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 (unlikely(current->total_link_count >= 40)) {
757 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
758 path_put_conditional(link, nd);
763 current->total_link_count++;
765 touch_atime(link->mnt, dentry);
766 nd_set_link(nd, NULL);
768 if (link->mnt == nd->path.mnt)
771 error = security_inode_follow_link(link->dentry, nd);
773 *p = ERR_PTR(error); /* no ->put_link(), please */
778 nd->last_type = LAST_BIND;
779 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
782 char *s = nd_get_link(nd);
785 error = __vfs_follow_link(nd, s);
786 else if (nd->last_type == LAST_BIND) {
787 nd->flags |= LOOKUP_JUMPED;
788 nd->inode = nd->path.dentry->d_inode;
789 if (nd->inode->i_op->follow_link) {
790 /* stepped on a _really_ weird one */
799 static int follow_up_rcu(struct path *path)
801 struct vfsmount *parent;
802 struct dentry *mountpoint;
804 parent = path->mnt->mnt_parent;
805 if (parent == path->mnt)
807 mountpoint = path->mnt->mnt_mountpoint;
808 path->dentry = mountpoint;
813 int follow_up(struct path *path)
815 struct vfsmount *parent;
816 struct dentry *mountpoint;
818 br_read_lock(vfsmount_lock);
819 parent = path->mnt->mnt_parent;
820 if (parent == path->mnt) {
821 br_read_unlock(vfsmount_lock);
825 mountpoint = dget(path->mnt->mnt_mountpoint);
826 br_read_unlock(vfsmount_lock);
828 path->dentry = mountpoint;
835 * Perform an automount
836 * - return -EISDIR to tell follow_managed() to stop and return the path we
839 static int follow_automount(struct path *path, unsigned flags,
842 struct vfsmount *mnt;
845 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
848 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
849 * and this is the terminal part of the path.
851 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
852 return -EISDIR; /* we actually want to stop here */
854 /* We want to mount if someone is trying to open/create a file of any
855 * type under the mountpoint, wants to traverse through the mountpoint
856 * or wants to open the mounted directory.
858 * We don't want to mount if someone's just doing a stat and they've
859 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
860 * appended a '/' to the name.
862 if (!(flags & LOOKUP_FOLLOW) &&
863 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
864 LOOKUP_OPEN | LOOKUP_CREATE)))
867 current->total_link_count++;
868 if (current->total_link_count >= 40)
871 mnt = path->dentry->d_op->d_automount(path);
874 * The filesystem is allowed to return -EISDIR here to indicate
875 * it doesn't want to automount. For instance, autofs would do
876 * this so that its userspace daemon can mount on this dentry.
878 * However, we can only permit this if it's a terminal point in
879 * the path being looked up; if it wasn't then the remainder of
880 * the path is inaccessible and we should say so.
882 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
887 if (!mnt) /* mount collision */
890 err = finish_automount(mnt, path);
894 /* Someone else made a mount here whilst we were busy */
901 path->dentry = dget(mnt->mnt_root);
911 * Handle a dentry that is managed in some way.
912 * - Flagged for transit management (autofs)
913 * - Flagged as mountpoint
914 * - Flagged as automount point
916 * This may only be called in refwalk mode.
918 * Serialization is taken care of in namespace.c
920 static int follow_managed(struct path *path, unsigned flags)
923 bool need_mntput = false;
926 /* Given that we're not holding a lock here, we retain the value in a
927 * local variable for each dentry as we look at it so that we don't see
928 * the components of that value change under us */
929 while (managed = ACCESS_ONCE(path->dentry->d_flags),
930 managed &= DCACHE_MANAGED_DENTRY,
931 unlikely(managed != 0)) {
932 /* Allow the filesystem to manage the transit without i_mutex
934 if (managed & DCACHE_MANAGE_TRANSIT) {
935 BUG_ON(!path->dentry->d_op);
936 BUG_ON(!path->dentry->d_op->d_manage);
937 ret = path->dentry->d_op->d_manage(path->dentry,
940 return ret == -EISDIR ? 0 : ret;
943 /* Transit to a mounted filesystem. */
944 if (managed & DCACHE_MOUNTED) {
945 struct vfsmount *mounted = lookup_mnt(path);
951 path->dentry = dget(mounted->mnt_root);
956 /* Something is mounted on this dentry in another
957 * namespace and/or whatever was mounted there in this
958 * namespace got unmounted before we managed to get the
962 /* Handle an automount point */
963 if (managed & DCACHE_NEED_AUTOMOUNT) {
964 ret = follow_automount(path, flags, &need_mntput);
966 return ret == -EISDIR ? 0 : ret;
970 /* We didn't change the current path point */
976 int follow_down_one(struct path *path)
978 struct vfsmount *mounted;
980 mounted = lookup_mnt(path);
985 path->dentry = dget(mounted->mnt_root);
992 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
993 * meet a managed dentry and we're not walking to "..". True is returned to
994 * continue, false to abort.
996 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
997 struct inode **inode, bool reverse_transit)
999 while (d_mountpoint(path->dentry)) {
1000 struct vfsmount *mounted;
1001 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1003 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1005 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1008 path->mnt = mounted;
1009 path->dentry = mounted->mnt_root;
1010 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1011 *inode = path->dentry->d_inode;
1014 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1015 return reverse_transit;
1019 static int follow_dotdot_rcu(struct nameidata *nd)
1021 struct inode *inode = nd->inode;
1026 if (nd->path.dentry == nd->root.dentry &&
1027 nd->path.mnt == nd->root.mnt) {
1030 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1031 struct dentry *old = nd->path.dentry;
1032 struct dentry *parent = old->d_parent;
1035 seq = read_seqcount_begin(&parent->d_seq);
1036 if (read_seqcount_retry(&old->d_seq, nd->seq))
1038 inode = parent->d_inode;
1039 nd->path.dentry = parent;
1043 if (!follow_up_rcu(&nd->path))
1045 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1046 inode = nd->path.dentry->d_inode;
1048 __follow_mount_rcu(nd, &nd->path, &inode, true);
1053 nd->flags &= ~LOOKUP_RCU;
1054 if (!(nd->flags & LOOKUP_ROOT))
1055 nd->root.mnt = NULL;
1057 br_read_unlock(vfsmount_lock);
1062 * Follow down to the covering mount currently visible to userspace. At each
1063 * point, the filesystem owning that dentry may be queried as to whether the
1064 * caller is permitted to proceed or not.
1066 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1069 int follow_down(struct path *path, bool mounting_here)
1074 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1075 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1076 /* Allow the filesystem to manage the transit without i_mutex
1079 * We indicate to the filesystem if someone is trying to mount
1080 * something here. This gives autofs the chance to deny anyone
1081 * other than its daemon the right to mount on its
1084 * The filesystem may sleep at this point.
