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) {
268 if (nd->flags & LOOKUP_OPEN)
271 retval = inode->i_op->permission(inode, mask | extra);
274 * Exec permission on a regular file is denied if none
275 * of the execute bits are set.
277 * This check should be done by the ->permission()
280 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode) &&
281 !(inode->i_mode & S_IXUGO))
285 retval = generic_permission(inode, mask, NULL);
290 retval = devcgroup_inode_permission(inode, mask);
294 return security_inode_permission(inode,
295 mask & (MAY_READ|MAY_WRITE|MAY_EXEC), nd);
299 * vfs_permission - check for access rights to a given path
300 * @nd: lookup result that describes the path
301 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
303 * Used to check for read/write/execute permissions on a path.
304 * We use "fsuid" for this, letting us set arbitrary permissions
305 * for filesystem access without changing the "normal" uids which
306 * are used for other things.
308 int vfs_permission(struct nameidata *nd, int mask)
310 return permission(nd->path.dentry->d_inode, mask, nd);
314 * file_permission - check for additional access rights to a given file
315 * @file: file to check access rights for
316 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
318 * Used to check for read/write/execute permissions on an already opened
322 * Do not use this function in new code. All access checks should
323 * be done using vfs_permission().
325 int file_permission(struct file *file, int mask)
327 return permission(file->f_path.dentry->d_inode, mask, NULL);
331 * get_write_access() gets write permission for a file.
332 * put_write_access() releases this write permission.
333 * This is used for regular files.
334 * We cannot support write (and maybe mmap read-write shared) accesses and
335 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
336 * can have the following values:
337 * 0: no writers, no VM_DENYWRITE mappings
338 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
339 * > 0: (i_writecount) users are writing to the file.
341 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
342 * except for the cases where we don't hold i_writecount yet. Then we need to
343 * use {get,deny}_write_access() - these functions check the sign and refuse
344 * to do the change if sign is wrong. Exclusion between them is provided by
345 * the inode->i_lock spinlock.
348 int get_write_access(struct inode * inode)
350 spin_lock(&inode->i_lock);
351 if (atomic_read(&inode->i_writecount) < 0) {
352 spin_unlock(&inode->i_lock);
355 atomic_inc(&inode->i_writecount);
356 spin_unlock(&inode->i_lock);
361 int deny_write_access(struct file * file)
363 struct inode *inode = file->f_path.dentry->d_inode;
365 spin_lock(&inode->i_lock);
366 if (atomic_read(&inode->i_writecount) > 0) {
367 spin_unlock(&inode->i_lock);
370 atomic_dec(&inode->i_writecount);
371 spin_unlock(&inode->i_lock);
377 * path_get - get a reference to a path
378 * @path: path to get the reference to
380 * Given a path increment the reference count to the dentry and the vfsmount.
382 void path_get(struct path *path)
387 EXPORT_SYMBOL(path_get);
390 * path_put - put a reference to a path
391 * @path: path to put the reference to
393 * Given a path decrement the reference count to the dentry and the vfsmount.
395 void path_put(struct path *path)
400 EXPORT_SYMBOL(path_put);
403 * release_open_intent - free up open intent resources
404 * @nd: pointer to nameidata
406 void release_open_intent(struct nameidata *nd)
408 if (nd->intent.open.file->f_path.dentry == NULL)
409 put_filp(nd->intent.open.file);
411 fput(nd->intent.open.file);
414 static inline struct dentry *
415 do_revalidate(struct dentry *dentry, struct nameidata *nd)
417 int status = dentry->d_op->d_revalidate(dentry, nd);
418 if (unlikely(status <= 0)) {
420 * The dentry failed validation.
421 * If d_revalidate returned 0 attempt to invalidate
422 * the dentry otherwise d_revalidate is asking us
423 * to return a fail status.
426 if (!d_invalidate(dentry)) {
432 dentry = ERR_PTR(status);
439 * Internal lookup() using the new generic dcache.
442 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
444 struct dentry * dentry = __d_lookup(parent, name);
446 /* lockess __d_lookup may fail due to concurrent d_move()
447 * in some unrelated directory, so try with d_lookup
450 dentry = d_lookup(parent, name);
452 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
453 dentry = do_revalidate(dentry, nd);
459 * Short-cut version of permission(), for calling by
460 * path_walk(), when dcache lock is held. Combines parts
461 * of permission() and generic_permission(), and tests ONLY for
462 * MAY_EXEC permission.
464 * If appropriate, check DAC only. If not appropriate, or
465 * short-cut DAC fails, then call permission() to do more
466 * complete permission check.
468 static int exec_permission_lite(struct inode *inode,
469 struct nameidata *nd)
471 umode_t mode = inode->i_mode;
473 if (inode->i_op && inode->i_op->permission)
476 if (current->fsuid == inode->i_uid)
478 else if (in_group_p(inode->i_gid))
484 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
487 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
490 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
495 return security_inode_permission(inode, MAY_EXEC, nd);
499 * This is called when everything else fails, and we actually have
500 * to go to the low-level filesystem to find out what we should do..
502 * We get the directory semaphore, and after getting that we also
503 * make sure that nobody added the entry to the dcache in the meantime..
506 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
508 struct dentry * result;
509 struct inode *dir = parent->d_inode;
511 mutex_lock(&dir->i_mutex);
513 * First re-do the cached lookup just in case it was created
514 * while we waited for the directory semaphore..
516 * FIXME! This could use version numbering or similar to
517 * avoid unnecessary cache lookups.
519 * The "dcache_lock" is purely to protect the RCU list walker
520 * from concurrent renames at this point (we mustn't get false
521 * negatives from the RCU list walk here, unlike the optimistic
524 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
526 result = d_lookup(parent, name);
528 struct dentry *dentry;
530 /* Don't create child dentry for a dead directory. */
531 result = ERR_PTR(-ENOENT);
535 dentry = d_alloc(parent, name);
536 result = ERR_PTR(-ENOMEM);
538 result = dir->i_op->lookup(dir, dentry, nd);
545 mutex_unlock(&dir->i_mutex);
550 * Uhhuh! Nasty case: the cache was re-populated while
551 * we waited on the semaphore. Need to revalidate.
553 mutex_unlock(&dir->i_mutex);
554 if (result->d_op && result->d_op->d_revalidate) {
555 result = do_revalidate(result, nd);
557 result = ERR_PTR(-ENOENT);
563 static __always_inline void
564 walk_init_root(const char *name, struct nameidata *nd)
566 struct fs_struct *fs = current->fs;
568 read_lock(&fs->lock);
571 read_unlock(&fs->lock);
575 * Wrapper to retry pathname resolution whenever the underlying
576 * file system returns an ESTALE.
