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 if (vx_check(inode->i_xid, VX_ADMIN|VX_WATCH|VX_IDENT))
250 vxwprintk(1, "xid=%d denied access to %p[#%d,%lu] »%s«.",
251 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
256 int permission(struct inode *inode, int mask, struct nameidata *nd)
260 if (mask & MAY_WRITE) {
261 umode_t mode = inode->i_mode;
264 * Nobody gets write access to a read-only fs.
266 if ((IS_RDONLY(inode) || (nd && MNT_IS_RDONLY(nd->mnt))) &&
267 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
271 * Nobody gets write access to an immutable file.
273 if (IS_IMMUTABLE(inode))
278 /* Ordinary permission routines do not understand MAY_APPEND. */
279 submask = mask & ~MAY_APPEND;
280 if ((retval = xid_permission(inode, mask, nd)))
282 if (inode->i_op && inode->i_op->permission)
283 retval = inode->i_op->permission(inode, submask, nd);
285 retval = generic_permission(inode, submask, NULL);
289 return security_inode_permission(inode, mask, nd);
293 * vfs_permission - check for access rights to a given path
294 * @nd: lookup result that describes the path
295 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
297 * Used to check for read/write/execute permissions on a path.
298 * We use "fsuid" for this, letting us set arbitrary permissions
299 * for filesystem access without changing the "normal" uids which
300 * are used for other things.
302 int vfs_permission(struct nameidata *nd, int mask)
304 return permission(nd->dentry->d_inode, mask, nd);
308 * file_permission - check for additional access rights to a given file
309 * @file: file to check access rights for
310 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
312 * Used to check for read/write/execute permissions on an already opened
316 * Do not use this function in new code. All access checks should
317 * be done using vfs_permission().
319 int file_permission(struct file *file, int mask)
321 return permission(file->f_dentry->d_inode, mask, NULL);
325 * get_write_access() gets write permission for a file.
326 * put_write_access() releases this write permission.
327 * This is used for regular files.
328 * We cannot support write (and maybe mmap read-write shared) accesses and
329 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
330 * can have the following values:
331 * 0: no writers, no VM_DENYWRITE mappings
332 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
333 * > 0: (i_writecount) users are writing to the file.
335 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
336 * except for the cases where we don't hold i_writecount yet. Then we need to
337 * use {get,deny}_write_access() - these functions check the sign and refuse
338 * to do the change if sign is wrong. Exclusion between them is provided by
339 * the inode->i_lock spinlock.
342 int get_write_access(struct inode * inode)
344 spin_lock(&inode->i_lock);
345 if (atomic_read(&inode->i_writecount) < 0) {
346 spin_unlock(&inode->i_lock);
349 atomic_inc(&inode->i_writecount);
350 spin_unlock(&inode->i_lock);
355 int deny_write_access(struct file * file)
357 struct inode *inode = file->f_dentry->d_inode;
359 spin_lock(&inode->i_lock);
360 if (atomic_read(&inode->i_writecount) > 0) {
361 spin_unlock(&inode->i_lock);
364 atomic_dec(&inode->i_writecount);
365 spin_unlock(&inode->i_lock);
370 void path_release(struct nameidata *nd)
377 * umount() mustn't call path_release()/mntput() as that would clear
380 void path_release_on_umount(struct nameidata *nd)
383 mntput_no_expire(nd->mnt);
387 * release_open_intent - free up open intent resources
388 * @nd: pointer to nameidata
390 void release_open_intent(struct nameidata *nd)
392 if (nd->intent.open.file->f_dentry == NULL)
393 put_filp(nd->intent.open.file);
395 fput(nd->intent.open.file);
399 * Internal lookup() using the new generic dcache.
402 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
404 struct dentry * dentry = __d_lookup(parent, name);
406 /* lockess __d_lookup may fail due to concurrent d_move()
407 * in some unrelated directory, so try with d_lookup
410 dentry = d_lookup(parent, name);
412 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
413 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
422 * Short-cut version of permission(), for calling by
423 * path_walk(), when dcache lock is held. Combines parts
424 * of permission() and generic_permission(), and tests ONLY for
425 * MAY_EXEC permission.
427 * If appropriate, check DAC only. If not appropriate, or
428 * short-cut DAC fails, then call permission() to do more
429 * complete permission check.