1086 if (managed & DCACHE_MANAGE_TRANSIT) {
1087 BUG_ON(!path->dentry->d_op);
1088 BUG_ON(!path->dentry->d_op->d_manage);
1089 ret = path->dentry->d_op->d_manage(
1090 path->dentry, mounting_here, false);
1092 return ret == -EISDIR ? 0 : ret;
1095 /* Transit to a mounted filesystem. */
1096 if (managed & DCACHE_MOUNTED) {
1097 struct vfsmount *mounted = lookup_mnt(path);
1102 path->mnt = mounted;
1103 path->dentry = dget(mounted->mnt_root);
1107 /* Don't handle automount points here */
1114 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1116 static void follow_mount(struct path *path)
1118 while (d_mountpoint(path->dentry)) {
1119 struct vfsmount *mounted = lookup_mnt(path);
1124 path->mnt = mounted;
1125 path->dentry = dget(mounted->mnt_root);
1129 static void follow_dotdot(struct nameidata *nd)
1134 struct dentry *old = nd->path.dentry;
1136 if (nd->path.dentry == nd->root.dentry &&
1137 nd->path.mnt == nd->root.mnt) {
1140 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1141 /* rare case of legitimate dget_parent()... */
1142 nd->path.dentry = dget_parent(nd->path.dentry);
1146 if (!follow_up(&nd->path))
1149 follow_mount(&nd->path);
1150 nd->inode = nd->path.dentry->d_inode;
1154 * Allocate a dentry with name and parent, and perform a parent
1155 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1156 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1157 * have verified that no child exists while under i_mutex.
1159 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1160 struct qstr *name, struct nameidata *nd)
1162 struct inode *inode = parent->d_inode;
1163 struct dentry *dentry;
1166 /* Don't create child dentry for a dead directory. */
1167 if (unlikely(IS_DEADDIR(inode)))
1168 return ERR_PTR(-ENOENT);
1170 dentry = d_alloc(parent, name);
1171 if (unlikely(!dentry))
1172 return ERR_PTR(-ENOMEM);
1174 old = inode->i_op->lookup(inode, dentry, nd);
1175 if (unlikely(old)) {
1183 * It's more convoluted than I'd like it to be, but... it's still fairly
1184 * small and for now I'd prefer to have fast path as straight as possible.
1185 * It _is_ time-critical.
1187 static int do_lookup(struct nameidata *nd, struct qstr *name,
1188 struct path *path, struct inode **inode)
1190 struct vfsmount *mnt = nd->path.mnt;
1191 struct dentry *dentry, *parent = nd->path.dentry;
1197 * Rename seqlock is not required here because in the off chance
1198 * of a false negative due to a concurrent rename, we're going to
1199 * do the non-racy lookup, below.
1201 if (nd->flags & LOOKUP_RCU) {
1204 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1208 /* Memory barrier in read_seqcount_begin of child is enough */
1209 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1213 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1214 status = d_revalidate(dentry, nd);
1215 if (unlikely(status <= 0)) {
1216 if (status != -ECHILD)
1222 path->dentry = dentry;
1223 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1227 if (nameidata_dentry_drop_rcu(nd, dentry))
1230 if (nameidata_drop_rcu(nd))
1234 dentry = __d_lookup(parent, name);
1238 if (unlikely(!dentry)) {
1239 struct inode *dir = parent->d_inode;
1240 BUG_ON(nd->inode != dir);
1242 mutex_lock(&dir->i_mutex);
1243 dentry = d_lookup(parent, name);
1244 if (likely(!dentry)) {
1245 dentry = d_alloc_and_lookup(parent, name, nd);
1246 if (IS_ERR(dentry)) {
1247 mutex_unlock(&dir->i_mutex);
1248 return PTR_ERR(dentry);
1254 mutex_unlock(&dir->i_mutex);
1256 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1257 status = d_revalidate(dentry, nd);
1258 if (unlikely(status <= 0)) {
1263 if (!d_invalidate(dentry)) {
1272 path->dentry = dentry;
1273 err = follow_managed(path, nd->flags);
1274 if (unlikely(err < 0)) {
1275 path_put_conditional(path, nd);
1278 *inode = path->dentry->d_inode;
1282 static inline int may_lookup(struct nameidata *nd)
1284 if (nd->flags & LOOKUP_RCU) {
1285 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1288 if (nameidata_drop_rcu(nd))
1291 return exec_permission(nd->inode, 0);
1294 static inline int handle_dots(struct nameidata *nd, int type)
1296 if (type == LAST_DOTDOT) {
1297 if (nd->flags & LOOKUP_RCU) {
1298 if (follow_dotdot_rcu(nd))
1306 static void terminate_walk(struct nameidata *nd)
1308 if (!(nd->flags & LOOKUP_RCU)) {
1309 path_put(&nd->path);
1311 nd->flags &= ~LOOKUP_RCU;
1312 if (!(nd->flags & LOOKUP_ROOT))
1313 nd->root.mnt = NULL;
1315 br_read_unlock(vfsmount_lock);
1319 static inline int walk_component(struct nameidata *nd, struct path *path,
1320 struct qstr *name, int type, int follow)
1322 struct inode *inode;
1325 * "." and ".." are special - ".." especially so because it has
1326 * to be able to know about the current root directory and
1327 * parent relationships.
1329 if (unlikely(type != LAST_NORM))
1330 return handle_dots(nd, type);
1331 err = do_lookup(nd, name, path, &inode);
1332 if (unlikely(err)) {
1337 path_to_nameidata(path, nd);
1341 if (unlikely(inode->i_op->follow_link) && follow) {
1342 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
1344 BUG_ON(inode != path->dentry->d_inode);
1347 path_to_nameidata(path, nd);
1353 * This limits recursive symlink follows to 8, while
1354 * limiting consecutive symlinks to 40.
1356 * Without that kind of total limit, nasty chains of consecutive
1357 * symlinks can cause almost arbitrarily long lookups.
1359 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1363 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1364 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1365 path_put_conditional(path, nd);
1366 path_put(&nd->path);
1371 current->link_count++;
1374 struct path link = *path;
1377 res = follow_link(&link, nd, &cookie);
1379 res = walk_component(nd, path, &nd->last,
1380 nd->last_type, LOOKUP_FOLLOW);
1381 put_link(nd, &link, cookie);
1384 current->link_count--;
1391 * This is the basic name resolution function, turning a pathname into
1392 * the final dentry. We expect 'base' to be positive and a directory.
