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/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user *filename, char *page)
120 unsigned long len = PATH_MAX;
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
129 retval = strncpy_from_user(page, filename, len);
133 return -ENAMETOOLONG;
139 static char *getname_flags(const char __user * filename, int flags)
143 result = ERR_PTR(-ENOMEM);
146 int retval = do_getname(filename, tmp);
150 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
152 result = ERR_PTR(retval);
156 audit_getname(result);
160 char *getname(const char __user * filename)
162 return getname_flags(filename, 0);
165 #ifdef CONFIG_AUDITSYSCALL
166 void putname(const char *name)
168 if (unlikely(!audit_dummy_context()))
173 EXPORT_SYMBOL(putname);
177 * This does basic POSIX ACL permission checking
179 static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
180 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
182 umode_t mode = inode->i_mode;
184 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
186 if (current_fsuid() == inode->i_uid)
189 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
190 int error = check_acl(inode, mask, flags);
191 if (error != -EAGAIN)
195 if (in_group_p(inode->i_gid))
200 * If the DACs are ok we don't need any capability check.
202 if ((mask & ~mode) == 0)
208 * generic_permission - check for access rights on a Posix-like filesystem
209 * @inode: inode to check access rights for
210 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
211 * @check_acl: optional callback to check for Posix ACLs
212 * @flags: IPERM_FLAG_ flags.
214 * Used to check for read/write/execute permissions on a file.
215 * We use "fsuid" for this, letting us set arbitrary permissions
216 * for filesystem access without changing the "normal" uids which
217 * are used for other things.
219 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
220 * request cannot be satisfied (eg. requires blocking or too much complexity).
221 * It would then be called again in ref-walk mode.
223 int generic_permission(struct inode *inode, int mask, unsigned int flags,
224 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
229 * Do the basic POSIX ACL permission checks.
231 ret = acl_permission_check(inode, mask, flags, check_acl);
236 * Read/write DACs are always overridable.
237 * Executable DACs are overridable if at least one exec bit is set.
239 if (!(mask & MAY_EXEC) || execute_ok(inode))
240 if (capable(CAP_DAC_OVERRIDE))
244 * Searching includes executable on directories, else just read.
246 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
247 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
248 if (capable(CAP_DAC_READ_SEARCH))
255 * inode_permission - check for access rights to a given inode
256 * @inode: inode to check permission on
257 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
259 * Used to check for read/write/execute permissions on an inode.
260 * We use "fsuid" for this, letting us set arbitrary permissions
261 * for filesystem access without changing the "normal" uids which
262 * are used for other things.
264 int inode_permission(struct inode *inode, int mask)
268 if (mask & MAY_WRITE) {
269 umode_t mode = inode->i_mode;
272 * Nobody gets write access to a read-only fs.
274 if (IS_RDONLY(inode) &&
275 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
279 * Nobody gets write access to an immutable file.
281 if (IS_IMMUTABLE(inode))
285 if (inode->i_op->permission)
286 retval = inode->i_op->permission(inode, mask, 0);
288 retval = generic_permission(inode, mask, 0,
289 inode->i_op->check_acl);
294 retval = devcgroup_inode_permission(inode, mask);
298 return security_inode_permission(inode, mask);
302 * file_permission - check for additional access rights to a given file
303 * @file: file to check access rights for
304 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
306 * Used to check for read/write/execute permissions on an already opened
310 * Do not use this function in new code. All access checks should
311 * be done using inode_permission().
313 int file_permission(struct file *file, int mask)
315 return inode_permission(file->f_path.dentry->d_inode, mask);
319 * get_write_access() gets write permission for a file.
320 * put_write_access() releases this write permission.
321 * This is used for regular files.
322 * We cannot support write (and maybe mmap read-write shared) accesses and
323 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
324 * can have the following values:
325 * 0: no writers, no VM_DENYWRITE mappings
326 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
327 * > 0: (i_writecount) users are writing to the file.
329 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
330 * except for the cases where we don't hold i_writecount yet. Then we need to
331 * use {get,deny}_write_access() - these functions check the sign and refuse
332 * to do the change if sign is wrong. Exclusion between them is provided by
333 * the inode->i_lock spinlock.
336 int get_write_access(struct inode * inode)
338 spin_lock(&inode->i_lock);
339 if (atomic_read(&inode->i_writecount) < 0) {
340 spin_unlock(&inode->i_lock);
343 atomic_inc(&inode->i_writecount);
344 spin_unlock(&inode->i_lock);
349 int deny_write_access(struct file * file)
351 struct inode *inode = file->f_path.dentry->d_inode;
353 spin_lock(&inode->i_lock);
354 if (atomic_read(&inode->i_writecount) > 0) {
355 spin_unlock(&inode->i_lock);
358 atomic_dec(&inode->i_writecount);
359 spin_unlock(&inode->i_lock);
365 * path_get - get a reference to a path
366 * @path: path to get the reference to
368 * Given a path increment the reference count to the dentry and the vfsmount.
370 void path_get(struct path *path)
375 EXPORT_SYMBOL(path_get);
378 * path_put - put a reference to a path
379 * @path: path to put the reference to
381 * Given a path decrement the reference count to the dentry and the vfsmount.
383 void path_put(struct path *path)
388 EXPORT_SYMBOL(path_put);
391 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
392 * @nd: nameidata pathwalk data to drop
393 * Returns: 0 on success, -ECHILD on failure
395 * Path walking has 2 modes, rcu-walk and ref-walk (see
396 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
397 * to drop out of rcu-walk mode and take normal reference counts on dentries
398 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
399 * refcounts at the last known good point before rcu-walk got stuck, so
400 * ref-walk may continue from there. If this is not successful (eg. a seqcount
401 * has changed), then failure is returned and path walk restarts from the
402 * beginning in ref-walk mode.
404 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
405 * ref-walk. Must be called from rcu-walk context.
407 static int nameidata_drop_rcu(struct nameidata *nd)
409 struct fs_struct *fs = current->fs;
410 struct dentry *dentry = nd->path.dentry;
413 BUG_ON(!(nd->flags & LOOKUP_RCU));
414 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
416 spin_lock(&fs->lock);
417 if (nd->root.mnt != fs->root.mnt ||
418 nd->root.dentry != fs->root.dentry)
421 spin_lock(&dentry->d_lock);
422 if (!__d_rcu_to_refcount(dentry, nd->seq))
424 BUG_ON(nd->inode != dentry->d_inode);
425 spin_unlock(&dentry->d_lock);
428 spin_unlock(&fs->lock);
430 mntget(nd->path.mnt);
433 br_read_unlock(vfsmount_lock);
434 nd->flags &= ~LOOKUP_RCU;
437 spin_unlock(&dentry->d_lock);
440 spin_unlock(&fs->lock);
444 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
445 static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
447 if (nd->flags & LOOKUP_RCU)
448 return nameidata_drop_rcu(nd);
453 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
454 * @nd: nameidata pathwalk data to drop
455 * @dentry: dentry to drop
456 * Returns: 0 on success, -ECHILD on failure
458 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
459 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
460 * @nd. Must be called from rcu-walk context.
462 static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
464 struct fs_struct *fs = current->fs;
465 struct dentry *parent = nd->path.dentry;
468 BUG_ON(!(nd->flags & LOOKUP_RCU));
469 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
471 spin_lock(&fs->lock);
472 if (nd->root.mnt != fs->root.mnt ||
473 nd->root.dentry != fs->root.dentry)
476 spin_lock(&parent->d_lock);
477 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
478 if (!__d_rcu_to_refcount(dentry, nd->seq))
481 * If the sequence check on the child dentry passed, then the child has
482 * not been removed from its parent. This means the parent dentry must
483 * be valid and able to take a reference at this point.
485 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
486 BUG_ON(!parent->d_count);
488 spin_unlock(&dentry->d_lock);
489 spin_unlock(&parent->d_lock);
492 spin_unlock(&fs->lock);
494 mntget(nd->path.mnt);
497 br_read_unlock(vfsmount_lock);
498 nd->flags &= ~LOOKUP_RCU;
501 spin_unlock(&dentry->d_lock);
502 spin_unlock(&parent->d_lock);
505 spin_unlock(&fs->lock);
509 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
510 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
512 if (nd->flags & LOOKUP_RCU) {
513 if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) {
514 nd->flags &= ~LOOKUP_RCU;
515 if (!(nd->flags & LOOKUP_ROOT))
518 br_read_unlock(vfsmount_lock);
526 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
527 * @nd: nameidata pathwalk data to drop
528 * Returns: 0 on success, -ECHILD on failure
530 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
531 * nd->path should be the final element of the lookup, so nd->root is discarded.
