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/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.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 <asm/uaccess.h>
36 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
38 /* [Feb-1997 T. Schoebel-Theuer]
39 * Fundamental changes in the pathname lookup mechanisms (namei)
40 * were necessary because of omirr. The reason is that omirr needs
41 * to know the _real_ pathname, not the user-supplied one, in case
42 * of symlinks (and also when transname replacements occur).
44 * The new code replaces the old recursive symlink resolution with
45 * an iterative one (in case of non-nested symlink chains). It does
46 * this with calls to <fs>_follow_link().
47 * As a side effect, dir_namei(), _namei() and follow_link() are now
48 * replaced with a single function lookup_dentry() that can handle all
49 * the special cases of the former code.
51 * With the new dcache, the pathname is stored at each inode, at least as
52 * long as the refcount of the inode is positive. As a side effect, the
53 * size of the dcache depends on the inode cache and thus is dynamic.
55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
56 * resolution to correspond with current state of the code.
58 * Note that the symlink resolution is not *completely* iterative.
59 * There is still a significant amount of tail- and mid- recursion in
60 * the algorithm. Also, note that <fs>_readlink() is not used in
61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
62 * may return different results than <fs>_follow_link(). Many virtual
63 * filesystems (including /proc) exhibit this behavior.
66 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
68 * and the name already exists in form of a symlink, try to create the new
69 * name indicated by the symlink. The old code always complained that the
70 * name already exists, due to not following the symlink even if its target
71 * is nonexistent. The new semantics affects also mknod() and link() when
72 * the name is a symlink pointing to a non-existant name.
74 * I don't know which semantics is the right one, since I have no access
75 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
77 * "old" one. Personally, I think the new semantics is much more logical.
78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
79 * file does succeed in both HP-UX and SunOs, but not in Solaris
80 * and in the old Linux semantics.
83 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
84 * semantics. See the comments in "open_namei" and "do_link" below.
86 * [10-Sep-98 Alan Modra] Another symlink change.
89 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
90 * inside the path - always follow.
91 * in the last component in creation/removal/renaming - never follow.
92 * if LOOKUP_FOLLOW passed - follow.
93 * if the pathname has trailing slashes - follow.
94 * otherwise - don't follow.
95 * (applied in that order).
97 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
98 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
99 * During the 2.4 we need to fix the userland stuff depending on it -
100 * hopefully we will be able to get rid of that wart in 2.5. So far only
101 * XEmacs seems to be relying on it...
104 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
105 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
106 * any extra contention...
109 static int __link_path_walk(const char *name, struct nameidata *nd);
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user *filename, char *page)
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
130 retval = strncpy_from_user(page, filename, len);
134 return -ENAMETOOLONG;
140 char * getname(const char __user * filename)
144 result = ERR_PTR(-ENOMEM);
147 int retval = do_getname(filename, tmp);
152 result = ERR_PTR(retval);
155 audit_getname(result);
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name)
162 if (unlikely(!audit_dummy_context()))
167 EXPORT_SYMBOL(putname);
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
182 int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
185 umode_t mode = inode->i_mode;
187 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
189 if (current->fsuid == inode->i_uid)
192 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
193 int error = check_acl(inode, mask);
194 if (error == -EACCES)
195 goto check_capabilities;
196 else if (error != -EAGAIN)
200 if (in_group_p(inode->i_gid))
205 * If the DACs are ok we don't need any capability check.
207 if ((mask & ~mode) == 0)
212 * Read/write DACs are always overridable.
213 * Executable DACs are overridable if at least one exec bit is set.
215 if (!(mask & MAY_EXEC) ||
216 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
217 if (capable(CAP_DAC_OVERRIDE))
221 * Searching includes executable on directories, else just read.
223 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
224 if (capable(CAP_DAC_READ_SEARCH))
230 int permission(struct inode *inode, int mask, struct nameidata *nd)
233 struct vfsmount *mnt = NULL;
238 if (mask & MAY_WRITE) {
239 umode_t mode = inode->i_mode;
242 * Nobody gets write access to a read-only fs.
244 if (IS_RDONLY(inode) &&
245 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
249 * Nobody gets write access to an immutable file.
251 if (IS_IMMUTABLE(inode))
255 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
257 * MAY_EXEC on regular files is denied if the fs is mounted
258 * with the "noexec" flag.
260 if (mnt && (mnt->mnt_flags & MNT_NOEXEC))
264 /* Ordinary permission routines do not understand MAY_APPEND. */
265 if (inode->i_op && inode->i_op->permission) {
266 retval = inode->i_op->permission(inode, mask);
269 * Exec permission on a regular file is denied if none
270 * of the execute bits are set.
272 * This check should be done by the ->permission()
275 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode) &&
276 !(inode->i_mode & S_IXUGO))
280 retval = generic_permission(inode, mask, NULL);
285 retval = devcgroup_inode_permission(inode, mask);
289 return security_inode_permission(inode,
290 mask & (MAY_READ|MAY_WRITE|MAY_EXEC));
294 * vfs_permission - check for access rights to a given path
295 * @nd: lookup result that describes the path
296 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
298 * Used to check for read/write/execute permissions on a path.
299 * We use "fsuid" for this, letting us set arbitrary permissions
300 * for filesystem access without changing the "normal" uids which
301 * are used for other things.
303 int vfs_permission(struct nameidata *nd, int mask)
305 return permission(nd->path.dentry->d_inode, mask, nd);
309 * file_permission - check for additional access rights to a given file
310 * @file: file to check access rights for
311 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
313 * Used to check for read/write/execute permissions on an already opened
317 * Do not use this function in new code. All access checks should
318 * be done using vfs_permission().
320 int file_permission(struct file *file, int mask)
322 return permission(file->f_path.dentry->d_inode, mask, NULL);
326 * get_write_access() gets write permission for a file.
327 * put_write_access() releases this write permission.
328 * This is used for regular files.
329 * We cannot support write (and maybe mmap read-write shared) accesses and
330 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
331 * can have the following values:
332 * 0: no writers, no VM_DENYWRITE mappings
333 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
334 * > 0: (i_writecount) users are writing to the file.
336 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
337 * except for the cases where we don't hold i_writecount yet. Then we need to
338 * use {get,deny}_write_access() - these functions check the sign and refuse
339 * to do the change if sign is wrong. Exclusion between them is provided by
340 * the inode->i_lock spinlock.
343 int get_write_access(struct inode * inode)
345 spin_lock(&inode->i_lock);
346 if (atomic_read(&inode->i_writecount) < 0) {
347 spin_unlock(&inode->i_lock);
350 atomic_inc(&inode->i_writecount);
351 spin_unlock(&inode->i_lock);
356 int deny_write_access(struct file * file)
358 struct inode *inode = file->f_path.dentry->d_inode;
360 spin_lock(&inode->i_lock);
361 if (atomic_read(&inode->i_writecount) > 0) {
362 spin_unlock(&inode->i_lock);
365 atomic_dec(&inode->i_writecount);
366 spin_unlock(&inode->i_lock);
372 * path_get - get a reference to a path
373 * @path: path to get the reference to
375 * Given a path increment the reference count to the dentry and the vfsmount.
