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/smp_lock.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/namei.h>
35 #include <linux/proc_fs.h>
36 #include <linux/vserver/inode.h>
37 #include <linux/vserver/debug.h>
38 #include <asm/namei.h>
39 #include <asm/uaccess.h>
41 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existant name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
121 static int do_getname(const char __user *filename, char *page)
124 unsigned long len = PATH_MAX;
126 if (!segment_eq(get_fs(), KERNEL_DS)) {
127 if ((unsigned long) filename >= TASK_SIZE)
129 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
130 len = TASK_SIZE - (unsigned long) filename;
133 retval = strncpy_from_user(page, filename, len);
137 return -ENAMETOOLONG;
143 char * getname(const char __user * filename)
147 result = ERR_PTR(-ENOMEM);
150 int retval = do_getname(filename, tmp);
155 result = ERR_PTR(retval);
158 audit_getname(result);
162 #ifdef CONFIG_AUDITSYSCALL
163 void putname(const char *name)
165 if (unlikely(current->audit_context))
170 EXPORT_SYMBOL(putname);
175 * generic_permission - check for access rights on a Posix-like filesystem
176 * @inode: inode to check access rights for
177 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
178 * @check_acl: optional callback to check for Posix ACLs
180 * Used to check for read/write/execute permissions on a file.
181 * We use "fsuid" for this, letting us set arbitrary permissions
182 * for filesystem access without changing the "normal" uids which
183 * are used for other things..
185 int generic_permission(struct inode *inode, int mask,
186 int (*check_acl)(struct inode *inode, int mask))
188 umode_t mode = inode->i_mode;
190 if (current->fsuid == inode->i_uid)
193 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
194 int error = check_acl(inode, mask);
195 if (error == -EACCES)
196 goto check_capabilities;
197 else if (error != -EAGAIN)
201 if (in_group_p(inode->i_gid))
206 * If the DACs are ok we don't need any capability check.
208 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
213 * Read/write DACs are always overridable.
214 * Executable DACs are overridable if at least one exec bit is set.
216 if (!(mask & MAY_EXEC) ||
217 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
218 if (capable(CAP_DAC_OVERRIDE))
222 * Searching includes executable on directories, else just read.
224 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
225 if (capable(CAP_DAC_READ_SEARCH))
231 static inline int vx_barrier(struct inode *inode)
233 if (IS_BARRIER(inode) && !vx_check(0, VX_ADMIN)) {
234 vxwprintk(1, "xid=%d did hit the barrier.",
241 static inline int xid_permission(struct inode *inode, int mask, struct nameidata *nd)
243 if (vx_barrier(inode))
245 if (inode->i_xid == 0)
247 #ifdef CONFIG_VSERVER_FILESHARING
248 /* MEF: PlanetLab FS module assumes that any file that can be
249 * named (e.g., via a cross mount) is not hidden from another
250 * context or the admin context.
252 if (vx_check(inode->i_xid,VX_STATIC|VX_DYNAMIC))
255 if (vx_check(inode->i_xid, VX_ADMIN|VX_WATCH|VX_IDENT))
258 vxwprintk(1, "xid=%d denied access to %p[#%d,%lu] »%s«.",
259 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
264 int permission(struct inode *inode, int mask, struct nameidata *nd)
268 if (mask & MAY_WRITE) {
269 umode_t mode = inode->i_mode;
272 * Nobody gets write access to a read-only fs.
274 if ((IS_RDONLY(inode) || (nd && MNT_IS_RDONLY(nd->mnt))) &&
275 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
279 * Nobody gets write access to an immutable file.
281 if (IS_IMMUTABLE(inode))
286 /* Ordinary permission routines do not understand MAY_APPEND. */
287 submask = mask & ~MAY_APPEND;
288 if ((retval = xid_permission(inode, mask, nd)))
290 if (inode->i_op && inode->i_op->permission)
291 retval = inode->i_op->permission(inode, submask, nd);
293 retval = generic_permission(inode, submask, NULL);
297 return security_inode_permission(inode, mask, nd);
301 * vfs_permission - check for access rights to a given path
302 * @nd: lookup result that describes the path
303 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
305 * Used to check for read/write/execute permissions on a path.
306 * We use "fsuid" for this, letting us set arbitrary permissions
307 * for filesystem access without changing the "normal" uids which
308 * are used for other things.
310 int vfs_permission(struct nameidata *nd, int mask)
312 return permission(nd->dentry->d_inode, mask, nd);
316 * file_permission - check for additional access rights to a given file
317 * @file: file to check access rights for
318 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
320 * Used to check for read/write/execute permissions on an already opened
324 * Do not use this function in new code. All access checks should
325 * be done using vfs_permission().
327 int file_permission(struct file *file, int mask)
329 return permission(file->f_dentry->d_inode, mask, NULL);
333 * get_write_access() gets write permission for a file.
334 * put_write_access() releases this write permission.
335 * This is used for regular files.
336 * We cannot support write (and maybe mmap read-write shared) accesses and
337 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
338 * can have the following values:
339 * 0: no writers, no VM_DENYWRITE mappings
340 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
341 * > 0: (i_writecount) users are writing to the file.
343 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
344 * except for the cases where we don't hold i_writecount yet. Then we need to
345 * use {get,deny}_write_access() - these functions check the sign and refuse
346 * to do the change if sign is wrong. Exclusion between them is provided by
347 * the inode->i_lock spinlock.
350 int get_write_access(struct inode * inode)
352 spin_lock(&inode->i_lock);
353 if (atomic_read(&inode->i_writecount) < 0) {
354 spin_unlock(&inode->i_lock);
357 atomic_inc(&inode->i_writecount);
358 spin_unlock(&inode->i_lock);
363 int deny_write_access(struct file * file)
365 struct inode *inode = file->f_dentry->d_inode;
367 spin_lock(&inode->i_lock);
368 if (atomic_read(&inode->i_writecount) > 0) {
369 spin_unlock(&inode->i_lock);
372 atomic_dec(&inode->i_writecount);
373 spin_unlock(&inode->i_lock);
378 void path_release(struct nameidata *nd)
385 * umount() mustn't call path_release()/mntput() as that would clear
388 void path_release_on_umount(struct nameidata *nd)
391 mntput_no_expire(nd->mnt);
395 * release_open_intent - free up open intent resources
396 * @nd: pointer to nameidata
398 void release_open_intent(struct nameidata *nd)
400 if (nd->intent.open.file->f_dentry == NULL)
401 put_filp(nd->intent.open.file);
403 fput(nd->intent.open.file);
407 * Internal lookup() using the new generic dcache.
410 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
412 struct dentry * dentry = __d_lookup(parent, name);
414 /* lockess __d_lookup may fail due to concurrent d_move()
415 * in some unrelated directory, so try with d_lookup
418 dentry = d_lookup(parent, name);
420 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
421 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
430 * Short-cut version of permission(), for calling by
431 * path_walk(), when dcache lock is held. Combines parts
432 * of permission() and generic_permission(), and tests ONLY for
433 * MAY_EXEC permission.