532 * Must be called from rcu-walk context.
534 static int nameidata_drop_rcu_last(struct nameidata *nd)
536 struct dentry *dentry = nd->path.dentry;
538 BUG_ON(!(nd->flags & LOOKUP_RCU));
539 nd->flags &= ~LOOKUP_RCU;
540 if (!(nd->flags & LOOKUP_ROOT))
542 spin_lock(&dentry->d_lock);
543 if (!__d_rcu_to_refcount(dentry, nd->seq))
545 BUG_ON(nd->inode != dentry->d_inode);
546 spin_unlock(&dentry->d_lock);
548 mntget(nd->path.mnt);
551 br_read_unlock(vfsmount_lock);
556 spin_unlock(&dentry->d_lock);
558 br_read_unlock(vfsmount_lock);
563 * release_open_intent - free up open intent resources
564 * @nd: pointer to nameidata
566 void release_open_intent(struct nameidata *nd)
568 struct file *file = nd->intent.open.file;
570 if (file && !IS_ERR(file)) {
571 if (file->f_path.dentry == NULL)
578 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
580 return dentry->d_op->d_revalidate(dentry, nd);
583 static struct dentry *
584 do_revalidate(struct dentry *dentry, struct nameidata *nd)
586 int status = d_revalidate(dentry, nd);
587 if (unlikely(status <= 0)) {
589 * The dentry failed validation.
590 * If d_revalidate returned 0 attempt to invalidate
591 * the dentry otherwise d_revalidate is asking us
592 * to return a fail status.
596 dentry = ERR_PTR(status);
597 } else if (!d_invalidate(dentry)) {
606 * handle_reval_path - force revalidation of a dentry
608 * In some situations the path walking code will trust dentries without
609 * revalidating them. This causes problems for filesystems that depend on
610 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
611 * (which indicates that it's possible for the dentry to go stale), force
612 * a d_revalidate call before proceeding.
614 * Returns 0 if the revalidation was successful. If the revalidation fails,
615 * either return the error returned by d_revalidate or -ESTALE if the
616 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
617 * invalidate the dentry. It's up to the caller to handle putting references
618 * to the path if necessary.
620 static inline int handle_reval_path(struct nameidata *nd)
622 struct dentry *dentry = nd->path.dentry;
625 if (likely(!(nd->flags & LOOKUP_JUMPED)))
628 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
631 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
634 /* Note: we do not d_invalidate() */
635 status = d_revalidate(dentry, nd);
646 * Short-cut version of permission(), for calling on directories
647 * during pathname resolution. Combines parts of permission()
648 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
650 * If appropriate, check DAC only. If not appropriate, or
651 * short-cut DAC fails, then call ->permission() to do more
652 * complete permission check.
654 static inline int exec_permission(struct inode *inode, unsigned int flags)
658 if (inode->i_op->permission) {
659 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
661 ret = acl_permission_check(inode, MAY_EXEC, flags,
662 inode->i_op->check_acl);
669 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
674 return security_inode_exec_permission(inode, flags);
677 static __always_inline void set_root(struct nameidata *nd)
680 get_fs_root(current->fs, &nd->root);
683 static int link_path_walk(const char *, struct nameidata *);
685 static __always_inline void set_root_rcu(struct nameidata *nd)
688 struct fs_struct *fs = current->fs;
692 seq = read_seqcount_begin(&fs->seq);
694 } while (read_seqcount_retry(&fs->seq, seq));
698 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
710 nd->flags |= LOOKUP_JUMPED;
712 nd->inode = nd->path.dentry->d_inode;
714 ret = link_path_walk(link, nd);
718 return PTR_ERR(link);
721 static void path_put_conditional(struct path *path, struct nameidata *nd)
724 if (path->mnt != nd->path.mnt)
728 static inline void path_to_nameidata(const struct path *path,
729 struct nameidata *nd)
731 if (!(nd->flags & LOOKUP_RCU)) {
732 dput(nd->path.dentry);
733 if (nd->path.mnt != path->mnt)
734 mntput(nd->path.mnt);
736 nd->path.mnt = path->mnt;
737 nd->path.dentry = path->dentry;
740 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
742 struct inode *inode = link->dentry->d_inode;
743 if (!IS_ERR(cookie) && inode->i_op->put_link)
744 inode->i_op->put_link(link->dentry, nd, cookie);
748 static __always_inline int
749 follow_link(struct path *link, struct nameidata *nd, void **p)
752 struct dentry *dentry = link->dentry;
754 BUG_ON(nd->flags & LOOKUP_RCU);
756 if (link->mnt == nd->path.mnt)
759 if (unlikely(current->total_link_count >= 40)) {
760 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
765 current->total_link_count++;
767 touch_atime(link->mnt, dentry);
768 nd_set_link(nd, NULL);
770 error = security_inode_follow_link(link->dentry, nd);
772 *p = ERR_PTR(error); /* no ->put_link(), please */
777 nd->last_type = LAST_BIND;
778 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
781 char *s = nd_get_link(nd);
784 error = __vfs_follow_link(nd, s);
785 else if (nd->last_type == LAST_BIND) {
786 nd->flags |= LOOKUP_JUMPED;
787 nd->inode = nd->path.dentry->d_inode;
788 if (nd->inode->i_op->follow_link) {
789 /* stepped on a _really_ weird one */
798 static int follow_up_rcu(struct path *path)
800 struct vfsmount *parent;
801 struct dentry *mountpoint;
803 parent = path->mnt->mnt_parent;
804 if (parent == path->mnt)
806 mountpoint = path->mnt->mnt_mountpoint;
807 path->dentry = mountpoint;
812 int follow_up(struct path *path)
814 struct vfsmount *parent;
815 struct dentry *mountpoint;
817 br_read_lock(vfsmount_lock);
818 parent = path->mnt->mnt_parent;
819 if (parent == path->mnt) {
820 br_read_unlock(vfsmount_lock);
824 mountpoint = dget(path->mnt->mnt_mountpoint);
825 br_read_unlock(vfsmount_lock);
827 path->dentry = mountpoint;
834 * Perform an automount
835 * - return -EISDIR to tell follow_managed() to stop and return the path we
838 static int follow_automount(struct path *path, unsigned flags,
841 struct vfsmount *mnt;
844 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
847 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
848 * and this is the terminal part of the path.
850 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
851 return -EISDIR; /* we actually want to stop here */
853 /* We want to mount if someone is trying to open/create a file of any
854 * type under the mountpoint, wants to traverse through the mountpoint
855 * or wants to open the mounted directory.
857 * We don't want to mount if someone's just doing a stat and they've
858 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
859 * appended a '/' to the name.
861 if (!(flags & LOOKUP_FOLLOW) &&
862 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
863 LOOKUP_OPEN | LOOKUP_CREATE)))
866 current->total_link_count++;
867 if (current->total_link_count >= 40)
870 mnt = path->dentry->d_op->d_automount(path);
873 * The filesystem is allowed to return -EISDIR here to indicate
874 * it doesn't want to automount. For instance, autofs would do
875 * this so that its userspace daemon can mount on this dentry.
877 * However, we can only permit this if it's a terminal point in
878 * the path being looked up; if it wasn't then the remainder of
879 * the path is inaccessible and we should say so.
881 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
886 if (!mnt) /* mount collision */
889 err = finish_automount(mnt, path);
893 /* Someone else made a mount here whilst we were busy */
900 path->dentry = dget(mnt->mnt_root);
910 * Handle a dentry that is managed in some way.
911 * - Flagged for transit management (autofs)
912 * - Flagged as mountpoint
913 * - Flagged as automount point
915 * This may only be called in refwalk mode.
