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(struct path *path, struct nameidata *nd)
784 if (!(nd->flags & LOOKUP_RCU)) {
785 dput(nd->path.dentry);
786 if (nd->path.mnt != path->mnt)
787 mntput(nd->path.mnt);
789 nd->path.mnt = path->mnt;
790 nd->path.dentry = path->dentry;
793 static __always_inline int
794 __do_follow_link(struct path *path, struct nameidata *nd, void **p)
797 struct dentry *dentry = path->dentry;
799 touch_atime(path->mnt, dentry);
800 nd_set_link(nd, NULL);
802 if (path->mnt != nd->path.mnt) {
803 path_to_nameidata(path, nd);
804 nd->inode = nd->path.dentry->d_inode;
809 nd->last_type = LAST_BIND;
810 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
813 char *s = nd_get_link(nd);
816 error = __vfs_follow_link(nd, s);
817 else if (nd->last_type == LAST_BIND) {
818 error = force_reval_path(&nd->path, nd);
827 * This limits recursive symlink follows to 8, while
828 * limiting consecutive symlinks to 40.
830 * Without that kind of total limit, nasty chains of consecutive
831 * symlinks can cause almost arbitrarily long lookups.
833 static inline int do_follow_link(struct path *path, struct nameidata *nd)
837 if (current->link_count >= MAX_NESTED_LINKS)
839 if (current->total_link_count >= 40)
841 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
843 err = security_inode_follow_link(path->dentry, nd);
846 current->link_count++;
847 current->total_link_count++;
849 err = __do_follow_link(path, nd, &cookie);
850 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
851 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
853 current->link_count--;
857 path_put_conditional(path, nd);
862 static int follow_up_rcu(struct path *path)
864 struct vfsmount *parent;
865 struct dentry *mountpoint;
867 parent = path->mnt->mnt_parent;
868 if (parent == path->mnt)
870 mountpoint = path->mnt->mnt_mountpoint;
871 path->dentry = mountpoint;
876 int follow_up(struct path *path)
878 struct vfsmount *parent;
879 struct dentry *mountpoint;
881 br_read_lock(vfsmount_lock);
882 parent = path->mnt->mnt_parent;
883 if (parent == path->mnt) {
884 br_read_unlock(vfsmount_lock);
888 mountpoint = dget(path->mnt->mnt_mountpoint);
889 br_read_unlock(vfsmount_lock);
891 path->dentry = mountpoint;
898 * serialization is taken care of in namespace.c
900 static void __follow_mount_rcu(struct nameidata *nd, struct path *path,
901 struct inode **inode)
903 while (d_mountpoint(path->dentry)) {
904 struct vfsmount *mounted;
905 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
909 path->dentry = mounted->mnt_root;
910 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
911 *inode = path->dentry->d_inode;
915 static int __follow_mount(struct path *path)
918 while (d_mountpoint(path->dentry)) {
919 struct vfsmount *mounted = lookup_mnt(path);
926 path->dentry = dget(mounted->mnt_root);
932 static void follow_mount(struct path *path)
934 while (d_mountpoint(path->dentry)) {
935 struct vfsmount *mounted = lookup_mnt(path);
941 path->dentry = dget(mounted->mnt_root);
945 int follow_down(struct path *path)
947 struct vfsmount *mounted;
949 mounted = lookup_mnt(path);
954 path->dentry = dget(mounted->mnt_root);
960 static int follow_dotdot_rcu(struct nameidata *nd)
962 struct inode *inode = nd->inode;
967 if (nd->path.dentry == nd->root.dentry &&
968 nd->path.mnt == nd->root.mnt) {
971 if (nd->path.dentry != nd->path.mnt->mnt_root) {
972 struct dentry *old = nd->path.dentry;
973 struct dentry *parent = old->d_parent;
976 seq = read_seqcount_begin(&parent->d_seq);
977 if (read_seqcount_retry(&old->d_seq, nd->seq))
979 inode = parent->d_inode;
980 nd->path.dentry = parent;
984 if (!follow_up_rcu(&nd->path))
986 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
987 inode = nd->path.dentry->d_inode;
989 __follow_mount_rcu(nd, &nd->path, &inode);
995 static void follow_dotdot(struct nameidata *nd)
1000 struct dentry *old = nd->path.dentry;
1002 if (nd->path.dentry == nd->root.dentry &&
1003 nd->path.mnt == nd->root.mnt) {
1006 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1007 /* rare case of legitimate dget_parent()... */
1008 nd->path.dentry = dget_parent(nd->path.dentry);
1012 if (!follow_up(&nd->path))
1015 follow_mount(&nd->path);
1016 nd->inode = nd->path.dentry->d_inode;
1020 * Allocate a dentry with name and parent, and perform a parent
1021 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1022 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1023 * have verified that no child exists while under i_mutex.
1025 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1026 struct qstr *name, struct nameidata *nd)
1028 struct inode *inode = parent->d_inode;
1029 struct dentry *dentry;
1032 /* Don't create child dentry for a dead directory. */
1033 if (unlikely(IS_DEADDIR(inode)))
1034 return ERR_PTR(-ENOENT);
1036 dentry = d_alloc(parent, name);
1037 if (unlikely(!dentry))
1038 return ERR_PTR(-ENOMEM);
1040 old = inode->i_op->lookup(inode, dentry, nd);
1041 if (unlikely(old)) {
1049 * It's more convoluted than I'd like it to be, but... it's still fairly
1050 * small and for now I'd prefer to have fast path as straight as possible.
1051 * It _is_ time-critical.
1053 static int do_lookup(struct nameidata *nd, struct qstr *name,
1054 struct path *path, struct inode **inode)
1056 struct vfsmount *mnt = nd->path.mnt;
1057 struct dentry *dentry, *parent = nd->path.dentry;
1060 * See if the low-level filesystem might want
1061 * to use its own hash..
1063 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1064 int err = parent->d_op->d_hash(parent, nd->inode, name);
1070 * Rename seqlock is not required here because in the off chance
1071 * of a false negative due to a concurrent rename, we're going to
1072 * do the non-racy lookup, below.
1074 if (nd->flags & LOOKUP_RCU) {
1078 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1080 if (nameidata_drop_rcu(nd))
1084 /* Memory barrier in read_seqcount_begin of child is enough */
1085 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1089 if (dentry->d_flags & DCACHE_OP_REVALIDATE)
1090 goto need_revalidate;
1092 path->dentry = dentry;
1093 __follow_mount_rcu(nd, path, inode);
1095 dentry = __d_lookup(parent, name);
1099 if (dentry->d_flags & DCACHE_OP_REVALIDATE)
1100 goto need_revalidate;
1103 path->dentry = dentry;
1104 __follow_mount(path);
1105 *inode = path->dentry->d_inode;
1110 dir = parent->d_inode;
1111 BUG_ON(nd->inode != dir);
1113 mutex_lock(&dir->i_mutex);
1115 * First re-do the cached lookup just in case it was created
1116 * while we waited for the directory semaphore, or the first
1117 * lookup failed due to an unrelated rename.
