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);
1163 error = nd_jump_root(nd);
1164 if (unlikely(error))
1165 return ERR_PTR(error);
1166 while (unlikely(*++res == '/'))
1171 all_done: // pure jump
1177 * follow_up - Find the mountpoint of path's vfsmount
1179 * Given a path, find the mountpoint of its source file system.
1180 * Replace @path with the path of the mountpoint in the parent mount.
1183 * Return 1 if we went up a level and 0 if we were already at the
1186 int follow_up(struct path *path)
1188 struct mount *mnt = real_mount(path->mnt);
1189 struct mount *parent;
1190 struct dentry *mountpoint;
1192 read_seqlock_excl(&mount_lock);
1193 parent = mnt->mnt_parent;
1194 if (parent == mnt) {
1195 read_sequnlock_excl(&mount_lock);
1198 mntget(&parent->mnt);
1199 mountpoint = dget(mnt->mnt_mountpoint);
1200 read_sequnlock_excl(&mount_lock);
1202 path->dentry = mountpoint;
1204 path->mnt = &parent->mnt;
1207 EXPORT_SYMBOL(follow_up);
1210 * Perform an automount
1211 * - return -EISDIR to tell follow_managed() to stop and return the path we
1214 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1216 struct dentry *dentry = path->dentry;
1218 /* We don't want to mount if someone's just doing a stat -
1219 * unless they're stat'ing a directory and appended a '/' to
1222 * We do, however, want to mount if someone wants to open or
1223 * create a file of any type under the mountpoint, wants to
1224 * traverse through the mountpoint or wants to open the
1225 * mounted directory. Also, autofs may mark negative dentries
1226 * as being automount points. These will need the attentions
1227 * of the daemon to instantiate them before they can be used.
1229 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1230 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1234 if (count && (*count)++ >= MAXSYMLINKS)
1237 return finish_automount(dentry->d_op->d_automount(path), path);
1241 * Handle a dentry that is managed in some way.
1242 * - Flagged for transit management (autofs)
1243 * - Flagged as mountpoint
1244 * - Flagged as automount point
1246 * This may only be called in refwalk mode.
1247 * On success path->dentry is known positive.
1249 * Serialization is taken care of in namespace.c
1251 static int follow_managed(struct path *path, struct nameidata *nd)
1253 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1255 bool need_mntput = false;
1258 /* Given that we're not holding a lock here, we retain the value in a
1259 * local variable for each dentry as we look at it so that we don't see
1260 * the components of that value change under us */
1261 while (flags = smp_load_acquire(&path->dentry->d_flags),
1262 unlikely(flags & DCACHE_MANAGED_DENTRY)) {
1263 /* Allow the filesystem to manage the transit without i_mutex
1265 if (flags & DCACHE_MANAGE_TRANSIT) {
1266 BUG_ON(!path->dentry->d_op);
1267 BUG_ON(!path->dentry->d_op->d_manage);
1268 ret = path->dentry->d_op->d_manage(path, false);
1269 flags = smp_load_acquire(&path->dentry->d_flags);
1274 /* Transit to a mounted filesystem. */
1275 if (flags & DCACHE_MOUNTED) {
1276 struct vfsmount *mounted = lookup_mnt(path);
1281 path->mnt = mounted;
1282 path->dentry = dget(mounted->mnt_root);
1287 /* Something is mounted on this dentry in another
1288 * namespace and/or whatever was mounted there in this
1289 * namespace got unmounted before lookup_mnt() could
1293 /* Handle an automount point */
1294 if (flags & DCACHE_NEED_AUTOMOUNT) {
1295 ret = follow_automount(path, &nd->total_link_count,
1302 /* We didn't change the current path point */
1307 if (path->mnt == mnt)
1309 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1312 nd->flags |= LOOKUP_JUMPED;
1314 if (ret == -EISDIR || !ret)
1316 if (ret > 0 && unlikely(d_flags_negative(flags)))
1318 if (unlikely(ret < 0)) {
1320 if (path->mnt != nd->path.mnt)
1326 int follow_down_one(struct path *path)
1328 struct vfsmount *mounted;
1330 mounted = lookup_mnt(path);
1334 path->mnt = mounted;
1335 path->dentry = dget(mounted->mnt_root);
1340 EXPORT_SYMBOL(follow_down_one);
1342 static inline int managed_dentry_rcu(const struct path *path)
1344 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1345 path->dentry->d_op->d_manage(path, true) : 0;
1349 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1350 * we meet a managed dentry that would need blocking.
1352 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1353 struct inode **inode, unsigned *seqp)
1356 struct mount *mounted;
1358 * Don't forget we might have a non-mountpoint managed dentry
1359 * that wants to block transit.
1361 switch (managed_dentry_rcu(path)) {
1371 if (!d_mountpoint(path->dentry))
1372 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1374 mounted = __lookup_mnt(path->mnt, path->dentry);
1377 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1379 path->mnt = &mounted->mnt;
1380 path->dentry = mounted->mnt.mnt_root;
1381 nd->flags |= LOOKUP_JUMPED;
1382 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1384 * Update the inode too. We don't need to re-check the
1385 * dentry sequence number here after this d_inode read,
1386 * because a mount-point is always pinned.
1388 *inode = path->dentry->d_inode;
1390 return !read_seqretry(&mount_lock, nd->m_seq) &&
1391 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1394 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1395 struct path *path, struct inode **inode,
1400 path->mnt = nd->path.mnt;
1401 path->dentry = dentry;
1402 if (nd->flags & LOOKUP_RCU) {
1403 unsigned int seq = *seqp;
1404 if (unlikely(!*inode))
1406 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1408 if (unlazy_child(nd, dentry, seq))
1410 // *path might've been clobbered by __follow_mount_rcu()
1411 path->mnt = nd->path.mnt;
1412 path->dentry = dentry;
1414 ret = follow_managed(path, nd);
1415 if (likely(ret >= 0)) {
1416 *inode = d_backing_inode(path->dentry);
1417 *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1422 static int follow_dotdot_rcu(struct nameidata *nd)
1424 struct inode *inode = nd->inode;
1427 if (path_equal(&nd->path, &nd->root)) {
1428 if (unlikely(nd->flags & LOOKUP_BENEATH))
1432 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1433 struct dentry *old = nd->path.dentry;
1434 struct dentry *parent = old->d_parent;
1437 inode = parent->d_inode;
1438 seq = read_seqcount_begin(&parent->d_seq);
1439 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1441 nd->path.dentry = parent;
1443 if (unlikely(!path_connected(&nd->path)))
1447 struct mount *mnt = real_mount(nd->path.mnt);
1448 struct mount *mparent = mnt->mnt_parent;
1449 struct dentry *mountpoint = mnt->mnt_mountpoint;
1450 struct inode *inode2 = mountpoint->d_inode;
1451 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1452 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1454 if (&mparent->mnt == nd->path.mnt)
1456 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1458 /* we know that mountpoint was pinned */
1459 nd->path.dentry = mountpoint;
1460 nd->path.mnt = &mparent->mnt;
1465 while (unlikely(d_mountpoint(nd->path.dentry))) {
1466 struct mount *mounted;
1467 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1468 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1472 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1474 nd->path.mnt = &mounted->mnt;
1475 nd->path.dentry = mounted->mnt.mnt_root;
1476 inode = nd->path.dentry->d_inode;
1477 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1484 * Follow down to the covering mount currently visible to userspace. At each
1485 * point, the filesystem owning that dentry may be queried as to whether the
1486 * caller is permitted to proceed or not.
1488 int follow_down(struct path *path)
1493 while (managed = READ_ONCE(path->dentry->d_flags),
1494 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1495 /* Allow the filesystem to manage the transit without i_mutex
1498 * We indicate to the filesystem if someone is trying to mount
1499 * something here. This gives autofs the chance to deny anyone
1500 * other than its daemon the right to mount on its
1503 * The filesystem may sleep at this point.
1505 if (managed & DCACHE_MANAGE_TRANSIT) {
1506 BUG_ON(!path->dentry->d_op);
1507 BUG_ON(!path->dentry->d_op->d_manage);
1508 ret = path->dentry->d_op->d_manage(path, false);
1510 return ret == -EISDIR ? 0 : ret;
1513 /* Transit to a mounted filesystem. */
1514 if (managed & DCACHE_MOUNTED) {
1515 struct vfsmount *mounted = lookup_mnt(path);
1520 path->mnt = mounted;
1521 path->dentry = dget(mounted->mnt_root);
1525 /* Don't handle automount points here */
1530 EXPORT_SYMBOL(follow_down);
1533 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1535 static void follow_mount(struct path *path)
1537 while (d_mountpoint(path->dentry)) {
1538 struct vfsmount *mounted = lookup_mnt(path);
1543 path->mnt = mounted;
1544 path->dentry = dget(mounted->mnt_root);
1548 static int path_parent_directory(struct path *path)
1550 struct dentry *old = path->dentry;
1551 /* rare case of legitimate dget_parent()... */
1552 path->dentry = dget_parent(path->dentry);
1554 if (unlikely(!path_connected(path)))
1559 static int follow_dotdot(struct nameidata *nd)
1562 if (path_equal(&nd->path, &nd->root)) {
1563 if (unlikely(nd->flags & LOOKUP_BENEATH))
1567 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1568 int ret = path_parent_directory(&nd->path);
1573 if (!follow_up(&nd->path))
1575 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1578 follow_mount(&nd->path);
1579 nd->inode = nd->path.dentry->d_inode;
1584 * This looks up the name in dcache and possibly revalidates the found dentry.
1585 * NULL is returned if the dentry does not exist in the cache.