377 void path_get(struct path *path)
382 EXPORT_SYMBOL(path_get);
385 * path_put - put a reference to a path
386 * @path: path to put the reference to
388 * Given a path decrement the reference count to the dentry and the vfsmount.
390 void path_put(struct path *path)
395 EXPORT_SYMBOL(path_put);
398 * release_open_intent - free up open intent resources
399 * @nd: pointer to nameidata
401 void release_open_intent(struct nameidata *nd)
403 if (nd->intent.open.file->f_path.dentry == NULL)
404 put_filp(nd->intent.open.file);
406 fput(nd->intent.open.file);
409 static inline struct dentry *
410 do_revalidate(struct dentry *dentry, struct nameidata *nd)
412 int status = dentry->d_op->d_revalidate(dentry, nd);
413 if (unlikely(status <= 0)) {
415 * The dentry failed validation.
416 * If d_revalidate returned 0 attempt to invalidate
417 * the dentry otherwise d_revalidate is asking us
418 * to return a fail status.
421 if (!d_invalidate(dentry)) {
427 dentry = ERR_PTR(status);
434 * Internal lookup() using the new generic dcache.
437 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
439 struct dentry * dentry = __d_lookup(parent, name);
441 /* lockess __d_lookup may fail due to concurrent d_move()
442 * in some unrelated directory, so try with d_lookup
445 dentry = d_lookup(parent, name);
447 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
448 dentry = do_revalidate(dentry, nd);
454 * Short-cut version of permission(), for calling by
455 * path_walk(), when dcache lock is held. Combines parts
456 * of permission() and generic_permission(), and tests ONLY for
457 * MAY_EXEC permission.
459 * If appropriate, check DAC only. If not appropriate, or
460 * short-cut DAC fails, then call permission() to do more
461 * complete permission check.
463 static int exec_permission_lite(struct inode *inode)
465 umode_t mode = inode->i_mode;
467 if (inode->i_op && inode->i_op->permission)
470 if (current->fsuid == inode->i_uid)
472 else if (in_group_p(inode->i_gid))
478 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
481 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
484 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
489 return security_inode_permission(inode, MAY_EXEC);
493 * This is called when everything else fails, and we actually have
494 * to go to the low-level filesystem to find out what we should do..
496 * We get the directory semaphore, and after getting that we also
497 * make sure that nobody added the entry to the dcache in the meantime..
500 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
502 struct dentry * result;
503 struct inode *dir = parent->d_inode;
505 mutex_lock(&dir->i_mutex);
507 * First re-do the cached lookup just in case it was created
508 * while we waited for the directory semaphore..
510 * FIXME! This could use version numbering or similar to
511 * avoid unnecessary cache lookups.
513 * The "dcache_lock" is purely to protect the RCU list walker
514 * from concurrent renames at this point (we mustn't get false
515 * negatives from the RCU list walk here, unlike the optimistic
518 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
520 result = d_lookup(parent, name);
522 struct dentry *dentry;
524 /* Don't create child dentry for a dead directory. */
525 result = ERR_PTR(-ENOENT);
529 dentry = d_alloc(parent, name);
530 result = ERR_PTR(-ENOMEM);
532 result = dir->i_op->lookup(dir, dentry, nd);
539 mutex_unlock(&dir->i_mutex);
544 * Uhhuh! Nasty case: the cache was re-populated while
545 * we waited on the semaphore. Need to revalidate.
547 mutex_unlock(&dir->i_mutex);
548 if (result->d_op && result->d_op->d_revalidate) {
549 result = do_revalidate(result, nd);
551 result = ERR_PTR(-ENOENT);
557 static __always_inline void
558 walk_init_root(const char *name, struct nameidata *nd)
560 struct fs_struct *fs = current->fs;
562 read_lock(&fs->lock);
565 read_unlock(&fs->lock);
569 * Wrapper to retry pathname resolution whenever the underlying
570 * file system returns an ESTALE.
572 * Retry the whole path once, forcing real lookup requests
573 * instead of relying on the dcache.
575 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
577 struct path save = nd->path;
580 /* make sure the stuff we saved doesn't go away */
583 result = __link_path_walk(name, nd);
584 if (result == -ESTALE) {
585 /* nd->path had been dropped */
588 nd->flags |= LOOKUP_REVAL;
589 result = __link_path_walk(name, nd);
597 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
606 walk_init_root(link, nd);
608 res = link_path_walk(link, nd);
609 if (nd->depth || res || nd->last_type!=LAST_NORM)
612 * If it is an iterative symlinks resolution in open_namei() we
613 * have to copy the last component. And all that crap because of
614 * bloody create() on broken symlinks. Furrfu...
617 if (unlikely(!name)) {
621 strcpy(name, nd->last.name);
622 nd->last.name = name;
626 return PTR_ERR(link);
629 static void path_put_conditional(struct path *path, struct nameidata *nd)
632 if (path->mnt != nd->path.mnt)
636 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
638 dput(nd->path.dentry);
639 if (nd->path.mnt != path->mnt)
640 mntput(nd->path.mnt);
641 nd->path.mnt = path->mnt;
642 nd->path.dentry = path->dentry;
645 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
649 struct dentry *dentry = path->dentry;
651 touch_atime(path->mnt, dentry);
652 nd_set_link(nd, NULL);
654 if (path->mnt != nd->path.mnt) {
655 path_to_nameidata(path, nd);
659 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
660 error = PTR_ERR(cookie);
661 if (!IS_ERR(cookie)) {
662 char *s = nd_get_link(nd);
665 error = __vfs_follow_link(nd, s);
666 if (dentry->d_inode->i_op->put_link)
667 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
675 * This limits recursive symlink follows to 8, while
676 * limiting consecutive symlinks to 40.
678 * Without that kind of total limit, nasty chains of consecutive
679 * symlinks can cause almost arbitrarily long lookups.