1119 * This could use version numbering or similar to avoid unnecessary
1120 * cache lookups, but then we'd have to do the first lookup in the
1121 * non-racy way. However in the common case here, everything should
1122 * be hot in cache, so would it be a big win?
1124 dentry = d_lookup(parent, name);
1125 if (likely(!dentry)) {
1126 dentry = d_alloc_and_lookup(parent, name, nd);
1127 mutex_unlock(&dir->i_mutex);
1133 * Uhhuh! Nasty case: the cache was re-populated while
1134 * we waited on the semaphore. Need to revalidate.
1136 mutex_unlock(&dir->i_mutex);
1140 dentry = do_revalidate(dentry, nd);
1148 return PTR_ERR(dentry);
1152 * This is a temporary kludge to deal with "automount" symlinks; proper
1153 * solution is to trigger them on follow_mount(), so that do_lookup()
1154 * would DTRT. To be killed before 2.6.34-final.
1156 static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
1158 return inode && unlikely(inode->i_op->follow_link) &&
1159 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
1164 * This is the basic name resolution function, turning a pathname into
1165 * the final dentry. We expect 'base' to be positive and a directory.
1167 * Returns 0 and nd will have valid dentry and mnt on success.
1168 * Returns error and drops reference to input namei data on failure.
1170 static int link_path_walk(const char *name, struct nameidata *nd)
1174 unsigned int lookup_flags = nd->flags;
1182 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1184 /* At this point we know we have a real path component. */
1186 struct inode *inode;
1191 nd->flags |= LOOKUP_CONTINUE;
1192 if (nd->flags & LOOKUP_RCU) {
1193 err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1194 if (err == -ECHILD) {
1195 if (nameidata_drop_rcu(nd))
1201 err = exec_permission(nd->inode, 0);
1207 c = *(const unsigned char *)name;
1209 hash = init_name_hash();
1212 hash = partial_name_hash(c, hash);
1213 c = *(const unsigned char *)name;
1214 } while (c && (c != '/'));
1215 this.len = name - (const char *) this.name;
1216 this.hash = end_name_hash(hash);
1218 /* remove trailing slashes? */
1220 goto last_component;
1221 while (*++name == '/');
1223 goto last_with_slashes;
1226 * "." and ".." are special - ".." especially so because it has
1227 * to be able to know about the current root directory and
1228 * parent relationships.
1230 if (this.name[0] == '.') switch (this.len) {
1234 if (this.name[1] != '.')
1236 if (nd->flags & LOOKUP_RCU) {
1237 if (follow_dotdot_rcu(nd))
1245 /* This does the actual lookups.. */
1246 err = do_lookup(nd, &this, &next, &inode);
1253 if (inode->i_op->follow_link) {
1254 /* We commonly drop rcu-walk here */
1255 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1257 BUG_ON(inode != next.dentry->d_inode);
1258 err = do_follow_link(&next, nd);
1261 nd->inode = nd->path.dentry->d_inode;
1266 path_to_nameidata(&next, nd);
1270 if (!nd->inode->i_op->lookup)
1273 /* here ends the main loop */
1276 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1278 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1279 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1280 if (lookup_flags & LOOKUP_PARENT)
1282 if (this.name[0] == '.') switch (this.len) {
1286 if (this.name[1] != '.')
1288 if (nd->flags & LOOKUP_RCU) {
1289 if (follow_dotdot_rcu(nd))
1297 err = do_lookup(nd, &this, &next, &inode);
1300 if (follow_on_final(inode, lookup_flags)) {
1301 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1303 BUG_ON(inode != next.dentry->d_inode);
1304 err = do_follow_link(&next, nd);
1307 nd->inode = nd->path.dentry->d_inode;
1309 path_to_nameidata(&next, nd);
1315 if (lookup_flags & LOOKUP_DIRECTORY) {
1317 if (!nd->inode->i_op->lookup)
1323 nd->last_type = LAST_NORM;
1324 if (this.name[0] != '.')
1327 nd->last_type = LAST_DOT;
1328 else if (this.len == 2 && this.name[1] == '.')
1329 nd->last_type = LAST_DOTDOT;
1334 * We bypassed the ordinary revalidation routines.
1335 * We may need to check the cached dentry for staleness.
1337 if (need_reval_dot(nd->path.dentry)) {
1338 /* Note: we do not d_invalidate() */
1339 err = d_revalidate(nd->path.dentry, nd);
1346 if (nameidata_drop_rcu_last_maybe(nd))
1350 if (!(nd->flags & LOOKUP_RCU))
1351 path_put_conditional(&next, nd);
1354 if (!(nd->flags & LOOKUP_RCU))
1355 path_put(&nd->path);
1360 static inline int path_walk_rcu(const char *name, struct nameidata *nd)
1362 current->total_link_count = 0;
1364 return link_path_walk(name, nd);
1367 static inline int path_walk_simple(const char *name, struct nameidata *nd)
1369 current->total_link_count = 0;
1371 return link_path_walk(name, nd);
1374 static int path_walk(const char *name, struct nameidata *nd)
1376 struct path save = nd->path;
1379 current->total_link_count = 0;
1381 /* make sure the stuff we saved doesn't go away */
1384 result = link_path_walk(name, nd);
1385 if (result == -ESTALE) {
1386 /* nd->path had been dropped */
1387 current->total_link_count = 0;
1389 path_get(&nd->path);
1390 nd->flags |= LOOKUP_REVAL;
1391 result = link_path_walk(name, nd);
1399 static void path_finish_rcu(struct nameidata *nd)
1401 if (nd->flags & LOOKUP_RCU) {
1402 /* RCU dangling. Cancel it. */
1403 nd->flags &= ~LOOKUP_RCU;
1404 nd->root.mnt = NULL;
1406 br_read_unlock(vfsmount_lock);
1412 static int path_init_rcu(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1418 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1419 nd->flags = flags | LOOKUP_RCU;
1421 nd->root.mnt = NULL;
1425 struct fs_struct *fs = current->fs;
1428 br_read_lock(vfsmount_lock);
1432 seq = read_seqcount_begin(&fs->seq);
1433 nd->root = fs->root;
1434 nd->path = nd->root;
1435 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1436 } while (read_seqcount_retry(&fs->seq, seq));
1438 } else if (dfd == AT_FDCWD) {
1439 struct fs_struct *fs = current->fs;
1442 br_read_lock(vfsmount_lock);
1446 seq = read_seqcount_begin(&fs->seq);
1448 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1449 } while (read_seqcount_retry(&fs->seq, seq));
1452 struct dentry *dentry;
1454 file = fget_light(dfd, &fput_needed);
1459 dentry = file->f_path.dentry;
1462 if (!S_ISDIR(dentry->d_inode->i_mode))
1465 retval = file_permission(file, MAY_EXEC);
1469 nd->path = file->f_path;
1473 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1474 br_read_lock(vfsmount_lock);
1477 nd->inode = nd->path.dentry->d_inode;
1481 fput_light(file, fput_needed);
1486 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1492 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1495 nd->root.mnt = NULL;
1499 nd->path = nd->root;
1500 path_get(&nd->root);
1501 } else if (dfd == AT_FDCWD) {
1502 get_fs_pwd(current->fs, &nd->path);
1504 struct dentry *dentry;
1506 file = fget_light(dfd, &fput_needed);
1511 dentry = file->f_path.dentry;
1514 if (!S_ISDIR(dentry->d_inode->i_mode))
1517 retval = file_permission(file, MAY_EXEC);
1521 nd->path = file->f_path;
1522 path_get(&file->f_path);
1524 fput_light(file, fput_needed);
1526 nd->inode = nd->path.dentry->d_inode;
1530 fput_light(file, fput_needed);
1535 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1536 static int do_path_lookup(int dfd, const char *name,
1537 unsigned int flags, struct nameidata *nd)
1542 * Path walking is largely split up into 2 different synchronisation
1543 * schemes, rcu-walk and ref-walk (explained in
1544 * Documentation/filesystems/path-lookup.txt). These share much of the
1545 * path walk code, but some things particularly setup, cleanup, and
1546 * following mounts are sufficiently divergent that functions are
1547 * duplicated. Typically there is a function foo(), and its RCU
1548 * analogue, foo_rcu().