1587 static struct dentry *lookup_dcache(const struct qstr *name,
1591 struct dentry *dentry = d_lookup(dir, name);
1593 int error = d_revalidate(dentry, flags);
1594 if (unlikely(error <= 0)) {
1596 d_invalidate(dentry);
1598 return ERR_PTR(error);
1605 * Parent directory has inode locked exclusive. This is one
1606 * and only case when ->lookup() gets called on non in-lookup
1607 * dentries - as the matter of fact, this only gets called
1608 * when directory is guaranteed to have no in-lookup children
1611 static struct dentry *__lookup_hash(const struct qstr *name,
1612 struct dentry *base, unsigned int flags)
1614 struct dentry *dentry = lookup_dcache(name, base, flags);
1616 struct inode *dir = base->d_inode;
1621 /* Don't create child dentry for a dead directory. */
1622 if (unlikely(IS_DEADDIR(dir)))
1623 return ERR_PTR(-ENOENT);
1625 dentry = d_alloc(base, name);
1626 if (unlikely(!dentry))
1627 return ERR_PTR(-ENOMEM);
1629 old = dir->i_op->lookup(dir, dentry, flags);
1630 if (unlikely(old)) {
1637 static struct dentry *lookup_fast(struct nameidata *nd,
1638 struct inode **inode,
1641 struct dentry *dentry, *parent = nd->path.dentry;
1645 * Rename seqlock is not required here because in the off chance
1646 * of a false negative due to a concurrent rename, the caller is
1647 * going to fall back to non-racy lookup.
1649 if (nd->flags & LOOKUP_RCU) {
1651 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1652 if (unlikely(!dentry)) {
1653 if (unlazy_walk(nd))
1654 return ERR_PTR(-ECHILD);
1659 * This sequence count validates that the inode matches
1660 * the dentry name information from lookup.
1662 *inode = d_backing_inode(dentry);
1663 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1664 return ERR_PTR(-ECHILD);
1667 * This sequence count validates that the parent had no
1668 * changes while we did the lookup of the dentry above.
1670 * The memory barrier in read_seqcount_begin of child is
1671 * enough, we can use __read_seqcount_retry here.
1673 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1674 return ERR_PTR(-ECHILD);
1677 status = d_revalidate(dentry, nd->flags);
1678 if (likely(status > 0))
1680 if (unlazy_child(nd, dentry, seq))
1681 return ERR_PTR(-ECHILD);
1682 if (unlikely(status == -ECHILD))
1683 /* we'd been told to redo it in non-rcu mode */
1684 status = d_revalidate(dentry, nd->flags);
1686 dentry = __d_lookup(parent, &nd->last);
1687 if (unlikely(!dentry))
1689 status = d_revalidate(dentry, nd->flags);
1691 if (unlikely(status <= 0)) {
1693 d_invalidate(dentry);
1695 return ERR_PTR(status);
1700 /* Fast lookup failed, do it the slow way */
1701 static struct dentry *__lookup_slow(const struct qstr *name,
1705 struct dentry *dentry, *old;
1706 struct inode *inode = dir->d_inode;
1707 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1709 /* Don't go there if it's already dead */
1710 if (unlikely(IS_DEADDIR(inode)))
1711 return ERR_PTR(-ENOENT);
1713 dentry = d_alloc_parallel(dir, name, &wq);
1716 if (unlikely(!d_in_lookup(dentry))) {
1717 int error = d_revalidate(dentry, flags);
1718 if (unlikely(error <= 0)) {
1720 d_invalidate(dentry);
1725 dentry = ERR_PTR(error);
1728 old = inode->i_op->lookup(inode, dentry, flags);
1729 d_lookup_done(dentry);
1730 if (unlikely(old)) {
1738 static struct dentry *lookup_slow(const struct qstr *name,
1742 struct inode *inode = dir->d_inode;
1744 inode_lock_shared(inode);
1745 res = __lookup_slow(name, dir, flags);
1746 inode_unlock_shared(inode);
1750 static inline int may_lookup(struct nameidata *nd)
1752 if (nd->flags & LOOKUP_RCU) {
1753 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1756 if (unlazy_walk(nd))
1759 return inode_permission(nd->inode, MAY_EXEC);
1762 static inline int handle_dots(struct nameidata *nd, int type)
1764 if (type == LAST_DOTDOT) {
1767 if (!nd->root.mnt) {
1768 error = set_root(nd);
1772 if (nd->flags & LOOKUP_RCU)
1773 error = follow_dotdot_rcu(nd);
1775 error = follow_dotdot(nd);
1779 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1781 * If there was a racing rename or mount along our
1782 * path, then we can't be sure that ".." hasn't jumped
1783 * above nd->root (and so userspace should retry or use
1787 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1789 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1796 static int pick_link(struct nameidata *nd, struct path *link,
1797 struct inode *inode, unsigned seq)
1801 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1802 path_to_nameidata(link, nd);
1805 if (!(nd->flags & LOOKUP_RCU)) {
1806 if (link->mnt == nd->path.mnt)
1809 error = nd_alloc_stack(nd);
1810 if (unlikely(error)) {
1811 if (error == -ECHILD) {
1812 if (unlikely(!legitimize_path(nd, link, seq))) {
1815 nd->flags &= ~LOOKUP_RCU;
1816 nd->path.mnt = NULL;
1817 nd->path.dentry = NULL;
1819 } else if (likely(unlazy_walk(nd)) == 0)
1820 error = nd_alloc_stack(nd);
1828 last = nd->stack + nd->depth++;
1830 clear_delayed_call(&last->done);
1831 nd->link_inode = inode;
1836 enum {WALK_FOLLOW = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1839 * Do we need to follow links? We _really_ want to be able
1840 * to do this check without having to look at inode->i_op,
1841 * so we keep a cache of "no, this doesn't need follow_link"
1842 * for the common case.
1844 static const char *step_into(struct nameidata *nd, int flags,
1845 struct dentry *dentry, struct inode *inode, unsigned seq)
1848 int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1851 return ERR_PTR(err);
1852 if (likely(!d_is_symlink(path.dentry)) ||
1853 !((flags & WALK_FOLLOW) || (nd->flags & LOOKUP_FOLLOW)) ||
1854 (flags & WALK_NOFOLLOW)) {
1855 /* not a symlink or should not follow */
1856 path_to_nameidata(&path, nd);
1861 /* make sure that d_is_symlink above matches inode */
1862 if (nd->flags & LOOKUP_RCU) {
1863 if (read_seqcount_retry(&path.dentry->d_seq, seq))
1864 return ERR_PTR(-ECHILD);
1866 err = pick_link(nd, &path, inode, seq);
1868 return get_link(nd);
1869 return ERR_PTR(err);
1872 static const char *walk_component(struct nameidata *nd, int flags)
1874 struct dentry *dentry;
1875 struct inode *inode;
1879 * "." and ".." are special - ".." especially so because it has
1880 * to be able to know about the current root directory and
1881 * parent relationships.
1883 if (unlikely(nd->last_type != LAST_NORM)) {
1884 if (!(flags & WALK_MORE) && nd->depth)
1886 err = handle_dots(nd, nd->last_type);
1887 return ERR_PTR(err);
1889 dentry = lookup_fast(nd, &inode, &seq);
1891 return ERR_CAST(dentry);
1892 if (unlikely(!dentry)) {
1893 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1895 return ERR_CAST(dentry);
1897 if (!(flags & WALK_MORE) && nd->depth)
1899 return step_into(nd, flags, dentry, inode, seq);
1903 * We can do the critical dentry name comparison and hashing
1904 * operations one word at a time, but we are limited to:
1906 * - Architectures with fast unaligned word accesses. We could
1907 * do a "get_unaligned()" if this helps and is sufficiently
1910 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1911 * do not trap on the (extremely unlikely) case of a page
1912 * crossing operation.
1914 * - Furthermore, we need an efficient 64-bit compile for the
1915 * 64-bit case in order to generate the "number of bytes in
1916 * the final mask". Again, that could be replaced with a
1917 * efficient population count instruction or similar.
1919 #ifdef CONFIG_DCACHE_WORD_ACCESS
1921 #include <asm/word-at-a-time.h>
1925 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1927 #elif defined(CONFIG_64BIT)
1929 * Register pressure in the mixing function is an issue, particularly
1930 * on 32-bit x86, but almost any function requires one state value and
1931 * one temporary. Instead, use a function designed for two state values
1932 * and no temporaries.
1934 * This function cannot create a collision in only two iterations, so
1935 * we have two iterations to achieve avalanche. In those two iterations,
1936 * we have six layers of mixing, which is enough to spread one bit's
1937 * influence out to 2^6 = 64 state bits.
1939 * Rotate constants are scored by considering either 64 one-bit input
1940 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1941 * probability of that delta causing a change to each of the 128 output
1942 * bits, using a sample of random initial states.
1944 * The Shannon entropy of the computed probabilities is then summed
1945 * to produce a score. Ideally, any input change has a 50% chance of
1946 * toggling any given output bit.
1948 * Mixing scores (in bits) for (12,45):
1949 * Input delta: 1-bit 2-bit
1950 * 1 round: 713.3 42542.6
1951 * 2 rounds: 2753.7 140389.8
1952 * 3 rounds: 5954.1 233458.2
1953 * 4 rounds: 7862.6 256672.2
1954 * Perfect: 8192 258048
1955 * (64*128) (64*63/2 * 128)
1957 #define HASH_MIX(x, y, a) \
1959 y ^= x, x = rol64(x,12),\
1960 x += y, y = rol64(y,45),\
1964 * Fold two longs into one 32-bit hash value. This must be fast, but
1965 * latency isn't quite as critical, as there is a fair bit of additional
1966 * work done before the hash value is used.