578 * Retry the whole path once, forcing real lookup requests
579 * instead of relying on the dcache.
581 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
583 struct path save = nd->path;
586 /* make sure the stuff we saved doesn't go away */
589 result = __link_path_walk(name, nd);
590 if (result == -ESTALE) {
591 /* nd->path had been dropped */
594 nd->flags |= LOOKUP_REVAL;
595 result = __link_path_walk(name, nd);
603 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
612 walk_init_root(link, nd);
614 res = link_path_walk(link, nd);
615 if (nd->depth || res || nd->last_type!=LAST_NORM)
618 * If it is an iterative symlinks resolution in open_namei() we
619 * have to copy the last component. And all that crap because of
620 * bloody create() on broken symlinks. Furrfu...
623 if (unlikely(!name)) {
627 strcpy(name, nd->last.name);
628 nd->last.name = name;
632 return PTR_ERR(link);
635 static void path_put_conditional(struct path *path, struct nameidata *nd)
638 if (path->mnt != nd->path.mnt)
642 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
644 dput(nd->path.dentry);
645 if (nd->path.mnt != path->mnt)
646 mntput(nd->path.mnt);
647 nd->path.mnt = path->mnt;
648 nd->path.dentry = path->dentry;
651 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
655 struct dentry *dentry = path->dentry;
657 touch_atime(path->mnt, dentry);
658 nd_set_link(nd, NULL);
660 if (path->mnt != nd->path.mnt) {
661 path_to_nameidata(path, nd);
665 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
666 error = PTR_ERR(cookie);
667 if (!IS_ERR(cookie)) {
668 char *s = nd_get_link(nd);
671 error = __vfs_follow_link(nd, s);
672 if (dentry->d_inode->i_op->put_link)
673 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
681 * This limits recursive symlink follows to 8, while
682 * limiting consecutive symlinks to 40.
684 * Without that kind of total limit, nasty chains of consecutive
685 * symlinks can cause almost arbitrarily long lookups.
687 static inline int do_follow_link(struct path *path, struct nameidata *nd)
690 if (current->link_count >= MAX_NESTED_LINKS)
692 if (current->total_link_count >= 40)
694 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
696 err = security_inode_follow_link(path->dentry, nd);
699 current->link_count++;
700 current->total_link_count++;
702 err = __do_follow_link(path, nd);
703 current->link_count--;
707 path_put_conditional(path, nd);
712 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
714 struct vfsmount *parent;
715 struct dentry *mountpoint;
716 spin_lock(&vfsmount_lock);
717 parent=(*mnt)->mnt_parent;
718 if (parent == *mnt) {
719 spin_unlock(&vfsmount_lock);
723 mountpoint=dget((*mnt)->mnt_mountpoint);
724 spin_unlock(&vfsmount_lock);
726 *dentry = mountpoint;
732 /* no need for dcache_lock, as serialization is taken care in
735 static int __follow_mount(struct path *path)
738 while (d_mountpoint(path->dentry)) {
739 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
746 path->dentry = dget(mounted->mnt_root);
752 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
754 while (d_mountpoint(*dentry)) {
755 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
761 *dentry = dget(mounted->mnt_root);
765 /* no need for dcache_lock, as serialization is taken care in
768 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
770 struct vfsmount *mounted;
772 mounted = lookup_mnt(*mnt, *dentry);
777 *dentry = dget(mounted->mnt_root);
783 static __always_inline void follow_dotdot(struct nameidata *nd)
785 struct fs_struct *fs = current->fs;
788 struct vfsmount *parent;
789 struct dentry *old = nd->path.dentry;
791 read_lock(&fs->lock);
792 if (nd->path.dentry == fs->root.dentry &&
793 nd->path.mnt == fs->root.mnt) {
794 read_unlock(&fs->lock);
797 read_unlock(&fs->lock);
798 spin_lock(&dcache_lock);
799 if (nd->path.dentry != nd->path.mnt->mnt_root) {
800 nd->path.dentry = dget(nd->path.dentry->d_parent);
801 spin_unlock(&dcache_lock);
805 spin_unlock(&dcache_lock);
806 spin_lock(&vfsmount_lock);
807 parent = nd->path.mnt->mnt_parent;
808 if (parent == nd->path.mnt) {
809 spin_unlock(&vfsmount_lock);
813 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
814 spin_unlock(&vfsmount_lock);
816 mntput(nd->path.mnt);
817 nd->path.mnt = parent;
819 follow_mount(&nd->path.mnt, &nd->path.dentry);
823 * It's more convoluted than I'd like it to be, but... it's still fairly
824 * small and for now I'd prefer to have fast path as straight as possible.
825 * It _is_ time-critical.
827 static int do_lookup(struct nameidata *nd, struct qstr *name,
830 struct vfsmount *mnt = nd->path.mnt;
831 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
835 if (dentry->d_op && dentry->d_op->d_revalidate)
836 goto need_revalidate;
839 path->dentry = dentry;
840 __follow_mount(path);
844 dentry = real_lookup(nd->path.dentry, name, nd);
850 dentry = do_revalidate(dentry, nd);
858 return PTR_ERR(dentry);
863 * This is the basic name resolution function, turning a pathname into
864 * the final dentry. We expect 'base' to be positive and a directory.
866 * Returns 0 and nd will have valid dentry and mnt on success.
867 * Returns error and drops reference to input namei data on failure.
869 static int __link_path_walk(const char *name, struct nameidata *nd)
874 unsigned int lookup_flags = nd->flags;
881 inode = nd->path.dentry->d_inode;
883 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
885 /* At this point we know we have a real path component. */
891 nd->flags |= LOOKUP_CONTINUE;
892 err = exec_permission_lite(inode, nd);
894 err = vfs_permission(nd, MAY_EXEC);
899 c = *(const unsigned char *)name;
901 hash = init_name_hash();
904 hash = partial_name_hash(c, hash);
905 c = *(const unsigned char *)name;
906 } while (c && (c != '/'));
907 this.len = name - (const char *) this.name;
908 this.hash = end_name_hash(hash);
910 /* remove trailing slashes? */
913 while (*++name == '/');
915 goto last_with_slashes;
918 * "." and ".." are special - ".." especially so because it has
919 * to be able to know about the current root directory and
920 * parent relationships.
922 if (this.name[0] == '.') switch (this.len) {
926 if (this.name[1] != '.')
929 inode = nd->path.dentry->d_inode;
935 * See if the low-level filesystem might want
936 * to use its own hash..
