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_user_ns() != inode_userns(inode))
189 if (current_fsuid() == inode->i_uid)
192 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
193 int error = check_acl(inode, mask, flags);
194 if (error != -EAGAIN)
198 if (in_group_p(inode->i_gid))
204 * If the DACs are ok we don't need any capability check.
206 if ((mask & ~mode) == 0)
212 * generic_permission - check for access rights on a Posix-like filesystem
213 * @inode: inode to check access rights for
214 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
215 * @check_acl: optional callback to check for Posix ACLs
216 * @flags: IPERM_FLAG_ flags.
218 * Used to check for read/write/execute permissions on a file.
219 * We use "fsuid" for this, letting us set arbitrary permissions
220 * for filesystem access without changing the "normal" uids which
221 * are used for other things.
223 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
224 * request cannot be satisfied (eg. requires blocking or too much complexity).
225 * It would then be called again in ref-walk mode.
227 int generic_permission(struct inode *inode, int mask, unsigned int flags,
228 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
233 * Do the basic POSIX ACL permission checks.
235 ret = acl_permission_check(inode, mask, flags, check_acl);
240 * Read/write DACs are always overridable.
241 * Executable DACs are overridable if at least one exec bit is set.
243 if (!(mask & MAY_EXEC) || execute_ok(inode))
244 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
248 * Searching includes executable on directories, else just read.
250 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
251 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
252 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
259 * inode_permission - check for access rights to a given inode
260 * @inode: inode to check permission on
261 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
263 * Used to check for read/write/execute permissions on an inode.
264 * We use "fsuid" for this, letting us set arbitrary permissions
265 * for filesystem access without changing the "normal" uids which
266 * are used for other things.
268 int inode_permission(struct inode *inode, int mask)
272 if (mask & MAY_WRITE) {
273 umode_t mode = inode->i_mode;
276 * Nobody gets write access to a read-only fs.
278 if (IS_RDONLY(inode) &&
279 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
283 * Nobody gets write access to an immutable file.
285 if (IS_IMMUTABLE(inode))
289 if (inode->i_op->permission)
290 retval = inode->i_op->permission(inode, mask, 0);
292 retval = generic_permission(inode, mask, 0,
293 inode->i_op->check_acl);
298 retval = devcgroup_inode_permission(inode, mask);
302 return security_inode_permission(inode, mask);
306 * file_permission - check for additional access rights to a given file
307 * @file: file to check access rights for
308 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
310 * Used to check for read/write/execute permissions on an already opened
314 * Do not use this function in new code. All access checks should
315 * be done using inode_permission().
317 int file_permission(struct file *file, int mask)
319 return inode_permission(file->f_path.dentry->d_inode, mask);
323 * get_write_access() gets write permission for a file.
324 * put_write_access() releases this write permission.
325 * This is used for regular files.
326 * We cannot support write (and maybe mmap read-write shared) accesses and
327 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
328 * can have the following values:
329 * 0: no writers, no VM_DENYWRITE mappings
330 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
331 * > 0: (i_writecount) users are writing to the file.
333 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
334 * except for the cases where we don't hold i_writecount yet. Then we need to
335 * use {get,deny}_write_access() - these functions check the sign and refuse
336 * to do the change if sign is wrong. Exclusion between them is provided by
337 * the inode->i_lock spinlock.
340 int get_write_access(struct inode * inode)
342 spin_lock(&inode->i_lock);
343 if (atomic_read(&inode->i_writecount) < 0) {
344 spin_unlock(&inode->i_lock);
347 atomic_inc(&inode->i_writecount);
348 spin_unlock(&inode->i_lock);
353 int deny_write_access(struct file * file)
355 struct inode *inode = file->f_path.dentry->d_inode;
357 spin_lock(&inode->i_lock);
358 if (atomic_read(&inode->i_writecount) > 0) {
359 spin_unlock(&inode->i_lock);
362 atomic_dec(&inode->i_writecount);
363 spin_unlock(&inode->i_lock);
369 * path_get - get a reference to a path
370 * @path: path to get the reference to
372 * Given a path increment the reference count to the dentry and the vfsmount.
374 void path_get(struct path *path)
379 EXPORT_SYMBOL(path_get);
382 * path_put - put a reference to a path
383 * @path: path to put the reference to
385 * Given a path decrement the reference count to the dentry and the vfsmount.
387 void path_put(struct path *path)
392 EXPORT_SYMBOL(path_put);
395 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
396 * @nd: nameidata pathwalk data to drop
397 * Returns: 0 on success, -ECHILD on failure
399 * Path walking has 2 modes, rcu-walk and ref-walk (see
400 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
401 * to drop out of rcu-walk mode and take normal reference counts on dentries
402 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
403 * refcounts at the last known good point before rcu-walk got stuck, so
404 * ref-walk may continue from there. If this is not successful (eg. a seqcount
405 * has changed), then failure is returned and path walk restarts from the
406 * beginning in ref-walk mode.
408 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
409 * ref-walk. Must be called from rcu-walk context.
411 static int nameidata_drop_rcu(struct nameidata *nd)
413 struct fs_struct *fs = current->fs;
414 struct dentry *dentry = nd->path.dentry;
417 BUG_ON(!(nd->flags & LOOKUP_RCU));
418 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
420 spin_lock(&fs->lock);
421 if (nd->root.mnt != fs->root.mnt ||
422 nd->root.dentry != fs->root.dentry)
425 spin_lock(&dentry->d_lock);
426 if (!__d_rcu_to_refcount(dentry, nd->seq))
428 BUG_ON(nd->inode != dentry->d_inode);
429 spin_unlock(&dentry->d_lock);
432 spin_unlock(&fs->lock);
434 mntget(nd->path.mnt);
437 br_read_unlock(vfsmount_lock);
438 nd->flags &= ~LOOKUP_RCU;
441 spin_unlock(&dentry->d_lock);
444 spin_unlock(&fs->lock);
448 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
449 static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
451 if (nd->flags & LOOKUP_RCU)
452 return nameidata_drop_rcu(nd);
457 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
458 * @nd: nameidata pathwalk data to drop
459 * @dentry: dentry to drop
460 * Returns: 0 on success, -ECHILD on failure
462 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
463 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
464 * @nd. Must be called from rcu-walk context.
466 static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
468 struct fs_struct *fs = current->fs;
469 struct dentry *parent = nd->path.dentry;
472 BUG_ON(!(nd->flags & LOOKUP_RCU));
473 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
475 spin_lock(&fs->lock);
476 if (nd->root.mnt != fs->root.mnt ||
477 nd->root.dentry != fs->root.dentry)
480 spin_lock(&parent->d_lock);
481 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
482 if (!__d_rcu_to_refcount(dentry, nd->seq))
485 * If the sequence check on the child dentry passed, then the child has
486 * not been removed from its parent. This means the parent dentry must
487 * be valid and able to take a reference at this point.
489 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
490 BUG_ON(!parent->d_count);
492 spin_unlock(&dentry->d_lock);
493 spin_unlock(&parent->d_lock);
496 spin_unlock(&fs->lock);
498 mntget(nd->path.mnt);
501 br_read_unlock(vfsmount_lock);
502 nd->flags &= ~LOOKUP_RCU;
505 spin_unlock(&dentry->d_lock);
506 spin_unlock(&parent->d_lock);
509 spin_unlock(&fs->lock);
513 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
514 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
516 if (nd->flags & LOOKUP_RCU) {
517 if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) {
518 nd->flags &= ~LOOKUP_RCU;
519 if (!(nd->flags & LOOKUP_ROOT))
522 br_read_unlock(vfsmount_lock);
530 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
531 * @nd: nameidata pathwalk data to drop
532 * Returns: 0 on success, -ECHILD on failure
534 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
535 * nd->path should be the final element of the lookup, so nd->root is discarded.
536 * Must be called from rcu-walk context.
