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_get_long - get a long reference to a path
372 * @path: path to get the reference to
374 * Given a path increment the reference count to the dentry and the vfsmount.
376 void path_get_long(struct path *path)
378 mntget_long(path->mnt);
383 * path_put - put a reference to a path
384 * @path: path to put the reference to
386 * Given a path decrement the reference count to the dentry and the vfsmount.
388 void path_put(struct path *path)
393 EXPORT_SYMBOL(path_put);
396 * path_put_long - put a long reference to a path
397 * @path: path to put the reference to
399 * Given a path decrement the reference count to the dentry and the vfsmount.
401 void path_put_long(struct path *path)
404 mntput_long(path->mnt);
408 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
409 * @nd: nameidata pathwalk data to drop
410 * Returns: 0 on success, -ECHILD on failure
412 * Path walking has 2 modes, rcu-walk and ref-walk (see
413 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
414 * to drop out of rcu-walk mode and take normal reference counts on dentries
415 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
416 * refcounts at the last known good point before rcu-walk got stuck, so
417 * ref-walk may continue from there. If this is not successful (eg. a seqcount
418 * has changed), then failure is returned and path walk restarts from the
419 * beginning in ref-walk mode.
421 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
422 * ref-walk. Must be called from rcu-walk context.
424 static int nameidata_drop_rcu(struct nameidata *nd)
426 struct fs_struct *fs = current->fs;
427 struct dentry *dentry = nd->path.dentry;
429 BUG_ON(!(nd->flags & LOOKUP_RCU));
431 spin_lock(&fs->lock);
432 if (nd->root.mnt != fs->root.mnt ||
433 nd->root.dentry != fs->root.dentry)
436 spin_lock(&dentry->d_lock);
437 if (!__d_rcu_to_refcount(dentry, nd->seq))
439 BUG_ON(nd->inode != dentry->d_inode);
440 spin_unlock(&dentry->d_lock);
443 spin_unlock(&fs->lock);
445 mntget(nd->path.mnt);
448 br_read_unlock(vfsmount_lock);
449 nd->flags &= ~LOOKUP_RCU;
452 spin_unlock(&dentry->d_lock);
455 spin_unlock(&fs->lock);
459 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
460 static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
462 if (nd->flags & LOOKUP_RCU)
463 return nameidata_drop_rcu(nd);
468 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
469 * @nd: nameidata pathwalk data to drop
470 * @dentry: dentry to drop
471 * Returns: 0 on success, -ECHILD on failure
473 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
474 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
475 * @nd. Must be called from rcu-walk context.
477 static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
479 struct fs_struct *fs = current->fs;
480 struct dentry *parent = nd->path.dentry;
483 * It can be possible to revalidate the dentry that we started
484 * the path walk with. force_reval_path may also revalidate the
485 * dentry already committed to the nameidata.
487 if (unlikely(parent == dentry))
488 return nameidata_drop_rcu(nd);
490 BUG_ON(!(nd->flags & LOOKUP_RCU));
492 spin_lock(&fs->lock);
493 if (nd->root.mnt != fs->root.mnt ||
494 nd->root.dentry != fs->root.dentry)
497 spin_lock(&parent->d_lock);
498 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
499 if (!__d_rcu_to_refcount(dentry, nd->seq))
502 * If the sequence check on the child dentry passed, then the child has
503 * not been removed from its parent. This means the parent dentry must
504 * be valid and able to take a reference at this point.
506 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
507 BUG_ON(!parent->d_count);
509 spin_unlock(&dentry->d_lock);
510 spin_unlock(&parent->d_lock);
513 spin_unlock(&fs->lock);
515 mntget(nd->path.mnt);
518 br_read_unlock(vfsmount_lock);
519 nd->flags &= ~LOOKUP_RCU;
522 spin_unlock(&dentry->d_lock);
523 spin_unlock(&parent->d_lock);
526 spin_unlock(&fs->lock);
530 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
531 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
533 if (nd->flags & LOOKUP_RCU)
534 return nameidata_dentry_drop_rcu(nd, dentry);
539 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
540 * @nd: nameidata pathwalk data to drop
541 * Returns: 0 on success, -ECHILD on failure
543 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
544 * nd->path should be the final element of the lookup, so nd->root is discarded.
545 * Must be called from rcu-walk context.
547 static int nameidata_drop_rcu_last(struct nameidata *nd)
549 struct dentry *dentry = nd->path.dentry;
551 BUG_ON(!(nd->flags & LOOKUP_RCU));
552 nd->flags &= ~LOOKUP_RCU;
554 spin_lock(&dentry->d_lock);
555 if (!__d_rcu_to_refcount(dentry, nd->seq))
557 BUG_ON(nd->inode != dentry->d_inode);
558 spin_unlock(&dentry->d_lock);
560 mntget(nd->path.mnt);
563 br_read_unlock(vfsmount_lock);
568 spin_unlock(&dentry->d_lock);
570 br_read_unlock(vfsmount_lock);
574 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
575 static inline int nameidata_drop_rcu_last_maybe(struct nameidata *nd)
577 if (likely(nd->flags & LOOKUP_RCU))
578 return nameidata_drop_rcu_last(nd);
583 * release_open_intent - free up open intent resources
584 * @nd: pointer to nameidata
586 void release_open_intent(struct nameidata *nd)
588 if (nd->intent.open.file->f_path.dentry == NULL)
589 put_filp(nd->intent.open.file);
591 fput(nd->intent.open.file);
595 * Call d_revalidate and handle filesystems that request rcu-walk
596 * to be dropped. This may be called and return in rcu-walk mode,
597 * regardless of success or error. If -ECHILD is returned, the caller
598 * must return -ECHILD back up the path walk stack so path walk may
599 * be restarted in ref-walk mode.
601 static int d_revalidate(struct dentry *dentry, struct nameidata *nd)
605 status = dentry->d_op->d_revalidate(dentry, nd);
606 if (status == -ECHILD) {
607 if (nameidata_dentry_drop_rcu(nd, dentry))
609 status = dentry->d_op->d_revalidate(dentry, nd);
615 static inline struct dentry *
616 do_revalidate(struct dentry *dentry, struct nameidata *nd)
620 status = d_revalidate(dentry, nd);
621 if (unlikely(status <= 0)) {
623 * The dentry failed validation.
624 * If d_revalidate returned 0 attempt to invalidate
625 * the dentry otherwise d_revalidate is asking us
626 * to return a fail status.
629 /* If we're in rcu-walk, we don't have a ref */
630 if (!(nd->flags & LOOKUP_RCU))
632 dentry = ERR_PTR(status);
635 /* Don't d_invalidate in rcu-walk mode */
636 if (nameidata_dentry_drop_rcu_maybe(nd, dentry))
637 return ERR_PTR(-ECHILD);
638 if (!d_invalidate(dentry)) {
647 static inline int need_reval_dot(struct dentry *dentry)
649 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
652 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
659 * force_reval_path - force revalidation of a dentry
661 * In some situations the path walking code will trust dentries without
662 * revalidating them. This causes problems for filesystems that depend on
663 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
664 * (which indicates that it's possible for the dentry to go stale), force
665 * a d_revalidate call before proceeding.
667 * Returns 0 if the revalidation was successful. If the revalidation fails,
668 * either return the error returned by d_revalidate or -ESTALE if the
669 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
670 * invalidate the dentry. It's up to the caller to handle putting references
671 * to the path if necessary.
674 force_reval_path(struct path *path, struct nameidata *nd)
677 struct dentry *dentry = path->dentry;
680 * only check on filesystems where it's possible for the dentry to
683 if (!need_reval_dot(dentry))
686 status = d_revalidate(dentry, nd);
691 /* Don't d_invalidate in rcu-walk mode */
692 if (nameidata_drop_rcu(nd))
694 d_invalidate(dentry);
701 * Short-cut version of permission(), for calling on directories
702 * during pathname resolution. Combines parts of permission()
703 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
705 * If appropriate, check DAC only. If not appropriate, or
706 * short-cut DAC fails, then call ->permission() to do more
707 * complete permission check.