435 * If appropriate, check DAC only. If not appropriate, or
436 * short-cut DAC fails, then call permission() to do more
437 * complete permission check.
439 static int exec_permission_lite(struct inode *inode,
440 struct nameidata *nd)
442 umode_t mode = inode->i_mode;
444 if (vx_barrier(inode))
446 if (inode->i_op && inode->i_op->permission)
449 if (current->fsuid == inode->i_uid)
451 else if (in_group_p(inode->i_gid))
457 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
460 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
463 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
468 return security_inode_permission(inode, MAY_EXEC, nd);
472 * This is called when everything else fails, and we actually have
473 * to go to the low-level filesystem to find out what we should do..
475 * We get the directory semaphore, and after getting that we also
476 * make sure that nobody added the entry to the dcache in the meantime..
479 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
481 struct dentry * result;
482 struct inode *dir = parent->d_inode;
484 mutex_lock(&dir->i_mutex);
486 * First re-do the cached lookup just in case it was created
487 * while we waited for the directory semaphore..
489 * FIXME! This could use version numbering or similar to
490 * avoid unnecessary cache lookups.
492 * The "dcache_lock" is purely to protect the RCU list walker
493 * from concurrent renames at this point (we mustn't get false
494 * negatives from the RCU list walk here, unlike the optimistic
497 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
499 result = d_lookup(parent, name);
501 struct dentry * dentry = d_alloc(parent, name);
502 result = ERR_PTR(-ENOMEM);
504 result = dir->i_op->lookup(dir, dentry, nd);
510 mutex_unlock(&dir->i_mutex);
515 * Uhhuh! Nasty case: the cache was re-populated while
516 * we waited on the semaphore. Need to revalidate.
518 mutex_unlock(&dir->i_mutex);
519 if (result->d_op && result->d_op->d_revalidate) {
520 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
522 result = ERR_PTR(-ENOENT);
528 static int __emul_lookup_dentry(const char *, struct nameidata *);
531 static __always_inline int
532 walk_init_root(const char *name, struct nameidata *nd)
534 read_lock(¤t->fs->lock);
535 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
536 nd->mnt = mntget(current->fs->altrootmnt);
537 nd->dentry = dget(current->fs->altroot);
538 read_unlock(¤t->fs->lock);
539 if (__emul_lookup_dentry(name,nd))
541 read_lock(¤t->fs->lock);
543 nd->mnt = mntget(current->fs->rootmnt);
544 nd->dentry = dget(current->fs->root);
545 read_unlock(¤t->fs->lock);
549 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
558 if (!walk_init_root(link, nd))
559 /* weird __emul_prefix() stuff did it */
562 res = link_path_walk(link, nd);
564 if (nd->depth || res || nd->last_type!=LAST_NORM)
567 * If it is an iterative symlinks resolution in open_namei() we
568 * have to copy the last component. And all that crap because of
569 * bloody create() on broken symlinks. Furrfu...
572 if (unlikely(!name)) {
576 strcpy(name, nd->last.name);
577 nd->last.name = name;
581 return PTR_ERR(link);
585 struct vfsmount *mnt;
586 struct dentry *dentry;
589 static inline void dput_path(struct path *path, struct nameidata *nd)
592 if (path->mnt != nd->mnt)
596 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
599 if (nd->mnt != path->mnt)
602 nd->dentry = path->dentry;
605 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
609 struct dentry *dentry = path->dentry;
611 touch_atime(path->mnt, dentry);
612 nd_set_link(nd, NULL);
614 if (path->mnt != nd->mnt) {
615 path_to_nameidata(path, nd);
619 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
620 error = PTR_ERR(cookie);
621 if (!IS_ERR(cookie)) {
622 char *s = nd_get_link(nd);
625 error = __vfs_follow_link(nd, s);
626 if (dentry->d_inode->i_op->put_link)
627 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
636 * This limits recursive symlink follows to 8, while
637 * limiting consecutive symlinks to 40.
639 * Without that kind of total limit, nasty chains of consecutive
640 * symlinks can cause almost arbitrarily long lookups.
642 static inline int do_follow_link(struct path *path, struct nameidata *nd)
645 if (current->link_count >= MAX_NESTED_LINKS)
647 if (current->total_link_count >= 40)
649 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
651 err = security_inode_follow_link(path->dentry, nd);
654 current->link_count++;
655 current->total_link_count++;
657 err = __do_follow_link(path, nd);
658 current->link_count--;
667 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
669 struct vfsmount *parent;
670 struct dentry *mountpoint;
671 spin_lock(&vfsmount_lock);
672 parent=(*mnt)->mnt_parent;
673 if (parent == *mnt) {
674 spin_unlock(&vfsmount_lock);
678 mountpoint=dget((*mnt)->mnt_mountpoint);
679 spin_unlock(&vfsmount_lock);
681 *dentry = mountpoint;
687 /* no need for dcache_lock, as serialization is taken care in
690 static int __follow_mount(struct path *path)
693 while (d_mountpoint(path->dentry)) {
694 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
701 path->dentry = dget(mounted->mnt_root);
707 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
709 while (d_mountpoint(*dentry)) {
710 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
716 *dentry = dget(mounted->mnt_root);
720 /* no need for dcache_lock, as serialization is taken care in
723 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
725 struct vfsmount *mounted;
727 mounted = lookup_mnt(*mnt, *dentry);
732 *dentry = dget(mounted->mnt_root);
738 static __always_inline void follow_dotdot(struct nameidata *nd)
741 struct vfsmount *parent;
742 struct dentry *old = nd->dentry;
744 read_lock(¤t->fs->lock);
745 if (nd->dentry == current->fs->root &&
746 nd->mnt == current->fs->rootmnt) {
747 read_unlock(¤t->fs->lock);
748 /* for sane '/' avoid follow_mount() */
751 read_unlock(¤t->fs->lock);
752 spin_lock(&dcache_lock);
753 if (nd->dentry != nd->mnt->mnt_root) {
754 nd->dentry = dget(nd->dentry->d_parent);
755 spin_unlock(&dcache_lock);
759 spin_unlock(&dcache_lock);
760 spin_lock(&vfsmount_lock);
761 parent = nd->mnt->mnt_parent;
762 if (parent == nd->mnt) {
763 spin_unlock(&vfsmount_lock);
767 nd->dentry = dget(nd->mnt->mnt_mountpoint);
768 spin_unlock(&vfsmount_lock);
773 follow_mount(&nd->mnt, &nd->dentry);
777 * It's more convoluted than I'd like it to be, but... it's still fairly
778 * small and for now I'd prefer to have fast path as straight as possible.
779 * It _is_ time-critical.
781 static int do_lookup(struct nameidata *nd, struct qstr *name,
782 struct path *path, int atomic)
784 struct vfsmount *mnt = nd->mnt;
785 struct dentry *dentry = __d_lookup(nd->dentry, name);
790 if (dentry->d_op && dentry->d_op->d_revalidate)
791 goto need_revalidate;
792 inode = dentry->d_inode;
795 #ifdef CONFIG_VSERVER_FILESHARING
796 /* MEF: PlanetLab FS module assumes that any file that can be
797 * named (e.g., via a cross mount) is not hidden from another
798 * context or the admin context.
