPORT_SYMBOL(vfs_mkdir);
3758 long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3760 struct dentry *dentry;
3763 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3766 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3768 return PTR_ERR(dentry);
3770 if (!IS_POSIXACL(path.dentry->d_inode))
3771 mode &= ~current_umask();
3772 error = security_path_mkdir(&path, dentry, mode);
3774 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3775 done_path_create(&path, dentry);
3776 if (retry_estale(error, lookup_flags)) {
3777 lookup_flags |= LOOKUP_REVAL;
3783 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3785 return do_mkdirat(dfd, pathname, mode);
3788 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3790 return do_mkdirat(AT_FDCWD, pathname, mode);
3793 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3795 int error = may_delete(dir, dentry, 1);
3800 if (!dir->i_op->rmdir)
3804 inode_lock(dentry->d_inode);
3807 if (is_local_mountpoint(dentry))
3810 error = security_inode_rmdir(dir, dentry);
3814 error = dir->i_op->rmdir(dir, dentry);
3818 shrink_dcache_parent(dentry);
3819 dentry->d_inode->i_flags |= S_DEAD;
3821 detach_mounts(dentry);
3822 fsnotify_rmdir(dir, dentry);
3825 inode_unlock(dentry->d_inode);
3831 EXPORT_SYMBOL(vfs_rmdir);
3833 long do_rmdir(int dfd, const char __user *pathname)
3836 struct filename *name;
3837 struct dentry *dentry;
3841 unsigned int lookup_flags = 0;
3843 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3844 &path, &last, &type);
3846 return PTR_ERR(name);
3860 error = mnt_want_write(path.mnt);
3864 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3865 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3866 error = PTR_ERR(dentry);
3869 if (!dentry->d_inode) {
3873 error = security_path_rmdir(&path, dentry);
3876 error = vfs_rmdir(path.dentry->d_inode, dentry);
3880 inode_unlock(path.dentry->d_inode);
3881 mnt_drop_write(path.mnt);
3885 if (retry_estale(error, lookup_flags)) {
3886 lookup_flags |= LOOKUP_REVAL;
3892 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3894 return do_rmdir(AT_FDCWD, pathname);
3898 * vfs_unlink - unlink a filesystem object
3899 * @dir: parent directory
3901 * @delegated_inode: returns victim inode, if the inode is delegated.
3903 * The caller must hold dir->i_mutex.
3905 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3906 * return a reference to the inode in delegated_inode. The caller
3907 * should then break the delegation on that inode and retry. Because
3908 * breaking a delegation may take a long time, the caller should drop
3909 * dir->i_mutex before doing so.
3911 * Alternatively, a caller may pass NULL for delegated_inode. This may
3912 * be appropriate for callers that expect the underlying filesystem not
3913 * to be NFS exported.
3915 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3917 struct inode *target = dentry->d_inode;
3918 int error = may_delete(dir, dentry, 0);
3923 if (!dir->i_op->unlink)
3927 if (is_local_mountpoint(dentry))
3930 error = security_inode_unlink(dir, dentry);
3932 error = try_break_deleg(target, delegated_inode);
3935 error = dir->i_op->unlink(dir, dentry);
3938 detach_mounts(dentry);
3939 fsnotify_unlink(dir, dentry);
3944 inode_unlock(target);
3946 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3947 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3948 fsnotify_link_count(target);
3954 EXPORT_SYMBOL(vfs_unlink);
3957 * Make sure that the actual truncation of the file will occur outside its
3958 * directory's i_mutex. Truncate can take a long time if there is a lot of
3959 * writeout happening, and we don't want to prevent access to the directory
3960 * while waiting on the I/O.