1968 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1970 y ^= x * GOLDEN_RATIO_64;
1971 y *= GOLDEN_RATIO_64;
1975 #else /* 32-bit case */
1978 * Mixing scores (in bits) for (7,20):
1979 * Input delta: 1-bit 2-bit
1980 * 1 round: 330.3 9201.6
1981 * 2 rounds: 1246.4 25475.4
1982 * 3 rounds: 1907.1 31295.1
1983 * 4 rounds: 2042.3 31718.6
1984 * Perfect: 2048 31744
1985 * (32*64) (32*31/2 * 64)
1987 #define HASH_MIX(x, y, a) \
1989 y ^= x, x = rol32(x, 7),\
1990 x += y, y = rol32(y,20),\
1993 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1995 /* Use arch-optimized multiply if one exists */
1996 return __hash_32(y ^ __hash_32(x));
2002 * Return the hash of a string of known length. This is carfully
2003 * designed to match hash_name(), which is the more critical function.
2004 * In particular, we must end by hashing a final word containing 0..7
2005 * payload bytes, to match the way that hash_name() iterates until it
2006 * finds the delimiter after the name.
2008 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2010 unsigned long a, x = 0, y = (unsigned long)salt;
2015 a = load_unaligned_zeropad(name);
2016 if (len < sizeof(unsigned long))
2019 name += sizeof(unsigned long);
2020 len -= sizeof(unsigned long);
2022 x ^= a & bytemask_from_count(len);
2024 return fold_hash(x, y);
2026 EXPORT_SYMBOL(full_name_hash);
2028 /* Return the "hash_len" (hash and length) of a null-terminated string */
2029 u64 hashlen_string(const void *salt, const char *name)
2031 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2032 unsigned long adata, mask, len;
2033 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2040 len += sizeof(unsigned long);
2042 a = load_unaligned_zeropad(name+len);
2043 } while (!has_zero(a, &adata, &constants));
2045 adata = prep_zero_mask(a, adata, &constants);
2046 mask = create_zero_mask(adata);
2047 x ^= a & zero_bytemask(mask);
2049 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2051 EXPORT_SYMBOL(hashlen_string);
2054 * Calculate the length and hash of the path component, and
2055 * return the "hash_len" as the result.
2057 static inline u64 hash_name(const void *salt, const char *name)
2059 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2060 unsigned long adata, bdata, mask, len;
2061 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2068 len += sizeof(unsigned long);
2070 a = load_unaligned_zeropad(name+len);
2071 b = a ^ REPEAT_BYTE('/');
2072 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2074 adata = prep_zero_mask(a, adata, &constants);
2075 bdata = prep_zero_mask(b, bdata, &constants);
2076 mask = create_zero_mask(adata | bdata);
2077 x ^= a & zero_bytemask(mask);
2079 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2082 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2084 /* Return the hash of a string of known length */
2085 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2087 unsigned long hash = init_name_hash(salt);
2089 hash = partial_name_hash((unsigned char)*name++, hash);
2090 return end_name_hash(hash);
2092 EXPORT_SYMBOL(full_name_hash);
2094 /* Return the "hash_len" (hash and length) of a null-terminated string */
2095 u64 hashlen_string(const void *salt, const char *name)
2097 unsigned long hash = init_name_hash(salt);
2098 unsigned long len = 0, c;
2100 c = (unsigned char)*name;
2103 hash = partial_name_hash(c, hash);
2104 c = (unsigned char)name[len];
2106 return hashlen_create(end_name_hash(hash), len);
2108 EXPORT_SYMBOL(hashlen_string);
2111 * We know there's a real path component here of at least
2114 static inline u64 hash_name(const void *salt, const char *name)
2116 unsigned long hash = init_name_hash(salt);
2117 unsigned long len = 0, c;
2119 c = (unsigned char)*name;
2122 hash = partial_name_hash(c, hash);
2123 c = (unsigned char)name[len];
2124 } while (c && c != '/');
2125 return hashlen_create(end_name_hash(hash), len);
2132 * This is the basic name resolution function, turning a pathname into
2133 * the final dentry. We expect 'base' to be positive and a directory.
2135 * Returns 0 and nd will have valid dentry and mnt on success.
2136 * Returns error and drops reference to input namei data on failure.
2138 static int link_path_walk(const char *name, struct nameidata *nd)
2143 return PTR_ERR(name);
2149 /* At this point we know we have a real path component. */
2155 err = may_lookup(nd);
2159 hash_len = hash_name(nd->path.dentry, name);
2162 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2164 if (name[1] == '.') {
2166 nd->flags |= LOOKUP_JUMPED;
2172 if (likely(type == LAST_NORM)) {
2173 struct dentry *parent = nd->path.dentry;
2174 nd->flags &= ~LOOKUP_JUMPED;
2175 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2176 struct qstr this = { { .hash_len = hash_len }, .name = name };
2177 err = parent->d_op->d_hash(parent, &this);
2180 hash_len = this.hash_len;
2185 nd->last.hash_len = hash_len;
2186 nd->last.name = name;
2187 nd->last_type = type;
2189 name += hashlen_len(hash_len);
2193 * If it wasn't NUL, we know it was '/'. Skip that
2194 * slash, and continue until no more slashes.
2198 } while (unlikely(*name == '/'));
2199 if (unlikely(!*name)) {
2201 /* pathname body, done */
2204 name = nd->stack[nd->depth - 1].name;
2205 /* trailing symlink, done */
2208 /* last component of nested symlink */
2209 link = walk_component(nd, WALK_FOLLOW);
2211 /* not the last component */
2212 link = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2214 if (unlikely(link)) {
2216 return PTR_ERR(link);
2217 /* a symlink to follow */
2218 nd->stack[nd->depth - 1].name = name;
2222 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2223 if (nd->flags & LOOKUP_RCU) {
2224 if (unlazy_walk(nd))
2232 /* must be paired with terminate_walk() */
2233 static const char *path_init(struct nameidata *nd, unsigned flags)
2236 const char *s = nd->name->name;
2239 flags &= ~LOOKUP_RCU;
2240 if (flags & LOOKUP_RCU)
2243 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2244 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2247 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2248 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2251 if (flags & LOOKUP_ROOT) {
2252 struct dentry *root = nd->root.dentry;
2253 struct inode *inode = root->d_inode;
2254 if (*s && unlikely(!d_can_lookup(root)))
2255 return ERR_PTR(-ENOTDIR);
2256 nd->path = nd->root;
2258 if (flags & LOOKUP_RCU) {
2259 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2260 nd->root_seq = nd->seq;
2262 path_get(&nd->path);
2267 nd->root.mnt = NULL;
2268 nd->path.mnt = NULL;
2269 nd->path.dentry = NULL;
2271 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2272 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2273 error = nd_jump_root(nd);
2274 if (unlikely(error))
2275 return ERR_PTR(error);
2279 /* Relative pathname -- get the starting-point it is relative to. */
2280 if (nd->dfd == AT_FDCWD) {
2281 if (flags & LOOKUP_RCU) {
2282 struct fs_struct *fs = current->fs;
2286 seq = read_seqcount_begin(&fs->seq);
2288 nd->inode = nd->path.dentry->d_inode;
2289 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2290 } while (read_seqcount_retry(&fs->seq, seq));
2292 get_fs_pwd(current->fs, &nd->path);
2293 nd->inode = nd->path.dentry->d_inode;
2296 /* Caller must check execute permissions on the starting path component */
2297 struct fd f = fdget_raw(nd->dfd);
2298 struct dentry *dentry;
2301 return ERR_PTR(-EBADF);
2303 dentry = f.file->f_path.dentry;
2305 if (*s && unlikely(!d_can_lookup(dentry))) {
2307 return ERR_PTR(-ENOTDIR);
2310 nd->path = f.file->f_path;
2311 if (flags & LOOKUP_RCU) {
2312 nd->inode = nd->path.dentry->d_inode;
2313 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2315 path_get(&nd->path);
2316 nd->inode = nd->path.dentry->d_inode;
2321 /* For scoped-lookups we need to set the root to the dirfd as well. */
2322 if (flags & LOOKUP_IS_SCOPED) {
2323 nd->root = nd->path;
2324 if (flags & LOOKUP_RCU) {
2325 nd->root_seq = nd->seq;
2327 path_get(&nd->root);
2328 nd->flags |= LOOKUP_ROOT_GRABBED;
2334 static inline const char *lookup_last(struct nameidata *nd)
2337 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2338 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2340 nd->flags &= ~LOOKUP_PARENT;
2341 link = walk_component(nd, 0);
2343 nd->flags |= LOOKUP_PARENT;
2344 nd->stack[0].name = NULL;
2349 static int handle_lookup_down(struct nameidata *nd)
2351 if (!(nd->flags & LOOKUP_RCU))
2352 dget(nd->path.dentry);
2353 return PTR_ERR(step_into(nd, WALK_NOFOLLOW,
2354 nd->path.dentry, nd->inode, nd->seq));
2357 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2358 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2360 const char *s = path_init(nd, flags);
2363 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2364 err = handle_lookup_down(nd);
2365 if (unlikely(err < 0))
2369 while (!(err = link_path_walk(s, nd)) &&
2370 (s = lookup_last(nd)) != NULL)
2373 err = complete_walk(nd);
2375 if (!err && nd->flags & LOOKUP_DIRECTORY)
2376 if (!d_can_lookup(nd->path.dentry))
2378 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2379 err = handle_lookup_down(nd);
2380 nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please...