681 static inline int do_follow_link(struct path *path, struct nameidata *nd)
684 if (current->link_count >= MAX_NESTED_LINKS)
686 if (current->total_link_count >= 40)
688 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
690 err = security_inode_follow_link(path->dentry, nd);
693 current->link_count++;
694 current->total_link_count++;
696 err = __do_follow_link(path, nd);
697 current->link_count--;
701 path_put_conditional(path, nd);
706 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
708 struct vfsmount *parent;
709 struct dentry *mountpoint;
710 spin_lock(&vfsmount_lock);
711 parent=(*mnt)->mnt_parent;
712 if (parent == *mnt) {
713 spin_unlock(&vfsmount_lock);
717 mountpoint=dget((*mnt)->mnt_mountpoint);
718 spin_unlock(&vfsmount_lock);
720 *dentry = mountpoint;
726 /* no need for dcache_lock, as serialization is taken care in
729 static int __follow_mount(struct path *path)
732 while (d_mountpoint(path->dentry)) {
733 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
740 path->dentry = dget(mounted->mnt_root);
746 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
748 while (d_mountpoint(*dentry)) {
749 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
755 *dentry = dget(mounted->mnt_root);
759 /* no need for dcache_lock, as serialization is taken care in
762 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
764 struct vfsmount *mounted;
766 mounted = lookup_mnt(*mnt, *dentry);
771 *dentry = dget(mounted->mnt_root);
777 static __always_inline void follow_dotdot(struct nameidata *nd)
779 struct fs_struct *fs = current->fs;
782 struct vfsmount *parent;
783 struct dentry *old = nd->path.dentry;
785 read_lock(&fs->lock);
786 if (nd->path.dentry == fs->root.dentry &&
787 nd->path.mnt == fs->root.mnt) {
788 read_unlock(&fs->lock);
791 read_unlock(&fs->lock);
792 spin_lock(&dcache_lock);
793 if (nd->path.dentry != nd->path.mnt->mnt_root) {
794 nd->path.dentry = dget(nd->path.dentry->d_parent);
795 spin_unlock(&dcache_lock);
799 spin_unlock(&dcache_lock);
800 spin_lock(&vfsmount_lock);
801 parent = nd->path.mnt->mnt_parent;
802 if (parent == nd->path.mnt) {
803 spin_unlock(&vfsmount_lock);
807 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
808 spin_unlock(&vfsmount_lock);
810 mntput(nd->path.mnt);
811 nd->path.mnt = parent;
813 follow_mount(&nd->path.mnt, &nd->path.dentry);
817 * It's more convoluted than I'd like it to be, but... it's still fairly
818 * small and for now I'd prefer to have fast path as straight as possible.
819 * It _is_ time-critical.
821 static int do_lookup(struct nameidata *nd, struct qstr *name,
824 struct vfsmount *mnt = nd->path.mnt;
825 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
829 if (dentry->d_op && dentry->d_op->d_revalidate)
830 goto need_revalidate;
833 path->dentry = dentry;
834 __follow_mount(path);
838 dentry = real_lookup(nd->path.dentry, name, nd);
844 dentry = do_revalidate(dentry, nd);
852 return PTR_ERR(dentry);
857 * This is the basic name resolution function, turning a pathname into
858 * the final dentry. We expect 'base' to be positive and a directory.
860 * Returns 0 and nd will have valid dentry and mnt on success.
861 * Returns error and drops reference to input namei data on failure.
863 static int __link_path_walk(const char *name, struct nameidata *nd)
868 unsigned int lookup_flags = nd->flags;
875 inode = nd->path.dentry->d_inode;
877 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
879 /* At this point we know we have a real path component. */
885 nd->flags |= LOOKUP_CONTINUE;
886 err = exec_permission_lite(inode);
888 err = vfs_permission(nd, MAY_EXEC);
893 c = *(const unsigned char *)name;
895 hash = init_name_hash();
898 hash = partial_name_hash(c, hash);
899 c = *(const unsigned char *)name;
900 } while (c && (c != '/'));
901 this.len = name - (const char *) this.name;
902 this.hash = end_name_hash(hash);
904 /* remove trailing slashes? */
907 while (*++name == '/');
909 goto last_with_slashes;
912 * "." and ".." are special - ".." especially so because it has
913 * to be able to know about the current root directory and
914 * parent relationships.
916 if (this.name[0] == '.') switch (this.len) {
920 if (this.name[1] != '.')
923 inode = nd->path.dentry->d_inode;
929 * See if the low-level filesystem might want
930 * to use its own hash..
932 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
933 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
938 /* This does the actual lookups.. */
939 err = do_lookup(nd, &this, &next);
944 inode = next.dentry->d_inode;
951 if (inode->i_op->follow_link) {
952 err = do_follow_link(&next, nd);
956 inode = nd->path.dentry->d_inode;
963 path_to_nameidata(&next, nd);
965 if (!inode->i_op->lookup)
968 /* here ends the main loop */
971 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
973 /* Clear LOOKUP_CONTINUE iff it was previously unset */
974 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
975 if (lookup_flags & LOOKUP_PARENT)
977 if (this.name[0] == '.') switch (this.len) {
981 if (this.name[1] != '.')
984 inode = nd->path.dentry->d_inode;
989 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
990 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
995 err = do_lookup(nd, &this, &next);
998 inode = next.dentry->d_inode;
999 if ((lookup_flags & LOOKUP_FOLLOW)
1000 && inode && inode->i_op && inode->i_op->follow_link) {
1001 err = do_follow_link(&next, nd);
1004 inode = nd->path.dentry->d_inode;
1006 path_to_nameidata(&next, nd);
1010 if (lookup_flags & LOOKUP_DIRECTORY) {
1012 if (!inode->i_op || !inode->i_op->lookup)
1018 nd->last_type = LAST_NORM;
1019 if (this.name[0] != '.')
1022 nd->last_type = LAST_DOT;
1023 else if (this.len == 2 && this.name[1] == '.')
1024 nd->last_type = LAST_DOTDOT;
1029 * We bypassed the ordinary revalidation routines.
1030 * We may need to check the cached dentry for staleness.
1032 if (nd->path.dentry && nd->path.dentry->d_sb &&
1033 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1035 /* Note: we do not d_invalidate() */
1036 if (!nd->path.dentry->d_op->d_revalidate(
1037 nd->path.dentry, nd))
1043 path_put_conditional(&next, nd);
1046 path_put(&nd->path);
1051 static int path_walk(const char *name, struct nameidata *nd)
1053 current->total_link_count = 0;
1054 return link_path_walk(name, nd);
1057 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1058 static int do_path_lookup(int dfd, const char *name,
1059 unsigned int flags, struct nameidata *nd)
1064 struct fs_struct *fs = current->fs;
1066 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1071 read_lock(&fs->lock);
1072 nd->path = fs->root;
1073 path_get(&fs->root);
1074 read_unlock(&fs->lock);
1075 } else if (dfd == AT_FDCWD) {
1076 read_lock(&fs->lock);
1079 read_unlock(&fs->lock);
1081 struct dentry *dentry;
1083 file = fget_light(dfd, &fput_needed);
1088 dentry = file->f_path.dentry;
1091 if (!S_ISDIR(dentry->d_inode->i_mode))
1094 retval = file_permission(file, MAY_EXEC);
1098 nd->path = file->f_path;
1099 path_get(&file->f_path);
1101 fput_light(file, fput_needed);
1104 retval = path_walk(name, nd);
1105 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1106 nd->path.dentry->d_inode))
1107 audit_inode(name, nd->path.dentry);
1112 fput_light(file, fput_needed);
1116 int path_lookup(const char *name, unsigned int flags,
1117 struct nameidata *nd)
1119 return do_path_lookup(AT_FDCWD, name, flags, nd);
1123 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1124 * @dentry: pointer to dentry of the base directory
1125 * @mnt: pointer to vfs mount of the base directory
1126 * @name: pointer to file name
1127 * @flags: lookup flags
1128 * @nd: pointer to nameidata
1130 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1131 const char *name, unsigned int flags,
1132 struct nameidata *nd)
1136 /* same as do_path_lookup */
1137 nd->last_type = LAST_ROOT;
1141 nd->path.dentry = dentry;
1143 path_get(&nd->path);
1145 retval = path_walk(name, nd);
1146 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1147 nd->path.dentry->d_inode))
1148 audit_inode(name, nd->path.dentry);
1154 static int __path_lookup_intent_open(int dfd, const char *name,
1155 unsigned int lookup_flags, struct nameidata *nd,
1156 int open_flags, int create_mode)
1158 struct file *filp = get_empty_filp();
1163 nd->intent.open.file = filp;
1164 nd->intent.open.flags = open_flags;
1165 nd->intent.open.create_mode = create_mode;
1166 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1167 if (IS_ERR(nd->intent.open.file)) {
1169 err = PTR_ERR(nd->intent.open.file);
1170 path_put(&nd->path);
1172 } else if (err != 0)
1173 release_open_intent(nd);
1178 * path_lookup_open - lookup a file path with open intent
1179 * @dfd: the directory to use as base, or AT_FDCWD
1180 * @name: pointer to file name
1181 * @lookup_flags: lookup intent flags
1182 * @nd: pointer to nameidata
1183 * @open_flags: open intent flags
1185 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1186 struct nameidata *nd, int open_flags)
1188 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1193 * path_lookup_create - lookup a file path with open + create intent
1194 * @dfd: the directory to use as base, or AT_FDCWD
1195 * @name: pointer to file name
1196 * @lookup_flags: lookup intent flags
1197 * @nd: pointer to nameidata
1198 * @open_flags: open intent flags
1199 * @create_mode: create intent flags
1201 static int path_lookup_create(int dfd, const char *name,
1202 unsigned int lookup_flags, struct nameidata *nd,
1203 int open_flags, int create_mode)
1205 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1206 nd, open_flags, create_mode);
1209 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1210 struct nameidata *nd, int open_flags)
1212 char *tmp = getname(name);
1213 int err = PTR_ERR(tmp);
1216 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1222 static struct dentry *__lookup_hash(struct qstr *name,
1223 struct dentry *base, struct nameidata *nd)
1225 struct dentry *dentry;
1226 struct inode *inode;
1229 inode = base->d_inode;
1232 * See if the low-level filesystem might want
1233 * to use its own hash..