1394 * Returns 0 and nd will have valid dentry and mnt on success.
1395 * Returns error and drops reference to input namei data on failure.
1397 static int link_path_walk(const char *name, struct nameidata *nd)
1401 unsigned int lookup_flags = nd->flags;
1408 /* At this point we know we have a real path component. */
1415 nd->flags |= LOOKUP_CONTINUE;
1417 err = may_lookup(nd);
1422 c = *(const unsigned char *)name;
1424 hash = init_name_hash();
1427 hash = partial_name_hash(c, hash);
1428 c = *(const unsigned char *)name;
1429 } while (c && (c != '/'));
1430 this.len = name - (const char *) this.name;
1431 this.hash = end_name_hash(hash);
1434 if (this.name[0] == '.') switch (this.len) {
1436 if (this.name[1] == '.') {
1438 nd->flags |= LOOKUP_JUMPED;
1444 if (likely(type == LAST_NORM)) {
1445 struct dentry *parent = nd->path.dentry;
1446 nd->flags &= ~LOOKUP_JUMPED;
1447 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1448 err = parent->d_op->d_hash(parent, nd->inode,
1455 /* remove trailing slashes? */
1457 goto last_component;
1458 while (*++name == '/');
1460 goto last_component;
1462 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1467 err = nested_symlink(&next, nd);
1472 if (!nd->inode->i_op->lookup)
1475 /* here ends the main loop */
1478 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1479 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1481 nd->last_type = type;
1488 static int path_init(int dfd, const char *name, unsigned int flags,
1489 struct nameidata *nd, struct file **fp)
1495 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1496 nd->flags = flags | LOOKUP_JUMPED;
1498 if (flags & LOOKUP_ROOT) {
1499 struct inode *inode = nd->root.dentry->d_inode;
1501 if (!inode->i_op->lookup)
1503 retval = inode_permission(inode, MAY_EXEC);
1507 nd->path = nd->root;
1509 if (flags & LOOKUP_RCU) {
1510 br_read_lock(vfsmount_lock);
1512 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1514 path_get(&nd->path);
1519 nd->root.mnt = NULL;
1522 if (flags & LOOKUP_RCU) {
1523 br_read_lock(vfsmount_lock);
1528 path_get(&nd->root);
1530 nd->path = nd->root;
1531 } else if (dfd == AT_FDCWD) {
1532 if (flags & LOOKUP_RCU) {
1533 struct fs_struct *fs = current->fs;
1536 br_read_lock(vfsmount_lock);
1540 seq = read_seqcount_begin(&fs->seq);
1542 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1543 } while (read_seqcount_retry(&fs->seq, seq));
1545 get_fs_pwd(current->fs, &nd->path);
1548 struct dentry *dentry;
1550 file = fget_raw_light(dfd, &fput_needed);
1555 dentry = file->f_path.dentry;
1559 if (!S_ISDIR(dentry->d_inode->i_mode))
1562 retval = file_permission(file, MAY_EXEC);
1567 nd->path = file->f_path;
1568 if (flags & LOOKUP_RCU) {
1571 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1572 br_read_lock(vfsmount_lock);
1575 path_get(&file->f_path);
1576 fput_light(file, fput_needed);
1580 nd->inode = nd->path.dentry->d_inode;
1584 fput_light(file, fput_needed);
1589 static inline int lookup_last(struct nameidata *nd, struct path *path)
1591 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1592 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1594 nd->flags &= ~LOOKUP_PARENT;
1595 return walk_component(nd, path, &nd->last, nd->last_type,
1596 nd->flags & LOOKUP_FOLLOW);
1599 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1600 static int path_lookupat(int dfd, const char *name,
1601 unsigned int flags, struct nameidata *nd)
1603 struct file *base = NULL;
1608 * Path walking is largely split up into 2 different synchronisation
1609 * schemes, rcu-walk and ref-walk (explained in
1610 * Documentation/filesystems/path-lookup.txt). These share much of the
1611 * path walk code, but some things particularly setup, cleanup, and
1612 * following mounts are sufficiently divergent that functions are
1613 * duplicated. Typically there is a function foo(), and its RCU
1614 * analogue, foo_rcu().
1616 * -ECHILD is the error number of choice (just to avoid clashes) that
1617 * is returned if some aspect of an rcu-walk fails. Such an error must
1618 * be handled by restarting a traditional ref-walk (which will always
1619 * be able to complete).
1621 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1626 current->total_link_count = 0;
1627 err = link_path_walk(name, nd);
1629 if (!err && !(flags & LOOKUP_PARENT)) {
1630 err = lookup_last(nd, &path);
1633 struct path link = path;
1634 nd->flags |= LOOKUP_PARENT;
1635 err = follow_link(&link, nd, &cookie);
1637 err = lookup_last(nd, &path);
1638 put_link(nd, &link, cookie);
1642 if (nd->flags & LOOKUP_RCU) {
1643 /* went all way through without dropping RCU */
1645 if (nameidata_drop_rcu_last(nd))
1650 err = handle_reval_path(nd);
1652 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1653 if (!nd->inode->i_op->lookup) {
1654 path_put(&nd->path);
1662 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1663 path_put(&nd->root);
1664 nd->root.mnt = NULL;
1669 static int do_path_lookup(int dfd, const char *name,
1670 unsigned int flags, struct nameidata *nd)
1672 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1673 if (unlikely(retval == -ECHILD))
1674 retval = path_lookupat(dfd, name, flags, nd);
1675 if (unlikely(retval == -ESTALE))
1676 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1678 if (likely(!retval)) {
1679 if (unlikely(!audit_dummy_context())) {
1680 if (nd->path.dentry && nd->inode)
1681 audit_inode(name, nd->path.dentry);
1687 int kern_path_parent(const char *name, struct nameidata *nd)
1689 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1692 int kern_path(const char *name, unsigned int flags, struct path *path)
1694 struct nameidata nd;
1695 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1702 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1703 * @dentry: pointer to dentry of the base directory
1704 * @mnt: pointer to vfs mount of the base directory
1705 * @name: pointer to file name
1706 * @flags: lookup flags
1707 * @nd: pointer to nameidata
1709 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1710 const char *name, unsigned int flags,
1711 struct nameidata *nd)
1713 nd->root.dentry = dentry;
1715 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1716 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
1719 static struct dentry *__lookup_hash(struct qstr *name,
1720 struct dentry *base, struct nameidata *nd)
1722 struct inode *inode = base->d_inode;
1723 struct dentry *dentry;
1726 err = exec_permission(inode, 0);
1728 return ERR_PTR(err);
1731 * Don't bother with __d_lookup: callers are for creat as
1732 * well as unlink, so a lot of the time it would cost
1735 dentry = d_lookup(base, name);
1737 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
1738 dentry = do_revalidate(dentry, nd);
1741 dentry = d_alloc_and_lookup(base, name, nd);
1747 * Restricted form of lookup. Doesn't follow links, single-component only,
1748 * needs parent already locked. Doesn't follow mounts.
