4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <asm/uaccess.h>
42 /* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
90 * [10-Sep-98 Alan Modra] Another symlink change.
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 static char *getname_flags(const char __user *filename, int flags, int *empty)
122 char *result = __getname(), *err;
125 if (unlikely(!result))
126 return ERR_PTR(-ENOMEM);
128 len = strncpy_from_user(result, filename, PATH_MAX);
130 if (unlikely(len < 0))
133 /* The empty path is special. */
134 if (unlikely(!len)) {
137 err = ERR_PTR(-ENOENT);
138 if (!(flags & LOOKUP_EMPTY))
142 err = ERR_PTR(-ENAMETOOLONG);
143 if (likely(len < PATH_MAX)) {
144 audit_getname(result);
153 char *getname(const char __user * filename)
155 return getname_flags(filename, 0, NULL);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname);
169 static int check_acl(struct inode *inode, int mask)
171 #ifdef CONFIG_FS_POSIX_ACL
172 struct posix_acl *acl;
174 if (mask & MAY_NOT_BLOCK) {
175 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
178 /* no ->get_acl() calls in RCU mode... */
179 if (acl == ACL_NOT_CACHED)
181 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
184 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
187 * A filesystem can force a ACL callback by just never filling the
188 * ACL cache. But normally you'd fill the cache either at inode
189 * instantiation time, or on the first ->get_acl call.
191 * If the filesystem doesn't have a get_acl() function at all, we'll
192 * just create the negative cache entry.
194 if (acl == ACL_NOT_CACHED) {
195 if (inode->i_op->get_acl) {
196 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
200 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
206 int error = posix_acl_permission(inode, acl, mask);
207 posix_acl_release(acl);
216 * This does the basic permission checking
218 static int acl_permission_check(struct inode *inode, int mask)
220 unsigned int mode = inode->i_mode;
222 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
225 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
226 int error = check_acl(inode, mask);
227 if (error != -EAGAIN)
231 if (in_group_p(inode->i_gid))
236 * If the DACs are ok we don't need any capability check.
238 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
244 * generic_permission - check for access rights on a Posix-like filesystem
245 * @inode: inode to check access rights for
246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
248 * Used to check for read/write/execute permissions on a file.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
251 * are used for other things.
253 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
254 * request cannot be satisfied (eg. requires blocking or too much complexity).
255 * It would then be called again in ref-walk mode.
257 int generic_permission(struct inode *inode, int mask)
262 * Do the basic permission checks.
264 ret = acl_permission_check(inode, mask);
268 if (S_ISDIR(inode->i_mode)) {
269 /* DACs are overridable for directories */
270 if (inode_capable(inode, CAP_DAC_OVERRIDE))
272 if (!(mask & MAY_WRITE))
273 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
278 * Read/write DACs are always overridable.
279 * Executable DACs are overridable when there is
280 * at least one exec bit set.
282 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
283 if (inode_capable(inode, CAP_DAC_OVERRIDE))
287 * Searching includes executable on directories, else just read.
289 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
290 if (mask == MAY_READ)
291 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
298 * We _really_ want to just do "generic_permission()" without
299 * even looking at the inode->i_op values. So we keep a cache
300 * flag in inode->i_opflags, that says "this has not special
301 * permission function, use the fast case".
303 static inline int do_inode_permission(struct inode *inode, int mask)
305 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
306 if (likely(inode->i_op->permission))
307 return inode->i_op->permission(inode, mask);
309 /* This gets set once for the inode lifetime */
310 spin_lock(&inode->i_lock);
311 inode->i_opflags |= IOP_FASTPERM;
312 spin_unlock(&inode->i_lock);
314 return generic_permission(inode, mask);
318 * inode_permission - check for access rights to a given inode
319 * @inode: inode to check permission on
320 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
322 * Used to check for read/write/execute permissions on an inode.
323 * We use "fsuid" for this, letting us set arbitrary permissions
324 * for filesystem access without changing the "normal" uids which
325 * are used for other things.
327 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
329 int inode_permission(struct inode *inode, int mask)
333 if (unlikely(mask & MAY_WRITE)) {
334 umode_t mode = inode->i_mode;
337 * Nobody gets write access to a read-only fs.
339 if (IS_RDONLY(inode) &&
340 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
344 * Nobody gets write access to an immutable file.
346 if (IS_IMMUTABLE(inode))
350 retval = do_inode_permission(inode, mask);
354 retval = devcgroup_inode_permission(inode, mask);
358 return security_inode_permission(inode, mask);
362 * path_get - get a reference to a path
363 * @path: path to get the reference to
365 * Given a path increment the reference count to the dentry and the vfsmount.
367 void path_get(struct path *path)
372 EXPORT_SYMBOL(path_get);
375 * path_put - put a reference to a path
376 * @path: path to put the reference to
378 * Given a path decrement the reference count to the dentry and the vfsmount.
380 void path_put(struct path *path)
385 EXPORT_SYMBOL(path_put);
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). In situations when we can't
390 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
391 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
392 * mode. Refcounts are grabbed at the last known good point before rcu-walk
393 * got stuck, so ref-walk may continue from there. If this is not successful
394 * (eg. a seqcount has changed), then failure is returned and it's up to caller
395 * to restart the path walk from the beginning in ref-walk mode.
399 * unlazy_walk - try to switch to ref-walk mode.
400 * @nd: nameidata pathwalk data
401 * @dentry: child of nd->path.dentry or NULL
402 * Returns: 0 on success, -ECHILD on failure
404 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
405 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
406 * @nd or NULL. Must be called from rcu-walk context.
408 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
410 struct fs_struct *fs = current->fs;
411 struct dentry *parent = nd->path.dentry;
414 BUG_ON(!(nd->flags & LOOKUP_RCU));
415 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
417 spin_lock(&fs->lock);
418 if (nd->root.mnt != fs->root.mnt ||
419 nd->root.dentry != fs->root.dentry)
422 spin_lock(&parent->d_lock);
424 if (!__d_rcu_to_refcount(parent, nd->seq))
426 BUG_ON(nd->inode != parent->d_inode);
428 if (dentry->d_parent != parent)
430 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
431 if (!__d_rcu_to_refcount(dentry, nd->seq))
434 * If the sequence check on the child dentry passed, then
435 * the child has not been removed from its parent. This
436 * means the parent dentry must be valid and able to take
437 * a reference at this point.
439 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
440 BUG_ON(!parent->d_count);
442 spin_unlock(&dentry->d_lock);
444 spin_unlock(&parent->d_lock);
447 spin_unlock(&fs->lock);
449 mntget(nd->path.mnt);
452 br_read_unlock(&vfsmount_lock);
453 nd->flags &= ~LOOKUP_RCU;
457 spin_unlock(&dentry->d_lock);
459 spin_unlock(&parent->d_lock);
462 spin_unlock(&fs->lock);
467 * release_open_intent - free up open intent resources
468 * @nd: pointer to nameidata
470 void release_open_intent(struct nameidata *nd)
472 struct file *file = nd->intent.open.file;
474 if (file && !IS_ERR(file)) {
475 if (file->f_path.dentry == NULL)
482 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
484 return dentry->d_op->d_revalidate(dentry, nd);
488 * complete_walk - successful completion of path walk
489 * @nd: pointer nameidata
491 * If we had been in RCU mode, drop out of it and legitimize nd->path.
492 * Revalidate the final result, unless we'd already done that during
493 * the path walk or the filesystem doesn't ask for it. Return 0 on
494 * success, -error on failure. In case of failure caller does not
495 * need to drop nd->path.
