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 char * getname(const char __user * filename)
143 result = ERR_PTR(-ENOMEM);
146 int retval = do_getname(filename, tmp);
151 result = ERR_PTR(retval);
154 audit_getname(result);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname);
170 * This does basic POSIX ACL permission checking
172 static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
173 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
175 umode_t mode = inode->i_mode;
177 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
179 if (current_fsuid() == inode->i_uid)
182 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
183 int error = check_acl(inode, mask, flags);
184 if (error != -EAGAIN)
188 if (in_group_p(inode->i_gid))
193 * If the DACs are ok we don't need any capability check.
195 if ((mask & ~mode) == 0)
201 * generic_permission - check for access rights on a Posix-like filesystem
202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
205 * @flags: IPERM_FLAG_ flags.
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
210 * are used for other things.
212 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
213 * request cannot be satisfied (eg. requires blocking or too much complexity).
214 * It would then be called again in ref-walk mode.
216 int generic_permission(struct inode *inode, int mask, unsigned int flags,
217 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
222 * Do the basic POSIX ACL permission checks.
224 ret = acl_permission_check(inode, mask, flags, check_acl);
229 * Read/write DACs are always overridable.
230 * Executable DACs are overridable if at least one exec bit is set.
232 if (!(mask & MAY_EXEC) || execute_ok(inode))
233 if (capable(CAP_DAC_OVERRIDE))
237 * Searching includes executable on directories, else just read.
239 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
240 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
241 if (capable(CAP_DAC_READ_SEARCH))
248 * inode_permission - check for access rights to a given inode
249 * @inode: inode to check permission on
250 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
252 * Used to check for read/write/execute permissions on an inode.
253 * We use "fsuid" for this, letting us set arbitrary permissions
254 * for filesystem access without changing the "normal" uids which
255 * are used for other things.
257 int inode_permission(struct inode *inode, int mask)
261 if (mask & MAY_WRITE) {
262 umode_t mode = inode->i_mode;
265 * Nobody gets write access to a read-only fs.
267 if (IS_RDONLY(inode) &&
268 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
272 * Nobody gets write access to an immutable file.
274 if (IS_IMMUTABLE(inode))
278 if (inode->i_op->permission)
279 retval = inode->i_op->permission(inode, mask, 0);
281 retval = generic_permission(inode, mask, 0,
282 inode->i_op->check_acl);
287 retval = devcgroup_inode_permission(inode, mask);
291 return security_inode_permission(inode, mask);
295 * file_permission - check for additional access rights to a given file
296 * @file: file to check access rights for
297 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
299 * Used to check for read/write/execute permissions on an already opened
303 * Do not use this function in new code. All access checks should
304 * be done using inode_permission().
306 int file_permission(struct file *file, int mask)
308 return inode_permission(file->f_path.dentry->d_inode, mask);
312 * get_write_access() gets write permission for a file.
313 * put_write_access() releases this write permission.
314 * This is used for regular files.
315 * We cannot support write (and maybe mmap read-write shared) accesses and
316 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
317 * can have the following values:
318 * 0: no writers, no VM_DENYWRITE mappings
319 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
320 * > 0: (i_writecount) users are writing to the file.
322 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
323 * except for the cases where we don't hold i_writecount yet. Then we need to
324 * use {get,deny}_write_access() - these functions check the sign and refuse
325 * to do the change if sign is wrong. Exclusion between them is provided by
326 * the inode->i_lock spinlock.
329 int get_write_access(struct inode * inode)
331 spin_lock(&inode->i_lock);
332 if (atomic_read(&inode->i_writecount) < 0) {
333 spin_unlock(&inode->i_lock);
336 atomic_inc(&inode->i_writecount);
337 spin_unlock(&inode->i_lock);
342 int deny_write_access(struct file * file)
344 struct inode *inode = file->f_path.dentry->d_inode;
346 spin_lock(&inode->i_lock);
347 if (atomic_read(&inode->i_writecount) > 0) {
348 spin_unlock(&inode->i_lock);
351 atomic_dec(&inode->i_writecount);
352 spin_unlock(&inode->i_lock);
358 * path_get - get a reference to a path
359 * @path: path to get the reference to
361 * Given a path increment the reference count to the dentry and the vfsmount.
363 void path_get(struct path *path)
368 EXPORT_SYMBOL(path_get);
371 * path_put - put a reference to a path
372 * @path: path to put the reference to
374 * Given a path decrement the reference count to the dentry and the vfsmount.
376 void path_put(struct path *path)
381 EXPORT_SYMBOL(path_put);
384 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
385 * @nd: nameidata pathwalk data to drop
386 * Returns: 0 on success, -ECHILD on failure
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
390 * to drop out of rcu-walk mode and take normal reference counts on dentries
391 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
392 * refcounts at the last known good point before rcu-walk got stuck, so
393 * ref-walk may continue from there. If this is not successful (eg. a seqcount
394 * has changed), then failure is returned and path walk restarts from the
395 * beginning in ref-walk mode.
397 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
398 * ref-walk. Must be called from rcu-walk context.
400 static int nameidata_drop_rcu(struct nameidata *nd)
402 struct fs_struct *fs = current->fs;
403 struct dentry *dentry = nd->path.dentry;
405 BUG_ON(!(nd->flags & LOOKUP_RCU));
407 spin_lock(&fs->lock);
408 if (nd->root.mnt != fs->root.mnt ||
409 nd->root.dentry != fs->root.dentry)
412 spin_lock(&dentry->d_lock);
413 if (!__d_rcu_to_refcount(dentry, nd->seq))
415 BUG_ON(nd->inode != dentry->d_inode);
416 spin_unlock(&dentry->d_lock);
419 spin_unlock(&fs->lock);
421 mntget(nd->path.mnt);
424 br_read_unlock(vfsmount_lock);
425 nd->flags &= ~LOOKUP_RCU;
428 spin_unlock(&dentry->d_lock);
431 spin_unlock(&fs->lock);
435 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
436 static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
438 if (nd->flags & LOOKUP_RCU)
439 return nameidata_drop_rcu(nd);
444 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
445 * @nd: nameidata pathwalk data to drop
446 * @dentry: dentry to drop
447 * Returns: 0 on success, -ECHILD on failure
449 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
450 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
451 * @nd. Must be called from rcu-walk context.
453 static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
455 struct fs_struct *fs = current->fs;
456 struct dentry *parent = nd->path.dentry;
458 BUG_ON(!(nd->flags & LOOKUP_RCU));
460 spin_lock(&fs->lock);
461 if (nd->root.mnt != fs->root.mnt ||
462 nd->root.dentry != fs->root.dentry)
465 spin_lock(&parent->d_lock);
466 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
467 if (!__d_rcu_to_refcount(dentry, nd->seq))
470 * If the sequence check on the child dentry passed, then the child has
471 * not been removed from its parent. This means the parent dentry must
472 * be valid and able to take a reference at this point.
474 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
475 BUG_ON(!parent->d_count);
477 spin_unlock(&dentry->d_lock);
478 spin_unlock(&parent->d_lock);
481 spin_unlock(&fs->lock);
483 mntget(nd->path.mnt);
486 br_read_unlock(vfsmount_lock);
487 nd->flags &= ~LOOKUP_RCU;
490 spin_unlock(&dentry->d_lock);
491 spin_unlock(&parent->d_lock);
494 spin_unlock(&fs->lock);
498 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
499 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
501 if (nd->flags & LOOKUP_RCU)
502 return nameidata_dentry_drop_rcu(nd, dentry);
507 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
508 * @nd: nameidata pathwalk data to drop
509 * Returns: 0 on success, -ECHILD on failure
511 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
512 * nd->path should be the final element of the lookup, so nd->root is discarded.
513 * Must be called from rcu-walk context.