431 static int exec_permission_lite(struct inode *inode,
432 struct nameidata *nd)
434 umode_t mode = inode->i_mode;
436 if (vx_barrier(inode))
438 if (inode->i_op && inode->i_op->permission)
441 if (current->fsuid == inode->i_uid)
443 else if (in_group_p(inode->i_gid))
449 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
452 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
455 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
460 return security_inode_permission(inode, MAY_EXEC, nd);
464 * This is called when everything else fails, and we actually have
465 * to go to the low-level filesystem to find out what we should do..
467 * We get the directory semaphore, and after getting that we also
468 * make sure that nobody added the entry to the dcache in the meantime..
471 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
473 struct dentry * result;
474 struct inode *dir = parent->d_inode;
476 mutex_lock(&dir->i_mutex);
478 * First re-do the cached lookup just in case it was created
479 * while we waited for the directory semaphore..
481 * FIXME! This could use version numbering or similar to
482 * avoid unnecessary cache lookups.
484 * The "dcache_lock" is purely to protect the RCU list walker
485 * from concurrent renames at this point (we mustn't get false
486 * negatives from the RCU list walk here, unlike the optimistic
489 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
491 result = d_lookup(parent, name);
493 struct dentry * dentry = d_alloc(parent, name);
494 result = ERR_PTR(-ENOMEM);
496 result = dir->i_op->lookup(dir, dentry, nd);
502 mutex_unlock(&dir->i_mutex);
507 * Uhhuh! Nasty case: the cache was re-populated while
508 * we waited on the semaphore. Need to revalidate.
510 mutex_unlock(&dir->i_mutex);
511 if (result->d_op && result->d_op->d_revalidate) {
512 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
514 result = ERR_PTR(-ENOENT);
520 static int __emul_lookup_dentry(const char *, struct nameidata *);
523 static __always_inline int
524 walk_init_root(const char *name, struct nameidata *nd)
526 read_lock(¤t->fs->lock);
527 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
528 nd->mnt = mntget(current->fs->altrootmnt);
529 nd->dentry = dget(current->fs->altroot);
530 read_unlock(¤t->fs->lock);
531 if (__emul_lookup_dentry(name,nd))
533 read_lock(¤t->fs->lock);
535 nd->mnt = mntget(current->fs->rootmnt);
536 nd->dentry = dget(current->fs->root);
537 read_unlock(¤t->fs->lock);
541 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
550 if (!walk_init_root(link, nd))
551 /* weird __emul_prefix() stuff did it */
554 res = link_path_walk(link, nd);
556 if (nd->depth || res || nd->last_type!=LAST_NORM)
559 * If it is an iterative symlinks resolution in open_namei() we
560 * have to copy the last component. And all that crap because of
561 * bloody create() on broken symlinks. Furrfu...
564 if (unlikely(!name)) {
568 strcpy(name, nd->last.name);
569 nd->last.name = name;
573 return PTR_ERR(link);
577 struct vfsmount *mnt;
578 struct dentry *dentry;
581 static inline void dput_path(struct path *path, struct nameidata *nd)
584 if (path->mnt != nd->mnt)
588 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
591 if (nd->mnt != path->mnt)
594 nd->dentry = path->dentry;
597 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
601 struct dentry *dentry = path->dentry;
603 touch_atime(path->mnt, dentry);
604 nd_set_link(nd, NULL);
606 if (path->mnt != nd->mnt) {
607 path_to_nameidata(path, nd);
611 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
612 error = PTR_ERR(cookie);
613 if (!IS_ERR(cookie)) {
614 char *s = nd_get_link(nd);
617 error = __vfs_follow_link(nd, s);
618 if (dentry->d_inode->i_op->put_link)
619 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
628 * This limits recursive symlink follows to 8, while
629 * limiting consecutive symlinks to 40.
631 * Without that kind of total limit, nasty chains of consecutive
632 * symlinks can cause almost arbitrarily long lookups.