497 static int complete_walk(struct nameidata *nd)
499 struct dentry *dentry = nd->path.dentry;
502 if (nd->flags & LOOKUP_RCU) {
503 nd->flags &= ~LOOKUP_RCU;
504 if (!(nd->flags & LOOKUP_ROOT))
506 spin_lock(&dentry->d_lock);
507 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
508 spin_unlock(&dentry->d_lock);
510 br_read_unlock(&vfsmount_lock);
513 BUG_ON(nd->inode != dentry->d_inode);
514 spin_unlock(&dentry->d_lock);
515 mntget(nd->path.mnt);
517 br_read_unlock(&vfsmount_lock);
520 if (likely(!(nd->flags & LOOKUP_JUMPED)))
523 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
526 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
529 /* Note: we do not d_invalidate() */
530 status = d_revalidate(dentry, nd);
541 static __always_inline void set_root(struct nameidata *nd)
544 get_fs_root(current->fs, &nd->root);
547 static int link_path_walk(const char *, struct nameidata *);
549 static __always_inline void set_root_rcu(struct nameidata *nd)
552 struct fs_struct *fs = current->fs;
556 seq = read_seqcount_begin(&fs->seq);
558 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
559 } while (read_seqcount_retry(&fs->seq, seq));
563 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
575 nd->flags |= LOOKUP_JUMPED;
577 nd->inode = nd->path.dentry->d_inode;
579 ret = link_path_walk(link, nd);
583 return PTR_ERR(link);
586 static void path_put_conditional(struct path *path, struct nameidata *nd)
589 if (path->mnt != nd->path.mnt)
593 static inline void path_to_nameidata(const struct path *path,
594 struct nameidata *nd)
596 if (!(nd->flags & LOOKUP_RCU)) {
597 dput(nd->path.dentry);
598 if (nd->path.mnt != path->mnt)
599 mntput(nd->path.mnt);
601 nd->path.mnt = path->mnt;
602 nd->path.dentry = path->dentry;
605 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
607 struct inode *inode = link->dentry->d_inode;
608 if (!IS_ERR(cookie) && inode->i_op->put_link)
609 inode->i_op->put_link(link->dentry, nd, cookie);
613 static __always_inline int
614 follow_link(struct path *link, struct nameidata *nd, void **p)
617 struct dentry *dentry = link->dentry;
619 BUG_ON(nd->flags & LOOKUP_RCU);
621 if (link->mnt == nd->path.mnt)
624 if (unlikely(current->total_link_count >= 40)) {
625 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
630 current->total_link_count++;
633 nd_set_link(nd, NULL);
635 error = security_inode_follow_link(link->dentry, nd);
637 *p = ERR_PTR(error); /* no ->put_link(), please */
642 nd->last_type = LAST_BIND;
643 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
646 char *s = nd_get_link(nd);
649 error = __vfs_follow_link(nd, s);
650 else if (nd->last_type == LAST_BIND) {
651 nd->flags |= LOOKUP_JUMPED;
652 nd->inode = nd->path.dentry->d_inode;
653 if (nd->inode->i_op->follow_link) {
654 /* stepped on a _really_ weird one */
663 static int follow_up_rcu(struct path *path)
665 struct mount *mnt = real_mount(path->mnt);
666 struct mount *parent;
667 struct dentry *mountpoint;
669 parent = mnt->mnt_parent;
670 if (&parent->mnt == path->mnt)
672 mountpoint = mnt->mnt_mountpoint;
673 path->dentry = mountpoint;
674 path->mnt = &parent->mnt;
678 int follow_up(struct path *path)
680 struct mount *mnt = real_mount(path->mnt);
681 struct mount *parent;
682 struct dentry *mountpoint;
684 br_read_lock(&vfsmount_lock);
685 parent = mnt->mnt_parent;
686 if (&parent->mnt == path->mnt) {
687 br_read_unlock(&vfsmount_lock);
690 mntget(&parent->mnt);
691 mountpoint = dget(mnt->mnt_mountpoint);
692 br_read_unlock(&vfsmount_lock);
694 path->dentry = mountpoint;
696 path->mnt = &parent->mnt;
701 * Perform an automount
702 * - return -EISDIR to tell follow_managed() to stop and return the path we
705 static int follow_automount(struct path *path, unsigned flags,
708 struct vfsmount *mnt;
711 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
714 /* We don't want to mount if someone's just doing a stat -
715 * unless they're stat'ing a directory and appended a '/' to
718 * We do, however, want to mount if someone wants to open or
719 * create a file of any type under the mountpoint, wants to
720 * traverse through the mountpoint or wants to open the
721 * mounted directory. Also, autofs may mark negative dentries
722 * as being automount points. These will need the attentions
723 * of the daemon to instantiate them before they can be used.
725 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
726 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
727 path->dentry->d_inode)
730 current->total_link_count++;
731 if (current->total_link_count >= 40)
734 mnt = path->dentry->d_op->d_automount(path);
737 * The filesystem is allowed to return -EISDIR here to indicate
738 * it doesn't want to automount. For instance, autofs would do
739 * this so that its userspace daemon can mount on this dentry.
741 * However, we can only permit this if it's a terminal point in
742 * the path being looked up; if it wasn't then the remainder of
743 * the path is inaccessible and we should say so.
745 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
750 if (!mnt) /* mount collision */
754 /* lock_mount() may release path->mnt on error */
758 err = finish_automount(mnt, path);
762 /* Someone else made a mount here whilst we were busy */
767 path->dentry = dget(mnt->mnt_root);
776 * Handle a dentry that is managed in some way.
777 * - Flagged for transit management (autofs)
778 * - Flagged as mountpoint
779 * - Flagged as automount point
781 * This may only be called in refwalk mode.
783 * Serialization is taken care of in namespace.c
785 static int follow_managed(struct path *path, unsigned flags)
787 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
789 bool need_mntput = false;
792 /* Given that we're not holding a lock here, we retain the value in a
793 * local variable for each dentry as we look at it so that we don't see
794 * the components of that value change under us */
795 while (managed = ACCESS_ONCE(path->dentry->d_flags),
796 managed &= DCACHE_MANAGED_DENTRY,
797 unlikely(managed != 0)) {
798 /* Allow the filesystem to manage the transit without i_mutex
800 if (managed & DCACHE_MANAGE_TRANSIT) {
801 BUG_ON(!path->dentry->d_op);
802 BUG_ON(!path->dentry->d_op->d_manage);
803 ret = path->dentry->d_op->d_manage(path->dentry, false);
808 /* Transit to a mounted filesystem. */
809 if (managed & DCACHE_MOUNTED) {
810 struct vfsmount *mounted = lookup_mnt(path);
816 path->dentry = dget(mounted->mnt_root);
821 /* Something is mounted on this dentry in another
822 * namespace and/or whatever was mounted there in this
823 * namespace got unmounted before we managed to get the
827 /* Handle an automount point */
828 if (managed & DCACHE_NEED_AUTOMOUNT) {
829 ret = follow_automount(path, flags, &need_mntput);
835 /* We didn't change the current path point */
839 if (need_mntput && path->mnt == mnt)
843 return ret < 0 ? ret : need_mntput;
846 int follow_down_one(struct path *path)
848 struct vfsmount *mounted;
850 mounted = lookup_mnt(path);
855 path->dentry = dget(mounted->mnt_root);
861 static inline bool managed_dentry_might_block(struct dentry *dentry)
863 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
864 dentry->d_op->d_manage(dentry, true) < 0);
868 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
869 * we meet a managed dentry that would need blocking.
871 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
872 struct inode **inode)
875 struct mount *mounted;
877 * Don't forget we might have a non-mountpoint managed dentry
878 * that wants to block transit.
880 if (unlikely(managed_dentry_might_block(path->dentry)))
883 if (!d_mountpoint(path->dentry))
886 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
889 path->mnt = &mounted->mnt;
890 path->dentry = mounted->mnt.mnt_root;
891 nd->flags |= LOOKUP_JUMPED;
892 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
894 * Update the inode too. We don't need to re-check the
895 * dentry sequence number here after this d_inode read,
896 * because a mount-point is always pinned.
898 *inode = path->dentry->d_inode;
903 static void follow_mount_rcu(struct nameidata *nd)
905 while (d_mountpoint(nd->path.dentry)) {
906 struct mount *mounted;
907 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
910 nd->path.mnt = &mounted->mnt;
911 nd->path.dentry = mounted->mnt.mnt_root;
912 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
916 static int follow_dotdot_rcu(struct nameidata *nd)
921 if (nd->path.dentry == nd->root.dentry &&
922 nd->path.mnt == nd->root.mnt) {
925 if (nd->path.dentry != nd->path.mnt->mnt_root) {
926 struct dentry *old = nd->path.dentry;
927 struct dentry *parent = old->d_parent;
930 seq = read_seqcount_begin(&parent->d_seq);
931 if (read_seqcount_retry(&old->d_seq, nd->seq))
933 nd->path.dentry = parent;
937 if (!follow_up_rcu(&nd->path))
939 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
941 follow_mount_rcu(nd);
942 nd->inode = nd->path.dentry->d_inode;
946 nd->flags &= ~LOOKUP_RCU;
947 if (!(nd->flags & LOOKUP_ROOT))
950 br_read_unlock(&vfsmount_lock);
955 * Follow down to the covering mount currently visible to userspace. At each
956 * point, the filesystem owning that dentry may be queried as to whether the
957 * caller is permitted to proceed or not.