515 static int nameidata_drop_rcu_last(struct nameidata *nd)
517 struct dentry *dentry = nd->path.dentry;
519 BUG_ON(!(nd->flags & LOOKUP_RCU));
520 nd->flags &= ~LOOKUP_RCU;
522 spin_lock(&dentry->d_lock);
523 if (!__d_rcu_to_refcount(dentry, nd->seq))
525 BUG_ON(nd->inode != dentry->d_inode);
526 spin_unlock(&dentry->d_lock);
528 mntget(nd->path.mnt);
531 br_read_unlock(vfsmount_lock);
536 spin_unlock(&dentry->d_lock);
538 br_read_unlock(vfsmount_lock);
543 * release_open_intent - free up open intent resources
544 * @nd: pointer to nameidata
546 void release_open_intent(struct nameidata *nd)
548 struct file *file = nd->intent.open.file;
550 if (file && !IS_ERR(file)) {
551 if (file->f_path.dentry == NULL)
558 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
560 return dentry->d_op->d_revalidate(dentry, nd);
563 static struct dentry *
564 do_revalidate(struct dentry *dentry, struct nameidata *nd)
566 int status = d_revalidate(dentry, nd);
567 if (unlikely(status <= 0)) {
569 * The dentry failed validation.
570 * If d_revalidate returned 0 attempt to invalidate
571 * the dentry otherwise d_revalidate is asking us
572 * to return a fail status.
576 dentry = ERR_PTR(status);
577 } else if (!d_invalidate(dentry)) {
585 static inline struct dentry *
586 do_revalidate_rcu(struct dentry *dentry, struct nameidata *nd)
588 int status = d_revalidate(dentry, nd);
589 if (likely(status > 0))
591 if (status == -ECHILD) {
592 if (nameidata_dentry_drop_rcu(nd, dentry))
593 return ERR_PTR(-ECHILD);
594 return do_revalidate(dentry, nd);
597 return ERR_PTR(status);
598 /* Don't d_invalidate in rcu-walk mode */
599 if (nameidata_dentry_drop_rcu(nd, dentry))
600 return ERR_PTR(-ECHILD);
601 if (!d_invalidate(dentry)) {
609 * handle_reval_path - force revalidation of a dentry
611 * In some situations the path walking code will trust dentries without
612 * revalidating them. This causes problems for filesystems that depend on
613 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
614 * (which indicates that it's possible for the dentry to go stale), force
615 * a d_revalidate call before proceeding.
617 * Returns 0 if the revalidation was successful. If the revalidation fails,
618 * either return the error returned by d_revalidate or -ESTALE if the
619 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
620 * invalidate the dentry. It's up to the caller to handle putting references
621 * to the path if necessary.
623 static inline int handle_reval_path(struct nameidata *nd)
625 struct dentry *dentry = nd->path.dentry;
628 if (likely(!(nd->flags & LOOKUP_JUMPED)))
631 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
634 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
637 /* Note: we do not d_invalidate() */
638 status = d_revalidate(dentry, nd);
649 * Short-cut version of permission(), for calling on directories
650 * during pathname resolution. Combines parts of permission()
651 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
653 * If appropriate, check DAC only. If not appropriate, or
654 * short-cut DAC fails, then call ->permission() to do more
655 * complete permission check.
657 static inline int exec_permission(struct inode *inode, unsigned int flags)
661 if (inode->i_op->permission) {
662 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
664 ret = acl_permission_check(inode, MAY_EXEC, flags,
665 inode->i_op->check_acl);
672 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
677 return security_inode_exec_permission(inode, flags);
680 static __always_inline void set_root(struct nameidata *nd)
683 get_fs_root(current->fs, &nd->root);
686 static int link_path_walk(const char *, struct nameidata *);
688 static __always_inline void set_root_rcu(struct nameidata *nd)
691 struct fs_struct *fs = current->fs;
695 seq = read_seqcount_begin(&fs->seq);
697 } while (read_seqcount_retry(&fs->seq, seq));
701 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
713 nd->flags |= LOOKUP_JUMPED;
715 nd->inode = nd->path.dentry->d_inode;
717 ret = link_path_walk(link, nd);
721 return PTR_ERR(link);
724 static void path_put_conditional(struct path *path, struct nameidata *nd)
727 if (path->mnt != nd->path.mnt)
731 static inline void path_to_nameidata(const struct path *path,
732 struct nameidata *nd)
734 if (!(nd->flags & LOOKUP_RCU)) {
735 dput(nd->path.dentry);
736 if (nd->path.mnt != path->mnt)
737 mntput(nd->path.mnt);
739 nd->path.mnt = path->mnt;
740 nd->path.dentry = path->dentry;
743 static __always_inline int
744 __do_follow_link(const struct path *link, struct nameidata *nd, void **p)
747 struct dentry *dentry = link->dentry;
749 BUG_ON(nd->flags & LOOKUP_RCU);
751 touch_atime(link->mnt, dentry);
752 nd_set_link(nd, NULL);
754 if (link->mnt == nd->path.mnt)
757 error = security_inode_follow_link(link->dentry, nd);
759 *p = ERR_PTR(error); /* no ->put_link(), please */
764 nd->last_type = LAST_BIND;
765 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
768 char *s = nd_get_link(nd);
771 error = __vfs_follow_link(nd, s);
772 else if (nd->last_type == LAST_BIND)
773 nd->flags |= LOOKUP_JUMPED;
779 * This limits recursive symlink follows to 8, while
780 * limiting consecutive symlinks to 40.
782 * Without that kind of total limit, nasty chains of consecutive
783 * symlinks can cause almost arbitrarily long lookups.
785 static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd)
790 /* We drop rcu-walk here */
791 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
793 BUG_ON(inode != path->dentry->d_inode);
795 if (current->link_count >= MAX_NESTED_LINKS)
797 if (current->total_link_count >= 40)
799 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
801 current->link_count++;
802 current->total_link_count++;
804 err = __do_follow_link(path, nd, &cookie);
805 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
806 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
808 current->link_count--;
812 path_put_conditional(path, nd);
817 static int follow_up_rcu(struct path *path)
819 struct vfsmount *parent;
820 struct dentry *mountpoint;
822 parent = path->mnt->mnt_parent;
823 if (parent == path->mnt)
825 mountpoint = path->mnt->mnt_mountpoint;
826 path->dentry = mountpoint;
831 int follow_up(struct path *path)
833 struct vfsmount *parent;
834 struct dentry *mountpoint;
836 br_read_lock(vfsmount_lock);
837 parent = path->mnt->mnt_parent;
838 if (parent == path->mnt) {
839 br_read_unlock(vfsmount_lock);
843 mountpoint = dget(path->mnt->mnt_mountpoint);
844 br_read_unlock(vfsmount_lock);
846 path->dentry = mountpoint;
853 * Perform an automount
854 * - return -EISDIR to tell follow_managed() to stop and return the path we
857 static int follow_automount(struct path *path, unsigned flags,
860 struct vfsmount *mnt;
863 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
866 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
867 * and this is the terminal part of the path.
869 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
870 return -EISDIR; /* we actually want to stop here */
872 /* We want to mount if someone is trying to open/create a file of any
873 * type under the mountpoint, wants to traverse through the mountpoint
874 * or wants to open the mounted directory.
876 * We don't want to mount if someone's just doing a stat and they've
877 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
878 * appended a '/' to the name.
880 if (!(flags & LOOKUP_FOLLOW) &&
881 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
882 LOOKUP_OPEN | LOOKUP_CREATE)))
885 current->total_link_count++;
886 if (current->total_link_count >= 40)
889 mnt = path->dentry->d_op->d_automount(path);
892 * The filesystem is allowed to return -EISDIR here to indicate
893 * it doesn't want to automount. For instance, autofs would do
894 * this so that its userspace daemon can mount on this dentry.
896 * However, we can only permit this if it's a terminal point in
897 * the path being looked up; if it wasn't then the remainder of
898 * the path is inaccessible and we should say so.
900 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
905 if (!mnt) /* mount collision */
908 err = finish_automount(mnt, path);
912 /* Someone else made a mount here whilst we were busy */
919 path->dentry = dget(mnt->mnt_root);
929 * Handle a dentry that is managed in some way.
930 * - Flagged for transit management (autofs)
931 * - Flagged as mountpoint
932 * - Flagged as automount point
934 * This may only be called in refwalk mode.