538 static int nameidata_drop_rcu_last(struct nameidata *nd)
540 struct dentry *dentry = nd->path.dentry;
542 BUG_ON(!(nd->flags & LOOKUP_RCU));
543 nd->flags &= ~LOOKUP_RCU;
544 if (!(nd->flags & LOOKUP_ROOT))
546 spin_lock(&dentry->d_lock);
547 if (!__d_rcu_to_refcount(dentry, nd->seq))
549 BUG_ON(nd->inode != dentry->d_inode);
550 spin_unlock(&dentry->d_lock);
552 mntget(nd->path.mnt);
555 br_read_unlock(vfsmount_lock);
560 spin_unlock(&dentry->d_lock);
562 br_read_unlock(vfsmount_lock);
567 * release_open_intent - free up open intent resources
568 * @nd: pointer to nameidata
570 void release_open_intent(struct nameidata *nd)
572 struct file *file = nd->intent.open.file;
574 if (file && !IS_ERR(file)) {
575 if (file->f_path.dentry == NULL)
582 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
584 return dentry->d_op->d_revalidate(dentry, nd);
587 static struct dentry *
588 do_revalidate(struct dentry *dentry, struct nameidata *nd)
590 int status = d_revalidate(dentry, nd);
591 if (unlikely(status <= 0)) {
593 * The dentry failed validation.
594 * If d_revalidate returned 0 attempt to invalidate
595 * the dentry otherwise d_revalidate is asking us
596 * to return a fail status.
600 dentry = ERR_PTR(status);
601 } else if (!d_invalidate(dentry)) {
610 * handle_reval_path - force revalidation of a dentry
612 * In some situations the path walking code will trust dentries without
613 * revalidating them. This causes problems for filesystems that depend on
614 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
615 * (which indicates that it's possible for the dentry to go stale), force
616 * a d_revalidate call before proceeding.
618 * Returns 0 if the revalidation was successful. If the revalidation fails,
619 * either return the error returned by d_revalidate or -ESTALE if the
620 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
621 * invalidate the dentry. It's up to the caller to handle putting references
622 * to the path if necessary.
624 static inline int handle_reval_path(struct nameidata *nd)
626 struct dentry *dentry = nd->path.dentry;
629 if (likely(!(nd->flags & LOOKUP_JUMPED)))
632 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
635 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
638 /* Note: we do not d_invalidate() */
639 status = d_revalidate(dentry, nd);
650 * Short-cut version of permission(), for calling on directories
651 * during pathname resolution. Combines parts of permission()
652 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
654 * If appropriate, check DAC only. If not appropriate, or
655 * short-cut DAC fails, then call ->permission() to do more
656 * complete permission check.
658 static inline int exec_permission(struct inode *inode, unsigned int flags)
661 struct user_namespace *ns = inode_userns(inode);
663 if (inode->i_op->permission) {
664 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
666 ret = acl_permission_check(inode, MAY_EXEC, flags,
667 inode->i_op->check_acl);
674 if (ns_capable(ns, CAP_DAC_OVERRIDE) ||
675 ns_capable(ns, CAP_DAC_READ_SEARCH))
680 return security_inode_exec_permission(inode, flags);
683 static __always_inline void set_root(struct nameidata *nd)
686 get_fs_root(current->fs, &nd->root);
689 static int link_path_walk(const char *, struct nameidata *);
691 static __always_inline void set_root_rcu(struct nameidata *nd)
694 struct fs_struct *fs = current->fs;
698 seq = read_seqcount_begin(&fs->seq);
700 } while (read_seqcount_retry(&fs->seq, seq));
704 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
716 nd->flags |= LOOKUP_JUMPED;
718 nd->inode = nd->path.dentry->d_inode;
720 ret = link_path_walk(link, nd);
724 return PTR_ERR(link);
727 static void path_put_conditional(struct path *path, struct nameidata *nd)
730 if (path->mnt != nd->path.mnt)
734 static inline void path_to_nameidata(const struct path *path,
735 struct nameidata *nd)
737 if (!(nd->flags & LOOKUP_RCU)) {
738 dput(nd->path.dentry);
739 if (nd->path.mnt != path->mnt)
740 mntput(nd->path.mnt);
742 nd->path.mnt = path->mnt;
743 nd->path.dentry = path->dentry;
746 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
748 struct inode *inode = link->dentry->d_inode;
749 if (!IS_ERR(cookie) && inode->i_op->put_link)
750 inode->i_op->put_link(link->dentry, nd, cookie);
754 static __always_inline int
755 follow_link(struct path *link, struct nameidata *nd, void **p)
758 struct dentry *dentry = link->dentry;
760 BUG_ON(nd->flags & LOOKUP_RCU);
762 if (link->mnt == nd->path.mnt)
765 if (unlikely(current->total_link_count >= 40)) {
766 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
771 current->total_link_count++;
773 touch_atime(link->mnt, dentry);
774 nd_set_link(nd, NULL);
776 error = security_inode_follow_link(link->dentry, nd);
778 *p = ERR_PTR(error); /* no ->put_link(), please */
783 nd->last_type = LAST_BIND;
784 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
787 char *s = nd_get_link(nd);
790 error = __vfs_follow_link(nd, s);
791 else if (nd->last_type == LAST_BIND) {
792 nd->flags |= LOOKUP_JUMPED;
793 nd->inode = nd->path.dentry->d_inode;
794 if (nd->inode->i_op->follow_link) {
795 /* stepped on a _really_ weird one */
804 static int follow_up_rcu(struct path *path)
806 struct vfsmount *parent;
807 struct dentry *mountpoint;
809 parent = path->mnt->mnt_parent;
810 if (parent == path->mnt)
812 mountpoint = path->mnt->mnt_mountpoint;
813 path->dentry = mountpoint;
818 int follow_up(struct path *path)
820 struct vfsmount *parent;
821 struct dentry *mountpoint;
823 br_read_lock(vfsmount_lock);
824 parent = path->mnt->mnt_parent;
825 if (parent == path->mnt) {
826 br_read_unlock(vfsmount_lock);
830 mountpoint = dget(path->mnt->mnt_mountpoint);
831 br_read_unlock(vfsmount_lock);
833 path->dentry = mountpoint;
840 * Perform an automount
841 * - return -EISDIR to tell follow_managed() to stop and return the path we
844 static int follow_automount(struct path *path, unsigned flags,
847 struct vfsmount *mnt;
850 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
853 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
854 * and this is the terminal part of the path.
856 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
857 return -EISDIR; /* we actually want to stop here */
859 /* We want to mount if someone is trying to open/create a file of any
860 * type under the mountpoint, wants to traverse through the mountpoint
861 * or wants to open the mounted directory.
863 * We don't want to mount if someone's just doing a stat and they've
864 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
865 * appended a '/' to the name.
867 if (!(flags & LOOKUP_FOLLOW) &&
868 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
869 LOOKUP_OPEN | LOOKUP_CREATE)))
872 current->total_link_count++;
873 if (current->total_link_count >= 40)
876 mnt = path->dentry->d_op->d_automount(path);
879 * The filesystem is allowed to return -EISDIR here to indicate
880 * it doesn't want to automount. For instance, autofs would do
881 * this so that its userspace daemon can mount on this dentry.
883 * However, we can only permit this if it's a terminal point in
884 * the path being looked up; if it wasn't then the remainder of
885 * the path is inaccessible and we should say so.
887 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
892 if (!mnt) /* mount collision */
895 err = finish_automount(mnt, path);
899 /* Someone else made a mount here whilst we were busy */
906 path->dentry = dget(mnt->mnt_root);
916 * Handle a dentry that is managed in some way.
917 * - Flagged for transit management (autofs)
918 * - Flagged as mountpoint
919 * - Flagged as automount point
921 * This may only be called in refwalk mode.