800 if (vx_check(inode->i_xid,VX_STATIC|VX_DYNAMIC|VX_ADMIN)) {
803 else /* do the following check */
805 if (!vx_check(inode->i_xid, VX_WATCH|VX_ADMIN|VX_HOSTID|VX_IDENT))
807 if (inode->i_sb->s_magic == PROC_SUPER_MAGIC) {
808 struct proc_dir_entry *de = PDE(inode);
810 if (de && !vx_hide_check(0, de->vx_flags))
815 path->dentry = dentry;
816 __follow_mount(path);
819 vxwprintk(1, "xid=%d did lookup hidden %p[#%d,%lu] »%s«.",
820 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
821 vxd_path(dentry, mnt));
827 return -EWOULDBLOCKIO;
828 dentry = real_lookup(nd->dentry, name, nd);
835 return -EWOULDBLOCKIO;
836 if (dentry->d_op->d_revalidate(dentry, nd))
838 if (d_invalidate(dentry))
844 return PTR_ERR(dentry);
849 * This is the basic name resolution function, turning a pathname into
850 * the final dentry. We expect 'base' to be positive and a directory.
852 * Returns 0 and nd will have valid dentry and mnt on success.
853 * Returns error and drops reference to input namei data on failure.
855 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
860 unsigned int lookup_flags = nd->flags;
862 atomic = (lookup_flags & LOOKUP_ATOMIC);
869 inode = nd->dentry->d_inode;
871 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
873 /* At this point we know we have a real path component. */
879 nd->flags |= LOOKUP_CONTINUE;
880 err = exec_permission_lite(inode, nd);
882 err = vfs_permission(nd, MAY_EXEC);
887 c = *(const unsigned char *)name;
889 hash = init_name_hash();
892 hash = partial_name_hash(c, hash);
893 c = *(const unsigned char *)name;
894 } while (c && (c != '/'));
895 this.len = name - (const char *) this.name;
896 this.hash = end_name_hash(hash);
898 /* remove trailing slashes? */
901 while (*++name == '/');
903 goto last_with_slashes;
906 * "." and ".." are special - ".." especially so because it has
907 * to be able to know about the current root directory and
908 * parent relationships.
910 if (this.name[0] == '.') switch (this.len) {
914 if (this.name[1] != '.')
917 inode = nd->dentry->d_inode;
923 * See if the low-level filesystem might want
924 * to use its own hash..
926 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
927 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
931 /* This does the actual lookups.. */
932 err = do_lookup(nd, &this, &next, atomic);
937 inode = next.dentry->d_inode;
944 if (inode->i_op->follow_link) {
945 err = do_follow_link(&next, nd);
949 inode = nd->dentry->d_inode;
956 path_to_nameidata(&next, nd);
958 if (!inode->i_op->lookup)
961 /* here ends the main loop */
964 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
966 /* Clear LOOKUP_CONTINUE iff it was previously unset */
967 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
968 if (lookup_flags & LOOKUP_PARENT)
970 if (this.name[0] == '.') switch (this.len) {
974 if (this.name[1] != '.')
977 inode = nd->dentry->d_inode;
982 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
983 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
987 err = do_lookup(nd, &this, &next, atomic);
990 inode = next.dentry->d_inode;
991 if ((lookup_flags & LOOKUP_FOLLOW)
992 && inode && inode->i_op && inode->i_op->follow_link) {
993 err = do_follow_link(&next, nd);
996 inode = nd->dentry->d_inode;
998 path_to_nameidata(&next, nd);
1002 if (lookup_flags & LOOKUP_DIRECTORY) {
1004 if (!inode->i_op || !inode->i_op->lookup)
1010 nd->last_type = LAST_NORM;
1011 if (this.name[0] != '.')
1014 nd->last_type = LAST_DOT;
1015 else if (this.len == 2 && this.name[1] == '.')
1016 nd->last_type = LAST_DOTDOT;
1021 * We bypassed the ordinary revalidation routines.
1022 * We may need to check the cached dentry for staleness.
1024 if (nd->dentry && nd->dentry->d_sb &&
1025 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1027 /* Note: we do not d_invalidate() */
1028 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
1034 dput_path(&next, nd);
1043 * Wrapper to retry pathname resolution whenever the underlying
1044 * file system returns an ESTALE.
1046 * Retry the whole path once, forcing real lookup requests
1047 * instead of relying on the dcache.
1049 int fastcall link_path_walk(const char *name, struct nameidata *nd)
1051 struct nameidata save = *nd;
1054 /* make sure the stuff we saved doesn't go away */
1058 result = __link_path_walk(name, nd);
1059 if (result == -ESTALE) {
1063 nd->flags |= LOOKUP_REVAL;
1064 result = __link_path_walk(name, nd);
1073 int fastcall path_walk(const char * name, struct nameidata *nd)
1075 current->total_link_count = 0;
1076 return link_path_walk(name, nd);
1080 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1081 * everything is done. Returns 0 and drops input nd, if lookup failed;
1083 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1085 if (path_walk(name, nd))
1086 return 0; /* something went wrong... */
1088 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1089 struct dentry *old_dentry = nd->dentry;
1090 struct vfsmount *old_mnt = nd->mnt;
1091 struct qstr last = nd->last;
1092 int last_type = nd->last_type;
1094 * NAME was not found in alternate root or it's a directory. Try to find
1095 * it in the normal root:
1097 nd->last_type = LAST_ROOT;
1098 read_lock(¤t->fs->lock);
1099 nd->mnt = mntget(current->fs->rootmnt);
1100 nd->dentry = dget(current->fs->root);
1101 read_unlock(¤t->fs->lock);
1102 if (path_walk(name, nd) == 0) {
1103 if (nd->dentry->d_inode) {
1110 nd->dentry = old_dentry;
1113 nd->last_type = last_type;
1118 void set_fs_altroot(void)
1120 char *emul = __emul_prefix();
1121 struct nameidata nd;
1122 struct vfsmount *mnt = NULL, *oldmnt;
1123 struct dentry *dentry = NULL, *olddentry;
1128 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1134 write_lock(¤t->fs->lock);
1135 oldmnt = current->fs->altrootmnt;
1136 olddentry = current->fs->altroot;
1137 current->fs->altrootmnt = mnt;
1138 current->fs->altroot = dentry;
1139 write_unlock(¤t->fs->lock);
1146 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1147 static int fastcall do_path_lookup(int dfd, const char *name,
1148 unsigned int flags, struct nameidata *nd)
1154 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1159 read_lock(¤t->fs->lock);
1160 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1161 nd->mnt = mntget(current->fs->altrootmnt);
1162 nd->dentry = dget(current->fs->altroot);
1163 read_unlock(¤t->fs->lock);
1164 if (__emul_lookup_dentry(name,nd))
1165 goto out; /* found in altroot */
1166 read_lock(¤t->fs->lock);
1168 nd->mnt = mntget(current->fs->rootmnt);
1169 nd->dentry = dget(current->fs->root);
1170 read_unlock(¤t->fs->lock);
1171 } else if (dfd == AT_FDCWD) {
1172 read_lock(¤t->fs->lock);
1173 nd->mnt = mntget(current->fs->pwdmnt);