709 static inline int exec_permission(struct inode *inode, unsigned int flags)
713 if (inode->i_op->permission) {
714 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
716 ret = acl_permission_check(inode, MAY_EXEC, flags,
717 inode->i_op->check_acl);
724 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
729 return security_inode_exec_permission(inode, flags);
732 static __always_inline void set_root(struct nameidata *nd)
735 get_fs_root(current->fs, &nd->root);
738 static int link_path_walk(const char *, struct nameidata *);
740 static __always_inline void set_root_rcu(struct nameidata *nd)
743 struct fs_struct *fs = current->fs;
747 seq = read_seqcount_begin(&fs->seq);
749 } while (read_seqcount_retry(&fs->seq, seq));
753 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
766 nd->inode = nd->path.dentry->d_inode;
768 ret = link_path_walk(link, nd);
772 return PTR_ERR(link);
775 static void path_put_conditional(struct path *path, struct nameidata *nd)
778 if (path->mnt != nd->path.mnt)
782 static inline void path_to_nameidata(const struct path *path,
783 struct nameidata *nd)
785 if (!(nd->flags & LOOKUP_RCU)) {
786 dput(nd->path.dentry);
787 if (nd->path.mnt != path->mnt)
788 mntput(nd->path.mnt);
790 nd->path.mnt = path->mnt;
791 nd->path.dentry = path->dentry;
794 static __always_inline int
795 __do_follow_link(const struct path *link, struct nameidata *nd, void **p)
798 struct dentry *dentry = link->dentry;
800 touch_atime(link->mnt, dentry);
801 nd_set_link(nd, NULL);
803 if (link->mnt != nd->path.mnt) {
804 path_to_nameidata(link, nd);
805 nd->inode = nd->path.dentry->d_inode;
810 nd->last_type = LAST_BIND;
811 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
814 char *s = nd_get_link(nd);
817 error = __vfs_follow_link(nd, s);
818 else if (nd->last_type == LAST_BIND) {
819 error = force_reval_path(&nd->path, nd);
828 * This limits recursive symlink follows to 8, while
829 * limiting consecutive symlinks to 40.
831 * Without that kind of total limit, nasty chains of consecutive
832 * symlinks can cause almost arbitrarily long lookups.
834 static inline int do_follow_link(struct path *path, struct nameidata *nd)
838 if (current->link_count >= MAX_NESTED_LINKS)
840 if (current->total_link_count >= 40)
842 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
844 err = security_inode_follow_link(path->dentry, nd);
847 current->link_count++;
848 current->total_link_count++;
850 err = __do_follow_link(path, nd, &cookie);
851 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
852 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
854 current->link_count--;
858 path_put_conditional(path, nd);
863 static int follow_up_rcu(struct path *path)
865 struct vfsmount *parent;
866 struct dentry *mountpoint;
868 parent = path->mnt->mnt_parent;
869 if (parent == path->mnt)
871 mountpoint = path->mnt->mnt_mountpoint;
872 path->dentry = mountpoint;
877 int follow_up(struct path *path)
879 struct vfsmount *parent;
880 struct dentry *mountpoint;
882 br_read_lock(vfsmount_lock);
883 parent = path->mnt->mnt_parent;
884 if (parent == path->mnt) {
885 br_read_unlock(vfsmount_lock);
889 mountpoint = dget(path->mnt->mnt_mountpoint);
890 br_read_unlock(vfsmount_lock);
892 path->dentry = mountpoint;
899 * Perform an automount
900 * - return -EISDIR to tell follow_managed() to stop and return the path we
903 static int follow_automount(struct path *path, unsigned flags,
906 struct vfsmount *mnt;
908 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
911 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
912 * and this is the terminal part of the path.
914 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
915 return -EISDIR; /* we actually want to stop here */
917 /* We want to mount if someone is trying to open/create a file of any
918 * type under the mountpoint, wants to traverse through the mountpoint
919 * or wants to open the mounted directory.
921 * We don't want to mount if someone's just doing a stat and they've
922 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
923 * appended a '/' to the name.
925 if (!(flags & LOOKUP_FOLLOW) &&
926 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
927 LOOKUP_OPEN | LOOKUP_CREATE)))
930 current->total_link_count++;
931 if (current->total_link_count >= 40)
934 mnt = path->dentry->d_op->d_automount(path);
937 * The filesystem is allowed to return -EISDIR here to indicate
938 * it doesn't want to automount. For instance, autofs would do
939 * this so that its userspace daemon can mount on this dentry.
941 * However, we can only permit this if it's a terminal point in
942 * the path being looked up; if it wasn't then the remainder of
943 * the path is inaccessible and we should say so.
945 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
949 if (!mnt) /* mount collision */
952 if (mnt->mnt_sb == path->mnt->mnt_sb &&
953 mnt->mnt_root == path->dentry) {
962 path->dentry = dget(mnt->mnt_root);
968 * Handle a dentry that is managed in some way.
969 * - Flagged for transit management (autofs)
970 * - Flagged as mountpoint
971 * - Flagged as automount point
973 * This may only be called in refwalk mode.
975 * Serialization is taken care of in namespace.c
977 static int follow_managed(struct path *path, unsigned flags)
980 bool need_mntput = false;
983 /* Given that we're not holding a lock here, we retain the value in a
984 * local variable for each dentry as we look at it so that we don't see
985 * the components of that value change under us */
986 while (managed = ACCESS_ONCE(path->dentry->d_flags),
987 managed &= DCACHE_MANAGED_DENTRY,
988 unlikely(managed != 0)) {
989 /* Allow the filesystem to manage the transit without i_mutex
991 if (managed & DCACHE_MANAGE_TRANSIT) {
992 BUG_ON(!path->dentry->d_op);
993 BUG_ON(!path->dentry->d_op->d_manage);
994 ret = path->dentry->d_op->d_manage(path->dentry, false);
996 return ret == -EISDIR ? 0 : ret;
999 /* Transit to a mounted filesystem. */
1000 if (managed & DCACHE_MOUNTED) {
1001 struct vfsmount *mounted = lookup_mnt(path);
1006 path->mnt = mounted;
1007 path->dentry = dget(mounted->mnt_root);
1012 /* Something is mounted on this dentry in another
1013 * namespace and/or whatever was mounted there in this
1014 * namespace got unmounted before we managed to get the
1018 /* Handle an automount point */
1019 if (managed & DCACHE_NEED_AUTOMOUNT) {
1020 ret = follow_automount(path, flags, &need_mntput);
1022 return ret == -EISDIR ? 0 : ret;
1026 /* We didn't change the current path point */
1032 int follow_down_one(struct path *path)
1034 struct vfsmount *mounted;
1036 mounted = lookup_mnt(path);
1040 path->mnt = mounted;
1041 path->dentry = dget(mounted->mnt_root);
1048 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
1049 * meet a managed dentry and we're not walking to "..". True is returned to
1050 * continue, false to abort.
1052 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1053 struct inode **inode, bool reverse_transit)
1055 unsigned abort_mask =
1056 reverse_transit ? 0 : DCACHE_MANAGE_TRANSIT;
1058 while (d_mountpoint(path->dentry)) {
1059 struct vfsmount *mounted;
1060 if (path->dentry->d_flags & abort_mask)
1062 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1065 path->mnt = mounted;
1066 path->dentry = mounted->mnt_root;
1067 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1068 *inode = path->dentry->d_inode;
1071 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1072 return reverse_transit;
1076 static int follow_dotdot_rcu(struct nameidata *nd)
1078 struct inode *inode = nd->inode;
1083 if (nd->path.dentry == nd->root.dentry &&
1084 nd->path.mnt == nd->root.mnt) {
1087 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1088 struct dentry *old = nd->path.dentry;
1089 struct dentry *parent = old->d_parent;
1092 seq = read_seqcount_begin(&parent->d_seq);
1093 if (read_seqcount_retry(&old->d_seq, nd->seq))
1095 inode = parent->d_inode;
1096 nd->path.dentry = parent;
1100 if (!follow_up_rcu(&nd->path))
1102 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1103 inode = nd->path.dentry->d_inode;
1105 __follow_mount_rcu(nd, &nd->path, &inode, true);
1112 * Follow down to the covering mount currently visible to userspace. At each
1113 * point, the filesystem owning that dentry may be queried as to whether the
1114 * caller is permitted to proceed or not.
1116 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1119 int follow_down(struct path *path, bool mounting_here)
1124 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1125 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1126 /* Allow the filesystem to manage the transit without i_mutex
1129 * We indicate to the filesystem if someone is trying to mount
1130 * something here. This gives autofs the chance to deny anyone
1131 * other than its daemon the right to mount on its
1134 * The filesystem may sleep at this point.