3962 long do_unlinkat(int dfd, struct filename *name)
3965 struct dentry *dentry;
3969 struct inode *inode = NULL;
3970 struct inode *delegated_inode = NULL;
3971 unsigned int lookup_flags = 0;
3973 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
3975 return PTR_ERR(name);
3978 if (type != LAST_NORM)
3981 error = mnt_want_write(path.mnt);
3985 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3986 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3987 error = PTR_ERR(dentry);
3988 if (!IS_ERR(dentry)) {
3989 /* Why not before? Because we want correct error value */
3990 if (last.name[last.len])
3992 inode = dentry->d_inode;
3993 if (d_is_negative(dentry))
3996 error = security_path_unlink(&path, dentry);
3999 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4003 inode_unlock(path.dentry->d_inode);
4005 iput(inode); /* truncate the inode here */
4007 if (delegated_inode) {
4008 error = break_deleg_wait(&delegated_inode);
4012 mnt_drop_write(path.mnt);
4015 if (retry_estale(error, lookup_flags)) {
4016 lookup_flags |= LOOKUP_REVAL;
4024 if (d_is_negative(dentry))
4026 else if (d_is_dir(dentry))
4033 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4035 if ((flag & ~AT_REMOVEDIR) != 0)
4038 if (flag & AT_REMOVEDIR)
4039 return do_rmdir(dfd, pathname);
4041 return do_unlinkat(dfd, getname(pathname));
4044 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4046 return do_unlinkat(AT_FDCWD, getname(pathname));
4049 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4051 int error = may_create(dir, dentry);
4056 if (!dir->i_op->symlink)
4059 error = security_inode_symlink(dir, dentry, oldname);
4063 error = dir->i_op->symlink(dir, dentry, oldname);
4065 fsnotify_create(dir, dentry);
4068 EXPORT_SYMBOL(vfs_symlink);
4070 long do_symlinkat(const char __user *oldname, int newdfd,
4071 const char __user *newname)
4074 struct filename *from;
4075 struct dentry *dentry;
4077 unsigned int lookup_flags = 0;
4079 from = getname(oldname);
4081 return PTR_ERR(from);
4083 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4084 error = PTR_ERR(dentry);
4088 error = security_path_symlink(&path, dentry, from->name);
4090 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4091 done_path_create(&path, dentry);
4092 if (retry_estale(error, lookup_flags)) {
4093 lookup_flags |= LOOKUP_REVAL;
4101 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4102 int, newdfd, const char __user *, newname)
4104 return do_symlinkat(oldname, newdfd, newname);
4107 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4109 return do_symlinkat(oldname, AT_FDCWD, newname);
4113 * vfs_link - create a new link
4114 * @old_dentry: object to be linked
4116 * @new_dentry: where to create the new link
4117 * @delegated_inode: returns inode needing a delegation break
4119 * The caller must hold dir->i_mutex
4121 * If vfs_link discovers a delegation on the to-be-linked file in need
4122 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4123 * inode in delegated_inode. The caller should then break the delegation
4124 * and retry. Because breaking a delegation may take a long time, the
4125 * caller should drop the i_mutex before doing so.
4127 * Alternatively, a caller may pass NULL for delegated_inode. This may
4128 * be appropriate for callers that expect the underlying filesystem not
4129 * to be NFS exported.
4131 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4133 struct inode *inode = old_dentry->d_inode;
4134 unsigned max_links = dir->i_sb->s_max_links;
4140 error = may_create(dir, new_dentry);
4144 if (dir->i_sb != inode->i_sb)
4148 * A link to an append-only or immutable file cannot be created.
4150 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4153 * Updating the link count will likely cause i_uid and i_gid to
4154 * be writen back improperly if their true value is unknown to
4157 if (HAS_UNMAPPED_ID(inode))
4159 if (!dir->i_op->link)
4161 if (S_ISDIR(inode->i_mode))
4164 error = security_inode_link(old_dentry, dir, new_dentry);
4169 /* Make sure we don't allow creating hardlink to an unlinked file */
4170 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4172 else if (max_links && inode->i_nlink >= max_links)
4175 error = try_break_deleg(inode, delegated_inode);
4177 error = dir->i_op->link(old_dentry, dir, new_dentry);
4180 if (!error && (inode->i_state & I_LINKABLE)) {
4181 spin_lock(&inode->i_lock);
4182 inode->i_state &= ~I_LINKABLE;
4183 spin_unlock(&inode->i_lock);
4185 inode_unlock(inode);
4187 fsnotify_link(dir, inode, new_dentry);
4190 EXPORT_SYMBOL(vfs_link);
4193 * Hardlinks are often used in delicate situations. We avoid
4194 * security-related surprises by not following symlinks on the
4197 * We don't follow them on the oldname either to be compatible
4198 * with linux 2.0, and to avoid hard-linking to directories
4199 * and other special files. --ADM
4201 int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4202 const char __user *newname, int flags)
4204 struct dentry *new_dentry;
4205 struct path old_path, new_path;
4206 struct inode *delegated_inode = NULL;
4210 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4213 * To use null names we require CAP_DAC_READ_SEARCH
4214 * This ensures that not everyone will be able to create
4215 * handlink using the passed filedescriptor.