1174 nd->dentry = dget(current->fs->pwd);
1175 read_unlock(¤t->fs->lock);
1177 struct dentry *dentry;
1179 file = fget_light(dfd, &fput_needed);
1184 dentry = file->f_dentry;
1187 if (!S_ISDIR(dentry->d_inode->i_mode))
1190 retval = file_permission(file, MAY_EXEC);
1194 nd->mnt = mntget(file->f_vfsmnt);
1195 nd->dentry = dget(dentry);
1197 fput_light(file, fput_needed);
1199 current->total_link_count = 0;
1200 retval = link_path_walk(name, nd);
1202 if (likely(retval == 0)) {
1203 if (unlikely(current->audit_context && nd && nd->dentry &&
1204 nd->dentry->d_inode))
1205 audit_inode(name, nd->dentry->d_inode, flags);
1211 fput_light(file, fput_needed);
1215 int fastcall path_lookup(const char *name, unsigned int flags,
1216 struct nameidata *nd)
1218 return do_path_lookup(AT_FDCWD, name, flags, nd);
1221 static int __path_lookup_intent_open(int dfd, const char *name,
1222 unsigned int lookup_flags, struct nameidata *nd,
1223 int open_flags, int create_mode)
1225 struct file *filp = get_empty_filp();
1230 nd->intent.open.file = filp;
1231 nd->intent.open.flags = open_flags;
1232 nd->intent.open.create_mode = create_mode;
1233 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1234 if (IS_ERR(nd->intent.open.file)) {
1236 err = PTR_ERR(nd->intent.open.file);
1239 } else if (err != 0)
1240 release_open_intent(nd);
1245 * path_lookup_open - lookup a file path with open intent
1246 * @dfd: the directory to use as base, or AT_FDCWD
1247 * @name: pointer to file name
1248 * @lookup_flags: lookup intent flags
1249 * @nd: pointer to nameidata
1250 * @open_flags: open intent flags
1252 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1253 struct nameidata *nd, int open_flags)
1255 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1260 * path_lookup_create - lookup a file path with open + create intent
1261 * @dfd: the directory to use as base, or AT_FDCWD
1262 * @name: pointer to file name
1263 * @lookup_flags: lookup intent flags
1264 * @nd: pointer to nameidata
1265 * @open_flags: open intent flags
1266 * @create_mode: create intent flags
1268 static int path_lookup_create(int dfd, const char *name,
1269 unsigned int lookup_flags, struct nameidata *nd,
1270 int open_flags, int create_mode)
1272 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1273 nd, open_flags, create_mode);
1276 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1277 struct nameidata *nd, int open_flags)
1279 char *tmp = getname(name);
1280 int err = PTR_ERR(tmp);
1283 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1290 * Restricted form of lookup. Doesn't follow links, single-component only,
1291 * needs parent already locked. Doesn't follow mounts.
1294 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1296 struct dentry * dentry;
1297 struct inode *inode;
1300 inode = base->d_inode;
1301 err = permission(inode, MAY_EXEC, nd);
1302 dentry = ERR_PTR(err);
1307 * See if the low-level filesystem might want
1308 * to use its own hash..
1310 if (base->d_op && base->d_op->d_hash) {
1311 err = base->d_op->d_hash(base, name);
1312 dentry = ERR_PTR(err);
1317 dentry = cached_lookup(base, name, nd);
1319 struct dentry *new = d_alloc(base, name);
1320 dentry = ERR_PTR(-ENOMEM);
1323 dentry = inode->i_op->lookup(inode, new, nd);
1333 static struct dentry *lookup_hash(struct nameidata *nd)
1335 return __lookup_hash(&nd->last, nd->dentry, nd);
1339 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1350 hash = init_name_hash();
1352 c = *(const unsigned char *)name++;
1353 if (c == '/' || c == '\0')
1355 hash = partial_name_hash(c, hash);
1357 this.hash = end_name_hash(hash);
1359 return __lookup_hash(&this, base, NULL);
1361 return ERR_PTR(-EACCES);
1367 * is used by most simple commands to get the inode of a specified name.
1368 * Open, link etc use their own routines, but this is enough for things
1371 * namei exists in two versions: namei/lnamei. The only difference is
1372 * that namei follows links, while lnamei does not.
1375 int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1376 struct nameidata *nd)
1378 char *tmp = getname(name);
1379 int err = PTR_ERR(tmp);
1382 err = do_path_lookup(dfd, tmp, flags, nd);
1388 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1390 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1394 * It's inline, so penalty for filesystems that don't use sticky bit is
1397 static inline int check_sticky(struct inode *dir, struct inode *inode)
1399 if (!(dir->i_mode & S_ISVTX))
1401 if (inode->i_uid == current->fsuid)
1403 if (dir->i_uid == current->fsuid)
1405 return !capable(CAP_FOWNER);
1409 * Check whether we can remove a link victim from directory dir, check
1410 * whether the type of victim is right.
1411 * 1. We can't do it if dir is read-only (done in permission())
1412 * 2. We should have write and exec permissions on dir
1413 * 3. We can't remove anything from append-only dir
1414 * 4. We can't do anything with immutable dir (done in permission())
1415 * 5. If the sticky bit on dir is set we should either
1416 * a. be owner of dir, or
1417 * b. be owner of victim, or
1418 * c. have CAP_FOWNER capability
1419 * 6. If the victim is append-only or immutable we can't do antyhing with
1420 * links pointing to it.
1421 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1422 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1423 * 9. We can't remove a root or mountpoint.
1424 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1425 * nfs_async_unlink().
1427 static int may_delete(struct inode *dir, struct dentry *victim,
1428 int isdir, struct nameidata *nd)
1432 if (!victim->d_inode)
1435 BUG_ON(victim->d_parent->d_inode != dir);
1436 audit_inode_child(victim->d_name.name, victim->d_inode, dir->i_ino);
1438 error = permission(dir,MAY_WRITE | MAY_EXEC, nd);
1443 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1444 IS_IXORUNLINK(victim->d_inode))
1447 if (!S_ISDIR(victim->d_inode->i_mode))
1449 if (IS_ROOT(victim))
1451 } else if (S_ISDIR(victim->d_inode->i_mode))
1453 if (IS_DEADDIR(dir))
1455 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1460 /* Check whether we can create an object with dentry child in directory
1462 * 1. We can't do it if child already exists (open has special treatment for
1463 * this case, but since we are inlined it's OK)
1464 * 2. We can't do it if dir is read-only (done in permission())
1465 * 3. We should have write and exec permissions on dir
1466 * 4. We can't do it if dir is immutable (done in permission())
1468 static inline int may_create(struct inode *dir, struct dentry *child,
1469 struct nameidata *nd)
1473 if (IS_DEADDIR(dir))
1475 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1479 * O_DIRECTORY translates into forcing a directory lookup.