917 * Serialization is taken care of in namespace.c
919 static int follow_managed(struct path *path, unsigned flags)
922 bool need_mntput = false;
925 /* Given that we're not holding a lock here, we retain the value in a
926 * local variable for each dentry as we look at it so that we don't see
927 * the components of that value change under us */
928 while (managed = ACCESS_ONCE(path->dentry->d_flags),
929 managed &= DCACHE_MANAGED_DENTRY,
930 unlikely(managed != 0)) {
931 /* Allow the filesystem to manage the transit without i_mutex
933 if (managed & DCACHE_MANAGE_TRANSIT) {
934 BUG_ON(!path->dentry->d_op);
935 BUG_ON(!path->dentry->d_op->d_manage);
936 ret = path->dentry->d_op->d_manage(path->dentry, false);
938 return ret == -EISDIR ? 0 : ret;
941 /* Transit to a mounted filesystem. */
942 if (managed & DCACHE_MOUNTED) {
943 struct vfsmount *mounted = lookup_mnt(path);
949 path->dentry = dget(mounted->mnt_root);
954 /* Something is mounted on this dentry in another
955 * namespace and/or whatever was mounted there in this
956 * namespace got unmounted before we managed to get the
960 /* Handle an automount point */
961 if (managed & DCACHE_NEED_AUTOMOUNT) {
962 ret = follow_automount(path, flags, &need_mntput);
964 return ret == -EISDIR ? 0 : ret;
968 /* We didn't change the current path point */
974 int follow_down_one(struct path *path)
976 struct vfsmount *mounted;
978 mounted = lookup_mnt(path);
983 path->dentry = dget(mounted->mnt_root);
990 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
991 * meet a managed dentry and we're not walking to "..". True is returned to
992 * continue, false to abort.
994 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
995 struct inode **inode, bool reverse_transit)
997 while (d_mountpoint(path->dentry)) {
998 struct vfsmount *mounted;
999 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1001 path->dentry->d_op->d_manage(path->dentry, true) < 0)
1003 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1006 path->mnt = mounted;
1007 path->dentry = mounted->mnt_root;
1008 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1009 *inode = path->dentry->d_inode;
1012 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1013 return reverse_transit;
1017 static int follow_dotdot_rcu(struct nameidata *nd)
1019 struct inode *inode = nd->inode;
1024 if (nd->path.dentry == nd->root.dentry &&
1025 nd->path.mnt == nd->root.mnt) {
1028 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1029 struct dentry *old = nd->path.dentry;
1030 struct dentry *parent = old->d_parent;
1033 seq = read_seqcount_begin(&parent->d_seq);
1034 if (read_seqcount_retry(&old->d_seq, nd->seq))
1036 inode = parent->d_inode;
1037 nd->path.dentry = parent;
1041 if (!follow_up_rcu(&nd->path))
1043 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1044 inode = nd->path.dentry->d_inode;
1046 __follow_mount_rcu(nd, &nd->path, &inode, true);
1051 nd->flags &= ~LOOKUP_RCU;
1052 if (!(nd->flags & LOOKUP_ROOT))
1053 nd->root.mnt = NULL;
1055 br_read_unlock(vfsmount_lock);
1060 * Follow down to the covering mount currently visible to userspace. At each
1061 * point, the filesystem owning that dentry may be queried as to whether the
1062 * caller is permitted to proceed or not.
1064 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1067 int follow_down(struct path *path)
1072 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1073 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1074 /* Allow the filesystem to manage the transit without i_mutex
1077 * We indicate to the filesystem if someone is trying to mount
1078 * something here. This gives autofs the chance to deny anyone
1079 * other than its daemon the right to mount on its
1082 * The filesystem may sleep at this point.
1084 if (managed & DCACHE_MANAGE_TRANSIT) {
1085 BUG_ON(!path->dentry->d_op);
1086 BUG_ON(!path->dentry->d_op->d_manage);
1087 ret = path->dentry->d_op->d_manage(
1088 path->dentry, false);
1090 return ret == -EISDIR ? 0 : ret;
1093 /* Transit to a mounted filesystem. */
1094 if (managed & DCACHE_MOUNTED) {
1095 struct vfsmount *mounted = lookup_mnt(path);
1100 path->mnt = mounted;
1101 path->dentry = dget(mounted->mnt_root);
1105 /* Don't handle automount points here */
1112 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1114 static void follow_mount(struct path *path)
1116 while (d_mountpoint(path->dentry)) {
1117 struct vfsmount *mounted = lookup_mnt(path);
1122 path->mnt = mounted;
1123 path->dentry = dget(mounted->mnt_root);
1127 static void follow_dotdot(struct nameidata *nd)
1132 struct dentry *old = nd->path.dentry;
1134 if (nd->path.dentry == nd->root.dentry &&
1135 nd->path.mnt == nd->root.mnt) {
1138 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1139 /* rare case of legitimate dget_parent()... */
1140 nd->path.dentry = dget_parent(nd->path.dentry);
1144 if (!follow_up(&nd->path))
1147 follow_mount(&nd->path);
1148 nd->inode = nd->path.dentry->d_inode;
1152 * Allocate a dentry with name and parent, and perform a parent
1153 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1154 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1155 * have verified that no child exists while under i_mutex.
1157 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1158 struct qstr *name, struct nameidata *nd)
1160 struct inode *inode = parent->d_inode;
1161 struct dentry *dentry;
1164 /* Don't create child dentry for a dead directory. */
1165 if (unlikely(IS_DEADDIR(inode)))
1166 return ERR_PTR(-ENOENT);
1168 dentry = d_alloc(parent, name);
1169 if (unlikely(!dentry))
1170 return ERR_PTR(-ENOMEM);
1172 old = inode->i_op->lookup(inode, dentry, nd);
1173 if (unlikely(old)) {
1181 * It's more convoluted than I'd like it to be, but... it's still fairly
1182 * small and for now I'd prefer to have fast path as straight as possible.
1183 * It _is_ time-critical.
1185 static int do_lookup(struct nameidata *nd, struct qstr *name,
1186 struct path *path, struct inode **inode)
1188 struct vfsmount *mnt = nd->path.mnt;
1189 struct dentry *dentry, *parent = nd->path.dentry;
1195 * Rename seqlock is not required here because in the off chance
1196 * of a false negative due to a concurrent rename, we're going to
1197 * do the non-racy lookup, below.
1199 if (nd->flags & LOOKUP_RCU) {
1202 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1206 /* Memory barrier in read_seqcount_begin of child is enough */
1207 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1211 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1212 status = d_revalidate(dentry, nd);
1213 if (unlikely(status <= 0)) {
1214 if (status != -ECHILD)
1220 path->dentry = dentry;
1221 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1225 if (nameidata_dentry_drop_rcu(nd, dentry))
1228 if (nameidata_drop_rcu(nd))
1232 dentry = __d_lookup(parent, name);
1236 if (unlikely(!dentry)) {
1237 struct inode *dir = parent->d_inode;
1238 BUG_ON(nd->inode != dir);
1240 mutex_lock(&dir->i_mutex);
1241 dentry = d_lookup(parent, name);
1242 if (likely(!dentry)) {
1243 dentry = d_alloc_and_lookup(parent, name, nd);
1244 if (IS_ERR(dentry)) {
1245 mutex_unlock(&dir->i_mutex);
1246 return PTR_ERR(dentry);
1252 mutex_unlock(&dir->i_mutex);
1254 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1255 status = d_revalidate(dentry, nd);
1256 if (unlikely(status <= 0)) {
1261 if (!d_invalidate(dentry)) {
1270 path->dentry = dentry;
1271 err = follow_managed(path, nd->flags);
1272 if (unlikely(err < 0)) {
1273 path_put_conditional(path, nd);
1276 *inode = path->dentry->d_inode;
1280 static inline int may_lookup(struct nameidata *nd)
1282 if (nd->flags & LOOKUP_RCU) {
1283 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1286 if (nameidata_drop_rcu(nd))
1289 return exec_permission(nd->inode, 0);
1292 static inline int handle_dots(struct nameidata *nd, int type)
1294 if (type == LAST_DOTDOT) {
1295 if (nd->flags & LOOKUP_RCU) {
1296 if (follow_dotdot_rcu(nd))
1304 static void terminate_walk(struct nameidata *nd)
1306 if (!(nd->flags & LOOKUP_RCU)) {
1307 path_put(&nd->path);
1309 nd->flags &= ~LOOKUP_RCU;
1310 if (!(nd->flags & LOOKUP_ROOT))
1311 nd->root.mnt = NULL;
1313 br_read_unlock(vfsmount_lock);
1317 static inline int walk_component(struct nameidata *nd, struct path *path,
1318 struct qstr *name, int type, int follow)
1320 struct inode *inode;
1323 * "." and ".." are special - ".." especially so because it has
1324 * to be able to know about the current root directory and
1325 * parent relationships.