938 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
939 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
944 /* This does the actual lookups.. */
945 err = do_lookup(nd, &this, &next);
950 inode = next.dentry->d_inode;
957 if (inode->i_op->follow_link) {
958 err = do_follow_link(&next, nd);
962 inode = nd->path.dentry->d_inode;
969 path_to_nameidata(&next, nd);
971 if (!inode->i_op->lookup)
974 /* here ends the main loop */
977 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
979 /* Clear LOOKUP_CONTINUE iff it was previously unset */
980 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
981 if (lookup_flags & LOOKUP_PARENT)
983 if (this.name[0] == '.') switch (this.len) {
987 if (this.name[1] != '.')
990 inode = nd->path.dentry->d_inode;
995 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
996 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
1001 err = do_lookup(nd, &this, &next);
1004 inode = next.dentry->d_inode;
1005 if ((lookup_flags & LOOKUP_FOLLOW)
1006 && inode && inode->i_op && inode->i_op->follow_link) {
1007 err = do_follow_link(&next, nd);
1010 inode = nd->path.dentry->d_inode;
1012 path_to_nameidata(&next, nd);
1016 if (lookup_flags & LOOKUP_DIRECTORY) {
1018 if (!inode->i_op || !inode->i_op->lookup)
1024 nd->last_type = LAST_NORM;
1025 if (this.name[0] != '.')
1028 nd->last_type = LAST_DOT;
1029 else if (this.len == 2 && this.name[1] == '.')
1030 nd->last_type = LAST_DOTDOT;
1035 * We bypassed the ordinary revalidation routines.
1036 * We may need to check the cached dentry for staleness.
1038 if (nd->path.dentry && nd->path.dentry->d_sb &&
1039 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1041 /* Note: we do not d_invalidate() */
1042 if (!nd->path.dentry->d_op->d_revalidate(
1043 nd->path.dentry, nd))
1049 path_put_conditional(&next, nd);
1052 path_put(&nd->path);
1057 static int path_walk(const char *name, struct nameidata *nd)
1059 current->total_link_count = 0;
1060 return link_path_walk(name, nd);
1063 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1064 static int do_path_lookup(int dfd, const char *name,
1065 unsigned int flags, struct nameidata *nd)
1070 struct fs_struct *fs = current->fs;
1072 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1077 read_lock(&fs->lock);
1078 nd->path = fs->root;
1079 path_get(&fs->root);
1080 read_unlock(&fs->lock);
1081 } else if (dfd == AT_FDCWD) {
1082 read_lock(&fs->lock);
1085 read_unlock(&fs->lock);
1087 struct dentry *dentry;
1089 file = fget_light(dfd, &fput_needed);
1094 dentry = file->f_path.dentry;
1097 if (!S_ISDIR(dentry->d_inode->i_mode))
1100 retval = file_permission(file, MAY_EXEC);
1104 nd->path = file->f_path;
1105 path_get(&file->f_path);
1107 fput_light(file, fput_needed);
1110 retval = path_walk(name, nd);
1111 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1112 nd->path.dentry->d_inode))
1113 audit_inode(name, nd->path.dentry);
1118 fput_light(file, fput_needed);
1122 int path_lookup(const char *name, unsigned int flags,
1123 struct nameidata *nd)
1125 return do_path_lookup(AT_FDCWD, name, flags, nd);
1129 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1130 * @dentry: pointer to dentry of the base directory
1131 * @mnt: pointer to vfs mount of the base directory
1132 * @name: pointer to file name
1133 * @flags: lookup flags
1134 * @nd: pointer to nameidata
1136 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1137 const char *name, unsigned int flags,
1138 struct nameidata *nd)
1142 /* same as do_path_lookup */
1143 nd->last_type = LAST_ROOT;
1147 nd->path.dentry = dentry;
1149 path_get(&nd->path);
1151 retval = path_walk(name, nd);
1152 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1153 nd->path.dentry->d_inode))
1154 audit_inode(name, nd->path.dentry);
1160 static int __path_lookup_intent_open(int dfd, const char *name,
1161 unsigned int lookup_flags, struct nameidata *nd,
1162 int open_flags, int create_mode)
1164 struct file *filp = get_empty_filp();
1169 nd->intent.open.file = filp;
1170 nd->intent.open.flags = open_flags;
1171 nd->intent.open.create_mode = create_mode;
1172 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1173 if (IS_ERR(nd->intent.open.file)) {
1175 err = PTR_ERR(nd->intent.open.file);
1176 path_put(&nd->path);
1178 } else if (err != 0)
1179 release_open_intent(nd);
1184 * path_lookup_open - lookup a file path with open intent
1185 * @dfd: the directory to use as base, or AT_FDCWD
1186 * @name: pointer to file name
1187 * @lookup_flags: lookup intent flags
1188 * @nd: pointer to nameidata
1189 * @open_flags: open intent flags
1191 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1192 struct nameidata *nd, int open_flags)
1194 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1199 * path_lookup_create - lookup a file path with open + create intent
1200 * @dfd: the directory to use as base, or AT_FDCWD
1201 * @name: pointer to file name
1202 * @lookup_flags: lookup intent flags
1203 * @nd: pointer to nameidata
1204 * @open_flags: open intent flags
1205 * @create_mode: create intent flags
1207 static int path_lookup_create(int dfd, const char *name,
1208 unsigned int lookup_flags, struct nameidata *nd,
1209 int open_flags, int create_mode)
1211 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1212 nd, open_flags, create_mode);
1215 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1216 struct nameidata *nd, int open_flags)
1218 char *tmp = getname(name);
1219 int err = PTR_ERR(tmp);
1222 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1228 static struct dentry *__lookup_hash(struct qstr *name,
1229 struct dentry *base, struct nameidata *nd)
1231 struct dentry *dentry;
1232 struct inode *inode;
1235 inode = base->d_inode;
1238 * See if the low-level filesystem might want
1239 * to use its own hash..
1241 if (base->d_op && base->d_op->d_hash) {
1242 err = base->d_op->d_hash(base, name);
1243 dentry = ERR_PTR(err);
1248 dentry = cached_lookup(base, name, nd);
1252 /* Don't create child dentry for a dead directory. */
1253 dentry = ERR_PTR(-ENOENT);
1254 if (IS_DEADDIR(inode))
1257 new = d_alloc(base, name);
1258 dentry = ERR_PTR(-ENOMEM);
1261 dentry = inode->i_op->lookup(inode, new, nd);
1272 * Restricted form of lookup. Doesn't follow links, single-component only,
1273 * needs parent already locked. Doesn't follow mounts.