923 * Serialization is taken care of in namespace.c
925 static int follow_managed(struct path *path, unsigned flags)
928 bool need_mntput = false;
931 /* Given that we're not holding a lock here, we retain the value in a
932 * local variable for each dentry as we look at it so that we don't see
933 * the components of that value change under us */
934 while (managed = ACCESS_ONCE(path->dentry->d_flags),
935 managed &= DCACHE_MANAGED_DENTRY,
936 unlikely(managed != 0)) {
937 /* Allow the filesystem to manage the transit without i_mutex
939 if (managed & DCACHE_MANAGE_TRANSIT) {
940 BUG_ON(!path->dentry->d_op);
941 BUG_ON(!path->dentry->d_op->d_manage);
942 ret = path->dentry->d_op->d_manage(path->dentry, false);
944 return ret == -EISDIR ? 0 : ret;
947 /* Transit to a mounted filesystem. */
948 if (managed & DCACHE_MOUNTED) {
949 struct vfsmount *mounted = lookup_mnt(path);
955 path->dentry = dget(mounted->mnt_root);
960 /* Something is mounted on this dentry in another
961 * namespace and/or whatever was mounted there in this
962 * namespace got unmounted before we managed to get the
966 /* Handle an automount point */
967 if (managed & DCACHE_NEED_AUTOMOUNT) {
968 ret = follow_automount(path, flags, &need_mntput);
970 return ret == -EISDIR ? 0 : ret;
974 /* We didn't change the current path point */
980 int follow_down_one(struct path *path)
982 struct vfsmount *mounted;
984 mounted = lookup_mnt(path);
989 path->dentry = dget(mounted->mnt_root);
996 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
997 * meet a managed dentry and we're not walking to "..". True is returned to
998 * continue, false to abort.
1000 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1001 struct inode **inode, bool reverse_transit)
1003 while (d_mountpoint(path->dentry)) {
1004 struct vfsmount *mounted;
1005 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1007 path->dentry->d_op->d_manage(path->dentry, true) < 0)
1009 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1012 path->mnt = mounted;
1013 path->dentry = mounted->mnt_root;
1014 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1015 *inode = path->dentry->d_inode;
1018 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1019 return reverse_transit;
1023 static int follow_dotdot_rcu(struct nameidata *nd)
1025 struct inode *inode = nd->inode;
1030 if (nd->path.dentry == nd->root.dentry &&
1031 nd->path.mnt == nd->root.mnt) {
1034 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1035 struct dentry *old = nd->path.dentry;
1036 struct dentry *parent = old->d_parent;
1039 seq = read_seqcount_begin(&parent->d_seq);
1040 if (read_seqcount_retry(&old->d_seq, nd->seq))
1042 inode = parent->d_inode;
1043 nd->path.dentry = parent;
1047 if (!follow_up_rcu(&nd->path))
1049 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1050 inode = nd->path.dentry->d_inode;
1052 __follow_mount_rcu(nd, &nd->path, &inode, true);
1057 nd->flags &= ~LOOKUP_RCU;
1058 if (!(nd->flags & LOOKUP_ROOT))
1059 nd->root.mnt = NULL;
1061 br_read_unlock(vfsmount_lock);
1066 * Follow down to the covering mount currently visible to userspace. At each
1067 * point, the filesystem owning that dentry may be queried as to whether the
1068 * caller is permitted to proceed or not.
1070 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1073 int follow_down(struct path *path)
1078 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1079 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1080 /* Allow the filesystem to manage the transit without i_mutex
1083 * We indicate to the filesystem if someone is trying to mount
1084 * something here. This gives autofs the chance to deny anyone
1085 * other than its daemon the right to mount on its
1088 * The filesystem may sleep at this point.
1090 if (managed & DCACHE_MANAGE_TRANSIT) {
1091 BUG_ON(!path->dentry->d_op);
1092 BUG_ON(!path->dentry->d_op->d_manage);
1093 ret = path->dentry->d_op->d_manage(
1094 path->dentry, false);
1096 return ret == -EISDIR ? 0 : ret;
1099 /* Transit to a mounted filesystem. */
1100 if (managed & DCACHE_MOUNTED) {
1101 struct vfsmount *mounted = lookup_mnt(path);
1106 path->mnt = mounted;
1107 path->dentry = dget(mounted->mnt_root);
1111 /* Don't handle automount points here */
1118 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1120 static void follow_mount(struct path *path)
1122 while (d_mountpoint(path->dentry)) {
1123 struct vfsmount *mounted = lookup_mnt(path);
1128 path->mnt = mounted;
1129 path->dentry = dget(mounted->mnt_root);
1133 static void follow_dotdot(struct nameidata *nd)
1138 struct dentry *old = nd->path.dentry;
1140 if (nd->path.dentry == nd->root.dentry &&
1141 nd->path.mnt == nd->root.mnt) {
1144 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1145 /* rare case of legitimate dget_parent()... */
1146 nd->path.dentry = dget_parent(nd->path.dentry);
1150 if (!follow_up(&nd->path))
1153 follow_mount(&nd->path);
1154 nd->inode = nd->path.dentry->d_inode;
1158 * Allocate a dentry with name and parent, and perform a parent
1159 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1160 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1161 * have verified that no child exists while under i_mutex.
1163 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1164 struct qstr *name, struct nameidata *nd)
1166 struct inode *inode = parent->d_inode;
1167 struct dentry *dentry;
1170 /* Don't create child dentry for a dead directory. */
1171 if (unlikely(IS_DEADDIR(inode)))
1172 return ERR_PTR(-ENOENT);
1174 dentry = d_alloc(parent, name);
1175 if (unlikely(!dentry))
1176 return ERR_PTR(-ENOMEM);
1178 old = inode->i_op->lookup(inode, dentry, nd);
1179 if (unlikely(old)) {
1187 * It's more convoluted than I'd like it to be, but... it's still fairly
1188 * small and for now I'd prefer to have fast path as straight as possible.
1189 * It _is_ time-critical.
1191 static int do_lookup(struct nameidata *nd, struct qstr *name,
1192 struct path *path, struct inode **inode)
1194 struct vfsmount *mnt = nd->path.mnt;
1195 struct dentry *dentry, *parent = nd->path.dentry;
1201 * Rename seqlock is not required here because in the off chance
1202 * of a false negative due to a concurrent rename, we're going to
1203 * do the non-racy lookup, below.
1205 if (nd->flags & LOOKUP_RCU) {
1208 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1212 /* Memory barrier in read_seqcount_begin of child is enough */
1213 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1217 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1218 status = d_revalidate(dentry, nd);
1219 if (unlikely(status <= 0)) {
1220 if (status != -ECHILD)
1226 path->dentry = dentry;
1227 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1231 if (nameidata_dentry_drop_rcu(nd, dentry))
1234 if (nameidata_drop_rcu(nd))
1238 dentry = __d_lookup(parent, name);
1242 if (unlikely(!dentry)) {
1243 struct inode *dir = parent->d_inode;
1244 BUG_ON(nd->inode != dir);
1246 mutex_lock(&dir->i_mutex);
1247 dentry = d_lookup(parent, name);
1248 if (likely(!dentry)) {
1249 dentry = d_alloc_and_lookup(parent, name, nd);
1250 if (IS_ERR(dentry)) {
1251 mutex_unlock(&dir->i_mutex);
1252 return PTR_ERR(dentry);
1258 mutex_unlock(&dir->i_mutex);
1260 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1261 status = d_revalidate(dentry, nd);
1262 if (unlikely(status <= 0)) {
1267 if (!d_invalidate(dentry)) {
1276 path->dentry = dentry;
1277 err = follow_managed(path, nd->flags);
1278 if (unlikely(err < 0)) {
1279 path_put_conditional(path, nd);
1282 *inode = path->dentry->d_inode;
1286 static inline int may_lookup(struct nameidata *nd)
1288 if (nd->flags & LOOKUP_RCU) {
1289 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1292 if (nameidata_drop_rcu(nd))
1295 return exec_permission(nd->inode, 0);
1298 static inline int handle_dots(struct nameidata *nd, int type)
1300 if (type == LAST_DOTDOT) {
1301 if (nd->flags & LOOKUP_RCU) {
1302 if (follow_dotdot_rcu(nd))
1310 static void terminate_walk(struct nameidata *nd)
1312 if (!(nd->flags & LOOKUP_RCU)) {
1313 path_put(&nd->path);
1315 nd->flags &= ~LOOKUP_RCU;
1316 if (!(nd->flags & LOOKUP_ROOT))
1317 nd->root.mnt = NULL;
1319 br_read_unlock(vfsmount_lock);
1323 static inline int walk_component(struct nameidata *nd, struct path *path,
1324 struct qstr *name, int type, int follow)
1326 struct inode *inode;
1329 * "." and ".." are special - ".." especially so because it has
1330 * to be able to know about the current root directory and
1331 * parent relationships.