1550 * -ECHILD is the error number of choice (just to avoid clashes) that
1551 * is returned if some aspect of an rcu-walk fails. Such an error must
1552 * be handled by restarting a traditional ref-walk (which will always
1553 * be able to complete).
1555 retval = path_init_rcu(dfd, name, flags, nd);
1556 if (unlikely(retval))
1558 retval = path_walk_rcu(name, nd);
1559 path_finish_rcu(nd);
1561 path_put(&nd->root);
1562 nd->root.mnt = NULL;
1565 if (unlikely(retval == -ECHILD || retval == -ESTALE)) {
1566 /* slower, locked walk */
1567 if (retval == -ESTALE)
1568 flags |= LOOKUP_REVAL;
1569 retval = path_init(dfd, name, flags, nd);
1570 if (unlikely(retval))
1572 retval = path_walk(name, nd);
1574 path_put(&nd->root);
1575 nd->root.mnt = NULL;
1579 if (likely(!retval)) {
1580 if (unlikely(!audit_dummy_context())) {
1581 if (nd->path.dentry && nd->inode)
1582 audit_inode(name, nd->path.dentry);
1589 int path_lookup(const char *name, unsigned int flags,
1590 struct nameidata *nd)
1592 return do_path_lookup(AT_FDCWD, name, flags, nd);
1595 int kern_path(const char *name, unsigned int flags, struct path *path)
1597 struct nameidata nd;
1598 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1605 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1606 * @dentry: pointer to dentry of the base directory
1607 * @mnt: pointer to vfs mount of the base directory
1608 * @name: pointer to file name
1609 * @flags: lookup flags
1610 * @nd: pointer to nameidata
1612 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1613 const char *name, unsigned int flags,
1614 struct nameidata *nd)
1618 /* same as do_path_lookup */
1619 nd->last_type = LAST_ROOT;
1623 nd->path.dentry = dentry;
1625 path_get(&nd->path);
1626 nd->root = nd->path;
1627 path_get(&nd->root);
1628 nd->inode = nd->path.dentry->d_inode;
1630 retval = path_walk(name, nd);
1631 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1633 audit_inode(name, nd->path.dentry);
1635 path_put(&nd->root);
1636 nd->root.mnt = NULL;
1641 static struct dentry *__lookup_hash(struct qstr *name,
1642 struct dentry *base, struct nameidata *nd)
1644 struct inode *inode = base->d_inode;
1645 struct dentry *dentry;
1648 err = exec_permission(inode, 0);
1650 return ERR_PTR(err);
1653 * See if the low-level filesystem might want
1654 * to use its own hash..
1656 if (base->d_flags & DCACHE_OP_HASH) {
1657 err = base->d_op->d_hash(base, inode, name);
1658 dentry = ERR_PTR(err);
1664 * Don't bother with __d_lookup: callers are for creat as
1665 * well as unlink, so a lot of the time it would cost
1668 dentry = d_lookup(base, name);
1670 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
1671 dentry = do_revalidate(dentry, nd);
1674 dentry = d_alloc_and_lookup(base, name, nd);
1680 * Restricted form of lookup. Doesn't follow links, single-component only,
1681 * needs parent already locked. Doesn't follow mounts.
1684 static struct dentry *lookup_hash(struct nameidata *nd)
1686 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1689 static int __lookup_one_len(const char *name, struct qstr *this,
1690 struct dentry *base, int len)
1700 hash = init_name_hash();
1702 c = *(const unsigned char *)name++;
1703 if (c == '/' || c == '\0')
1705 hash = partial_name_hash(c, hash);
1707 this->hash = end_name_hash(hash);
1712 * lookup_one_len - filesystem helper to lookup single pathname component
1713 * @name: pathname component to lookup
1714 * @base: base directory to lookup from
1715 * @len: maximum length @len should be interpreted to
1717 * Note that this routine is purely a helper for filesystem usage and should
1718 * not be called by generic code. Also note that by using this function the
1719 * nameidata argument is passed to the filesystem methods and a filesystem
1720 * using this helper needs to be prepared for that.
1722 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1727 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1729 err = __lookup_one_len(name, &this, base, len);
1731 return ERR_PTR(err);
1733 return __lookup_hash(&this, base, NULL);
1736 int user_path_at(int dfd, const char __user *name, unsigned flags,
1739 struct nameidata nd;
1740 char *tmp = getname(name);
1741 int err = PTR_ERR(tmp);
1744 BUG_ON(flags & LOOKUP_PARENT);
1746 err = do_path_lookup(dfd, tmp, flags, &nd);
1754 static int user_path_parent(int dfd, const char __user *path,
1755 struct nameidata *nd, char **name)
1757 char *s = getname(path);
1763 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1773 * It's inline, so penalty for filesystems that don't use sticky bit is
1776 static inline int check_sticky(struct inode *dir, struct inode *inode)
1778 uid_t fsuid = current_fsuid();
1780 if (!(dir->i_mode & S_ISVTX))
1782 if (inode->i_uid == fsuid)
1784 if (dir->i_uid == fsuid)
1786 return !capable(CAP_FOWNER);
1790 * Check whether we can remove a link victim from directory dir, check
1791 * whether the type of victim is right.