1136 if (managed & DCACHE_MANAGE_TRANSIT) {
1137 BUG_ON(!path->dentry->d_op);
1138 BUG_ON(!path->dentry->d_op->d_manage);
1139 ret = path->dentry->d_op->d_manage(path->dentry, mounting_here);
1141 return ret == -EISDIR ? 0 : ret;
1144 /* Transit to a mounted filesystem. */
1145 if (managed & DCACHE_MOUNTED) {
1146 struct vfsmount *mounted = lookup_mnt(path);
1151 path->mnt = mounted;
1152 path->dentry = dget(mounted->mnt_root);
1156 /* Don't handle automount points here */
1163 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1165 static void follow_mount(struct path *path)
1167 while (d_mountpoint(path->dentry)) {
1168 struct vfsmount *mounted = lookup_mnt(path);
1173 path->mnt = mounted;
1174 path->dentry = dget(mounted->mnt_root);
1178 static void follow_dotdot(struct nameidata *nd)
1183 struct dentry *old = nd->path.dentry;
1185 if (nd->path.dentry == nd->root.dentry &&
1186 nd->path.mnt == nd->root.mnt) {
1189 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1190 /* rare case of legitimate dget_parent()... */
1191 nd->path.dentry = dget_parent(nd->path.dentry);
1195 if (!follow_up(&nd->path))
1198 follow_mount(&nd->path);
1199 nd->inode = nd->path.dentry->d_inode;
1203 * Allocate a dentry with name and parent, and perform a parent
1204 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1205 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1206 * have verified that no child exists while under i_mutex.
1208 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1209 struct qstr *name, struct nameidata *nd)
1211 struct inode *inode = parent->d_inode;
1212 struct dentry *dentry;
1215 /* Don't create child dentry for a dead directory. */
1216 if (unlikely(IS_DEADDIR(inode)))
1217 return ERR_PTR(-ENOENT);
1219 dentry = d_alloc(parent, name);
1220 if (unlikely(!dentry))
1221 return ERR_PTR(-ENOMEM);
1223 old = inode->i_op->lookup(inode, dentry, nd);
1224 if (unlikely(old)) {
1232 * It's more convoluted than I'd like it to be, but... it's still fairly
1233 * small and for now I'd prefer to have fast path as straight as possible.
1234 * It _is_ time-critical.
1236 static int do_lookup(struct nameidata *nd, struct qstr *name,
1237 struct path *path, struct inode **inode)
1239 struct vfsmount *mnt = nd->path.mnt;
1240 struct dentry *dentry, *parent = nd->path.dentry;
1245 * See if the low-level filesystem might want
1246 * to use its own hash..
1248 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1249 err = parent->d_op->d_hash(parent, nd->inode, name);
1255 * Rename seqlock is not required here because in the off chance
1256 * of a false negative due to a concurrent rename, we're going to
1257 * do the non-racy lookup, below.
1259 if (nd->flags & LOOKUP_RCU) {
1263 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1265 if (nameidata_drop_rcu(nd))
1269 /* Memory barrier in read_seqcount_begin of child is enough */
1270 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1274 if (dentry->d_flags & DCACHE_OP_REVALIDATE)
1275 goto need_revalidate;
1278 path->dentry = dentry;
1279 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1281 if (nameidata_drop_rcu(nd))
1285 dentry = __d_lookup(parent, name);
1289 if (dentry->d_flags & DCACHE_OP_REVALIDATE)
1290 goto need_revalidate;
1293 path->dentry = dentry;
1294 err = follow_managed(path, nd->flags);
1295 if (unlikely(err < 0))
1297 *inode = path->dentry->d_inode;
1301 dir = parent->d_inode;
1302 BUG_ON(nd->inode != dir);
1304 mutex_lock(&dir->i_mutex);
1306 * First re-do the cached lookup just in case it was created
1307 * while we waited for the directory semaphore, or the first
1308 * lookup failed due to an unrelated rename.
1310 * This could use version numbering or similar to avoid unnecessary
1311 * cache lookups, but then we'd have to do the first lookup in the
1312 * non-racy way. However in the common case here, everything should
1313 * be hot in cache, so would it be a big win?
1315 dentry = d_lookup(parent, name);
1316 if (likely(!dentry)) {
1317 dentry = d_alloc_and_lookup(parent, name, nd);
1318 mutex_unlock(&dir->i_mutex);
1324 * Uhhuh! Nasty case: the cache was re-populated while
1325 * we waited on the semaphore. Need to revalidate.
1327 mutex_unlock(&dir->i_mutex);
1331 dentry = do_revalidate(dentry, nd);
1336 if (nd->flags & LOOKUP_RCU)
1341 return PTR_ERR(dentry);
1345 * This is a temporary kludge to deal with "automount" symlinks; proper
1346 * solution is to trigger them on follow_mount(), so that do_lookup()
1347 * would DTRT. To be killed before 2.6.34-final.
1349 static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
1351 return inode && unlikely(inode->i_op->follow_link) &&
1352 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
1357 * This is the basic name resolution function, turning a pathname into
1358 * the final dentry. We expect 'base' to be positive and a directory.
1360 * Returns 0 and nd will have valid dentry and mnt on success.
1361 * Returns error and drops reference to input namei data on failure.
1363 static int link_path_walk(const char *name, struct nameidata *nd)
1367 unsigned int lookup_flags = nd->flags;
1375 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1377 /* At this point we know we have a real path component. */
1379 struct inode *inode;
1384 nd->flags |= LOOKUP_CONTINUE;
1385 if (nd->flags & LOOKUP_RCU) {
1386 err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1387 if (err == -ECHILD) {
1388 if (nameidata_drop_rcu(nd))
1394 err = exec_permission(nd->inode, 0);
1400 c = *(const unsigned char *)name;
1402 hash = init_name_hash();
1405 hash = partial_name_hash(c, hash);
1406 c = *(const unsigned char *)name;
1407 } while (c && (c != '/'));
1408 this.len = name - (const char *) this.name;
1409 this.hash = end_name_hash(hash);
1411 /* remove trailing slashes? */
1413 goto last_component;
1414 while (*++name == '/');
1416 goto last_with_slashes;
1419 * "." and ".." are special - ".." especially so because it has
1420 * to be able to know about the current root directory and
1421 * parent relationships.
1423 if (this.name[0] == '.') switch (this.len) {
1427 if (this.name[1] != '.')
1429 if (nd->flags & LOOKUP_RCU) {
1430 if (follow_dotdot_rcu(nd))
1438 /* This does the actual lookups.. */
1439 err = do_lookup(nd, &this, &next, &inode);
1446 if (inode->i_op->follow_link) {
1447 /* We commonly drop rcu-walk here */
1448 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1450 BUG_ON(inode != next.dentry->d_inode);
1451 err = do_follow_link(&next, nd);
1454 nd->inode = nd->path.dentry->d_inode;
1459 path_to_nameidata(&next, nd);
1463 if (!nd->inode->i_op->lookup)
1466 /* here ends the main loop */
1469 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1471 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1472 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1473 if (lookup_flags & LOOKUP_PARENT)
1475 if (this.name[0] == '.') switch (this.len) {
1479 if (this.name[1] != '.')
1481 if (nd->flags & LOOKUP_RCU) {
1482 if (follow_dotdot_rcu(nd))
1490 err = do_lookup(nd, &this, &next, &inode);
1493 if (follow_on_final(inode, lookup_flags)) {
1494 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1496 BUG_ON(inode != next.dentry->d_inode);
1497 err = do_follow_link(&next, nd);
1500 nd->inode = nd->path.dentry->d_inode;
1502 path_to_nameidata(&next, nd);
1508 if (lookup_flags & LOOKUP_DIRECTORY) {
1510 if (!nd->inode->i_op->lookup)
1516 nd->last_type = LAST_NORM;
1517 if (this.name[0] != '.')
1520 nd->last_type = LAST_DOT;
1521 else if (this.len == 2 && this.name[1] == '.')
1522 nd->last_type = LAST_DOTDOT;
1527 * We bypassed the ordinary revalidation routines.
1528 * We may need to check the cached dentry for staleness.
