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 __always_inline int
741 __do_follow_link(const struct path *link, struct nameidata *nd, void **p)
744 struct dentry *dentry = link->dentry;
746 BUG_ON(nd->flags & LOOKUP_RCU);
748 touch_atime(link->mnt, dentry);
749 nd_set_link(nd, NULL);
751 if (link->mnt == nd->path.mnt)
754 error = security_inode_follow_link(link->dentry, nd);
756 *p = ERR_PTR(error); /* no ->put_link(), please */
761 nd->last_type = LAST_BIND;
762 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
765 char *s = nd_get_link(nd);
768 error = __vfs_follow_link(nd, s);
769 else if (nd->last_type == LAST_BIND)
770 nd->flags |= LOOKUP_JUMPED;
776 * This limits recursive symlink follows to 8, while
777 * limiting consecutive symlinks to 40.
779 * Without that kind of total limit, nasty chains of consecutive
780 * symlinks can cause almost arbitrarily long lookups.
782 static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd)
787 /* We drop rcu-walk here */
788 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
790 BUG_ON(inode != path->dentry->d_inode);
792 if (current->link_count >= MAX_NESTED_LINKS)
794 if (current->total_link_count >= 40)
796 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
798 current->link_count++;
799 current->total_link_count++;
801 err = __do_follow_link(path, nd, &cookie);
802 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
803 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
805 current->link_count--;
809 path_put_conditional(path, nd);
814 static int follow_up_rcu(struct path *path)
816 struct vfsmount *parent;
817 struct dentry *mountpoint;
819 parent = path->mnt->mnt_parent;
820 if (parent == path->mnt)
822 mountpoint = path->mnt->mnt_mountpoint;
823 path->dentry = mountpoint;
828 int follow_up(struct path *path)
830 struct vfsmount *parent;
831 struct dentry *mountpoint;
833 br_read_lock(vfsmount_lock);
834 parent = path->mnt->mnt_parent;
835 if (parent == path->mnt) {
836 br_read_unlock(vfsmount_lock);
840 mountpoint = dget(path->mnt->mnt_mountpoint);
841 br_read_unlock(vfsmount_lock);
843 path->dentry = mountpoint;
850 * Perform an automount
851 * - return -EISDIR to tell follow_managed() to stop and return the path we
854 static int follow_automount(struct path *path, unsigned flags,
857 struct vfsmount *mnt;
860 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
863 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
864 * and this is the terminal part of the path.
866 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
867 return -EISDIR; /* we actually want to stop here */
869 /* We want to mount if someone is trying to open/create a file of any
870 * type under the mountpoint, wants to traverse through the mountpoint
871 * or wants to open the mounted directory.
873 * We don't want to mount if someone's just doing a stat and they've
874 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
875 * appended a '/' to the name.
877 if (!(flags & LOOKUP_FOLLOW) &&
878 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
879 LOOKUP_OPEN | LOOKUP_CREATE)))
882 current->total_link_count++;
883 if (current->total_link_count >= 40)
886 mnt = path->dentry->d_op->d_automount(path);
889 * The filesystem is allowed to return -EISDIR here to indicate
890 * it doesn't want to automount. For instance, autofs would do
891 * this so that its userspace daemon can mount on this dentry.
893 * However, we can only permit this if it's a terminal point in
894 * the path being looked up; if it wasn't then the remainder of
895 * the path is inaccessible and we should say so.
897 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
902 if (!mnt) /* mount collision */
905 err = finish_automount(mnt, path);
909 /* Someone else made a mount here whilst we were busy */
916 path->dentry = dget(mnt->mnt_root);
926 * Handle a dentry that is managed in some way.
927 * - Flagged for transit management (autofs)
928 * - Flagged as mountpoint
929 * - Flagged as automount point
931 * This may only be called in refwalk mode.
933 * Serialization is taken care of in namespace.c
935 static int follow_managed(struct path *path, unsigned flags)
938 bool need_mntput = false;
941 /* Given that we're not holding a lock here, we retain the value in a
942 * local variable for each dentry as we look at it so that we don't see
943 * the components of that value change under us */
944 while (managed = ACCESS_ONCE(path->dentry->d_flags),
945 managed &= DCACHE_MANAGED_DENTRY,
946 unlikely(managed != 0)) {
947 /* Allow the filesystem to manage the transit without i_mutex
949 if (managed & DCACHE_MANAGE_TRANSIT) {
950 BUG_ON(!path->dentry->d_op);
951 BUG_ON(!path->dentry->d_op->d_manage);
952 ret = path->dentry->d_op->d_manage(path->dentry,
955 return ret == -EISDIR ? 0 : ret;
958 /* Transit to a mounted filesystem. */
959 if (managed & DCACHE_MOUNTED) {
960 struct vfsmount *mounted = lookup_mnt(path);
966 path->dentry = dget(mounted->mnt_root);
971 /* Something is mounted on this dentry in another
972 * namespace and/or whatever was mounted there in this
973 * namespace got unmounted before we managed to get the
977 /* Handle an automount point */
978 if (managed & DCACHE_NEED_AUTOMOUNT) {
979 ret = follow_automount(path, flags, &need_mntput);
981 return ret == -EISDIR ? 0 : ret;
985 /* We didn't change the current path point */
991 int follow_down_one(struct path *path)
993 struct vfsmount *mounted;
995 mounted = lookup_mnt(path);
1000 path->dentry = dget(mounted->mnt_root);
1007 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
1008 * meet a managed dentry and we're not walking to "..". True is returned to
1009 * continue, false to abort.
1011 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1012 struct inode **inode, bool reverse_transit)
1014 while (d_mountpoint(path->dentry)) {
1015 struct vfsmount *mounted;
1016 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1018 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1020 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1023 path->mnt = mounted;
1024 path->dentry = mounted->mnt_root;
1025 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1026 *inode = path->dentry->d_inode;
1029 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1030 return reverse_transit;
1034 static int follow_dotdot_rcu(struct nameidata *nd)
1036 struct inode *inode = nd->inode;
1041 if (nd->path.dentry == nd->root.dentry &&
1042 nd->path.mnt == nd->root.mnt) {
1045 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1046 struct dentry *old = nd->path.dentry;
1047 struct dentry *parent = old->d_parent;
1050 seq = read_seqcount_begin(&parent->d_seq);
1051 if (read_seqcount_retry(&old->d_seq, nd->seq))
1053 inode = parent->d_inode;
1054 nd->path.dentry = parent;
1058 if (!follow_up_rcu(&nd->path))
1060 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1061 inode = nd->path.dentry->d_inode;
1063 __follow_mount_rcu(nd, &nd->path, &inode, true);
1068 nd->flags &= ~LOOKUP_RCU;
1069 if (!(nd->flags & LOOKUP_ROOT))
1070 nd->root.mnt = NULL;
1072 br_read_unlock(vfsmount_lock);
1077 * Follow down to the covering mount currently visible to userspace. At each
1078 * point, the filesystem owning that dentry may be queried as to whether the
1079 * caller is permitted to proceed or not.