1333 if (unlikely(type != LAST_NORM))
1334 return handle_dots(nd, type);
1335 err = do_lookup(nd, name, path, &inode);
1336 if (unlikely(err)) {
1341 path_to_nameidata(path, nd);
1345 if (unlikely(inode->i_op->follow_link) && follow) {
1346 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
1348 BUG_ON(inode != path->dentry->d_inode);
1351 path_to_nameidata(path, nd);
1357 * This limits recursive symlink follows to 8, while
1358 * limiting consecutive symlinks to 40.
1360 * Without that kind of total limit, nasty chains of consecutive
1361 * symlinks can cause almost arbitrarily long lookups.
1363 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1367 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1368 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1369 path_put_conditional(path, nd);
1370 path_put(&nd->path);
1375 current->link_count++;
1378 struct path link = *path;
1381 res = follow_link(&link, nd, &cookie);
1383 res = walk_component(nd, path, &nd->last,
1384 nd->last_type, LOOKUP_FOLLOW);
1385 put_link(nd, &link, cookie);
1388 current->link_count--;
1395 * This is the basic name resolution function, turning a pathname into
1396 * the final dentry. We expect 'base' to be positive and a directory.
1398 * Returns 0 and nd will have valid dentry and mnt on success.
1399 * Returns error and drops reference to input namei data on failure.
1401 static int link_path_walk(const char *name, struct nameidata *nd)
1405 unsigned int lookup_flags = nd->flags;
1412 /* At this point we know we have a real path component. */
1419 nd->flags |= LOOKUP_CONTINUE;
1421 err = may_lookup(nd);
1426 c = *(const unsigned char *)name;
1428 hash = init_name_hash();
1431 hash = partial_name_hash(c, hash);
1432 c = *(const unsigned char *)name;
1433 } while (c && (c != '/'));
1434 this.len = name - (const char *) this.name;
1435 this.hash = end_name_hash(hash);
1438 if (this.name[0] == '.') switch (this.len) {
1440 if (this.name[1] == '.') {
1442 nd->flags |= LOOKUP_JUMPED;
1448 if (likely(type == LAST_NORM)) {
1449 struct dentry *parent = nd->path.dentry;
1450 nd->flags &= ~LOOKUP_JUMPED;
1451 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1452 err = parent->d_op->d_hash(parent, nd->inode,
1459 /* remove trailing slashes? */
1461 goto last_component;
1462 while (*++name == '/');
1464 goto last_component;
1466 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1471 err = nested_symlink(&next, nd);
1476 if (!nd->inode->i_op->lookup)
1479 /* here ends the main loop */
1482 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1483 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1485 nd->last_type = type;
1492 static int path_init(int dfd, const char *name, unsigned int flags,
1493 struct nameidata *nd, struct file **fp)
1499 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1500 nd->flags = flags | LOOKUP_JUMPED;
1502 if (flags & LOOKUP_ROOT) {
1503 struct inode *inode = nd->root.dentry->d_inode;
1505 if (!inode->i_op->lookup)
1507 retval = inode_permission(inode, MAY_EXEC);
1511 nd->path = nd->root;
1513 if (flags & LOOKUP_RCU) {
1514 br_read_lock(vfsmount_lock);
1516 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1518 path_get(&nd->path);
1523 nd->root.mnt = NULL;
1526 if (flags & LOOKUP_RCU) {
1527 br_read_lock(vfsmount_lock);
1532 path_get(&nd->root);
1534 nd->path = nd->root;
1535 } else if (dfd == AT_FDCWD) {
1536 if (flags & LOOKUP_RCU) {
1537 struct fs_struct *fs = current->fs;
1540 br_read_lock(vfsmount_lock);
1544 seq = read_seqcount_begin(&fs->seq);
1546 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1547 } while (read_seqcount_retry(&fs->seq, seq));
1549 get_fs_pwd(current->fs, &nd->path);
1552 struct dentry *dentry;
1554 file = fget_raw_light(dfd, &fput_needed);
1559 dentry = file->f_path.dentry;
1563 if (!S_ISDIR(dentry->d_inode->i_mode))
1566 retval = file_permission(file, MAY_EXEC);
1571 nd->path = file->f_path;
1572 if (flags & LOOKUP_RCU) {
1575 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1576 br_read_lock(vfsmount_lock);
1579 path_get(&file->f_path);
1580 fput_light(file, fput_needed);
1584 nd->inode = nd->path.dentry->d_inode;
1588 fput_light(file, fput_needed);
1593 static inline int lookup_last(struct nameidata *nd, struct path *path)
1595 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1596 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1598 nd->flags &= ~LOOKUP_PARENT;
1599 return walk_component(nd, path, &nd->last, nd->last_type,
1600 nd->flags & LOOKUP_FOLLOW);
1603 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1604 static int path_lookupat(int dfd, const char *name,
1605 unsigned int flags, struct nameidata *nd)
1607 struct file *base = NULL;
1612 * Path walking is largely split up into 2 different synchronisation
1613 * schemes, rcu-walk and ref-walk (explained in
1614 * Documentation/filesystems/path-lookup.txt). These share much of the
1615 * path walk code, but some things particularly setup, cleanup, and
1616 * following mounts are sufficiently divergent that functions are
1617 * duplicated. Typically there is a function foo(), and its RCU
1618 * analogue, foo_rcu().
1620 * -ECHILD is the error number of choice (just to avoid clashes) that
1621 * is returned if some aspect of an rcu-walk fails. Such an error must
1622 * be handled by restarting a traditional ref-walk (which will always
1623 * be able to complete).
1625 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1630 current->total_link_count = 0;
1631 err = link_path_walk(name, nd);
1633 if (!err && !(flags & LOOKUP_PARENT)) {
1634 err = lookup_last(nd, &path);
1637 struct path link = path;
1638 nd->flags |= LOOKUP_PARENT;
1639 err = follow_link(&link, nd, &cookie);
1641 err = lookup_last(nd, &path);
1642 put_link(nd, &link, cookie);
1646 if (nd->flags & LOOKUP_RCU) {
1647 /* went all way through without dropping RCU */
1649 if (nameidata_drop_rcu_last(nd))
1654 err = handle_reval_path(nd);
1656 path_put(&nd->path);
1659 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1660 if (!nd->inode->i_op->lookup) {
1661 path_put(&nd->path);
1669 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1670 path_put(&nd->root);
1671 nd->root.mnt = NULL;
1676 static int do_path_lookup(int dfd, const char *name,
1677 unsigned int flags, struct nameidata *nd)
1679 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1680 if (unlikely(retval == -ECHILD))
1681 retval = path_lookupat(dfd, name, flags, nd);
1682 if (unlikely(retval == -ESTALE))
1683 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1685 if (likely(!retval)) {
1686 if (unlikely(!audit_dummy_context())) {
1687 if (nd->path.dentry && nd->inode)
1688 audit_inode(name, nd->path.dentry);
1694 int kern_path_parent(const char *name, struct nameidata *nd)
1696 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1699 int kern_path(const char *name, unsigned int flags, struct path *path)
1701 struct nameidata nd;
1702 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1709 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1710 * @dentry: pointer to dentry of the base directory
1711 * @mnt: pointer to vfs mount of the base directory
1712 * @name: pointer to file name
1713 * @flags: lookup flags
1714 * @nd: pointer to nameidata
1716 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1717 const char *name, unsigned int flags,
1718 struct nameidata *nd)
1720 nd->root.dentry = dentry;
1722 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1723 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
1726 static struct dentry *__lookup_hash(struct qstr *name,
1727 struct dentry *base, struct nameidata *nd)
1729 struct inode *inode = base->d_inode;
1730 struct dentry *dentry;
1733 err = exec_permission(inode, 0);
1735 return ERR_PTR(err);
1738 * Don't bother with __d_lookup: callers are for creat as
1739 * well as unlink, so a lot of the time it would cost
1742 dentry = d_lookup(base, name);
1744 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
1745 dentry = do_revalidate(dentry, nd);
1748 dentry = d_alloc_and_lookup(base, name, nd);
1754 * Restricted form of lookup. Doesn't follow links, single-component only,
1755 * needs parent already locked. Doesn't follow mounts.