1235 if (base->d_op && base->d_op->d_hash) {
1236 err = base->d_op->d_hash(base, name);
1237 dentry = ERR_PTR(err);
1242 dentry = cached_lookup(base, name, nd);
1246 /* Don't create child dentry for a dead directory. */
1247 dentry = ERR_PTR(-ENOENT);
1248 if (IS_DEADDIR(inode))
1251 new = d_alloc(base, name);
1252 dentry = ERR_PTR(-ENOMEM);
1255 dentry = inode->i_op->lookup(inode, new, nd);
1266 * Restricted form of lookup. Doesn't follow links, single-component only,
1267 * needs parent already locked. Doesn't follow mounts.
1270 static struct dentry *lookup_hash(struct nameidata *nd)
1274 err = permission(nd->path.dentry->d_inode, MAY_EXEC, nd);
1276 return ERR_PTR(err);
1277 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1280 static int __lookup_one_len(const char *name, struct qstr *this,
1281 struct dentry *base, int len)
1291 hash = init_name_hash();
1293 c = *(const unsigned char *)name++;
1294 if (c == '/' || c == '\0')
1296 hash = partial_name_hash(c, hash);
1298 this->hash = end_name_hash(hash);
1303 * lookup_one_len - filesystem helper to lookup single pathname component
1304 * @name: pathname component to lookup
1305 * @base: base directory to lookup from
1306 * @len: maximum length @len should be interpreted to
1308 * Note that this routine is purely a helper for filesystem usage and should
1309 * not be called by generic code. Also note that by using this function the
1310 * nameidata argument is passed to the filesystem methods and a filesystem
1311 * using this helper needs to be prepared for that.
1313 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1318 err = __lookup_one_len(name, &this, base, len);
1320 return ERR_PTR(err);
1322 err = permission(base->d_inode, MAY_EXEC, NULL);
1324 return ERR_PTR(err);
1325 return __lookup_hash(&this, base, NULL);
1329 * lookup_one_noperm - bad hack for sysfs
1330 * @name: pathname component to lookup
1331 * @base: base directory to lookup from
1333 * This is a variant of lookup_one_len that doesn't perform any permission
1334 * checks. It's a horrible hack to work around the braindead sysfs
1335 * architecture and should not be used anywhere else.
1337 * DON'T USE THIS FUNCTION EVER, thanks.
1339 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1344 err = __lookup_one_len(name, &this, base, strlen(name));
1346 return ERR_PTR(err);
1347 return __lookup_hash(&this, base, NULL);
1350 int __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1351 struct nameidata *nd)
1353 char *tmp = getname(name);
1354 int err = PTR_ERR(tmp);
1357 err = do_path_lookup(dfd, tmp, flags, nd);
1363 int __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1365 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1369 * It's inline, so penalty for filesystems that don't use sticky bit is
1372 static inline int check_sticky(struct inode *dir, struct inode *inode)
1374 if (!(dir->i_mode & S_ISVTX))
1376 if (inode->i_uid == current->fsuid)
1378 if (dir->i_uid == current->fsuid)
1380 return !capable(CAP_FOWNER);
1384 * Check whether we can remove a link victim from directory dir, check
1385 * whether the type of victim is right.
1386 * 1. We can't do it if dir is read-only (done in permission())
1387 * 2. We should have write and exec permissions on dir
1388 * 3. We can't remove anything from append-only dir
1389 * 4. We can't do anything with immutable dir (done in permission())
1390 * 5. If the sticky bit on dir is set we should either
1391 * a. be owner of dir, or
1392 * b. be owner of victim, or
1393 * c. have CAP_FOWNER capability
1394 * 6. If the victim is append-only or immutable we can't do antyhing with
1395 * links pointing to it.
1396 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1397 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1398 * 9. We can't remove a root or mountpoint.
1399 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1400 * nfs_async_unlink().
1402 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1406 if (!victim->d_inode)
1409 BUG_ON(victim->d_parent->d_inode != dir);
1410 audit_inode_child(victim->d_name.name, victim, dir);
1412 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1417 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1418 IS_IMMUTABLE(victim->d_inode))
1421 if (!S_ISDIR(victim->d_inode->i_mode))
1423 if (IS_ROOT(victim))
1425 } else if (S_ISDIR(victim->d_inode->i_mode))
1427 if (IS_DEADDIR(dir))
1429 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1434 /* Check whether we can create an object with dentry child in directory
1436 * 1. We can't do it if child already exists (open has special treatment for
1437 * this case, but since we are inlined it's OK)
1438 * 2. We can't do it if dir is read-only (done in permission())
1439 * 3. We should have write and exec permissions on dir
1440 * 4. We can't do it if dir is immutable (done in permission())
1442 static inline int may_create(struct inode *dir, struct dentry *child,
1443 struct nameidata *nd)
1447 if (IS_DEADDIR(dir))
1449 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1453 * O_DIRECTORY translates into forcing a directory lookup.
1455 static inline int lookup_flags(unsigned int f)
1457 unsigned long retval = LOOKUP_FOLLOW;
1460 retval &= ~LOOKUP_FOLLOW;
1462 if (f & O_DIRECTORY)
1463 retval |= LOOKUP_DIRECTORY;
1469 * p1 and p2 should be directories on the same fs.