634 static inline int do_follow_link(struct path *path, struct nameidata *nd)
637 if (current->link_count >= MAX_NESTED_LINKS)
639 if (current->total_link_count >= 40)
641 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
643 err = security_inode_follow_link(path->dentry, nd);
646 current->link_count++;
647 current->total_link_count++;
649 err = __do_follow_link(path, nd);
650 current->link_count--;
659 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
661 struct vfsmount *parent;
662 struct dentry *mountpoint;
663 spin_lock(&vfsmount_lock);
664 parent=(*mnt)->mnt_parent;
665 if (parent == *mnt) {
666 spin_unlock(&vfsmount_lock);
670 mountpoint=dget((*mnt)->mnt_mountpoint);
671 spin_unlock(&vfsmount_lock);
673 *dentry = mountpoint;
679 /* no need for dcache_lock, as serialization is taken care in
682 static int __follow_mount(struct path *path)
685 while (d_mountpoint(path->dentry)) {
686 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
693 path->dentry = dget(mounted->mnt_root);
699 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
701 while (d_mountpoint(*dentry)) {
702 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
708 *dentry = dget(mounted->mnt_root);
712 /* no need for dcache_lock, as serialization is taken care in
715 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
717 struct vfsmount *mounted;
719 mounted = lookup_mnt(*mnt, *dentry);
724 *dentry = dget(mounted->mnt_root);
730 static __always_inline void follow_dotdot(struct nameidata *nd)
733 struct vfsmount *parent;
734 struct dentry *old = nd->dentry;
736 read_lock(¤t->fs->lock);
737 if (nd->dentry == current->fs->root &&
738 nd->mnt == current->fs->rootmnt) {
739 read_unlock(¤t->fs->lock);
740 /* for sane '/' avoid follow_mount() */
743 read_unlock(¤t->fs->lock);
744 spin_lock(&dcache_lock);
745 if (nd->dentry != nd->mnt->mnt_root) {
746 nd->dentry = dget(nd->dentry->d_parent);
747 spin_unlock(&dcache_lock);
751 spin_unlock(&dcache_lock);
752 spin_lock(&vfsmount_lock);
753 parent = nd->mnt->mnt_parent;
754 if (parent == nd->mnt) {
755 spin_unlock(&vfsmount_lock);
759 nd->dentry = dget(nd->mnt->mnt_mountpoint);
760 spin_unlock(&vfsmount_lock);
765 follow_mount(&nd->mnt, &nd->dentry);
769 * It's more convoluted than I'd like it to be, but... it's still fairly
770 * small and for now I'd prefer to have fast path as straight as possible.
771 * It _is_ time-critical.
773 static int do_lookup(struct nameidata *nd, struct qstr *name,
774 struct path *path, int atomic)
776 struct vfsmount *mnt = nd->mnt;
777 struct dentry *dentry = __d_lookup(nd->dentry, name);
782 if (dentry->d_op && dentry->d_op->d_revalidate)
783 goto need_revalidate;
784 inode = dentry->d_inode;
787 if (!vx_check(inode->i_xid, VX_WATCH|VX_ADMIN|VX_HOSTID|VX_IDENT))
789 if (inode->i_sb->s_magic == PROC_SUPER_MAGIC) {
790 struct proc_dir_entry *de = PDE(inode);
792 if (de && !vx_hide_check(0, de->vx_flags))
797 path->dentry = dentry;
798 __follow_mount(path);
801 vxwprintk(1, "xid=%d did lookup hidden %p[#%d,%lu] »%s«.",
802 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
803 vxd_path(dentry, mnt));
809 return -EWOULDBLOCKIO;
810 dentry = real_lookup(nd->dentry, name, nd);
817 return -EWOULDBLOCKIO;
818 if (dentry->d_op->d_revalidate(dentry, nd))
820 if (d_invalidate(dentry))
826 return PTR_ERR(dentry);
831 * This is the basic name resolution function, turning a pathname into
832 * the final dentry. We expect 'base' to be positive and a directory.
834 * Returns 0 and nd will have valid dentry and mnt on success.
835 * Returns error and drops reference to input namei data on failure.
837 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
842 unsigned int lookup_flags = nd->flags;
844 atomic = (lookup_flags & LOOKUP_ATOMIC);
851 inode = nd->dentry->d_inode;
853 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
855 /* At this point we know we have a real path component. */
861 nd->flags |= LOOKUP_CONTINUE;
862 err = exec_permission_lite(inode, nd);
864 err = vfs_permission(nd, MAY_EXEC);
869 c = *(const unsigned char *)name;
871 hash = init_name_hash();
874 hash = partial_name_hash(c, hash);
875 c = *(const unsigned char *)name;
876 } while (c && (c != '/'));
877 this.len = name - (const char *) this.name;
878 this.hash = end_name_hash(hash);
880 /* remove trailing slashes? */
883 while (*++name == '/');
885 goto last_with_slashes;
888 * "." and ".." are special - ".." especially so because it has
889 * to be able to know about the current root directory and
890 * parent relationships.