4217 if (flags & AT_EMPTY_PATH) {
4218 if (!capable(CAP_DAC_READ_SEARCH))
4223 if (flags & AT_SYMLINK_FOLLOW)
4224 how |= LOOKUP_FOLLOW;
4226 error = user_path_at(olddfd, oldname, how, &old_path);
4230 new_dentry = user_path_create(newdfd, newname, &new_path,
4231 (how & LOOKUP_REVAL));
4232 error = PTR_ERR(new_dentry);
4233 if (IS_ERR(new_dentry))
4237 if (old_path.mnt != new_path.mnt)
4239 error = may_linkat(&old_path);
4240 if (unlikely(error))
4242 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4245 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4247 done_path_create(&new_path, new_dentry);
4248 if (delegated_inode) {
4249 error = break_deleg_wait(&delegated_inode);
4251 path_put(&old_path);
4255 if (retry_estale(error, how)) {
4256 path_put(&old_path);
4257 how |= LOOKUP_REVAL;
4261 path_put(&old_path);
4266 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4267 int, newdfd, const char __user *, newname, int, flags)
4269 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4272 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4274 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4278 * vfs_rename - rename a filesystem object
4279 * @old_dir: parent of source
4280 * @old_dentry: source
4281 * @new_dir: parent of destination
4282 * @new_dentry: destination
4283 * @delegated_inode: returns an inode needing a delegation break
4284 * @flags: rename flags
4286 * The caller must hold multiple mutexes--see lock_rename()).
4288 * If vfs_rename discovers a delegation in need of breaking at either
4289 * the source or destination, it will return -EWOULDBLOCK and return a
4290 * reference to the inode in delegated_inode. The caller should then
4291 * break the delegation and retry. Because breaking a delegation may
4292 * take a long time, the caller should drop all locks before doing
4295 * Alternatively, a caller may pass NULL for delegated_inode. This may
4296 * be appropriate for callers that expect the underlying filesystem not
4297 * to be NFS exported.
4299 * The worst of all namespace operations - renaming directory. "Perverted"
4300 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4303 * a) we can get into loop creation.
4304 * b) race potential - two innocent renames can create a loop together.
4305 * That's where 4.4 screws up. Current fix: serialization on
4306 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4308 * c) we have to lock _four_ objects - parents and victim (if it exists),
4309 * and source (if it is not a directory).
4310 * And that - after we got ->i_mutex on parents (until then we don't know
4311 * whether the target exists). Solution: try to be smart with locking
4312 * order for inodes. We rely on the fact that tree topology may change
4313 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4314 * move will be locked. Thus we can rank directories by the tree
4315 * (ancestors first) and rank all non-directories after them.
4316 * That works since everybody except rename does "lock parent, lookup,
4317 * lock child" and rename is under ->s_vfs_rename_mutex.
4318 * HOWEVER, it relies on the assumption that any object with ->lookup()
4319 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4320 * we'd better make sure that there's no link(2) for them.
4321 * d) conversion from fhandle to dentry may come in the wrong moment - when
4322 * we are removing the target. Solution: we will have to grab ->i_mutex
4323 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4324 * ->i_mutex on parents, which works but leads to some truly excessive
4327 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4328 struct inode *new_dir, struct dentry *new_dentry,
4329 struct inode **delegated_inode, unsigned int flags)
4332 bool is_dir = d_is_dir(old_dentry);
4333 struct inode *source = old_dentry->d_inode;
4334 struct inode *target = new_dentry->d_inode;
4335 bool new_is_dir = false;
4336 unsigned max_links = new_dir->i_sb->s_max_links;
4337 struct name_snapshot old_name;
4339 if (source == target)
4342 error = may_delete(old_dir, old_dentry, is_dir);
4347 error = may_create(new_dir, new_dentry);
4349 new_is_dir = d_is_dir(new_dentry);
4351 if (!(flags & RENAME_EXCHANGE))
4352 error = may_delete(new_dir, new_dentry, is_dir);
4354 error = may_delete(new_dir, new_dentry, new_is_dir);
4359 if (!old_dir->i_op->rename)
4363 * If we are going to change the parent - check write permissions,
4364 * we'll need to flip '..'.