1792 * 1. We can't do it if dir is read-only (done in permission())
1793 * 2. We should have write and exec permissions on dir
1794 * 3. We can't remove anything from append-only dir
1795 * 4. We can't do anything with immutable dir (done in permission())
1796 * 5. If the sticky bit on dir is set we should either
1797 * a. be owner of dir, or
1798 * b. be owner of victim, or
1799 * c. have CAP_FOWNER capability
1800 * 6. If the victim is append-only or immutable we can't do antyhing with
1801 * links pointing to it.
1802 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1803 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1804 * 9. We can't remove a root or mountpoint.
1805 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1806 * nfs_async_unlink().
1808 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1812 if (!victim->d_inode)
1815 BUG_ON(victim->d_parent->d_inode != dir);
1816 audit_inode_child(victim, dir);
1818 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1823 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1824 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1827 if (!S_ISDIR(victim->d_inode->i_mode))
1829 if (IS_ROOT(victim))
1831 } else if (S_ISDIR(victim->d_inode->i_mode))
1833 if (IS_DEADDIR(dir))
1835 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1840 /* Check whether we can create an object with dentry child in directory
1842 * 1. We can't do it if child already exists (open has special treatment for
1843 * this case, but since we are inlined it's OK)
1844 * 2. We can't do it if dir is read-only (done in permission())
1845 * 3. We should have write and exec permissions on dir
1846 * 4. We can't do it if dir is immutable (done in permission())
1848 static inline int may_create(struct inode *dir, struct dentry *child)
1852 if (IS_DEADDIR(dir))
1854 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1858 * p1 and p2 should be directories on the same fs.
1860 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1865 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1869 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1871 p = d_ancestor(p2, p1);
1873 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1874 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1878 p = d_ancestor(p1, p2);
1880 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1881 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1885 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1886 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1890 void unlock_rename(struct dentry *p1, struct dentry *p2)
1892 mutex_unlock(&p1->d_inode->i_mutex);
1894 mutex_unlock(&p2->d_inode->i_mutex);
1895 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1899 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1900 struct nameidata *nd)
1902 int error = may_create(dir, dentry);
1907 if (!dir->i_op->create)
1908 return -EACCES; /* shouldn't it be ENOSYS? */
1911 error = security_inode_create(dir, dentry, mode);
1914 error = dir->i_op->create(dir, dentry, mode, nd);
1916 fsnotify_create(dir, dentry);
1920 int may_open(struct path *path, int acc_mode, int flag)
1922 struct dentry *dentry = path->dentry;
1923 struct inode *inode = dentry->d_inode;
1929 switch (inode->i_mode & S_IFMT) {
1933 if (acc_mode & MAY_WRITE)
1938 if (path->mnt->mnt_flags & MNT_NODEV)
1947 error = inode_permission(inode, acc_mode);
1952 * An append-only file must be opened in append mode for writing.
1954 if (IS_APPEND(inode)) {
1955 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
1961 /* O_NOATIME can only be set by the owner or superuser */
1962 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1966 * Ensure there are no outstanding leases on the file.
1968 return break_lease(inode, flag);
1971 static int handle_truncate(struct file *filp)
1973 struct path *path = &filp->f_path;
1974 struct inode *inode = path->dentry->d_inode;
1975 int error = get_write_access(inode);
1979 * Refuse to truncate files with mandatory locks held on them.
1981 error = locks_verify_locked(inode);
1983 error = security_path_truncate(path);
1985 error = do_truncate(path->dentry, 0,
1986 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1989 put_write_access(inode);
1994 * Be careful about ever adding any more callers of this
1995 * function. Its flags must be in the namei format, not
1996 * what get passed to sys_open().
1998 static int __open_namei_create(struct nameidata *nd, struct path *path,
1999 int open_flag, int mode)
2002 struct dentry *dir = nd->path.dentry;
2004 if (!IS_POSIXACL(dir->d_inode))
2005 mode &= ~current_umask();
2006 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
2009 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
2011 mutex_unlock(&dir->d_inode->i_mutex);
2012 dput(nd->path.dentry);
2013 nd->path.dentry = path->dentry;
2017 /* Don't check for write permission, don't truncate */
2018 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
2022 * Note that while the flag value (low two bits) for sys_open means:
2027 * it is changed into
2028 * 00 - no permissions needed
2029 * 01 - read-permission
2030 * 10 - write-permission
2032 * for the internal routines (ie open_namei()/follow_link() etc)
2033 * This is more logical, and also allows the 00 "no perm needed"
2034 * to be used for symlinks (where the permissions are checked
2038 static inline int open_to_namei_flags(int flag)
2040 if ((flag+1) & O_ACCMODE)
2045 static int open_will_truncate(int flag, struct inode *inode)
2048 * We'll never write to the fs underlying
2051 if (special_file(inode->i_mode))
2053 return (flag & O_TRUNC);
2056 static struct file *finish_open(struct nameidata *nd,
2057 int open_flag, int acc_mode)
2063 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
2064 if (will_truncate) {
2065 error = mnt_want_write(nd->path.mnt);
2069 error = may_open(&nd->path, acc_mode, open_flag);
2072 mnt_drop_write(nd->path.mnt);
2075 filp = nameidata_to_filp(nd);
2076 if (!IS_ERR(filp)) {
2077 error = ima_file_check(filp, acc_mode);
2080 filp = ERR_PTR(error);
2083 if (!IS_ERR(filp)) {
2084 if (will_truncate) {
2085 error = handle_truncate(filp);
2088 filp = ERR_PTR(error);
2093 * It is now safe to drop the mnt write
2094 * because the filp has had a write taken
2098 mnt_drop_write(nd->path.mnt);
2099 path_put(&nd->path);
2103 if (!IS_ERR(nd->intent.open.file))
2104 release_open_intent(nd);
2105 path_put(&nd->path);
2106 return ERR_PTR(error);
2110 * Handle O_CREAT case for do_filp_open
2112 static struct file *do_last(struct nameidata *nd, struct path *path,
2113 int open_flag, int acc_mode,
2114 int mode, const char *pathname)
2116 struct dentry *dir = nd->path.dentry;
2118 int error = -EISDIR;
2120 switch (nd->last_type) {
2123 dir = nd->path.dentry;
2125 if (need_reval_dot(dir)) {
2126 int status = d_revalidate(nd->path.dentry, nd);
2138 audit_inode(pathname, dir);
2142 /* trailing slashes? */
2143 if (nd->last.name[nd->last.len])
2146 mutex_lock(&dir->d_inode->i_mutex);
2148 path->dentry = lookup_hash(nd);
2149 path->mnt = nd->path.mnt;
2151 error = PTR_ERR(path->dentry);
2152 if (IS_ERR(path->dentry)) {
2153 mutex_unlock(&dir->d_inode->i_mutex);
2157 if (IS_ERR(nd->intent.open.file)) {
2158 error = PTR_ERR(nd->intent.open.file);
2159 goto exit_mutex_unlock;
2162 /* Negative dentry, just create the file */
2163 if (!path->dentry->d_inode) {
2165 * This write is needed to ensure that a
2166 * ro->rw transition does not occur between
2167 * the time when the file is created and when
2168 * a permanent write count is taken through
2169 * the 'struct file' in nameidata_to_filp().