936 * Serialization is taken care of in namespace.c
938 static int follow_managed(struct path *path, unsigned flags)
941 bool need_mntput = false;
944 /* Given that we're not holding a lock here, we retain the value in a
945 * local variable for each dentry as we look at it so that we don't see
946 * the components of that value change under us */
947 while (managed = ACCESS_ONCE(path->dentry->d_flags),
948 managed &= DCACHE_MANAGED_DENTRY,
949 unlikely(managed != 0)) {
950 /* Allow the filesystem to manage the transit without i_mutex
952 if (managed & DCACHE_MANAGE_TRANSIT) {
953 BUG_ON(!path->dentry->d_op);
954 BUG_ON(!path->dentry->d_op->d_manage);
955 ret = path->dentry->d_op->d_manage(path->dentry,
958 return ret == -EISDIR ? 0 : ret;
961 /* Transit to a mounted filesystem. */
962 if (managed & DCACHE_MOUNTED) {
963 struct vfsmount *mounted = lookup_mnt(path);
969 path->dentry = dget(mounted->mnt_root);
974 /* Something is mounted on this dentry in another
975 * namespace and/or whatever was mounted there in this
976 * namespace got unmounted before we managed to get the
980 /* Handle an automount point */
981 if (managed & DCACHE_NEED_AUTOMOUNT) {
982 ret = follow_automount(path, flags, &need_mntput);
984 return ret == -EISDIR ? 0 : ret;
988 /* We didn't change the current path point */
994 int follow_down_one(struct path *path)
996 struct vfsmount *mounted;
998 mounted = lookup_mnt(path);
1002 path->mnt = mounted;
1003 path->dentry = dget(mounted->mnt_root);
1010 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
1011 * meet a managed dentry and we're not walking to "..". True is returned to
1012 * continue, false to abort.
1014 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1015 struct inode **inode, bool reverse_transit)
1017 while (d_mountpoint(path->dentry)) {
1018 struct vfsmount *mounted;
1019 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1021 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1023 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1026 path->mnt = mounted;
1027 path->dentry = mounted->mnt_root;
1028 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1029 *inode = path->dentry->d_inode;
1032 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1033 return reverse_transit;
1037 static int follow_dotdot_rcu(struct nameidata *nd)
1039 struct inode *inode = nd->inode;
1044 if (nd->path.dentry == nd->root.dentry &&
1045 nd->path.mnt == nd->root.mnt) {
1048 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1049 struct dentry *old = nd->path.dentry;
1050 struct dentry *parent = old->d_parent;
1053 seq = read_seqcount_begin(&parent->d_seq);
1054 if (read_seqcount_retry(&old->d_seq, nd->seq))
1056 inode = parent->d_inode;
1057 nd->path.dentry = parent;
1061 if (!follow_up_rcu(&nd->path))
1063 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1064 inode = nd->path.dentry->d_inode;
1066 __follow_mount_rcu(nd, &nd->path, &inode, true);
1071 nd->flags &= ~LOOKUP_RCU;
1072 nd->root.mnt = NULL;
1074 br_read_unlock(vfsmount_lock);
1079 * Follow down to the covering mount currently visible to userspace. At each
1080 * point, the filesystem owning that dentry may be queried as to whether the
1081 * caller is permitted to proceed or not.
1083 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1086 int follow_down(struct path *path, bool mounting_here)
1091 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1092 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1093 /* Allow the filesystem to manage the transit without i_mutex
1096 * We indicate to the filesystem if someone is trying to mount
1097 * something here. This gives autofs the chance to deny anyone
1098 * other than its daemon the right to mount on its
1101 * The filesystem may sleep at this point.
1103 if (managed & DCACHE_MANAGE_TRANSIT) {
1104 BUG_ON(!path->dentry->d_op);
1105 BUG_ON(!path->dentry->d_op->d_manage);
1106 ret = path->dentry->d_op->d_manage(
1107 path->dentry, mounting_here, false);
1109 return ret == -EISDIR ? 0 : ret;
1112 /* Transit to a mounted filesystem. */
1113 if (managed & DCACHE_MOUNTED) {
1114 struct vfsmount *mounted = lookup_mnt(path);
1119 path->mnt = mounted;
1120 path->dentry = dget(mounted->mnt_root);
1124 /* Don't handle automount points here */
1131 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1133 static void follow_mount(struct path *path)
1135 while (d_mountpoint(path->dentry)) {
1136 struct vfsmount *mounted = lookup_mnt(path);
1141 path->mnt = mounted;
1142 path->dentry = dget(mounted->mnt_root);
1146 static void follow_dotdot(struct nameidata *nd)
1151 struct dentry *old = nd->path.dentry;
1153 if (nd->path.dentry == nd->root.dentry &&
1154 nd->path.mnt == nd->root.mnt) {
1157 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1158 /* rare case of legitimate dget_parent()... */
1159 nd->path.dentry = dget_parent(nd->path.dentry);
1163 if (!follow_up(&nd->path))
1166 follow_mount(&nd->path);
1167 nd->inode = nd->path.dentry->d_inode;
1171 * Allocate a dentry with name and parent, and perform a parent
1172 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1173 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1174 * have verified that no child exists while under i_mutex.
1176 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1177 struct qstr *name, struct nameidata *nd)
1179 struct inode *inode = parent->d_inode;
1180 struct dentry *dentry;
1183 /* Don't create child dentry for a dead directory. */
1184 if (unlikely(IS_DEADDIR(inode)))
1185 return ERR_PTR(-ENOENT);
1187 dentry = d_alloc(parent, name);
1188 if (unlikely(!dentry))
1189 return ERR_PTR(-ENOMEM);
1191 old = inode->i_op->lookup(inode, dentry, nd);
1192 if (unlikely(old)) {
1200 * It's more convoluted than I'd like it to be, but... it's still fairly
1201 * small and for now I'd prefer to have fast path as straight as possible.
1202 * It _is_ time-critical.
1204 static int do_lookup(struct nameidata *nd, struct qstr *name,
1205 struct path *path, struct inode **inode)
1207 struct vfsmount *mnt = nd->path.mnt;
1208 struct dentry *dentry, *parent = nd->path.dentry;
1213 * See if the low-level filesystem might want
1214 * to use its own hash..
1216 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1217 err = parent->d_op->d_hash(parent, nd->inode, name);
1223 * Rename seqlock is not required here because in the off chance
1224 * of a false negative due to a concurrent rename, we're going to
1225 * do the non-racy lookup, below.
1227 if (nd->flags & LOOKUP_RCU) {
1231 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1233 if (nameidata_drop_rcu(nd))
1237 /* Memory barrier in read_seqcount_begin of child is enough */
1238 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1242 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1243 dentry = do_revalidate_rcu(dentry, nd);
1248 if (!(nd->flags & LOOKUP_RCU))
1252 path->dentry = dentry;
1253 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1255 if (nameidata_drop_rcu(nd))
1259 dentry = __d_lookup(parent, name);
1263 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1264 dentry = do_revalidate(dentry, nd);
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 dir = parent->d_inode;
1283 BUG_ON(nd->inode != dir);
1285 mutex_lock(&dir->i_mutex);
1287 * First re-do the cached lookup just in case it was created
1288 * while we waited for the directory semaphore, or the first
1289 * lookup failed due to an unrelated rename.
1291 * This could use version numbering or similar to avoid unnecessary
1292 * cache lookups, but then we'd have to do the first lookup in the
1293 * non-racy way. However in the common case here, everything should
1294 * be hot in cache, so would it be a big win?
1296 dentry = d_lookup(parent, name);
1297 if (likely(!dentry)) {
1298 dentry = d_alloc_and_lookup(parent, name, nd);
1299 mutex_unlock(&dir->i_mutex);
1305 * Uhhuh! Nasty case: the cache was re-populated while
1306 * we waited on the semaphore. Need to revalidate.
1308 mutex_unlock(&dir->i_mutex);
1312 return PTR_ERR(dentry);
1315 static inline int may_lookup(struct nameidata *nd)
1317 if (nd->flags & LOOKUP_RCU) {
1318 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1321 if (nameidata_drop_rcu(nd))
1324 return exec_permission(nd->inode, 0);
1327 static inline int handle_dots(struct nameidata *nd, int type)
1329 if (type == LAST_DOTDOT) {
1330 if (nd->flags & LOOKUP_RCU) {
1331 if (follow_dotdot_rcu(nd))
1341 * This is the basic name resolution function, turning a pathname into
1342 * the final dentry. We expect 'base' to be positive and a directory.