1276 static struct dentry *lookup_hash(struct nameidata *nd)
1280 err = permission(nd->path.dentry->d_inode, MAY_EXEC, nd);
1282 return ERR_PTR(err);
1283 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1286 static int __lookup_one_len(const char *name, struct qstr *this,
1287 struct dentry *base, int len)
1297 hash = init_name_hash();
1299 c = *(const unsigned char *)name++;
1300 if (c == '/' || c == '\0')
1302 hash = partial_name_hash(c, hash);
1304 this->hash = end_name_hash(hash);
1309 * lookup_one_len - filesystem helper to lookup single pathname component
1310 * @name: pathname component to lookup
1311 * @base: base directory to lookup from
1312 * @len: maximum length @len should be interpreted to
1314 * Note that this routine is purely a helper for filesystem usage and should
1315 * not be called by generic code. Also note that by using this function the
1316 * nameidata argument is passed to the filesystem methods and a filesystem
1317 * using this helper needs to be prepared for that.
1319 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1324 err = __lookup_one_len(name, &this, base, len);
1326 return ERR_PTR(err);
1328 err = permission(base->d_inode, MAY_EXEC, NULL);
1330 return ERR_PTR(err);
1331 return __lookup_hash(&this, base, NULL);
1335 * lookup_one_noperm - bad hack for sysfs
1336 * @name: pathname component to lookup
1337 * @base: base directory to lookup from
1339 * This is a variant of lookup_one_len that doesn't perform any permission
1340 * checks. It's a horrible hack to work around the braindead sysfs
1341 * architecture and should not be used anywhere else.
1343 * DON'T USE THIS FUNCTION EVER, thanks.
1345 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1350 err = __lookup_one_len(name, &this, base, strlen(name));
1352 return ERR_PTR(err);
1353 return __lookup_hash(&this, base, NULL);
1356 int __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1357 struct nameidata *nd)
1359 char *tmp = getname(name);
1360 int err = PTR_ERR(tmp);
1363 err = do_path_lookup(dfd, tmp, flags, nd);
1369 int __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1371 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1375 * It's inline, so penalty for filesystems that don't use sticky bit is
1378 static inline int check_sticky(struct inode *dir, struct inode *inode)
1380 if (!(dir->i_mode & S_ISVTX))
1382 if (inode->i_uid == current->fsuid)
1384 if (dir->i_uid == current->fsuid)
1386 return !capable(CAP_FOWNER);
1390 * Check whether we can remove a link victim from directory dir, check
1391 * whether the type of victim is right.
1392 * 1. We can't do it if dir is read-only (done in permission())
1393 * 2. We should have write and exec permissions on dir
1394 * 3. We can't remove anything from append-only dir
1395 * 4. We can't do anything with immutable dir (done in permission())
1396 * 5. If the sticky bit on dir is set we should either
1397 * a. be owner of dir, or
1398 * b. be owner of victim, or
1399 * c. have CAP_FOWNER capability
1400 * 6. If the victim is append-only or immutable we can't do antyhing with
1401 * links pointing to it.
1402 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1403 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1404 * 9. We can't remove a root or mountpoint.
1405 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1406 * nfs_async_unlink().
1408 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1412 if (!victim->d_inode)
1415 BUG_ON(victim->d_parent->d_inode != dir);
1416 audit_inode_child(victim->d_name.name, victim, dir);
1418 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1423 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1424 IS_IMMUTABLE(victim->d_inode))
1427 if (!S_ISDIR(victim->d_inode->i_mode))
1429 if (IS_ROOT(victim))
1431 } else if (S_ISDIR(victim->d_inode->i_mode))
1433 if (IS_DEADDIR(dir))
1435 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1440 /* Check whether we can create an object with dentry child in directory
1442 * 1. We can't do it if child already exists (open has special treatment for
1443 * this case, but since we are inlined it's OK)
1444 * 2. We can't do it if dir is read-only (done in permission())
1445 * 3. We should have write and exec permissions on dir
1446 * 4. We can't do it if dir is immutable (done in permission())
1448 static inline int may_create(struct inode *dir, struct dentry *child,
1449 struct nameidata *nd)
1453 if (IS_DEADDIR(dir))
1455 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1459 * O_DIRECTORY translates into forcing a directory lookup.
1461 static inline int lookup_flags(unsigned int f)
1463 unsigned long retval = LOOKUP_FOLLOW;
1466 retval &= ~LOOKUP_FOLLOW;
1468 if (f & O_DIRECTORY)
1469 retval |= LOOKUP_DIRECTORY;
1475 * p1 and p2 should be directories on the same fs.
1477 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1482 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1486 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1488 for (p = p1; p->d_parent != p; p = p->d_parent) {
1489 if (p->d_parent == p2) {
1490 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1491 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1496 for (p = p2; p->d_parent != p; p = p->d_parent) {
1497 if (p->d_parent == p1) {
1498 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1499 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1504 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1505 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1509 void unlock_rename(struct dentry *p1, struct dentry *p2)
1511 mutex_unlock(&p1->d_inode->i_mutex);
1513 mutex_unlock(&p2->d_inode->i_mutex);
1514 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1518 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1519 struct nameidata *nd)
1521 int error = may_create(dir, dentry, nd);
1526 if (!dir->i_op || !dir->i_op->create)
1527 return -EACCES; /* shouldn't it be ENOSYS? */
1530 error = security_inode_create(dir, dentry, mode);
1534 error = dir->i_op->create(dir, dentry, mode, nd);
1536 fsnotify_create(dir, dentry);
1540 int may_open(struct nameidata *nd, int acc_mode, int flag)
1542 struct dentry *dentry = nd->path.dentry;
1543 struct inode *inode = dentry->d_inode;
1549 if (S_ISLNK(inode->i_mode))
1552 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
1556 * FIFO's, sockets and device files are special: they don't
1557 * actually live on the filesystem itself, and as such you
1558 * can write to them even if the filesystem is read-only.
1560 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1562 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1563 if (nd->path.mnt->mnt_flags & MNT_NODEV)
1569 error = vfs_permission(nd, acc_mode);
1573 * An append-only file must be opened in append mode for writing.
1575 if (IS_APPEND(inode)) {
1576 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1582 /* O_NOATIME can only be set by the owner or superuser */
1583 if (flag & O_NOATIME)
1584 if (!is_owner_or_cap(inode))
1588 * Ensure there are no outstanding leases on the file.
1590 error = break_lease(inode, flag);
1594 if (flag & O_TRUNC) {
1595 error = get_write_access(inode);
1600 * Refuse to truncate files with mandatory locks held on them.