892 if (this.name[0] == '.') switch (this.len) {
896 if (this.name[1] != '.')
899 inode = nd->dentry->d_inode;
905 * See if the low-level filesystem might want
906 * to use its own hash..
908 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
909 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
913 /* This does the actual lookups.. */
914 err = do_lookup(nd, &this, &next, atomic);
919 inode = next.dentry->d_inode;
926 if (inode->i_op->follow_link) {
927 err = do_follow_link(&next, nd);
931 inode = nd->dentry->d_inode;
938 path_to_nameidata(&next, nd);
940 if (!inode->i_op->lookup)
943 /* here ends the main loop */
946 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
948 /* Clear LOOKUP_CONTINUE iff it was previously unset */
949 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
950 if (lookup_flags & LOOKUP_PARENT)
952 if (this.name[0] == '.') switch (this.len) {
956 if (this.name[1] != '.')
959 inode = nd->dentry->d_inode;
964 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
965 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
969 err = do_lookup(nd, &this, &next, atomic);
972 inode = next.dentry->d_inode;
973 if ((lookup_flags & LOOKUP_FOLLOW)
974 && inode && inode->i_op && inode->i_op->follow_link) {
975 err = do_follow_link(&next, nd);
978 inode = nd->dentry->d_inode;
980 path_to_nameidata(&next, nd);
984 if (lookup_flags & LOOKUP_DIRECTORY) {
986 if (!inode->i_op || !inode->i_op->lookup)
992 nd->last_type = LAST_NORM;
993 if (this.name[0] != '.')
996 nd->last_type = LAST_DOT;
997 else if (this.len == 2 && this.name[1] == '.')
998 nd->last_type = LAST_DOTDOT;
1003 * We bypassed the ordinary revalidation routines.
1004 * We may need to check the cached dentry for staleness.
1006 if (nd->dentry && nd->dentry->d_sb &&
1007 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1009 /* Note: we do not d_invalidate() */
1010 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
1016 dput_path(&next, nd);
1025 * Wrapper to retry pathname resolution whenever the underlying
1026 * file system returns an ESTALE.
1028 * Retry the whole path once, forcing real lookup requests
1029 * instead of relying on the dcache.
1031 int fastcall link_path_walk(const char *name, struct nameidata *nd)
1033 struct nameidata save = *nd;
1036 /* make sure the stuff we saved doesn't go away */
1040 result = __link_path_walk(name, nd);
1041 if (result == -ESTALE) {
1045 nd->flags |= LOOKUP_REVAL;
1046 result = __link_path_walk(name, nd);
1055 int fastcall path_walk(const char * name, struct nameidata *nd)
1057 current->total_link_count = 0;
1058 return link_path_walk(name, nd);
1062 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1063 * everything is done. Returns 0 and drops input nd, if lookup failed;
1065 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1067 if (path_walk(name, nd))
1068 return 0; /* something went wrong... */
1070 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1071 struct dentry *old_dentry = nd->dentry;
1072 struct vfsmount *old_mnt = nd->mnt;
1073 struct qstr last = nd->last;
1074 int last_type = nd->last_type;
1076 * NAME was not found in alternate root or it's a directory. Try to find
1077 * it in the normal root:
1079 nd->last_type = LAST_ROOT;
1080 read_lock(¤t->fs->lock);
1081 nd->mnt = mntget(current->fs->rootmnt);
1082 nd->dentry = dget(current->fs->root);
1083 read_unlock(¤t->fs->lock);
1084 if (path_walk(name, nd) == 0) {
1085 if (nd->dentry->d_inode) {
1092 nd->dentry = old_dentry;
1095 nd->last_type = last_type;
1100 void set_fs_altroot(void)
1102 char *emul = __emul_prefix();
1103 struct nameidata nd;
1104 struct vfsmount *mnt = NULL, *oldmnt;
1105 struct dentry *dentry = NULL, *olddentry;
1110 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1116 write_lock(¤t->fs->lock);
1117 oldmnt = current->fs->altrootmnt;
1118 olddentry = current->fs->altroot;
1119 current->fs->altrootmnt = mnt;
1120 current->fs->altroot = dentry;
1121 write_unlock(¤t->fs->lock);