959 int follow_down(struct path *path)
964 while (managed = ACCESS_ONCE(path->dentry->d_flags),
965 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
966 /* Allow the filesystem to manage the transit without i_mutex
969 * We indicate to the filesystem if someone is trying to mount
970 * something here. This gives autofs the chance to deny anyone
971 * other than its daemon the right to mount on its
974 * The filesystem may sleep at this point.
976 if (managed & DCACHE_MANAGE_TRANSIT) {
977 BUG_ON(!path->dentry->d_op);
978 BUG_ON(!path->dentry->d_op->d_manage);
979 ret = path->dentry->d_op->d_manage(
980 path->dentry, false);
982 return ret == -EISDIR ? 0 : ret;
985 /* Transit to a mounted filesystem. */
986 if (managed & DCACHE_MOUNTED) {
987 struct vfsmount *mounted = lookup_mnt(path);
993 path->dentry = dget(mounted->mnt_root);
997 /* Don't handle automount points here */
1004 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1006 static void follow_mount(struct path *path)
1008 while (d_mountpoint(path->dentry)) {
1009 struct vfsmount *mounted = lookup_mnt(path);
1014 path->mnt = mounted;
1015 path->dentry = dget(mounted->mnt_root);
1019 static void follow_dotdot(struct nameidata *nd)
1024 struct dentry *old = nd->path.dentry;
1026 if (nd->path.dentry == nd->root.dentry &&
1027 nd->path.mnt == nd->root.mnt) {
1030 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1031 /* rare case of legitimate dget_parent()... */
1032 nd->path.dentry = dget_parent(nd->path.dentry);
1036 if (!follow_up(&nd->path))
1039 follow_mount(&nd->path);
1040 nd->inode = nd->path.dentry->d_inode;
1044 * This looks up the name in dcache, possibly revalidates the old dentry and
1045 * allocates a new one if not found or not valid. In the need_lookup argument
1046 * returns whether i_op->lookup is necessary.
1048 * dir->d_inode->i_mutex must be held
1050 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1051 struct nameidata *nd, bool *need_lookup)
1053 struct dentry *dentry;
1056 *need_lookup = false;
1057 dentry = d_lookup(dir, name);
1059 if (d_need_lookup(dentry)) {
1060 *need_lookup = true;
1061 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1062 error = d_revalidate(dentry, nd);
1063 if (unlikely(error <= 0)) {
1066 return ERR_PTR(error);
1067 } else if (!d_invalidate(dentry)) {
1076 dentry = d_alloc(dir, name);
1077 if (unlikely(!dentry))
1078 return ERR_PTR(-ENOMEM);
1080 *need_lookup = true;
1086 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1087 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1089 * dir->d_inode->i_mutex must be held
1091 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1092 struct nameidata *nd)
1096 /* Don't create child dentry for a dead directory. */
1097 if (unlikely(IS_DEADDIR(dir))) {
1099 return ERR_PTR(-ENOENT);
1102 old = dir->i_op->lookup(dir, dentry, nd);
1103 if (unlikely(old)) {
1110 static struct dentry *__lookup_hash(struct qstr *name,
1111 struct dentry *base, struct nameidata *nd)
1114 struct dentry *dentry;
1116 dentry = lookup_dcache(name, base, nd, &need_lookup);
1120 return lookup_real(base->d_inode, dentry, nd);
1124 * It's more convoluted than I'd like it to be, but... it's still fairly
1125 * small and for now I'd prefer to have fast path as straight as possible.
1126 * It _is_ time-critical.
1128 static int lookup_fast(struct nameidata *nd, struct qstr *name,
1129 struct path *path, struct inode **inode)
1131 struct vfsmount *mnt = nd->path.mnt;
1132 struct dentry *dentry, *parent = nd->path.dentry;
1138 * Rename seqlock is not required here because in the off chance
1139 * of a false negative due to a concurrent rename, we're going to
1140 * do the non-racy lookup, below.
1142 if (nd->flags & LOOKUP_RCU) {
1144 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
1149 * This sequence count validates that the inode matches
1150 * the dentry name information from lookup.
1152 *inode = dentry->d_inode;
1153 if (read_seqcount_retry(&dentry->d_seq, seq))
1157 * This sequence count validates that the parent had no
1158 * changes while we did the lookup of the dentry above.
1160 * The memory barrier in read_seqcount_begin of child is
1161 * enough, we can use __read_seqcount_retry here.
1163 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1167 if (unlikely(d_need_lookup(dentry)))
1169 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1170 status = d_revalidate(dentry, nd);
1171 if (unlikely(status <= 0)) {
1172 if (status != -ECHILD)
1178 path->dentry = dentry;
1179 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1181 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1185 if (unlazy_walk(nd, dentry))
1188 dentry = __d_lookup(parent, name);
1191 if (unlikely(!dentry))
1194 if (unlikely(d_need_lookup(dentry))) {
1199 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1200 status = d_revalidate(dentry, nd);
1201 if (unlikely(status <= 0)) {
1206 if (!d_invalidate(dentry)) {
1213 path->dentry = dentry;
1214 err = follow_managed(path, nd->flags);
1215 if (unlikely(err < 0)) {
1216 path_put_conditional(path, nd);
1220 nd->flags |= LOOKUP_JUMPED;
1221 *inode = path->dentry->d_inode;
1228 /* Fast lookup failed, do it the slow way */
1229 static int lookup_slow(struct nameidata *nd, struct qstr *name,
1232 struct dentry *dentry, *parent;
1235 parent = nd->path.dentry;
1236 BUG_ON(nd->inode != parent->d_inode);
1238 mutex_lock(&parent->d_inode->i_mutex);
1239 dentry = __lookup_hash(name, parent, nd);
1240 mutex_unlock(&parent->d_inode->i_mutex);
1242 return PTR_ERR(dentry);
1243 path->mnt = nd->path.mnt;
1244 path->dentry = dentry;
1245 err = follow_managed(path, nd->flags);
1246 if (unlikely(err < 0)) {
1247 path_put_conditional(path, nd);
1251 nd->flags |= LOOKUP_JUMPED;
1255 static inline int may_lookup(struct nameidata *nd)
1257 if (nd->flags & LOOKUP_RCU) {
1258 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1261 if (unlazy_walk(nd, NULL))
1264 return inode_permission(nd->inode, MAY_EXEC);
1267 static inline int handle_dots(struct nameidata *nd, int type)
1269 if (type == LAST_DOTDOT) {
1270 if (nd->flags & LOOKUP_RCU) {
1271 if (follow_dotdot_rcu(nd))
1279 static void terminate_walk(struct nameidata *nd)
1281 if (!(nd->flags & LOOKUP_RCU)) {
1282 path_put(&nd->path);
1284 nd->flags &= ~LOOKUP_RCU;
1285 if (!(nd->flags & LOOKUP_ROOT))
1286 nd->root.mnt = NULL;
1288 br_read_unlock(&vfsmount_lock);
1293 * Do we need to follow links? We _really_ want to be able
1294 * to do this check without having to look at inode->i_op,
1295 * so we keep a cache of "no, this doesn't need follow_link"
1296 * for the common case.
1298 static inline int should_follow_link(struct inode *inode, int follow)
1300 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1301 if (likely(inode->i_op->follow_link))
1304 /* This gets set once for the inode lifetime */
1305 spin_lock(&inode->i_lock);
1306 inode->i_opflags |= IOP_NOFOLLOW;
1307 spin_unlock(&inode->i_lock);
1312 static inline int walk_component(struct nameidata *nd, struct path *path,
1313 struct qstr *name, int type, int follow)
1315 struct inode *inode;
1318 * "." and ".." are special - ".." especially so because it has
1319 * to be able to know about the current root directory and
1320 * parent relationships.
1322 if (unlikely(type != LAST_NORM))
1323 return handle_dots(nd, type);
1324 err = lookup_fast(nd, name, path, &inode);
1325 if (unlikely(err)) {
1329 err = lookup_slow(nd, name, path);
1333 inode = path->dentry->d_inode;
1339 if (should_follow_link(inode, follow)) {
1340 if (nd->flags & LOOKUP_RCU) {
1341 if (unlikely(unlazy_walk(nd, path->dentry))) {
1346 BUG_ON(inode != path->dentry->d_inode);
1349 path_to_nameidata(path, nd);
1354 path_to_nameidata(path, nd);
1361 * This limits recursive symlink follows to 8, while
1362 * limiting consecutive symlinks to 40.
1364 * Without that kind of total limit, nasty chains of consecutive
1365 * symlinks can cause almost arbitrarily long lookups.