4366 if (new_dir != old_dir) {
4368 error = inode_permission(source, MAY_WRITE);
4372 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4373 error = inode_permission(target, MAY_WRITE);
4379 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4384 take_dentry_name_snapshot(&old_name, old_dentry);
4386 if (!is_dir || (flags & RENAME_EXCHANGE))
4387 lock_two_nondirectories(source, target);
4392 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4395 if (max_links && new_dir != old_dir) {
4397 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4399 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4400 old_dir->i_nlink >= max_links)
4404 error = try_break_deleg(source, delegated_inode);
4408 if (target && !new_is_dir) {
4409 error = try_break_deleg(target, delegated_inode);
4413 error = old_dir->i_op->rename(old_dir, old_dentry,
4414 new_dir, new_dentry, flags);
4418 if (!(flags & RENAME_EXCHANGE) && target) {
4420 shrink_dcache_parent(new_dentry);
4421 target->i_flags |= S_DEAD;
4423 dont_mount(new_dentry);
4424 detach_mounts(new_dentry);
4426 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4427 if (!(flags & RENAME_EXCHANGE))
4428 d_move(old_dentry, new_dentry);
4430 d_exchange(old_dentry, new_dentry);
4433 if (!is_dir || (flags & RENAME_EXCHANGE))
4434 unlock_two_nondirectories(source, target);
4436 inode_unlock(target);
4439 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4440 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4441 if (flags & RENAME_EXCHANGE) {
4442 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4443 new_is_dir, NULL, new_dentry);
4446 release_dentry_name_snapshot(&old_name);
4450 EXPORT_SYMBOL(vfs_rename);
4452 static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4453 const char __user *newname, unsigned int flags)
4455 struct dentry *old_dentry, *new_dentry;
4456 struct dentry *trap;
4457 struct path old_path, new_path;
4458 struct qstr old_last, new_last;
4459 int old_type, new_type;
4460 struct inode *delegated_inode = NULL;
4461 struct filename *from;
4462 struct filename *to;
4463 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4464 bool should_retry = false;
4467 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4470 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4471 (flags & RENAME_EXCHANGE))
4474 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4477 if (flags & RENAME_EXCHANGE)
4481 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4482 &old_path, &old_last, &old_type);
4484 error = PTR_ERR(from);
4488 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4489 &new_path, &new_last, &new_type);
4491 error = PTR_ERR(to);
4496 if (old_path.mnt != new_path.mnt)
4500 if (old_type != LAST_NORM)
4503 if (flags & RENAME_NOREPLACE)
4505 if (new_type != LAST_NORM)
4508 error = mnt_want_write(old_path.mnt);
4513 trap = lock_rename(new_path.dentry, old_path.dentry);
4515 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4516 error = PTR_ERR(old_dentry);
4517 if (IS_ERR(old_dentry))
4519 /* source must exist */
4521 if (d_is_negative(old_dentry))
4523 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4524 error = PTR_ERR(new_dentry);
4525 if (IS_ERR(new_dentry))
4528 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4530 if (flags & RENAME_EXCHANGE) {
4532 if (d_is_negative(new_dentry))
4535 if (!d_is_dir(new_dentry)) {
4537 if (new_last.name[new_last.len])
4541 /* unless the source is a directory trailing slashes give -ENOTDIR */
4542 if (!d_is_dir(old_dentry)) {
4544 if (old_last.name[old_last.len])
4546 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4549 /* source should not be ancestor of target */
4551 if (old_dentry == trap)
4553 /* target should not be an ancestor of source */
4554 if (!(flags & RENAME_EXCHANGE))
4556 if (new_dentry == trap)
4559 error = security_path_rename(&old_path, old_dentry,
4560 &new_path, new_dentry, flags);
4563 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4564 new_path.dentry->d_inode, new_dentry,
4565 &delegated_inode, flags);
4571 unlock_rename(new_path.dentry, old_path.dentry);
4572 if (delegated_inode) {
4573 error = break_deleg_wait(&delegated_inode);
4577 mnt_drop_write(old_path.mnt);
4579 if (retry_estale(error, lookup_flags))
4580 should_retry = true;
4581 path_put(&new_path);
4584 path_put(&old_path);
4587 should_retry = false;
4588 lookup_flags |= LOOKUP_REVAL;
4595 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4596 int, newdfd, const char __user *, newname, unsigned int, flags)
4598 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4601 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4602 int, newdfd, const char __user *, newname)
4604 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
4607 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4609 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4612 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4614 int error = may_create(dir, dentry);
4618 if (!dir->i_op->mknod)
4621 return dir->i_op->mknod(dir, dentry,
4622 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4624 EXPORT_SYMBOL(vfs_whiteout);
4626 int readlink_copy(char __user *buffer, int buflen, const char *link)
4628 int len = PTR_ERR(link);
4633 if (len > (unsigned) buflen)
4635 if (copy_to_user(buffer, link, len))
4642 * vfs_readlink - copy symlink body into userspace buffer
4643 * @dentry: dentry on which to get symbolic link
4644 * @buffer: user memory pointer
4645 * @buflen: size of buffer
4647 * Does not touch atime. That's up to the caller if necessary
4649 * Does not call security hook.