2171 error = mnt_want_write(nd->path.mnt);
2173 goto exit_mutex_unlock;
2174 error = __open_namei_create(nd, path, open_flag, mode);
2176 mnt_drop_write(nd->path.mnt);
2179 filp = nameidata_to_filp(nd);
2180 mnt_drop_write(nd->path.mnt);
2181 path_put(&nd->path);
2182 if (!IS_ERR(filp)) {
2183 error = ima_file_check(filp, acc_mode);
2186 filp = ERR_PTR(error);
2193 * It already exists.
2195 mutex_unlock(&dir->d_inode->i_mutex);
2196 audit_inode(pathname, path->dentry);
2199 if (open_flag & O_EXCL)
2202 if (__follow_mount(path)) {
2204 if (open_flag & O_NOFOLLOW)
2209 if (!path->dentry->d_inode)
2212 if (path->dentry->d_inode->i_op->follow_link)
2215 path_to_nameidata(path, nd);
2216 nd->inode = path->dentry->d_inode;
2218 if (S_ISDIR(nd->inode->i_mode))
2221 filp = finish_open(nd, open_flag, acc_mode);
2225 mutex_unlock(&dir->d_inode->i_mutex);
2227 path_put_conditional(path, nd);
2229 if (!IS_ERR(nd->intent.open.file))
2230 release_open_intent(nd);
2231 path_put(&nd->path);
2232 return ERR_PTR(error);
2236 * Note that the low bits of the passed in "open_flag"
2237 * are not the same as in the local variable "flag". See
2238 * open_to_namei_flags() for more details.
2240 struct file *do_filp_open(int dfd, const char *pathname,
2241 int open_flag, int mode, int acc_mode)
2244 struct nameidata nd;
2248 int flag = open_to_namei_flags(open_flag);
2251 if (!(open_flag & O_CREAT))
2254 /* Must never be set by userspace */
2255 open_flag &= ~FMODE_NONOTIFY;
2258 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2259 * check for O_DSYNC if the need any syncing at all we enforce it's
2260 * always set instead of having to deal with possibly weird behaviour
2261 * for malicious applications setting only __O_SYNC.
2263 if (open_flag & __O_SYNC)
2264 open_flag |= O_DSYNC;
2267 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
2269 /* O_TRUNC implies we need access checks for write permissions */
2270 if (open_flag & O_TRUNC)
2271 acc_mode |= MAY_WRITE;
2273 /* Allow the LSM permission hook to distinguish append
2274 access from general write access. */
2275 if (open_flag & O_APPEND)
2276 acc_mode |= MAY_APPEND;
2278 flags = LOOKUP_OPEN;
2279 if (open_flag & O_CREAT) {
2280 flags |= LOOKUP_CREATE;
2281 if (open_flag & O_EXCL)
2282 flags |= LOOKUP_EXCL;
2284 if (open_flag & O_DIRECTORY)
2285 flags |= LOOKUP_DIRECTORY;
2286 if (!(open_flag & O_NOFOLLOW))
2287 flags |= LOOKUP_FOLLOW;
2289 filp = get_empty_filp();
2291 return ERR_PTR(-ENFILE);
2293 filp->f_flags = open_flag;
2294 nd.intent.open.file = filp;
2295 nd.intent.open.flags = flag;
2296 nd.intent.open.create_mode = mode;
2298 if (open_flag & O_CREAT)
2301 /* !O_CREAT, simple open */
2302 error = do_path_lookup(dfd, pathname, flags, &nd);
2303 if (unlikely(error))
2306 if (!(nd.flags & LOOKUP_FOLLOW)) {
2307 if (nd.inode->i_op->follow_link)
2311 if (nd.flags & LOOKUP_DIRECTORY) {
2312 if (!nd.inode->i_op->lookup)
2315 audit_inode(pathname, nd.path.dentry);
2316 filp = finish_open(&nd, open_flag, acc_mode);
2320 /* OK, have to create the file. Find the parent. */
2321 error = path_init_rcu(dfd, pathname,
2322 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2325 error = path_walk_rcu(pathname, &nd);
2326 path_finish_rcu(&nd);
2327 if (unlikely(error == -ECHILD || error == -ESTALE)) {
2328 /* slower, locked walk */
2329 if (error == -ESTALE) {
2331 flags |= LOOKUP_REVAL;
2333 error = path_init(dfd, pathname,
2334 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2338 error = path_walk_simple(pathname, &nd);
2340 if (unlikely(error))
2342 if (unlikely(!audit_dummy_context()))
2343 audit_inode(pathname, nd.path.dentry);
2346 * We have the parent and last component.
2349 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
2350 while (unlikely(!filp)) { /* trailing symlink */
2354 /* S_ISDIR part is a temporary automount kludge */
2355 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(nd.inode->i_mode))
2360 * This is subtle. Instead of calling do_follow_link() we do
2361 * the thing by hands. The reason is that this way we have zero
2362 * link_count and path_walk() (called from ->follow_link)
2363 * honoring LOOKUP_PARENT. After that we have the parent and
2364 * last component, i.e. we are in the same situation as after
2365 * the first path_walk(). Well, almost - if the last component
2366 * is normal we get its copy stored in nd->last.name and we will
2367 * have to putname() it when we are done. Procfs-like symlinks
2368 * just set LAST_BIND.