1081 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1084 int follow_down(struct path *path, bool mounting_here)
1089 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1090 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1091 /* Allow the filesystem to manage the transit without i_mutex
1094 * We indicate to the filesystem if someone is trying to mount
1095 * something here. This gives autofs the chance to deny anyone
1096 * other than its daemon the right to mount on its
1099 * The filesystem may sleep at this point.
1101 if (managed & DCACHE_MANAGE_TRANSIT) {
1102 BUG_ON(!path->dentry->d_op);
1103 BUG_ON(!path->dentry->d_op->d_manage);
1104 ret = path->dentry->d_op->d_manage(
1105 path->dentry, mounting_here, false);
1107 return ret == -EISDIR ? 0 : ret;
1110 /* Transit to a mounted filesystem. */
1111 if (managed & DCACHE_MOUNTED) {
1112 struct vfsmount *mounted = lookup_mnt(path);
1117 path->mnt = mounted;
1118 path->dentry = dget(mounted->mnt_root);
1122 /* Don't handle automount points here */
1129 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1131 static void follow_mount(struct path *path)
1133 while (d_mountpoint(path->dentry)) {
1134 struct vfsmount *mounted = lookup_mnt(path);
1139 path->mnt = mounted;
1140 path->dentry = dget(mounted->mnt_root);
1144 static void follow_dotdot(struct nameidata *nd)
1149 struct dentry *old = nd->path.dentry;
1151 if (nd->path.dentry == nd->root.dentry &&
1152 nd->path.mnt == nd->root.mnt) {
1155 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1156 /* rare case of legitimate dget_parent()... */
1157 nd->path.dentry = dget_parent(nd->path.dentry);
1161 if (!follow_up(&nd->path))
1164 follow_mount(&nd->path);
1165 nd->inode = nd->path.dentry->d_inode;
1169 * Allocate a dentry with name and parent, and perform a parent
1170 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1171 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1172 * have verified that no child exists while under i_mutex.
1174 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1175 struct qstr *name, struct nameidata *nd)
1177 struct inode *inode = parent->d_inode;
1178 struct dentry *dentry;
1181 /* Don't create child dentry for a dead directory. */
1182 if (unlikely(IS_DEADDIR(inode)))
1183 return ERR_PTR(-ENOENT);
1185 dentry = d_alloc(parent, name);
1186 if (unlikely(!dentry))
1187 return ERR_PTR(-ENOMEM);
1189 old = inode->i_op->lookup(inode, dentry, nd);
1190 if (unlikely(old)) {
1198 * It's more convoluted than I'd like it to be, but... it's still fairly
1199 * small and for now I'd prefer to have fast path as straight as possible.
1200 * It _is_ time-critical.
1202 static int do_lookup(struct nameidata *nd, struct qstr *name,
1203 struct path *path, struct inode **inode)
1205 struct vfsmount *mnt = nd->path.mnt;
1206 struct dentry *dentry, *parent = nd->path.dentry;
1212 * Rename seqlock is not required here because in the off chance
1213 * of a false negative due to a concurrent rename, we're going to
1214 * do the non-racy lookup, below.
1216 if (nd->flags & LOOKUP_RCU) {
1219 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1223 /* Memory barrier in read_seqcount_begin of child is enough */
1224 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1228 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1229 status = d_revalidate(dentry, nd);
1230 if (unlikely(status <= 0)) {
1231 if (status != -ECHILD)
1237 path->dentry = dentry;
1238 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1242 if (nameidata_dentry_drop_rcu(nd, dentry))
1245 if (nameidata_drop_rcu(nd))
1249 dentry = __d_lookup(parent, name);
1253 if (unlikely(!dentry)) {
1254 struct inode *dir = parent->d_inode;
1255 BUG_ON(nd->inode != dir);
1257 mutex_lock(&dir->i_mutex);
1258 dentry = d_lookup(parent, name);
1259 if (likely(!dentry)) {
1260 dentry = d_alloc_and_lookup(parent, name, nd);
1261 if (IS_ERR(dentry)) {
1262 mutex_unlock(&dir->i_mutex);
1263 return PTR_ERR(dentry);
1269 mutex_unlock(&dir->i_mutex);
1271 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1272 status = d_revalidate(dentry, nd);
1273 if (unlikely(status <= 0)) {
1278 if (!d_invalidate(dentry)) {
1287 path->dentry = dentry;
1288 err = follow_managed(path, nd->flags);
1289 if (unlikely(err < 0)) {
1290 path_put_conditional(path, nd);
1293 *inode = path->dentry->d_inode;
1297 static inline int may_lookup(struct nameidata *nd)
1299 if (nd->flags & LOOKUP_RCU) {
1300 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1303 if (nameidata_drop_rcu(nd))
1306 return exec_permission(nd->inode, 0);
1309 static inline int handle_dots(struct nameidata *nd, int type)
1311 if (type == LAST_DOTDOT) {
1312 if (nd->flags & LOOKUP_RCU) {
1313 if (follow_dotdot_rcu(nd))
1321 static void terminate_walk(struct nameidata *nd)
1323 if (!(nd->flags & LOOKUP_RCU)) {
1324 path_put(&nd->path);
1326 nd->flags &= ~LOOKUP_RCU;
1327 if (!(nd->flags & LOOKUP_ROOT))
1328 nd->root.mnt = NULL;
1330 br_read_unlock(vfsmount_lock);
1336 * This is the basic name resolution function, turning a pathname into
1337 * the final dentry. We expect 'base' to be positive and a directory.
1339 * Returns 0 and nd will have valid dentry and mnt on success.
1340 * Returns error and drops reference to input namei data on failure.
1342 static int link_path_walk(const char *name, struct nameidata *nd)
1346 unsigned int lookup_flags = nd->flags;
1354 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1356 /* At this point we know we have a real path component. */
1358 struct inode *inode;
1364 nd->flags |= LOOKUP_CONTINUE;
1366 err = may_lookup(nd);
1371 c = *(const unsigned char *)name;
1373 hash = init_name_hash();
1376 hash = partial_name_hash(c, hash);
1377 c = *(const unsigned char *)name;
1378 } while (c && (c != '/'));
1379 this.len = name - (const char *) this.name;
1380 this.hash = end_name_hash(hash);
1383 if (this.name[0] == '.') switch (this.len) {
1385 if (this.name[1] == '.') {
1387 nd->flags |= LOOKUP_JUMPED;
1393 if (likely(type == LAST_NORM)) {
1394 struct dentry *parent = nd->path.dentry;
1395 nd->flags &= ~LOOKUP_JUMPED;
1396 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1397 err = parent->d_op->d_hash(parent, nd->inode,
1404 /* remove trailing slashes? */
1406 goto last_component;
1407 while (*++name == '/');
1409 goto last_with_slashes;
1412 * "." and ".." are special - ".." especially so because it has
1413 * to be able to know about the current root directory and
1414 * parent relationships.