1327 if (unlikely(type != LAST_NORM))
1328 return handle_dots(nd, type);
1329 err = do_lookup(nd, name, path, &inode);
1330 if (unlikely(err)) {
1335 path_to_nameidata(path, nd);
1339 if (unlikely(inode->i_op->follow_link) && follow) {
1340 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
1342 BUG_ON(inode != path->dentry->d_inode);
1345 path_to_nameidata(path, nd);
1351 * This limits recursive symlink follows to 8, while
1352 * limiting consecutive symlinks to 40.
1354 * Without that kind of total limit, nasty chains of consecutive
1355 * symlinks can cause almost arbitrarily long lookups.
1357 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1361 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1362 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1363 path_put_conditional(path, nd);
1364 path_put(&nd->path);
1369 current->link_count++;
1372 struct path link = *path;
1375 res = follow_link(&link, nd, &cookie);
1377 res = walk_component(nd, path, &nd->last,
1378 nd->last_type, LOOKUP_FOLLOW);
1379 put_link(nd, &link, cookie);
1382 current->link_count--;
1389 * This is the basic name resolution function, turning a pathname into
1390 * the final dentry. We expect 'base' to be positive and a directory.
1392 * Returns 0 and nd will have valid dentry and mnt on success.
1393 * Returns error and drops reference to input namei data on failure.
1395 static int link_path_walk(const char *name, struct nameidata *nd)
1399 unsigned int lookup_flags = nd->flags;
1406 /* At this point we know we have a real path component. */
1413 nd->flags |= LOOKUP_CONTINUE;
1415 err = may_lookup(nd);
1420 c = *(const unsigned char *)name;
1422 hash = init_name_hash();
1425 hash = partial_name_hash(c, hash);
1426 c = *(const unsigned char *)name;
1427 } while (c && (c != '/'));
1428 this.len = name - (const char *) this.name;
1429 this.hash = end_name_hash(hash);
1432 if (this.name[0] == '.') switch (this.len) {
1434 if (this.name[1] == '.') {
1436 nd->flags |= LOOKUP_JUMPED;
1442 if (likely(type == LAST_NORM)) {
1443 struct dentry *parent = nd->path.dentry;
1444 nd->flags &= ~LOOKUP_JUMPED;
1445 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1446 err = parent->d_op->d_hash(parent, nd->inode,
1453 /* remove trailing slashes? */
1455 goto last_component;
1456 while (*++name == '/');
1458 goto last_component;
1460 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1465 err = nested_symlink(&next, nd);
1470 if (!nd->inode->i_op->lookup)
1473 /* here ends the main loop */
1476 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1477 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1479 nd->last_type = type;
1486 static int path_init(int dfd, const char *name, unsigned int flags,
1487 struct nameidata *nd, struct file **fp)
1493 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1494 nd->flags = flags | LOOKUP_JUMPED;
1496 if (flags & LOOKUP_ROOT) {
1497 struct inode *inode = nd->root.dentry->d_inode;
1499 if (!inode->i_op->lookup)
1501 retval = inode_permission(inode, MAY_EXEC);
1505 nd->path = nd->root;
1507 if (flags & LOOKUP_RCU) {
1508 br_read_lock(vfsmount_lock);
1510 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1512 path_get(&nd->path);
1517 nd->root.mnt = NULL;
1520 if (flags & LOOKUP_RCU) {
1521 br_read_lock(vfsmount_lock);
1526 path_get(&nd->root);
1528 nd->path = nd->root;
1529 } else if (dfd == AT_FDCWD) {
1530 if (flags & LOOKUP_RCU) {
1531 struct fs_struct *fs = current->fs;
1534 br_read_lock(vfsmount_lock);
1538 seq = read_seqcount_begin(&fs->seq);
1540 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1541 } while (read_seqcount_retry(&fs->seq, seq));
1543 get_fs_pwd(current->fs, &nd->path);
1546 struct dentry *dentry;
1548 file = fget_raw_light(dfd, &fput_needed);
1553 dentry = file->f_path.dentry;
1557 if (!S_ISDIR(dentry->d_inode->i_mode))
1560 retval = file_permission(file, MAY_EXEC);
1565 nd->path = file->f_path;
1566 if (flags & LOOKUP_RCU) {
1569 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1570 br_read_lock(vfsmount_lock);
1573 path_get(&file->f_path);
1574 fput_light(file, fput_needed);
1578 nd->inode = nd->path.dentry->d_inode;
1582 fput_light(file, fput_needed);
1587 static inline int lookup_last(struct nameidata *nd, struct path *path)
1589 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1590 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1592 nd->flags &= ~LOOKUP_PARENT;
1593 return walk_component(nd, path, &nd->last, nd->last_type,
1594 nd->flags & LOOKUP_FOLLOW);
1597 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1598 static int path_lookupat(int dfd, const char *name,
1599 unsigned int flags, struct nameidata *nd)
1601 struct file *base = NULL;
1606 * Path walking is largely split up into 2 different synchronisation
1607 * schemes, rcu-walk and ref-walk (explained in
1608 * Documentation/filesystems/path-lookup.txt). These share much of the
1609 * path walk code, but some things particularly setup, cleanup, and
1610 * following mounts are sufficiently divergent that functions are
1611 * duplicated. Typically there is a function foo(), and its RCU
1612 * analogue, foo_rcu().
1614 * -ECHILD is the error number of choice (just to avoid clashes) that
1615 * is returned if some aspect of an rcu-walk fails. Such an error must
1616 * be handled by restarting a traditional ref-walk (which will always
1617 * be able to complete).
1619 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1624 current->total_link_count = 0;
1625 err = link_path_walk(name, nd);
1627 if (!err && !(flags & LOOKUP_PARENT)) {
1628 err = lookup_last(nd, &path);
1631 struct path link = path;
1632 nd->flags |= LOOKUP_PARENT;
1633 err = follow_link(&link, nd, &cookie);
1635 err = lookup_last(nd, &path);
1636 put_link(nd, &link, cookie);
1640 if (nd->flags & LOOKUP_RCU) {
1641 /* went all way through without dropping RCU */
1643 if (nameidata_drop_rcu_last(nd))
1648 err = handle_reval_path(nd);
1650 path_put(&nd->path);
1653 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1654 if (!nd->inode->i_op->lookup) {
1655 path_put(&nd->path);
1663 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1664 path_put(&nd->root);
1665 nd->root.mnt = NULL;
1670 static int do_path_lookup(int dfd, const char *name,
1671 unsigned int flags, struct nameidata *nd)
1673 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1674 if (unlikely(retval == -ECHILD))
1675 retval = path_lookupat(dfd, name, flags, nd);
1676 if (unlikely(retval == -ESTALE))
1677 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1679 if (likely(!retval)) {
1680 if (unlikely(!audit_dummy_context())) {
1681 if (nd->path.dentry && nd->inode)
1682 audit_inode(name, nd->path.dentry);
1688 int kern_path_parent(const char *name, struct nameidata *nd)
1690 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1693 int kern_path(const char *name, unsigned int flags, struct path *path)
1695 struct nameidata nd;
1696 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1703 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1704 * @dentry: pointer to dentry of the base directory
1705 * @mnt: pointer to vfs mount of the base directory
1706 * @name: pointer to file name
1707 * @flags: lookup flags
1708 * @nd: pointer to nameidata
1710 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1711 const char *name, unsigned int flags,
1712 struct nameidata *nd)
1714 nd->root.dentry = dentry;
1716 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1717 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
1720 static struct dentry *__lookup_hash(struct qstr *name,
1721 struct dentry *base, struct nameidata *nd)
1723 struct inode *inode = base->d_inode;
1724 struct dentry *dentry;
1727 err = exec_permission(inode, 0);
1729 return ERR_PTR(err);
1732 * Don't bother with __d_lookup: callers are for creat as
1733 * well as unlink, so a lot of the time it would cost
1736 dentry = d_lookup(base, name);
1738 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
1739 dentry = do_revalidate(dentry, nd);
1742 dentry = d_alloc_and_lookup(base, name, nd);
1748 * Restricted form of lookup. Doesn't follow links, single-component only,
1749 * needs parent already locked. Doesn't follow mounts.