1751 static struct dentry *lookup_hash(struct nameidata *nd)
1753 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1757 * lookup_one_len - filesystem helper to lookup single pathname component
1758 * @name: pathname component to lookup
1759 * @base: base directory to lookup from
1760 * @len: maximum length @len should be interpreted to
1762 * Note that this routine is purely a helper for filesystem usage and should
1763 * not be called by generic code. Also note that by using this function the
1764 * nameidata argument is passed to the filesystem methods and a filesystem
1765 * using this helper needs to be prepared for that.
1767 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1773 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1778 return ERR_PTR(-EACCES);
1780 hash = init_name_hash();
1782 c = *(const unsigned char *)name++;
1783 if (c == '/' || c == '\0')
1784 return ERR_PTR(-EACCES);
1785 hash = partial_name_hash(c, hash);
1787 this.hash = end_name_hash(hash);
1789 * See if the low-level filesystem might want
1790 * to use its own hash..
1792 if (base->d_flags & DCACHE_OP_HASH) {
1793 int err = base->d_op->d_hash(base, base->d_inode, &this);
1795 return ERR_PTR(err);
1798 return __lookup_hash(&this, base, NULL);
1801 int user_path_at(int dfd, const char __user *name, unsigned flags,
1804 struct nameidata nd;
1805 char *tmp = getname_flags(name, flags);
1806 int err = PTR_ERR(tmp);
1809 BUG_ON(flags & LOOKUP_PARENT);
1811 err = do_path_lookup(dfd, tmp, flags, &nd);
1819 static int user_path_parent(int dfd, const char __user *path,
1820 struct nameidata *nd, char **name)
1822 char *s = getname(path);
1828 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1838 * It's inline, so penalty for filesystems that don't use sticky bit is
1841 static inline int check_sticky(struct inode *dir, struct inode *inode)
1843 uid_t fsuid = current_fsuid();
1845 if (!(dir->i_mode & S_ISVTX))
1847 if (inode->i_uid == fsuid)
1849 if (dir->i_uid == fsuid)
1851 return !capable(CAP_FOWNER);
1855 * Check whether we can remove a link victim from directory dir, check
1856 * whether the type of victim is right.
1857 * 1. We can't do it if dir is read-only (done in permission())
1858 * 2. We should have write and exec permissions on dir
1859 * 3. We can't remove anything from append-only dir
1860 * 4. We can't do anything with immutable dir (done in permission())
1861 * 5. If the sticky bit on dir is set we should either
1862 * a. be owner of dir, or
1863 * b. be owner of victim, or
1864 * c. have CAP_FOWNER capability
1865 * 6. If the victim is append-only or immutable we can't do antyhing with
1866 * links pointing to it.
1867 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1868 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1869 * 9. We can't remove a root or mountpoint.
1870 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1871 * nfs_async_unlink().
1873 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1877 if (!victim->d_inode)
1880 BUG_ON(victim->d_parent->d_inode != dir);
1881 audit_inode_child(victim, dir);
1883 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1888 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1889 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1892 if (!S_ISDIR(victim->d_inode->i_mode))
1894 if (IS_ROOT(victim))
1896 } else if (S_ISDIR(victim->d_inode->i_mode))
1898 if (IS_DEADDIR(dir))
1900 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1905 /* Check whether we can create an object with dentry child in directory
1907 * 1. We can't do it if child already exists (open has special treatment for
1908 * this case, but since we are inlined it's OK)
1909 * 2. We can't do it if dir is read-only (done in permission())
1910 * 3. We should have write and exec permissions on dir
1911 * 4. We can't do it if dir is immutable (done in permission())
1913 static inline int may_create(struct inode *dir, struct dentry *child)
1917 if (IS_DEADDIR(dir))
1919 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1923 * p1 and p2 should be directories on the same fs.
1925 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1930 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1934 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1936 p = d_ancestor(p2, p1);
1938 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1939 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1943 p = d_ancestor(p1, p2);
1945 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1946 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1950 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1951 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1955 void unlock_rename(struct dentry *p1, struct dentry *p2)
1957 mutex_unlock(&p1->d_inode->i_mutex);
1959 mutex_unlock(&p2->d_inode->i_mutex);
1960 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1964 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1965 struct nameidata *nd)
1967 int error = may_create(dir, dentry);
1972 if (!dir->i_op->create)
1973 return -EACCES; /* shouldn't it be ENOSYS? */
1976 error = security_inode_create(dir, dentry, mode);
1979 error = dir->i_op->create(dir, dentry, mode, nd);
1981 fsnotify_create(dir, dentry);
1985 static int may_open(struct path *path, int acc_mode, int flag)
1987 struct dentry *dentry = path->dentry;
1988 struct inode *inode = dentry->d_inode;
1998 switch (inode->i_mode & S_IFMT) {
2002 if (acc_mode & MAY_WRITE)
2007 if (path->mnt->mnt_flags & MNT_NODEV)
2016 error = inode_permission(inode, acc_mode);
2021 * An append-only file must be opened in append mode for writing.
2023 if (IS_APPEND(inode)) {
2024 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2030 /* O_NOATIME can only be set by the owner or superuser */
2031 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2035 * Ensure there are no outstanding leases on the file.
2037 return break_lease(inode, flag);
2040 static int handle_truncate(struct file *filp)
2042 struct path *path = &filp->f_path;
2043 struct inode *inode = path->dentry->d_inode;
2044 int error = get_write_access(inode);
2048 * Refuse to truncate files with mandatory locks held on them.