2384 nd->path.mnt = NULL;
2385 nd->path.dentry = NULL;
2391 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2392 struct path *path, struct path *root)
2395 struct nameidata nd;
2397 return PTR_ERR(name);
2398 if (unlikely(root)) {
2400 flags |= LOOKUP_ROOT;
2402 set_nameidata(&nd, dfd, name);
2403 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2404 if (unlikely(retval == -ECHILD))
2405 retval = path_lookupat(&nd, flags, path);
2406 if (unlikely(retval == -ESTALE))
2407 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2409 if (likely(!retval))
2410 audit_inode(name, path->dentry,
2411 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2412 restore_nameidata();
2417 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2418 static int path_parentat(struct nameidata *nd, unsigned flags,
2419 struct path *parent)
2421 const char *s = path_init(nd, flags);
2422 int err = link_path_walk(s, nd);
2424 err = complete_walk(nd);
2427 nd->path.mnt = NULL;
2428 nd->path.dentry = NULL;
2434 static struct filename *filename_parentat(int dfd, struct filename *name,
2435 unsigned int flags, struct path *parent,
2436 struct qstr *last, int *type)
2439 struct nameidata nd;
2443 set_nameidata(&nd, dfd, name);
2444 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2445 if (unlikely(retval == -ECHILD))
2446 retval = path_parentat(&nd, flags, parent);
2447 if (unlikely(retval == -ESTALE))
2448 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2449 if (likely(!retval)) {
2451 *type = nd.last_type;
2452 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2455 name = ERR_PTR(retval);
2457 restore_nameidata();
2461 /* does lookup, returns the object with parent locked */
2462 struct dentry *kern_path_locked(const char *name, struct path *path)
2464 struct filename *filename;
2469 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2471 if (IS_ERR(filename))
2472 return ERR_CAST(filename);
2473 if (unlikely(type != LAST_NORM)) {
2476 return ERR_PTR(-EINVAL);
2478 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2479 d = __lookup_hash(&last, path->dentry, 0);
2481 inode_unlock(path->dentry->d_inode);
2488 int kern_path(const char *name, unsigned int flags, struct path *path)
2490 return filename_lookup(AT_FDCWD, getname_kernel(name),
2493 EXPORT_SYMBOL(kern_path);
2496 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2497 * @dentry: pointer to dentry of the base directory
2498 * @mnt: pointer to vfs mount of the base directory
2499 * @name: pointer to file name
2500 * @flags: lookup flags
2501 * @path: pointer to struct path to fill
2503 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2504 const char *name, unsigned int flags,
2507 struct path root = {.mnt = mnt, .dentry = dentry};
2508 /* the first argument of filename_lookup() is ignored with root */
2509 return filename_lookup(AT_FDCWD, getname_kernel(name),
2510 flags , path, &root);
2512 EXPORT_SYMBOL(vfs_path_lookup);
2514 static int lookup_one_len_common(const char *name, struct dentry *base,
2515 int len, struct qstr *this)
2519 this->hash = full_name_hash(base, name, len);
2523 if (unlikely(name[0] == '.')) {
2524 if (len < 2 || (len == 2 && name[1] == '.'))
2529 unsigned int c = *(const unsigned char *)name++;
2530 if (c == '/' || c == '\0')
2534 * See if the low-level filesystem might want
2535 * to use its own hash..
2537 if (base->d_flags & DCACHE_OP_HASH) {
2538 int err = base->d_op->d_hash(base, this);
2543 return inode_permission(base->d_inode, MAY_EXEC);
2547 * try_lookup_one_len - filesystem helper to lookup single pathname component
2548 * @name: pathname component to lookup
2549 * @base: base directory to lookup from
2550 * @len: maximum length @len should be interpreted to
2552 * Look up a dentry by name in the dcache, returning NULL if it does not
2553 * currently exist. The function does not try to create a dentry.
2555 * Note that this routine is purely a helper for filesystem usage and should
2556 * not be called by generic code.
2558 * The caller must hold base->i_mutex.
2560 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2565 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2567 err = lookup_one_len_common(name, base, len, &this);
2569 return ERR_PTR(err);
2571 return lookup_dcache(&this, base, 0);
2573 EXPORT_SYMBOL(try_lookup_one_len);
2576 * lookup_one_len - filesystem helper to lookup single pathname component
2577 * @name: pathname component to lookup
2578 * @base: base directory to lookup from
2579 * @len: maximum length @len should be interpreted to
2581 * Note that this routine is purely a helper for filesystem usage and should
2582 * not be called by generic code.
2584 * The caller must hold base->i_mutex.
2586 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2588 struct dentry *dentry;
2592 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2594 err = lookup_one_len_common(name, base, len, &this);
2596 return ERR_PTR(err);
2598 dentry = lookup_dcache(&this, base, 0);
2599 return dentry ? dentry : __lookup_slow(&this, base, 0);
2601 EXPORT_SYMBOL(lookup_one_len);
2604 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2605 * @name: pathname component to lookup
2606 * @base: base directory to lookup from
2607 * @len: maximum length @len should be interpreted to
2609 * Note that this routine is purely a helper for filesystem usage and should
2610 * not be called by generic code.
2612 * Unlike lookup_one_len, it should be called without the parent
2613 * i_mutex held, and will take the i_mutex itself if necessary.
2615 struct dentry *lookup_one_len_unlocked(const char *name,
2616 struct dentry *base, int len)
2622 err = lookup_one_len_common(name, base, len, &this);
2624 return ERR_PTR(err);
2626 ret = lookup_dcache(&this, base, 0);
2628 ret = lookup_slow(&this, base, 0);
2631 EXPORT_SYMBOL(lookup_one_len_unlocked);
2634 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2635 * on negatives. Returns known positive or ERR_PTR(); that's what
2636 * most of the users want. Note that pinned negative with unlocked parent
2637 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2638 * need to be very careful; pinned positives have ->d_inode stable, so
2639 * this one avoids such problems.
2641 struct dentry *lookup_positive_unlocked(const char *name,
2642 struct dentry *base, int len)
2644 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2645 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2647 ret = ERR_PTR(-ENOENT);
2651 EXPORT_SYMBOL(lookup_positive_unlocked);
2653 #ifdef CONFIG_UNIX98_PTYS
2654 int path_pts(struct path *path)
2656 /* Find something mounted on "pts" in the same directory as
2659 struct dentry *child, *parent;
2663 ret = path_parent_directory(path);
2667 parent = path->dentry;
2670 child = d_hash_and_lookup(parent, &this);
2674 path->dentry = child;
2681 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2682 struct path *path, int *empty)
2684 return filename_lookup(dfd, getname_flags(name, flags, empty),
2687 EXPORT_SYMBOL(user_path_at_empty);
2689 int __check_sticky(struct inode *dir, struct inode *inode)
2691 kuid_t fsuid = current_fsuid();
2693 if (uid_eq(inode->i_uid, fsuid))
2695 if (uid_eq(dir->i_uid, fsuid))
2697 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2699 EXPORT_SYMBOL(__check_sticky);
2702 * Check whether we can remove a link victim from directory dir, check
2703 * whether the type of victim is right.
2704 * 1. We can't do it if dir is read-only (done in permission())
2705 * 2. We should have write and exec permissions on dir
2706 * 3. We can't remove anything from append-only dir
2707 * 4. We can't do anything with immutable dir (done in permission())
2708 * 5. If the sticky bit on dir is set we should either
2709 * a. be owner of dir, or
2710 * b. be owner of victim, or
2711 * c. have CAP_FOWNER capability
2712 * 6. If the victim is append-only or immutable we can't do antyhing with
2713 * links pointing to it.
2714 * 7. If the victim has an unknown uid or gid we can't change the inode.
2715 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2716 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2717 * 10. We can't remove a root or mountpoint.
2718 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2719 * nfs_async_unlink().
2721 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2723 struct inode *inode = d_backing_inode(victim);
2726 if (d_is_negative(victim))
2730 BUG_ON(victim->d_parent->d_inode != dir);
2732 /* Inode writeback is not safe when the uid or gid are invalid. */
2733 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2736 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2738 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2744 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2745 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2748 if (!d_is_dir(victim))
2750 if (IS_ROOT(victim))
2752 } else if (d_is_dir(victim))
2754 if (IS_DEADDIR(dir))
2756 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2761 /* Check whether we can create an object with dentry child in directory
2763 * 1. We can't do it if child already exists (open has special treatment for
2764 * this case, but since we are inlined it's OK)
2765 * 2. We can't do it if dir is read-only (done in permission())
2766 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2767 * 4. We should have write and exec permissions on dir
2768 * 5. We can't do it if dir is immutable (done in permission())
2770 static inline int may_create(struct inode *dir, struct dentry *child)
2772 struct user_namespace *s_user_ns;
2773 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2776 if (IS_DEADDIR(dir))
2778 s_user_ns = dir->i_sb->s_user_ns;
2779 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2780 !kgid_has_mapping(s_user_ns, current_fsgid()))
2782 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2786 * p1 and p2 should be directories on the same fs.
2788 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2793 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2797 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2799 p = d_ancestor(p2, p1);
2801 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2802 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2806 p = d_ancestor(p1, p2);
2808 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2809 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2813 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2814 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2817 EXPORT_SYMBOL(lock_rename);
2819 void unlock_rename(struct dentry *p1, struct dentry *p2)
2821 inode_unlock(p1->d_inode);
2823 inode_unlock(p2->d_inode);
2824 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2827 EXPORT_SYMBOL(unlock_rename);
2829 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2832 int error = may_create(dir, dentry);
2836 if (!dir->i_op->create)
2837 return -EACCES; /* shouldn't it be ENOSYS? */
2840 error = security_inode_create(dir, dentry, mode);
2843 error = dir->i_op->create(dir, dentry, mode, want_excl);
2845 fsnotify_create(dir, dentry);
2848 EXPORT_SYMBOL(vfs_create);
2850 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2851 int (*f)(struct dentry *, umode_t, void *),
2854 struct inode *dir = dentry->d_parent->d_inode;
2855 int error = may_create(dir, dentry);
2861 error = security_inode_create(dir, dentry, mode);
2864 error = f(dentry, mode, arg);
2866 fsnotify_create(dir, dentry);
2869 EXPORT_SYMBOL(vfs_mkobj);
2871 bool may_open_dev(const struct path *path)
2873 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2874 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2877 static int may_open(const struct path *path, int acc_mode, int flag)
2879 struct dentry *dentry = path->dentry;
2880 struct inode *inode = dentry->d_inode;
2886 switch (inode->i_mode & S_IFMT) {
2890 if (acc_mode & MAY_WRITE)
2895 if (!may_open_dev(path))
2904 error = inode_permission(inode, MAY_OPEN | acc_mode);
2909 * An append-only file must be opened in append mode for writing.