1758 static struct dentry *lookup_hash(struct nameidata *nd)
1760 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1764 * lookup_one_len - filesystem helper to lookup single pathname component
1765 * @name: pathname component to lookup
1766 * @base: base directory to lookup from
1767 * @len: maximum length @len should be interpreted to
1769 * Note that this routine is purely a helper for filesystem usage and should
1770 * not be called by generic code. Also note that by using this function the
1771 * nameidata argument is passed to the filesystem methods and a filesystem
1772 * using this helper needs to be prepared for that.
1774 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1780 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1785 return ERR_PTR(-EACCES);
1787 hash = init_name_hash();
1789 c = *(const unsigned char *)name++;
1790 if (c == '/' || c == '\0')
1791 return ERR_PTR(-EACCES);
1792 hash = partial_name_hash(c, hash);
1794 this.hash = end_name_hash(hash);
1796 * See if the low-level filesystem might want
1797 * to use its own hash..
1799 if (base->d_flags & DCACHE_OP_HASH) {
1800 int err = base->d_op->d_hash(base, base->d_inode, &this);
1802 return ERR_PTR(err);
1805 return __lookup_hash(&this, base, NULL);
1808 int user_path_at(int dfd, const char __user *name, unsigned flags,
1811 struct nameidata nd;
1812 char *tmp = getname_flags(name, flags);
1813 int err = PTR_ERR(tmp);
1816 BUG_ON(flags & LOOKUP_PARENT);
1818 err = do_path_lookup(dfd, tmp, flags, &nd);
1826 static int user_path_parent(int dfd, const char __user *path,
1827 struct nameidata *nd, char **name)
1829 char *s = getname(path);
1835 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1845 * It's inline, so penalty for filesystems that don't use sticky bit is
1848 static inline int check_sticky(struct inode *dir, struct inode *inode)
1850 uid_t fsuid = current_fsuid();
1852 if (!(dir->i_mode & S_ISVTX))
1854 if (current_user_ns() != inode_userns(inode))
1856 if (inode->i_uid == fsuid)
1858 if (dir->i_uid == fsuid)
1862 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1866 * Check whether we can remove a link victim from directory dir, check
1867 * whether the type of victim is right.
1868 * 1. We can't do it if dir is read-only (done in permission())
1869 * 2. We should have write and exec permissions on dir
1870 * 3. We can't remove anything from append-only dir
1871 * 4. We can't do anything with immutable dir (done in permission())
1872 * 5. If the sticky bit on dir is set we should either
1873 * a. be owner of dir, or
1874 * b. be owner of victim, or
1875 * c. have CAP_FOWNER capability
1876 * 6. If the victim is append-only or immutable we can't do antyhing with
1877 * links pointing to it.
1878 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1879 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1880 * 9. We can't remove a root or mountpoint.
1881 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1882 * nfs_async_unlink().
1884 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1888 if (!victim->d_inode)
1891 BUG_ON(victim->d_parent->d_inode != dir);
1892 audit_inode_child(victim, dir);
1894 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1899 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1900 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1903 if (!S_ISDIR(victim->d_inode->i_mode))
1905 if (IS_ROOT(victim))
1907 } else if (S_ISDIR(victim->d_inode->i_mode))
1909 if (IS_DEADDIR(dir))
1911 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1916 /* Check whether we can create an object with dentry child in directory
1918 * 1. We can't do it if child already exists (open has special treatment for
1919 * this case, but since we are inlined it's OK)
1920 * 2. We can't do it if dir is read-only (done in permission())
1921 * 3. We should have write and exec permissions on dir
1922 * 4. We can't do it if dir is immutable (done in permission())
1924 static inline int may_create(struct inode *dir, struct dentry *child)
1928 if (IS_DEADDIR(dir))
1930 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1934 * p1 and p2 should be directories on the same fs.
1936 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1941 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1945 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1947 p = d_ancestor(p2, p1);
1949 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1950 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1954 p = d_ancestor(p1, p2);
1956 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1957 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1961 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1962 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1966 void unlock_rename(struct dentry *p1, struct dentry *p2)
1968 mutex_unlock(&p1->d_inode->i_mutex);
1970 mutex_unlock(&p2->d_inode->i_mutex);
1971 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1975 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1976 struct nameidata *nd)
1978 int error = may_create(dir, dentry);
1983 if (!dir->i_op->create)
1984 return -EACCES; /* shouldn't it be ENOSYS? */
1987 error = security_inode_create(dir, dentry, mode);
1990 error = dir->i_op->create(dir, dentry, mode, nd);
1992 fsnotify_create(dir, dentry);
1996 static int may_open(struct path *path, int acc_mode, int flag)
1998 struct dentry *dentry = path->dentry;
1999 struct inode *inode = dentry->d_inode;
2009 switch (inode->i_mode & S_IFMT) {
2013 if (acc_mode & MAY_WRITE)
2018 if (path->mnt->mnt_flags & MNT_NODEV)
2027 error = inode_permission(inode, acc_mode);
2032 * An append-only file must be opened in append mode for writing.
2034 if (IS_APPEND(inode)) {
2035 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2041 /* O_NOATIME can only be set by the owner or superuser */
2042 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2046 * Ensure there are no outstanding leases on the file.
2048 return break_lease(inode, flag);
2051 static int handle_truncate(struct file *filp)
2053 struct path *path = &filp->f_path;
2054 struct inode *inode = path->dentry->d_inode;
2055 int error = get_write_access(inode);
2059 * Refuse to truncate files with mandatory locks held on them.
2061 error = locks_verify_locked(inode);
2063 error = security_path_truncate(path);
2065 error = do_truncate(path->dentry, 0,
2066 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2069 put_write_access(inode);
2074 * Note that while the flag value (low two bits) for sys_open means:
2079 * it is changed into
2080 * 00 - no permissions needed
2081 * 01 - read-permission
2082 * 10 - write-permission
2084 * for the internal routines (ie open_namei()/follow_link() etc)
2085 * This is more logical, and also allows the 00 "no perm needed"
2086 * to be used for symlinks (where the permissions are checked
2090 static inline int open_to_namei_flags(int flag)
2092 if ((flag+1) & O_ACCMODE)
2098 * Handle the last step of open()
2100 static struct file *do_last(struct nameidata *nd, struct path *path,
2101 const struct open_flags *op, const char *pathname)
2103 struct dentry *dir = nd->path.dentry;
2104 struct dentry *dentry;
2105 int open_flag = op->open_flag;
2106 int will_truncate = open_flag & O_TRUNC;
2108 int acc_mode = op->acc_mode;
2112 nd->flags &= ~LOOKUP_PARENT;
2113 nd->flags |= op->intent;
2115 switch (nd->last_type) {
2118 error = handle_dots(nd, nd->last_type);
2120 return ERR_PTR(error);
2123 if (nd->flags & LOOKUP_RCU) {
2124 if (nameidata_drop_rcu_last(nd))
2125 return ERR_PTR(-ECHILD);
2127 error = handle_reval_path(nd);
2130 audit_inode(pathname, nd->path.dentry);
2131 if (open_flag & O_CREAT) {
2137 /* can't be RCU mode here */
2138 error = handle_reval_path(nd);
2141 audit_inode(pathname, dir);
2145 if (!(open_flag & O_CREAT)) {
2147 if (nd->last.name[nd->last.len])
2148 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2149 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2151 /* we _can_ be in RCU mode here */
2152 error = walk_component(nd, path, &nd->last, LAST_NORM,
2155 return ERR_PTR(error);
2156 if (error) /* symlink */
2159 if (nd->flags & LOOKUP_RCU) {
2160 if (nameidata_drop_rcu_last(nd))
2161 return ERR_PTR(-ECHILD);
2165 if (nd->flags & LOOKUP_DIRECTORY) {
2166 if (!nd->inode->i_op->lookup)
2169 audit_inode(pathname, nd->path.dentry);
2173 /* create side of things */
2175 if (nd->flags & LOOKUP_RCU) {
2176 if (nameidata_drop_rcu_last(nd))
2177 return ERR_PTR(-ECHILD);
2180 audit_inode(pathname, dir);
2182 /* trailing slashes? */
2183 if (nd->last.name[nd->last.len])
2186 mutex_lock(&dir->d_inode->i_mutex);
2188 dentry = lookup_hash(nd);
2189 error = PTR_ERR(dentry);
2190 if (IS_ERR(dentry)) {
2191 mutex_unlock(&dir->d_inode->i_mutex);
2195 path->dentry = dentry;
2196 path->mnt = nd->path.mnt;
2198 /* Negative dentry, just create the file */
2199 if (!dentry->d_inode) {
2200 int mode = op->mode;
2201 if (!IS_POSIXACL(dir->d_inode))
2202 mode &= ~current_umask();
2204 * This write is needed to ensure that a
2205 * rw->ro transition does not occur between
2206 * the time when the file is created and when
2207 * a permanent write count is taken through
2208 * the 'struct file' in nameidata_to_filp().