1367 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1371 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1372 path_put_conditional(path, nd);
1373 path_put(&nd->path);
1376 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1379 current->link_count++;
1382 struct path link = *path;
1385 res = follow_link(&link, nd, &cookie);
1387 res = walk_component(nd, path, &nd->last,
1388 nd->last_type, LOOKUP_FOLLOW);
1389 put_link(nd, &link, cookie);
1392 current->link_count--;
1398 * We really don't want to look at inode->i_op->lookup
1399 * when we don't have to. So we keep a cache bit in
1400 * the inode ->i_opflags field that says "yes, we can
1401 * do lookup on this inode".
1403 static inline int can_lookup(struct inode *inode)
1405 if (likely(inode->i_opflags & IOP_LOOKUP))
1407 if (likely(!inode->i_op->lookup))
1410 /* We do this once for the lifetime of the inode */
1411 spin_lock(&inode->i_lock);
1412 inode->i_opflags |= IOP_LOOKUP;
1413 spin_unlock(&inode->i_lock);
1418 * We can do the critical dentry name comparison and hashing
1419 * operations one word at a time, but we are limited to:
1421 * - Architectures with fast unaligned word accesses. We could
1422 * do a "get_unaligned()" if this helps and is sufficiently
1425 * - Little-endian machines (so that we can generate the mask
1426 * of low bytes efficiently). Again, we *could* do a byte
1427 * swapping load on big-endian architectures if that is not
1428 * expensive enough to make the optimization worthless.
1430 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1431 * do not trap on the (extremely unlikely) case of a page
1432 * crossing operation.
1434 * - Furthermore, we need an efficient 64-bit compile for the
1435 * 64-bit case in order to generate the "number of bytes in
1436 * the final mask". Again, that could be replaced with a
1437 * efficient population count instruction or similar.
1439 #ifdef CONFIG_DCACHE_WORD_ACCESS
1441 #include <asm/word-at-a-time.h>
1445 static inline unsigned int fold_hash(unsigned long hash)
1447 hash += hash >> (8*sizeof(int));
1451 #else /* 32-bit case */
1453 #define fold_hash(x) (x)
1457 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1459 unsigned long a, mask;
1460 unsigned long hash = 0;
1463 a = load_unaligned_zeropad(name);
1464 if (len < sizeof(unsigned long))
1468 name += sizeof(unsigned long);
1469 len -= sizeof(unsigned long);
1473 mask = ~(~0ul << len*8);
1476 return fold_hash(hash);
1478 EXPORT_SYMBOL(full_name_hash);
1481 * Calculate the length and hash of the path component, and
1482 * return the length of the component;
1484 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1486 unsigned long a, b, adata, bdata, mask, hash, len;
1487 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1490 len = -sizeof(unsigned long);
1492 hash = (hash + a) * 9;
1493 len += sizeof(unsigned long);
1494 a = load_unaligned_zeropad(name+len);
1495 b = a ^ REPEAT_BYTE('/');
1496 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1498 adata = prep_zero_mask(a, adata, &constants);
1499 bdata = prep_zero_mask(b, bdata, &constants);
1501 mask = create_zero_mask(adata | bdata);
1503 hash += a & zero_bytemask(mask);
1504 *hashp = fold_hash(hash);
1506 return len + find_zero(mask);
1511 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1513 unsigned long hash = init_name_hash();
1515 hash = partial_name_hash(*name++, hash);
1516 return end_name_hash(hash);
1518 EXPORT_SYMBOL(full_name_hash);
1521 * We know there's a real path component here of at least
1524 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1526 unsigned long hash = init_name_hash();
1527 unsigned long len = 0, c;
1529 c = (unsigned char)*name;
1532 hash = partial_name_hash(c, hash);
1533 c = (unsigned char)name[len];
1534 } while (c && c != '/');
1535 *hashp = end_name_hash(hash);
1543 * This is the basic name resolution function, turning a pathname into
1544 * the final dentry. We expect 'base' to be positive and a directory.
1546 * Returns 0 and nd will have valid dentry and mnt on success.
1547 * Returns error and drops reference to input namei data on failure.
1549 static int link_path_walk(const char *name, struct nameidata *nd)
1559 /* At this point we know we have a real path component. */
1565 err = may_lookup(nd);
1569 len = hash_name(name, &this.hash);
1574 if (name[0] == '.') switch (len) {
1576 if (name[1] == '.') {
1578 nd->flags |= LOOKUP_JUMPED;
1584 if (likely(type == LAST_NORM)) {
1585 struct dentry *parent = nd->path.dentry;
1586 nd->flags &= ~LOOKUP_JUMPED;
1587 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1588 err = parent->d_op->d_hash(parent, nd->inode,
1596 goto last_component;
1598 * If it wasn't NUL, we know it was '/'. Skip that
1599 * slash, and continue until no more slashes.
1603 } while (unlikely(name[len] == '/'));
1605 goto last_component;
1608 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1613 err = nested_symlink(&next, nd);
1617 if (can_lookup(nd->inode))
1621 /* here ends the main loop */
1625 nd->last_type = type;
1632 static int path_init(int dfd, const char *name, unsigned int flags,
1633 struct nameidata *nd, struct file **fp)
1639 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1640 nd->flags = flags | LOOKUP_JUMPED;
1642 if (flags & LOOKUP_ROOT) {
1643 struct inode *inode = nd->root.dentry->d_inode;
1645 if (!inode->i_op->lookup)
1647 retval = inode_permission(inode, MAY_EXEC);
1651 nd->path = nd->root;
1653 if (flags & LOOKUP_RCU) {
1654 br_read_lock(&vfsmount_lock);
1656 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1658 path_get(&nd->path);
1663 nd->root.mnt = NULL;
1666 if (flags & LOOKUP_RCU) {
1667 br_read_lock(&vfsmount_lock);
1672 path_get(&nd->root);
1674 nd->path = nd->root;
1675 } else if (dfd == AT_FDCWD) {
1676 if (flags & LOOKUP_RCU) {
1677 struct fs_struct *fs = current->fs;
1680 br_read_lock(&vfsmount_lock);
1684 seq = read_seqcount_begin(&fs->seq);
1686 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1687 } while (read_seqcount_retry(&fs->seq, seq));
1689 get_fs_pwd(current->fs, &nd->path);
1692 struct dentry *dentry;
1694 file = fget_raw_light(dfd, &fput_needed);
1699 dentry = file->f_path.dentry;
1703 if (!S_ISDIR(dentry->d_inode->i_mode))
1706 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1711 nd->path = file->f_path;
1712 if (flags & LOOKUP_RCU) {
1715 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1716 br_read_lock(&vfsmount_lock);
1719 path_get(&file->f_path);
1720 fput_light(file, fput_needed);
1724 nd->inode = nd->path.dentry->d_inode;
1728 fput_light(file, fput_needed);
1733 static inline int lookup_last(struct nameidata *nd, struct path *path)
1735 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1736 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1738 nd->flags &= ~LOOKUP_PARENT;
1739 return walk_component(nd, path, &nd->last, nd->last_type,
1740 nd->flags & LOOKUP_FOLLOW);
1743 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1744 static int path_lookupat(int dfd, const char *name,
1745 unsigned int flags, struct nameidata *nd)
1747 struct file *base = NULL;
1752 * Path walking is largely split up into 2 different synchronisation
1753 * schemes, rcu-walk and ref-walk (explained in
1754 * Documentation/filesystems/path-lookup.txt). These share much of the
1755 * path walk code, but some things particularly setup, cleanup, and
1756 * following mounts are sufficiently divergent that functions are
1757 * duplicated. Typically there is a function foo(), and its RCU
1758 * analogue, foo_rcu().
1760 * -ECHILD is the error number of choice (just to avoid clashes) that
1761 * is returned if some aspect of an rcu-walk fails. Such an error must
1762 * be handled by restarting a traditional ref-walk (which will always
1763 * be able to complete).