2911 if (IS_APPEND(inode)) {
2912 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2918 /* O_NOATIME can only be set by the owner or superuser */
2919 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2925 static int handle_truncate(struct file *filp)
2927 const struct path *path = &filp->f_path;
2928 struct inode *inode = path->dentry->d_inode;
2929 int error = get_write_access(inode);
2933 * Refuse to truncate files with mandatory locks held on them.
2935 error = locks_verify_locked(filp);
2937 error = security_path_truncate(path);
2939 error = do_truncate(path->dentry, 0,
2940 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2943 put_write_access(inode);
2947 static inline int open_to_namei_flags(int flag)
2949 if ((flag & O_ACCMODE) == 3)
2954 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2956 struct user_namespace *s_user_ns;
2957 int error = security_path_mknod(dir, dentry, mode, 0);
2961 s_user_ns = dir->dentry->d_sb->s_user_ns;
2962 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2963 !kgid_has_mapping(s_user_ns, current_fsgid()))
2966 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2970 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2974 * Attempt to atomically look up, create and open a file from a negative
2977 * Returns 0 if successful. The file will have been created and attached to
2978 * @file by the filesystem calling finish_open().
2980 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
2981 * be set. The caller will need to perform the open themselves. @path will
2982 * have been updated to point to the new dentry. This may be negative.
2984 * Returns an error code otherwise.
2986 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
2988 const struct open_flags *op,
2989 int open_flag, umode_t mode)
2991 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2992 struct inode *dir = nd->path.dentry->d_inode;
2995 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
2996 open_flag &= ~O_TRUNC;
2998 if (nd->flags & LOOKUP_DIRECTORY)
2999 open_flag |= O_DIRECTORY;
3001 file->f_path.dentry = DENTRY_NOT_SET;
3002 file->f_path.mnt = nd->path.mnt;
3003 error = dir->i_op->atomic_open(dir, dentry, file,
3004 open_to_namei_flags(open_flag), mode);
3005 d_lookup_done(dentry);
3007 if (file->f_mode & FMODE_OPENED) {
3009 * We didn't have the inode before the open, so check open
3012 int acc_mode = op->acc_mode;
3013 if (file->f_mode & FMODE_CREATED) {
3014 WARN_ON(!(open_flag & O_CREAT));
3015 fsnotify_create(dir, dentry);
3018 error = may_open(&file->f_path, acc_mode, open_flag);
3019 if (WARN_ON(error > 0))
3021 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3024 if (file->f_path.dentry) {
3026 dentry = file->f_path.dentry;
3028 if (file->f_mode & FMODE_CREATED)
3029 fsnotify_create(dir, dentry);
3030 if (unlikely(d_is_negative(dentry)))
3036 dentry = ERR_PTR(error);
3042 * Look up and maybe create and open the last component.
3044 * Must be called with parent locked (exclusive in O_CREAT case).
3046 * Returns 0 on success, that is, if
3047 * the file was successfully atomically created (if necessary) and opened, or
3048 * the file was not completely opened at this time, though lookups and
3049 * creations were performed.
3050 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3051 * In the latter case dentry returned in @path might be negative if O_CREAT
3052 * hadn't been specified.
3054 * An error code is returned on failure.
3056 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3057 const struct open_flags *op,
3060 struct dentry *dir = nd->path.dentry;
3061 struct inode *dir_inode = dir->d_inode;
3062 int open_flag = op->open_flag;
3063 struct dentry *dentry;
3064 int error, create_error = 0;
3065 umode_t mode = op->mode;
3066 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3068 if (unlikely(IS_DEADDIR(dir_inode)))
3069 return ERR_PTR(-ENOENT);
3071 file->f_mode &= ~FMODE_CREATED;
3072 dentry = d_lookup(dir, &nd->last);
3075 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3079 if (d_in_lookup(dentry))
3082 error = d_revalidate(dentry, nd->flags);
3083 if (likely(error > 0))
3087 d_invalidate(dentry);
3091 if (dentry->d_inode) {
3092 /* Cached positive dentry: will open in f_op->open */
3097 * Checking write permission is tricky, bacuse we don't know if we are
3098 * going to actually need it: O_CREAT opens should work as long as the
3099 * file exists. But checking existence breaks atomicity. The trick is
3100 * to check access and if not granted clear O_CREAT from the flags.
3102 * Another problem is returing the "right" error value (e.g. for an
3103 * O_EXCL open we want to return EEXIST not EROFS).
3105 if (open_flag & O_CREAT) {
3106 if (!IS_POSIXACL(dir->d_inode))
3107 mode &= ~current_umask();
3108 if (unlikely(!got_write)) {
3109 create_error = -EROFS;
3110 open_flag &= ~O_CREAT;
3111 if (open_flag & (O_EXCL | O_TRUNC))
3113 /* No side effects, safe to clear O_CREAT */
3115 create_error = may_o_create(&nd->path, dentry, mode);
3117 open_flag &= ~O_CREAT;
3118 if (open_flag & O_EXCL)
3122 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3123 unlikely(!got_write)) {
3125 * No O_CREATE -> atomicity not a requirement -> fall
3126 * back to lookup + open
3131 if (dir_inode->i_op->atomic_open) {
3132 dentry = atomic_open(nd, dentry, file, op, open_flag, mode);
3133 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3134 dentry = ERR_PTR(create_error);
3139 if (d_in_lookup(dentry)) {
3140 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3142 d_lookup_done(dentry);
3143 if (unlikely(res)) {
3145 error = PTR_ERR(res);
3153 /* Negative dentry, just create the file */
3154 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3155 file->f_mode |= FMODE_CREATED;
3156 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3157 if (!dir_inode->i_op->create) {
3161 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3162 open_flag & O_EXCL);
3165 fsnotify_create(dir_inode, dentry);
3167 if (unlikely(create_error) && !dentry->d_inode) {
3168 error = create_error;
3175 return ERR_PTR(error);
3179 * Handle the last step of open()
3181 static const char *do_last(struct nameidata *nd,
3182 struct file *file, const struct open_flags *op)
3184 struct dentry *dir = nd->path.dentry;
3185 kuid_t dir_uid = nd->inode->i_uid;
3186 umode_t dir_mode = nd->inode->i_mode;
3187 int open_flag = op->open_flag;
3188 bool will_truncate = (open_flag & O_TRUNC) != 0;
3189 bool got_write = false;
3190 int acc_mode = op->acc_mode;
3192 struct inode *inode;
3193 struct dentry *dentry;
3197 nd->flags &= ~LOOKUP_PARENT;
3198 nd->flags |= op->intent;
3200 if (nd->last_type != LAST_NORM) {
3203 error = handle_dots(nd, nd->last_type);
3204 if (unlikely(error))
3205 return ERR_PTR(error);
3209 if (!(open_flag & O_CREAT)) {
3210 if (nd->last.name[nd->last.len])
3211 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3212 /* we _can_ be in RCU mode here */
3213 dentry = lookup_fast(nd, &inode, &seq);
3215 return ERR_CAST(dentry);
3219 BUG_ON(nd->inode != dir->d_inode);
3220 BUG_ON(nd->flags & LOOKUP_RCU);
3222 /* create side of things */
3224 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3225 * has been cleared when we got to the last component we are
3228 error = complete_walk(nd);
3230 return ERR_PTR(error);
3232 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3233 /* trailing slashes? */
3234 if (unlikely(nd->last.name[nd->last.len]))
3235 return ERR_PTR(-EISDIR);
3238 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3239 error = mnt_want_write(nd->path.mnt);
3243 * do _not_ fail yet - we might not need that or fail with
3244 * a different error; let lookup_open() decide; we'll be
3245 * dropping this one anyway.