4651 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4653 struct inode *inode = d_inode(dentry);
4654 DEFINE_DELAYED_CALL(done);
4658 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4659 if (unlikely(inode->i_op->readlink))
4660 return inode->i_op->readlink(dentry, buffer, buflen);
4662 if (!d_is_symlink(dentry))
4665 spin_lock(&inode->i_lock);
4666 inode->i_opflags |= IOP_DEFAULT_READLINK;
4667 spin_unlock(&inode->i_lock);
4670 link = READ_ONCE(inode->i_link);
4672 link = inode->i_op->get_link(dentry, inode, &done);
4674 return PTR_ERR(link);
4676 res = readlink_copy(buffer, buflen, link);
4677 do_delayed_call(&done);
4680 EXPORT_SYMBOL(vfs_readlink);
4683 * vfs_get_link - get symlink body
4684 * @dentry: dentry on which to get symbolic link
4685 * @done: caller needs to free returned data with this
4687 * Calls security hook and i_op->get_link() on the supplied inode.
4689 * It does not touch atime. That's up to the caller if necessary.
4691 * Does not work on "special" symlinks like /proc/$$/fd/N
4693 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4695 const char *res = ERR_PTR(-EINVAL);
4696 struct inode *inode = d_inode(dentry);
4698 if (d_is_symlink(dentry)) {
4699 res = ERR_PTR(security_inode_readlink(dentry));
4701 res = inode->i_op->get_link(dentry, inode, done);
4705 EXPORT_SYMBOL(vfs_get_link);
4707 /* get the link contents into pagecache */
4708 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4709 struct delayed_call *callback)
4713 struct address_space *mapping = inode->i_mapping;
4716 page = find_get_page(mapping, 0);
4718 return ERR_PTR(-ECHILD);
4719 if (!PageUptodate(page)) {
4721 return ERR_PTR(-ECHILD);
4724 page = read_mapping_page(mapping, 0, NULL);
4728 set_delayed_call(callback, page_put_link, page);
4729 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4730 kaddr = page_address(page);
4731 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4735 EXPORT_SYMBOL(page_get_link);
4737 void page_put_link(void *arg)
4741 EXPORT_SYMBOL(page_put_link);
4743 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4745 DEFINE_DELAYED_CALL(done);
4746 int res = readlink_copy(buffer, buflen,
4747 page_get_link(dentry, d_inode(dentry),
4749 do_delayed_call(&done);
4752 EXPORT_SYMBOL(page_readlink);
4755 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4757 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4759 struct address_space *mapping = inode->i_mapping;
4763 unsigned int flags = 0;
4765 flags |= AOP_FLAG_NOFS;
4768 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4769 flags, &page, &fsdata);
4773 memcpy(page_address(page), symname, len-1);
4775 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4782 mark_inode_dirty(inode);
4787 EXPORT_SYMBOL(__page_symlink);
4789 int page_symlink(struct inode *inode, const char *symname, int len)
4791 return __page_symlink(inode, symname, len,
4792 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4794 EXPORT_SYMBOL(page_symlink);
4796 const struct inode_operations page_symlink_inode_operations = {
4797 .get_link = page_get_link,
4799 EXPORT_SYMBOL(page_symlink_inode_operations);