1530 if (need_reval_dot(nd->path.dentry)) {
1531 /* Note: we do not d_invalidate() */
1532 err = d_revalidate(nd->path.dentry, nd);
1539 if (nameidata_drop_rcu_last_maybe(nd))
1543 if (!(nd->flags & LOOKUP_RCU))
1544 path_put_conditional(&next, nd);
1547 if (!(nd->flags & LOOKUP_RCU))
1548 path_put(&nd->path);
1553 static inline int path_walk_rcu(const char *name, struct nameidata *nd)
1555 current->total_link_count = 0;
1557 return link_path_walk(name, nd);
1560 static inline int path_walk_simple(const char *name, struct nameidata *nd)
1562 current->total_link_count = 0;
1564 return link_path_walk(name, nd);
1567 static int path_walk(const char *name, struct nameidata *nd)
1569 struct path save = nd->path;
1572 current->total_link_count = 0;
1574 /* make sure the stuff we saved doesn't go away */
1577 result = link_path_walk(name, nd);
1578 if (result == -ESTALE) {
1579 /* nd->path had been dropped */
1580 current->total_link_count = 0;
1582 path_get(&nd->path);
1583 nd->flags |= LOOKUP_REVAL;
1584 result = link_path_walk(name, nd);
1592 static void path_finish_rcu(struct nameidata *nd)
1594 if (nd->flags & LOOKUP_RCU) {
1595 /* RCU dangling. Cancel it. */
1596 nd->flags &= ~LOOKUP_RCU;
1597 nd->root.mnt = NULL;
1599 br_read_unlock(vfsmount_lock);
1605 static int path_init_rcu(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1611 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1612 nd->flags = flags | LOOKUP_RCU;
1614 nd->root.mnt = NULL;
1618 struct fs_struct *fs = current->fs;
1621 br_read_lock(vfsmount_lock);
1625 seq = read_seqcount_begin(&fs->seq);
1626 nd->root = fs->root;
1627 nd->path = nd->root;
1628 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1629 } while (read_seqcount_retry(&fs->seq, seq));
1631 } else if (dfd == AT_FDCWD) {
1632 struct fs_struct *fs = current->fs;
1635 br_read_lock(vfsmount_lock);
1639 seq = read_seqcount_begin(&fs->seq);
1641 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1642 } while (read_seqcount_retry(&fs->seq, seq));
1645 struct dentry *dentry;
1647 file = fget_light(dfd, &fput_needed);
1652 dentry = file->f_path.dentry;
1655 if (!S_ISDIR(dentry->d_inode->i_mode))
1658 retval = file_permission(file, MAY_EXEC);
1662 nd->path = file->f_path;
1666 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1667 br_read_lock(vfsmount_lock);
1670 nd->inode = nd->path.dentry->d_inode;
1674 fput_light(file, fput_needed);
1679 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1685 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1688 nd->root.mnt = NULL;
1692 nd->path = nd->root;
1693 path_get(&nd->root);
1694 } else if (dfd == AT_FDCWD) {
1695 get_fs_pwd(current->fs, &nd->path);
1697 struct dentry *dentry;
1699 file = fget_light(dfd, &fput_needed);
1704 dentry = file->f_path.dentry;
1707 if (!S_ISDIR(dentry->d_inode->i_mode))
1710 retval = file_permission(file, MAY_EXEC);
1714 nd->path = file->f_path;
1715 path_get(&file->f_path);
1717 fput_light(file, fput_needed);
1719 nd->inode = nd->path.dentry->d_inode;
1723 fput_light(file, fput_needed);
1728 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1729 static int do_path_lookup(int dfd, const char *name,
1730 unsigned int flags, struct nameidata *nd)
1735 * Path walking is largely split up into 2 different synchronisation
1736 * schemes, rcu-walk and ref-walk (explained in
1737 * Documentation/filesystems/path-lookup.txt). These share much of the
1738 * path walk code, but some things particularly setup, cleanup, and
1739 * following mounts are sufficiently divergent that functions are
1740 * duplicated. Typically there is a function foo(), and its RCU
1741 * analogue, foo_rcu().
1743 * -ECHILD is the error number of choice (just to avoid clashes) that
1744 * is returned if some aspect of an rcu-walk fails. Such an error must
1745 * be handled by restarting a traditional ref-walk (which will always
1746 * be able to complete).
1748 retval = path_init_rcu(dfd, name, flags, nd);
1749 if (unlikely(retval))
1751 retval = path_walk_rcu(name, nd);
1752 path_finish_rcu(nd);
1754 path_put(&nd->root);
1755 nd->root.mnt = NULL;
1758 if (unlikely(retval == -ECHILD || retval == -ESTALE)) {
1759 /* slower, locked walk */
1760 if (retval == -ESTALE)
1761 flags |= LOOKUP_REVAL;
1762 retval = path_init(dfd, name, flags, nd);
1763 if (unlikely(retval))
1765 retval = path_walk(name, nd);
1767 path_put(&nd->root);
1768 nd->root.mnt = NULL;
1772 if (likely(!retval)) {
1773 if (unlikely(!audit_dummy_context())) {
1774 if (nd->path.dentry && nd->inode)
1775 audit_inode(name, nd->path.dentry);
1782 int path_lookup(const char *name, unsigned int flags,
1783 struct nameidata *nd)
1785 return do_path_lookup(AT_FDCWD, name, flags, nd);
1788 int kern_path(const char *name, unsigned int flags, struct path *path)
1790 struct nameidata nd;
1791 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1798 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1799 * @dentry: pointer to dentry of the base directory
1800 * @mnt: pointer to vfs mount of the base directory
1801 * @name: pointer to file name
1802 * @flags: lookup flags
1803 * @nd: pointer to nameidata
1805 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1806 const char *name, unsigned int flags,
1807 struct nameidata *nd)
1811 /* same as do_path_lookup */
1812 nd->last_type = LAST_ROOT;
1816 nd->path.dentry = dentry;
1818 path_get(&nd->path);
1819 nd->root = nd->path;
1820 path_get(&nd->root);
1821 nd->inode = nd->path.dentry->d_inode;
1823 retval = path_walk(name, nd);
1824 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1826 audit_inode(name, nd->path.dentry);
1828 path_put(&nd->root);
1829 nd->root.mnt = NULL;
1834 static struct dentry *__lookup_hash(struct qstr *name,
1835 struct dentry *base, struct nameidata *nd)
1837 struct inode *inode = base->d_inode;
1838 struct dentry *dentry;
1841 err = exec_permission(inode, 0);
1843 return ERR_PTR(err);
1846 * See if the low-level filesystem might want
1847 * to use its own hash..
1849 if (base->d_flags & DCACHE_OP_HASH) {
1850 err = base->d_op->d_hash(base, inode, name);
1851 dentry = ERR_PTR(err);
1857 * Don't bother with __d_lookup: callers are for creat as
1858 * well as unlink, so a lot of the time it would cost
1861 dentry = d_lookup(base, name);
1863 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
1864 dentry = do_revalidate(dentry, nd);
1867 dentry = d_alloc_and_lookup(base, name, nd);
1873 * Restricted form of lookup. Doesn't follow links, single-component only,
1874 * needs parent already locked. Doesn't follow mounts.
1877 static struct dentry *lookup_hash(struct nameidata *nd)
1879 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1882 static int __lookup_one_len(const char *name, struct qstr *this,
1883 struct dentry *base, int len)
1893 hash = init_name_hash();
1895 c = *(const unsigned char *)name++;
1896 if (c == '/' || c == '\0')
1898 hash = partial_name_hash(c, hash);
1900 this->hash = end_name_hash(hash);
1905 * lookup_one_len - filesystem helper to lookup single pathname component
1906 * @name: pathname component to lookup
1907 * @base: base directory to lookup from
1908 * @len: maximum length @len should be interpreted to
1910 * Note that this routine is purely a helper for filesystem usage and should
1911 * not be called by generic code. Also note that by using this function the
1912 * nameidata argument is passed to the filesystem methods and a filesystem
1913 * using this helper needs to be prepared for that.