1344 * Returns 0 and nd will have valid dentry and mnt on success.
1345 * Returns error and drops reference to input namei data on failure.
1347 static int link_path_walk(const char *name, struct nameidata *nd)
1351 unsigned int lookup_flags = nd->flags;
1359 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1361 /* At this point we know we have a real path component. */
1363 struct inode *inode;
1369 nd->flags |= LOOKUP_CONTINUE;
1371 err = may_lookup(nd);
1376 c = *(const unsigned char *)name;
1378 hash = init_name_hash();
1381 hash = partial_name_hash(c, hash);
1382 c = *(const unsigned char *)name;
1383 } while (c && (c != '/'));
1384 this.len = name - (const char *) this.name;
1385 this.hash = end_name_hash(hash);
1388 if (this.name[0] == '.') switch (this.len) {
1390 if (this.name[1] == '.') {
1392 nd->flags |= LOOKUP_JUMPED;
1398 if (likely(type == LAST_NORM))
1399 nd->flags &= ~LOOKUP_JUMPED;
1401 /* remove trailing slashes? */
1403 goto last_component;
1404 while (*++name == '/');
1406 goto last_with_slashes;
1409 * "." and ".." are special - ".." especially so because it has
1410 * to be able to know about the current root directory and
1411 * parent relationships.
1413 if (unlikely(type != LAST_NORM)) {
1414 if (handle_dots(nd, type))
1419 /* This does the actual lookups.. */
1420 err = do_lookup(nd, &this, &next, &inode);
1424 if (inode && inode->i_op->follow_link) {
1425 err = do_follow_link(inode, &next, nd);
1428 nd->inode = nd->path.dentry->d_inode;
1430 path_to_nameidata(&next, nd);
1437 if (!nd->inode->i_op->lookup)
1440 /* here ends the main loop */
1443 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1445 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1446 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1447 if (lookup_flags & LOOKUP_PARENT)
1449 if (unlikely(type != LAST_NORM))
1450 return handle_dots(nd, type);
1451 err = do_lookup(nd, &this, &next, &inode);
1454 if (inode && unlikely(inode->i_op->follow_link) &&
1455 (lookup_flags & LOOKUP_FOLLOW)) {
1456 err = do_follow_link(inode, &next, nd);
1459 nd->inode = nd->path.dentry->d_inode;
1461 path_to_nameidata(&next, nd);
1467 if (lookup_flags & LOOKUP_DIRECTORY) {
1469 if (!nd->inode->i_op->lookup)
1475 nd->last_type = type;
1478 if (!(nd->flags & LOOKUP_RCU))
1479 path_put(&nd->path);
1481 if (nd->flags & LOOKUP_RCU) {
1482 nd->flags &= ~LOOKUP_RCU;
1483 nd->root.mnt = NULL;
1485 br_read_unlock(vfsmount_lock);
1490 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1496 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1497 nd->flags = flags | LOOKUP_JUMPED;
1499 nd->root.mnt = NULL;
1503 if (flags & LOOKUP_RCU) {
1504 br_read_lock(vfsmount_lock);
1509 path_get(&nd->root);
1511 nd->path = nd->root;
1512 } else if (dfd == AT_FDCWD) {
1513 if (flags & LOOKUP_RCU) {
1514 struct fs_struct *fs = current->fs;
1517 br_read_lock(vfsmount_lock);
1521 seq = read_seqcount_begin(&fs->seq);
1523 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1524 } while (read_seqcount_retry(&fs->seq, seq));
1526 get_fs_pwd(current->fs, &nd->path);
1529 struct dentry *dentry;
1531 file = fget_light(dfd, &fput_needed);
1536 dentry = file->f_path.dentry;
1539 if (!S_ISDIR(dentry->d_inode->i_mode))
1542 retval = file_permission(file, MAY_EXEC);
1546 nd->path = file->f_path;
1547 if (flags & LOOKUP_RCU) {
1550 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1551 br_read_lock(vfsmount_lock);
1554 path_get(&file->f_path);
1555 fput_light(file, fput_needed);
1559 nd->inode = nd->path.dentry->d_inode;
1563 fput_light(file, fput_needed);
1568 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1569 static int path_lookupat(int dfd, const char *name,
1570 unsigned int flags, struct nameidata *nd)
1575 * Path walking is largely split up into 2 different synchronisation
1576 * schemes, rcu-walk and ref-walk (explained in
1577 * Documentation/filesystems/path-lookup.txt). These share much of the
1578 * path walk code, but some things particularly setup, cleanup, and
1579 * following mounts are sufficiently divergent that functions are
1580 * duplicated. Typically there is a function foo(), and its RCU
1581 * analogue, foo_rcu().
1583 * -ECHILD is the error number of choice (just to avoid clashes) that
1584 * is returned if some aspect of an rcu-walk fails. Such an error must
1585 * be handled by restarting a traditional ref-walk (which will always
1586 * be able to complete).
1588 retval = path_init(dfd, name, flags, nd);
1590 if (unlikely(retval))
1593 current->total_link_count = 0;
1594 retval = link_path_walk(name, nd);
1596 if (nd->flags & LOOKUP_RCU) {
1597 /* went all way through without dropping RCU */
1599 if (nameidata_drop_rcu_last(nd))
1604 retval = handle_reval_path(nd);
1612 path_put(&nd->root);
1613 nd->root.mnt = NULL;
1618 static int do_path_lookup(int dfd, const char *name,
1619 unsigned int flags, struct nameidata *nd)
1621 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1622 if (unlikely(retval == -ECHILD))
1623 retval = path_lookupat(dfd, name, flags, nd);
1624 if (unlikely(retval == -ESTALE))
1625 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1627 if (likely(!retval)) {
1628 if (unlikely(!audit_dummy_context())) {
1629 if (nd->path.dentry && nd->inode)
1630 audit_inode(name, nd->path.dentry);
1636 int kern_path_parent(const char *name, struct nameidata *nd)
1638 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1641 int kern_path(const char *name, unsigned int flags, struct path *path)
1643 struct nameidata nd;
1644 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1651 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1652 * @dentry: pointer to dentry of the base directory
1653 * @mnt: pointer to vfs mount of the base directory
1654 * @name: pointer to file name
1655 * @flags: lookup flags
1656 * @nd: pointer to nameidata
1658 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1659 const char *name, unsigned int flags,
1660 struct nameidata *nd)
1664 /* same as do_path_lookup */
1665 nd->last_type = LAST_ROOT;
1666 nd->flags = flags | LOOKUP_JUMPED;
1669 nd->path.dentry = dentry;
1671 path_get(&nd->path);
1672 nd->root = nd->path;
1673 path_get(&nd->root);
1674 nd->inode = nd->path.dentry->d_inode;
1676 current->total_link_count = 0;
1678 result = link_path_walk(name, nd);
1680 result = handle_reval_path(nd);
1681 if (result == -ESTALE) {
1682 /* nd->path had been dropped */
1683 current->total_link_count = 0;
1684 nd->path.dentry = dentry;
1686 nd->inode = dentry->d_inode;
1687 path_get(&nd->path);
1688 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_REVAL;
1690 result = link_path_walk(name, nd);
1692 result = handle_reval_path(nd);
1694 if (unlikely(!result && !audit_dummy_context() && nd->path.dentry &&
1696 audit_inode(name, nd->path.dentry);
1698 path_put(&nd->root);
1699 nd->root.mnt = NULL;
1704 static struct dentry *__lookup_hash(struct qstr *name,
1705 struct dentry *base, struct nameidata *nd)
1707 struct inode *inode = base->d_inode;
1708 struct dentry *dentry;
1711 err = exec_permission(inode, 0);
1713 return ERR_PTR(err);
1716 * See if the low-level filesystem might want
1717 * to use its own hash..
1719 if (base->d_flags & DCACHE_OP_HASH) {
1720 err = base->d_op->d_hash(base, inode, name);
1721 dentry = ERR_PTR(err);
1727 * Don't bother with __d_lookup: callers are for creat as
1728 * well as unlink, so a lot of the time it would cost
1731 dentry = d_lookup(base, name);
1733 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
1734 dentry = do_revalidate(dentry, nd);
1737 dentry = d_alloc_and_lookup(base, name, nd);
1743 * Restricted form of lookup. Doesn't follow links, single-component only,
1744 * needs parent already locked. Doesn't follow mounts.