1752 static struct dentry *lookup_hash(struct nameidata *nd)
1754 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1758 * lookup_one_len - filesystem helper to lookup single pathname component
1759 * @name: pathname component to lookup
1760 * @base: base directory to lookup from
1761 * @len: maximum length @len should be interpreted to
1763 * Note that this routine is purely a helper for filesystem usage and should
1764 * not be called by generic code. Also note that by using this function the
1765 * nameidata argument is passed to the filesystem methods and a filesystem
1766 * using this helper needs to be prepared for that.
1768 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1774 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1779 return ERR_PTR(-EACCES);
1781 hash = init_name_hash();
1783 c = *(const unsigned char *)name++;
1784 if (c == '/' || c == '\0')
1785 return ERR_PTR(-EACCES);
1786 hash = partial_name_hash(c, hash);
1788 this.hash = end_name_hash(hash);
1790 * See if the low-level filesystem might want
1791 * to use its own hash..
1793 if (base->d_flags & DCACHE_OP_HASH) {
1794 int err = base->d_op->d_hash(base, base->d_inode, &this);
1796 return ERR_PTR(err);
1799 return __lookup_hash(&this, base, NULL);
1802 int user_path_at(int dfd, const char __user *name, unsigned flags,
1805 struct nameidata nd;
1806 char *tmp = getname_flags(name, flags);
1807 int err = PTR_ERR(tmp);
1810 BUG_ON(flags & LOOKUP_PARENT);
1812 err = do_path_lookup(dfd, tmp, flags, &nd);
1820 static int user_path_parent(int dfd, const char __user *path,
1821 struct nameidata *nd, char **name)
1823 char *s = getname(path);
1829 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1839 * It's inline, so penalty for filesystems that don't use sticky bit is
1842 static inline int check_sticky(struct inode *dir, struct inode *inode)
1844 uid_t fsuid = current_fsuid();
1846 if (!(dir->i_mode & S_ISVTX))
1848 if (inode->i_uid == fsuid)
1850 if (dir->i_uid == fsuid)
1852 return !capable(CAP_FOWNER);
1856 * Check whether we can remove a link victim from directory dir, check
1857 * whether the type of victim is right.
1858 * 1. We can't do it if dir is read-only (done in permission())
1859 * 2. We should have write and exec permissions on dir
1860 * 3. We can't remove anything from append-only dir
1861 * 4. We can't do anything with immutable dir (done in permission())
1862 * 5. If the sticky bit on dir is set we should either
1863 * a. be owner of dir, or
1864 * b. be owner of victim, or
1865 * c. have CAP_FOWNER capability
1866 * 6. If the victim is append-only or immutable we can't do antyhing with
1867 * links pointing to it.
1868 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1869 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1870 * 9. We can't remove a root or mountpoint.
1871 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1872 * nfs_async_unlink().
1874 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1878 if (!victim->d_inode)
1881 BUG_ON(victim->d_parent->d_inode != dir);
1882 audit_inode_child(victim, dir);
1884 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1889 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1890 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1893 if (!S_ISDIR(victim->d_inode->i_mode))
1895 if (IS_ROOT(victim))
1897 } else if (S_ISDIR(victim->d_inode->i_mode))
1899 if (IS_DEADDIR(dir))
1901 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1906 /* Check whether we can create an object with dentry child in directory
1908 * 1. We can't do it if child already exists (open has special treatment for
1909 * this case, but since we are inlined it's OK)
1910 * 2. We can't do it if dir is read-only (done in permission())
1911 * 3. We should have write and exec permissions on dir
1912 * 4. We can't do it if dir is immutable (done in permission())
1914 static inline int may_create(struct inode *dir, struct dentry *child)
1918 if (IS_DEADDIR(dir))
1920 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1924 * p1 and p2 should be directories on the same fs.
1926 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1931 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1935 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1937 p = d_ancestor(p2, p1);
1939 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1940 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1944 p = d_ancestor(p1, p2);
1946 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1947 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1951 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1952 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1956 void unlock_rename(struct dentry *p1, struct dentry *p2)
1958 mutex_unlock(&p1->d_inode->i_mutex);
1960 mutex_unlock(&p2->d_inode->i_mutex);
1961 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1965 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1966 struct nameidata *nd)
1968 int error = may_create(dir, dentry);
1973 if (!dir->i_op->create)
1974 return -EACCES; /* shouldn't it be ENOSYS? */
1977 error = security_inode_create(dir, dentry, mode);
1980 error = dir->i_op->create(dir, dentry, mode, nd);
1982 fsnotify_create(dir, dentry);
1986 static int may_open(struct path *path, int acc_mode, int flag)
1988 struct dentry *dentry = path->dentry;
1989 struct inode *inode = dentry->d_inode;
1999 switch (inode->i_mode & S_IFMT) {
2003 if (acc_mode & MAY_WRITE)
2008 if (path->mnt->mnt_flags & MNT_NODEV)
2017 error = inode_permission(inode, acc_mode);
2022 * An append-only file must be opened in append mode for writing.
2024 if (IS_APPEND(inode)) {
2025 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2031 /* O_NOATIME can only be set by the owner or superuser */
2032 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2036 * Ensure there are no outstanding leases on the file.
2038 return break_lease(inode, flag);
2041 static int handle_truncate(struct file *filp)
2043 struct path *path = &filp->f_path;
2044 struct inode *inode = path->dentry->d_inode;
2045 int error = get_write_access(inode);
2049 * Refuse to truncate files with mandatory locks held on them.
2051 error = locks_verify_locked(inode);
2053 error = security_path_truncate(path);
2055 error = do_truncate(path->dentry, 0,
2056 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2059 put_write_access(inode);
2064 * Note that while the flag value (low two bits) for sys_open means:
2069 * it is changed into
2070 * 00 - no permissions needed
2071 * 01 - read-permission
2072 * 10 - write-permission
2074 * for the internal routines (ie open_namei()/follow_link() etc)
2075 * This is more logical, and also allows the 00 "no perm needed"
2076 * to be used for symlinks (where the permissions are checked
2080 static inline int open_to_namei_flags(int flag)
2082 if ((flag+1) & O_ACCMODE)
2088 * Handle the last step of open()
2090 static struct file *do_last(struct nameidata *nd, struct path *path,
2091 const struct open_flags *op, const char *pathname)
2093 struct dentry *dir = nd->path.dentry;
2094 struct dentry *dentry;
2095 int open_flag = op->open_flag;
2096 int will_truncate = open_flag & O_TRUNC;
2098 int acc_mode = op->acc_mode;
2102 nd->flags &= ~LOOKUP_PARENT;
2103 nd->flags |= op->intent;
2105 switch (nd->last_type) {
2108 error = handle_dots(nd, nd->last_type);
2110 return ERR_PTR(error);
2113 if (nd->flags & LOOKUP_RCU) {
2114 if (nameidata_drop_rcu_last(nd))
2115 return ERR_PTR(-ECHILD);
2117 error = handle_reval_path(nd);
2120 audit_inode(pathname, nd->path.dentry);
2121 if (open_flag & O_CREAT) {
2127 /* can't be RCU mode here */
2128 error = handle_reval_path(nd);
2131 audit_inode(pathname, dir);
2135 if (!(open_flag & O_CREAT)) {
2137 if (nd->last.name[nd->last.len])
2138 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2139 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2141 /* we _can_ be in RCU mode here */
2142 error = walk_component(nd, path, &nd->last, LAST_NORM,
2145 return ERR_PTR(error);
2146 if (error) /* symlink */
2149 if (nd->flags & LOOKUP_RCU) {
2150 if (nameidata_drop_rcu_last(nd))
2151 return ERR_PTR(-ECHILD);
2155 if (nd->flags & LOOKUP_DIRECTORY) {
2156 if (!nd->inode->i_op->lookup)
2159 audit_inode(pathname, nd->path.dentry);
2163 /* create side of things */
2165 if (nd->flags & LOOKUP_RCU) {
2166 if (nameidata_drop_rcu_last(nd))
2167 return ERR_PTR(-ECHILD);
2170 audit_inode(pathname, dir);
2172 /* trailing slashes? */
2173 if (nd->last.name[nd->last.len])
2176 mutex_lock(&dir->d_inode->i_mutex);
2178 dentry = lookup_hash(nd);
2179 error = PTR_ERR(dentry);
2180 if (IS_ERR(dentry)) {
2181 mutex_unlock(&dir->d_inode->i_mutex);
2185 path->dentry = dentry;
2186 path->mnt = nd->path.mnt;
2188 /* Negative dentry, just create the file */
2189 if (!dentry->d_inode) {
2190 int mode = op->mode;
2191 if (!IS_POSIXACL(dir->d_inode))
2192 mode &= ~current_umask();
2194 * This write is needed to ensure that a
2195 * rw->ro transition does not occur between
2196 * the time when the file is created and when
2197 * a permanent write count is taken through
2198 * the 'struct file' in nameidata_to_filp().