1471 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1476 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1480 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1482 for (p = p1; p->d_parent != p; p = p->d_parent) {
1483 if (p->d_parent == p2) {
1484 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1485 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1490 for (p = p2; p->d_parent != p; p = p->d_parent) {
1491 if (p->d_parent == p1) {
1492 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1493 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1498 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1499 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1503 void unlock_rename(struct dentry *p1, struct dentry *p2)
1505 mutex_unlock(&p1->d_inode->i_mutex);
1507 mutex_unlock(&p2->d_inode->i_mutex);
1508 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1512 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1513 struct nameidata *nd)
1515 int error = may_create(dir, dentry, nd);
1520 if (!dir->i_op || !dir->i_op->create)
1521 return -EACCES; /* shouldn't it be ENOSYS? */
1524 error = security_inode_create(dir, dentry, mode);
1528 error = dir->i_op->create(dir, dentry, mode, nd);
1530 fsnotify_create(dir, dentry);
1534 int may_open(struct nameidata *nd, int acc_mode, int flag)
1536 struct dentry *dentry = nd->path.dentry;
1537 struct inode *inode = dentry->d_inode;
1543 if (S_ISLNK(inode->i_mode))
1546 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
1550 * FIFO's, sockets and device files are special: they don't
1551 * actually live on the filesystem itself, and as such you
1552 * can write to them even if the filesystem is read-only.
1554 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1556 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1557 if (nd->path.mnt->mnt_flags & MNT_NODEV)
1563 error = vfs_permission(nd, acc_mode);
1567 * An append-only file must be opened in append mode for writing.
1569 if (IS_APPEND(inode)) {
1570 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1576 /* O_NOATIME can only be set by the owner or superuser */
1577 if (flag & O_NOATIME)
1578 if (!is_owner_or_cap(inode))
1582 * Ensure there are no outstanding leases on the file.
1584 error = break_lease(inode, flag);
1588 if (flag & O_TRUNC) {
1589 error = get_write_access(inode);
1594 * Refuse to truncate files with mandatory locks held on them.
1596 error = locks_verify_locked(inode);
1600 error = do_truncate(dentry, 0,
1601 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1604 put_write_access(inode);
1608 if (flag & FMODE_WRITE)
1615 * Be careful about ever adding any more callers of this
1616 * function. Its flags must be in the namei format, not
1617 * what get passed to sys_open().
1619 static int __open_namei_create(struct nameidata *nd, struct path *path,
1623 struct dentry *dir = nd->path.dentry;
1625 if (!IS_POSIXACL(dir->d_inode))
1626 mode &= ~current->fs->umask;
1627 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1628 mutex_unlock(&dir->d_inode->i_mutex);
1629 dput(nd->path.dentry);
1630 nd->path.dentry = path->dentry;
1633 /* Don't check for write permission, don't truncate */
1634 return may_open(nd, 0, flag & ~O_TRUNC);
1638 * Note that while the flag value (low two bits) for sys_open means:
1643 * it is changed into
1644 * 00 - no permissions needed
1645 * 01 - read-permission
1646 * 10 - write-permission
1648 * for the internal routines (ie open_namei()/follow_link() etc)
1649 * This is more logical, and also allows the 00 "no perm needed"
1650 * to be used for symlinks (where the permissions are checked
1654 static inline int open_to_namei_flags(int flag)
1656 if ((flag+1) & O_ACCMODE)
1661 static int open_will_write_to_fs(int flag, struct inode *inode)
1664 * We'll never write to the fs underlying
1667 if (special_file(inode->i_mode))
1669 return (flag & O_TRUNC);
1673 * Note that the low bits of the passed in "open_flag"
1674 * are not the same as in the local variable "flag". See
1675 * open_to_namei_flags() for more details.
1677 struct file *do_filp_open(int dfd, const char *pathname,
1678 int open_flag, int mode)
1681 struct nameidata nd;
1682 int acc_mode, error;
1687 int flag = open_to_namei_flags(open_flag);
1689 acc_mode = MAY_OPEN | ACC_MODE(flag);
1691 /* O_TRUNC implies we need access checks for write permissions */
1693 acc_mode |= MAY_WRITE;
1695 /* Allow the LSM permission hook to distinguish append
1696 access from general write access. */
1697 if (flag & O_APPEND)
1698 acc_mode |= MAY_APPEND;
1701 * The simplest case - just a plain lookup.
1703 if (!(flag & O_CREAT)) {
1704 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1707 return ERR_PTR(error);
1712 * Create - we need to know the parent.
1714 error = path_lookup_create(dfd, pathname, LOOKUP_PARENT,
1717 return ERR_PTR(error);
1720 * We have the parent and last component. First of all, check
1721 * that we are not asked to creat(2) an obvious directory - that
1725 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1728 dir = nd.path.dentry;
1729 nd.flags &= ~LOOKUP_PARENT;
1730 mutex_lock(&dir->d_inode->i_mutex);
1731 path.dentry = lookup_hash(&nd);
1732 path.mnt = nd.path.mnt;
1735 error = PTR_ERR(path.dentry);
1736 if (IS_ERR(path.dentry)) {
1737 mutex_unlock(&dir->d_inode->i_mutex);
1741 if (IS_ERR(nd.intent.open.file)) {
1742 error = PTR_ERR(nd.intent.open.file);
1743 goto exit_mutex_unlock;
1746 /* Negative dentry, just create the file */
1747 if (!path.dentry->d_inode) {
1749 * This write is needed to ensure that a
1750 * ro->rw transition does not occur between
1751 * the time when the file is created and when
1752 * a permanent write count is taken through
1753 * the 'struct file' in nameidata_to_filp().
1755 error = mnt_want_write(nd.path.mnt);
1757 goto exit_mutex_unlock;
1758 error = __open_namei_create(&nd, &path, flag, mode);
1760 mnt_drop_write(nd.path.mnt);
1763 filp = nameidata_to_filp(&nd, open_flag);
1764 mnt_drop_write(nd.path.mnt);
1769 * It already exists.
1771 mutex_unlock(&dir->d_inode->i_mutex);
1772 audit_inode(pathname, path.dentry);
1778 if (__follow_mount(&path)) {
1780 if (flag & O_NOFOLLOW)
1785 if (!path.dentry->d_inode)
1787 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1790 path_to_nameidata(&path, &nd);
1792 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1797 * 1. may_open() truncates a file
1798 * 2. a rw->ro mount transition occurs
1799 * 3. nameidata_to_filp() fails due to
1801 * That would be inconsistent, and should
1802 * be avoided. Taking this mnt write here
1803 * ensures that (2) can not occur.