1765 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1770 current->total_link_count = 0;
1771 err = link_path_walk(name, nd);
1773 if (!err && !(flags & LOOKUP_PARENT)) {
1774 err = lookup_last(nd, &path);
1777 struct path link = path;
1778 nd->flags |= LOOKUP_PARENT;
1779 err = follow_link(&link, nd, &cookie);
1781 err = lookup_last(nd, &path);
1782 put_link(nd, &link, cookie);
1787 err = complete_walk(nd);
1789 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1790 if (!nd->inode->i_op->lookup) {
1791 path_put(&nd->path);
1799 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1800 path_put(&nd->root);
1801 nd->root.mnt = NULL;
1806 static int do_path_lookup(int dfd, const char *name,
1807 unsigned int flags, struct nameidata *nd)
1809 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1810 if (unlikely(retval == -ECHILD))
1811 retval = path_lookupat(dfd, name, flags, nd);
1812 if (unlikely(retval == -ESTALE))
1813 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1815 if (likely(!retval)) {
1816 if (unlikely(!audit_dummy_context())) {
1817 if (nd->path.dentry && nd->inode)
1818 audit_inode(name, nd->path.dentry);
1824 int kern_path_parent(const char *name, struct nameidata *nd)
1826 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1829 int kern_path(const char *name, unsigned int flags, struct path *path)
1831 struct nameidata nd;
1832 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1839 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1840 * @dentry: pointer to dentry of the base directory
1841 * @mnt: pointer to vfs mount of the base directory
1842 * @name: pointer to file name
1843 * @flags: lookup flags
1844 * @path: pointer to struct path to fill
1846 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1847 const char *name, unsigned int flags,
1850 struct nameidata nd;
1852 nd.root.dentry = dentry;
1854 BUG_ON(flags & LOOKUP_PARENT);
1855 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1856 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1863 * Restricted form of lookup. Doesn't follow links, single-component only,
1864 * needs parent already locked. Doesn't follow mounts.
1867 static struct dentry *lookup_hash(struct nameidata *nd)
1869 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1873 * lookup_one_len - filesystem helper to lookup single pathname component
1874 * @name: pathname component to lookup
1875 * @base: base directory to lookup from
1876 * @len: maximum length @len should be interpreted to
1878 * Note that this routine is purely a helper for filesystem usage and should
1879 * not be called by generic code. Also note that by using this function the
1880 * nameidata argument is passed to the filesystem methods and a filesystem
1881 * using this helper needs to be prepared for that.
1883 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1889 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1893 this.hash = full_name_hash(name, len);
1895 return ERR_PTR(-EACCES);
1898 c = *(const unsigned char *)name++;
1899 if (c == '/' || c == '\0')
1900 return ERR_PTR(-EACCES);
1903 * See if the low-level filesystem might want
1904 * to use its own hash..
1906 if (base->d_flags & DCACHE_OP_HASH) {
1907 int err = base->d_op->d_hash(base, base->d_inode, &this);
1909 return ERR_PTR(err);
1912 err = inode_permission(base->d_inode, MAY_EXEC);
1914 return ERR_PTR(err);
1916 return __lookup_hash(&this, base, NULL);
1919 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1920 struct path *path, int *empty)
1922 struct nameidata nd;
1923 char *tmp = getname_flags(name, flags, empty);
1924 int err = PTR_ERR(tmp);
1927 BUG_ON(flags & LOOKUP_PARENT);
1929 err = do_path_lookup(dfd, tmp, flags, &nd);
1937 int user_path_at(int dfd, const char __user *name, unsigned flags,
1940 return user_path_at_empty(dfd, name, flags, path, NULL);
1943 static int user_path_parent(int dfd, const char __user *path,
1944 struct nameidata *nd, char **name)
1946 char *s = getname(path);
1952 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1962 * It's inline, so penalty for filesystems that don't use sticky bit is
1965 static inline int check_sticky(struct inode *dir, struct inode *inode)
1967 kuid_t fsuid = current_fsuid();
1969 if (!(dir->i_mode & S_ISVTX))
1971 if (uid_eq(inode->i_uid, fsuid))
1973 if (uid_eq(dir->i_uid, fsuid))
1975 return !inode_capable(inode, CAP_FOWNER);
1979 * Check whether we can remove a link victim from directory dir, check
1980 * whether the type of victim is right.
1981 * 1. We can't do it if dir is read-only (done in permission())
1982 * 2. We should have write and exec permissions on dir
1983 * 3. We can't remove anything from append-only dir
1984 * 4. We can't do anything with immutable dir (done in permission())
1985 * 5. If the sticky bit on dir is set we should either
1986 * a. be owner of dir, or
1987 * b. be owner of victim, or
1988 * c. have CAP_FOWNER capability
1989 * 6. If the victim is append-only or immutable we can't do antyhing with
1990 * links pointing to it.
1991 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1992 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1993 * 9. We can't remove a root or mountpoint.
1994 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1995 * nfs_async_unlink().
1997 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2001 if (!victim->d_inode)
2004 BUG_ON(victim->d_parent->d_inode != dir);
2005 audit_inode_child(victim, dir);
2007 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2012 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2013 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2016 if (!S_ISDIR(victim->d_inode->i_mode))
2018 if (IS_ROOT(victim))
2020 } else if (S_ISDIR(victim->d_inode->i_mode))
2022 if (IS_DEADDIR(dir))
2024 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2029 /* Check whether we can create an object with dentry child in directory
2031 * 1. We can't do it if child already exists (open has special treatment for
2032 * this case, but since we are inlined it's OK)
2033 * 2. We can't do it if dir is read-only (done in permission())
2034 * 3. We should have write and exec permissions on dir
2035 * 4. We can't do it if dir is immutable (done in permission())
2037 static inline int may_create(struct inode *dir, struct dentry *child)
2041 if (IS_DEADDIR(dir))
2043 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2047 * p1 and p2 should be directories on the same fs.
2049 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2054 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2058 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2060 p = d_ancestor(p2, p1);
2062 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2063 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2067 p = d_ancestor(p1, p2);
2069 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2070 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2074 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2075 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2079 void unlock_rename(struct dentry *p1, struct dentry *p2)
2081 mutex_unlock(&p1->d_inode->i_mutex);
2083 mutex_unlock(&p2->d_inode->i_mutex);
2084 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2088 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2089 struct nameidata *nd)
2091 int error = may_create(dir, dentry);
2096 if (!dir->i_op->create)
2097 return -EACCES; /* shouldn't it be ENOSYS? */
2100 error = security_inode_create(dir, dentry, mode);
2103 error = dir->i_op->create(dir, dentry, mode, nd);
2105 fsnotify_create(dir, dentry);
2109 static int may_open(struct path *path, int acc_mode, int flag)
2111 struct dentry *dentry = path->dentry;
2112 struct inode *inode = dentry->d_inode;
2122 switch (inode->i_mode & S_IFMT) {
2126 if (acc_mode & MAY_WRITE)
2131 if (path->mnt->mnt_flags & MNT_NODEV)
2140 error = inode_permission(inode, acc_mode);
2145 * An append-only file must be opened in append mode for writing.
2147 if (IS_APPEND(inode)) {
2148 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2154 /* O_NOATIME can only be set by the owner or superuser */
2155 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2161 static int handle_truncate(struct file *filp)
2163 struct path *path = &filp->f_path;
2164 struct inode *inode = path->dentry->d_inode;
2165 int error = get_write_access(inode);
2169 * Refuse to truncate files with mandatory locks held on them.
2171 error = locks_verify_locked(inode);
2173 error = security_path_truncate(path);
2175 error = do_truncate(path->dentry, 0,
2176 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2179 put_write_access(inode);
2183 static inline int open_to_namei_flags(int flag)
2185 if ((flag & O_ACCMODE) == 3)
2191 * Handle the last step of open()
2193 static struct file *do_last(struct nameidata *nd, struct path *path,
2194 const struct open_flags *op, const char *pathname)
2196 struct dentry *dir = nd->path.dentry;
2197 struct dentry *dentry;
2198 int open_flag = op->open_flag;
2199 int will_truncate = open_flag & O_TRUNC;
2201 int acc_mode = op->acc_mode;
2203 struct inode *inode;
2205 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2206 bool retried = false;
2209 nd->flags &= ~LOOKUP_PARENT;
2210 nd->flags |= op->intent;
2212 switch (nd->last_type) {
2215 error = handle_dots(nd, nd->last_type);
2217 return ERR_PTR(error);
2220 error = complete_walk(nd);
2222 return ERR_PTR(error);
2223 audit_inode(pathname, nd->path.dentry);
2224 if (open_flag & O_CREAT) {
2230 error = complete_walk(nd);
2232 return ERR_PTR(error);
2233 audit_inode(pathname, dir);
2237 if (!(open_flag & O_CREAT)) {
2238 if (nd->last.name[nd->last.len])
2239 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2240 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2242 /* we _can_ be in RCU mode here */
2243 error = lookup_fast(nd, &nd->last, path, &inode);
2244 if (unlikely(error)) {
2248 error = lookup_slow(nd, &nd->last, path);
2252 inode = path->dentry->d_inode;
2257 /* create side of things */
2259 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2260 * cleared when we got to the last component we are about to look up
2262 error = complete_walk(nd);
2264 return ERR_PTR(error);
2266 audit_inode(pathname, dir);
2268 /* trailing slashes? */
2269 if (nd->last.name[nd->last.len])
2273 mutex_lock(&dir->d_inode->i_mutex);
2275 dentry = lookup_hash(nd);
2276 error = PTR_ERR(dentry);
2277 if (IS_ERR(dentry)) {
2278 mutex_unlock(&dir->d_inode->i_mutex);
2282 path->dentry = dentry;
2283 path->mnt = nd->path.mnt;
2285 /* Negative dentry, just create the file */
2286 if (!dentry->d_inode) {
2287 umode_t mode = op->mode;
2288 if (!IS_POSIXACL(dir->d_inode))
2289 mode &= ~current_umask();
2291 * This write is needed to ensure that a
2292 * rw->ro transition does not occur between
2293 * the time when the file is created and when
2294 * a permanent write count is taken through
2295 * the 'struct file' in nameidata_to_filp().