1602 error = locks_verify_locked(inode);
1606 error = do_truncate(dentry, 0,
1607 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1610 put_write_access(inode);
1614 if (flag & FMODE_WRITE)
1621 * Be careful about ever adding any more callers of this
1622 * function. Its flags must be in the namei format, not
1623 * what get passed to sys_open().
1625 static int __open_namei_create(struct nameidata *nd, struct path *path,
1629 struct dentry *dir = nd->path.dentry;
1631 if (!IS_POSIXACL(dir->d_inode))
1632 mode &= ~current->fs->umask;
1633 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1634 mutex_unlock(&dir->d_inode->i_mutex);
1635 dput(nd->path.dentry);
1636 nd->path.dentry = path->dentry;
1639 /* Don't check for write permission, don't truncate */
1640 return may_open(nd, 0, flag & ~O_TRUNC);
1644 * Note that while the flag value (low two bits) for sys_open means:
1649 * it is changed into
1650 * 00 - no permissions needed
1651 * 01 - read-permission
1652 * 10 - write-permission
1654 * for the internal routines (ie open_namei()/follow_link() etc)
1655 * This is more logical, and also allows the 00 "no perm needed"
1656 * to be used for symlinks (where the permissions are checked
1660 static inline int open_to_namei_flags(int flag)
1662 if ((flag+1) & O_ACCMODE)
1667 static int open_will_write_to_fs(int flag, struct inode *inode)
1670 * We'll never write to the fs underlying
1673 if (special_file(inode->i_mode))
1675 return (flag & O_TRUNC);
1679 * Note that the low bits of the passed in "open_flag"
1680 * are not the same as in the local variable "flag". See
1681 * open_to_namei_flags() for more details.
1683 struct file *do_filp_open(int dfd, const char *pathname,
1684 int open_flag, int mode)
1687 struct nameidata nd;
1688 int acc_mode, error;
1693 int flag = open_to_namei_flags(open_flag);
1695 acc_mode = ACC_MODE(flag);
1697 /* O_TRUNC implies we need access checks for write permissions */
1699 acc_mode |= MAY_WRITE;
1701 /* Allow the LSM permission hook to distinguish append
1702 access from general write access. */
1703 if (flag & O_APPEND)
1704 acc_mode |= MAY_APPEND;
1707 * The simplest case - just a plain lookup.
1709 if (!(flag & O_CREAT)) {
1710 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1713 return ERR_PTR(error);
1718 * Create - we need to know the parent.
1720 error = path_lookup_create(dfd, pathname, LOOKUP_PARENT,
1723 return ERR_PTR(error);
1726 * We have the parent and last component. First of all, check
1727 * that we are not asked to creat(2) an obvious directory - that
1731 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1734 dir = nd.path.dentry;
1735 nd.flags &= ~LOOKUP_PARENT;
1736 mutex_lock(&dir->d_inode->i_mutex);
1737 path.dentry = lookup_hash(&nd);
1738 path.mnt = nd.path.mnt;
1741 error = PTR_ERR(path.dentry);
1742 if (IS_ERR(path.dentry)) {
1743 mutex_unlock(&dir->d_inode->i_mutex);
1747 if (IS_ERR(nd.intent.open.file)) {
1748 error = PTR_ERR(nd.intent.open.file);
1749 goto exit_mutex_unlock;
1752 /* Negative dentry, just create the file */
1753 if (!path.dentry->d_inode) {
1755 * This write is needed to ensure that a
1756 * ro->rw transition does not occur between
1757 * the time when the file is created and when
1758 * a permanent write count is taken through
1759 * the 'struct file' in nameidata_to_filp().
1761 error = mnt_want_write(nd.path.mnt);
1763 goto exit_mutex_unlock;
1764 error = __open_namei_create(&nd, &path, flag, mode);
1766 mnt_drop_write(nd.path.mnt);
1769 filp = nameidata_to_filp(&nd, open_flag);
1770 mnt_drop_write(nd.path.mnt);
1775 * It already exists.
1777 mutex_unlock(&dir->d_inode->i_mutex);
1778 audit_inode(pathname, path.dentry);
1784 if (__follow_mount(&path)) {
1786 if (flag & O_NOFOLLOW)
1791 if (!path.dentry->d_inode)
1793 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1796 path_to_nameidata(&path, &nd);
1798 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1803 * 1. may_open() truncates a file
1804 * 2. a rw->ro mount transition occurs
1805 * 3. nameidata_to_filp() fails due to
1807 * That would be inconsistent, and should
1808 * be avoided. Taking this mnt write here
1809 * ensures that (2) can not occur.
1811 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1813 error = mnt_want_write(nd.path.mnt);
1817 error = may_open(&nd, acc_mode, flag);
1820 mnt_drop_write(nd.path.mnt);
1823 filp = nameidata_to_filp(&nd, open_flag);
1825 * It is now safe to drop the mnt write
1826 * because the filp has had a write taken
1830 mnt_drop_write(nd.path.mnt);
1834 mutex_unlock(&dir->d_inode->i_mutex);
1836 path_put_conditional(&path, &nd);
1838 if (!IS_ERR(nd.intent.open.file))
1839 release_open_intent(&nd);
1841 return ERR_PTR(error);
1845 if (flag & O_NOFOLLOW)
1848 * This is subtle. Instead of calling do_follow_link() we do the
1849 * thing by hands. The reason is that this way we have zero link_count
1850 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1851 * After that we have the parent and last component, i.e.
1852 * we are in the same situation as after the first path_walk().
1853 * Well, almost - if the last component is normal we get its copy
1854 * stored in nd->last.name and we will have to putname() it when we
1855 * are done. Procfs-like symlinks just set LAST_BIND.
1857 nd.flags |= LOOKUP_PARENT;
1858 error = security_inode_follow_link(path.dentry, &nd);
1861 error = __do_follow_link(&path, &nd);
1863 /* Does someone understand code flow here? Or it is only
1864 * me so stupid? Anathema to whoever designed this non-sense
1865 * with "intent.open".
1867 release_open_intent(&nd);
1868 return ERR_PTR(error);
1870 nd.flags &= ~LOOKUP_PARENT;
1871 if (nd.last_type == LAST_BIND)
1874 if (nd.last_type != LAST_NORM)
1876 if (nd.last.name[nd.last.len]) {
1877 __putname(nd.last.name);
1882 __putname(nd.last.name);
1885 dir = nd.path.dentry;
1886 mutex_lock(&dir->d_inode->i_mutex);
1887 path.dentry = lookup_hash(&nd);
1888 path.mnt = nd.path.mnt;
1889 __putname(nd.last.name);
1894 * filp_open - open file and return file pointer
1896 * @filename: path to open
1897 * @flags: open flags as per the open(2) second argument
1898 * @mode: mode for the new file if O_CREAT is set, else ignored
1900 * This is the helper to open a file from kernelspace if you really
1901 * have to. But in generally you should not do this, so please move
1902 * along, nothing to see here..