1805 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1807 error = mnt_want_write(nd.path.mnt);
1811 error = may_open(&nd, acc_mode, flag);
1814 mnt_drop_write(nd.path.mnt);
1817 filp = nameidata_to_filp(&nd, open_flag);
1819 * It is now safe to drop the mnt write
1820 * because the filp has had a write taken
1824 mnt_drop_write(nd.path.mnt);
1828 mutex_unlock(&dir->d_inode->i_mutex);
1830 path_put_conditional(&path, &nd);
1832 if (!IS_ERR(nd.intent.open.file))
1833 release_open_intent(&nd);
1835 return ERR_PTR(error);
1839 if (flag & O_NOFOLLOW)
1842 * This is subtle. Instead of calling do_follow_link() we do the
1843 * thing by hands. The reason is that this way we have zero link_count
1844 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1845 * After that we have the parent and last component, i.e.
1846 * we are in the same situation as after the first path_walk().
1847 * Well, almost - if the last component is normal we get its copy
1848 * stored in nd->last.name and we will have to putname() it when we
1849 * are done. Procfs-like symlinks just set LAST_BIND.
1851 nd.flags |= LOOKUP_PARENT;
1852 error = security_inode_follow_link(path.dentry, &nd);
1855 error = __do_follow_link(&path, &nd);
1857 /* Does someone understand code flow here? Or it is only
1858 * me so stupid? Anathema to whoever designed this non-sense
1859 * with "intent.open".
1861 release_open_intent(&nd);
1862 return ERR_PTR(error);
1864 nd.flags &= ~LOOKUP_PARENT;
1865 if (nd.last_type == LAST_BIND)
1868 if (nd.last_type != LAST_NORM)
1870 if (nd.last.name[nd.last.len]) {
1871 __putname(nd.last.name);
1876 __putname(nd.last.name);
1879 dir = nd.path.dentry;
1880 mutex_lock(&dir->d_inode->i_mutex);
1881 path.dentry = lookup_hash(&nd);
1882 path.mnt = nd.path.mnt;
1883 __putname(nd.last.name);
1888 * filp_open - open file and return file pointer
1890 * @filename: path to open
1891 * @flags: open flags as per the open(2) second argument
1892 * @mode: mode for the new file if O_CREAT is set, else ignored
1894 * This is the helper to open a file from kernelspace if you really
1895 * have to. But in generally you should not do this, so please move
1896 * along, nothing to see here..
1898 struct file *filp_open(const char *filename, int flags, int mode)
1900 return do_filp_open(AT_FDCWD, filename, flags, mode);
1902 EXPORT_SYMBOL(filp_open);
1905 * lookup_create - lookup a dentry, creating it if it doesn't exist
1906 * @nd: nameidata info
1907 * @is_dir: directory flag
1909 * Simple function to lookup and return a dentry and create it
1910 * if it doesn't exist. Is SMP-safe.
1912 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1914 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1916 struct dentry *dentry = ERR_PTR(-EEXIST);
1918 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1920 * Yucky last component or no last component at all?
1921 * (foo/., foo/.., /////)
1923 if (nd->last_type != LAST_NORM)
1925 nd->flags &= ~LOOKUP_PARENT;
1926 nd->flags |= LOOKUP_CREATE;
1927 nd->intent.open.flags = O_EXCL;
1930 * Do the final lookup.
1932 dentry = lookup_hash(nd);
1936 if (dentry->d_inode)
1939 * Special case - lookup gave negative, but... we had foo/bar/
1940 * From the vfs_mknod() POV we just have a negative dentry -
1941 * all is fine. Let's be bastards - you had / on the end, you've
1942 * been asking for (non-existent) directory. -ENOENT for you.
1944 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1946 dentry = ERR_PTR(-ENOENT);
1951 dentry = ERR_PTR(-EEXIST);
1955 EXPORT_SYMBOL_GPL(lookup_create);
1957 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1959 int error = may_create(dir, dentry, NULL);
1964 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1967 if (!dir->i_op || !dir->i_op->mknod)
1970 error = devcgroup_inode_mknod(mode, dev);
1974 error = security_inode_mknod(dir, dentry, mode, dev);
1979 error = dir->i_op->mknod(dir, dentry, mode, dev);
1981 fsnotify_create(dir, dentry);
1985 static int may_mknod(mode_t mode)
1987 switch (mode & S_IFMT) {
1993 case 0: /* zero mode translates to S_IFREG */
2002 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
2007 struct dentry * dentry;
2008 struct nameidata nd;
2012 tmp = getname(filename);
2014 return PTR_ERR(tmp);
2016 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
2019 dentry = lookup_create(&nd, 0);
2020 if (IS_ERR(dentry)) {
2021 error = PTR_ERR(dentry);
2024 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2025 mode &= ~current->fs->umask;
2026 error = may_mknod(mode);
2029 error = mnt_want_write(nd.path.mnt);
2032 switch (mode & S_IFMT) {
2033 case 0: case S_IFREG:
2034 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2036 case S_IFCHR: case S_IFBLK:
2037 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2038 new_decode_dev(dev));
2040 case S_IFIFO: case S_IFSOCK:
2041 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2044 mnt_drop_write(nd.path.mnt);
2048 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2056 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
2058 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2061 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2063 int error = may_create(dir, dentry, NULL);
2068 if (!dir->i_op || !dir->i_op->mkdir)
2071 mode &= (S_IRWXUGO|S_ISVTX);
2072 error = security_inode_mkdir(dir, dentry, mode);
2077 error = dir->i_op->mkdir(dir, dentry, mode);
2079 fsnotify_mkdir(dir, dentry);
2083 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
2087 struct dentry *dentry;
2088 struct nameidata nd;
2090 tmp = getname(pathname);
2091 error = PTR_ERR(tmp);
2095 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
2098 dentry = lookup_create(&nd, 1);
2099 error = PTR_ERR(dentry);
2103 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2104 mode &= ~current->fs->umask;
2105 error = mnt_want_write(nd.path.mnt);
2108 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2109 mnt_drop_write(nd.path.mnt);
2113 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2121 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2123 return sys_mkdirat(AT_FDCWD, pathname, mode);
2127 * We try to drop the dentry early: we should have
2128 * a usage count of 2 if we're the only user of this
2129 * dentry, and if that is true (possibly after pruning
2130 * the dcache), then we drop the dentry now.
2132 * A low-level filesystem can, if it choses, legally
2135 * if (!d_unhashed(dentry))
2138 * if it cannot handle the case of removing a directory
2139 * that is still in use by something else..