2210 error = mnt_want_write(nd->path.mnt);
2212 goto exit_mutex_unlock;
2214 /* Don't check for write permission, don't truncate */
2215 open_flag &= ~O_TRUNC;
2217 acc_mode = MAY_OPEN;
2218 error = security_path_mknod(&nd->path, dentry, mode, 0);
2220 goto exit_mutex_unlock;
2221 error = vfs_create(dir->d_inode, dentry, mode, nd);
2223 goto exit_mutex_unlock;
2224 mutex_unlock(&dir->d_inode->i_mutex);
2225 dput(nd->path.dentry);
2226 nd->path.dentry = dentry;
2231 * It already exists.
2233 mutex_unlock(&dir->d_inode->i_mutex);
2234 audit_inode(pathname, path->dentry);
2237 if (open_flag & O_EXCL)
2240 error = follow_managed(path, nd->flags);
2245 if (!path->dentry->d_inode)
2248 if (path->dentry->d_inode->i_op->follow_link)
2251 path_to_nameidata(path, nd);
2252 nd->inode = path->dentry->d_inode;
2254 if (S_ISDIR(nd->inode->i_mode))
2257 if (!S_ISREG(nd->inode->i_mode))
2260 if (will_truncate) {
2261 error = mnt_want_write(nd->path.mnt);
2267 error = may_open(&nd->path, acc_mode, open_flag);
2270 filp = nameidata_to_filp(nd);
2271 if (!IS_ERR(filp)) {
2272 error = ima_file_check(filp, op->acc_mode);
2275 filp = ERR_PTR(error);
2278 if (!IS_ERR(filp)) {
2279 if (will_truncate) {
2280 error = handle_truncate(filp);
2283 filp = ERR_PTR(error);
2289 mnt_drop_write(nd->path.mnt);
2290 path_put(&nd->path);
2294 mutex_unlock(&dir->d_inode->i_mutex);
2296 path_put_conditional(path, nd);
2298 filp = ERR_PTR(error);
2302 static struct file *path_openat(int dfd, const char *pathname,
2303 struct nameidata *nd, const struct open_flags *op, int flags)
2305 struct file *base = NULL;
2310 filp = get_empty_filp();
2312 return ERR_PTR(-ENFILE);
2314 filp->f_flags = op->open_flag;
2315 nd->intent.open.file = filp;
2316 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2317 nd->intent.open.create_mode = op->mode;
2319 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2320 if (unlikely(error))
2323 current->total_link_count = 0;
2324 error = link_path_walk(pathname, nd);
2325 if (unlikely(error))
2328 filp = do_last(nd, &path, op, pathname);
2329 while (unlikely(!filp)) { /* trailing symlink */
2330 struct path link = path;
2332 if (!(nd->flags & LOOKUP_FOLLOW)) {
2333 path_put_conditional(&path, nd);
2334 path_put(&nd->path);
2335 filp = ERR_PTR(-ELOOP);
2338 nd->flags |= LOOKUP_PARENT;
2339 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2340 error = follow_link(&link, nd, &cookie);
2341 if (unlikely(error))
2342 filp = ERR_PTR(error);
2344 filp = do_last(nd, &path, op, pathname);
2345 put_link(nd, &link, cookie);
2348 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2349 path_put(&nd->root);
2352 release_open_intent(nd);
2356 filp = ERR_PTR(error);
2360 struct file *do_filp_open(int dfd, const char *pathname,
2361 const struct open_flags *op, int flags)
2363 struct nameidata nd;
2366 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2367 if (unlikely(filp == ERR_PTR(-ECHILD)))
2368 filp = path_openat(dfd, pathname, &nd, op, flags);
2369 if (unlikely(filp == ERR_PTR(-ESTALE)))
2370 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2374 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2375 const char *name, const struct open_flags *op, int flags)
2377 struct nameidata nd;
2381 nd.root.dentry = dentry;
2383 flags |= LOOKUP_ROOT;
2385 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2386 return ERR_PTR(-ELOOP);
2388 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2389 if (unlikely(file == ERR_PTR(-ECHILD)))
2390 file = path_openat(-1, name, &nd, op, flags);
2391 if (unlikely(file == ERR_PTR(-ESTALE)))
2392 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2397 * lookup_create - lookup a dentry, creating it if it doesn't exist
2398 * @nd: nameidata info
2399 * @is_dir: directory flag
2401 * Simple function to lookup and return a dentry and create it
2402 * if it doesn't exist. Is SMP-safe.
2404 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2406 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2408 struct dentry *dentry = ERR_PTR(-EEXIST);
2410 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2412 * Yucky last component or no last component at all?
2413 * (foo/., foo/.., /////)
2415 if (nd->last_type != LAST_NORM)
2417 nd->flags &= ~LOOKUP_PARENT;
2418 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2419 nd->intent.open.flags = O_EXCL;
2422 * Do the final lookup.
2424 dentry = lookup_hash(nd);
2428 if (dentry->d_inode)
2431 * Special case - lookup gave negative, but... we had foo/bar/
2432 * From the vfs_mknod() POV we just have a negative dentry -
2433 * all is fine. Let's be bastards - you had / on the end, you've
2434 * been asking for (non-existent) directory. -ENOENT for you.
2436 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2438 dentry = ERR_PTR(-ENOENT);
2443 dentry = ERR_PTR(-EEXIST);
2447 EXPORT_SYMBOL_GPL(lookup_create);
2449 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2451 int error = may_create(dir, dentry);
2456 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2457 !ns_capable(inode_userns(dir), CAP_MKNOD))
2460 if (!dir->i_op->mknod)
2463 error = devcgroup_inode_mknod(mode, dev);
2467 error = security_inode_mknod(dir, dentry, mode, dev);
2471 error = dir->i_op->mknod(dir, dentry, mode, dev);
2473 fsnotify_create(dir, dentry);
2477 static int may_mknod(mode_t mode)
2479 switch (mode & S_IFMT) {
2485 case 0: /* zero mode translates to S_IFREG */
2494 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2499 struct dentry *dentry;
2500 struct nameidata nd;
2505 error = user_path_parent(dfd, filename, &nd, &tmp);
2509 dentry = lookup_create(&nd, 0);
2510 if (IS_ERR(dentry)) {
2511 error = PTR_ERR(dentry);
2514 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2515 mode &= ~current_umask();
2516 error = may_mknod(mode);
2519 error = mnt_want_write(nd.path.mnt);
2522 error = security_path_mknod(&nd.path, dentry, mode, dev);
2524 goto out_drop_write;
2525 switch (mode & S_IFMT) {
2526 case 0: case S_IFREG:
2527 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2529 case S_IFCHR: case S_IFBLK:
2530 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2531 new_decode_dev(dev));
2533 case S_IFIFO: case S_IFSOCK:
2534 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2538 mnt_drop_write(nd.path.mnt);
2542 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2549 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2551 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2554 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2556 int error = may_create(dir, dentry);
2561 if (!dir->i_op->mkdir)
2564 mode &= (S_IRWXUGO|S_ISVTX);
2565 error = security_inode_mkdir(dir, dentry, mode);
2569 error = dir->i_op->mkdir(dir, dentry, mode);
2571 fsnotify_mkdir(dir, dentry);
2575 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2579 struct dentry *dentry;
2580 struct nameidata nd;
2582 error = user_path_parent(dfd, pathname, &nd, &tmp);
2586 dentry = lookup_create(&nd, 1);
2587 error = PTR_ERR(dentry);
2591 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2592 mode &= ~current_umask();
2593 error = mnt_want_write(nd.path.mnt);
2596 error = security_path_mkdir(&nd.path, dentry, mode);
2598 goto out_drop_write;
2599 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2601 mnt_drop_write(nd.path.mnt);
2605 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2612 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2614 return sys_mkdirat(AT_FDCWD, pathname, mode);
2618 * We try to drop the dentry early: we should have
2619 * a usage count of 2 if we're the only user of this
2620 * dentry, and if that is true (possibly after pruning
2621 * the dcache), then we drop the dentry now.