1481 static inline int lookup_flags(unsigned int f)
1483 unsigned long retval = LOOKUP_FOLLOW;
1486 retval &= ~LOOKUP_FOLLOW;
1488 if (f & O_DIRECTORY)
1489 retval |= LOOKUP_DIRECTORY;
1490 if (f & O_ATOMICLOOKUP)
1491 retval |= LOOKUP_ATOMIC;
1497 * p1 and p2 should be directories on the same fs.
1499 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1504 mutex_lock(&p1->d_inode->i_mutex);
1508 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1510 for (p = p1; p->d_parent != p; p = p->d_parent) {
1511 if (p->d_parent == p2) {
1512 mutex_lock(&p2->d_inode->i_mutex);
1513 mutex_lock(&p1->d_inode->i_mutex);
1518 for (p = p2; p->d_parent != p; p = p->d_parent) {
1519 if (p->d_parent == p1) {
1520 mutex_lock(&p1->d_inode->i_mutex);
1521 mutex_lock(&p2->d_inode->i_mutex);
1526 mutex_lock(&p1->d_inode->i_mutex);
1527 mutex_lock(&p2->d_inode->i_mutex);
1531 void unlock_rename(struct dentry *p1, struct dentry *p2)
1533 mutex_unlock(&p1->d_inode->i_mutex);
1535 mutex_unlock(&p2->d_inode->i_mutex);
1536 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1540 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1541 struct nameidata *nd)
1543 int error = may_create(dir, dentry, nd);
1548 if (!dir->i_op || !dir->i_op->create)
1549 return -EACCES; /* shouldn't it be ENOSYS? */
1552 error = security_inode_create(dir, dentry, mode);
1556 error = dir->i_op->create(dir, dentry, mode, nd);
1558 fsnotify_create(dir, dentry);
1562 int may_open(struct nameidata *nd, int acc_mode, int flag)
1564 struct dentry *dentry = nd->dentry;
1565 struct inode *inode = dentry->d_inode;
1571 if (S_ISLNK(inode->i_mode))
1574 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1577 error = vfs_permission(nd, acc_mode);
1582 * FIFO's, sockets and device files are special: they don't
1583 * actually live on the filesystem itself, and as such you
1584 * can write to them even if the filesystem is read-only.
1586 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1588 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1589 if (nd->mnt->mnt_flags & MNT_NODEV)
1593 } else if ((IS_RDONLY(inode) || MNT_IS_RDONLY(nd->mnt))
1594 && (flag & FMODE_WRITE))
1597 * An append-only file must be opened in append mode for writing.
1599 if (IS_APPEND(inode)) {
1600 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1606 /* O_NOATIME can only be set by the owner or superuser */
1607 if (flag & O_NOATIME)
1608 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1612 * Ensure there are no outstanding leases on the file.
1614 error = break_lease(inode, flag);
1618 if (flag & O_TRUNC) {
1619 error = get_write_access(inode);
1624 * Refuse to truncate files with mandatory locks held on them.
1626 error = locks_verify_locked(inode);
1630 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
1632 put_write_access(inode);
1636 if (flag & FMODE_WRITE)
1645 * namei for open - this is in fact almost the whole open-routine.
1647 * Note that the low bits of "flag" aren't the same as in the open
1648 * system call - they are 00 - no permissions needed
1649 * 01 - read permission needed
1650 * 10 - write permission needed
1651 * 11 - read/write permissions needed
1652 * which is a lot more logical, and also allows the "no perm" needed
1653 * for symlinks (where the permissions are checked later).
1656 int open_namei(int dfd, const char *pathname, int flag,
1657 int mode, struct nameidata *nd)
1659 int acc_mode, error;
1664 acc_mode = ACC_MODE(flag);
1666 /* O_TRUNC implies we need access checks for write permissions */
1668 acc_mode |= MAY_WRITE;
1670 /* Allow the LSM permission hook to distinguish append
1671 access from general write access. */
1672 if (flag & O_APPEND)
1673 acc_mode |= MAY_APPEND;
1676 * The simplest case - just a plain lookup.
1678 if (!(flag & O_CREAT)) {
1679 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1687 * Create - we need to know the parent.
1689 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
1694 * We have the parent and last component. First of all, check
1695 * that we are not asked to creat(2) an obvious directory - that
1699 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1703 nd->flags &= ~LOOKUP_PARENT;
1704 mutex_lock(&dir->d_inode->i_mutex);
1705 path.dentry = lookup_hash(nd);
1709 error = PTR_ERR(path.dentry);
1710 if (IS_ERR(path.dentry)) {
1711 mutex_unlock(&dir->d_inode->i_mutex);
1715 if (IS_ERR(nd->intent.open.file)) {
1716 mutex_unlock(&dir->d_inode->i_mutex);
1717 error = PTR_ERR(nd->intent.open.file);
1721 /* Negative dentry, just create the file */
1722 if (!path.dentry->d_inode) {
1723 if (!IS_POSIXACL(dir->d_inode))
1724 mode &= ~current->fs->umask;
1725 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1726 mutex_unlock(&dir->d_inode->i_mutex);
1728 nd->dentry = path.dentry;
1731 /* Don't check for write permission, don't truncate */
1738 * It already exists.
1740 mutex_unlock(&dir->d_inode->i_mutex);
1746 if (__follow_mount(&path)) {
1748 if (flag & O_NOFOLLOW)
1752 if (!path.dentry->d_inode)
1754 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1757 path_to_nameidata(&path, nd);
1759 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1762 error = may_open(nd, acc_mode, flag);
1768 dput_path(&path, nd);
1770 if (!IS_ERR(nd->intent.open.file))
1771 release_open_intent(nd);
1777 if (flag & O_NOFOLLOW)
1780 * This is subtle. Instead of calling do_follow_link() we do the
1781 * thing by hands. The reason is that this way we have zero link_count
1782 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1783 * After that we have the parent and last component, i.e.
1784 * we are in the same situation as after the first path_walk().
1785 * Well, almost - if the last component is normal we get its copy
1786 * stored in nd->last.name and we will have to putname() it when we
1787 * are done. Procfs-like symlinks just set LAST_BIND.
1789 nd->flags |= LOOKUP_PARENT;
1790 error = security_inode_follow_link(path.dentry, nd);
1793 error = __do_follow_link(&path, nd);
1795 /* Does someone understand code flow here? Or it is only
1796 * me so stupid? Anathema to whoever designed this non-sense
1797 * with "intent.open".