3248 if (open_flag & O_CREAT)
3249 inode_lock(dir->d_inode);
3251 inode_lock_shared(dir->d_inode);
3252 dentry = lookup_open(nd, file, op, got_write);
3253 if (open_flag & O_CREAT)
3254 inode_unlock(dir->d_inode);
3256 inode_unlock_shared(dir->d_inode);
3258 if (IS_ERR(dentry)) {
3259 error = PTR_ERR(dentry);
3263 if (file->f_mode & FMODE_OPENED) {
3264 if ((file->f_mode & FMODE_CREATED) ||
3265 !S_ISREG(file_inode(file)->i_mode))
3266 will_truncate = false;
3268 audit_inode(nd->name, file->f_path.dentry, 0);
3273 if (file->f_mode & FMODE_CREATED) {
3274 /* Don't check for write permission, don't truncate */
3275 open_flag &= ~O_TRUNC;
3276 will_truncate = false;
3278 dput(nd->path.dentry);
3279 nd->path.dentry = dentry;
3280 goto finish_open_created;
3284 * If atomic_open() acquired write access it is dropped now due to
3285 * possible mount and symlink following (this might be optimized away if
3289 mnt_drop_write(nd->path.mnt);
3296 res = step_into(nd, 0, dentry, inode, seq);
3297 if (unlikely(res)) {
3298 nd->flags |= LOOKUP_PARENT;
3299 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3300 nd->stack[0].name = NULL;
3304 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3305 audit_inode(nd->name, nd->path.dentry, 0);
3306 return ERR_PTR(-EEXIST);
3309 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3310 error = complete_walk(nd);
3312 return ERR_PTR(error);
3313 audit_inode(nd->name, nd->path.dentry, 0);
3314 if (open_flag & O_CREAT) {
3316 if (d_is_dir(nd->path.dentry))
3318 error = may_create_in_sticky(dir_mode, dir_uid,
3319 d_backing_inode(nd->path.dentry));
3320 if (unlikely(error))
3324 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3326 if (!d_is_reg(nd->path.dentry))
3327 will_truncate = false;
3329 if (will_truncate) {
3330 error = mnt_want_write(nd->path.mnt);
3335 finish_open_created:
3336 error = may_open(&nd->path, acc_mode, open_flag);
3339 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
3340 error = vfs_open(&nd->path, file);
3344 error = ima_file_check(file, op->acc_mode);
3345 if (!error && will_truncate)
3346 error = handle_truncate(file);
3348 if (unlikely(error > 0)) {
3353 mnt_drop_write(nd->path.mnt);
3354 return ERR_PTR(error);
3357 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3359 struct dentry *child = NULL;
3360 struct inode *dir = dentry->d_inode;
3361 struct inode *inode;
3364 /* we want directory to be writable */
3365 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3368 error = -EOPNOTSUPP;
3369 if (!dir->i_op->tmpfile)
3372 child = d_alloc(dentry, &slash_name);
3373 if (unlikely(!child))
3375 error = dir->i_op->tmpfile(dir, child, mode);
3379 inode = child->d_inode;
3380 if (unlikely(!inode))
3382 if (!(open_flag & O_EXCL)) {
3383 spin_lock(&inode->i_lock);
3384 inode->i_state |= I_LINKABLE;
3385 spin_unlock(&inode->i_lock);
3387 ima_post_create_tmpfile(inode);
3392 return ERR_PTR(error);
3394 EXPORT_SYMBOL(vfs_tmpfile);
3396 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3397 const struct open_flags *op,
3400 struct dentry *child;
3402 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3403 if (unlikely(error))
3405 error = mnt_want_write(path.mnt);
3406 if (unlikely(error))
3408 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3409 error = PTR_ERR(child);
3413 path.dentry = child;
3414 audit_inode(nd->name, child, 0);
3415 /* Don't check for other permissions, the inode was just created */
3416 error = may_open(&path, 0, op->open_flag);
3419 file->f_path.mnt = path.mnt;
3420 error = finish_open(file, child, NULL);
3422 mnt_drop_write(path.mnt);
3428 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3431 int error = path_lookupat(nd, flags, &path);
3433 audit_inode(nd->name, path.dentry, 0);
3434 error = vfs_open(&path, file);
3440 static struct file *path_openat(struct nameidata *nd,
3441 const struct open_flags *op, unsigned flags)
3446 file = alloc_empty_file(op->open_flag, current_cred());
3450 if (unlikely(file->f_flags & __O_TMPFILE)) {
3451 error = do_tmpfile(nd, flags, op, file);
3452 } else if (unlikely(file->f_flags & O_PATH)) {
3453 error = do_o_path(nd, flags, file);
3455 const char *s = path_init(nd, flags);
3456 while (!(error = link_path_walk(s, nd)) &&
3457 (s = do_last(nd, file, op)) != NULL)
3461 if (likely(!error)) {
3462 if (likely(file->f_mode & FMODE_OPENED))
3468 if (error == -EOPENSTALE) {
3469 if (flags & LOOKUP_RCU)
3474 return ERR_PTR(error);
3477 struct file *do_filp_open(int dfd, struct filename *pathname,
3478 const struct open_flags *op)
3480 struct nameidata nd;
3481 int flags = op->lookup_flags;
3484 set_nameidata(&nd, dfd, pathname);
3485 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3486 if (unlikely(filp == ERR_PTR(-ECHILD)))
3487 filp = path_openat(&nd, op, flags);
3488 if (unlikely(filp == ERR_PTR(-ESTALE)))
3489 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3490 restore_nameidata();
3494 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3495 const char *name, const struct open_flags *op)
3497 struct nameidata nd;
3499 struct filename *filename;
3500 int flags = op->lookup_flags | LOOKUP_ROOT;
3503 nd.root.dentry = dentry;
3505 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3506 return ERR_PTR(-ELOOP);
3508 filename = getname_kernel(name);
3509 if (IS_ERR(filename))
3510 return ERR_CAST(filename);
3512 set_nameidata(&nd, -1, filename);
3513 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3514 if (unlikely(file == ERR_PTR(-ECHILD)))
3515 file = path_openat(&nd, op, flags);
3516 if (unlikely(file == ERR_PTR(-ESTALE)))
3517 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3518 restore_nameidata();
3523 static struct dentry *filename_create(int dfd, struct filename *name,
3524 struct path *path, unsigned int lookup_flags)
3526 struct dentry *dentry = ERR_PTR(-EEXIST);
3531 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3534 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3535 * other flags passed in are ignored!
3537 lookup_flags &= LOOKUP_REVAL;
3539 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3541 return ERR_CAST(name);
3544 * Yucky last component or no last component at all?
3545 * (foo/., foo/.., /////)
3547 if (unlikely(type != LAST_NORM))
3550 /* don't fail immediately if it's r/o, at least try to report other errors */
3551 err2 = mnt_want_write(path->mnt);
3553 * Do the final lookup.
3555 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3556 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3557 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3562 if (d_is_positive(dentry))
3566 * Special case - lookup gave negative, but... we had foo/bar/
3567 * From the vfs_mknod() POV we just have a negative dentry -
3568 * all is fine. Let's be bastards - you had / on the end, you've
3569 * been asking for (non-existent) directory. -ENOENT for you.
3571 if (unlikely(!is_dir && last.name[last.len])) {
3575 if (unlikely(err2)) {
3583 dentry = ERR_PTR(error);
3585 inode_unlock(path->dentry->d_inode);
3587 mnt_drop_write(path->mnt);
3594 struct dentry *kern_path_create(int dfd, const char *pathname,
3595 struct path *path, unsigned int lookup_flags)
3597 return filename_create(dfd, getname_kernel(pathname),
3598 path, lookup_flags);
3600 EXPORT_SYMBOL(kern_path_create);
3602 void done_path_create(struct path *path, struct dentry *dentry)
3605 inode_unlock(path->dentry->d_inode);
3606 mnt_drop_write(path->mnt);
3609 EXPORT_SYMBOL(done_path_create);
3611 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3612 struct path *path, unsigned int lookup_flags)
3614 return filename_create(dfd, getname(pathname), path, lookup_flags);
3616 EXPORT_SYMBOL(user_path_create);
3618 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3620 int error = may_create(dir, dentry);
3625 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3628 if (!dir->i_op->mknod)
3631 error = devcgroup_inode_mknod(mode, dev);
3635 error = security_inode_mknod(dir, dentry, mode, dev);
3639 error = dir->i_op->mknod(dir, dentry, mode, dev);
3641 fsnotify_create(dir, dentry);
3644 EXPORT_SYMBOL(vfs_mknod);
3646 static int may_mknod(umode_t mode)
3648 switch (mode & S_IFMT) {
3654 case 0: /* zero mode translates to S_IFREG */
3663 long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3666 struct dentry *dentry;
3669 unsigned int lookup_flags = 0;
3671 error = may_mknod(mode);
3675 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3677 return PTR_ERR(dentry);
3679 if (!IS_POSIXACL(path.dentry->d_inode))
3680 mode &= ~current_umask();
3681 error = security_path_mknod(&path, dentry, mode, dev);
3684 switch (mode & S_IFMT) {
3685 case 0: case S_IFREG:
3686 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3688 ima_post_path_mknod(dentry);
3690 case S_IFCHR: case S_IFBLK:
3691 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3692 new_decode_dev(dev));
3694 case S_IFIFO: case S_IFSOCK:
3695 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3699 done_path_create(&path, dentry);
3700 if (retry_estale(error, lookup_flags)) {
3701 lookup_flags |= LOOKUP_REVAL;
3707 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3710 return do_mknodat(dfd, filename, mode, dev);
3713 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3715 return do_mknodat(AT_FDCWD, filename, mode, dev);
3718 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3720 int error = may_create(dir, dentry);
3721 unsigned max_links = dir->i_sb->s_max_links;
3726 if (!dir->i_op->mkdir)
3729 mode &= (S_IRWXUGO|S_ISVTX);
3730 error = security_inode_mkdir(dir, dentry, mode);
3734 if (max_links && dir->i_nlink >= max_links)
3737 error = dir->i_op->mkdir(dir, dentry, mode);
3739 fsnotify_mkdir(dir, dentry);
3742 EXPORT_SYMBOL(vfs_mkdir);
3744 long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3746 struct dentry *dentry;
3749 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3752 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3754 return PTR_ERR(dentry);
3756 if (!IS_POSIXACL(path.dentry->d_inode))
3757 mode &= ~current_umask();
3758 error = security_path_mkdir(&path, dentry, mode);
3760 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3761 done_path_create(&path, dentry);
3762 if (retry_estale(error, lookup_flags)) {
3763 lookup_flags |= LOOKUP_REVAL;
3769 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3771 return do_mkdirat(dfd, pathname, mode);
3774 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3776 return do_mkdirat(AT_FDCWD, pathname, mode);
3779 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3781 int error = may_delete(dir, dentry, 1);
3786 if (!dir->i_op->rmdir)
3790 inode_lock(dentry->d_inode);
3793 if (is_local_mountpoint(dentry))
3796 error = security_inode_rmdir(dir, dentry);
3800 error = dir->i_op->rmdir(dir, dentry);
3804 shrink_dcache_parent(dentry);
3805 dentry->d_inode->i_flags |= S_DEAD;
3807 detach_mounts(dentry);
3808 fsnotify_rmdir(dir, dentry);
3811 inode_unlock(dentry->d_inode);
3817 EXPORT_SYMBOL(vfs_rmdir);
3819 long do_rmdir(int dfd, const char __user *pathname)
3822 struct filename *name;
3823 struct dentry *dentry;
3827 unsigned int lookup_flags = 0;
3829 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3830 &path, &last, &type);
3832 return PTR_ERR(name);
3846 error = mnt_want_write(path.mnt);
3850 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3851 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3852 error = PTR_ERR(dentry);
3855 if (!dentry->d_inode) {
3859 error = security_path_rmdir(&path, dentry);
3862 error = vfs_rmdir(path.dentry->d_inode, dentry);
3866 inode_unlock(path.dentry->d_inode);
3867 mnt_drop_write(path.mnt);
3871 if (retry_estale(error, lookup_flags)) {
3872 lookup_flags |= LOOKUP_REVAL;
3878 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3880 return do_rmdir(AT_FDCWD, pathname);
3884 * vfs_unlink - unlink a filesystem object
3885 * @dir: parent directory
3887 * @delegated_inode: returns victim inode, if the inode is delegated.