1747 static struct dentry *lookup_hash(struct nameidata *nd)
1749 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1752 static int __lookup_one_len(const char *name, struct qstr *this,
1753 struct dentry *base, int len)
1763 hash = init_name_hash();
1765 c = *(const unsigned char *)name++;
1766 if (c == '/' || c == '\0')
1768 hash = partial_name_hash(c, hash);
1770 this->hash = end_name_hash(hash);
1775 * lookup_one_len - filesystem helper to lookup single pathname component
1776 * @name: pathname component to lookup
1777 * @base: base directory to lookup from
1778 * @len: maximum length @len should be interpreted to
1780 * Note that this routine is purely a helper for filesystem usage and should
1781 * not be called by generic code. Also note that by using this function the
1782 * nameidata argument is passed to the filesystem methods and a filesystem
1783 * using this helper needs to be prepared for that.
1785 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1790 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1792 err = __lookup_one_len(name, &this, base, len);
1794 return ERR_PTR(err);
1796 return __lookup_hash(&this, base, NULL);
1799 int user_path_at(int dfd, const char __user *name, unsigned flags,
1802 struct nameidata nd;
1803 char *tmp = getname(name);
1804 int err = PTR_ERR(tmp);
1807 BUG_ON(flags & LOOKUP_PARENT);
1809 err = do_path_lookup(dfd, tmp, flags, &nd);
1817 static int user_path_parent(int dfd, const char __user *path,
1818 struct nameidata *nd, char **name)
1820 char *s = getname(path);
1826 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1836 * It's inline, so penalty for filesystems that don't use sticky bit is
1839 static inline int check_sticky(struct inode *dir, struct inode *inode)
1841 uid_t fsuid = current_fsuid();
1843 if (!(dir->i_mode & S_ISVTX))
1845 if (inode->i_uid == fsuid)
1847 if (dir->i_uid == fsuid)
1849 return !capable(CAP_FOWNER);
1853 * Check whether we can remove a link victim from directory dir, check
1854 * whether the type of victim is right.
1855 * 1. We can't do it if dir is read-only (done in permission())
1856 * 2. We should have write and exec permissions on dir
1857 * 3. We can't remove anything from append-only dir
1858 * 4. We can't do anything with immutable dir (done in permission())
1859 * 5. If the sticky bit on dir is set we should either
1860 * a. be owner of dir, or
1861 * b. be owner of victim, or
1862 * c. have CAP_FOWNER capability
1863 * 6. If the victim is append-only or immutable we can't do antyhing with
1864 * links pointing to it.
1865 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1866 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1867 * 9. We can't remove a root or mountpoint.
1868 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1869 * nfs_async_unlink().
1871 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1875 if (!victim->d_inode)
1878 BUG_ON(victim->d_parent->d_inode != dir);
1879 audit_inode_child(victim, dir);
1881 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1886 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1887 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1890 if (!S_ISDIR(victim->d_inode->i_mode))
1892 if (IS_ROOT(victim))
1894 } else if (S_ISDIR(victim->d_inode->i_mode))
1896 if (IS_DEADDIR(dir))
1898 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1903 /* Check whether we can create an object with dentry child in directory
1905 * 1. We can't do it if child already exists (open has special treatment for
1906 * this case, but since we are inlined it's OK)
1907 * 2. We can't do it if dir is read-only (done in permission())
1908 * 3. We should have write and exec permissions on dir
1909 * 4. We can't do it if dir is immutable (done in permission())
1911 static inline int may_create(struct inode *dir, struct dentry *child)
1915 if (IS_DEADDIR(dir))
1917 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1921 * p1 and p2 should be directories on the same fs.
1923 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1928 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1932 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1934 p = d_ancestor(p2, p1);
1936 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1937 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1941 p = d_ancestor(p1, p2);
1943 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1944 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1948 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1949 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1953 void unlock_rename(struct dentry *p1, struct dentry *p2)
1955 mutex_unlock(&p1->d_inode->i_mutex);
1957 mutex_unlock(&p2->d_inode->i_mutex);
1958 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1962 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1963 struct nameidata *nd)
1965 int error = may_create(dir, dentry);
1970 if (!dir->i_op->create)
1971 return -EACCES; /* shouldn't it be ENOSYS? */
1974 error = security_inode_create(dir, dentry, mode);
1977 error = dir->i_op->create(dir, dentry, mode, nd);
1979 fsnotify_create(dir, dentry);
1983 int may_open(struct path *path, int acc_mode, int flag)
1985 struct dentry *dentry = path->dentry;
1986 struct inode *inode = dentry->d_inode;
1992 switch (inode->i_mode & S_IFMT) {
1996 if (acc_mode & MAY_WRITE)
2001 if (path->mnt->mnt_flags & MNT_NODEV)
2010 error = inode_permission(inode, acc_mode);
2015 * An append-only file must be opened in append mode for writing.
2017 if (IS_APPEND(inode)) {
2018 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2024 /* O_NOATIME can only be set by the owner or superuser */
2025 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2029 * Ensure there are no outstanding leases on the file.
2031 return break_lease(inode, flag);
2034 static int handle_truncate(struct file *filp)
2036 struct path *path = &filp->f_path;
2037 struct inode *inode = path->dentry->d_inode;
2038 int error = get_write_access(inode);
2042 * Refuse to truncate files with mandatory locks held on them.
2044 error = locks_verify_locked(inode);
2046 error = security_path_truncate(path);
2048 error = do_truncate(path->dentry, 0,
2049 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2052 put_write_access(inode);
2057 * Be careful about ever adding any more callers of this
2058 * function. Its flags must be in the namei format, not
2059 * what get passed to sys_open().
2061 static int __open_namei_create(struct nameidata *nd, struct path *path,
2062 int open_flag, int mode)
2065 struct dentry *dir = nd->path.dentry;
2067 if (!IS_POSIXACL(dir->d_inode))
2068 mode &= ~current_umask();
2069 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
2072 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
2074 mutex_unlock(&dir->d_inode->i_mutex);
2075 dput(nd->path.dentry);
2076 nd->path.dentry = path->dentry;
2080 /* Don't check for write permission, don't truncate */
2081 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
2085 * Note that while the flag value (low two bits) for sys_open means:
2090 * it is changed into
2091 * 00 - no permissions needed
2092 * 01 - read-permission
2093 * 10 - write-permission
2095 * for the internal routines (ie open_namei()/follow_link() etc)
2096 * This is more logical, and also allows the 00 "no perm needed"
2097 * to be used for symlinks (where the permissions are checked
2101 static inline int open_to_namei_flags(int flag)
2103 if ((flag+1) & O_ACCMODE)
2108 static int open_will_truncate(int flag, struct inode *inode)
2111 * We'll never write to the fs underlying
2114 if (special_file(inode->i_mode))
2116 return (flag & O_TRUNC);
2119 static struct file *finish_open(struct nameidata *nd,
2120 int open_flag, int acc_mode)
2126 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
2127 if (will_truncate) {
2128 error = mnt_want_write(nd->path.mnt);
2132 error = may_open(&nd->path, acc_mode, open_flag);
2135 mnt_drop_write(nd->path.mnt);
2138 filp = nameidata_to_filp(nd);
2139 if (!IS_ERR(filp)) {
2140 error = ima_file_check(filp, acc_mode);
2143 filp = ERR_PTR(error);
2146 if (!IS_ERR(filp)) {
2147 if (will_truncate) {
2148 error = handle_truncate(filp);
2151 filp = ERR_PTR(error);
2156 * It is now safe to drop the mnt write
2157 * because the filp has had a write taken
2161 mnt_drop_write(nd->path.mnt);
2162 path_put(&nd->path);
2166 path_put(&nd->path);
2167 return ERR_PTR(error);
2171 * Handle O_CREAT case for do_filp_open
2173 static struct file *do_last(struct nameidata *nd, struct path *path,
2174 const struct open_flags *op, const char *pathname)
2176 struct dentry *dir = nd->path.dentry;
2180 nd->flags &= ~LOOKUP_PARENT;
2181 nd->flags |= op->intent;
2183 switch (nd->last_type) {
2186 dir = nd->path.dentry;
2190 error = handle_reval_path(nd);
2196 error = handle_reval_path(nd);
2199 audit_inode(pathname, dir);
2204 /* trailing slashes? */
2205 if (nd->last.name[nd->last.len])
2208 mutex_lock(&dir->d_inode->i_mutex);
2210 path->dentry = lookup_hash(nd);
2211 path->mnt = nd->path.mnt;
2213 error = PTR_ERR(path->dentry);
2214 if (IS_ERR(path->dentry)) {
2215 mutex_unlock(&dir->d_inode->i_mutex);
2219 if (IS_ERR(nd->intent.open.file)) {
2220 error = PTR_ERR(nd->intent.open.file);
2221 goto exit_mutex_unlock;
2224 /* Negative dentry, just create the file */
2225 if (!path->dentry->d_inode) {
2227 * This write is needed to ensure that a
2228 * ro->rw transition does not occur between
2229 * the time when the file is created and when
2230 * a permanent write count is taken through
2231 * the 'struct file' in nameidata_to_filp().