1799 release_open_intent(nd);
1802 nd->flags &= ~LOOKUP_PARENT;
1803 if (nd->last_type == LAST_BIND)
1806 if (nd->last_type != LAST_NORM)
1808 if (nd->last.name[nd->last.len]) {
1809 __putname(nd->last.name);
1814 __putname(nd->last.name);
1818 mutex_lock(&dir->d_inode->i_mutex);
1819 path.dentry = lookup_hash(nd);
1821 __putname(nd->last.name);
1826 * lookup_create - lookup a dentry, creating it if it doesn't exist
1827 * @nd: nameidata info
1828 * @is_dir: directory flag
1830 * Simple function to lookup and return a dentry and create it
1831 * if it doesn't exist. Is SMP-safe.
1833 * Returns with nd->dentry->d_inode->i_mutex locked.
1835 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1837 struct dentry *dentry = ERR_PTR(-EEXIST);
1839 mutex_lock(&nd->dentry->d_inode->i_mutex);
1841 * Yucky last component or no last component at all?
1842 * (foo/., foo/.., /////)
1844 if (nd->last_type != LAST_NORM)
1846 nd->flags &= ~LOOKUP_PARENT;
1849 * Do the final lookup.
1851 dentry = lookup_hash(nd);
1856 * Special case - lookup gave negative, but... we had foo/bar/
1857 * From the vfs_mknod() POV we just have a negative dentry -
1858 * all is fine. Let's be bastards - you had / on the end, you've
1859 * been asking for (non-existent) directory. -ENOENT for you.
1861 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1866 dentry = ERR_PTR(-ENOENT);
1870 EXPORT_SYMBOL_GPL(lookup_create);
1872 int vfs_mknod(struct inode *dir, struct dentry *dentry,
1873 int mode, dev_t dev, struct nameidata *nd)
1875 int error = may_create(dir, dentry, nd);
1880 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1883 if (!dir->i_op || !dir->i_op->mknod)
1886 error = security_inode_mknod(dir, dentry, mode, dev);
1891 error = dir->i_op->mknod(dir, dentry, mode, dev);
1893 fsnotify_create(dir, dentry);
1897 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1902 struct dentry * dentry;
1903 struct nameidata nd;
1907 tmp = getname(filename);
1909 return PTR_ERR(tmp);
1911 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1914 dentry = lookup_create(&nd, 0);
1915 error = PTR_ERR(dentry);
1917 if (!IS_POSIXACL(nd.dentry->d_inode))
1918 mode &= ~current->fs->umask;
1919 if (!IS_ERR(dentry)) {
1920 switch (mode & S_IFMT) {
1921 case 0: case S_IFREG:
1922 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1924 case S_IFCHR: case S_IFBLK:
1925 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1926 new_decode_dev(dev), &nd);
1928 case S_IFIFO: case S_IFSOCK:
1929 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1940 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1948 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1950 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1953 int vfs_mkdir(struct inode *dir, struct dentry *dentry,
1954 int mode, struct nameidata *nd)
1956 int error = may_create(dir, dentry, nd);
1961 if (!dir->i_op || !dir->i_op->mkdir)
1964 mode &= (S_IRWXUGO|S_ISVTX);
1965 error = security_inode_mkdir(dir, dentry, mode);
1970 error = dir->i_op->mkdir(dir, dentry, mode);
1972 fsnotify_mkdir(dir, dentry);
1976 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
1981 tmp = getname(pathname);
1982 error = PTR_ERR(tmp);
1984 struct dentry *dentry;
1985 struct nameidata nd;
1987 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1990 dentry = lookup_create(&nd, 1);
1991 error = PTR_ERR(dentry);
1992 if (!IS_ERR(dentry)) {
1993 if (!IS_POSIXACL(nd.dentry->d_inode))
1994 mode &= ~current->fs->umask;
1995 error = vfs_mkdir(nd.dentry->d_inode, dentry,
1999 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2008 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2010 return sys_mkdirat(AT_FDCWD, pathname, mode);
2014 * We try to drop the dentry early: we should have
2015 * a usage count of 2 if we're the only user of this
2016 * dentry, and if that is true (possibly after pruning
2017 * the dcache), then we drop the dentry now.
2019 * A low-level filesystem can, if it choses, legally
2022 * if (!d_unhashed(dentry))
2025 * if it cannot handle the case of removing a directory
2026 * that is still in use by something else..
2028 void dentry_unhash(struct dentry *dentry)
2031 if (atomic_read(&dentry->d_count))
2032 shrink_dcache_parent(dentry);
2033 spin_lock(&dcache_lock);
2034 spin_lock(&dentry->d_lock);
2035 if (atomic_read(&dentry->d_count) == 2)
2037 spin_unlock(&dentry->d_lock);
2038 spin_unlock(&dcache_lock);
2041 int vfs_rmdir(struct inode *dir, struct dentry *dentry,
2042 struct nameidata *nd)
2044 int error = may_delete(dir, dentry, 1, nd);
2049 if (!dir->i_op || !dir->i_op->rmdir)
2054 mutex_lock(&dentry->d_inode->i_mutex);
2055 dentry_unhash(dentry);
2056 if (d_mountpoint(dentry))
2059 error = security_inode_rmdir(dir, dentry);
2061 error = dir->i_op->rmdir(dir, dentry);
2063 dentry->d_inode->i_flags |= S_DEAD;
2066 mutex_unlock(&dentry->d_inode->i_mutex);
2075 static long do_rmdir(int dfd, const char __user *pathname)
2079 struct dentry *dentry;
2080 struct nameidata nd;
2082 name = getname(pathname);
2084 return PTR_ERR(name);
2086 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2090 switch(nd.last_type) {
2101 mutex_lock(&nd.dentry->d_inode->i_mutex);
2102 dentry = lookup_hash(&nd);
2103 error = PTR_ERR(dentry);
2104 if (!IS_ERR(dentry)) {
2105 error = vfs_rmdir(nd.dentry->d_inode, dentry, &nd);
2108 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2116 asmlinkage long sys_rmdir(const char __user *pathname)
2118 return do_rmdir(AT_FDCWD, pathname);
2121 int vfs_unlink(struct inode *dir, struct dentry *dentry,
2122 struct nameidata *nd)
2124 int error = may_delete(dir, dentry, 0, nd);
2129 if (!dir->i_op || !dir->i_op->unlink)
2134 mutex_lock(&dentry->d_inode->i_mutex);
2135 if (d_mountpoint(dentry))
2138 error = security_inode_unlink(dir, dentry);
2140 error = dir->i_op->unlink(dir, dentry);
2142 mutex_unlock(&dentry->d_inode->i_mutex);
2144 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2145 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2153 * Make sure that the actual truncation of the file will occur outside its
2154 * directory's i_mutex. Truncate can take a long time if there is a lot of
2155 * writeout happening, and we don't want to prevent access to the directory
2156 * while waiting on the I/O.