2050 error = locks_verify_locked(inode);
2052 error = security_path_truncate(path);
2054 error = do_truncate(path->dentry, 0,
2055 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2058 put_write_access(inode);
2063 * Note that while the flag value (low two bits) for sys_open means:
2068 * it is changed into
2069 * 00 - no permissions needed
2070 * 01 - read-permission
2071 * 10 - write-permission
2073 * for the internal routines (ie open_namei()/follow_link() etc)
2074 * This is more logical, and also allows the 00 "no perm needed"
2075 * to be used for symlinks (where the permissions are checked
2079 static inline int open_to_namei_flags(int flag)
2081 if ((flag+1) & O_ACCMODE)
2087 * Handle the last step of open()
2089 static struct file *do_last(struct nameidata *nd, struct path *path,
2090 const struct open_flags *op, const char *pathname)
2092 struct dentry *dir = nd->path.dentry;
2093 struct dentry *dentry;
2094 int open_flag = op->open_flag;
2095 int will_truncate = open_flag & O_TRUNC;
2097 int acc_mode = op->acc_mode;
2101 nd->flags &= ~LOOKUP_PARENT;
2102 nd->flags |= op->intent;
2104 switch (nd->last_type) {
2107 error = handle_dots(nd, nd->last_type);
2109 return ERR_PTR(error);
2112 if (nd->flags & LOOKUP_RCU) {
2113 if (nameidata_drop_rcu_last(nd))
2114 return ERR_PTR(-ECHILD);
2116 error = handle_reval_path(nd);
2119 audit_inode(pathname, nd->path.dentry);
2120 if (open_flag & O_CREAT) {
2126 /* can't be RCU mode here */
2127 error = handle_reval_path(nd);
2130 audit_inode(pathname, dir);
2134 if (!(open_flag & O_CREAT)) {
2136 if (nd->last.name[nd->last.len])
2137 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2138 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2140 /* we _can_ be in RCU mode here */
2141 error = walk_component(nd, path, &nd->last, LAST_NORM,
2144 return ERR_PTR(error);
2145 if (error) /* symlink */
2148 if (nd->flags & LOOKUP_RCU) {
2149 if (nameidata_drop_rcu_last(nd))
2150 return ERR_PTR(-ECHILD);
2154 if (nd->flags & LOOKUP_DIRECTORY) {
2155 if (!nd->inode->i_op->lookup)
2158 audit_inode(pathname, nd->path.dentry);
2162 /* create side of things */
2164 if (nd->flags & LOOKUP_RCU) {
2165 if (nameidata_drop_rcu_last(nd))
2166 return ERR_PTR(-ECHILD);
2169 audit_inode(pathname, dir);
2171 /* trailing slashes? */
2172 if (nd->last.name[nd->last.len])
2175 mutex_lock(&dir->d_inode->i_mutex);
2177 dentry = lookup_hash(nd);
2178 error = PTR_ERR(dentry);
2179 if (IS_ERR(dentry)) {
2180 mutex_unlock(&dir->d_inode->i_mutex);
2184 path->dentry = dentry;
2185 path->mnt = nd->path.mnt;
2187 /* Negative dentry, just create the file */
2188 if (!dentry->d_inode) {
2189 int mode = op->mode;
2190 if (!IS_POSIXACL(dir->d_inode))
2191 mode &= ~current_umask();
2193 * This write is needed to ensure that a
2194 * rw->ro transition does not occur between
2195 * the time when the file is created and when
2196 * a permanent write count is taken through
2197 * the 'struct file' in nameidata_to_filp().
2199 error = mnt_want_write(nd->path.mnt);
2201 goto exit_mutex_unlock;
2203 /* Don't check for write permission, don't truncate */
2204 open_flag &= ~O_TRUNC;
2206 acc_mode = MAY_OPEN;
2207 error = security_path_mknod(&nd->path, dentry, mode, 0);
2209 goto exit_mutex_unlock;
2210 error = vfs_create(dir->d_inode, dentry, mode, nd);
2212 goto exit_mutex_unlock;
2213 mutex_unlock(&dir->d_inode->i_mutex);
2214 dput(nd->path.dentry);
2215 nd->path.dentry = dentry;
2220 * It already exists.
2222 mutex_unlock(&dir->d_inode->i_mutex);
2223 audit_inode(pathname, path->dentry);
2226 if (open_flag & O_EXCL)
2229 error = follow_managed(path, nd->flags);
2234 if (!path->dentry->d_inode)
2237 if (path->dentry->d_inode->i_op->follow_link)
2240 path_to_nameidata(path, nd);
2241 nd->inode = path->dentry->d_inode;
2243 if (S_ISDIR(nd->inode->i_mode))
2246 if (!S_ISREG(nd->inode->i_mode))
2249 if (will_truncate) {
2250 error = mnt_want_write(nd->path.mnt);
2256 error = may_open(&nd->path, acc_mode, open_flag);
2259 filp = nameidata_to_filp(nd);
2260 if (!IS_ERR(filp)) {
2261 error = ima_file_check(filp, op->acc_mode);
2264 filp = ERR_PTR(error);
2267 if (!IS_ERR(filp)) {
2268 if (will_truncate) {
2269 error = handle_truncate(filp);
2272 filp = ERR_PTR(error);
2278 mnt_drop_write(nd->path.mnt);
2279 path_put(&nd->path);
2283 mutex_unlock(&dir->d_inode->i_mutex);
2285 path_put_conditional(path, nd);
2287 filp = ERR_PTR(error);
2291 static struct file *path_openat(int dfd, const char *pathname,
2292 struct nameidata *nd, const struct open_flags *op, int flags)
2294 struct file *base = NULL;
2299 filp = get_empty_filp();
2301 return ERR_PTR(-ENFILE);
2303 filp->f_flags = op->open_flag;
2304 nd->intent.open.file = filp;
2305 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2306 nd->intent.open.create_mode = op->mode;
2308 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2309 if (unlikely(error))
2312 current->total_link_count = 0;
2313 error = link_path_walk(pathname, nd);
2314 if (unlikely(error))
2317 filp = do_last(nd, &path, op, pathname);
2318 while (unlikely(!filp)) { /* trailing symlink */
2319 struct path link = path;
2321 if (!(nd->flags & LOOKUP_FOLLOW)) {
2322 path_put_conditional(&path, nd);
2323 path_put(&nd->path);
2324 filp = ERR_PTR(-ELOOP);
2327 nd->flags |= LOOKUP_PARENT;
2328 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2329 error = follow_link(&link, nd, &cookie);
2330 if (unlikely(error))
2331 filp = ERR_PTR(error);
2333 filp = do_last(nd, &path, op, pathname);
2334 put_link(nd, &link, cookie);
2337 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2338 path_put(&nd->root);
2341 release_open_intent(nd);
2345 filp = ERR_PTR(error);
2349 struct file *do_filp_open(int dfd, const char *pathname,
2350 const struct open_flags *op, int flags)
2352 struct nameidata nd;
2355 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2356 if (unlikely(filp == ERR_PTR(-ECHILD)))
2357 filp = path_openat(dfd, pathname, &nd, op, flags);
2358 if (unlikely(filp == ERR_PTR(-ESTALE)))
2359 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2363 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2364 const char *name, const struct open_flags *op, int flags)
2366 struct nameidata nd;
2370 nd.root.dentry = dentry;
2372 flags |= LOOKUP_ROOT;
2374 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2375 return ERR_PTR(-ELOOP);
2377 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2378 if (unlikely(file == ERR_PTR(-ECHILD)))
2379 file = path_openat(-1, name, &nd, op, flags);
2380 if (unlikely(file == ERR_PTR(-ESTALE)))
2381 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2386 * lookup_create - lookup a dentry, creating it if it doesn't exist
2387 * @nd: nameidata info
2388 * @is_dir: directory flag
2390 * Simple function to lookup and return a dentry and create it
2391 * if it doesn't exist. Is SMP-safe.