2370 nd.flags |= LOOKUP_PARENT;
2371 error = security_inode_follow_link(path.dentry, &nd);
2374 error = __do_follow_link(&path, &nd, &cookie);
2375 if (unlikely(error)) {
2376 if (!IS_ERR(cookie) && nd.inode->i_op->put_link)
2377 nd.inode->i_op->put_link(path.dentry, &nd, cookie);
2378 /* nd.path had been dropped */
2383 nd.flags &= ~LOOKUP_PARENT;
2384 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
2385 if (nd.inode->i_op->put_link)
2386 nd.inode->i_op->put_link(holder.dentry, &nd, cookie);
2392 if (filp == ERR_PTR(-ESTALE) && !(flags & LOOKUP_REVAL))
2397 path_put_conditional(&path, &nd);
2401 if (!IS_ERR(nd.intent.open.file))
2402 release_open_intent(&nd);
2403 filp = ERR_PTR(error);
2408 * filp_open - open file and return file pointer
2410 * @filename: path to open
2411 * @flags: open flags as per the open(2) second argument
2412 * @mode: mode for the new file if O_CREAT is set, else ignored
2414 * This is the helper to open a file from kernelspace if you really
2415 * have to. But in generally you should not do this, so please move
2416 * along, nothing to see here..
2418 struct file *filp_open(const char *filename, int flags, int mode)
2420 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
2422 EXPORT_SYMBOL(filp_open);
2425 * lookup_create - lookup a dentry, creating it if it doesn't exist
2426 * @nd: nameidata info
2427 * @is_dir: directory flag
2429 * Simple function to lookup and return a dentry and create it
2430 * if it doesn't exist. Is SMP-safe.
2432 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2434 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2436 struct dentry *dentry = ERR_PTR(-EEXIST);
2438 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2440 * Yucky last component or no last component at all?
2441 * (foo/., foo/.., /////)
2443 if (nd->last_type != LAST_NORM)
2445 nd->flags &= ~LOOKUP_PARENT;
2446 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2447 nd->intent.open.flags = O_EXCL;
2450 * Do the final lookup.
2452 dentry = lookup_hash(nd);
2456 if (dentry->d_inode)
2459 * Special case - lookup gave negative, but... we had foo/bar/
2460 * From the vfs_mknod() POV we just have a negative dentry -
2461 * all is fine. Let's be bastards - you had / on the end, you've
2462 * been asking for (non-existent) directory. -ENOENT for you.
2464 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2466 dentry = ERR_PTR(-ENOENT);
2471 dentry = ERR_PTR(-EEXIST);
2475 EXPORT_SYMBOL_GPL(lookup_create);
2477 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2479 int error = may_create(dir, dentry);
2484 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2487 if (!dir->i_op->mknod)
2490 error = devcgroup_inode_mknod(mode, dev);
2494 error = security_inode_mknod(dir, dentry, mode, dev);
2498 error = dir->i_op->mknod(dir, dentry, mode, dev);
2500 fsnotify_create(dir, dentry);
2504 static int may_mknod(mode_t mode)
2506 switch (mode & S_IFMT) {
2512 case 0: /* zero mode translates to S_IFREG */
2521 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2526 struct dentry *dentry;
2527 struct nameidata nd;
2532 error = user_path_parent(dfd, filename, &nd, &tmp);
2536 dentry = lookup_create(&nd, 0);
2537 if (IS_ERR(dentry)) {
2538 error = PTR_ERR(dentry);
2541 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2542 mode &= ~current_umask();
2543 error = may_mknod(mode);
2546 error = mnt_want_write(nd.path.mnt);
2549 error = security_path_mknod(&nd.path, dentry, mode, dev);
2551 goto out_drop_write;
2552 switch (mode & S_IFMT) {
2553 case 0: case S_IFREG:
2554 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2556 case S_IFCHR: case S_IFBLK:
2557 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2558 new_decode_dev(dev));
2560 case S_IFIFO: case S_IFSOCK:
2561 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2565 mnt_drop_write(nd.path.mnt);
2569 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2576 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2578 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2581 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2583 int error = may_create(dir, dentry);
2588 if (!dir->i_op->mkdir)
2591 mode &= (S_IRWXUGO|S_ISVTX);
2592 error = security_inode_mkdir(dir, dentry, mode);
2596 error = dir->i_op->mkdir(dir, dentry, mode);
2598 fsnotify_mkdir(dir, dentry);
2602 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2606 struct dentry *dentry;
2607 struct nameidata nd;
2609 error = user_path_parent(dfd, pathname, &nd, &tmp);
2613 dentry = lookup_create(&nd, 1);
2614 error = PTR_ERR(dentry);
2618 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2619 mode &= ~current_umask();
2620 error = mnt_want_write(nd.path.mnt);
2623 error = security_path_mkdir(&nd.path, dentry, mode);
2625 goto out_drop_write;
2626 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2628 mnt_drop_write(nd.path.mnt);
2632 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2639 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2641 return sys_mkdirat(AT_FDCWD, pathname, mode);
2645 * We try to drop the dentry early: we should have
2646 * a usage count of 2 if we're the only user of this
2647 * dentry, and if that is true (possibly after pruning
2648 * the dcache), then we drop the dentry now.
2650 * A low-level filesystem can, if it choses, legally
2653 * if (!d_unhashed(dentry))
2656 * if it cannot handle the case of removing a directory
2657 * that is still in use by something else..
2659 void dentry_unhash(struct dentry *dentry)
2662 shrink_dcache_parent(dentry);
2663 spin_lock(&dentry->d_lock);
2664 if (dentry->d_count == 2)
2666 spin_unlock(&dentry->d_lock);
2669 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2671 int error = may_delete(dir, dentry, 1);
2676 if (!dir->i_op->rmdir)
2679 mutex_lock(&dentry->d_inode->i_mutex);
2680 dentry_unhash(dentry);
2681 if (d_mountpoint(dentry))
2684 error = security_inode_rmdir(dir, dentry);
2686 error = dir->i_op->rmdir(dir, dentry);
2688 dentry->d_inode->i_flags |= S_DEAD;
2693 mutex_unlock(&dentry->d_inode->i_mutex);
2702 static long do_rmdir(int dfd, const char __user *pathname)
2706 struct dentry *dentry;
2707 struct nameidata nd;
2709 error = user_path_parent(dfd, pathname, &nd, &name);
2713 switch(nd.last_type) {
2725 nd.flags &= ~LOOKUP_PARENT;
2727 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2728 dentry = lookup_hash(&nd);
2729 error = PTR_ERR(dentry);
2732 error = mnt_want_write(nd.path.mnt);
2735 error = security_path_rmdir(&nd.path, dentry);
2738 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2740 mnt_drop_write(nd.path.mnt);
2744 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2751 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2753 return do_rmdir(AT_FDCWD, pathname);
2756 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2758 int error = may_delete(dir, dentry, 0);
2763 if (!dir->i_op->unlink)
2766 mutex_lock(&dentry->d_inode->i_mutex);
2767 if (d_mountpoint(dentry))
2770 error = security_inode_unlink(dir, dentry);
2772 error = dir->i_op->unlink(dir, dentry);
2777 mutex_unlock(&dentry->d_inode->i_mutex);
2779 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2780 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2781 fsnotify_link_count(dentry->d_inode);
2789 * Make sure that the actual truncation of the file will occur outside its
2790 * directory's i_mutex. Truncate can take a long time if there is a lot of
2791 * writeout happening, and we don't want to prevent access to the directory
2792 * while waiting on the I/O.