1416 if (unlikely(type != LAST_NORM)) {
1417 if (handle_dots(nd, type))
1422 /* This does the actual lookups.. */
1423 err = do_lookup(nd, &this, &next, &inode);
1427 if (inode && inode->i_op->follow_link) {
1428 err = do_follow_link(inode, &next, nd);
1431 nd->inode = nd->path.dentry->d_inode;
1433 path_to_nameidata(&next, nd);
1440 if (!nd->inode->i_op->lookup)
1443 /* here ends the main loop */
1446 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1448 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1449 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1450 if (lookup_flags & LOOKUP_PARENT)
1452 if (unlikely(type != LAST_NORM))
1453 return handle_dots(nd, type);
1454 err = do_lookup(nd, &this, &next, &inode);
1457 if (inode && unlikely(inode->i_op->follow_link) &&
1458 (lookup_flags & LOOKUP_FOLLOW)) {
1459 err = do_follow_link(inode, &next, nd);
1462 nd->inode = nd->path.dentry->d_inode;
1464 path_to_nameidata(&next, nd);
1470 if (lookup_flags & LOOKUP_DIRECTORY) {
1472 if (!nd->inode->i_op->lookup)
1478 nd->last_type = type;
1485 static int path_init(int dfd, const char *name, unsigned int flags,
1486 struct nameidata *nd, struct file **fp)
1492 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1493 nd->flags = flags | LOOKUP_JUMPED;
1495 if (flags & LOOKUP_ROOT) {
1496 struct inode *inode = nd->root.dentry->d_inode;
1498 if (!inode->i_op->lookup)
1500 retval = inode_permission(inode, MAY_EXEC);
1504 nd->path = nd->root;
1506 if (flags & LOOKUP_RCU) {
1507 br_read_lock(vfsmount_lock);
1509 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1511 path_get(&nd->path);
1516 nd->root.mnt = NULL;
1519 if (flags & LOOKUP_RCU) {
1520 br_read_lock(vfsmount_lock);
1525 path_get(&nd->root);
1527 nd->path = nd->root;
1528 } else if (dfd == AT_FDCWD) {
1529 if (flags & LOOKUP_RCU) {
1530 struct fs_struct *fs = current->fs;
1533 br_read_lock(vfsmount_lock);
1537 seq = read_seqcount_begin(&fs->seq);
1539 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1540 } while (read_seqcount_retry(&fs->seq, seq));
1542 get_fs_pwd(current->fs, &nd->path);
1545 struct dentry *dentry;
1547 file = fget_light(dfd, &fput_needed);
1552 dentry = file->f_path.dentry;
1556 if (!S_ISDIR(dentry->d_inode->i_mode))
1559 retval = file_permission(file, MAY_EXEC);
1564 nd->path = file->f_path;
1565 if (flags & LOOKUP_RCU) {
1568 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1569 br_read_lock(vfsmount_lock);
1572 path_get(&file->f_path);
1573 fput_light(file, fput_needed);
1577 nd->inode = nd->path.dentry->d_inode;
1581 fput_light(file, fput_needed);
1586 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1587 static int path_lookupat(int dfd, const char *name,
1588 unsigned int flags, struct nameidata *nd)
1590 struct file *base = NULL;
1594 * Path walking is largely split up into 2 different synchronisation
1595 * schemes, rcu-walk and ref-walk (explained in
1596 * Documentation/filesystems/path-lookup.txt). These share much of the
1597 * path walk code, but some things particularly setup, cleanup, and
1598 * following mounts are sufficiently divergent that functions are
1599 * duplicated. Typically there is a function foo(), and its RCU
1600 * analogue, foo_rcu().
1602 * -ECHILD is the error number of choice (just to avoid clashes) that
1603 * is returned if some aspect of an rcu-walk fails. Such an error must
1604 * be handled by restarting a traditional ref-walk (which will always
1605 * be able to complete).
1607 retval = path_init(dfd, name, flags, nd, &base);
1609 if (unlikely(retval))
1612 current->total_link_count = 0;
1613 retval = link_path_walk(name, nd);
1615 if (nd->flags & LOOKUP_RCU) {
1616 /* went all way through without dropping RCU */
1618 if (nameidata_drop_rcu_last(nd))
1623 retval = handle_reval_path(nd);
1628 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1629 path_put(&nd->root);
1630 nd->root.mnt = NULL;
1635 static int do_path_lookup(int dfd, const char *name,
1636 unsigned int flags, struct nameidata *nd)
1638 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1639 if (unlikely(retval == -ECHILD))
1640 retval = path_lookupat(dfd, name, flags, nd);
1641 if (unlikely(retval == -ESTALE))
1642 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1644 if (likely(!retval)) {
1645 if (unlikely(!audit_dummy_context())) {
1646 if (nd->path.dentry && nd->inode)
1647 audit_inode(name, nd->path.dentry);
1653 int kern_path_parent(const char *name, struct nameidata *nd)
1655 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1658 int kern_path(const char *name, unsigned int flags, struct path *path)
1660 struct nameidata nd;
1661 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1668 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1669 * @dentry: pointer to dentry of the base directory
1670 * @mnt: pointer to vfs mount of the base directory
1671 * @name: pointer to file name
1672 * @flags: lookup flags
1673 * @nd: pointer to nameidata
1675 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1676 const char *name, unsigned int flags,
1677 struct nameidata *nd)
1679 nd->root.dentry = dentry;
1681 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1682 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
1685 static struct dentry *__lookup_hash(struct qstr *name,
1686 struct dentry *base, struct nameidata *nd)
1688 struct inode *inode = base->d_inode;
1689 struct dentry *dentry;
1692 err = exec_permission(inode, 0);
1694 return ERR_PTR(err);
1697 * Don't bother with __d_lookup: callers are for creat as
1698 * well as unlink, so a lot of the time it would cost
1701 dentry = d_lookup(base, name);
1703 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
1704 dentry = do_revalidate(dentry, nd);
1707 dentry = d_alloc_and_lookup(base, name, nd);
1713 * Restricted form of lookup. Doesn't follow links, single-component only,
1714 * needs parent already locked. Doesn't follow mounts.