2393 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2395 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2397 struct dentry *dentry = ERR_PTR(-EEXIST);
2399 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2401 * Yucky last component or no last component at all?
2402 * (foo/., foo/.., /////)
2404 if (nd->last_type != LAST_NORM)
2406 nd->flags &= ~LOOKUP_PARENT;
2407 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2408 nd->intent.open.flags = O_EXCL;
2411 * Do the final lookup.
2413 dentry = lookup_hash(nd);
2417 if (dentry->d_inode)
2420 * Special case - lookup gave negative, but... we had foo/bar/
2421 * From the vfs_mknod() POV we just have a negative dentry -
2422 * all is fine. Let's be bastards - you had / on the end, you've
2423 * been asking for (non-existent) directory. -ENOENT for you.
2425 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2427 dentry = ERR_PTR(-ENOENT);
2432 dentry = ERR_PTR(-EEXIST);
2436 EXPORT_SYMBOL_GPL(lookup_create);
2438 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2440 int error = may_create(dir, dentry);
2445 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2448 if (!dir->i_op->mknod)
2451 error = devcgroup_inode_mknod(mode, dev);
2455 error = security_inode_mknod(dir, dentry, mode, dev);
2459 error = dir->i_op->mknod(dir, dentry, mode, dev);
2461 fsnotify_create(dir, dentry);
2465 static int may_mknod(mode_t mode)
2467 switch (mode & S_IFMT) {
2473 case 0: /* zero mode translates to S_IFREG */
2482 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2487 struct dentry *dentry;
2488 struct nameidata nd;
2493 error = user_path_parent(dfd, filename, &nd, &tmp);
2497 dentry = lookup_create(&nd, 0);
2498 if (IS_ERR(dentry)) {
2499 error = PTR_ERR(dentry);
2502 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2503 mode &= ~current_umask();
2504 error = may_mknod(mode);
2507 error = mnt_want_write(nd.path.mnt);
2510 error = security_path_mknod(&nd.path, dentry, mode, dev);
2512 goto out_drop_write;
2513 switch (mode & S_IFMT) {
2514 case 0: case S_IFREG:
2515 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2517 case S_IFCHR: case S_IFBLK:
2518 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2519 new_decode_dev(dev));
2521 case S_IFIFO: case S_IFSOCK:
2522 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2526 mnt_drop_write(nd.path.mnt);
2530 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2537 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2539 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2542 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2544 int error = may_create(dir, dentry);
2549 if (!dir->i_op->mkdir)
2552 mode &= (S_IRWXUGO|S_ISVTX);
2553 error = security_inode_mkdir(dir, dentry, mode);
2557 error = dir->i_op->mkdir(dir, dentry, mode);
2559 fsnotify_mkdir(dir, dentry);
2563 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2567 struct dentry *dentry;
2568 struct nameidata nd;
2570 error = user_path_parent(dfd, pathname, &nd, &tmp);
2574 dentry = lookup_create(&nd, 1);
2575 error = PTR_ERR(dentry);
2579 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2580 mode &= ~current_umask();
2581 error = mnt_want_write(nd.path.mnt);
2584 error = security_path_mkdir(&nd.path, dentry, mode);
2586 goto out_drop_write;
2587 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2589 mnt_drop_write(nd.path.mnt);
2593 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2600 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2602 return sys_mkdirat(AT_FDCWD, pathname, mode);
2606 * We try to drop the dentry early: we should have
2607 * a usage count of 2 if we're the only user of this
2608 * dentry, and if that is true (possibly after pruning
2609 * the dcache), then we drop the dentry now.
2611 * A low-level filesystem can, if it choses, legally
2614 * if (!d_unhashed(dentry))
2617 * if it cannot handle the case of removing a directory
2618 * that is still in use by something else..
2620 void dentry_unhash(struct dentry *dentry)
2623 shrink_dcache_parent(dentry);
2624 spin_lock(&dentry->d_lock);
2625 if (dentry->d_count == 2)
2627 spin_unlock(&dentry->d_lock);
2630 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2632 int error = may_delete(dir, dentry, 1);
2637 if (!dir->i_op->rmdir)
2640 mutex_lock(&dentry->d_inode->i_mutex);
2641 dentry_unhash(dentry);
2642 if (d_mountpoint(dentry))
2645 error = security_inode_rmdir(dir, dentry);
2647 error = dir->i_op->rmdir(dir, dentry);
2649 dentry->d_inode->i_flags |= S_DEAD;
2654 mutex_unlock(&dentry->d_inode->i_mutex);
2663 static long do_rmdir(int dfd, const char __user *pathname)
2667 struct dentry *dentry;
2668 struct nameidata nd;
2670 error = user_path_parent(dfd, pathname, &nd, &name);
2674 switch(nd.last_type) {
2686 nd.flags &= ~LOOKUP_PARENT;
2688 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2689 dentry = lookup_hash(&nd);
2690 error = PTR_ERR(dentry);
2693 error = mnt_want_write(nd.path.mnt);
2696 error = security_path_rmdir(&nd.path, dentry);
2699 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2701 mnt_drop_write(nd.path.mnt);
2705 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2712 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2714 return do_rmdir(AT_FDCWD, pathname);
2717 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2719 int error = may_delete(dir, dentry, 0);
2724 if (!dir->i_op->unlink)
2727 mutex_lock(&dentry->d_inode->i_mutex);
2728 if (d_mountpoint(dentry))
2731 error = security_inode_unlink(dir, dentry);
2733 error = dir->i_op->unlink(dir, dentry);
2738 mutex_unlock(&dentry->d_inode->i_mutex);
2740 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2741 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2742 fsnotify_link_count(dentry->d_inode);
2750 * Make sure that the actual truncation of the file will occur outside its
2751 * directory's i_mutex. Truncate can take a long time if there is a lot of
2752 * writeout happening, and we don't want to prevent access to the directory
2753 * while waiting on the I/O.