3889 * The caller must hold dir->i_mutex.
3891 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3892 * return a reference to the inode in delegated_inode. The caller
3893 * should then break the delegation on that inode and retry. Because
3894 * breaking a delegation may take a long time, the caller should drop
3895 * dir->i_mutex before doing so.
3897 * Alternatively, a caller may pass NULL for delegated_inode. This may
3898 * be appropriate for callers that expect the underlying filesystem not
3899 * to be NFS exported.
3901 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3903 struct inode *target = dentry->d_inode;
3904 int error = may_delete(dir, dentry, 0);
3909 if (!dir->i_op->unlink)
3913 if (is_local_mountpoint(dentry))
3916 error = security_inode_unlink(dir, dentry);
3918 error = try_break_deleg(target, delegated_inode);
3921 error = dir->i_op->unlink(dir, dentry);
3924 detach_mounts(dentry);
3925 fsnotify_unlink(dir, dentry);
3930 inode_unlock(target);
3932 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3933 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3934 fsnotify_link_count(target);
3940 EXPORT_SYMBOL(vfs_unlink);
3943 * Make sure that the actual truncation of the file will occur outside its
3944 * directory's i_mutex. Truncate can take a long time if there is a lot of
3945 * writeout happening, and we don't want to prevent access to the directory
3946 * while waiting on the I/O.
3948 long do_unlinkat(int dfd, struct filename *name)
3951 struct dentry *dentry;
3955 struct inode *inode = NULL;
3956 struct inode *delegated_inode = NULL;
3957 unsigned int lookup_flags = 0;
3959 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
3961 return PTR_ERR(name);
3964 if (type != LAST_NORM)
3967 error = mnt_want_write(path.mnt);
3971 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3972 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3973 error = PTR_ERR(dentry);
3974 if (!IS_ERR(dentry)) {
3975 /* Why not before? Because we want correct error value */
3976 if (last.name[last.len])
3978 inode = dentry->d_inode;
3979 if (d_is_negative(dentry))
3982 error = security_path_unlink(&path, dentry);
3985 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3989 inode_unlock(path.dentry->d_inode);
3991 iput(inode); /* truncate the inode here */
3993 if (delegated_inode) {
3994 error = break_deleg_wait(&delegated_inode);
3998 mnt_drop_write(path.mnt);
4001 if (retry_estale(error, lookup_flags)) {
4002 lookup_flags |= LOOKUP_REVAL;
4010 if (d_is_negative(dentry))
4012 else if (d_is_dir(dentry))
4019 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4021 if ((flag & ~AT_REMOVEDIR) != 0)
4024 if (flag & AT_REMOVEDIR)
4025 return do_rmdir(dfd, pathname);
4027 return do_unlinkat(dfd, getname(pathname));
4030 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4032 return do_unlinkat(AT_FDCWD, getname(pathname));
4035 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4037 int error = may_create(dir, dentry);
4042 if (!dir->i_op->symlink)
4045 error = security_inode_symlink(dir, dentry, oldname);
4049 error = dir->i_op->symlink(dir, dentry, oldname);
4051 fsnotify_create(dir, dentry);
4054 EXPORT_SYMBOL(vfs_symlink);
4056 long do_symlinkat(const char __user *oldname, int newdfd,
4057 const char __user *newname)
4060 struct filename *from;
4061 struct dentry *dentry;
4063 unsigned int lookup_flags = 0;
4065 from = getname(oldname);
4067 return PTR_ERR(from);
4069 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4070 error = PTR_ERR(dentry);
4074 error = security_path_symlink(&path, dentry, from->name);
4076 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4077 done_path_create(&path, dentry);
4078 if (retry_estale(error, lookup_flags)) {
4079 lookup_flags |= LOOKUP_REVAL;
4087 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4088 int, newdfd, const char __user *, newname)
4090 return do_symlinkat(oldname, newdfd, newname);
4093 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4095 return do_symlinkat(oldname, AT_FDCWD, newname);
4099 * vfs_link - create a new link
4100 * @old_dentry: object to be linked
4102 * @new_dentry: where to create the new link
4103 * @delegated_inode: returns inode needing a delegation break
4105 * The caller must hold dir->i_mutex
4107 * If vfs_link discovers a delegation on the to-be-linked file in need
4108 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4109 * inode in delegated_inode. The caller should then break the delegation
4110 * and retry. Because breaking a delegation may take a long time, the
4111 * caller should drop the i_mutex before doing so.
4113 * Alternatively, a caller may pass NULL for delegated_inode. This may
4114 * be appropriate for callers that expect the underlying filesystem not
4115 * to be NFS exported.
4117 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4119 struct inode *inode = old_dentry->d_inode;
4120 unsigned max_links = dir->i_sb->s_max_links;
4126 error = may_create(dir, new_dentry);
4130 if (dir->i_sb != inode->i_sb)
4134 * A link to an append-only or immutable file cannot be created.
4136 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4139 * Updating the link count will likely cause i_uid and i_gid to
4140 * be writen back improperly if their true value is unknown to
4143 if (HAS_UNMAPPED_ID(inode))
4145 if (!dir->i_op->link)
4147 if (S_ISDIR(inode->i_mode))
4150 error = security_inode_link(old_dentry, dir, new_dentry);
4155 /* Make sure we don't allow creating hardlink to an unlinked file */
4156 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4158 else if (max_links && inode->i_nlink >= max_links)
4161 error = try_break_deleg(inode, delegated_inode);
4163 error = dir->i_op->link(old_dentry, dir, new_dentry);
4166 if (!error && (inode->i_state & I_LINKABLE)) {
4167 spin_lock(&inode->i_lock);
4168 inode->i_state &= ~I_LINKABLE;
4169 spin_unlock(&inode->i_lock);
4171 inode_unlock(inode);
4173 fsnotify_link(dir, inode, new_dentry);
4176 EXPORT_SYMBOL(vfs_link);
4179 * Hardlinks are often used in delicate situations. We avoid
4180 * security-related surprises by not following symlinks on the
4183 * We don't follow them on the oldname either to be compatible
4184 * with linux 2.0, and to avoid hard-linking to directories
4185 * and other special files. --ADM
4187 int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4188 const char __user *newname, int flags)
4190 struct dentry *new_dentry;
4191 struct path old_path, new_path;
4192 struct inode *delegated_inode = NULL;
4196 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4199 * To use null names we require CAP_DAC_READ_SEARCH
4200 * This ensures that not everyone will be able to create
4201 * handlink using the passed filedescriptor.
4203 if (flags & AT_EMPTY_PATH) {
4204 if (!capable(CAP_DAC_READ_SEARCH))
4209 if (flags & AT_SYMLINK_FOLLOW)
4210 how |= LOOKUP_FOLLOW;
4212 error = user_path_at(olddfd, oldname, how, &old_path);
4216 new_dentry = user_path_create(newdfd, newname, &new_path,
4217 (how & LOOKUP_REVAL));
4218 error = PTR_ERR(new_dentry);
4219 if (IS_ERR(new_dentry))
4223 if (old_path.mnt != new_path.mnt)
4225 error = may_linkat(&old_path);
4226 if (unlikely(error))
4228 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4231 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4233 done_path_create(&new_path, new_dentry);
4234 if (delegated_inode) {
4235 error = break_deleg_wait(&delegated_inode);
4237 path_put(&old_path);
4241 if (retry_estale(error, how)) {
4242 path_put(&old_path);
4243 how |= LOOKUP_REVAL;
4247 path_put(&old_path);
4252 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4253 int, newdfd, const char __user *, newname, int, flags)
4255 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4258 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4260 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4264 * vfs_rename - rename a filesystem object
4265 * @old_dir: parent of source
4266 * @old_dentry: source
4267 * @new_dir: parent of destination
4268 * @new_dentry: destination
4269 * @delegated_inode: returns an inode needing a delegation break
4270 * @flags: rename flags
4272 * The caller must hold multiple mutexes--see lock_rename()).
4274 * If vfs_rename discovers a delegation in need of breaking at either
4275 * the source or destination, it will return -EWOULDBLOCK and return a
4276 * reference to the inode in delegated_inode. The caller should then
4277 * break the delegation and retry. Because breaking a delegation may
4278 * take a long time, the caller should drop all locks before doing
4281 * Alternatively, a caller may pass NULL for delegated_inode. This may
4282 * be appropriate for callers that expect the underlying filesystem not
4283 * to be NFS exported.
4285 * The worst of all namespace operations - renaming directory. "Perverted"
4286 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4289 * a) we can get into loop creation.
4290 * b) race potential - two innocent renames can create a loop together.
4291 * That's where 4.4 screws up. Current fix: serialization on
4292 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4294 * c) we have to lock _four_ objects - parents and victim (if it exists),
4295 * and source (if it is not a directory).
4296 * And that - after we got ->i_mutex on parents (until then we don't know
4297 * whether the target exists). Solution: try to be smart with locking
4298 * order for inodes. We rely on the fact that tree topology may change
4299 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4300 * move will be locked. Thus we can rank directories by the tree
4301 * (ancestors first) and rank all non-directories after them.