2200 error = mnt_want_write(nd->path.mnt);
2202 goto exit_mutex_unlock;
2204 /* Don't check for write permission, don't truncate */
2205 open_flag &= ~O_TRUNC;
2207 acc_mode = MAY_OPEN;
2208 error = security_path_mknod(&nd->path, dentry, mode, 0);
2210 goto exit_mutex_unlock;
2211 error = vfs_create(dir->d_inode, dentry, mode, nd);
2213 goto exit_mutex_unlock;
2214 mutex_unlock(&dir->d_inode->i_mutex);
2215 dput(nd->path.dentry);
2216 nd->path.dentry = dentry;
2221 * It already exists.
2223 mutex_unlock(&dir->d_inode->i_mutex);
2224 audit_inode(pathname, path->dentry);
2227 if (open_flag & O_EXCL)
2230 error = follow_managed(path, nd->flags);
2235 if (!path->dentry->d_inode)
2238 if (path->dentry->d_inode->i_op->follow_link)
2241 path_to_nameidata(path, nd);
2242 nd->inode = path->dentry->d_inode;
2244 if (S_ISDIR(nd->inode->i_mode))
2247 if (!S_ISREG(nd->inode->i_mode))
2250 if (will_truncate) {
2251 error = mnt_want_write(nd->path.mnt);
2257 error = may_open(&nd->path, acc_mode, open_flag);
2260 filp = nameidata_to_filp(nd);
2261 if (!IS_ERR(filp)) {
2262 error = ima_file_check(filp, op->acc_mode);
2265 filp = ERR_PTR(error);
2268 if (!IS_ERR(filp)) {
2269 if (will_truncate) {
2270 error = handle_truncate(filp);
2273 filp = ERR_PTR(error);
2279 mnt_drop_write(nd->path.mnt);
2280 path_put(&nd->path);
2284 mutex_unlock(&dir->d_inode->i_mutex);
2286 path_put_conditional(path, nd);
2288 filp = ERR_PTR(error);
2292 static struct file *path_openat(int dfd, const char *pathname,
2293 struct nameidata *nd, const struct open_flags *op, int flags)
2295 struct file *base = NULL;
2300 filp = get_empty_filp();
2302 return ERR_PTR(-ENFILE);
2304 filp->f_flags = op->open_flag;
2305 nd->intent.open.file = filp;
2306 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2307 nd->intent.open.create_mode = op->mode;
2309 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2310 if (unlikely(error))
2313 current->total_link_count = 0;
2314 error = link_path_walk(pathname, nd);
2315 if (unlikely(error))
2318 filp = do_last(nd, &path, op, pathname);
2319 while (unlikely(!filp)) { /* trailing symlink */
2320 struct path link = path;
2322 if (!(nd->flags & LOOKUP_FOLLOW)) {
2323 path_put_conditional(&path, nd);
2324 path_put(&nd->path);
2325 filp = ERR_PTR(-ELOOP);
2328 nd->flags |= LOOKUP_PARENT;
2329 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2330 error = follow_link(&link, nd, &cookie);
2331 if (unlikely(error))
2332 filp = ERR_PTR(error);
2334 filp = do_last(nd, &path, op, pathname);
2335 put_link(nd, &link, cookie);
2338 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2339 path_put(&nd->root);
2342 release_open_intent(nd);
2346 filp = ERR_PTR(error);
2350 struct file *do_filp_open(int dfd, const char *pathname,
2351 const struct open_flags *op, int flags)
2353 struct nameidata nd;
2356 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2357 if (unlikely(filp == ERR_PTR(-ECHILD)))
2358 filp = path_openat(dfd, pathname, &nd, op, flags);
2359 if (unlikely(filp == ERR_PTR(-ESTALE)))
2360 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2364 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2365 const char *name, const struct open_flags *op, int flags)
2367 struct nameidata nd;
2371 nd.root.dentry = dentry;
2373 flags |= LOOKUP_ROOT;
2375 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2376 return ERR_PTR(-ELOOP);
2378 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2379 if (unlikely(file == ERR_PTR(-ECHILD)))
2380 file = path_openat(-1, name, &nd, op, flags);
2381 if (unlikely(file == ERR_PTR(-ESTALE)))
2382 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2387 * lookup_create - lookup a dentry, creating it if it doesn't exist
2388 * @nd: nameidata info
2389 * @is_dir: directory flag
2391 * Simple function to lookup and return a dentry and create it
2392 * if it doesn't exist. Is SMP-safe.
2394 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2396 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2398 struct dentry *dentry = ERR_PTR(-EEXIST);
2400 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2402 * Yucky last component or no last component at all?
2403 * (foo/., foo/.., /////)
2405 if (nd->last_type != LAST_NORM)
2407 nd->flags &= ~LOOKUP_PARENT;
2408 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2409 nd->intent.open.flags = O_EXCL;
2412 * Do the final lookup.
2414 dentry = lookup_hash(nd);
2418 if (dentry->d_inode)
2421 * Special case - lookup gave negative, but... we had foo/bar/
2422 * From the vfs_mknod() POV we just have a negative dentry -
2423 * all is fine. Let's be bastards - you had / on the end, you've
2424 * been asking for (non-existent) directory. -ENOENT for you.
2426 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2428 dentry = ERR_PTR(-ENOENT);
2433 dentry = ERR_PTR(-EEXIST);
2437 EXPORT_SYMBOL_GPL(lookup_create);
2439 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2441 int error = may_create(dir, dentry);
2446 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2449 if (!dir->i_op->mknod)
2452 error = devcgroup_inode_mknod(mode, dev);
2456 error = security_inode_mknod(dir, dentry, mode, dev);
2460 error = dir->i_op->mknod(dir, dentry, mode, dev);
2462 fsnotify_create(dir, dentry);
2466 static int may_mknod(mode_t mode)
2468 switch (mode & S_IFMT) {
2474 case 0: /* zero mode translates to S_IFREG */
2483 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2488 struct dentry *dentry;
2489 struct nameidata nd;
2494 error = user_path_parent(dfd, filename, &nd, &tmp);
2498 dentry = lookup_create(&nd, 0);
2499 if (IS_ERR(dentry)) {
2500 error = PTR_ERR(dentry);
2503 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2504 mode &= ~current_umask();
2505 error = may_mknod(mode);
2508 error = mnt_want_write(nd.path.mnt);
2511 error = security_path_mknod(&nd.path, dentry, mode, dev);
2513 goto out_drop_write;
2514 switch (mode & S_IFMT) {
2515 case 0: case S_IFREG:
2516 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2518 case S_IFCHR: case S_IFBLK:
2519 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2520 new_decode_dev(dev));
2522 case S_IFIFO: case S_IFSOCK:
2523 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2527 mnt_drop_write(nd.path.mnt);
2531 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2538 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2540 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2543 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2545 int error = may_create(dir, dentry);
2550 if (!dir->i_op->mkdir)
2553 mode &= (S_IRWXUGO|S_ISVTX);
2554 error = security_inode_mkdir(dir, dentry, mode);
2558 error = dir->i_op->mkdir(dir, dentry, mode);
2560 fsnotify_mkdir(dir, dentry);
2564 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2568 struct dentry *dentry;
2569 struct nameidata nd;
2571 error = user_path_parent(dfd, pathname, &nd, &tmp);
2575 dentry = lookup_create(&nd, 1);
2576 error = PTR_ERR(dentry);
2580 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2581 mode &= ~current_umask();
2582 error = mnt_want_write(nd.path.mnt);
2585 error = security_path_mkdir(&nd.path, dentry, mode);
2587 goto out_drop_write;
2588 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2590 mnt_drop_write(nd.path.mnt);
2594 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2601 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2603 return sys_mkdirat(AT_FDCWD, pathname, mode);
2607 * We try to drop the dentry early: we should have
2608 * a usage count of 2 if we're the only user of this
2609 * dentry, and if that is true (possibly after pruning
2610 * the dcache), then we drop the dentry now.