2794 static long do_unlinkat(int dfd, const char __user *pathname)
2798 struct dentry *dentry;
2799 struct nameidata nd;
2800 struct inode *inode = NULL;
2802 error = user_path_parent(dfd, pathname, &nd, &name);
2807 if (nd.last_type != LAST_NORM)
2810 nd.flags &= ~LOOKUP_PARENT;
2812 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2813 dentry = lookup_hash(&nd);
2814 error = PTR_ERR(dentry);
2815 if (!IS_ERR(dentry)) {
2816 /* Why not before? Because we want correct error value */
2817 if (nd.last.name[nd.last.len])
2819 inode = dentry->d_inode;
2822 error = mnt_want_write(nd.path.mnt);
2825 error = security_path_unlink(&nd.path, dentry);
2828 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2830 mnt_drop_write(nd.path.mnt);
2834 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2836 iput(inode); /* truncate the inode here */
2843 error = !dentry->d_inode ? -ENOENT :
2844 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2848 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2850 if ((flag & ~AT_REMOVEDIR) != 0)
2853 if (flag & AT_REMOVEDIR)
2854 return do_rmdir(dfd, pathname);
2856 return do_unlinkat(dfd, pathname);
2859 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2861 return do_unlinkat(AT_FDCWD, pathname);
2864 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2866 int error = may_create(dir, dentry);
2871 if (!dir->i_op->symlink)
2874 error = security_inode_symlink(dir, dentry, oldname);
2878 error = dir->i_op->symlink(dir, dentry, oldname);
2880 fsnotify_create(dir, dentry);
2884 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2885 int, newdfd, const char __user *, newname)
2890 struct dentry *dentry;
2891 struct nameidata nd;
2893 from = getname(oldname);
2895 return PTR_ERR(from);
2897 error = user_path_parent(newdfd, newname, &nd, &to);
2901 dentry = lookup_create(&nd, 0);
2902 error = PTR_ERR(dentry);
2906 error = mnt_want_write(nd.path.mnt);
2909 error = security_path_symlink(&nd.path, dentry, from);
2911 goto out_drop_write;
2912 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2914 mnt_drop_write(nd.path.mnt);
2918 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2926 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2928 return sys_symlinkat(oldname, AT_FDCWD, newname);
2931 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2933 struct inode *inode = old_dentry->d_inode;
2939 error = may_create(dir, new_dentry);
2943 if (dir->i_sb != inode->i_sb)
2947 * A link to an append-only or immutable file cannot be created.
2949 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2951 if (!dir->i_op->link)
2953 if (S_ISDIR(inode->i_mode))
2956 error = security_inode_link(old_dentry, dir, new_dentry);
2960 mutex_lock(&inode->i_mutex);
2961 error = dir->i_op->link(old_dentry, dir, new_dentry);
2962 mutex_unlock(&inode->i_mutex);
2964 fsnotify_link(dir, inode, new_dentry);
2969 * Hardlinks are often used in delicate situations. We avoid
2970 * security-related surprises by not following symlinks on the
2973 * We don't follow them on the oldname either to be compatible
2974 * with linux 2.0, and to avoid hard-linking to directories
2975 * and other special files. --ADM
2977 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2978 int, newdfd, const char __user *, newname, int, flags)
2980 struct dentry *new_dentry;
2981 struct nameidata nd;
2982 struct path old_path;
2986 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2989 error = user_path_at(olddfd, oldname,
2990 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2995 error = user_path_parent(newdfd, newname, &nd, &to);
2999 if (old_path.mnt != nd.path.mnt)
3001 new_dentry = lookup_create(&nd, 0);
3002 error = PTR_ERR(new_dentry);
3003 if (IS_ERR(new_dentry))
3005 error = mnt_want_write(nd.path.mnt);
3008 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
3010 goto out_drop_write;
3011 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
3013 mnt_drop_write(nd.path.mnt);
3017 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3022 path_put(&old_path);
3027 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3029 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3033 * The worst of all namespace operations - renaming directory. "Perverted"
3034 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3036 * a) we can get into loop creation. Check is done in is_subdir().
3037 * b) race potential - two innocent renames can create a loop together.
3038 * That's where 4.4 screws up. Current fix: serialization on
3039 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3041 * c) we have to lock _three_ objects - parents and victim (if it exists).
3042 * And that - after we got ->i_mutex on parents (until then we don't know
3043 * whether the target exists). Solution: try to be smart with locking
3044 * order for inodes. We rely on the fact that tree topology may change
3045 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3046 * move will be locked. Thus we can rank directories by the tree
3047 * (ancestors first) and rank all non-directories after them.
3048 * That works since everybody except rename does "lock parent, lookup,
3049 * lock child" and rename is under ->s_vfs_rename_mutex.
3050 * HOWEVER, it relies on the assumption that any object with ->lookup()
3051 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3052 * we'd better make sure that there's no link(2) for them.
3053 * d) some filesystems don't support opened-but-unlinked directories,
3054 * either because of layout or because they are not ready to deal with
3055 * all cases correctly. The latter will be fixed (taking this sort of
3056 * stuff into VFS), but the former is not going away. Solution: the same
3057 * trick as in rmdir().
3058 * e) conversion from fhandle to dentry may come in the wrong moment - when
3059 * we are removing the target. Solution: we will have to grab ->i_mutex
3060 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3061 * ->i_mutex on parents, which works but leads to some truly excessive
3064 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3065 struct inode *new_dir, struct dentry *new_dentry)
3068 struct inode *target;
3071 * If we are going to change the parent - check write permissions,
3072 * we'll need to flip '..'.