1915 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1920 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1922 err = __lookup_one_len(name, &this, base, len);
1924 return ERR_PTR(err);
1926 return __lookup_hash(&this, base, NULL);
1929 int user_path_at(int dfd, const char __user *name, unsigned flags,
1932 struct nameidata nd;
1933 char *tmp = getname(name);
1934 int err = PTR_ERR(tmp);
1937 BUG_ON(flags & LOOKUP_PARENT);
1939 err = do_path_lookup(dfd, tmp, flags, &nd);
1947 static int user_path_parent(int dfd, const char __user *path,
1948 struct nameidata *nd, char **name)
1950 char *s = getname(path);
1956 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1966 * It's inline, so penalty for filesystems that don't use sticky bit is
1969 static inline int check_sticky(struct inode *dir, struct inode *inode)
1971 uid_t fsuid = current_fsuid();
1973 if (!(dir->i_mode & S_ISVTX))
1975 if (inode->i_uid == fsuid)
1977 if (dir->i_uid == fsuid)
1979 return !capable(CAP_FOWNER);
1983 * Check whether we can remove a link victim from directory dir, check
1984 * whether the type of victim is right.
1985 * 1. We can't do it if dir is read-only (done in permission())
1986 * 2. We should have write and exec permissions on dir
1987 * 3. We can't remove anything from append-only dir
1988 * 4. We can't do anything with immutable dir (done in permission())
1989 * 5. If the sticky bit on dir is set we should either
1990 * a. be owner of dir, or
1991 * b. be owner of victim, or
1992 * c. have CAP_FOWNER capability
1993 * 6. If the victim is append-only or immutable we can't do antyhing with
1994 * links pointing to it.
1995 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1996 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1997 * 9. We can't remove a root or mountpoint.
1998 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1999 * nfs_async_unlink().
2001 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2005 if (!victim->d_inode)
2008 BUG_ON(victim->d_parent->d_inode != dir);
2009 audit_inode_child(victim, dir);
2011 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2016 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2017 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2020 if (!S_ISDIR(victim->d_inode->i_mode))
2022 if (IS_ROOT(victim))
2024 } else if (S_ISDIR(victim->d_inode->i_mode))
2026 if (IS_DEADDIR(dir))
2028 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2033 /* Check whether we can create an object with dentry child in directory
2035 * 1. We can't do it if child already exists (open has special treatment for
2036 * this case, but since we are inlined it's OK)
2037 * 2. We can't do it if dir is read-only (done in permission())
2038 * 3. We should have write and exec permissions on dir
2039 * 4. We can't do it if dir is immutable (done in permission())
2041 static inline int may_create(struct inode *dir, struct dentry *child)
2045 if (IS_DEADDIR(dir))
2047 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2051 * p1 and p2 should be directories on the same fs.
2053 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2058 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2062 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2064 p = d_ancestor(p2, p1);
2066 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2067 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2071 p = d_ancestor(p1, p2);
2073 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2074 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2078 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2079 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2083 void unlock_rename(struct dentry *p1, struct dentry *p2)
2085 mutex_unlock(&p1->d_inode->i_mutex);
2087 mutex_unlock(&p2->d_inode->i_mutex);
2088 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2092 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
2093 struct nameidata *nd)
2095 int error = may_create(dir, dentry);
2100 if (!dir->i_op->create)
2101 return -EACCES; /* shouldn't it be ENOSYS? */
2104 error = security_inode_create(dir, dentry, mode);
2107 error = dir->i_op->create(dir, dentry, mode, nd);
2109 fsnotify_create(dir, dentry);
2113 int may_open(struct path *path, int acc_mode, int flag)
2115 struct dentry *dentry = path->dentry;
2116 struct inode *inode = dentry->d_inode;
2122 switch (inode->i_mode & S_IFMT) {
2126 if (acc_mode & MAY_WRITE)
2131 if (path->mnt->mnt_flags & MNT_NODEV)
2140 error = inode_permission(inode, acc_mode);
2145 * An append-only file must be opened in append mode for writing.
2147 if (IS_APPEND(inode)) {
2148 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2154 /* O_NOATIME can only be set by the owner or superuser */
2155 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2159 * Ensure there are no outstanding leases on the file.
2161 return break_lease(inode, flag);
2164 static int handle_truncate(struct file *filp)
2166 struct path *path = &filp->f_path;
2167 struct inode *inode = path->dentry->d_inode;
2168 int error = get_write_access(inode);
2172 * Refuse to truncate files with mandatory locks held on them.
2174 error = locks_verify_locked(inode);
2176 error = security_path_truncate(path);
2178 error = do_truncate(path->dentry, 0,
2179 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2182 put_write_access(inode);
2187 * Be careful about ever adding any more callers of this
2188 * function. Its flags must be in the namei format, not
2189 * what get passed to sys_open().
2191 static int __open_namei_create(struct nameidata *nd, struct path *path,
2192 int open_flag, int mode)
2195 struct dentry *dir = nd->path.dentry;
2197 if (!IS_POSIXACL(dir->d_inode))
2198 mode &= ~current_umask();
2199 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
2202 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
2204 mutex_unlock(&dir->d_inode->i_mutex);
2205 dput(nd->path.dentry);
2206 nd->path.dentry = path->dentry;
2210 /* Don't check for write permission, don't truncate */
2211 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
2215 * Note that while the flag value (low two bits) for sys_open means:
2220 * it is changed into
2221 * 00 - no permissions needed
2222 * 01 - read-permission
2223 * 10 - write-permission
2225 * for the internal routines (ie open_namei()/follow_link() etc)
2226 * This is more logical, and also allows the 00 "no perm needed"
2227 * to be used for symlinks (where the permissions are checked
2231 static inline int open_to_namei_flags(int flag)
2233 if ((flag+1) & O_ACCMODE)
2238 static int open_will_truncate(int flag, struct inode *inode)
2241 * We'll never write to the fs underlying
2244 if (special_file(inode->i_mode))
2246 return (flag & O_TRUNC);
2249 static struct file *finish_open(struct nameidata *nd,
2250 int open_flag, int acc_mode)
2256 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
2257 if (will_truncate) {
2258 error = mnt_want_write(nd->path.mnt);
2262 error = may_open(&nd->path, acc_mode, open_flag);
2265 mnt_drop_write(nd->path.mnt);
2268 filp = nameidata_to_filp(nd);
2269 if (!IS_ERR(filp)) {
2270 error = ima_file_check(filp, acc_mode);
2273 filp = ERR_PTR(error);
2276 if (!IS_ERR(filp)) {
2277 if (will_truncate) {
2278 error = handle_truncate(filp);
2281 filp = ERR_PTR(error);
2286 * It is now safe to drop the mnt write
2287 * because the filp has had a write taken
2291 mnt_drop_write(nd->path.mnt);
2292 path_put(&nd->path);
2296 if (!IS_ERR(nd->intent.open.file))
2297 release_open_intent(nd);
2298 path_put(&nd->path);
2299 return ERR_PTR(error);
2303 * Handle O_CREAT case for do_filp_open
2305 static struct file *do_last(struct nameidata *nd, struct path *path,
2306 int open_flag, int acc_mode,
2307 int mode, const char *pathname)
2309 struct dentry *dir = nd->path.dentry;
2311 int error = -EISDIR;
2313 switch (nd->last_type) {
2316 dir = nd->path.dentry;
2318 if (need_reval_dot(dir)) {
2319 int status = d_revalidate(nd->path.dentry, nd);
2331 audit_inode(pathname, dir);
2335 /* trailing slashes? */
2336 if (nd->last.name[nd->last.len])
2339 mutex_lock(&dir->d_inode->i_mutex);
2341 path->dentry = lookup_hash(nd);
2342 path->mnt = nd->path.mnt;
2344 error = PTR_ERR(path->dentry);
2345 if (IS_ERR(path->dentry)) {
2346 mutex_unlock(&dir->d_inode->i_mutex);
2350 if (IS_ERR(nd->intent.open.file)) {
2351 error = PTR_ERR(nd->intent.open.file);
2352 goto exit_mutex_unlock;
2355 /* Negative dentry, just create the file */
2356 if (!path->dentry->d_inode) {
2358 * This write is needed to ensure that a
2359 * ro->rw transition does not occur between
2360 * the time when the file is created and when
2361 * a permanent write count is taken through
2362 * the 'struct file' in nameidata_to_filp().
2364 error = mnt_want_write(nd->path.mnt);
2366 goto exit_mutex_unlock;
2367 error = __open_namei_create(nd, path, open_flag, mode);
2369 mnt_drop_write(nd->path.mnt);
2372 filp = nameidata_to_filp(nd);
2373 mnt_drop_write(nd->path.mnt);
2374 path_put(&nd->path);
2375 if (!IS_ERR(filp)) {
2376 error = ima_file_check(filp, acc_mode);
2379 filp = ERR_PTR(error);
2386 * It already exists.