2612 * A low-level filesystem can, if it choses, legally
2615 * if (!d_unhashed(dentry))
2618 * if it cannot handle the case of removing a directory
2619 * that is still in use by something else..
2621 void dentry_unhash(struct dentry *dentry)
2624 shrink_dcache_parent(dentry);
2625 spin_lock(&dentry->d_lock);
2626 if (dentry->d_count == 2)
2628 spin_unlock(&dentry->d_lock);
2631 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2633 int error = may_delete(dir, dentry, 1);
2638 if (!dir->i_op->rmdir)
2641 mutex_lock(&dentry->d_inode->i_mutex);
2642 dentry_unhash(dentry);
2643 if (d_mountpoint(dentry))
2646 error = security_inode_rmdir(dir, dentry);
2648 error = dir->i_op->rmdir(dir, dentry);
2650 dentry->d_inode->i_flags |= S_DEAD;
2655 mutex_unlock(&dentry->d_inode->i_mutex);
2664 static long do_rmdir(int dfd, const char __user *pathname)
2668 struct dentry *dentry;
2669 struct nameidata nd;
2671 error = user_path_parent(dfd, pathname, &nd, &name);
2675 switch(nd.last_type) {
2687 nd.flags &= ~LOOKUP_PARENT;
2689 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2690 dentry = lookup_hash(&nd);
2691 error = PTR_ERR(dentry);
2694 error = mnt_want_write(nd.path.mnt);
2697 error = security_path_rmdir(&nd.path, dentry);
2700 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2702 mnt_drop_write(nd.path.mnt);
2706 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2713 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2715 return do_rmdir(AT_FDCWD, pathname);
2718 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2720 int error = may_delete(dir, dentry, 0);
2725 if (!dir->i_op->unlink)
2728 mutex_lock(&dentry->d_inode->i_mutex);
2729 if (d_mountpoint(dentry))
2732 error = security_inode_unlink(dir, dentry);
2734 error = dir->i_op->unlink(dir, dentry);
2739 mutex_unlock(&dentry->d_inode->i_mutex);
2741 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2742 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2743 fsnotify_link_count(dentry->d_inode);
2751 * Make sure that the actual truncation of the file will occur outside its
2752 * directory's i_mutex. Truncate can take a long time if there is a lot of
2753 * writeout happening, and we don't want to prevent access to the directory
2754 * while waiting on the I/O.
2756 static long do_unlinkat(int dfd, const char __user *pathname)
2760 struct dentry *dentry;
2761 struct nameidata nd;
2762 struct inode *inode = NULL;
2764 error = user_path_parent(dfd, pathname, &nd, &name);
2769 if (nd.last_type != LAST_NORM)
2772 nd.flags &= ~LOOKUP_PARENT;
2774 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2775 dentry = lookup_hash(&nd);
2776 error = PTR_ERR(dentry);
2777 if (!IS_ERR(dentry)) {
2778 /* Why not before? Because we want correct error value */
2779 if (nd.last.name[nd.last.len])
2781 inode = dentry->d_inode;
2784 error = mnt_want_write(nd.path.mnt);
2787 error = security_path_unlink(&nd.path, dentry);
2790 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2792 mnt_drop_write(nd.path.mnt);
2796 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2798 iput(inode); /* truncate the inode here */
2805 error = !dentry->d_inode ? -ENOENT :
2806 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2810 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2812 if ((flag & ~AT_REMOVEDIR) != 0)
2815 if (flag & AT_REMOVEDIR)
2816 return do_rmdir(dfd, pathname);
2818 return do_unlinkat(dfd, pathname);
2821 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2823 return do_unlinkat(AT_FDCWD, pathname);
2826 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2828 int error = may_create(dir, dentry);
2833 if (!dir->i_op->symlink)
2836 error = security_inode_symlink(dir, dentry, oldname);
2840 error = dir->i_op->symlink(dir, dentry, oldname);
2842 fsnotify_create(dir, dentry);
2846 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2847 int, newdfd, const char __user *, newname)
2852 struct dentry *dentry;
2853 struct nameidata nd;
2855 from = getname(oldname);
2857 return PTR_ERR(from);
2859 error = user_path_parent(newdfd, newname, &nd, &to);
2863 dentry = lookup_create(&nd, 0);
2864 error = PTR_ERR(dentry);
2868 error = mnt_want_write(nd.path.mnt);
2871 error = security_path_symlink(&nd.path, dentry, from);
2873 goto out_drop_write;
2874 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2876 mnt_drop_write(nd.path.mnt);
2880 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2888 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2890 return sys_symlinkat(oldname, AT_FDCWD, newname);
2893 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2895 struct inode *inode = old_dentry->d_inode;
2901 error = may_create(dir, new_dentry);
2905 if (dir->i_sb != inode->i_sb)
2909 * A link to an append-only or immutable file cannot be created.
2911 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2913 if (!dir->i_op->link)
2915 if (S_ISDIR(inode->i_mode))
2918 error = security_inode_link(old_dentry, dir, new_dentry);
2922 mutex_lock(&inode->i_mutex);
2923 /* Make sure we don't allow creating hardlink to an unlinked file */
2924 if (inode->i_nlink == 0)
2927 error = dir->i_op->link(old_dentry, dir, new_dentry);
2928 mutex_unlock(&inode->i_mutex);
2930 fsnotify_link(dir, inode, new_dentry);
2935 * Hardlinks are often used in delicate situations. We avoid
2936 * security-related surprises by not following symlinks on the
2939 * We don't follow them on the oldname either to be compatible
2940 * with linux 2.0, and to avoid hard-linking to directories
2941 * and other special files. --ADM
2943 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2944 int, newdfd, const char __user *, newname, int, flags)
2946 struct dentry *new_dentry;
2947 struct nameidata nd;
2948 struct path old_path;
2953 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
2956 * To use null names we require CAP_DAC_READ_SEARCH
2957 * This ensures that not everyone will be able to create
2958 * handlink using the passed filedescriptor.
2960 if (flags & AT_EMPTY_PATH) {
2961 if (!capable(CAP_DAC_READ_SEARCH))
2966 if (flags & AT_SYMLINK_FOLLOW)
2967 how |= LOOKUP_FOLLOW;
2969 error = user_path_at(olddfd, oldname, how, &old_path);
2973 error = user_path_parent(newdfd, newname, &nd, &to);
2977 if (old_path.mnt != nd.path.mnt)
2979 new_dentry = lookup_create(&nd, 0);
2980 error = PTR_ERR(new_dentry);
2981 if (IS_ERR(new_dentry))
2983 error = mnt_want_write(nd.path.mnt);
2986 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2988 goto out_drop_write;
2989 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2991 mnt_drop_write(nd.path.mnt);
2995 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3000 path_put(&old_path);
3005 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3007 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3011 * The worst of all namespace operations - renaming directory. "Perverted"
3012 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3014 * a) we can get into loop creation. Check is done in is_subdir().
3015 * b) race potential - two innocent renames can create a loop together.
3016 * That's where 4.4 screws up. Current fix: serialization on
3017 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3019 * c) we have to lock _three_ objects - parents and victim (if it exists).
3020 * And that - after we got ->i_mutex on parents (until then we don't know
3021 * whether the target exists). Solution: try to be smart with locking
3022 * order for inodes. We rely on the fact that tree topology may change
3023 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3024 * move will be locked. Thus we can rank directories by the tree
3025 * (ancestors first) and rank all non-directories after them.
3026 * That works since everybody except rename does "lock parent, lookup,
3027 * lock child" and rename is under ->s_vfs_rename_mutex.