2233 error = mnt_want_write(nd->path.mnt);
2235 goto exit_mutex_unlock;
2236 error = __open_namei_create(nd, path, op->open_flag, op->mode);
2238 mnt_drop_write(nd->path.mnt);
2241 filp = nameidata_to_filp(nd);
2242 mnt_drop_write(nd->path.mnt);
2243 path_put(&nd->path);
2244 if (!IS_ERR(filp)) {
2245 error = ima_file_check(filp, op->acc_mode);
2248 filp = ERR_PTR(error);
2255 * It already exists.
2257 mutex_unlock(&dir->d_inode->i_mutex);
2258 audit_inode(pathname, path->dentry);
2261 if (op->open_flag & O_EXCL)
2264 error = follow_managed(path, nd->flags);
2269 if (!path->dentry->d_inode)
2272 if (path->dentry->d_inode->i_op->follow_link)
2275 path_to_nameidata(path, nd);
2276 nd->inode = path->dentry->d_inode;
2278 if (S_ISDIR(nd->inode->i_mode))
2281 filp = finish_open(nd, op->open_flag, op->acc_mode);
2285 mutex_unlock(&dir->d_inode->i_mutex);
2287 path_put_conditional(path, nd);
2289 path_put(&nd->path);
2290 return ERR_PTR(error);
2293 static struct file *path_openat(int dfd, const char *pathname,
2294 const struct open_flags *op, int flags)
2297 struct nameidata nd;
2302 filp = get_empty_filp();
2304 return ERR_PTR(-ENFILE);
2306 filp->f_flags = op->open_flag;
2307 nd.intent.open.file = filp;
2308 nd.intent.open.flags = open_to_namei_flags(op->open_flag);
2309 nd.intent.open.create_mode = op->mode;
2311 if (op->open_flag & O_CREAT)
2314 /* !O_CREAT, simple open */
2315 error = path_lookupat(dfd, pathname, flags | op->intent, &nd);
2316 if (unlikely(error))
2319 if (!(nd.flags & LOOKUP_FOLLOW)) {
2320 if (nd.inode->i_op->follow_link)
2324 if (nd.flags & LOOKUP_DIRECTORY) {
2325 if (!nd.inode->i_op->lookup)
2328 audit_inode(pathname, nd.path.dentry);
2329 filp = finish_open(&nd, op->open_flag, op->acc_mode);
2330 release_open_intent(&nd);
2334 /* OK, have to create the file. Find the parent. */
2335 error = path_lookupat(dfd, pathname, LOOKUP_PARENT | flags, &nd);
2336 if (unlikely(error))
2338 if (unlikely(!audit_dummy_context()))
2339 audit_inode(pathname, nd.path.dentry);
2342 * We have the parent and last component.
2344 filp = do_last(&nd, &path, op, pathname);
2345 while (unlikely(!filp)) { /* trailing symlink */
2346 struct path link = path;
2347 struct inode *linki = link.dentry->d_inode;
2350 if (!(nd.flags & LOOKUP_FOLLOW))
2355 * This is subtle. Instead of calling do_follow_link() we do
2356 * the thing by hands. The reason is that this way we have zero
2357 * link_count and path_walk() (called from ->follow_link)
2358 * honoring LOOKUP_PARENT. After that we have the parent and
2359 * last component, i.e. we are in the same situation as after
2360 * the first path_walk(). Well, almost - if the last component
2361 * is normal we get its copy stored in nd->last.name and we will
2362 * have to putname() it when we are done. Procfs-like symlinks
2363 * just set LAST_BIND.
2365 nd.flags |= LOOKUP_PARENT;
2366 nd.flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2367 error = __do_follow_link(&link, &nd, &cookie);
2368 if (unlikely(error))
2369 filp = ERR_PTR(error);
2371 filp = do_last(&nd, &path, op, pathname);
2372 if (!IS_ERR(cookie) && linki->i_op->put_link)
2373 linki->i_op->put_link(link.dentry, &nd, cookie);
2379 release_open_intent(&nd);
2383 path_put_conditional(&path, &nd);
2387 filp = ERR_PTR(error);
2391 struct file *do_filp_open(int dfd, const char *pathname,
2392 const struct open_flags *op, int flags)
2396 filp = path_openat(dfd, pathname, op, flags | LOOKUP_RCU);
2397 if (unlikely(filp == ERR_PTR(-ECHILD)))
2398 filp = path_openat(dfd, pathname, op, flags);
2399 if (unlikely(filp == ERR_PTR(-ESTALE)))
2400 filp = path_openat(dfd, pathname, op, flags | LOOKUP_REVAL);
2405 * lookup_create - lookup a dentry, creating it if it doesn't exist
2406 * @nd: nameidata info
2407 * @is_dir: directory flag
2409 * Simple function to lookup and return a dentry and create it
2410 * if it doesn't exist. Is SMP-safe.
2412 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2414 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2416 struct dentry *dentry = ERR_PTR(-EEXIST);
2418 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2420 * Yucky last component or no last component at all?
2421 * (foo/., foo/.., /////)
2423 if (nd->last_type != LAST_NORM)
2425 nd->flags &= ~LOOKUP_PARENT;
2426 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2427 nd->intent.open.flags = O_EXCL;
2430 * Do the final lookup.
2432 dentry = lookup_hash(nd);
2436 if (dentry->d_inode)
2439 * Special case - lookup gave negative, but... we had foo/bar/
2440 * From the vfs_mknod() POV we just have a negative dentry -
2441 * all is fine. Let's be bastards - you had / on the end, you've
2442 * been asking for (non-existent) directory. -ENOENT for you.
2444 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2446 dentry = ERR_PTR(-ENOENT);
2451 dentry = ERR_PTR(-EEXIST);
2455 EXPORT_SYMBOL_GPL(lookup_create);
2457 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2459 int error = may_create(dir, dentry);
2464 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2467 if (!dir->i_op->mknod)
2470 error = devcgroup_inode_mknod(mode, dev);
2474 error = security_inode_mknod(dir, dentry, mode, dev);
2478 error = dir->i_op->mknod(dir, dentry, mode, dev);
2480 fsnotify_create(dir, dentry);
2484 static int may_mknod(mode_t mode)
2486 switch (mode & S_IFMT) {
2492 case 0: /* zero mode translates to S_IFREG */
2501 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2506 struct dentry *dentry;
2507 struct nameidata nd;
2512 error = user_path_parent(dfd, filename, &nd, &tmp);
2516 dentry = lookup_create(&nd, 0);
2517 if (IS_ERR(dentry)) {
2518 error = PTR_ERR(dentry);
2521 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2522 mode &= ~current_umask();
2523 error = may_mknod(mode);
2526 error = mnt_want_write(nd.path.mnt);
2529 error = security_path_mknod(&nd.path, dentry, mode, dev);
2531 goto out_drop_write;
2532 switch (mode & S_IFMT) {
2533 case 0: case S_IFREG:
2534 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2536 case S_IFCHR: case S_IFBLK:
2537 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2538 new_decode_dev(dev));
2540 case S_IFIFO: case S_IFSOCK:
2541 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2545 mnt_drop_write(nd.path.mnt);
2549 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2556 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2558 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2561 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2563 int error = may_create(dir, dentry);
2568 if (!dir->i_op->mkdir)
2571 mode &= (S_IRWXUGO|S_ISVTX);
2572 error = security_inode_mkdir(dir, dentry, mode);
2576 error = dir->i_op->mkdir(dir, dentry, mode);
2578 fsnotify_mkdir(dir, dentry);
2582 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2586 struct dentry *dentry;
2587 struct nameidata nd;
2589 error = user_path_parent(dfd, pathname, &nd, &tmp);
2593 dentry = lookup_create(&nd, 1);
2594 error = PTR_ERR(dentry);
2598 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2599 mode &= ~current_umask();
2600 error = mnt_want_write(nd.path.mnt);
2603 error = security_path_mkdir(&nd.path, dentry, mode);
2605 goto out_drop_write;
2606 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2608 mnt_drop_write(nd.path.mnt);
2612 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2619 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2621 return sys_mkdirat(AT_FDCWD, pathname, mode);
2625 * We try to drop the dentry early: we should have
2626 * a usage count of 2 if we're the only user of this
2627 * dentry, and if that is true (possibly after pruning
2628 * the dcache), then we drop the dentry now.