2755 static long do_unlinkat(int dfd, const char __user *pathname)
2759 struct dentry *dentry;
2760 struct nameidata nd;
2761 struct inode *inode = NULL;
2763 error = user_path_parent(dfd, pathname, &nd, &name);
2768 if (nd.last_type != LAST_NORM)
2771 nd.flags &= ~LOOKUP_PARENT;
2773 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2774 dentry = lookup_hash(&nd);
2775 error = PTR_ERR(dentry);
2776 if (!IS_ERR(dentry)) {
2777 /* Why not before? Because we want correct error value */
2778 if (nd.last.name[nd.last.len])
2780 inode = dentry->d_inode;
2783 error = mnt_want_write(nd.path.mnt);
2786 error = security_path_unlink(&nd.path, dentry);
2789 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2791 mnt_drop_write(nd.path.mnt);
2795 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2797 iput(inode); /* truncate the inode here */
2804 error = !dentry->d_inode ? -ENOENT :
2805 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2809 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2811 if ((flag & ~AT_REMOVEDIR) != 0)
2814 if (flag & AT_REMOVEDIR)
2815 return do_rmdir(dfd, pathname);
2817 return do_unlinkat(dfd, pathname);
2820 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2822 return do_unlinkat(AT_FDCWD, pathname);
2825 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2827 int error = may_create(dir, dentry);
2832 if (!dir->i_op->symlink)
2835 error = security_inode_symlink(dir, dentry, oldname);
2839 error = dir->i_op->symlink(dir, dentry, oldname);
2841 fsnotify_create(dir, dentry);
2845 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2846 int, newdfd, const char __user *, newname)
2851 struct dentry *dentry;
2852 struct nameidata nd;
2854 from = getname(oldname);
2856 return PTR_ERR(from);
2858 error = user_path_parent(newdfd, newname, &nd, &to);
2862 dentry = lookup_create(&nd, 0);
2863 error = PTR_ERR(dentry);
2867 error = mnt_want_write(nd.path.mnt);
2870 error = security_path_symlink(&nd.path, dentry, from);
2872 goto out_drop_write;
2873 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2875 mnt_drop_write(nd.path.mnt);
2879 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2887 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2889 return sys_symlinkat(oldname, AT_FDCWD, newname);
2892 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2894 struct inode *inode = old_dentry->d_inode;
2900 error = may_create(dir, new_dentry);
2904 if (dir->i_sb != inode->i_sb)
2908 * A link to an append-only or immutable file cannot be created.
2910 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2912 if (!dir->i_op->link)
2914 if (S_ISDIR(inode->i_mode))
2917 error = security_inode_link(old_dentry, dir, new_dentry);
2921 mutex_lock(&inode->i_mutex);
2922 /* Make sure we don't allow creating hardlink to an unlinked file */
2923 if (inode->i_nlink == 0)
2926 error = dir->i_op->link(old_dentry, dir, new_dentry);
2927 mutex_unlock(&inode->i_mutex);
2929 fsnotify_link(dir, inode, new_dentry);
2934 * Hardlinks are often used in delicate situations. We avoid
2935 * security-related surprises by not following symlinks on the
2938 * We don't follow them on the oldname either to be compatible
2939 * with linux 2.0, and to avoid hard-linking to directories
2940 * and other special files. --ADM
2942 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2943 int, newdfd, const char __user *, newname, int, flags)
2945 struct dentry *new_dentry;
2946 struct nameidata nd;
2947 struct path old_path;
2952 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
2955 * To use null names we require CAP_DAC_READ_SEARCH
2956 * This ensures that not everyone will be able to create
2957 * handlink using the passed filedescriptor.
2959 if (flags & AT_EMPTY_PATH) {
2960 if (!capable(CAP_DAC_READ_SEARCH))
2965 if (flags & AT_SYMLINK_FOLLOW)
2966 how |= LOOKUP_FOLLOW;
2968 error = user_path_at(olddfd, oldname, how, &old_path);
2972 error = user_path_parent(newdfd, newname, &nd, &to);
2976 if (old_path.mnt != nd.path.mnt)
2978 new_dentry = lookup_create(&nd, 0);
2979 error = PTR_ERR(new_dentry);
2980 if (IS_ERR(new_dentry))
2982 error = mnt_want_write(nd.path.mnt);
2985 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2987 goto out_drop_write;
2988 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2990 mnt_drop_write(nd.path.mnt);
2994 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2999 path_put(&old_path);
3004 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3006 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3010 * The worst of all namespace operations - renaming directory. "Perverted"
3011 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3013 * a) we can get into loop creation. Check is done in is_subdir().
3014 * b) race potential - two innocent renames can create a loop together.
3015 * That's where 4.4 screws up. Current fix: serialization on
3016 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3018 * c) we have to lock _three_ objects - parents and victim (if it exists).
3019 * And that - after we got ->i_mutex on parents (until then we don't know
3020 * whether the target exists). Solution: try to be smart with locking
3021 * order for inodes. We rely on the fact that tree topology may change
3022 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3023 * move will be locked. Thus we can rank directories by the tree
3024 * (ancestors first) and rank all non-directories after them.
3025 * That works since everybody except rename does "lock parent, lookup,
3026 * lock child" and rename is under ->s_vfs_rename_mutex.
3027 * HOWEVER, it relies on the assumption that any object with ->lookup()
3028 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3029 * we'd better make sure that there's no link(2) for them.
3030 * d) some filesystems don't support opened-but-unlinked directories,
3031 * either because of layout or because they are not ready to deal with
3032 * all cases correctly. The latter will be fixed (taking this sort of
3033 * stuff into VFS), but the former is not going away. Solution: the same
3034 * trick as in rmdir().
3035 * e) conversion from fhandle to dentry may come in the wrong moment - when
3036 * we are removing the target. Solution: we will have to grab ->i_mutex
3037 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3038 * ->i_mutex on parents, which works but leads to some truly excessive
3041 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3042 struct inode *new_dir, struct dentry *new_dentry)
3045 struct inode *target;
3048 * If we are going to change the parent - check write permissions,
3049 * we'll need to flip '..'.