2158 static long do_unlinkat(int dfd, const char __user *pathname)
2162 struct dentry *dentry;
2163 struct nameidata nd;
2164 struct inode *inode = NULL;
2166 name = getname(pathname);
2168 return PTR_ERR(name);
2170 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2174 if (nd.last_type != LAST_NORM)
2176 mutex_lock(&nd.dentry->d_inode->i_mutex);
2177 dentry = lookup_hash(&nd);
2178 error = PTR_ERR(dentry);
2179 if (!IS_ERR(dentry)) {
2180 /* Why not before? Because we want correct error value */
2181 if (nd.last.name[nd.last.len])
2183 inode = dentry->d_inode;
2185 atomic_inc(&inode->i_count);
2186 error = vfs_unlink(nd.dentry->d_inode, dentry, &nd);
2190 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2192 iput(inode); /* truncate the inode here */
2200 error = !dentry->d_inode ? -ENOENT :
2201 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2205 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2207 if ((flag & ~AT_REMOVEDIR) != 0)
2210 if (flag & AT_REMOVEDIR)
2211 return do_rmdir(dfd, pathname);
2213 return do_unlinkat(dfd, pathname);
2216 asmlinkage long sys_unlink(const char __user *pathname)
2218 return do_unlinkat(AT_FDCWD, pathname);
2221 int vfs_symlink(struct inode *dir, struct dentry *dentry,
2222 const char *oldname, int mode, struct nameidata *nd)
2224 int error = may_create(dir, dentry, nd);
2229 if (!dir->i_op || !dir->i_op->symlink)
2232 error = security_inode_symlink(dir, dentry, oldname);
2237 error = dir->i_op->symlink(dir, dentry, oldname);
2239 fsnotify_create(dir, dentry);
2243 asmlinkage long sys_symlinkat(const char __user *oldname,
2244 int newdfd, const char __user *newname)
2250 from = getname(oldname);
2252 return PTR_ERR(from);
2253 to = getname(newname);
2254 error = PTR_ERR(to);
2256 struct dentry *dentry;
2257 struct nameidata nd;
2259 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2262 dentry = lookup_create(&nd, 0);
2263 error = PTR_ERR(dentry);
2264 if (!IS_ERR(dentry)) {
2265 error = vfs_symlink(nd.dentry->d_inode, dentry,
2266 from, S_IALLUGO, &nd);
2269 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2278 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2280 return sys_symlinkat(oldname, AT_FDCWD, newname);
2283 int vfs_link(struct dentry *old_dentry, struct inode *dir,
2284 struct dentry *new_dentry, struct nameidata *nd)
2286 struct inode *inode = old_dentry->d_inode;
2292 error = may_create(dir, new_dentry, nd);
2296 if (dir->i_sb != inode->i_sb)
2300 * A link to an append-only or immutable file cannot be created.
2302 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
2304 if (!dir->i_op || !dir->i_op->link)
2306 if (S_ISDIR(old_dentry->d_inode->i_mode))
2309 error = security_inode_link(old_dentry, dir, new_dentry);
2313 mutex_lock(&old_dentry->d_inode->i_mutex);
2315 error = dir->i_op->link(old_dentry, dir, new_dentry);
2316 mutex_unlock(&old_dentry->d_inode->i_mutex);
2318 fsnotify_create(dir, new_dentry);
2323 * Hardlinks are often used in delicate situations. We avoid
2324 * security-related surprises by not following symlinks on the
2327 * We don't follow them on the oldname either to be compatible
2328 * with linux 2.0, and to avoid hard-linking to directories
2329 * and other special files. --ADM
2331 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2332 int newdfd, const char __user *newname,
2335 struct dentry *new_dentry;
2336 struct nameidata nd, old_nd;
2343 to = getname(newname);
2347 error = __user_walk_fd(olddfd, oldname, 0, &old_nd);
2350 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2354 * We allow hard-links to be created to a bind-mount as long
2355 * as the bind-mount is not read-only. Checking for cross-dev
2356 * links is subsumed by the superblock check in vfs_link().
2359 if (MNT_IS_RDONLY(old_nd.mnt))
2361 new_dentry = lookup_create(&nd, 0);
2362 error = PTR_ERR(new_dentry);
2363 if (!IS_ERR(new_dentry)) {
2364 error = vfs_link(old_nd.dentry, nd.dentry->d_inode,
2368 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2372 path_release(&old_nd);
2379 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2381 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2385 * The worst of all namespace operations - renaming directory. "Perverted"
2386 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2388 * a) we can get into loop creation. Check is done in is_subdir().
2389 * b) race potential - two innocent renames can create a loop together.
2390 * That's where 4.4 screws up. Current fix: serialization on
2391 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2393 * c) we have to lock _three_ objects - parents and victim (if it exists).
2394 * And that - after we got ->i_mutex on parents (until then we don't know
2395 * whether the target exists). Solution: try to be smart with locking
2396 * order for inodes. We rely on the fact that tree topology may change
2397 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2398 * move will be locked. Thus we can rank directories by the tree
2399 * (ancestors first) and rank all non-directories after them.
2400 * That works since everybody except rename does "lock parent, lookup,
2401 * lock child" and rename is under ->s_vfs_rename_mutex.
2402 * HOWEVER, it relies on the assumption that any object with ->lookup()
2403 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2404 * we'd better make sure that there's no link(2) for them.
2405 * d) some filesystems don't support opened-but-unlinked directories,
2406 * either because of layout or because they are not ready to deal with
2407 * all cases correctly. The latter will be fixed (taking this sort of
2408 * stuff into VFS), but the former is not going away. Solution: the same
2409 * trick as in rmdir().
2410 * e) conversion from fhandle to dentry may come in the wrong moment - when
2411 * we are removing the target. Solution: we will have to grab ->i_mutex
2412 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2413 * ->i_mutex on parents, which works but leads to some truely excessive
2416 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2417 struct inode *new_dir, struct dentry *new_dentry)
2420 struct inode *target;
2423 * If we are going to change the parent - check write permissions,
2424 * we'll need to flip '..'.