1717 static struct dentry *lookup_hash(struct nameidata *nd)
1719 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1723 * lookup_one_len - filesystem helper to lookup single pathname component
1724 * @name: pathname component to lookup
1725 * @base: base directory to lookup from
1726 * @len: maximum length @len should be interpreted to
1728 * Note that this routine is purely a helper for filesystem usage and should
1729 * not be called by generic code. Also note that by using this function the
1730 * nameidata argument is passed to the filesystem methods and a filesystem
1731 * using this helper needs to be prepared for that.
1733 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1739 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1744 return ERR_PTR(-EACCES);
1746 hash = init_name_hash();
1748 c = *(const unsigned char *)name++;
1749 if (c == '/' || c == '\0')
1750 return ERR_PTR(-EACCES);
1751 hash = partial_name_hash(c, hash);
1753 this.hash = end_name_hash(hash);
1755 * See if the low-level filesystem might want
1756 * to use its own hash..
1758 if (base->d_flags & DCACHE_OP_HASH) {
1759 int err = base->d_op->d_hash(base, base->d_inode, &this);
1761 return ERR_PTR(err);
1764 return __lookup_hash(&this, base, NULL);
1767 int user_path_at(int dfd, const char __user *name, unsigned flags,
1770 struct nameidata nd;
1771 char *tmp = getname_flags(name, flags);
1772 int err = PTR_ERR(tmp);
1775 BUG_ON(flags & LOOKUP_PARENT);
1777 err = do_path_lookup(dfd, tmp, flags, &nd);
1785 static int user_path_parent(int dfd, const char __user *path,
1786 struct nameidata *nd, char **name)
1788 char *s = getname(path);
1794 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1804 * It's inline, so penalty for filesystems that don't use sticky bit is
1807 static inline int check_sticky(struct inode *dir, struct inode *inode)
1809 uid_t fsuid = current_fsuid();
1811 if (!(dir->i_mode & S_ISVTX))
1813 if (inode->i_uid == fsuid)
1815 if (dir->i_uid == fsuid)
1817 return !capable(CAP_FOWNER);
1821 * Check whether we can remove a link victim from directory dir, check
1822 * whether the type of victim is right.
1823 * 1. We can't do it if dir is read-only (done in permission())
1824 * 2. We should have write and exec permissions on dir
1825 * 3. We can't remove anything from append-only dir
1826 * 4. We can't do anything with immutable dir (done in permission())
1827 * 5. If the sticky bit on dir is set we should either
1828 * a. be owner of dir, or
1829 * b. be owner of victim, or
1830 * c. have CAP_FOWNER capability
1831 * 6. If the victim is append-only or immutable we can't do antyhing with
1832 * links pointing to it.
1833 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1834 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1835 * 9. We can't remove a root or mountpoint.
1836 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1837 * nfs_async_unlink().
1839 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1843 if (!victim->d_inode)
1846 BUG_ON(victim->d_parent->d_inode != dir);
1847 audit_inode_child(victim, dir);
1849 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1854 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1855 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1858 if (!S_ISDIR(victim->d_inode->i_mode))
1860 if (IS_ROOT(victim))
1862 } else if (S_ISDIR(victim->d_inode->i_mode))
1864 if (IS_DEADDIR(dir))
1866 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1871 /* Check whether we can create an object with dentry child in directory
1873 * 1. We can't do it if child already exists (open has special treatment for
1874 * this case, but since we are inlined it's OK)
1875 * 2. We can't do it if dir is read-only (done in permission())
1876 * 3. We should have write and exec permissions on dir
1877 * 4. We can't do it if dir is immutable (done in permission())
1879 static inline int may_create(struct inode *dir, struct dentry *child)
1883 if (IS_DEADDIR(dir))
1885 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1889 * p1 and p2 should be directories on the same fs.
1891 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1896 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1900 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1902 p = d_ancestor(p2, p1);
1904 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1905 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1909 p = d_ancestor(p1, p2);
1911 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1912 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1916 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1917 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1921 void unlock_rename(struct dentry *p1, struct dentry *p2)
1923 mutex_unlock(&p1->d_inode->i_mutex);
1925 mutex_unlock(&p2->d_inode->i_mutex);
1926 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1930 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1931 struct nameidata *nd)
1933 int error = may_create(dir, dentry);
1938 if (!dir->i_op->create)
1939 return -EACCES; /* shouldn't it be ENOSYS? */
1942 error = security_inode_create(dir, dentry, mode);
1945 error = dir->i_op->create(dir, dentry, mode, nd);
1947 fsnotify_create(dir, dentry);
1951 static int may_open(struct path *path, int acc_mode, int flag)
1953 struct dentry *dentry = path->dentry;
1954 struct inode *inode = dentry->d_inode;
1960 switch (inode->i_mode & S_IFMT) {
1964 if (acc_mode & MAY_WRITE)
1969 if (path->mnt->mnt_flags & MNT_NODEV)
1978 error = inode_permission(inode, acc_mode);
1983 * An append-only file must be opened in append mode for writing.
1985 if (IS_APPEND(inode)) {
1986 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
1992 /* O_NOATIME can only be set by the owner or superuser */
1993 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1997 * Ensure there are no outstanding leases on the file.
1999 return break_lease(inode, flag);
2002 static int handle_truncate(struct file *filp)
2004 struct path *path = &filp->f_path;
2005 struct inode *inode = path->dentry->d_inode;
2006 int error = get_write_access(inode);
2010 * Refuse to truncate files with mandatory locks held on them.