2297 error = mnt_want_write(nd->path.mnt);
2299 goto exit_mutex_unlock;
2301 /* Don't check for write permission, don't truncate */
2302 open_flag &= ~O_TRUNC;
2304 acc_mode = MAY_OPEN;
2305 error = security_path_mknod(&nd->path, dentry, mode, 0);
2307 goto exit_mutex_unlock;
2308 error = vfs_create(dir->d_inode, dentry, mode, nd);
2310 goto exit_mutex_unlock;
2311 mutex_unlock(&dir->d_inode->i_mutex);
2312 dput(nd->path.dentry);
2313 nd->path.dentry = dentry;
2318 * It already exists.
2320 mutex_unlock(&dir->d_inode->i_mutex);
2321 audit_inode(pathname, path->dentry);
2324 if (open_flag & O_EXCL)
2327 error = follow_managed(path, nd->flags);
2332 nd->flags |= LOOKUP_JUMPED;
2334 BUG_ON(nd->flags & LOOKUP_RCU);
2335 inode = path->dentry->d_inode;
2337 /* we _can_ be in RCU mode here */
2340 path_to_nameidata(path, nd);
2344 if (should_follow_link(inode, !symlink_ok)) {
2345 if (nd->flags & LOOKUP_RCU) {
2346 if (unlikely(unlazy_walk(nd, path->dentry))) {
2351 BUG_ON(inode != path->dentry->d_inode);
2355 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2356 path_to_nameidata(path, nd);
2358 save_parent.dentry = nd->path.dentry;
2359 save_parent.mnt = mntget(path->mnt);
2360 nd->path.dentry = path->dentry;
2364 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2365 error = complete_walk(nd);
2367 path_put(&save_parent);
2368 return ERR_PTR(error);
2371 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
2374 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
2376 audit_inode(pathname, nd->path.dentry);
2378 if (!S_ISREG(nd->inode->i_mode))
2381 if (will_truncate) {
2382 error = mnt_want_write(nd->path.mnt);
2388 error = may_open(&nd->path, acc_mode, open_flag);
2391 filp = nameidata_to_filp(nd);
2392 if (filp == ERR_PTR(-EOPENSTALE) && save_parent.dentry && !retried) {
2393 BUG_ON(save_parent.dentry != dir);
2394 path_put(&nd->path);
2395 nd->path = save_parent;
2396 nd->inode = dir->d_inode;
2397 save_parent.mnt = NULL;
2398 save_parent.dentry = NULL;
2400 mnt_drop_write(nd->path.mnt);
2406 if (!IS_ERR(filp)) {
2407 error = ima_file_check(filp, op->acc_mode);
2410 filp = ERR_PTR(error);
2413 if (!IS_ERR(filp)) {
2414 if (will_truncate) {
2415 error = handle_truncate(filp);
2418 filp = ERR_PTR(error);
2424 mnt_drop_write(nd->path.mnt);
2425 path_put(&save_parent);
2430 mutex_unlock(&dir->d_inode->i_mutex);
2432 path_put_conditional(path, nd);
2434 filp = ERR_PTR(error);
2438 static struct file *path_openat(int dfd, const char *pathname,
2439 struct nameidata *nd, const struct open_flags *op, int flags)
2441 struct file *base = NULL;
2446 filp = get_empty_filp();
2448 return ERR_PTR(-ENFILE);
2450 filp->f_flags = op->open_flag;
2451 nd->intent.open.file = filp;
2452 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2453 nd->intent.open.create_mode = op->mode;
2455 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2456 if (unlikely(error))
2459 current->total_link_count = 0;
2460 error = link_path_walk(pathname, nd);
2461 if (unlikely(error))
2464 filp = do_last(nd, &path, op, pathname);
2465 while (unlikely(!filp)) { /* trailing symlink */
2466 struct path link = path;
2468 if (!(nd->flags & LOOKUP_FOLLOW)) {
2469 path_put_conditional(&path, nd);
2470 path_put(&nd->path);
2471 filp = ERR_PTR(-ELOOP);
2474 nd->flags |= LOOKUP_PARENT;
2475 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2476 error = follow_link(&link, nd, &cookie);
2477 if (unlikely(error))
2478 filp = ERR_PTR(error);
2480 filp = do_last(nd, &path, op, pathname);
2481 put_link(nd, &link, cookie);
2484 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2485 path_put(&nd->root);
2488 release_open_intent(nd);
2489 if (filp == ERR_PTR(-EOPENSTALE)) {
2490 if (flags & LOOKUP_RCU)
2491 filp = ERR_PTR(-ECHILD);
2493 filp = ERR_PTR(-ESTALE);
2498 filp = ERR_PTR(error);
2502 struct file *do_filp_open(int dfd, const char *pathname,
2503 const struct open_flags *op, int flags)
2505 struct nameidata nd;
2508 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2509 if (unlikely(filp == ERR_PTR(-ECHILD)))
2510 filp = path_openat(dfd, pathname, &nd, op, flags);
2511 if (unlikely(filp == ERR_PTR(-ESTALE)))
2512 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2516 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2517 const char *name, const struct open_flags *op, int flags)
2519 struct nameidata nd;
2523 nd.root.dentry = dentry;
2525 flags |= LOOKUP_ROOT;
2527 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2528 return ERR_PTR(-ELOOP);
2530 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2531 if (unlikely(file == ERR_PTR(-ECHILD)))
2532 file = path_openat(-1, name, &nd, op, flags);
2533 if (unlikely(file == ERR_PTR(-ESTALE)))
2534 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2538 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2540 struct dentry *dentry = ERR_PTR(-EEXIST);
2541 struct nameidata nd;
2542 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2544 return ERR_PTR(error);
2547 * Yucky last component or no last component at all?
2548 * (foo/., foo/.., /////)
2550 if (nd.last_type != LAST_NORM)
2552 nd.flags &= ~LOOKUP_PARENT;
2553 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2554 nd.intent.open.flags = O_EXCL;
2557 * Do the final lookup.
2559 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2560 dentry = lookup_hash(&nd);
2564 if (dentry->d_inode)
2567 * Special case - lookup gave negative, but... we had foo/bar/
2568 * From the vfs_mknod() POV we just have a negative dentry -
2569 * all is fine. Let's be bastards - you had / on the end, you've
2570 * been asking for (non-existent) directory. -ENOENT for you.