2426 if (new_dir != old_dir) {
2427 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2432 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2436 target = new_dentry->d_inode;
2438 mutex_lock(&target->i_mutex);
2439 dentry_unhash(new_dentry);
2441 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2444 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2447 target->i_flags |= S_DEAD;
2448 mutex_unlock(&target->i_mutex);
2449 if (d_unhashed(new_dentry))
2450 d_rehash(new_dentry);
2454 d_move(old_dentry,new_dentry);
2458 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2459 struct inode *new_dir, struct dentry *new_dentry)
2461 struct inode *target;
2464 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2469 target = new_dentry->d_inode;
2471 mutex_lock(&target->i_mutex);
2472 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2475 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2477 /* The following d_move() should become unconditional */
2478 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2479 d_move(old_dentry, new_dentry);
2482 mutex_unlock(&target->i_mutex);
2487 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2488 struct inode *new_dir, struct dentry *new_dentry)
2491 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2492 const char *old_name;
2494 if (old_dentry->d_inode == new_dentry->d_inode)
2497 error = may_delete(old_dir, old_dentry, is_dir, NULL);
2501 if (!new_dentry->d_inode)
2502 error = may_create(new_dir, new_dentry, NULL);
2504 error = may_delete(new_dir, new_dentry, is_dir, NULL);
2508 if (!old_dir->i_op || !old_dir->i_op->rename)
2511 DQUOT_INIT(old_dir);
2512 DQUOT_INIT(new_dir);
2514 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2517 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2519 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2521 const char *new_name = old_dentry->d_name.name;
2522 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2523 new_dentry->d_inode, old_dentry->d_inode);
2525 fsnotify_oldname_free(old_name);
2530 static int do_rename(int olddfd, const char *oldname,
2531 int newdfd, const char *newname)
2534 struct dentry * old_dir, * new_dir;
2535 struct dentry * old_dentry, *new_dentry;
2536 struct dentry * trap;
2537 struct nameidata oldnd, newnd;
2539 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2543 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2548 if (oldnd.mnt != newnd.mnt)
2551 old_dir = oldnd.dentry;
2553 if (oldnd.last_type != LAST_NORM)
2556 new_dir = newnd.dentry;
2557 if (newnd.last_type != LAST_NORM)
2560 trap = lock_rename(new_dir, old_dir);
2562 old_dentry = lookup_hash(&oldnd);
2563 error = PTR_ERR(old_dentry);
2564 if (IS_ERR(old_dentry))
2566 /* source must exist */
2568 if (!old_dentry->d_inode)
2570 /* unless the source is a directory trailing slashes give -ENOTDIR */
2571 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2573 if (oldnd.last.name[oldnd.last.len])
2575 if (newnd.last.name[newnd.last.len])
2578 /* source should not be ancestor of target */
2580 if (old_dentry == trap)
2583 if (MNT_IS_RDONLY(newnd.mnt))
2585 new_dentry = lookup_hash(&newnd);
2586 error = PTR_ERR(new_dentry);
2587 if (IS_ERR(new_dentry))
2589 /* target should not be an ancestor of source */
2591 if (new_dentry == trap)
2594 error = vfs_rename(old_dir->d_inode, old_dentry,
2595 new_dir->d_inode, new_dentry);
2601 unlock_rename(new_dir, old_dir);
2603 path_release(&newnd);
2605 path_release(&oldnd);
2610 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2611 int newdfd, const char __user *newname)
2617 from = getname(oldname);
2619 return PTR_ERR(from);
2620 to = getname(newname);
2621 error = PTR_ERR(to);
2623 error = do_rename(olddfd, from, newdfd, to);
2630 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2632 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2635 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2639 len = PTR_ERR(link);
2644 if (len > (unsigned) buflen)
2646 if (copy_to_user(buffer, link, len))
2653 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2654 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2655 * using) it for any given inode is up to filesystem.
2657 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2659 struct nameidata nd;
2663 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2664 if (!IS_ERR(cookie)) {
2665 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2666 if (dentry->d_inode->i_op->put_link)
2667 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2668 cookie = ERR_PTR(res);
2670 return PTR_ERR(cookie);
2673 int vfs_follow_link(struct nameidata *nd, const char *link)
2675 return __vfs_follow_link(nd, link);
2678 /* get the link contents into pagecache */
2679 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2682 struct address_space *mapping = dentry->d_inode->i_mapping;
2683 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2687 wait_on_page_locked(page);
2688 if (!PageUptodate(page))
2694 page_cache_release(page);
2695 return ERR_PTR(-EIO);
2701 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2703 struct page *page = NULL;
2704 char *s = page_getlink(dentry, &page);
2705 int res = vfs_readlink(dentry,buffer,buflen,s);
2708 page_cache_release(page);
2713 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2715 struct page *page = NULL;
2716 nd_set_link(nd, page_getlink(dentry, &page));
2720 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2722 struct page *page = cookie;
2726 page_cache_release(page);
2730 int __page_symlink(struct inode *inode, const char *symname, int len,
2733 struct address_space *mapping = inode->i_mapping;
2739 page = find_or_create_page(mapping, 0, gfp_mask);
2742 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2743 if (err == AOP_TRUNCATED_PAGE) {
2744 page_cache_release(page);
2749 kaddr = kmap_atomic(page, KM_USER0);
2750 memcpy(kaddr, symname, len-1);
2751 kunmap_atomic(kaddr, KM_USER0);
2752 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2753 if (err == AOP_TRUNCATED_PAGE) {
2754 page_cache_release(page);
2760 * Notice that we are _not_ going to block here - end of page is
2761 * unmapped, so this will only try to map the rest of page, see
2762 * that it is unmapped (typically even will not look into inode -
2763 * ->i_size will be enough for everything) and zero it out.
2764 * OTOH it's obviously correct and should make the page up-to-date.
2766 if (!PageUptodate(page)) {
2767 err = mapping->a_ops->readpage(NULL, page);
2768 if (err != AOP_TRUNCATED_PAGE)
2769 wait_on_page_locked(page);
2773 page_cache_release(page);
2776 mark_inode_dirty(inode);
2780 page_cache_release(page);
2785 int page_symlink(struct inode *inode, const char *symname, int len)
2787 return __page_symlink(inode, symname, len,
2788 mapping_gfp_mask(inode->i_mapping));
2791 struct inode_operations page_symlink_inode_operations = {
2792 .readlink = generic_readlink,
2793 .follow_link = page_follow_link_light,
2794 .put_link = page_put_link,
2797 EXPORT_SYMBOL(__user_walk);
2798 EXPORT_SYMBOL(__user_walk_fd);
2799 EXPORT_SYMBOL(follow_down);
2800 EXPORT_SYMBOL(follow_up);
2801 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2802 EXPORT_SYMBOL(getname);
2803 EXPORT_SYMBOL(lock_rename);
2804 EXPORT_SYMBOL(lookup_one_len);
2805 EXPORT_SYMBOL(page_follow_link_light);
2806 EXPORT_SYMBOL(page_put_link);
2807 EXPORT_SYMBOL(page_readlink);
2808 EXPORT_SYMBOL(__page_symlink);
2809 EXPORT_SYMBOL(page_symlink);
2810 EXPORT_SYMBOL(page_symlink_inode_operations);
2811 EXPORT_SYMBOL(path_lookup);
2812 EXPORT_SYMBOL(path_release);
2813 EXPORT_SYMBOL(path_walk);
2814 EXPORT_SYMBOL(permission);
2815 EXPORT_SYMBOL(vfs_permission);
2816 EXPORT_SYMBOL(file_permission);
2817 EXPORT_SYMBOL(unlock_rename);
2818 EXPORT_SYMBOL(vfs_create);
2819 EXPORT_SYMBOL(vfs_follow_link);
2820 EXPORT_SYMBOL(vfs_link);
2821 EXPORT_SYMBOL(vfs_mkdir);
2822 EXPORT_SYMBOL(vfs_mknod);
2823 EXPORT_SYMBOL(generic_permission);
2824 EXPORT_SYMBOL(vfs_readlink);
2825 EXPORT_SYMBOL(vfs_rename);
2826 EXPORT_SYMBOL(vfs_rmdir);
2827 EXPORT_SYMBOL(vfs_symlink);
2828 EXPORT_SYMBOL(vfs_unlink);
2829 EXPORT_SYMBOL(dentry_unhash);
2830 EXPORT_SYMBOL(generic_readlink);