3074 if (new_dir != old_dir) {
3075 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3080 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3084 target = new_dentry->d_inode;
3086 mutex_lock(&target->i_mutex);
3087 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3091 dentry_unhash(new_dentry);
3092 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3096 target->i_flags |= S_DEAD;
3097 dont_mount(new_dentry);
3099 mutex_unlock(&target->i_mutex);
3100 if (d_unhashed(new_dentry))
3101 d_rehash(new_dentry);
3105 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3106 d_move(old_dentry,new_dentry);
3110 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3111 struct inode *new_dir, struct dentry *new_dentry)
3113 struct inode *target;
3116 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3121 target = new_dentry->d_inode;
3123 mutex_lock(&target->i_mutex);
3124 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3127 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3130 dont_mount(new_dentry);
3131 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3132 d_move(old_dentry, new_dentry);
3135 mutex_unlock(&target->i_mutex);
3140 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3141 struct inode *new_dir, struct dentry *new_dentry)
3144 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3145 const unsigned char *old_name;
3147 if (old_dentry->d_inode == new_dentry->d_inode)
3150 error = may_delete(old_dir, old_dentry, is_dir);
3154 if (!new_dentry->d_inode)
3155 error = may_create(new_dir, new_dentry);
3157 error = may_delete(new_dir, new_dentry, is_dir);
3161 if (!old_dir->i_op->rename)
3164 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3167 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3169 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3171 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3172 new_dentry->d_inode, old_dentry);
3173 fsnotify_oldname_free(old_name);
3178 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3179 int, newdfd, const char __user *, newname)
3181 struct dentry *old_dir, *new_dir;
3182 struct dentry *old_dentry, *new_dentry;
3183 struct dentry *trap;
3184 struct nameidata oldnd, newnd;
3189 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3193 error = user_path_parent(newdfd, newname, &newnd, &to);
3198 if (oldnd.path.mnt != newnd.path.mnt)
3201 old_dir = oldnd.path.dentry;
3203 if (oldnd.last_type != LAST_NORM)
3206 new_dir = newnd.path.dentry;
3207 if (newnd.last_type != LAST_NORM)
3210 oldnd.flags &= ~LOOKUP_PARENT;
3211 newnd.flags &= ~LOOKUP_PARENT;
3212 newnd.flags |= LOOKUP_RENAME_TARGET;
3214 trap = lock_rename(new_dir, old_dir);
3216 old_dentry = lookup_hash(&oldnd);
3217 error = PTR_ERR(old_dentry);
3218 if (IS_ERR(old_dentry))
3220 /* source must exist */
3222 if (!old_dentry->d_inode)
3224 /* unless the source is a directory trailing slashes give -ENOTDIR */
3225 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3227 if (oldnd.last.name[oldnd.last.len])
3229 if (newnd.last.name[newnd.last.len])
3232 /* source should not be ancestor of target */
3234 if (old_dentry == trap)
3236 new_dentry = lookup_hash(&newnd);
3237 error = PTR_ERR(new_dentry);
3238 if (IS_ERR(new_dentry))
3240 /* target should not be an ancestor of source */
3242 if (new_dentry == trap)
3245 error = mnt_want_write(oldnd.path.mnt);
3248 error = security_path_rename(&oldnd.path, old_dentry,
3249 &newnd.path, new_dentry);
3252 error = vfs_rename(old_dir->d_inode, old_dentry,
3253 new_dir->d_inode, new_dentry);
3255 mnt_drop_write(oldnd.path.mnt);
3261 unlock_rename(new_dir, old_dir);
3263 path_put(&newnd.path);
3266 path_put(&oldnd.path);
3272 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3274 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3277 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3281 len = PTR_ERR(link);
3286 if (len > (unsigned) buflen)
3288 if (copy_to_user(buffer, link, len))
3295 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3296 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3297 * using) it for any given inode is up to filesystem.
3299 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3301 struct nameidata nd;
3306 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3308 return PTR_ERR(cookie);
3310 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3311 if (dentry->d_inode->i_op->put_link)
3312 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3316 int vfs_follow_link(struct nameidata *nd, const char *link)
3318 return __vfs_follow_link(nd, link);
3321 /* get the link contents into pagecache */
3322 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3326 struct address_space *mapping = dentry->d_inode->i_mapping;
3327 page = read_mapping_page(mapping, 0, NULL);
3332 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3336 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3338 struct page *page = NULL;
3339 char *s = page_getlink(dentry, &page);
3340 int res = vfs_readlink(dentry,buffer,buflen,s);
3343 page_cache_release(page);
3348 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3350 struct page *page = NULL;
3351 nd_set_link(nd, page_getlink(dentry, &page));
3355 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3357 struct page *page = cookie;
3361 page_cache_release(page);
3366 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3368 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3370 struct address_space *mapping = inode->i_mapping;
3375 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3377 flags |= AOP_FLAG_NOFS;
3380 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3381 flags, &page, &fsdata);
3385 kaddr = kmap_atomic(page, KM_USER0);
3386 memcpy(kaddr, symname, len-1);
3387 kunmap_atomic(kaddr, KM_USER0);
3389 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3396 mark_inode_dirty(inode);
3402 int page_symlink(struct inode *inode, const char *symname, int len)
3404 return __page_symlink(inode, symname, len,
3405 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3408 const struct inode_operations page_symlink_inode_operations = {
3409 .readlink = generic_readlink,
3410 .follow_link = page_follow_link_light,
3411 .put_link = page_put_link,
3414 EXPORT_SYMBOL(user_path_at);
3415 EXPORT_SYMBOL(follow_down);
3416 EXPORT_SYMBOL(follow_up);
3417 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3418 EXPORT_SYMBOL(getname);
3419 EXPORT_SYMBOL(lock_rename);
3420 EXPORT_SYMBOL(lookup_one_len);
3421 EXPORT_SYMBOL(page_follow_link_light);
3422 EXPORT_SYMBOL(page_put_link);
3423 EXPORT_SYMBOL(page_readlink);
3424 EXPORT_SYMBOL(__page_symlink);
3425 EXPORT_SYMBOL(page_symlink);
3426 EXPORT_SYMBOL(page_symlink_inode_operations);
3427 EXPORT_SYMBOL(path_lookup);
3428 EXPORT_SYMBOL(kern_path);
3429 EXPORT_SYMBOL(vfs_path_lookup);
3430 EXPORT_SYMBOL(inode_permission);
3431 EXPORT_SYMBOL(file_permission);
3432 EXPORT_SYMBOL(unlock_rename);
3433 EXPORT_SYMBOL(vfs_create);
3434 EXPORT_SYMBOL(vfs_follow_link);
3435 EXPORT_SYMBOL(vfs_link);
3436 EXPORT_SYMBOL(vfs_mkdir);
3437 EXPORT_SYMBOL(vfs_mknod);
3438 EXPORT_SYMBOL(generic_permission);
3439 EXPORT_SYMBOL(vfs_readlink);
3440 EXPORT_SYMBOL(vfs_rename);
3441 EXPORT_SYMBOL(vfs_rmdir);
3442 EXPORT_SYMBOL(vfs_symlink);
3443 EXPORT_SYMBOL(vfs_unlink);
3444 EXPORT_SYMBOL(dentry_unhash);
3445 EXPORT_SYMBOL(generic_readlink);