2388 mutex_unlock(&dir->d_inode->i_mutex);
2389 audit_inode(pathname, path->dentry);
2392 if (open_flag & O_EXCL)
2395 error = follow_managed(path, nd->flags);
2400 if (!path->dentry->d_inode)
2403 if (path->dentry->d_inode->i_op->follow_link)
2406 path_to_nameidata(path, nd);
2407 nd->inode = path->dentry->d_inode;
2409 if (S_ISDIR(nd->inode->i_mode))
2412 filp = finish_open(nd, open_flag, acc_mode);
2416 mutex_unlock(&dir->d_inode->i_mutex);
2418 path_put_conditional(path, nd);
2420 if (!IS_ERR(nd->intent.open.file))
2421 release_open_intent(nd);
2422 path_put(&nd->path);
2423 return ERR_PTR(error);
2427 * Note that the low bits of the passed in "open_flag"
2428 * are not the same as in the local variable "flag". See
2429 * open_to_namei_flags() for more details.
2431 struct file *do_filp_open(int dfd, const char *pathname,
2432 int open_flag, int mode, int acc_mode)
2435 struct nameidata nd;
2439 int flag = open_to_namei_flags(open_flag);
2442 if (!(open_flag & O_CREAT))
2445 /* Must never be set by userspace */
2446 open_flag &= ~FMODE_NONOTIFY;
2449 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2450 * check for O_DSYNC if the need any syncing at all we enforce it's
2451 * always set instead of having to deal with possibly weird behaviour
2452 * for malicious applications setting only __O_SYNC.
2454 if (open_flag & __O_SYNC)
2455 open_flag |= O_DSYNC;
2458 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
2460 /* O_TRUNC implies we need access checks for write permissions */
2461 if (open_flag & O_TRUNC)
2462 acc_mode |= MAY_WRITE;
2464 /* Allow the LSM permission hook to distinguish append
2465 access from general write access. */
2466 if (open_flag & O_APPEND)
2467 acc_mode |= MAY_APPEND;
2469 flags = LOOKUP_OPEN;
2470 if (open_flag & O_CREAT) {
2471 flags |= LOOKUP_CREATE;
2472 if (open_flag & O_EXCL)
2473 flags |= LOOKUP_EXCL;
2475 if (open_flag & O_DIRECTORY)
2476 flags |= LOOKUP_DIRECTORY;
2477 if (!(open_flag & O_NOFOLLOW))
2478 flags |= LOOKUP_FOLLOW;
2480 filp = get_empty_filp();
2482 return ERR_PTR(-ENFILE);
2484 filp->f_flags = open_flag;
2485 nd.intent.open.file = filp;
2486 nd.intent.open.flags = flag;
2487 nd.intent.open.create_mode = mode;
2489 if (open_flag & O_CREAT)
2492 /* !O_CREAT, simple open */
2493 error = do_path_lookup(dfd, pathname, flags, &nd);
2494 if (unlikely(error))
2497 if (!(nd.flags & LOOKUP_FOLLOW)) {
2498 if (nd.inode->i_op->follow_link)
2502 if (nd.flags & LOOKUP_DIRECTORY) {
2503 if (!nd.inode->i_op->lookup)
2506 audit_inode(pathname, nd.path.dentry);
2507 filp = finish_open(&nd, open_flag, acc_mode);
2511 /* OK, have to create the file. Find the parent. */
2512 error = path_init_rcu(dfd, pathname,
2513 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2516 error = path_walk_rcu(pathname, &nd);
2517 path_finish_rcu(&nd);
2518 if (unlikely(error == -ECHILD || error == -ESTALE)) {
2519 /* slower, locked walk */
2520 if (error == -ESTALE) {
2522 flags |= LOOKUP_REVAL;
2524 error = path_init(dfd, pathname,
2525 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2529 error = path_walk_simple(pathname, &nd);
2531 if (unlikely(error))
2533 if (unlikely(!audit_dummy_context()))
2534 audit_inode(pathname, nd.path.dentry);
2537 * We have the parent and last component.
2540 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
2541 while (unlikely(!filp)) { /* trailing symlink */
2542 struct path link = path;
2543 struct inode *linki = link.dentry->d_inode;
2546 /* S_ISDIR part is a temporary automount kludge */
2547 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(linki->i_mode))
2552 * This is subtle. Instead of calling do_follow_link() we do
2553 * the thing by hands. The reason is that this way we have zero
2554 * link_count and path_walk() (called from ->follow_link)
2555 * honoring LOOKUP_PARENT. After that we have the parent and
2556 * last component, i.e. we are in the same situation as after
2557 * the first path_walk(). Well, almost - if the last component
2558 * is normal we get its copy stored in nd->last.name and we will
2559 * have to putname() it when we are done. Procfs-like symlinks
2560 * just set LAST_BIND.
2562 nd.flags |= LOOKUP_PARENT;
2563 error = security_inode_follow_link(link.dentry, &nd);
2566 error = __do_follow_link(&link, &nd, &cookie);
2567 if (unlikely(error)) {
2568 if (!IS_ERR(cookie) && linki->i_op->put_link)
2569 linki->i_op->put_link(link.dentry, &nd, cookie);
2570 /* nd.path had been dropped */
2574 nd.flags &= ~LOOKUP_PARENT;
2575 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
2576 if (linki->i_op->put_link)
2577 linki->i_op->put_link(link.dentry, &nd, cookie);
2583 if (filp == ERR_PTR(-ESTALE) && !(flags & LOOKUP_REVAL))
2588 path_put_conditional(&path, &nd);
2592 if (!IS_ERR(nd.intent.open.file))
2593 release_open_intent(&nd);
2594 filp = ERR_PTR(error);
2599 * filp_open - open file and return file pointer
2601 * @filename: path to open
2602 * @flags: open flags as per the open(2) second argument
2603 * @mode: mode for the new file if O_CREAT is set, else ignored
2605 * This is the helper to open a file from kernelspace if you really
2606 * have to. But in generally you should not do this, so please move
2607 * along, nothing to see here..
2609 struct file *filp_open(const char *filename, int flags, int mode)
2611 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
2613 EXPORT_SYMBOL(filp_open);
2616 * lookup_create - lookup a dentry, creating it if it doesn't exist
2617 * @nd: nameidata info
2618 * @is_dir: directory flag
2620 * Simple function to lookup and return a dentry and create it
2621 * if it doesn't exist. Is SMP-safe.
2623 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2625 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2627 struct dentry *dentry = ERR_PTR(-EEXIST);
2629 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2631 * Yucky last component or no last component at all?
2632 * (foo/., foo/.., /////)
2634 if (nd->last_type != LAST_NORM)
2636 nd->flags &= ~LOOKUP_PARENT;
2637 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2638 nd->intent.open.flags = O_EXCL;
2641 * Do the final lookup.
2643 dentry = lookup_hash(nd);
2647 if (dentry->d_inode)
2650 * Special case - lookup gave negative, but... we had foo/bar/
2651 * From the vfs_mknod() POV we just have a negative dentry -
2652 * all is fine. Let's be bastards - you had / on the end, you've
2653 * been asking for (non-existent) directory. -ENOENT for you.