2141 void dentry_unhash(struct dentry *dentry)
2144 shrink_dcache_parent(dentry);
2145 spin_lock(&dcache_lock);
2146 spin_lock(&dentry->d_lock);
2147 if (atomic_read(&dentry->d_count) == 2)
2149 spin_unlock(&dentry->d_lock);
2150 spin_unlock(&dcache_lock);
2153 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2155 int error = may_delete(dir, dentry, 1);
2160 if (!dir->i_op || !dir->i_op->rmdir)
2165 mutex_lock(&dentry->d_inode->i_mutex);
2166 dentry_unhash(dentry);
2167 if (d_mountpoint(dentry))
2170 error = security_inode_rmdir(dir, dentry);
2172 error = dir->i_op->rmdir(dir, dentry);
2174 dentry->d_inode->i_flags |= S_DEAD;
2177 mutex_unlock(&dentry->d_inode->i_mutex);
2186 static long do_rmdir(int dfd, const char __user *pathname)
2190 struct dentry *dentry;
2191 struct nameidata nd;
2193 name = getname(pathname);
2195 return PTR_ERR(name);
2197 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2201 switch(nd.last_type) {
2212 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2213 dentry = lookup_hash(&nd);
2214 error = PTR_ERR(dentry);
2217 error = mnt_want_write(nd.path.mnt);
2220 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2221 mnt_drop_write(nd.path.mnt);
2225 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2233 asmlinkage long sys_rmdir(const char __user *pathname)
2235 return do_rmdir(AT_FDCWD, pathname);
2238 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2240 int error = may_delete(dir, dentry, 0);
2245 if (!dir->i_op || !dir->i_op->unlink)
2250 mutex_lock(&dentry->d_inode->i_mutex);
2251 if (d_mountpoint(dentry))
2254 error = security_inode_unlink(dir, dentry);
2256 error = dir->i_op->unlink(dir, dentry);
2258 mutex_unlock(&dentry->d_inode->i_mutex);
2260 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2261 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2262 fsnotify_link_count(dentry->d_inode);
2270 * Make sure that the actual truncation of the file will occur outside its
2271 * directory's i_mutex. Truncate can take a long time if there is a lot of
2272 * writeout happening, and we don't want to prevent access to the directory
2273 * while waiting on the I/O.
2275 static long do_unlinkat(int dfd, const char __user *pathname)
2279 struct dentry *dentry;
2280 struct nameidata nd;
2281 struct inode *inode = NULL;
2283 name = getname(pathname);
2285 return PTR_ERR(name);
2287 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2291 if (nd.last_type != LAST_NORM)
2293 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2294 dentry = lookup_hash(&nd);
2295 error = PTR_ERR(dentry);
2296 if (!IS_ERR(dentry)) {
2297 /* Why not before? Because we want correct error value */
2298 if (nd.last.name[nd.last.len])
2300 inode = dentry->d_inode;
2302 atomic_inc(&inode->i_count);
2303 error = mnt_want_write(nd.path.mnt);
2306 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2307 mnt_drop_write(nd.path.mnt);
2311 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2313 iput(inode); /* truncate the inode here */
2321 error = !dentry->d_inode ? -ENOENT :
2322 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2326 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2328 if ((flag & ~AT_REMOVEDIR) != 0)
2331 if (flag & AT_REMOVEDIR)
2332 return do_rmdir(dfd, pathname);
2334 return do_unlinkat(dfd, pathname);
2337 asmlinkage long sys_unlink(const char __user *pathname)
2339 return do_unlinkat(AT_FDCWD, pathname);
2342 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2344 int error = may_create(dir, dentry, NULL);
2349 if (!dir->i_op || !dir->i_op->symlink)
2352 error = security_inode_symlink(dir, dentry, oldname);
2357 error = dir->i_op->symlink(dir, dentry, oldname);
2359 fsnotify_create(dir, dentry);
2363 asmlinkage long sys_symlinkat(const char __user *oldname,
2364 int newdfd, const char __user *newname)
2369 struct dentry *dentry;
2370 struct nameidata nd;
2372 from = getname(oldname);
2374 return PTR_ERR(from);
2375 to = getname(newname);
2376 error = PTR_ERR(to);
2380 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2383 dentry = lookup_create(&nd, 0);
2384 error = PTR_ERR(dentry);
2388 error = mnt_want_write(nd.path.mnt);
2391 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2392 mnt_drop_write(nd.path.mnt);
2396 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2405 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2407 return sys_symlinkat(oldname, AT_FDCWD, newname);
2410 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2412 struct inode *inode = old_dentry->d_inode;
2418 error = may_create(dir, new_dentry, NULL);
2422 if (dir->i_sb != inode->i_sb)
2426 * A link to an append-only or immutable file cannot be created.
2428 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2430 if (!dir->i_op || !dir->i_op->link)
2432 if (S_ISDIR(inode->i_mode))
2435 error = security_inode_link(old_dentry, dir, new_dentry);
2439 mutex_lock(&inode->i_mutex);
2441 error = dir->i_op->link(old_dentry, dir, new_dentry);
2442 mutex_unlock(&inode->i_mutex);
2444 fsnotify_link(dir, inode, new_dentry);
2449 * Hardlinks are often used in delicate situations. We avoid
2450 * security-related surprises by not following symlinks on the
2453 * We don't follow them on the oldname either to be compatible
2454 * with linux 2.0, and to avoid hard-linking to directories
2455 * and other special files. --ADM
2457 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2458 int newdfd, const char __user *newname,
2461 struct dentry *new_dentry;
2462 struct nameidata nd, old_nd;
2466 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2469 to = getname(newname);
2473 error = __user_walk_fd(olddfd, oldname,
2474 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2478 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2482 if (old_nd.path.mnt != nd.path.mnt)
2484 new_dentry = lookup_create(&nd, 0);
2485 error = PTR_ERR(new_dentry);
2486 if (IS_ERR(new_dentry))
2488 error = mnt_want_write(nd.path.mnt);
2491 error = vfs_link(old_nd.path.dentry, nd.path.dentry->d_inode, new_dentry);
2492 mnt_drop_write(nd.path.mnt);
2496 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2500 path_put(&old_nd.path);
2507 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2509 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2513 * The worst of all namespace operations - renaming directory. "Perverted"
2514 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2516 * a) we can get into loop creation. Check is done in is_subdir().
2517 * b) race potential - two innocent renames can create a loop together.
2518 * That's where 4.4 screws up. Current fix: serialization on
2519 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2521 * c) we have to lock _three_ objects - parents and victim (if it exists).
2522 * And that - after we got ->i_mutex on parents (until then we don't know
2523 * whether the target exists). Solution: try to be smart with locking
2524 * order for inodes. We rely on the fact that tree topology may change
2525 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2526 * move will be locked. Thus we can rank directories by the tree
2527 * (ancestors first) and rank all non-directories after them.
2528 * That works since everybody except rename does "lock parent, lookup,
2529 * lock child" and rename is under ->s_vfs_rename_mutex.
2530 * HOWEVER, it relies on the assumption that any object with ->lookup()
2531 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2532 * we'd better make sure that there's no link(2) for them.
2533 * d) some filesystems don't support opened-but-unlinked directories,
2534 * either because of layout or because they are not ready to deal with
2535 * all cases correctly. The latter will be fixed (taking this sort of
2536 * stuff into VFS), but the former is not going away. Solution: the same
2537 * trick as in rmdir().
2538 * e) conversion from fhandle to dentry may come in the wrong moment - when
2539 * we are removing the target. Solution: we will have to grab ->i_mutex
2540 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2541 * ->i_mutex on parents, which works but leads to some truely excessive
2544 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2545 struct inode *new_dir, struct dentry *new_dentry)
2548 struct inode *target;
2551 * If we are going to change the parent - check write permissions,
2552 * we'll need to flip '..'.