3051 if (new_dir != old_dir) {
3052 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3057 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3061 target = new_dentry->d_inode;
3063 mutex_lock(&target->i_mutex);
3064 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3068 dentry_unhash(new_dentry);
3069 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3073 target->i_flags |= S_DEAD;
3074 dont_mount(new_dentry);
3076 mutex_unlock(&target->i_mutex);
3077 if (d_unhashed(new_dentry))
3078 d_rehash(new_dentry);
3082 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3083 d_move(old_dentry,new_dentry);
3087 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3088 struct inode *new_dir, struct dentry *new_dentry)
3090 struct inode *target;
3093 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3098 target = new_dentry->d_inode;
3100 mutex_lock(&target->i_mutex);
3101 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3104 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3107 dont_mount(new_dentry);
3108 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3109 d_move(old_dentry, new_dentry);
3112 mutex_unlock(&target->i_mutex);
3117 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3118 struct inode *new_dir, struct dentry *new_dentry)
3121 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3122 const unsigned char *old_name;
3124 if (old_dentry->d_inode == new_dentry->d_inode)
3127 error = may_delete(old_dir, old_dentry, is_dir);
3131 if (!new_dentry->d_inode)
3132 error = may_create(new_dir, new_dentry);
3134 error = may_delete(new_dir, new_dentry, is_dir);
3138 if (!old_dir->i_op->rename)
3141 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3144 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3146 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3148 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3149 new_dentry->d_inode, old_dentry);
3150 fsnotify_oldname_free(old_name);
3155 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3156 int, newdfd, const char __user *, newname)
3158 struct dentry *old_dir, *new_dir;
3159 struct dentry *old_dentry, *new_dentry;
3160 struct dentry *trap;
3161 struct nameidata oldnd, newnd;
3166 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3170 error = user_path_parent(newdfd, newname, &newnd, &to);
3175 if (oldnd.path.mnt != newnd.path.mnt)
3178 old_dir = oldnd.path.dentry;
3180 if (oldnd.last_type != LAST_NORM)
3183 new_dir = newnd.path.dentry;
3184 if (newnd.last_type != LAST_NORM)
3187 oldnd.flags &= ~LOOKUP_PARENT;
3188 newnd.flags &= ~LOOKUP_PARENT;
3189 newnd.flags |= LOOKUP_RENAME_TARGET;
3191 trap = lock_rename(new_dir, old_dir);
3193 old_dentry = lookup_hash(&oldnd);
3194 error = PTR_ERR(old_dentry);
3195 if (IS_ERR(old_dentry))
3197 /* source must exist */
3199 if (!old_dentry->d_inode)
3201 /* unless the source is a directory trailing slashes give -ENOTDIR */
3202 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3204 if (oldnd.last.name[oldnd.last.len])
3206 if (newnd.last.name[newnd.last.len])
3209 /* source should not be ancestor of target */
3211 if (old_dentry == trap)
3213 new_dentry = lookup_hash(&newnd);
3214 error = PTR_ERR(new_dentry);
3215 if (IS_ERR(new_dentry))
3217 /* target should not be an ancestor of source */
3219 if (new_dentry == trap)
3222 error = mnt_want_write(oldnd.path.mnt);
3225 error = security_path_rename(&oldnd.path, old_dentry,
3226 &newnd.path, new_dentry);
3229 error = vfs_rename(old_dir->d_inode, old_dentry,
3230 new_dir->d_inode, new_dentry);
3232 mnt_drop_write(oldnd.path.mnt);
3238 unlock_rename(new_dir, old_dir);
3240 path_put(&newnd.path);
3243 path_put(&oldnd.path);
3249 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3251 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3254 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3258 len = PTR_ERR(link);
3263 if (len > (unsigned) buflen)
3265 if (copy_to_user(buffer, link, len))
3272 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3273 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3274 * using) it for any given inode is up to filesystem.
3276 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3278 struct nameidata nd;
3283 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3285 return PTR_ERR(cookie);
3287 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3288 if (dentry->d_inode->i_op->put_link)
3289 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3293 int vfs_follow_link(struct nameidata *nd, const char *link)
3295 return __vfs_follow_link(nd, link);
3298 /* get the link contents into pagecache */
3299 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3303 struct address_space *mapping = dentry->d_inode->i_mapping;
3304 page = read_mapping_page(mapping, 0, NULL);
3309 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3313 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3315 struct page *page = NULL;
3316 char *s = page_getlink(dentry, &page);
3317 int res = vfs_readlink(dentry,buffer,buflen,s);
3320 page_cache_release(page);
3325 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3327 struct page *page = NULL;
3328 nd_set_link(nd, page_getlink(dentry, &page));
3332 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3334 struct page *page = cookie;
3338 page_cache_release(page);
3343 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3345 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3347 struct address_space *mapping = inode->i_mapping;
3352 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3354 flags |= AOP_FLAG_NOFS;
3357 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3358 flags, &page, &fsdata);
3362 kaddr = kmap_atomic(page, KM_USER0);
3363 memcpy(kaddr, symname, len-1);
3364 kunmap_atomic(kaddr, KM_USER0);
3366 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3373 mark_inode_dirty(inode);
3379 int page_symlink(struct inode *inode, const char *symname, int len)
3381 return __page_symlink(inode, symname, len,
3382 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3385 const struct inode_operations page_symlink_inode_operations = {
3386 .readlink = generic_readlink,
3387 .follow_link = page_follow_link_light,
3388 .put_link = page_put_link,
3391 EXPORT_SYMBOL(user_path_at);
3392 EXPORT_SYMBOL(follow_down_one);
3393 EXPORT_SYMBOL(follow_down);
3394 EXPORT_SYMBOL(follow_up);
3395 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3396 EXPORT_SYMBOL(getname);
3397 EXPORT_SYMBOL(lock_rename);
3398 EXPORT_SYMBOL(lookup_one_len);
3399 EXPORT_SYMBOL(page_follow_link_light);
3400 EXPORT_SYMBOL(page_put_link);
3401 EXPORT_SYMBOL(page_readlink);
3402 EXPORT_SYMBOL(__page_symlink);
3403 EXPORT_SYMBOL(page_symlink);
3404 EXPORT_SYMBOL(page_symlink_inode_operations);
3405 EXPORT_SYMBOL(kern_path_parent);
3406 EXPORT_SYMBOL(kern_path);
3407 EXPORT_SYMBOL(vfs_path_lookup);
3408 EXPORT_SYMBOL(inode_permission);
3409 EXPORT_SYMBOL(file_permission);
3410 EXPORT_SYMBOL(unlock_rename);
3411 EXPORT_SYMBOL(vfs_create);
3412 EXPORT_SYMBOL(vfs_follow_link);
3413 EXPORT_SYMBOL(vfs_link);
3414 EXPORT_SYMBOL(vfs_mkdir);
3415 EXPORT_SYMBOL(vfs_mknod);
3416 EXPORT_SYMBOL(generic_permission);
3417 EXPORT_SYMBOL(vfs_readlink);
3418 EXPORT_SYMBOL(vfs_rename);
3419 EXPORT_SYMBOL(vfs_rmdir);
3420 EXPORT_SYMBOL(vfs_symlink);
3421 EXPORT_SYMBOL(vfs_unlink);
3422 EXPORT_SYMBOL(dentry_unhash);
3423 EXPORT_SYMBOL(generic_readlink);