2572 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2574 dentry = ERR_PTR(-ENOENT);
2581 dentry = ERR_PTR(-EEXIST);
2583 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2588 EXPORT_SYMBOL(kern_path_create);
2590 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2592 char *tmp = getname(pathname);
2595 return ERR_CAST(tmp);
2596 res = kern_path_create(dfd, tmp, path, is_dir);
2600 EXPORT_SYMBOL(user_path_create);
2602 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2604 int error = may_create(dir, dentry);
2609 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2612 if (!dir->i_op->mknod)
2615 error = devcgroup_inode_mknod(mode, dev);
2619 error = security_inode_mknod(dir, dentry, mode, dev);
2623 error = dir->i_op->mknod(dir, dentry, mode, dev);
2625 fsnotify_create(dir, dentry);
2629 static int may_mknod(umode_t mode)
2631 switch (mode & S_IFMT) {
2637 case 0: /* zero mode translates to S_IFREG */
2646 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2649 struct dentry *dentry;
2656 dentry = user_path_create(dfd, filename, &path, 0);
2658 return PTR_ERR(dentry);
2660 if (!IS_POSIXACL(path.dentry->d_inode))
2661 mode &= ~current_umask();
2662 error = may_mknod(mode);
2665 error = mnt_want_write(path.mnt);
2668 error = security_path_mknod(&path, dentry, mode, dev);
2670 goto out_drop_write;
2671 switch (mode & S_IFMT) {
2672 case 0: case S_IFREG:
2673 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2675 case S_IFCHR: case S_IFBLK:
2676 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2677 new_decode_dev(dev));
2679 case S_IFIFO: case S_IFSOCK:
2680 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2684 mnt_drop_write(path.mnt);
2687 mutex_unlock(&path.dentry->d_inode->i_mutex);
2693 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
2695 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2698 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2700 int error = may_create(dir, dentry);
2701 unsigned max_links = dir->i_sb->s_max_links;
2706 if (!dir->i_op->mkdir)
2709 mode &= (S_IRWXUGO|S_ISVTX);
2710 error = security_inode_mkdir(dir, dentry, mode);
2714 if (max_links && dir->i_nlink >= max_links)
2717 error = dir->i_op->mkdir(dir, dentry, mode);
2719 fsnotify_mkdir(dir, dentry);
2723 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
2725 struct dentry *dentry;
2729 dentry = user_path_create(dfd, pathname, &path, 1);
2731 return PTR_ERR(dentry);
2733 if (!IS_POSIXACL(path.dentry->d_inode))
2734 mode &= ~current_umask();
2735 error = mnt_want_write(path.mnt);
2738 error = security_path_mkdir(&path, dentry, mode);
2740 goto out_drop_write;
2741 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2743 mnt_drop_write(path.mnt);
2746 mutex_unlock(&path.dentry->d_inode->i_mutex);
2751 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
2753 return sys_mkdirat(AT_FDCWD, pathname, mode);
2757 * The dentry_unhash() helper will try to drop the dentry early: we
2758 * should have a usage count of 1 if we're the only user of this
2759 * dentry, and if that is true (possibly after pruning the dcache),
2760 * then we drop the dentry now.
2762 * A low-level filesystem can, if it choses, legally
2765 * if (!d_unhashed(dentry))
2768 * if it cannot handle the case of removing a directory
2769 * that is still in use by something else..
2771 void dentry_unhash(struct dentry *dentry)
2773 shrink_dcache_parent(dentry);
2774 spin_lock(&dentry->d_lock);
2775 if (dentry->d_count == 1)
2777 spin_unlock(&dentry->d_lock);
2780 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2782 int error = may_delete(dir, dentry, 1);
2787 if (!dir->i_op->rmdir)
2791 mutex_lock(&dentry->d_inode->i_mutex);
2794 if (d_mountpoint(dentry))
2797 error = security_inode_rmdir(dir, dentry);
2801 shrink_dcache_parent(dentry);
2802 error = dir->i_op->rmdir(dir, dentry);
2806 dentry->d_inode->i_flags |= S_DEAD;
2810 mutex_unlock(&dentry->d_inode->i_mutex);
2817 static long do_rmdir(int dfd, const char __user *pathname)
2821 struct dentry *dentry;
2822 struct nameidata nd;
2824 error = user_path_parent(dfd, pathname, &nd, &name);
2828 switch(nd.last_type) {
2840 nd.flags &= ~LOOKUP_PARENT;
2842 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2843 dentry = lookup_hash(&nd);
2844 error = PTR_ERR(dentry);
2847 if (!dentry->d_inode) {
2851 error = mnt_want_write(nd.path.mnt);
2854 error = security_path_rmdir(&nd.path, dentry);
2857 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2859 mnt_drop_write(nd.path.mnt);
2863 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2870 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2872 return do_rmdir(AT_FDCWD, pathname);
2875 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2877 int error = may_delete(dir, dentry, 0);
2882 if (!dir->i_op->unlink)
2885 mutex_lock(&dentry->d_inode->i_mutex);
2886 if (d_mountpoint(dentry))
2889 error = security_inode_unlink(dir, dentry);
2891 error = dir->i_op->unlink(dir, dentry);
2896 mutex_unlock(&dentry->d_inode->i_mutex);
2898 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2899 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2900 fsnotify_link_count(dentry->d_inode);
2908 * Make sure that the actual truncation of the file will occur outside its
2909 * directory's i_mutex. Truncate can take a long time if there is a lot of
2910 * writeout happening, and we don't want to prevent access to the directory
2911 * while waiting on the I/O.
2913 static long do_unlinkat(int dfd, const char __user *pathname)
2917 struct dentry *dentry;
2918 struct nameidata nd;
2919 struct inode *inode = NULL;
2921 error = user_path_parent(dfd, pathname, &nd, &name);
2926 if (nd.last_type != LAST_NORM)
2929 nd.flags &= ~LOOKUP_PARENT;
2931 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2932 dentry = lookup_hash(&nd);
2933 error = PTR_ERR(dentry);
2934 if (!IS_ERR(dentry)) {
2935 /* Why not before? Because we want correct error value */
2936 if (nd.last.name[nd.last.len])
2938 inode = dentry->d_inode;
2942 error = mnt_want_write(nd.path.mnt);
2945 error = security_path_unlink(&nd.path, dentry);
2948 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2950 mnt_drop_write(nd.path.mnt);
2954 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2956 iput(inode); /* truncate the inode here */
2963 error = !dentry->d_inode ? -ENOENT :
2964 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2968 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2970 if ((flag & ~AT_REMOVEDIR) != 0)
2973 if (flag & AT_REMOVEDIR)
2974 return do_rmdir(dfd, pathname);
2976 return do_unlinkat(dfd, pathname);
2979 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2981 return do_unlinkat(AT_FDCWD, pathname);
2984 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2986 int error = may_create(dir, dentry);
2991 if (!dir->i_op->symlink)
2994 error = security_inode_symlink(dir, dentry, oldname);
2998 error = dir->i_op->symlink(dir, dentry, oldname);
3000 fsnotify_create(dir, dentry);
3004 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3005 int, newdfd, const char __user *, newname)
3009 struct dentry *dentry;
3012 from = getname(oldname);
3014 return PTR_ERR(from);
3016 dentry = user_path_create(newdfd, newname, &path, 0);
3017 error = PTR_ERR(dentry);
3021 error = mnt_want_write(path.mnt);
3024 error = security_path_symlink(&path, dentry, from);
3026 goto out_drop_write;
3027 error = vfs_symlink(path.dentry->d_inode, dentry, from);
3029 mnt_drop_write(path.mnt);
3032 mutex_unlock(&path.dentry->d_inode->i_mutex);
3039 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3041 return sys_symlinkat(oldname, AT_FDCWD, newname);
3044 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3046 struct inode *inode = old_dentry->d_inode;
3047 unsigned max_links = dir->i_sb->s_max_links;
3053 error = may_create(dir, new_dentry);
3057 if (dir->i_sb != inode->i_sb)
3061 * A link to an append-only or immutable file cannot be created.
3063 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3065 if (!dir->i_op->link)
3067 if (S_ISDIR(inode->i_mode))
3070 error = security_inode_link(old_dentry, dir, new_dentry);
3074 mutex_lock(&inode->i_mutex);
3075 /* Make sure we don't allow creating hardlink to an unlinked file */
3076 if (inode->i_nlink == 0)
3078 else if (max_links && inode->i_nlink >= max_links)
3081 error = dir->i_op->link(old_dentry, dir, new_dentry);
3082 mutex_unlock(&inode->i_mutex);
3084 fsnotify_link(dir, inode, new_dentry);
3089 * Hardlinks are often used in delicate situations. We avoid
3090 * security-related surprises by not following symlinks on the
3093 * We don't follow them on the oldname either to be compatible
3094 * with linux 2.0, and to avoid hard-linking to directories
3095 * and other special files. --ADM
3097 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3098 int, newdfd, const char __user *, newname, int, flags)
3100 struct dentry *new_dentry;
3101 struct path old_path, new_path;
3105 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3108 * To use null names we require CAP_DAC_READ_SEARCH
3109 * This ensures that not everyone will be able to create
3110 * handlink using the passed filedescriptor.
3112 if (flags & AT_EMPTY_PATH) {
3113 if (!capable(CAP_DAC_READ_SEARCH))
3118 if (flags & AT_SYMLINK_FOLLOW)
3119 how |= LOOKUP_FOLLOW;
3121 error = user_path_at(olddfd, oldname, how, &old_path);
3125 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3126 error = PTR_ERR(new_dentry);
3127 if (IS_ERR(new_dentry))
3131 if (old_path.mnt != new_path.mnt)
3133 error = mnt_want_write(new_path.mnt);
3136 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3138 goto out_drop_write;
3139 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3141 mnt_drop_write(new_path.mnt);
3144 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3145 path_put(&new_path);
3147 path_put(&old_path);
3152 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3154 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3158 * The worst of all namespace operations - renaming directory. "Perverted"
3159 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3161 * a) we can get into loop creation. Check is done in is_subdir().