2623 * A low-level filesystem can, if it choses, legally
2626 * if (!d_unhashed(dentry))
2629 * if it cannot handle the case of removing a directory
2630 * that is still in use by something else..
2632 void dentry_unhash(struct dentry *dentry)
2635 shrink_dcache_parent(dentry);
2636 spin_lock(&dentry->d_lock);
2637 if (dentry->d_count == 2)
2639 spin_unlock(&dentry->d_lock);
2642 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2644 int error = may_delete(dir, dentry, 1);
2649 if (!dir->i_op->rmdir)
2652 mutex_lock(&dentry->d_inode->i_mutex);
2653 dentry_unhash(dentry);
2654 if (d_mountpoint(dentry))
2657 error = security_inode_rmdir(dir, dentry);
2659 error = dir->i_op->rmdir(dir, dentry);
2661 dentry->d_inode->i_flags |= S_DEAD;
2666 mutex_unlock(&dentry->d_inode->i_mutex);
2675 static long do_rmdir(int dfd, const char __user *pathname)
2679 struct dentry *dentry;
2680 struct nameidata nd;
2682 error = user_path_parent(dfd, pathname, &nd, &name);
2686 switch(nd.last_type) {
2698 nd.flags &= ~LOOKUP_PARENT;
2700 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2701 dentry = lookup_hash(&nd);
2702 error = PTR_ERR(dentry);
2705 error = mnt_want_write(nd.path.mnt);
2708 error = security_path_rmdir(&nd.path, dentry);
2711 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2713 mnt_drop_write(nd.path.mnt);
2717 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2724 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2726 return do_rmdir(AT_FDCWD, pathname);
2729 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2731 int error = may_delete(dir, dentry, 0);
2736 if (!dir->i_op->unlink)
2739 mutex_lock(&dentry->d_inode->i_mutex);
2740 if (d_mountpoint(dentry))
2743 error = security_inode_unlink(dir, dentry);
2745 error = dir->i_op->unlink(dir, dentry);
2750 mutex_unlock(&dentry->d_inode->i_mutex);
2752 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2753 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2754 fsnotify_link_count(dentry->d_inode);
2762 * Make sure that the actual truncation of the file will occur outside its
2763 * directory's i_mutex. Truncate can take a long time if there is a lot of
2764 * writeout happening, and we don't want to prevent access to the directory
2765 * while waiting on the I/O.
2767 static long do_unlinkat(int dfd, const char __user *pathname)
2771 struct dentry *dentry;
2772 struct nameidata nd;
2773 struct inode *inode = NULL;
2775 error = user_path_parent(dfd, pathname, &nd, &name);
2780 if (nd.last_type != LAST_NORM)
2783 nd.flags &= ~LOOKUP_PARENT;
2785 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2786 dentry = lookup_hash(&nd);
2787 error = PTR_ERR(dentry);
2788 if (!IS_ERR(dentry)) {
2789 /* Why not before? Because we want correct error value */
2790 if (nd.last.name[nd.last.len])
2792 inode = dentry->d_inode;
2795 error = mnt_want_write(nd.path.mnt);
2798 error = security_path_unlink(&nd.path, dentry);
2801 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2803 mnt_drop_write(nd.path.mnt);
2807 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2809 iput(inode); /* truncate the inode here */
2816 error = !dentry->d_inode ? -ENOENT :
2817 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2821 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2823 if ((flag & ~AT_REMOVEDIR) != 0)
2826 if (flag & AT_REMOVEDIR)
2827 return do_rmdir(dfd, pathname);
2829 return do_unlinkat(dfd, pathname);
2832 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2834 return do_unlinkat(AT_FDCWD, pathname);
2837 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2839 int error = may_create(dir, dentry);
2844 if (!dir->i_op->symlink)
2847 error = security_inode_symlink(dir, dentry, oldname);
2851 error = dir->i_op->symlink(dir, dentry, oldname);
2853 fsnotify_create(dir, dentry);
2857 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2858 int, newdfd, const char __user *, newname)
2863 struct dentry *dentry;
2864 struct nameidata nd;
2866 from = getname(oldname);
2868 return PTR_ERR(from);
2870 error = user_path_parent(newdfd, newname, &nd, &to);
2874 dentry = lookup_create(&nd, 0);
2875 error = PTR_ERR(dentry);
2879 error = mnt_want_write(nd.path.mnt);
2882 error = security_path_symlink(&nd.path, dentry, from);
2884 goto out_drop_write;
2885 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2887 mnt_drop_write(nd.path.mnt);
2891 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2899 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2901 return sys_symlinkat(oldname, AT_FDCWD, newname);
2904 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2906 struct inode *inode = old_dentry->d_inode;
2912 error = may_create(dir, new_dentry);
2916 if (dir->i_sb != inode->i_sb)
2920 * A link to an append-only or immutable file cannot be created.
2922 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2924 if (!dir->i_op->link)
2926 if (S_ISDIR(inode->i_mode))
2929 error = security_inode_link(old_dentry, dir, new_dentry);
2933 mutex_lock(&inode->i_mutex);
2934 /* Make sure we don't allow creating hardlink to an unlinked file */
2935 if (inode->i_nlink == 0)
2938 error = dir->i_op->link(old_dentry, dir, new_dentry);
2939 mutex_unlock(&inode->i_mutex);
2941 fsnotify_link(dir, inode, new_dentry);
2946 * Hardlinks are often used in delicate situations. We avoid
2947 * security-related surprises by not following symlinks on the
2950 * We don't follow them on the oldname either to be compatible
2951 * with linux 2.0, and to avoid hard-linking to directories
2952 * and other special files. --ADM
2954 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2955 int, newdfd, const char __user *, newname, int, flags)
2957 struct dentry *new_dentry;
2958 struct nameidata nd;
2959 struct path old_path;
2964 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
2967 * To use null names we require CAP_DAC_READ_SEARCH
2968 * This ensures that not everyone will be able to create
2969 * handlink using the passed filedescriptor.
2971 if (flags & AT_EMPTY_PATH) {
2972 if (!capable(CAP_DAC_READ_SEARCH))
2977 if (flags & AT_SYMLINK_FOLLOW)
2978 how |= LOOKUP_FOLLOW;
2980 error = user_path_at(olddfd, oldname, how, &old_path);
2984 error = user_path_parent(newdfd, newname, &nd, &to);
2988 if (old_path.mnt != nd.path.mnt)
2990 new_dentry = lookup_create(&nd, 0);
2991 error = PTR_ERR(new_dentry);
2992 if (IS_ERR(new_dentry))
2994 error = mnt_want_write(nd.path.mnt);
2997 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2999 goto out_drop_write;
3000 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
3002 mnt_drop_write(nd.path.mnt);
3006 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3011 path_put(&old_path);
3016 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3018 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3022 * The worst of all namespace operations - renaming directory. "Perverted"
3023 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3025 * a) we can get into loop creation. Check is done in is_subdir().
3026 * b) race potential - two innocent renames can create a loop together.
3027 * That's where 4.4 screws up. Current fix: serialization on
3028 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3030 * c) we have to lock _three_ objects - parents and victim (if it exists).
3031 * And that - after we got ->i_mutex on parents (until then we don't know
3032 * whether the target exists). Solution: try to be smart with locking
3033 * order for inodes. We rely on the fact that tree topology may change
3034 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3035 * move will be locked. Thus we can rank directories by the tree
3036 * (ancestors first) and rank all non-directories after them.
3037 * That works since everybody except rename does "lock parent, lookup,
3038 * lock child" and rename is under ->s_vfs_rename_mutex.