1904 struct file *filp_open(const char *filename, int flags, int mode)
1906 return do_filp_open(AT_FDCWD, filename, flags, mode);
1908 EXPORT_SYMBOL(filp_open);
1911 * lookup_create - lookup a dentry, creating it if it doesn't exist
1912 * @nd: nameidata info
1913 * @is_dir: directory flag
1915 * Simple function to lookup and return a dentry and create it
1916 * if it doesn't exist. Is SMP-safe.
1918 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1920 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1922 struct dentry *dentry = ERR_PTR(-EEXIST);
1924 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1926 * Yucky last component or no last component at all?
1927 * (foo/., foo/.., /////)
1929 if (nd->last_type != LAST_NORM)
1931 nd->flags &= ~LOOKUP_PARENT;
1932 nd->flags |= LOOKUP_CREATE;
1933 nd->intent.open.flags = O_EXCL;
1936 * Do the final lookup.
1938 dentry = lookup_hash(nd);
1942 if (dentry->d_inode)
1945 * Special case - lookup gave negative, but... we had foo/bar/
1946 * From the vfs_mknod() POV we just have a negative dentry -
1947 * all is fine. Let's be bastards - you had / on the end, you've
1948 * been asking for (non-existent) directory. -ENOENT for you.
1950 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1952 dentry = ERR_PTR(-ENOENT);
1957 dentry = ERR_PTR(-EEXIST);
1961 EXPORT_SYMBOL_GPL(lookup_create);
1963 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1965 int error = may_create(dir, dentry, NULL);
1970 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1973 if (!dir->i_op || !dir->i_op->mknod)
1976 error = devcgroup_inode_mknod(mode, dev);
1980 error = security_inode_mknod(dir, dentry, mode, dev);
1985 error = dir->i_op->mknod(dir, dentry, mode, dev);
1987 fsnotify_create(dir, dentry);
1991 static int may_mknod(mode_t mode)
1993 switch (mode & S_IFMT) {
1999 case 0: /* zero mode translates to S_IFREG */
2008 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
2013 struct dentry * dentry;
2014 struct nameidata nd;
2018 tmp = getname(filename);
2020 return PTR_ERR(tmp);
2022 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
2025 dentry = lookup_create(&nd, 0);
2026 if (IS_ERR(dentry)) {
2027 error = PTR_ERR(dentry);
2030 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2031 mode &= ~current->fs->umask;
2032 error = may_mknod(mode);
2035 error = mnt_want_write(nd.path.mnt);
2038 switch (mode & S_IFMT) {
2039 case 0: case S_IFREG:
2040 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2042 case S_IFCHR: case S_IFBLK:
2043 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2044 new_decode_dev(dev));
2046 case S_IFIFO: case S_IFSOCK:
2047 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2050 mnt_drop_write(nd.path.mnt);
2054 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2062 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
2064 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2067 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2069 int error = may_create(dir, dentry, NULL);
2074 if (!dir->i_op || !dir->i_op->mkdir)
2077 mode &= (S_IRWXUGO|S_ISVTX);
2078 error = security_inode_mkdir(dir, dentry, mode);
2083 error = dir->i_op->mkdir(dir, dentry, mode);
2085 fsnotify_mkdir(dir, dentry);
2089 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
2093 struct dentry *dentry;
2094 struct nameidata nd;
2096 tmp = getname(pathname);
2097 error = PTR_ERR(tmp);
2101 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
2104 dentry = lookup_create(&nd, 1);
2105 error = PTR_ERR(dentry);
2109 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2110 mode &= ~current->fs->umask;
2111 error = mnt_want_write(nd.path.mnt);
2114 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2115 mnt_drop_write(nd.path.mnt);
2119 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2127 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2129 return sys_mkdirat(AT_FDCWD, pathname, mode);
2133 * We try to drop the dentry early: we should have
2134 * a usage count of 2 if we're the only user of this
2135 * dentry, and if that is true (possibly after pruning
2136 * the dcache), then we drop the dentry now.
2138 * A low-level filesystem can, if it choses, legally
2141 * if (!d_unhashed(dentry))
2144 * if it cannot handle the case of removing a directory
2145 * that is still in use by something else..
2147 void dentry_unhash(struct dentry *dentry)
2150 shrink_dcache_parent(dentry);
2151 spin_lock(&dcache_lock);
2152 spin_lock(&dentry->d_lock);
2153 if (atomic_read(&dentry->d_count) == 2)
2155 spin_unlock(&dentry->d_lock);
2156 spin_unlock(&dcache_lock);
2159 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2161 int error = may_delete(dir, dentry, 1);
2166 if (!dir->i_op || !dir->i_op->rmdir)
2171 mutex_lock(&dentry->d_inode->i_mutex);
2172 dentry_unhash(dentry);
2173 if (d_mountpoint(dentry))
2176 error = security_inode_rmdir(dir, dentry);
2178 error = dir->i_op->rmdir(dir, dentry);
2180 dentry->d_inode->i_flags |= S_DEAD;
2183 mutex_unlock(&dentry->d_inode->i_mutex);
2192 static long do_rmdir(int dfd, const char __user *pathname)
2196 struct dentry *dentry;
2197 struct nameidata nd;
2199 name = getname(pathname);
2201 return PTR_ERR(name);
2203 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2207 switch(nd.last_type) {
2218 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2219 dentry = lookup_hash(&nd);
2220 error = PTR_ERR(dentry);
2223 error = mnt_want_write(nd.path.mnt);
2226 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2227 mnt_drop_write(nd.path.mnt);
2231 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2239 asmlinkage long sys_rmdir(const char __user *pathname)
2241 return do_rmdir(AT_FDCWD, pathname);
2244 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2246 int error = may_delete(dir, dentry, 0);
2251 if (!dir->i_op || !dir->i_op->unlink)
2256 mutex_lock(&dentry->d_inode->i_mutex);
2257 if (d_mountpoint(dentry))
2260 error = security_inode_unlink(dir, dentry);
2262 error = dir->i_op->unlink(dir, dentry);
2264 mutex_unlock(&dentry->d_inode->i_mutex);
2266 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2267 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2268 fsnotify_link_count(dentry->d_inode);
2276 * Make sure that the actual truncation of the file will occur outside its
2277 * directory's i_mutex. Truncate can take a long time if there is a lot of
2278 * writeout happening, and we don't want to prevent access to the directory
2279 * while waiting on the I/O.