2012 error = locks_verify_locked(inode);
2014 error = security_path_truncate(path);
2016 error = do_truncate(path->dentry, 0,
2017 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2020 put_write_access(inode);
2025 * Note that while the flag value (low two bits) for sys_open means:
2030 * it is changed into
2031 * 00 - no permissions needed
2032 * 01 - read-permission
2033 * 10 - write-permission
2035 * for the internal routines (ie open_namei()/follow_link() etc)
2036 * This is more logical, and also allows the 00 "no perm needed"
2037 * to be used for symlinks (where the permissions are checked
2041 static inline int open_to_namei_flags(int flag)
2043 if ((flag+1) & O_ACCMODE)
2049 * Handle the last step of open()
2051 static struct file *do_last(struct nameidata *nd, struct path *path,
2052 const struct open_flags *op, const char *pathname)
2054 struct dentry *dir = nd->path.dentry;
2055 struct dentry *dentry;
2056 int open_flag = op->open_flag;
2057 int will_truncate = open_flag & O_TRUNC;
2061 struct inode *inode;
2064 nd->flags &= ~LOOKUP_PARENT;
2065 nd->flags |= op->intent;
2067 switch (nd->last_type) {
2070 error = handle_dots(nd, nd->last_type);
2072 return ERR_PTR(error);
2075 if (nd->flags & LOOKUP_RCU) {
2076 if (nameidata_drop_rcu_last(nd))
2077 return ERR_PTR(-ECHILD);
2079 error = handle_reval_path(nd);
2082 audit_inode(pathname, nd->path.dentry);
2083 if (open_flag & O_CREAT) {
2089 /* can't be RCU mode here */
2090 error = handle_reval_path(nd);
2093 audit_inode(pathname, dir);
2097 if (!(open_flag & O_CREAT)) {
2098 if (nd->last.name[nd->last.len])
2099 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2100 /* we _can_ be in RCU mode here */
2101 error = do_lookup(nd, &nd->last, path, &inode);
2104 return ERR_PTR(error);
2107 path_to_nameidata(path, nd);
2109 return ERR_PTR(-ENOENT);
2111 if (unlikely(inode->i_op->follow_link)) {
2112 /* We drop rcu-walk here */
2113 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
2114 return ERR_PTR(-ECHILD);
2117 path_to_nameidata(path, nd);
2120 if (nd->flags & LOOKUP_RCU) {
2121 if (nameidata_drop_rcu_last(nd))
2122 return ERR_PTR(-ECHILD);
2126 if (nd->flags & LOOKUP_DIRECTORY) {
2127 if (!inode->i_op->lookup)
2130 audit_inode(pathname, nd->path.dentry);
2134 /* create side of things */
2136 if (nd->flags & LOOKUP_RCU) {
2137 if (nameidata_drop_rcu_last(nd))
2138 return ERR_PTR(-ECHILD);
2141 audit_inode(pathname, dir);
2143 /* trailing slashes? */
2144 if (nd->last.name[nd->last.len])
2147 mutex_lock(&dir->d_inode->i_mutex);
2149 dentry = lookup_hash(nd);
2150 error = PTR_ERR(dentry);
2151 if (IS_ERR(dentry)) {
2152 mutex_unlock(&dir->d_inode->i_mutex);
2156 path->dentry = dentry;
2157 path->mnt = nd->path.mnt;
2159 /* Negative dentry, just create the file */
2160 if (!dentry->d_inode) {
2161 int mode = op->mode;
2162 if (!IS_POSIXACL(dir->d_inode))
2163 mode &= ~current_umask();
2165 * This write is needed to ensure that a
2166 * rw->ro transition does not occur between
2167 * the time when the file is created and when
2168 * a permanent write count is taken through
2169 * the 'struct file' in nameidata_to_filp().
2171 error = mnt_want_write(nd->path.mnt);
2173 goto exit_mutex_unlock;
2175 /* Don't check for write permission, don't truncate */
2176 open_flag &= ~O_TRUNC;
2179 error = security_path_mknod(&nd->path, dentry, mode, 0);
2181 goto exit_mutex_unlock;
2182 error = vfs_create(dir->d_inode, dentry, mode, nd);
2184 goto exit_mutex_unlock;
2185 mutex_unlock(&dir->d_inode->i_mutex);
2186 dput(nd->path.dentry);
2187 nd->path.dentry = dentry;
2192 * It already exists.
2194 mutex_unlock(&dir->d_inode->i_mutex);
2195 audit_inode(pathname, path->dentry);
2198 if (open_flag & O_EXCL)
2201 error = follow_managed(path, nd->flags);
2206 if (!path->dentry->d_inode)
2209 if (path->dentry->d_inode->i_op->follow_link)
2212 path_to_nameidata(path, nd);
2213 nd->inode = path->dentry->d_inode;
2215 if (S_ISDIR(nd->inode->i_mode))
2218 if (!S_ISREG(nd->inode->i_mode))
2221 if (will_truncate) {
2222 error = mnt_want_write(nd->path.mnt);
2228 error = may_open(&nd->path, skip_perm ? 0 : op->acc_mode, open_flag);
2231 filp = nameidata_to_filp(nd);
2232 if (!IS_ERR(filp)) {
2233 error = ima_file_check(filp, op->acc_mode);
2236 filp = ERR_PTR(error);
2239 if (!IS_ERR(filp)) {
2240 if (will_truncate) {
2241 error = handle_truncate(filp);
2244 filp = ERR_PTR(error);
2250 mnt_drop_write(nd->path.mnt);
2251 path_put(&nd->path);
2255 mutex_unlock(&dir->d_inode->i_mutex);
2257 path_put_conditional(path, nd);
2259 filp = ERR_PTR(error);
2263 static struct file *path_openat(int dfd, const char *pathname,
2264 struct nameidata *nd, const struct open_flags *op, int flags)
2266 struct file *base = NULL;
2272 filp = get_empty_filp();
2274 return ERR_PTR(-ENFILE);
2276 filp->f_flags = op->open_flag;
2277 nd->intent.open.file = filp;
2278 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2279 nd->intent.open.create_mode = op->mode;
2281 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2282 if (unlikely(error))
2285 current->total_link_count = 0;
2286 error = link_path_walk(pathname, nd);
2287 if (unlikely(error))
2290 filp = do_last(nd, &path, op, pathname);
2291 while (unlikely(!filp)) { /* trailing symlink */
2292 struct path link = path;
2293 struct inode *linki = link.dentry->d_inode;
2295 if (!(nd->flags & LOOKUP_FOLLOW) || count++ == 32) {
2296 path_put_conditional(&path, nd);
2297 path_put(&nd->path);
2298 filp = ERR_PTR(-ELOOP);
2302 * This is subtle. Instead of calling do_follow_link() we do
2303 * the thing by hands. The reason is that this way we have zero
2304 * link_count and path_walk() (called from ->follow_link)
2305 * honoring LOOKUP_PARENT. After that we have the parent and
2306 * last component, i.e. we are in the same situation as after
2307 * the first path_walk(). Well, almost - if the last component
2308 * is normal we get its copy stored in nd->last.name and we will
2309 * have to putname() it when we are done. Procfs-like symlinks
2310 * just set LAST_BIND.