3028 * HOWEVER, it relies on the assumption that any object with ->lookup()
3029 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3030 * we'd better make sure that there's no link(2) for them.
3031 * d) some filesystems don't support opened-but-unlinked directories,
3032 * either because of layout or because they are not ready to deal with
3033 * all cases correctly. The latter will be fixed (taking this sort of
3034 * stuff into VFS), but the former is not going away. Solution: the same
3035 * trick as in rmdir().
3036 * e) conversion from fhandle to dentry may come in the wrong moment - when
3037 * we are removing the target. Solution: we will have to grab ->i_mutex
3038 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3039 * ->i_mutex on parents, which works but leads to some truly excessive
3042 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3043 struct inode *new_dir, struct dentry *new_dentry)
3046 struct inode *target;
3049 * If we are going to change the parent - check write permissions,
3050 * we'll need to flip '..'.
3052 if (new_dir != old_dir) {
3053 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3058 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3062 target = new_dentry->d_inode;
3064 mutex_lock(&target->i_mutex);
3065 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3069 dentry_unhash(new_dentry);
3070 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3074 target->i_flags |= S_DEAD;
3075 dont_mount(new_dentry);
3077 mutex_unlock(&target->i_mutex);
3078 if (d_unhashed(new_dentry))
3079 d_rehash(new_dentry);
3083 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3084 d_move(old_dentry,new_dentry);
3088 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3089 struct inode *new_dir, struct dentry *new_dentry)
3091 struct inode *target;
3094 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3099 target = new_dentry->d_inode;
3101 mutex_lock(&target->i_mutex);
3102 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3105 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3108 dont_mount(new_dentry);
3109 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3110 d_move(old_dentry, new_dentry);
3113 mutex_unlock(&target->i_mutex);
3118 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3119 struct inode *new_dir, struct dentry *new_dentry)
3122 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3123 const unsigned char *old_name;
3125 if (old_dentry->d_inode == new_dentry->d_inode)
3128 error = may_delete(old_dir, old_dentry, is_dir);
3132 if (!new_dentry->d_inode)
3133 error = may_create(new_dir, new_dentry);
3135 error = may_delete(new_dir, new_dentry, is_dir);
3139 if (!old_dir->i_op->rename)
3142 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3145 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3147 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3149 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3150 new_dentry->d_inode, old_dentry);
3151 fsnotify_oldname_free(old_name);
3156 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3157 int, newdfd, const char __user *, newname)
3159 struct dentry *old_dir, *new_dir;
3160 struct dentry *old_dentry, *new_dentry;
3161 struct dentry *trap;
3162 struct nameidata oldnd, newnd;
3167 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3171 error = user_path_parent(newdfd, newname, &newnd, &to);
3176 if (oldnd.path.mnt != newnd.path.mnt)
3179 old_dir = oldnd.path.dentry;
3181 if (oldnd.last_type != LAST_NORM)
3184 new_dir = newnd.path.dentry;
3185 if (newnd.last_type != LAST_NORM)
3188 oldnd.flags &= ~LOOKUP_PARENT;
3189 newnd.flags &= ~LOOKUP_PARENT;
3190 newnd.flags |= LOOKUP_RENAME_TARGET;
3192 trap = lock_rename(new_dir, old_dir);
3194 old_dentry = lookup_hash(&oldnd);
3195 error = PTR_ERR(old_dentry);
3196 if (IS_ERR(old_dentry))
3198 /* source must exist */
3200 if (!old_dentry->d_inode)
3202 /* unless the source is a directory trailing slashes give -ENOTDIR */
3203 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3205 if (oldnd.last.name[oldnd.last.len])
3207 if (newnd.last.name[newnd.last.len])
3210 /* source should not be ancestor of target */
3212 if (old_dentry == trap)
3214 new_dentry = lookup_hash(&newnd);
3215 error = PTR_ERR(new_dentry);
3216 if (IS_ERR(new_dentry))
3218 /* target should not be an ancestor of source */
3220 if (new_dentry == trap)
3223 error = mnt_want_write(oldnd.path.mnt);
3226 error = security_path_rename(&oldnd.path, old_dentry,
3227 &newnd.path, new_dentry);
3230 error = vfs_rename(old_dir->d_inode, old_dentry,
3231 new_dir->d_inode, new_dentry);
3233 mnt_drop_write(oldnd.path.mnt);
3239 unlock_rename(new_dir, old_dir);
3241 path_put(&newnd.path);
3244 path_put(&oldnd.path);
3250 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3252 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3255 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3259 len = PTR_ERR(link);
3264 if (len > (unsigned) buflen)
3266 if (copy_to_user(buffer, link, len))
3273 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3274 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3275 * using) it for any given inode is up to filesystem.
3277 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3279 struct nameidata nd;
3284 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3286 return PTR_ERR(cookie);
3288 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3289 if (dentry->d_inode->i_op->put_link)
3290 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3294 int vfs_follow_link(struct nameidata *nd, const char *link)
3296 return __vfs_follow_link(nd, link);
3299 /* get the link contents into pagecache */
3300 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3304 struct address_space *mapping = dentry->d_inode->i_mapping;
3305 page = read_mapping_page(mapping, 0, NULL);
3310 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3314 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3316 struct page *page = NULL;
3317 char *s = page_getlink(dentry, &page);
3318 int res = vfs_readlink(dentry,buffer,buflen,s);
3321 page_cache_release(page);
3326 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3328 struct page *page = NULL;
3329 nd_set_link(nd, page_getlink(dentry, &page));
3333 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3335 struct page *page = cookie;
3339 page_cache_release(page);
3344 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3346 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3348 struct address_space *mapping = inode->i_mapping;
3353 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3355 flags |= AOP_FLAG_NOFS;
3358 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3359 flags, &page, &fsdata);
3363 kaddr = kmap_atomic(page, KM_USER0);
3364 memcpy(kaddr, symname, len-1);
3365 kunmap_atomic(kaddr, KM_USER0);
3367 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3374 mark_inode_dirty(inode);
3380 int page_symlink(struct inode *inode, const char *symname, int len)
3382 return __page_symlink(inode, symname, len,
3383 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3386 const struct inode_operations page_symlink_inode_operations = {
3387 .readlink = generic_readlink,
3388 .follow_link = page_follow_link_light,
3389 .put_link = page_put_link,
3392 EXPORT_SYMBOL(user_path_at);
3393 EXPORT_SYMBOL(follow_down_one);
3394 EXPORT_SYMBOL(follow_down);
3395 EXPORT_SYMBOL(follow_up);
3396 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3397 EXPORT_SYMBOL(getname);
3398 EXPORT_SYMBOL(lock_rename);
3399 EXPORT_SYMBOL(lookup_one_len);
3400 EXPORT_SYMBOL(page_follow_link_light);
3401 EXPORT_SYMBOL(page_put_link);
3402 EXPORT_SYMBOL(page_readlink);
3403 EXPORT_SYMBOL(__page_symlink);
3404 EXPORT_SYMBOL(page_symlink);
3405 EXPORT_SYMBOL(page_symlink_inode_operations);
3406 EXPORT_SYMBOL(kern_path_parent);
3407 EXPORT_SYMBOL(kern_path);
3408 EXPORT_SYMBOL(vfs_path_lookup);
3409 EXPORT_SYMBOL(inode_permission);
3410 EXPORT_SYMBOL(file_permission);
3411 EXPORT_SYMBOL(unlock_rename);
3412 EXPORT_SYMBOL(vfs_create);
3413 EXPORT_SYMBOL(vfs_follow_link);
3414 EXPORT_SYMBOL(vfs_link);
3415 EXPORT_SYMBOL(vfs_mkdir);
3416 EXPORT_SYMBOL(vfs_mknod);
3417 EXPORT_SYMBOL(generic_permission);
3418 EXPORT_SYMBOL(vfs_readlink);
3419 EXPORT_SYMBOL(vfs_rename);
3420 EXPORT_SYMBOL(vfs_rmdir);
3421 EXPORT_SYMBOL(vfs_symlink);
3422 EXPORT_SYMBOL(vfs_unlink);
3423 EXPORT_SYMBOL(dentry_unhash);
3424 EXPORT_SYMBOL(generic_readlink);