2630 * A low-level filesystem can, if it choses, legally
2633 * if (!d_unhashed(dentry))
2636 * if it cannot handle the case of removing a directory
2637 * that is still in use by something else..
2639 void dentry_unhash(struct dentry *dentry)
2642 shrink_dcache_parent(dentry);
2643 spin_lock(&dentry->d_lock);
2644 if (dentry->d_count == 2)
2646 spin_unlock(&dentry->d_lock);
2649 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2651 int error = may_delete(dir, dentry, 1);
2656 if (!dir->i_op->rmdir)
2659 mutex_lock(&dentry->d_inode->i_mutex);
2660 dentry_unhash(dentry);
2661 if (d_mountpoint(dentry))
2664 error = security_inode_rmdir(dir, dentry);
2666 error = dir->i_op->rmdir(dir, dentry);
2668 dentry->d_inode->i_flags |= S_DEAD;
2673 mutex_unlock(&dentry->d_inode->i_mutex);
2682 static long do_rmdir(int dfd, const char __user *pathname)
2686 struct dentry *dentry;
2687 struct nameidata nd;
2689 error = user_path_parent(dfd, pathname, &nd, &name);
2693 switch(nd.last_type) {
2705 nd.flags &= ~LOOKUP_PARENT;
2707 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2708 dentry = lookup_hash(&nd);
2709 error = PTR_ERR(dentry);
2712 error = mnt_want_write(nd.path.mnt);
2715 error = security_path_rmdir(&nd.path, dentry);
2718 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2720 mnt_drop_write(nd.path.mnt);
2724 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2731 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2733 return do_rmdir(AT_FDCWD, pathname);
2736 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2738 int error = may_delete(dir, dentry, 0);
2743 if (!dir->i_op->unlink)
2746 mutex_lock(&dentry->d_inode->i_mutex);
2747 if (d_mountpoint(dentry))
2750 error = security_inode_unlink(dir, dentry);
2752 error = dir->i_op->unlink(dir, dentry);
2757 mutex_unlock(&dentry->d_inode->i_mutex);
2759 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2760 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2761 fsnotify_link_count(dentry->d_inode);
2769 * Make sure that the actual truncation of the file will occur outside its
2770 * directory's i_mutex. Truncate can take a long time if there is a lot of
2771 * writeout happening, and we don't want to prevent access to the directory
2772 * while waiting on the I/O.
2774 static long do_unlinkat(int dfd, const char __user *pathname)
2778 struct dentry *dentry;
2779 struct nameidata nd;
2780 struct inode *inode = NULL;
2782 error = user_path_parent(dfd, pathname, &nd, &name);
2787 if (nd.last_type != LAST_NORM)
2790 nd.flags &= ~LOOKUP_PARENT;
2792 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2793 dentry = lookup_hash(&nd);
2794 error = PTR_ERR(dentry);
2795 if (!IS_ERR(dentry)) {
2796 /* Why not before? Because we want correct error value */
2797 if (nd.last.name[nd.last.len])
2799 inode = dentry->d_inode;
2802 error = mnt_want_write(nd.path.mnt);
2805 error = security_path_unlink(&nd.path, dentry);
2808 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2810 mnt_drop_write(nd.path.mnt);
2814 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2816 iput(inode); /* truncate the inode here */
2823 error = !dentry->d_inode ? -ENOENT :
2824 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2828 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2830 if ((flag & ~AT_REMOVEDIR) != 0)
2833 if (flag & AT_REMOVEDIR)
2834 return do_rmdir(dfd, pathname);
2836 return do_unlinkat(dfd, pathname);
2839 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2841 return do_unlinkat(AT_FDCWD, pathname);
2844 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2846 int error = may_create(dir, dentry);
2851 if (!dir->i_op->symlink)
2854 error = security_inode_symlink(dir, dentry, oldname);
2858 error = dir->i_op->symlink(dir, dentry, oldname);
2860 fsnotify_create(dir, dentry);
2864 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2865 int, newdfd, const char __user *, newname)
2870 struct dentry *dentry;
2871 struct nameidata nd;
2873 from = getname(oldname);
2875 return PTR_ERR(from);
2877 error = user_path_parent(newdfd, newname, &nd, &to);
2881 dentry = lookup_create(&nd, 0);
2882 error = PTR_ERR(dentry);
2886 error = mnt_want_write(nd.path.mnt);
2889 error = security_path_symlink(&nd.path, dentry, from);
2891 goto out_drop_write;
2892 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2894 mnt_drop_write(nd.path.mnt);
2898 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2906 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2908 return sys_symlinkat(oldname, AT_FDCWD, newname);
2911 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2913 struct inode *inode = old_dentry->d_inode;
2919 error = may_create(dir, new_dentry);
2923 if (dir->i_sb != inode->i_sb)
2927 * A link to an append-only or immutable file cannot be created.
2929 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2931 if (!dir->i_op->link)
2933 if (S_ISDIR(inode->i_mode))
2936 error = security_inode_link(old_dentry, dir, new_dentry);
2940 mutex_lock(&inode->i_mutex);
2941 error = dir->i_op->link(old_dentry, dir, new_dentry);
2942 mutex_unlock(&inode->i_mutex);
2944 fsnotify_link(dir, inode, new_dentry);
2949 * Hardlinks are often used in delicate situations. We avoid
2950 * security-related surprises by not following symlinks on the
2953 * We don't follow them on the oldname either to be compatible
2954 * with linux 2.0, and to avoid hard-linking to directories
2955 * and other special files. --ADM
2957 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2958 int, newdfd, const char __user *, newname, int, flags)
2960 struct dentry *new_dentry;
2961 struct nameidata nd;
2962 struct path old_path;
2966 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2969 error = user_path_at(olddfd, oldname,
2970 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2975 error = user_path_parent(newdfd, newname, &nd, &to);
2979 if (old_path.mnt != nd.path.mnt)
2981 new_dentry = lookup_create(&nd, 0);
2982 error = PTR_ERR(new_dentry);
2983 if (IS_ERR(new_dentry))
2985 error = mnt_want_write(nd.path.mnt);
2988 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2990 goto out_drop_write;
2991 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2993 mnt_drop_write(nd.path.mnt);
2997 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3002 path_put(&old_path);
3007 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3009 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3013 * The worst of all namespace operations - renaming directory. "Perverted"
3014 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3016 * a) we can get into loop creation. Check is done in is_subdir().
3017 * b) race potential - two innocent renames can create a loop together.
3018 * That's where 4.4 screws up. Current fix: serialization on
3019 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3021 * c) we have to lock _three_ objects - parents and victim (if it exists).
3022 * And that - after we got ->i_mutex on parents (until then we don't know
3023 * whether the target exists). Solution: try to be smart with locking
3024 * order for inodes. We rely on the fact that tree topology may change
3025 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3026 * move will be locked. Thus we can rank directories by the tree
3027 * (ancestors first) and rank all non-directories after them.
3028 * That works since everybody except rename does "lock parent, lookup,
3029 * lock child" and rename is under ->s_vfs_rename_mutex.