4302 * That works since everybody except rename does "lock parent, lookup,
4303 * lock child" and rename is under ->s_vfs_rename_mutex.
4304 * HOWEVER, it relies on the assumption that any object with ->lookup()
4305 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4306 * we'd better make sure that there's no link(2) for them.
4307 * d) conversion from fhandle to dentry may come in the wrong moment - when
4308 * we are removing the target. Solution: we will have to grab ->i_mutex
4309 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4310 * ->i_mutex on parents, which works but leads to some truly excessive
4313 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4314 struct inode *new_dir, struct dentry *new_dentry,
4315 struct inode **delegated_inode, unsigned int flags)
4318 bool is_dir = d_is_dir(old_dentry);
4319 struct inode *source = old_dentry->d_inode;
4320 struct inode *target = new_dentry->d_inode;
4321 bool new_is_dir = false;
4322 unsigned max_links = new_dir->i_sb->s_max_links;
4323 struct name_snapshot old_name;
4325 if (source == target)
4328 error = may_delete(old_dir, old_dentry, is_dir);
4333 error = may_create(new_dir, new_dentry);
4335 new_is_dir = d_is_dir(new_dentry);
4337 if (!(flags & RENAME_EXCHANGE))
4338 error = may_delete(new_dir, new_dentry, is_dir);
4340 error = may_delete(new_dir, new_dentry, new_is_dir);
4345 if (!old_dir->i_op->rename)
4349 * If we are going to change the parent - check write permissions,
4350 * we'll need to flip '..'.
4352 if (new_dir != old_dir) {
4354 error = inode_permission(source, MAY_WRITE);
4358 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4359 error = inode_permission(target, MAY_WRITE);
4365 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4370 take_dentry_name_snapshot(&old_name, old_dentry);
4372 if (!is_dir || (flags & RENAME_EXCHANGE))
4373 lock_two_nondirectories(source, target);
4378 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4381 if (max_links && new_dir != old_dir) {
4383 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4385 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4386 old_dir->i_nlink >= max_links)
4390 error = try_break_deleg(source, delegated_inode);
4394 if (target && !new_is_dir) {
4395 error = try_break_deleg(target, delegated_inode);
4399 error = old_dir->i_op->rename(old_dir, old_dentry,
4400 new_dir, new_dentry, flags);
4404 if (!(flags & RENAME_EXCHANGE) && target) {
4406 shrink_dcache_parent(new_dentry);
4407 target->i_flags |= S_DEAD;
4409 dont_mount(new_dentry);
4410 detach_mounts(new_dentry);
4412 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4413 if (!(flags & RENAME_EXCHANGE))
4414 d_move(old_dentry, new_dentry);
4416 d_exchange(old_dentry, new_dentry);
4419 if (!is_dir || (flags & RENAME_EXCHANGE))
4420 unlock_two_nondirectories(source, target);
4422 inode_unlock(target);
4425 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4426 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4427 if (flags & RENAME_EXCHANGE) {
4428 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4429 new_is_dir, NULL, new_dentry);
4432 release_dentry_name_snapshot(&old_name);
4436 EXPORT_SYMBOL(vfs_rename);
4438 static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4439 const char __user *newname, unsigned int flags)
4441 struct dentry *old_dentry, *new_dentry;
4442 struct dentry *trap;
4443 struct path old_path, new_path;
4444 struct qstr old_last, new_last;
4445 int old_type, new_type;
4446 struct inode *delegated_inode = NULL;
4447 struct filename *from;
4448 struct filename *to;
4449 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4450 bool should_retry = false;
4453 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4456 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4457 (flags & RENAME_EXCHANGE))
4460 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4463 if (flags & RENAME_EXCHANGE)
4467 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4468 &old_path, &old_last, &old_type);
4470 error = PTR_ERR(from);
4474 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4475 &new_path, &new_last, &new_type);
4477 error = PTR_ERR(to);
4482 if (old_path.mnt != new_path.mnt)
4486 if (old_type != LAST_NORM)
4489 if (flags & RENAME_NOREPLACE)
4491 if (new_type != LAST_NORM)
4494 error = mnt_want_write(old_path.mnt);
4499 trap = lock_rename(new_path.dentry, old_path.dentry);
4501 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4502 error = PTR_ERR(old_dentry);
4503 if (IS_ERR(old_dentry))
4505 /* source must exist */
4507 if (d_is_negative(old_dentry))
4509 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4510 error = PTR_ERR(new_dentry);
4511 if (IS_ERR(new_dentry))
4514 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4516 if (flags & RENAME_EXCHANGE) {
4518 if (d_is_negative(new_dentry))
4521 if (!d_is_dir(new_dentry)) {
4523 if (new_last.name[new_last.len])
4527 /* unless the source is a directory trailing slashes give -ENOTDIR */
4528 if (!d_is_dir(old_dentry)) {
4530 if (old_last.name[old_last.len])
4532 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4535 /* source should not be ancestor of target */
4537 if (old_dentry == trap)
4539 /* target should not be an ancestor of source */
4540 if (!(flags & RENAME_EXCHANGE))
4542 if (new_dentry == trap)
4545 error = security_path_rename(&old_path, old_dentry,
4546 &new_path, new_dentry, flags);
4549 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4550 new_path.dentry->d_inode, new_dentry,
4551 &delegated_inode, flags);
4557 unlock_rename(new_path.dentry, old_path.dentry);
4558 if (delegated_inode) {
4559 error = break_deleg_wait(&delegated_inode);
4563 mnt_drop_write(old_path.mnt);
4565 if (retry_estale(error, lookup_flags))
4566 should_retry = true;
4567 path_put(&new_path);
4570 path_put(&old_path);
4573 should_retry = false;
4574 lookup_flags |= LOOKUP_REVAL;
4581 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4582 int, newdfd, const char __user *, newname, unsigned int, flags)
4584 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4587 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4588 int, newdfd, const char __user *, newname)
4590 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
4593 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4595 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4598 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4600 int error = may_create(dir, dentry);
4604 if (!dir->i_op->mknod)
4607 return dir->i_op->mknod(dir, dentry,
4608 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4610 EXPORT_SYMBOL(vfs_whiteout);
4612 int readlink_copy(char __user *buffer, int buflen, const char *link)
4614 int len = PTR_ERR(link);
4619 if (len > (unsigned) buflen)
4621 if (copy_to_user(buffer, link, len))
4628 * vfs_readlink - copy symlink body into userspace buffer
4629 * @dentry: dentry on which to get symbolic link
4630 * @buffer: user memory pointer
4631 * @buflen: size of buffer
4633 * Does not touch atime. That's up to the caller if necessary
4635 * Does not call security hook.
4637 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4639 struct inode *inode = d_inode(dentry);
4640 DEFINE_DELAYED_CALL(done);
4644 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4645 if (unlikely(inode->i_op->readlink))
4646 return inode->i_op->readlink(dentry, buffer, buflen);
4648 if (!d_is_symlink(dentry))
4651 spin_lock(&inode->i_lock);
4652 inode->i_opflags |= IOP_DEFAULT_READLINK;
4653 spin_unlock(&inode->i_lock);
4656 link = READ_ONCE(inode->i_link);
4658 link = inode->i_op->get_link(dentry, inode, &done);
4660 return PTR_ERR(link);
4662 res = readlink_copy(buffer, buflen, link);
4663 do_delayed_call(&done);
4666 EXPORT_SYMBOL(vfs_readlink);
4669 * vfs_get_link - get symlink body
4670 * @dentry: dentry on which to get symbolic link
4671 * @done: caller needs to free returned data with this
4673 * Calls security hook and i_op->get_link() on the supplied inode.
4675 * It does not touch atime. That's up to the caller if necessary.
4677 * Does not work on "special" symlinks like /proc/$$/fd/N
4679 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4681 const char *res = ERR_PTR(-EINVAL);
4682 struct inode *inode = d_inode(dentry);
4684 if (d_is_symlink(dentry)) {
4685 res = ERR_PTR(security_inode_readlink(dentry));
4687 res = inode->i_op->get_link(dentry, inode, done);
4691 EXPORT_SYMBOL(vfs_get_link);
4693 /* get the link contents into pagecache */
4694 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4695 struct delayed_call *callback)
4699 struct address_space *mapping = inode->i_mapping;
4702 page = find_get_page(mapping, 0);
4704 return ERR_PTR(-ECHILD);
4705 if (!PageUptodate(page)) {
4707 return ERR_PTR(-ECHILD);
4710 page = read_mapping_page(mapping, 0, NULL);
4714 set_delayed_call(callback, page_put_link, page);
4715 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4716 kaddr = page_address(page);
4717 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4721 EXPORT_SYMBOL(page_get_link);
4723 void page_put_link(void *arg)
4727 EXPORT_SYMBOL(page_put_link);
4729 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4731 DEFINE_DELAYED_CALL(done);
4732 int res = readlink_copy(buffer, buflen,
4733 page_get_link(dentry, d_inode(dentry),
4735 do_delayed_call(&done);
4738 EXPORT_SYMBOL(page_readlink);
4741 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4743 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4745 struct address_space *mapping = inode->i_mapping;
4749 unsigned int flags = 0;
4751 flags |= AOP_FLAG_NOFS;
4754 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4755 flags, &page, &fsdata);
4759 memcpy(page_address(page), symname, len-1);
4761 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4768 mark_inode_dirty(inode);
4773 EXPORT_SYMBOL(__page_symlink);
4775 int page_symlink(struct inode *inode, const char *symname, int len)
4777 return __page_symlink(inode, symname, len,
4778 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4780 EXPORT_SYMBOL(page_symlink);
4782 const struct inode_operations page_symlink_inode_operations = {
4783 .get_link = page_get_link,
4785 EXPORT_SYMBOL(page_symlink_inode_operations);