2312 nd->flags |= LOOKUP_PARENT;
2313 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2314 error = __do_follow_link(&link, nd, &cookie);
2315 if (unlikely(error))
2316 filp = ERR_PTR(error);
2318 filp = do_last(nd, &path, op, pathname);
2319 if (!IS_ERR(cookie) && linki->i_op->put_link)
2320 linki->i_op->put_link(link.dentry, nd, cookie);
2324 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2325 path_put(&nd->root);
2328 release_open_intent(nd);
2332 filp = ERR_PTR(error);
2336 struct file *do_filp_open(int dfd, const char *pathname,
2337 const struct open_flags *op, int flags)
2339 struct nameidata nd;
2342 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2343 if (unlikely(filp == ERR_PTR(-ECHILD)))
2344 filp = path_openat(dfd, pathname, &nd, op, flags);
2345 if (unlikely(filp == ERR_PTR(-ESTALE)))
2346 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2350 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2351 const char *name, const struct open_flags *op, int flags)
2353 struct nameidata nd;
2357 nd.root.dentry = dentry;
2359 flags |= LOOKUP_ROOT;
2361 if (dentry->d_inode->i_op->follow_link)
2362 return ERR_PTR(-ELOOP);
2364 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2365 if (unlikely(file == ERR_PTR(-ECHILD)))
2366 file = path_openat(-1, name, &nd, op, flags);
2367 if (unlikely(file == ERR_PTR(-ESTALE)))
2368 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2373 * lookup_create - lookup a dentry, creating it if it doesn't exist
2374 * @nd: nameidata info
2375 * @is_dir: directory flag
2377 * Simple function to lookup and return a dentry and create it
2378 * if it doesn't exist. Is SMP-safe.
2380 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2382 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2384 struct dentry *dentry = ERR_PTR(-EEXIST);
2386 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2388 * Yucky last component or no last component at all?
2389 * (foo/., foo/.., /////)
2391 if (nd->last_type != LAST_NORM)
2393 nd->flags &= ~LOOKUP_PARENT;
2394 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2395 nd->intent.open.flags = O_EXCL;
2398 * Do the final lookup.
2400 dentry = lookup_hash(nd);
2404 if (dentry->d_inode)
2407 * Special case - lookup gave negative, but... we had foo/bar/
2408 * From the vfs_mknod() POV we just have a negative dentry -
2409 * all is fine. Let's be bastards - you had / on the end, you've
2410 * been asking for (non-existent) directory. -ENOENT for you.
2412 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2414 dentry = ERR_PTR(-ENOENT);
2419 dentry = ERR_PTR(-EEXIST);
2423 EXPORT_SYMBOL_GPL(lookup_create);
2425 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2427 int error = may_create(dir, dentry);
2432 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2435 if (!dir->i_op->mknod)
2438 error = devcgroup_inode_mknod(mode, dev);
2442 error = security_inode_mknod(dir, dentry, mode, dev);
2446 error = dir->i_op->mknod(dir, dentry, mode, dev);
2448 fsnotify_create(dir, dentry);
2452 static int may_mknod(mode_t mode)
2454 switch (mode & S_IFMT) {
2460 case 0: /* zero mode translates to S_IFREG */
2469 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2474 struct dentry *dentry;
2475 struct nameidata nd;
2480 error = user_path_parent(dfd, filename, &nd, &tmp);
2484 dentry = lookup_create(&nd, 0);
2485 if (IS_ERR(dentry)) {
2486 error = PTR_ERR(dentry);
2489 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2490 mode &= ~current_umask();
2491 error = may_mknod(mode);
2494 error = mnt_want_write(nd.path.mnt);
2497 error = security_path_mknod(&nd.path, dentry, mode, dev);
2499 goto out_drop_write;
2500 switch (mode & S_IFMT) {
2501 case 0: case S_IFREG:
2502 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2504 case S_IFCHR: case S_IFBLK:
2505 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2506 new_decode_dev(dev));
2508 case S_IFIFO: case S_IFSOCK:
2509 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2513 mnt_drop_write(nd.path.mnt);
2517 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2524 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2526 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2529 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2531 int error = may_create(dir, dentry);
2536 if (!dir->i_op->mkdir)
2539 mode &= (S_IRWXUGO|S_ISVTX);
2540 error = security_inode_mkdir(dir, dentry, mode);
2544 error = dir->i_op->mkdir(dir, dentry, mode);
2546 fsnotify_mkdir(dir, dentry);
2550 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2554 struct dentry *dentry;
2555 struct nameidata nd;
2557 error = user_path_parent(dfd, pathname, &nd, &tmp);
2561 dentry = lookup_create(&nd, 1);
2562 error = PTR_ERR(dentry);
2566 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2567 mode &= ~current_umask();
2568 error = mnt_want_write(nd.path.mnt);
2571 error = security_path_mkdir(&nd.path, dentry, mode);
2573 goto out_drop_write;
2574 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2576 mnt_drop_write(nd.path.mnt);
2580 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2587 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2589 return sys_mkdirat(AT_FDCWD, pathname, mode);
2593 * We try to drop the dentry early: we should have
2594 * a usage count of 2 if we're the only user of this
2595 * dentry, and if that is true (possibly after pruning
2596 * the dcache), then we drop the dentry now.
2598 * A low-level filesystem can, if it choses, legally
2601 * if (!d_unhashed(dentry))
2604 * if it cannot handle the case of removing a directory
2605 * that is still in use by something else..
2607 void dentry_unhash(struct dentry *dentry)
2610 shrink_dcache_parent(dentry);
2611 spin_lock(&dentry->d_lock);
2612 if (dentry->d_count == 2)
2614 spin_unlock(&dentry->d_lock);
2617 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2619 int error = may_delete(dir, dentry, 1);
2624 if (!dir->i_op->rmdir)
2627 mutex_lock(&dentry->d_inode->i_mutex);
2628 dentry_unhash(dentry);
2629 if (d_mountpoint(dentry))
2632 error = security_inode_rmdir(dir, dentry);
2634 error = dir->i_op->rmdir(dir, dentry);
2636 dentry->d_inode->i_flags |= S_DEAD;
2641 mutex_unlock(&dentry->d_inode->i_mutex);
2650 static long do_rmdir(int dfd, const char __user *pathname)
2654 struct dentry *dentry;
2655 struct nameidata nd;
2657 error = user_path_parent(dfd, pathname, &nd, &name);
2661 switch(nd.last_type) {
2673 nd.flags &= ~LOOKUP_PARENT;
2675 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2676 dentry = lookup_hash(&nd);
2677 error = PTR_ERR(dentry);
2680 error = mnt_want_write(nd.path.mnt);
2683 error = security_path_rmdir(&nd.path, dentry);
2686 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2688 mnt_drop_write(nd.path.mnt);
2692 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2699 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2701 return do_rmdir(AT_FDCWD, pathname);
2704 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2706 int error = may_delete(dir, dentry, 0);
2711 if (!dir->i_op->unlink)
2714 mutex_lock(&dentry->d_inode->i_mutex);
2715 if (d_mountpoint(dentry))
2718 error = security_inode_unlink(dir, dentry);
2720 error = dir->i_op->unlink(dir, dentry);
2725 mutex_unlock(&dentry->d_inode->i_mutex);
2727 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2728 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2729 fsnotify_link_count(dentry->d_inode);
2737 * Make sure that the actual truncation of the file will occur outside its
2738 * directory's i_mutex. Truncate can take a long time if there is a lot of
2739 * writeout happening, and we don't want to prevent access to the directory
2740 * while waiting on the I/O.