3030 * HOWEVER, it relies on the assumption that any object with ->lookup()
3031 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3032 * we'd better make sure that there's no link(2) for them.
3033 * d) some filesystems don't support opened-but-unlinked directories,
3034 * either because of layout or because they are not ready to deal with
3035 * all cases correctly. The latter will be fixed (taking this sort of
3036 * stuff into VFS), but the former is not going away. Solution: the same
3037 * trick as in rmdir().
3038 * e) conversion from fhandle to dentry may come in the wrong moment - when
3039 * we are removing the target. Solution: we will have to grab ->i_mutex
3040 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3041 * ->i_mutex on parents, which works but leads to some truly excessive
3044 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3045 struct inode *new_dir, struct dentry *new_dentry)
3048 struct inode *target;
3051 * If we are going to change the parent - check write permissions,
3052 * we'll need to flip '..'.
3054 if (new_dir != old_dir) {
3055 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3060 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3064 target = new_dentry->d_inode;
3066 mutex_lock(&target->i_mutex);
3067 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3071 dentry_unhash(new_dentry);
3072 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3076 target->i_flags |= S_DEAD;
3077 dont_mount(new_dentry);
3079 mutex_unlock(&target->i_mutex);
3080 if (d_unhashed(new_dentry))
3081 d_rehash(new_dentry);
3085 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3086 d_move(old_dentry,new_dentry);
3090 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3091 struct inode *new_dir, struct dentry *new_dentry)
3093 struct inode *target;
3096 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3101 target = new_dentry->d_inode;
3103 mutex_lock(&target->i_mutex);
3104 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3107 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3110 dont_mount(new_dentry);
3111 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3112 d_move(old_dentry, new_dentry);
3115 mutex_unlock(&target->i_mutex);
3120 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3121 struct inode *new_dir, struct dentry *new_dentry)
3124 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3125 const unsigned char *old_name;
3127 if (old_dentry->d_inode == new_dentry->d_inode)
3130 error = may_delete(old_dir, old_dentry, is_dir);
3134 if (!new_dentry->d_inode)
3135 error = may_create(new_dir, new_dentry);
3137 error = may_delete(new_dir, new_dentry, is_dir);
3141 if (!old_dir->i_op->rename)
3144 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3147 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3149 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3151 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3152 new_dentry->d_inode, old_dentry);
3153 fsnotify_oldname_free(old_name);
3158 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3159 int, newdfd, const char __user *, newname)
3161 struct dentry *old_dir, *new_dir;
3162 struct dentry *old_dentry, *new_dentry;
3163 struct dentry *trap;
3164 struct nameidata oldnd, newnd;
3169 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3173 error = user_path_parent(newdfd, newname, &newnd, &to);
3178 if (oldnd.path.mnt != newnd.path.mnt)
3181 old_dir = oldnd.path.dentry;
3183 if (oldnd.last_type != LAST_NORM)
3186 new_dir = newnd.path.dentry;
3187 if (newnd.last_type != LAST_NORM)
3190 oldnd.flags &= ~LOOKUP_PARENT;
3191 newnd.flags &= ~LOOKUP_PARENT;
3192 newnd.flags |= LOOKUP_RENAME_TARGET;
3194 trap = lock_rename(new_dir, old_dir);
3196 old_dentry = lookup_hash(&oldnd);
3197 error = PTR_ERR(old_dentry);
3198 if (IS_ERR(old_dentry))
3200 /* source must exist */
3202 if (!old_dentry->d_inode)
3204 /* unless the source is a directory trailing slashes give -ENOTDIR */
3205 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3207 if (oldnd.last.name[oldnd.last.len])
3209 if (newnd.last.name[newnd.last.len])
3212 /* source should not be ancestor of target */
3214 if (old_dentry == trap)
3216 new_dentry = lookup_hash(&newnd);
3217 error = PTR_ERR(new_dentry);
3218 if (IS_ERR(new_dentry))
3220 /* target should not be an ancestor of source */
3222 if (new_dentry == trap)
3225 error = mnt_want_write(oldnd.path.mnt);
3228 error = security_path_rename(&oldnd.path, old_dentry,
3229 &newnd.path, new_dentry);
3232 error = vfs_rename(old_dir->d_inode, old_dentry,
3233 new_dir->d_inode, new_dentry);
3235 mnt_drop_write(oldnd.path.mnt);
3241 unlock_rename(new_dir, old_dir);
3243 path_put(&newnd.path);
3246 path_put(&oldnd.path);
3252 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3254 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3257 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3261 len = PTR_ERR(link);
3266 if (len > (unsigned) buflen)
3268 if (copy_to_user(buffer, link, len))
3275 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3276 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3277 * using) it for any given inode is up to filesystem.
3279 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3281 struct nameidata nd;
3286 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3288 return PTR_ERR(cookie);
3290 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3291 if (dentry->d_inode->i_op->put_link)
3292 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3296 int vfs_follow_link(struct nameidata *nd, const char *link)
3298 return __vfs_follow_link(nd, link);
3301 /* get the link contents into pagecache */
3302 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3306 struct address_space *mapping = dentry->d_inode->i_mapping;
3307 page = read_mapping_page(mapping, 0, NULL);
3312 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3316 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3318 struct page *page = NULL;
3319 char *s = page_getlink(dentry, &page);
3320 int res = vfs_readlink(dentry,buffer,buflen,s);
3323 page_cache_release(page);
3328 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3330 struct page *page = NULL;
3331 nd_set_link(nd, page_getlink(dentry, &page));
3335 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3337 struct page *page = cookie;
3341 page_cache_release(page);
3346 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3348 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3350 struct address_space *mapping = inode->i_mapping;
3355 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3357 flags |= AOP_FLAG_NOFS;
3360 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3361 flags, &page, &fsdata);
3365 kaddr = kmap_atomic(page, KM_USER0);
3366 memcpy(kaddr, symname, len-1);
3367 kunmap_atomic(kaddr, KM_USER0);
3369 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3376 mark_inode_dirty(inode);
3382 int page_symlink(struct inode *inode, const char *symname, int len)
3384 return __page_symlink(inode, symname, len,
3385 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3388 const struct inode_operations page_symlink_inode_operations = {
3389 .readlink = generic_readlink,
3390 .follow_link = page_follow_link_light,
3391 .put_link = page_put_link,
3394 EXPORT_SYMBOL(user_path_at);
3395 EXPORT_SYMBOL(follow_down_one);
3396 EXPORT_SYMBOL(follow_down);
3397 EXPORT_SYMBOL(follow_up);
3398 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3399 EXPORT_SYMBOL(getname);
3400 EXPORT_SYMBOL(lock_rename);
3401 EXPORT_SYMBOL(lookup_one_len);
3402 EXPORT_SYMBOL(page_follow_link_light);
3403 EXPORT_SYMBOL(page_put_link);
3404 EXPORT_SYMBOL(page_readlink);
3405 EXPORT_SYMBOL(__page_symlink);
3406 EXPORT_SYMBOL(page_symlink);
3407 EXPORT_SYMBOL(page_symlink_inode_operations);
3408 EXPORT_SYMBOL(kern_path_parent);
3409 EXPORT_SYMBOL(kern_path);
3410 EXPORT_SYMBOL(vfs_path_lookup);
3411 EXPORT_SYMBOL(inode_permission);
3412 EXPORT_SYMBOL(file_permission);
3413 EXPORT_SYMBOL(unlock_rename);
3414 EXPORT_SYMBOL(vfs_create);
3415 EXPORT_SYMBOL(vfs_follow_link);
3416 EXPORT_SYMBOL(vfs_link);
3417 EXPORT_SYMBOL(vfs_mkdir);
3418 EXPORT_SYMBOL(vfs_mknod);
3419 EXPORT_SYMBOL(generic_permission);
3420 EXPORT_SYMBOL(vfs_readlink);
3421 EXPORT_SYMBOL(vfs_rename);
3422 EXPORT_SYMBOL(vfs_rmdir);
3423 EXPORT_SYMBOL(vfs_symlink);
3424 EXPORT_SYMBOL(vfs_unlink);
3425 EXPORT_SYMBOL(dentry_unhash);
3426 EXPORT_SYMBOL(generic_readlink);