2655 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2657 dentry = ERR_PTR(-ENOENT);
2662 dentry = ERR_PTR(-EEXIST);
2666 EXPORT_SYMBOL_GPL(lookup_create);
2668 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2670 int error = may_create(dir, dentry);
2675 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2678 if (!dir->i_op->mknod)
2681 error = devcgroup_inode_mknod(mode, dev);
2685 error = security_inode_mknod(dir, dentry, mode, dev);
2689 error = dir->i_op->mknod(dir, dentry, mode, dev);
2691 fsnotify_create(dir, dentry);
2695 static int may_mknod(mode_t mode)
2697 switch (mode & S_IFMT) {
2703 case 0: /* zero mode translates to S_IFREG */
2712 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2717 struct dentry *dentry;
2718 struct nameidata nd;
2723 error = user_path_parent(dfd, filename, &nd, &tmp);
2727 dentry = lookup_create(&nd, 0);
2728 if (IS_ERR(dentry)) {
2729 error = PTR_ERR(dentry);
2732 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2733 mode &= ~current_umask();
2734 error = may_mknod(mode);
2737 error = mnt_want_write(nd.path.mnt);
2740 error = security_path_mknod(&nd.path, dentry, mode, dev);
2742 goto out_drop_write;
2743 switch (mode & S_IFMT) {
2744 case 0: case S_IFREG:
2745 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2747 case S_IFCHR: case S_IFBLK:
2748 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2749 new_decode_dev(dev));
2751 case S_IFIFO: case S_IFSOCK:
2752 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2756 mnt_drop_write(nd.path.mnt);
2760 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2767 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2769 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2772 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2774 int error = may_create(dir, dentry);
2779 if (!dir->i_op->mkdir)
2782 mode &= (S_IRWXUGO|S_ISVTX);
2783 error = security_inode_mkdir(dir, dentry, mode);
2787 error = dir->i_op->mkdir(dir, dentry, mode);
2789 fsnotify_mkdir(dir, dentry);
2793 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2797 struct dentry *dentry;
2798 struct nameidata nd;
2800 error = user_path_parent(dfd, pathname, &nd, &tmp);
2804 dentry = lookup_create(&nd, 1);
2805 error = PTR_ERR(dentry);
2809 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2810 mode &= ~current_umask();
2811 error = mnt_want_write(nd.path.mnt);
2814 error = security_path_mkdir(&nd.path, dentry, mode);
2816 goto out_drop_write;
2817 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2819 mnt_drop_write(nd.path.mnt);
2823 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2830 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2832 return sys_mkdirat(AT_FDCWD, pathname, mode);
2836 * We try to drop the dentry early: we should have
2837 * a usage count of 2 if we're the only user of this
2838 * dentry, and if that is true (possibly after pruning
2839 * the dcache), then we drop the dentry now.
2841 * A low-level filesystem can, if it choses, legally
2844 * if (!d_unhashed(dentry))
2847 * if it cannot handle the case of removing a directory
2848 * that is still in use by something else..
2850 void dentry_unhash(struct dentry *dentry)
2853 shrink_dcache_parent(dentry);
2854 spin_lock(&dentry->d_lock);
2855 if (dentry->d_count == 2)
2857 spin_unlock(&dentry->d_lock);
2860 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2862 int error = may_delete(dir, dentry, 1);
2867 if (!dir->i_op->rmdir)
2870 mutex_lock(&dentry->d_inode->i_mutex);
2871 dentry_unhash(dentry);
2872 if (d_mountpoint(dentry))
2875 error = security_inode_rmdir(dir, dentry);
2877 error = dir->i_op->rmdir(dir, dentry);
2879 dentry->d_inode->i_flags |= S_DEAD;
2884 mutex_unlock(&dentry->d_inode->i_mutex);
2893 static long do_rmdir(int dfd, const char __user *pathname)
2897 struct dentry *dentry;
2898 struct nameidata nd;
2900 error = user_path_parent(dfd, pathname, &nd, &name);
2904 switch(nd.last_type) {
2916 nd.flags &= ~LOOKUP_PARENT;
2918 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2919 dentry = lookup_hash(&nd);
2920 error = PTR_ERR(dentry);
2923 error = mnt_want_write(nd.path.mnt);
2926 error = security_path_rmdir(&nd.path, dentry);
2929 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2931 mnt_drop_write(nd.path.mnt);
2935 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2942 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2944 return do_rmdir(AT_FDCWD, pathname);
2947 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2949 int error = may_delete(dir, dentry, 0);
2954 if (!dir->i_op->unlink)
2957 mutex_lock(&dentry->d_inode->i_mutex);
2958 if (d_mountpoint(dentry))
2961 error = security_inode_unlink(dir, dentry);
2963 error = dir->i_op->unlink(dir, dentry);
2968 mutex_unlock(&dentry->d_inode->i_mutex);
2970 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2971 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2972 fsnotify_link_count(dentry->d_inode);
2980 * Make sure that the actual truncation of the file will occur outside its
2981 * directory's i_mutex. Truncate can take a long time if there is a lot of
2982 * writeout happening, and we don't want to prevent access to the directory
2983 * while waiting on the I/O.
2985 static long do_unlinkat(int dfd, const char __user *pathname)
2989 struct dentry *dentry;
2990 struct nameidata nd;
2991 struct inode *inode = NULL;
2993 error = user_path_parent(dfd, pathname, &nd, &name);
2998 if (nd.last_type != LAST_NORM)
3001 nd.flags &= ~LOOKUP_PARENT;
3003 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3004 dentry = lookup_hash(&nd);
3005 error = PTR_ERR(dentry);
3006 if (!IS_ERR(dentry)) {
3007 /* Why not before? Because we want correct error value */
3008 if (nd.last.name[nd.last.len])
3010 inode = dentry->d_inode;
3013 error = mnt_want_write(nd.path.mnt);
3016 error = security_path_unlink(&nd.path, dentry);
3019 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
3021 mnt_drop_write(nd.path.mnt);
3025 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3027 iput(inode); /* truncate the inode here */
3034 error = !dentry->d_inode ? -ENOENT :
3035 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3039 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3041 if ((flag & ~AT_REMOVEDIR) != 0)
3044 if (flag & AT_REMOVEDIR)
3045 return do_rmdir(dfd, pathname);
3047 return do_unlinkat(dfd, pathname);
3050 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3052 return do_unlinkat(AT_FDCWD, pathname);
3055 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3057 int error = may_create(dir, dentry);
3062 if (!dir->i_op->symlink)
3065 error = security_inode_symlink(dir, dentry, oldname);
3069 error = dir->i_op->symlink(dir, dentry, oldname);
3071 fsnotify_create(dir, dentry);
3075 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3076 int, newdfd, const char __user *, newname)
3081 struct dentry *dentry;
3082 struct nameidata nd;
3084 from = getname(oldname);
3086 return PTR_ERR(from);
3088 error = user_path_parent(newdfd, newname, &nd, &to);
3092 dentry = lookup_create(&nd, 0);
3093 error = PTR_ERR(dentry);
3097 error = mnt_want_write(nd.path.mnt);
3100 error = security_path_symlink(&nd.path, dentry, from);
3102 goto out_drop_write;
3103 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
3105 mnt_drop_write(nd.path.mnt);
3109 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3117 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3119 return sys_symlinkat(oldname, AT_FDCWD, newname);
3122 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3124 struct inode *inode = old_dentry->d_inode;
3130 error = may_create(dir, new_dentry);
3134 if (dir->i_sb != inode->i_sb)
3138 * A link to an append-only or immutable file cannot be created.
3140 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3142 if (!dir->i_op->link)
3144 if (S_ISDIR(inode->i_mode))
3147 error = security_inode_link(old_dentry, dir, new_dentry);
3151 mutex_lock(&inode->i_mutex);
3152 error = dir->i_op->link(old_dentry, dir, new_dentry);
3153 mutex_unlock(&inode->i_mutex);
3155 fsnotify_link(dir, inode, new_dentry);
3160 * Hardlinks are often used in delicate situations. We avoid
3161 * security-related surprises by not following symlinks on the
3164 * We don't follow them on the oldname either to be compatible
3165 * with linux 2.0, and to avoid hard-linking to directories
3166 * and other special files. --ADM
3168 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3169 int, newdfd, const char __user *, newname, int, flags)
3171 struct dentry *new_dentry;
3172 struct nameidata nd;
3173 struct path old_path;
3177 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
3180 error = user_path_at(olddfd, oldname,
3181 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
3186 error = user_path_parent(newdfd, newname, &nd, &to);
3190 if (old_path.mnt != nd.path.mnt)
3192 new_dentry = lookup_create(&nd, 0);
3193 error = PTR_ERR(new_dentry);
3194 if (IS_ERR(new_dentry))
3196 error = mnt_want_write(nd.path.mnt);
3199 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
3201 goto out_drop_write;
3202 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
3204 mnt_drop_write(nd.path.mnt);
3208 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3213 path_put(&old_path);
3218 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3220 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3224 * The worst of all namespace operations - renaming directory. "Perverted"
3225 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3227 * a) we can get into loop creation. Check is done in is_subdir().
3228 * b) race potential - two innocent renames can create a loop together.
3229 * That's where 4.4 screws up. Current fix: serialization on
3230 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3232 * c) we have to lock _three_ objects - parents and victim (if it exists).
3233 * And that - after we got ->i_mutex on parents (until then we don't know
3234 * whether the target exists). Solution: try to be smart with locking
3235 * order for inodes. We rely on the fact that tree topology may change
3236 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3237 * move will be locked. Thus we can rank directories by the tree
3238 * (ancestors first) and rank all non-directories after them.