2281 static long do_unlinkat(int dfd, const char __user *pathname)
2285 struct dentry *dentry;
2286 struct nameidata nd;
2287 struct inode *inode = NULL;
2289 name = getname(pathname);
2291 return PTR_ERR(name);
2293 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2297 if (nd.last_type != LAST_NORM)
2299 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2300 dentry = lookup_hash(&nd);
2301 error = PTR_ERR(dentry);
2302 if (!IS_ERR(dentry)) {
2303 /* Why not before? Because we want correct error value */
2304 if (nd.last.name[nd.last.len])
2306 inode = dentry->d_inode;
2308 atomic_inc(&inode->i_count);
2309 error = mnt_want_write(nd.path.mnt);
2312 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2313 mnt_drop_write(nd.path.mnt);
2317 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2319 iput(inode); /* truncate the inode here */
2327 error = !dentry->d_inode ? -ENOENT :
2328 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2332 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2334 if ((flag & ~AT_REMOVEDIR) != 0)
2337 if (flag & AT_REMOVEDIR)
2338 return do_rmdir(dfd, pathname);
2340 return do_unlinkat(dfd, pathname);
2343 asmlinkage long sys_unlink(const char __user *pathname)
2345 return do_unlinkat(AT_FDCWD, pathname);
2348 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2350 int error = may_create(dir, dentry, NULL);
2355 if (!dir->i_op || !dir->i_op->symlink)
2358 error = security_inode_symlink(dir, dentry, oldname);
2363 error = dir->i_op->symlink(dir, dentry, oldname);
2365 fsnotify_create(dir, dentry);
2369 asmlinkage long sys_symlinkat(const char __user *oldname,
2370 int newdfd, const char __user *newname)
2375 struct dentry *dentry;
2376 struct nameidata nd;
2378 from = getname(oldname);
2380 return PTR_ERR(from);
2381 to = getname(newname);
2382 error = PTR_ERR(to);
2386 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2389 dentry = lookup_create(&nd, 0);
2390 error = PTR_ERR(dentry);
2394 error = mnt_want_write(nd.path.mnt);
2397 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2398 mnt_drop_write(nd.path.mnt);
2402 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2411 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2413 return sys_symlinkat(oldname, AT_FDCWD, newname);
2416 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2418 struct inode *inode = old_dentry->d_inode;
2424 error = may_create(dir, new_dentry, NULL);
2428 if (dir->i_sb != inode->i_sb)
2432 * A link to an append-only or immutable file cannot be created.
2434 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2436 if (!dir->i_op || !dir->i_op->link)
2438 if (S_ISDIR(inode->i_mode))
2441 error = security_inode_link(old_dentry, dir, new_dentry);
2445 mutex_lock(&inode->i_mutex);
2447 error = dir->i_op->link(old_dentry, dir, new_dentry);
2448 mutex_unlock(&inode->i_mutex);
2450 fsnotify_link(dir, inode, new_dentry);
2455 * Hardlinks are often used in delicate situations. We avoid
2456 * security-related surprises by not following symlinks on the
2459 * We don't follow them on the oldname either to be compatible
2460 * with linux 2.0, and to avoid hard-linking to directories
2461 * and other special files. --ADM
2463 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2464 int newdfd, const char __user *newname,
2467 struct dentry *new_dentry;
2468 struct nameidata nd, old_nd;
2472 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2475 to = getname(newname);
2479 error = __user_walk_fd(olddfd, oldname,
2480 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2484 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2488 if (old_nd.path.mnt != nd.path.mnt)
2490 new_dentry = lookup_create(&nd, 0);
2491 error = PTR_ERR(new_dentry);
2492 if (IS_ERR(new_dentry))
2494 error = mnt_want_write(nd.path.mnt);
2497 error = vfs_link(old_nd.path.dentry, nd.path.dentry->d_inode, new_dentry);
2498 mnt_drop_write(nd.path.mnt);
2502 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2506 path_put(&old_nd.path);
2513 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2515 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2519 * The worst of all namespace operations - renaming directory. "Perverted"
2520 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2522 * a) we can get into loop creation. Check is done in is_subdir().
2523 * b) race potential - two innocent renames can create a loop together.
2524 * That's where 4.4 screws up. Current fix: serialization on
2525 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2527 * c) we have to lock _three_ objects - parents and victim (if it exists).
2528 * And that - after we got ->i_mutex on parents (until then we don't know
2529 * whether the target exists). Solution: try to be smart with locking
2530 * order for inodes. We rely on the fact that tree topology may change
2531 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2532 * move will be locked. Thus we can rank directories by the tree
2533 * (ancestors first) and rank all non-directories after them.
2534 * That works since everybody except rename does "lock parent, lookup,
2535 * lock child" and rename is under ->s_vfs_rename_mutex.
2536 * HOWEVER, it relies on the assumption that any object with ->lookup()
2537 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2538 * we'd better make sure that there's no link(2) for them.
2539 * d) some filesystems don't support opened-but-unlinked directories,
2540 * either because of layout or because they are not ready to deal with
2541 * all cases correctly. The latter will be fixed (taking this sort of
2542 * stuff into VFS), but the former is not going away. Solution: the same
2543 * trick as in rmdir().
2544 * e) conversion from fhandle to dentry may come in the wrong moment - when
2545 * we are removing the target. Solution: we will have to grab ->i_mutex
2546 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2547 * ->i_mutex on parents, which works but leads to some truely excessive
2550 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2551 struct inode *new_dir, struct dentry *new_dentry)
2554 struct inode *target;
2557 * If we are going to change the parent - check write permissions,
2558 * we'll need to flip '..'.