3039 * HOWEVER, it relies on the assumption that any object with ->lookup()
3040 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3041 * we'd better make sure that there's no link(2) for them.
3042 * d) some filesystems don't support opened-but-unlinked directories,
3043 * either because of layout or because they are not ready to deal with
3044 * all cases correctly. The latter will be fixed (taking this sort of
3045 * stuff into VFS), but the former is not going away. Solution: the same
3046 * trick as in rmdir().
3047 * e) conversion from fhandle to dentry may come in the wrong moment - when
3048 * we are removing the target. Solution: we will have to grab ->i_mutex
3049 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3050 * ->i_mutex on parents, which works but leads to some truly excessive
3053 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3054 struct inode *new_dir, struct dentry *new_dentry)
3057 struct inode *target;
3060 * If we are going to change the parent - check write permissions,
3061 * we'll need to flip '..'.
3063 if (new_dir != old_dir) {
3064 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3069 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3073 target = new_dentry->d_inode;
3075 mutex_lock(&target->i_mutex);
3076 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3080 dentry_unhash(new_dentry);
3081 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3085 target->i_flags |= S_DEAD;
3086 dont_mount(new_dentry);
3088 mutex_unlock(&target->i_mutex);
3089 if (d_unhashed(new_dentry))
3090 d_rehash(new_dentry);
3094 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3095 d_move(old_dentry,new_dentry);
3099 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3100 struct inode *new_dir, struct dentry *new_dentry)
3102 struct inode *target;
3105 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3110 target = new_dentry->d_inode;
3112 mutex_lock(&target->i_mutex);
3113 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3116 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3119 dont_mount(new_dentry);
3120 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3121 d_move(old_dentry, new_dentry);
3124 mutex_unlock(&target->i_mutex);
3129 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3130 struct inode *new_dir, struct dentry *new_dentry)
3133 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3134 const unsigned char *old_name;
3136 if (old_dentry->d_inode == new_dentry->d_inode)
3139 error = may_delete(old_dir, old_dentry, is_dir);
3143 if (!new_dentry->d_inode)
3144 error = may_create(new_dir, new_dentry);
3146 error = may_delete(new_dir, new_dentry, is_dir);
3150 if (!old_dir->i_op->rename)
3153 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3156 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3158 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3160 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3161 new_dentry->d_inode, old_dentry);
3162 fsnotify_oldname_free(old_name);
3167 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3168 int, newdfd, const char __user *, newname)
3170 struct dentry *old_dir, *new_dir;
3171 struct dentry *old_dentry, *new_dentry;
3172 struct dentry *trap;
3173 struct nameidata oldnd, newnd;
3178 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3182 error = user_path_parent(newdfd, newname, &newnd, &to);
3187 if (oldnd.path.mnt != newnd.path.mnt)
3190 old_dir = oldnd.path.dentry;
3192 if (oldnd.last_type != LAST_NORM)
3195 new_dir = newnd.path.dentry;
3196 if (newnd.last_type != LAST_NORM)
3199 oldnd.flags &= ~LOOKUP_PARENT;
3200 newnd.flags &= ~LOOKUP_PARENT;
3201 newnd.flags |= LOOKUP_RENAME_TARGET;
3203 trap = lock_rename(new_dir, old_dir);
3205 old_dentry = lookup_hash(&oldnd);
3206 error = PTR_ERR(old_dentry);
3207 if (IS_ERR(old_dentry))
3209 /* source must exist */
3211 if (!old_dentry->d_inode)
3213 /* unless the source is a directory trailing slashes give -ENOTDIR */
3214 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3216 if (oldnd.last.name[oldnd.last.len])
3218 if (newnd.last.name[newnd.last.len])
3221 /* source should not be ancestor of target */
3223 if (old_dentry == trap)
3225 new_dentry = lookup_hash(&newnd);
3226 error = PTR_ERR(new_dentry);
3227 if (IS_ERR(new_dentry))
3229 /* target should not be an ancestor of source */
3231 if (new_dentry == trap)
3234 error = mnt_want_write(oldnd.path.mnt);
3237 error = security_path_rename(&oldnd.path, old_dentry,
3238 &newnd.path, new_dentry);
3241 error = vfs_rename(old_dir->d_inode, old_dentry,
3242 new_dir->d_inode, new_dentry);
3244 mnt_drop_write(oldnd.path.mnt);
3250 unlock_rename(new_dir, old_dir);
3252 path_put(&newnd.path);
3255 path_put(&oldnd.path);
3261 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3263 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3266 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3270 len = PTR_ERR(link);
3275 if (len > (unsigned) buflen)
3277 if (copy_to_user(buffer, link, len))
3284 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3285 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3286 * using) it for any given inode is up to filesystem.
3288 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3290 struct nameidata nd;
3295 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3297 return PTR_ERR(cookie);
3299 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3300 if (dentry->d_inode->i_op->put_link)
3301 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3305 int vfs_follow_link(struct nameidata *nd, const char *link)
3307 return __vfs_follow_link(nd, link);
3310 /* get the link contents into pagecache */
3311 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3315 struct address_space *mapping = dentry->d_inode->i_mapping;
3316 page = read_mapping_page(mapping, 0, NULL);
3321 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3325 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3327 struct page *page = NULL;
3328 char *s = page_getlink(dentry, &page);
3329 int res = vfs_readlink(dentry,buffer,buflen,s);
3332 page_cache_release(page);
3337 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3339 struct page *page = NULL;
3340 nd_set_link(nd, page_getlink(dentry, &page));
3344 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3346 struct page *page = cookie;
3350 page_cache_release(page);
3355 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3357 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3359 struct address_space *mapping = inode->i_mapping;
3364 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3366 flags |= AOP_FLAG_NOFS;
3369 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3370 flags, &page, &fsdata);
3374 kaddr = kmap_atomic(page, KM_USER0);
3375 memcpy(kaddr, symname, len-1);
3376 kunmap_atomic(kaddr, KM_USER0);
3378 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3385 mark_inode_dirty(inode);
3391 int page_symlink(struct inode *inode, const char *symname, int len)
3393 return __page_symlink(inode, symname, len,
3394 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3397 const struct inode_operations page_symlink_inode_operations = {
3398 .readlink = generic_readlink,
3399 .follow_link = page_follow_link_light,
3400 .put_link = page_put_link,
3403 EXPORT_SYMBOL(user_path_at);
3404 EXPORT_SYMBOL(follow_down_one);
3405 EXPORT_SYMBOL(follow_down);
3406 EXPORT_SYMBOL(follow_up);
3407 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3408 EXPORT_SYMBOL(getname);
3409 EXPORT_SYMBOL(lock_rename);
3410 EXPORT_SYMBOL(lookup_one_len);
3411 EXPORT_SYMBOL(page_follow_link_light);
3412 EXPORT_SYMBOL(page_put_link);
3413 EXPORT_SYMBOL(page_readlink);
3414 EXPORT_SYMBOL(__page_symlink);
3415 EXPORT_SYMBOL(page_symlink);
3416 EXPORT_SYMBOL(page_symlink_inode_operations);
3417 EXPORT_SYMBOL(kern_path_parent);
3418 EXPORT_SYMBOL(kern_path);
3419 EXPORT_SYMBOL(vfs_path_lookup);
3420 EXPORT_SYMBOL(inode_permission);
3421 EXPORT_SYMBOL(file_permission);
3422 EXPORT_SYMBOL(unlock_rename);
3423 EXPORT_SYMBOL(vfs_create);
3424 EXPORT_SYMBOL(vfs_follow_link);
3425 EXPORT_SYMBOL(vfs_link);
3426 EXPORT_SYMBOL(vfs_mkdir);
3427 EXPORT_SYMBOL(vfs_mknod);
3428 EXPORT_SYMBOL(generic_permission);
3429 EXPORT_SYMBOL(vfs_readlink);
3430 EXPORT_SYMBOL(vfs_rename);
3431 EXPORT_SYMBOL(vfs_rmdir);
3432 EXPORT_SYMBOL(vfs_symlink);
3433 EXPORT_SYMBOL(vfs_unlink);
3434 EXPORT_SYMBOL(dentry_unhash);
3435 EXPORT_SYMBOL(generic_readlink);