3162 * b) race potential - two innocent renames can create a loop together.
3163 * That's where 4.4 screws up. Current fix: serialization on
3164 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3166 * c) we have to lock _three_ objects - parents and victim (if it exists).
3167 * And that - after we got ->i_mutex on parents (until then we don't know
3168 * whether the target exists). Solution: try to be smart with locking
3169 * order for inodes. We rely on the fact that tree topology may change
3170 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3171 * move will be locked. Thus we can rank directories by the tree
3172 * (ancestors first) and rank all non-directories after them.
3173 * That works since everybody except rename does "lock parent, lookup,
3174 * lock child" and rename is under ->s_vfs_rename_mutex.
3175 * HOWEVER, it relies on the assumption that any object with ->lookup()
3176 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3177 * we'd better make sure that there's no link(2) for them.
3178 * d) conversion from fhandle to dentry may come in the wrong moment - when
3179 * we are removing the target. Solution: we will have to grab ->i_mutex
3180 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3181 * ->i_mutex on parents, which works but leads to some truly excessive
3184 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3185 struct inode *new_dir, struct dentry *new_dentry)
3188 struct inode *target = new_dentry->d_inode;
3189 unsigned max_links = new_dir->i_sb->s_max_links;
3192 * If we are going to change the parent - check write permissions,
3193 * we'll need to flip '..'.
3195 if (new_dir != old_dir) {
3196 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3201 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3207 mutex_lock(&target->i_mutex);
3210 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3214 if (max_links && !target && new_dir != old_dir &&
3215 new_dir->i_nlink >= max_links)
3219 shrink_dcache_parent(new_dentry);
3220 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3225 target->i_flags |= S_DEAD;
3226 dont_mount(new_dentry);
3230 mutex_unlock(&target->i_mutex);
3233 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3234 d_move(old_dentry,new_dentry);
3238 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3239 struct inode *new_dir, struct dentry *new_dentry)
3241 struct inode *target = new_dentry->d_inode;
3244 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3250 mutex_lock(&target->i_mutex);
3253 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3256 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3261 dont_mount(new_dentry);
3262 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3263 d_move(old_dentry, new_dentry);
3266 mutex_unlock(&target->i_mutex);
3271 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3272 struct inode *new_dir, struct dentry *new_dentry)
3275 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3276 const unsigned char *old_name;
3278 if (old_dentry->d_inode == new_dentry->d_inode)
3281 error = may_delete(old_dir, old_dentry, is_dir);
3285 if (!new_dentry->d_inode)
3286 error = may_create(new_dir, new_dentry);
3288 error = may_delete(new_dir, new_dentry, is_dir);
3292 if (!old_dir->i_op->rename)
3295 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3298 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3300 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3302 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3303 new_dentry->d_inode, old_dentry);
3304 fsnotify_oldname_free(old_name);
3309 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3310 int, newdfd, const char __user *, newname)
3312 struct dentry *old_dir, *new_dir;
3313 struct dentry *old_dentry, *new_dentry;
3314 struct dentry *trap;
3315 struct nameidata oldnd, newnd;
3320 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3324 error = user_path_parent(newdfd, newname, &newnd, &to);
3329 if (oldnd.path.mnt != newnd.path.mnt)
3332 old_dir = oldnd.path.dentry;
3334 if (oldnd.last_type != LAST_NORM)
3337 new_dir = newnd.path.dentry;
3338 if (newnd.last_type != LAST_NORM)
3341 oldnd.flags &= ~LOOKUP_PARENT;
3342 newnd.flags &= ~LOOKUP_PARENT;
3343 newnd.flags |= LOOKUP_RENAME_TARGET;
3345 trap = lock_rename(new_dir, old_dir);
3347 old_dentry = lookup_hash(&oldnd);
3348 error = PTR_ERR(old_dentry);
3349 if (IS_ERR(old_dentry))
3351 /* source must exist */
3353 if (!old_dentry->d_inode)
3355 /* unless the source is a directory trailing slashes give -ENOTDIR */
3356 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3358 if (oldnd.last.name[oldnd.last.len])
3360 if (newnd.last.name[newnd.last.len])
3363 /* source should not be ancestor of target */
3365 if (old_dentry == trap)
3367 new_dentry = lookup_hash(&newnd);
3368 error = PTR_ERR(new_dentry);
3369 if (IS_ERR(new_dentry))
3371 /* target should not be an ancestor of source */
3373 if (new_dentry == trap)
3376 error = mnt_want_write(oldnd.path.mnt);
3379 error = security_path_rename(&oldnd.path, old_dentry,
3380 &newnd.path, new_dentry);
3383 error = vfs_rename(old_dir->d_inode, old_dentry,
3384 new_dir->d_inode, new_dentry);
3386 mnt_drop_write(oldnd.path.mnt);
3392 unlock_rename(new_dir, old_dir);
3394 path_put(&newnd.path);
3397 path_put(&oldnd.path);
3403 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3405 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3408 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3412 len = PTR_ERR(link);
3417 if (len > (unsigned) buflen)
3419 if (copy_to_user(buffer, link, len))
3426 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3427 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3428 * using) it for any given inode is up to filesystem.
3430 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3432 struct nameidata nd;
3437 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3439 return PTR_ERR(cookie);
3441 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3442 if (dentry->d_inode->i_op->put_link)
3443 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3447 int vfs_follow_link(struct nameidata *nd, const char *link)
3449 return __vfs_follow_link(nd, link);
3452 /* get the link contents into pagecache */
3453 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3457 struct address_space *mapping = dentry->d_inode->i_mapping;
3458 page = read_mapping_page(mapping, 0, NULL);
3463 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3467 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3469 struct page *page = NULL;
3470 char *s = page_getlink(dentry, &page);
3471 int res = vfs_readlink(dentry,buffer,buflen,s);
3474 page_cache_release(page);
3479 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3481 struct page *page = NULL;
3482 nd_set_link(nd, page_getlink(dentry, &page));
3486 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3488 struct page *page = cookie;
3492 page_cache_release(page);
3497 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3499 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3501 struct address_space *mapping = inode->i_mapping;
3506 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3508 flags |= AOP_FLAG_NOFS;
3511 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3512 flags, &page, &fsdata);
3516 kaddr = kmap_atomic(page);
3517 memcpy(kaddr, symname, len-1);
3518 kunmap_atomic(kaddr);
3520 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3527 mark_inode_dirty(inode);
3533 int page_symlink(struct inode *inode, const char *symname, int len)
3535 return __page_symlink(inode, symname, len,
3536 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3539 const struct inode_operations page_symlink_inode_operations = {
3540 .readlink = generic_readlink,
3541 .follow_link = page_follow_link_light,
3542 .put_link = page_put_link,
3545 EXPORT_SYMBOL(user_path_at);
3546 EXPORT_SYMBOL(follow_down_one);
3547 EXPORT_SYMBOL(follow_down);
3548 EXPORT_SYMBOL(follow_up);
3549 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3550 EXPORT_SYMBOL(getname);
3551 EXPORT_SYMBOL(lock_rename);
3552 EXPORT_SYMBOL(lookup_one_len);
3553 EXPORT_SYMBOL(page_follow_link_light);
3554 EXPORT_SYMBOL(page_put_link);
3555 EXPORT_SYMBOL(page_readlink);
3556 EXPORT_SYMBOL(__page_symlink);
3557 EXPORT_SYMBOL(page_symlink);
3558 EXPORT_SYMBOL(page_symlink_inode_operations);
3559 EXPORT_SYMBOL(kern_path);
3560 EXPORT_SYMBOL(vfs_path_lookup);
3561 EXPORT_SYMBOL(inode_permission);
3562 EXPORT_SYMBOL(unlock_rename);
3563 EXPORT_SYMBOL(vfs_create);
3564 EXPORT_SYMBOL(vfs_follow_link);
3565 EXPORT_SYMBOL(vfs_link);
3566 EXPORT_SYMBOL(vfs_mkdir);
3567 EXPORT_SYMBOL(vfs_mknod);
3568 EXPORT_SYMBOL(generic_permission);
3569 EXPORT_SYMBOL(vfs_readlink);
3570 EXPORT_SYMBOL(vfs_rename);
3571 EXPORT_SYMBOL(vfs_rmdir);
3572 EXPORT_SYMBOL(vfs_symlink);
3573 EXPORT_SYMBOL(vfs_unlink);
3574 EXPORT_SYMBOL(dentry_unhash);
3575 EXPORT_SYMBOL(generic_readlink);