2560 if (new_dir != old_dir) {
2561 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2566 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2570 target = new_dentry->d_inode;
2572 mutex_lock(&target->i_mutex);
2573 dentry_unhash(new_dentry);
2575 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2578 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2581 target->i_flags |= S_DEAD;
2582 mutex_unlock(&target->i_mutex);
2583 if (d_unhashed(new_dentry))
2584 d_rehash(new_dentry);
2588 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2589 d_move(old_dentry,new_dentry);
2593 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2594 struct inode *new_dir, struct dentry *new_dentry)
2596 struct inode *target;
2599 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2604 target = new_dentry->d_inode;
2606 mutex_lock(&target->i_mutex);
2607 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2610 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2612 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2613 d_move(old_dentry, new_dentry);
2616 mutex_unlock(&target->i_mutex);
2621 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2622 struct inode *new_dir, struct dentry *new_dentry)
2625 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2626 const char *old_name;
2628 if (old_dentry->d_inode == new_dentry->d_inode)
2631 error = may_delete(old_dir, old_dentry, is_dir);
2635 if (!new_dentry->d_inode)
2636 error = may_create(new_dir, new_dentry, NULL);
2638 error = may_delete(new_dir, new_dentry, is_dir);
2642 if (!old_dir->i_op || !old_dir->i_op->rename)
2645 DQUOT_INIT(old_dir);
2646 DQUOT_INIT(new_dir);
2648 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2651 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2653 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2655 const char *new_name = old_dentry->d_name.name;
2656 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2657 new_dentry->d_inode, old_dentry);
2659 fsnotify_oldname_free(old_name);
2664 static int do_rename(int olddfd, const char *oldname,
2665 int newdfd, const char *newname)
2668 struct dentry * old_dir, * new_dir;
2669 struct dentry * old_dentry, *new_dentry;
2670 struct dentry * trap;
2671 struct nameidata oldnd, newnd;
2673 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2677 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2682 if (oldnd.path.mnt != newnd.path.mnt)
2685 old_dir = oldnd.path.dentry;
2687 if (oldnd.last_type != LAST_NORM)
2690 new_dir = newnd.path.dentry;
2691 if (newnd.last_type != LAST_NORM)
2694 trap = lock_rename(new_dir, old_dir);
2696 old_dentry = lookup_hash(&oldnd);
2697 error = PTR_ERR(old_dentry);
2698 if (IS_ERR(old_dentry))
2700 /* source must exist */
2702 if (!old_dentry->d_inode)
2704 /* unless the source is a directory trailing slashes give -ENOTDIR */
2705 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2707 if (oldnd.last.name[oldnd.last.len])
2709 if (newnd.last.name[newnd.last.len])
2712 /* source should not be ancestor of target */
2714 if (old_dentry == trap)
2716 new_dentry = lookup_hash(&newnd);
2717 error = PTR_ERR(new_dentry);
2718 if (IS_ERR(new_dentry))
2720 /* target should not be an ancestor of source */
2722 if (new_dentry == trap)
2725 error = mnt_want_write(oldnd.path.mnt);
2728 error = vfs_rename(old_dir->d_inode, old_dentry,
2729 new_dir->d_inode, new_dentry);
2730 mnt_drop_write(oldnd.path.mnt);
2736 unlock_rename(new_dir, old_dir);
2738 path_put(&newnd.path);
2740 path_put(&oldnd.path);
2745 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2746 int newdfd, const char __user *newname)
2752 from = getname(oldname);
2754 return PTR_ERR(from);
2755 to = getname(newname);
2756 error = PTR_ERR(to);
2758 error = do_rename(olddfd, from, newdfd, to);
2765 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2767 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2770 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2774 len = PTR_ERR(link);
2779 if (len > (unsigned) buflen)
2781 if (copy_to_user(buffer, link, len))
2788 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2789 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2790 * using) it for any given inode is up to filesystem.
2792 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2794 struct nameidata nd;
2799 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2801 return PTR_ERR(cookie);
2803 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2804 if (dentry->d_inode->i_op->put_link)
2805 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2809 int vfs_follow_link(struct nameidata *nd, const char *link)
2811 return __vfs_follow_link(nd, link);
2814 /* get the link contents into pagecache */
2815 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2818 struct address_space *mapping = dentry->d_inode->i_mapping;
2819 page = read_mapping_page(mapping, 0, NULL);
2826 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2828 struct page *page = NULL;
2829 char *s = page_getlink(dentry, &page);
2830 int res = vfs_readlink(dentry,buffer,buflen,s);
2833 page_cache_release(page);
2838 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2840 struct page *page = NULL;
2841 nd_set_link(nd, page_getlink(dentry, &page));
2845 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2847 struct page *page = cookie;
2851 page_cache_release(page);
2855 int __page_symlink(struct inode *inode, const char *symname, int len,
2858 struct address_space *mapping = inode->i_mapping;
2865 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2866 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
2870 kaddr = kmap_atomic(page, KM_USER0);
2871 memcpy(kaddr, symname, len-1);
2872 kunmap_atomic(kaddr, KM_USER0);
2874 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2881 mark_inode_dirty(inode);
2887 int page_symlink(struct inode *inode, const char *symname, int len)
2889 return __page_symlink(inode, symname, len,
2890 mapping_gfp_mask(inode->i_mapping));
2893 const struct inode_operations page_symlink_inode_operations = {
2894 .readlink = generic_readlink,
2895 .follow_link = page_follow_link_light,
2896 .put_link = page_put_link,
2899 EXPORT_SYMBOL(__user_walk);
2900 EXPORT_SYMBOL(__user_walk_fd);
2901 EXPORT_SYMBOL(follow_down);
2902 EXPORT_SYMBOL(follow_up);
2903 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2904 EXPORT_SYMBOL(getname);
2905 EXPORT_SYMBOL(lock_rename);
2906 EXPORT_SYMBOL(lookup_one_len);
2907 EXPORT_SYMBOL(page_follow_link_light);
2908 EXPORT_SYMBOL(page_put_link);
2909 EXPORT_SYMBOL(page_readlink);
2910 EXPORT_SYMBOL(__page_symlink);
2911 EXPORT_SYMBOL(page_symlink);
2912 EXPORT_SYMBOL(page_symlink_inode_operations);
2913 EXPORT_SYMBOL(path_lookup);
2914 EXPORT_SYMBOL(vfs_path_lookup);
2915 EXPORT_SYMBOL(permission);
2916 EXPORT_SYMBOL(vfs_permission);
2917 EXPORT_SYMBOL(file_permission);
2918 EXPORT_SYMBOL(unlock_rename);
2919 EXPORT_SYMBOL(vfs_create);
2920 EXPORT_SYMBOL(vfs_follow_link);
2921 EXPORT_SYMBOL(vfs_link);
2922 EXPORT_SYMBOL(vfs_mkdir);
2923 EXPORT_SYMBOL(vfs_mknod);
2924 EXPORT_SYMBOL(generic_permission);
2925 EXPORT_SYMBOL(vfs_readlink);
2926 EXPORT_SYMBOL(vfs_rename);
2927 EXPORT_SYMBOL(vfs_rmdir);
2928 EXPORT_SYMBOL(vfs_symlink);
2929 EXPORT_SYMBOL(vfs_unlink);
2930 EXPORT_SYMBOL(dentry_unhash);
2931 EXPORT_SYMBOL(generic_readlink);