2742 static long do_unlinkat(int dfd, const char __user *pathname)
2746 struct dentry *dentry;
2747 struct nameidata nd;
2748 struct inode *inode = NULL;
2750 error = user_path_parent(dfd, pathname, &nd, &name);
2755 if (nd.last_type != LAST_NORM)
2758 nd.flags &= ~LOOKUP_PARENT;
2760 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2761 dentry = lookup_hash(&nd);
2762 error = PTR_ERR(dentry);
2763 if (!IS_ERR(dentry)) {
2764 /* Why not before? Because we want correct error value */
2765 if (nd.last.name[nd.last.len])
2767 inode = dentry->d_inode;
2770 error = mnt_want_write(nd.path.mnt);
2773 error = security_path_unlink(&nd.path, dentry);
2776 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2778 mnt_drop_write(nd.path.mnt);
2782 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2784 iput(inode); /* truncate the inode here */
2791 error = !dentry->d_inode ? -ENOENT :
2792 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2796 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2798 if ((flag & ~AT_REMOVEDIR) != 0)
2801 if (flag & AT_REMOVEDIR)
2802 return do_rmdir(dfd, pathname);
2804 return do_unlinkat(dfd, pathname);
2807 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2809 return do_unlinkat(AT_FDCWD, pathname);
2812 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2814 int error = may_create(dir, dentry);
2819 if (!dir->i_op->symlink)
2822 error = security_inode_symlink(dir, dentry, oldname);
2826 error = dir->i_op->symlink(dir, dentry, oldname);
2828 fsnotify_create(dir, dentry);
2832 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2833 int, newdfd, const char __user *, newname)
2838 struct dentry *dentry;
2839 struct nameidata nd;
2841 from = getname(oldname);
2843 return PTR_ERR(from);
2845 error = user_path_parent(newdfd, newname, &nd, &to);
2849 dentry = lookup_create(&nd, 0);
2850 error = PTR_ERR(dentry);
2854 error = mnt_want_write(nd.path.mnt);
2857 error = security_path_symlink(&nd.path, dentry, from);
2859 goto out_drop_write;
2860 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2862 mnt_drop_write(nd.path.mnt);
2866 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2874 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2876 return sys_symlinkat(oldname, AT_FDCWD, newname);
2879 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2881 struct inode *inode = old_dentry->d_inode;
2887 error = may_create(dir, new_dentry);
2891 if (dir->i_sb != inode->i_sb)
2895 * A link to an append-only or immutable file cannot be created.
2897 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2899 if (!dir->i_op->link)
2901 if (S_ISDIR(inode->i_mode))
2904 error = security_inode_link(old_dentry, dir, new_dentry);
2908 mutex_lock(&inode->i_mutex);
2909 /* Make sure we don't allow creating hardlink to an unlinked file */
2910 if (inode->i_nlink == 0)
2913 error = dir->i_op->link(old_dentry, dir, new_dentry);
2914 mutex_unlock(&inode->i_mutex);
2916 fsnotify_link(dir, inode, new_dentry);
2921 * Hardlinks are often used in delicate situations. We avoid
2922 * security-related surprises by not following symlinks on the
2925 * We don't follow them on the oldname either to be compatible
2926 * with linux 2.0, and to avoid hard-linking to directories
2927 * and other special files. --ADM
2929 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2930 int, newdfd, const char __user *, newname, int, flags)
2932 struct dentry *new_dentry;
2933 struct nameidata nd;
2934 struct path old_path;
2938 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2941 error = user_path_at(olddfd, oldname,
2942 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2947 error = user_path_parent(newdfd, newname, &nd, &to);
2951 if (old_path.mnt != nd.path.mnt)
2953 new_dentry = lookup_create(&nd, 0);
2954 error = PTR_ERR(new_dentry);
2955 if (IS_ERR(new_dentry))
2957 error = mnt_want_write(nd.path.mnt);
2960 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2962 goto out_drop_write;
2963 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2965 mnt_drop_write(nd.path.mnt);
2969 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2974 path_put(&old_path);
2979 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2981 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2985 * The worst of all namespace operations - renaming directory. "Perverted"
2986 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2988 * a) we can get into loop creation. Check is done in is_subdir().
2989 * b) race potential - two innocent renames can create a loop together.
2990 * That's where 4.4 screws up. Current fix: serialization on
2991 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2993 * c) we have to lock _three_ objects - parents and victim (if it exists).
2994 * And that - after we got ->i_mutex on parents (until then we don't know
2995 * whether the target exists). Solution: try to be smart with locking
2996 * order for inodes. We rely on the fact that tree topology may change
2997 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2998 * move will be locked. Thus we can rank directories by the tree
2999 * (ancestors first) and rank all non-directories after them.
3000 * That works since everybody except rename does "lock parent, lookup,
3001 * lock child" and rename is under ->s_vfs_rename_mutex.
3002 * HOWEVER, it relies on the assumption that any object with ->lookup()
3003 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3004 * we'd better make sure that there's no link(2) for them.
3005 * d) some filesystems don't support opened-but-unlinked directories,
3006 * either because of layout or because they are not ready to deal with
3007 * all cases correctly. The latter will be fixed (taking this sort of
3008 * stuff into VFS), but the former is not going away. Solution: the same
3009 * trick as in rmdir().
3010 * e) conversion from fhandle to dentry may come in the wrong moment - when
3011 * we are removing the target. Solution: we will have to grab ->i_mutex
3012 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3013 * ->i_mutex on parents, which works but leads to some truly excessive
3016 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3017 struct inode *new_dir, struct dentry *new_dentry)
3020 struct inode *target;
3023 * If we are going to change the parent - check write permissions,
3024 * we'll need to flip '..'.