3239 * That works since everybody except rename does "lock parent, lookup,
3240 * lock child" and rename is under ->s_vfs_rename_mutex.
3241 * HOWEVER, it relies on the assumption that any object with ->lookup()
3242 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3243 * we'd better make sure that there's no link(2) for them.
3244 * d) some filesystems don't support opened-but-unlinked directories,
3245 * either because of layout or because they are not ready to deal with
3246 * all cases correctly. The latter will be fixed (taking this sort of
3247 * stuff into VFS), but the former is not going away. Solution: the same
3248 * trick as in rmdir().
3249 * e) conversion from fhandle to dentry may come in the wrong moment - when
3250 * we are removing the target. Solution: we will have to grab ->i_mutex
3251 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3252 * ->i_mutex on parents, which works but leads to some truly excessive
3255 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3256 struct inode *new_dir, struct dentry *new_dentry)
3259 struct inode *target;
3262 * If we are going to change the parent - check write permissions,
3263 * we'll need to flip '..'.
3265 if (new_dir != old_dir) {
3266 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3271 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3275 target = new_dentry->d_inode;
3277 mutex_lock(&target->i_mutex);
3278 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3282 dentry_unhash(new_dentry);
3283 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3287 target->i_flags |= S_DEAD;
3288 dont_mount(new_dentry);
3290 mutex_unlock(&target->i_mutex);
3291 if (d_unhashed(new_dentry))
3292 d_rehash(new_dentry);
3296 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3297 d_move(old_dentry,new_dentry);
3301 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3302 struct inode *new_dir, struct dentry *new_dentry)
3304 struct inode *target;
3307 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3312 target = new_dentry->d_inode;
3314 mutex_lock(&target->i_mutex);
3315 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3318 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3321 dont_mount(new_dentry);
3322 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3323 d_move(old_dentry, new_dentry);
3326 mutex_unlock(&target->i_mutex);
3331 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3332 struct inode *new_dir, struct dentry *new_dentry)
3335 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3336 const unsigned char *old_name;
3338 if (old_dentry->d_inode == new_dentry->d_inode)
3341 error = may_delete(old_dir, old_dentry, is_dir);
3345 if (!new_dentry->d_inode)
3346 error = may_create(new_dir, new_dentry);
3348 error = may_delete(new_dir, new_dentry, is_dir);
3352 if (!old_dir->i_op->rename)
3355 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3358 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3360 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3362 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3363 new_dentry->d_inode, old_dentry);
3364 fsnotify_oldname_free(old_name);
3369 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3370 int, newdfd, const char __user *, newname)
3372 struct dentry *old_dir, *new_dir;
3373 struct dentry *old_dentry, *new_dentry;
3374 struct dentry *trap;
3375 struct nameidata oldnd, newnd;
3380 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3384 error = user_path_parent(newdfd, newname, &newnd, &to);
3389 if (oldnd.path.mnt != newnd.path.mnt)
3392 old_dir = oldnd.path.dentry;
3394 if (oldnd.last_type != LAST_NORM)
3397 new_dir = newnd.path.dentry;
3398 if (newnd.last_type != LAST_NORM)
3401 oldnd.flags &= ~LOOKUP_PARENT;
3402 newnd.flags &= ~LOOKUP_PARENT;
3403 newnd.flags |= LOOKUP_RENAME_TARGET;
3405 trap = lock_rename(new_dir, old_dir);
3407 old_dentry = lookup_hash(&oldnd);
3408 error = PTR_ERR(old_dentry);
3409 if (IS_ERR(old_dentry))
3411 /* source must exist */
3413 if (!old_dentry->d_inode)
3415 /* unless the source is a directory trailing slashes give -ENOTDIR */
3416 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3418 if (oldnd.last.name[oldnd.last.len])
3420 if (newnd.last.name[newnd.last.len])
3423 /* source should not be ancestor of target */
3425 if (old_dentry == trap)
3427 new_dentry = lookup_hash(&newnd);
3428 error = PTR_ERR(new_dentry);
3429 if (IS_ERR(new_dentry))
3431 /* target should not be an ancestor of source */
3433 if (new_dentry == trap)
3436 error = mnt_want_write(oldnd.path.mnt);
3439 error = security_path_rename(&oldnd.path, old_dentry,
3440 &newnd.path, new_dentry);
3443 error = vfs_rename(old_dir->d_inode, old_dentry,
3444 new_dir->d_inode, new_dentry);
3446 mnt_drop_write(oldnd.path.mnt);
3452 unlock_rename(new_dir, old_dir);
3454 path_put(&newnd.path);
3457 path_put(&oldnd.path);
3463 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3465 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3468 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3472 len = PTR_ERR(link);
3477 if (len > (unsigned) buflen)
3479 if (copy_to_user(buffer, link, len))
3486 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3487 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3488 * using) it for any given inode is up to filesystem.
3490 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3492 struct nameidata nd;
3497 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3499 return PTR_ERR(cookie);
3501 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3502 if (dentry->d_inode->i_op->put_link)
3503 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3507 int vfs_follow_link(struct nameidata *nd, const char *link)
3509 return __vfs_follow_link(nd, link);
3512 /* get the link contents into pagecache */
3513 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3517 struct address_space *mapping = dentry->d_inode->i_mapping;
3518 page = read_mapping_page(mapping, 0, NULL);
3523 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3527 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3529 struct page *page = NULL;
3530 char *s = page_getlink(dentry, &page);
3531 int res = vfs_readlink(dentry,buffer,buflen,s);
3534 page_cache_release(page);
3539 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3541 struct page *page = NULL;
3542 nd_set_link(nd, page_getlink(dentry, &page));
3546 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3548 struct page *page = cookie;
3552 page_cache_release(page);
3557 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3559 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3561 struct address_space *mapping = inode->i_mapping;
3566 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3568 flags |= AOP_FLAG_NOFS;
3571 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3572 flags, &page, &fsdata);
3576 kaddr = kmap_atomic(page, KM_USER0);
3577 memcpy(kaddr, symname, len-1);
3578 kunmap_atomic(kaddr, KM_USER0);
3580 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3587 mark_inode_dirty(inode);
3593 int page_symlink(struct inode *inode, const char *symname, int len)
3595 return __page_symlink(inode, symname, len,
3596 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3599 const struct inode_operations page_symlink_inode_operations = {
3600 .readlink = generic_readlink,
3601 .follow_link = page_follow_link_light,
3602 .put_link = page_put_link,
3605 EXPORT_SYMBOL(user_path_at);
3606 EXPORT_SYMBOL(follow_down_one);
3607 EXPORT_SYMBOL(follow_down);
3608 EXPORT_SYMBOL(follow_up);
3609 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3610 EXPORT_SYMBOL(getname);
3611 EXPORT_SYMBOL(lock_rename);
3612 EXPORT_SYMBOL(lookup_one_len);
3613 EXPORT_SYMBOL(page_follow_link_light);
3614 EXPORT_SYMBOL(page_put_link);
3615 EXPORT_SYMBOL(page_readlink);
3616 EXPORT_SYMBOL(__page_symlink);
3617 EXPORT_SYMBOL(page_symlink);
3618 EXPORT_SYMBOL(page_symlink_inode_operations);
3619 EXPORT_SYMBOL(path_lookup);
3620 EXPORT_SYMBOL(kern_path);
3621 EXPORT_SYMBOL(vfs_path_lookup);
3622 EXPORT_SYMBOL(inode_permission);
3623 EXPORT_SYMBOL(file_permission);
3624 EXPORT_SYMBOL(unlock_rename);
3625 EXPORT_SYMBOL(vfs_create);
3626 EXPORT_SYMBOL(vfs_follow_link);
3627 EXPORT_SYMBOL(vfs_link);
3628 EXPORT_SYMBOL(vfs_mkdir);
3629 EXPORT_SYMBOL(vfs_mknod);
3630 EXPORT_SYMBOL(generic_permission);
3631 EXPORT_SYMBOL(vfs_readlink);
3632 EXPORT_SYMBOL(vfs_rename);
3633 EXPORT_SYMBOL(vfs_rmdir);
3634 EXPORT_SYMBOL(vfs_symlink);
3635 EXPORT_SYMBOL(vfs_unlink);
3636 EXPORT_SYMBOL(dentry_unhash);
3637 EXPORT_SYMBOL(generic_readlink);