2554 if (new_dir != old_dir) {
2555 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2560 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2564 target = new_dentry->d_inode;
2566 mutex_lock(&target->i_mutex);
2567 dentry_unhash(new_dentry);
2569 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2572 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2575 target->i_flags |= S_DEAD;
2576 mutex_unlock(&target->i_mutex);
2577 if (d_unhashed(new_dentry))
2578 d_rehash(new_dentry);
2582 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2583 d_move(old_dentry,new_dentry);
2587 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2588 struct inode *new_dir, struct dentry *new_dentry)
2590 struct inode *target;
2593 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2598 target = new_dentry->d_inode;
2600 mutex_lock(&target->i_mutex);
2601 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2604 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2606 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2607 d_move(old_dentry, new_dentry);
2610 mutex_unlock(&target->i_mutex);
2615 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2616 struct inode *new_dir, struct dentry *new_dentry)
2619 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2620 const char *old_name;
2622 if (old_dentry->d_inode == new_dentry->d_inode)
2625 error = may_delete(old_dir, old_dentry, is_dir);
2629 if (!new_dentry->d_inode)
2630 error = may_create(new_dir, new_dentry, NULL);
2632 error = may_delete(new_dir, new_dentry, is_dir);
2636 if (!old_dir->i_op || !old_dir->i_op->rename)
2639 DQUOT_INIT(old_dir);
2640 DQUOT_INIT(new_dir);
2642 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2645 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2647 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2649 const char *new_name = old_dentry->d_name.name;
2650 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2651 new_dentry->d_inode, old_dentry);
2653 fsnotify_oldname_free(old_name);
2658 static int do_rename(int olddfd, const char *oldname,
2659 int newdfd, const char *newname)
2662 struct dentry * old_dir, * new_dir;
2663 struct dentry * old_dentry, *new_dentry;
2664 struct dentry * trap;
2665 struct nameidata oldnd, newnd;
2667 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2671 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2676 if (oldnd.path.mnt != newnd.path.mnt)
2679 old_dir = oldnd.path.dentry;
2681 if (oldnd.last_type != LAST_NORM)
2684 new_dir = newnd.path.dentry;
2685 if (newnd.last_type != LAST_NORM)
2688 trap = lock_rename(new_dir, old_dir);
2690 old_dentry = lookup_hash(&oldnd);
2691 error = PTR_ERR(old_dentry);
2692 if (IS_ERR(old_dentry))
2694 /* source must exist */
2696 if (!old_dentry->d_inode)
2698 /* unless the source is a directory trailing slashes give -ENOTDIR */
2699 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2701 if (oldnd.last.name[oldnd.last.len])
2703 if (newnd.last.name[newnd.last.len])
2706 /* source should not be ancestor of target */
2708 if (old_dentry == trap)
2710 new_dentry = lookup_hash(&newnd);
2711 error = PTR_ERR(new_dentry);
2712 if (IS_ERR(new_dentry))
2714 /* target should not be an ancestor of source */
2716 if (new_dentry == trap)
2719 error = mnt_want_write(oldnd.path.mnt);
2722 error = vfs_rename(old_dir->d_inode, old_dentry,
2723 new_dir->d_inode, new_dentry);
2724 mnt_drop_write(oldnd.path.mnt);
2730 unlock_rename(new_dir, old_dir);
2732 path_put(&newnd.path);
2734 path_put(&oldnd.path);
2739 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2740 int newdfd, const char __user *newname)
2746 from = getname(oldname);
2748 return PTR_ERR(from);
2749 to = getname(newname);
2750 error = PTR_ERR(to);
2752 error = do_rename(olddfd, from, newdfd, to);
2759 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2761 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2764 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2768 len = PTR_ERR(link);
2773 if (len > (unsigned) buflen)
2775 if (copy_to_user(buffer, link, len))
2782 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2783 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2784 * using) it for any given inode is up to filesystem.
2786 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2788 struct nameidata nd;
2793 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2795 return PTR_ERR(cookie);
2797 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2798 if (dentry->d_inode->i_op->put_link)
2799 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2803 int vfs_follow_link(struct nameidata *nd, const char *link)
2805 return __vfs_follow_link(nd, link);
2808 /* get the link contents into pagecache */
2809 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2812 struct address_space *mapping = dentry->d_inode->i_mapping;
2813 page = read_mapping_page(mapping, 0, NULL);
2820 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2822 struct page *page = NULL;
2823 char *s = page_getlink(dentry, &page);
2824 int res = vfs_readlink(dentry,buffer,buflen,s);
2827 page_cache_release(page);
2832 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2834 struct page *page = NULL;
2835 nd_set_link(nd, page_getlink(dentry, &page));
2839 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2841 struct page *page = cookie;
2845 page_cache_release(page);
2849 int __page_symlink(struct inode *inode, const char *symname, int len,
2852 struct address_space *mapping = inode->i_mapping;
2859 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2860 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
2864 kaddr = kmap_atomic(page, KM_USER0);
2865 memcpy(kaddr, symname, len-1);
2866 kunmap_atomic(kaddr, KM_USER0);
2868 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2875 mark_inode_dirty(inode);
2881 int page_symlink(struct inode *inode, const char *symname, int len)
2883 return __page_symlink(inode, symname, len,
2884 mapping_gfp_mask(inode->i_mapping));
2887 const struct inode_operations page_symlink_inode_operations = {
2888 .readlink = generic_readlink,
2889 .follow_link = page_follow_link_light,
2890 .put_link = page_put_link,
2893 EXPORT_SYMBOL(__user_walk);
2894 EXPORT_SYMBOL(__user_walk_fd);
2895 EXPORT_SYMBOL(follow_down);
2896 EXPORT_SYMBOL(follow_up);
2897 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2898 EXPORT_SYMBOL(getname);
2899 EXPORT_SYMBOL(lock_rename);
2900 EXPORT_SYMBOL(lookup_one_len);
2901 EXPORT_SYMBOL(page_follow_link_light);
2902 EXPORT_SYMBOL(page_put_link);
2903 EXPORT_SYMBOL(page_readlink);
2904 EXPORT_SYMBOL(__page_symlink);
2905 EXPORT_SYMBOL(page_symlink);
2906 EXPORT_SYMBOL(page_symlink_inode_operations);
2907 EXPORT_SYMBOL(path_lookup);
2908 EXPORT_SYMBOL(vfs_path_lookup);
2909 EXPORT_SYMBOL(permission);
2910 EXPORT_SYMBOL(vfs_permission);
2911 EXPORT_SYMBOL(file_permission);
2912 EXPORT_SYMBOL(unlock_rename);
2913 EXPORT_SYMBOL(vfs_create);
2914 EXPORT_SYMBOL(vfs_follow_link);
2915 EXPORT_SYMBOL(vfs_link);
2916 EXPORT_SYMBOL(vfs_mkdir);
2917 EXPORT_SYMBOL(vfs_mknod);
2918 EXPORT_SYMBOL(generic_permission);
2919 EXPORT_SYMBOL(vfs_readlink);
2920 EXPORT_SYMBOL(vfs_rename);
2921 EXPORT_SYMBOL(vfs_rmdir);
2922 EXPORT_SYMBOL(vfs_symlink);
2923 EXPORT_SYMBOL(vfs_unlink);
2924 EXPORT_SYMBOL(dentry_unhash);
2925 EXPORT_SYMBOL(generic_readlink);