3026 if (new_dir != old_dir) {
3027 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3032 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3036 target = new_dentry->d_inode;
3038 mutex_lock(&target->i_mutex);
3039 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3043 dentry_unhash(new_dentry);
3044 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3048 target->i_flags |= S_DEAD;
3049 dont_mount(new_dentry);
3051 mutex_unlock(&target->i_mutex);
3052 if (d_unhashed(new_dentry))
3053 d_rehash(new_dentry);
3057 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3058 d_move(old_dentry,new_dentry);
3062 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3063 struct inode *new_dir, struct dentry *new_dentry)
3065 struct inode *target;
3068 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))
3079 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3082 dont_mount(new_dentry);
3083 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3084 d_move(old_dentry, new_dentry);
3087 mutex_unlock(&target->i_mutex);
3092 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3093 struct inode *new_dir, struct dentry *new_dentry)
3096 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3097 const unsigned char *old_name;
3099 if (old_dentry->d_inode == new_dentry->d_inode)
3102 error = may_delete(old_dir, old_dentry, is_dir);
3106 if (!new_dentry->d_inode)
3107 error = may_create(new_dir, new_dentry);
3109 error = may_delete(new_dir, new_dentry, is_dir);
3113 if (!old_dir->i_op->rename)
3116 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3119 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3121 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3123 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3124 new_dentry->d_inode, old_dentry);
3125 fsnotify_oldname_free(old_name);
3130 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3131 int, newdfd, const char __user *, newname)
3133 struct dentry *old_dir, *new_dir;
3134 struct dentry *old_dentry, *new_dentry;
3135 struct dentry *trap;
3136 struct nameidata oldnd, newnd;
3141 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3145 error = user_path_parent(newdfd, newname, &newnd, &to);
3150 if (oldnd.path.mnt != newnd.path.mnt)
3153 old_dir = oldnd.path.dentry;
3155 if (oldnd.last_type != LAST_NORM)
3158 new_dir = newnd.path.dentry;
3159 if (newnd.last_type != LAST_NORM)
3162 oldnd.flags &= ~LOOKUP_PARENT;
3163 newnd.flags &= ~LOOKUP_PARENT;
3164 newnd.flags |= LOOKUP_RENAME_TARGET;
3166 trap = lock_rename(new_dir, old_dir);
3168 old_dentry = lookup_hash(&oldnd);
3169 error = PTR_ERR(old_dentry);
3170 if (IS_ERR(old_dentry))
3172 /* source must exist */
3174 if (!old_dentry->d_inode)
3176 /* unless the source is a directory trailing slashes give -ENOTDIR */
3177 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3179 if (oldnd.last.name[oldnd.last.len])
3181 if (newnd.last.name[newnd.last.len])
3184 /* source should not be ancestor of target */
3186 if (old_dentry == trap)
3188 new_dentry = lookup_hash(&newnd);
3189 error = PTR_ERR(new_dentry);
3190 if (IS_ERR(new_dentry))
3192 /* target should not be an ancestor of source */
3194 if (new_dentry == trap)
3197 error = mnt_want_write(oldnd.path.mnt);
3200 error = security_path_rename(&oldnd.path, old_dentry,
3201 &newnd.path, new_dentry);
3204 error = vfs_rename(old_dir->d_inode, old_dentry,
3205 new_dir->d_inode, new_dentry);
3207 mnt_drop_write(oldnd.path.mnt);
3213 unlock_rename(new_dir, old_dir);
3215 path_put(&newnd.path);
3218 path_put(&oldnd.path);
3224 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3226 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3229 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3233 len = PTR_ERR(link);
3238 if (len > (unsigned) buflen)
3240 if (copy_to_user(buffer, link, len))
3247 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3248 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3249 * using) it for any given inode is up to filesystem.
3251 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3253 struct nameidata nd;
3258 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3260 return PTR_ERR(cookie);
3262 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3263 if (dentry->d_inode->i_op->put_link)
3264 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3268 int vfs_follow_link(struct nameidata *nd, const char *link)
3270 return __vfs_follow_link(nd, link);
3273 /* get the link contents into pagecache */
3274 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3278 struct address_space *mapping = dentry->d_inode->i_mapping;
3279 page = read_mapping_page(mapping, 0, NULL);
3284 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3288 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3290 struct page *page = NULL;
3291 char *s = page_getlink(dentry, &page);
3292 int res = vfs_readlink(dentry,buffer,buflen,s);
3295 page_cache_release(page);
3300 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3302 struct page *page = NULL;
3303 nd_set_link(nd, page_getlink(dentry, &page));
3307 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3309 struct page *page = cookie;
3313 page_cache_release(page);
3318 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3320 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3322 struct address_space *mapping = inode->i_mapping;
3327 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3329 flags |= AOP_FLAG_NOFS;
3332 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3333 flags, &page, &fsdata);
3337 kaddr = kmap_atomic(page, KM_USER0);
3338 memcpy(kaddr, symname, len-1);
3339 kunmap_atomic(kaddr, KM_USER0);
3341 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3348 mark_inode_dirty(inode);
3354 int page_symlink(struct inode *inode, const char *symname, int len)
3356 return __page_symlink(inode, symname, len,
3357 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3360 const struct inode_operations page_symlink_inode_operations = {
3361 .readlink = generic_readlink,
3362 .follow_link = page_follow_link_light,
3363 .put_link = page_put_link,
3366 EXPORT_SYMBOL(user_path_at);
3367 EXPORT_SYMBOL(follow_down_one);
3368 EXPORT_SYMBOL(follow_down);
3369 EXPORT_SYMBOL(follow_up);
3370 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3371 EXPORT_SYMBOL(getname);
3372 EXPORT_SYMBOL(lock_rename);
3373 EXPORT_SYMBOL(lookup_one_len);
3374 EXPORT_SYMBOL(page_follow_link_light);
3375 EXPORT_SYMBOL(page_put_link);
3376 EXPORT_SYMBOL(page_readlink);
3377 EXPORT_SYMBOL(__page_symlink);
3378 EXPORT_SYMBOL(page_symlink);
3379 EXPORT_SYMBOL(page_symlink_inode_operations);
3380 EXPORT_SYMBOL(kern_path_parent);
3381 EXPORT_SYMBOL(kern_path);
3382 EXPORT_SYMBOL(vfs_path_lookup);
3383 EXPORT_SYMBOL(inode_permission);
3384 EXPORT_SYMBOL(file_permission);
3385 EXPORT_SYMBOL(unlock_rename);
3386 EXPORT_SYMBOL(vfs_create);
3387 EXPORT_SYMBOL(vfs_follow_link);
3388 EXPORT_SYMBOL(vfs_link);
3389 EXPORT_SYMBOL(vfs_mkdir);
3390 EXPORT_SYMBOL(vfs_mknod);
3391 EXPORT_SYMBOL(generic_permission);
3392 EXPORT_SYMBOL(vfs_readlink);
3393 EXPORT_SYMBOL(vfs_rename);
3394 EXPORT_SYMBOL(vfs_rmdir);
3395 EXPORT_SYMBOL(vfs_symlink);
3396 EXPORT_SYMBOL(vfs_unlink);
3397 EXPORT_SYMBOL(dentry_unhash);
3398 EXPORT_SYMBOL(generic_readlink);