1128 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1129 static int fastcall do_path_lookup(int dfd, const char *name,
1130 unsigned int flags, struct nameidata *nd)
1136 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1141 read_lock(¤t->fs->lock);
1142 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1143 nd->mnt = mntget(current->fs->altrootmnt);
1144 nd->dentry = dget(current->fs->altroot);
1145 read_unlock(¤t->fs->lock);
1146 if (__emul_lookup_dentry(name,nd))
1147 goto out; /* found in altroot */
1148 read_lock(¤t->fs->lock);
1150 nd->mnt = mntget(current->fs->rootmnt);
1151 nd->dentry = dget(current->fs->root);
1152 read_unlock(¤t->fs->lock);
1153 } else if (dfd == AT_FDCWD) {
1154 read_lock(¤t->fs->lock);
1155 nd->mnt = mntget(current->fs->pwdmnt);
1156 nd->dentry = dget(current->fs->pwd);
1157 read_unlock(¤t->fs->lock);
1159 struct dentry *dentry;
1161 file = fget_light(dfd, &fput_needed);
1166 dentry = file->f_dentry;
1169 if (!S_ISDIR(dentry->d_inode->i_mode))
1172 retval = file_permission(file, MAY_EXEC);
1176 nd->mnt = mntget(file->f_vfsmnt);
1177 nd->dentry = dget(dentry);
1179 fput_light(file, fput_needed);
1181 current->total_link_count = 0;
1182 retval = link_path_walk(name, nd);
1184 if (likely(retval == 0)) {
1185 if (unlikely(current->audit_context && nd && nd->dentry &&
1186 nd->dentry->d_inode))
1187 audit_inode(name, nd->dentry->d_inode, flags);
1193 fput_light(file, fput_needed);
1197 int fastcall path_lookup(const char *name, unsigned int flags,
1198 struct nameidata *nd)
1200 return do_path_lookup(AT_FDCWD, name, flags, nd);
1203 static int __path_lookup_intent_open(int dfd, const char *name,
1204 unsigned int lookup_flags, struct nameidata *nd,
1205 int open_flags, int create_mode)
1207 struct file *filp = get_empty_filp();
1212 nd->intent.open.file = filp;
1213 nd->intent.open.flags = open_flags;
1214 nd->intent.open.create_mode = create_mode;
1215 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1216 if (IS_ERR(nd->intent.open.file)) {
1218 err = PTR_ERR(nd->intent.open.file);
1221 } else if (err != 0)
1222 release_open_intent(nd);
1227 * path_lookup_open - lookup a file path with open intent
1228 * @dfd: the directory to use as base, or AT_FDCWD
1229 * @name: pointer to file name
1230 * @lookup_flags: lookup intent flags
1231 * @nd: pointer to nameidata
1232 * @open_flags: open intent flags
1234 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1235 struct nameidata *nd, int open_flags)
1237 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1242 * path_lookup_create - lookup a file path with open + create intent
1243 * @dfd: the directory to use as base, or AT_FDCWD
1244 * @name: pointer to file name
1245 * @lookup_flags: lookup intent flags
1246 * @nd: pointer to nameidata
1247 * @open_flags: open intent flags
1248 * @create_mode: create intent flags
1250 static int path_lookup_create(int dfd, const char *name,
1251 unsigned int lookup_flags, struct nameidata *nd,
1252 int open_flags, int create_mode)
1254 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1255 nd, open_flags, create_mode);
1258 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1259 struct nameidata *nd, int open_flags)
1261 char *tmp = getname(name);
1262 int err = PTR_ERR(tmp);
1265 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
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 qstr *name, struct dentry * base, struct nameidata *nd)
1278 struct dentry * dentry;
1279 struct inode *inode;
1282 inode = base->d_inode;
1283 err = permission(inode, MAY_EXEC, nd);
1284 dentry = ERR_PTR(err);
1289 * See if the low-level filesystem might want
1290 * to use its own hash..
1292 if (base->d_op && base->d_op->d_hash) {
1293 err = base->d_op->d_hash(base, name);
1294 dentry = ERR_PTR(err);
1299 dentry = cached_lookup(base, name, nd);
1301 struct dentry *new = d_alloc(base, name);
1302 dentry = ERR_PTR(-ENOMEM);
1305 dentry = inode->i_op->lookup(inode, new, nd);
1315 static struct dentry *lookup_hash(struct nameidata *nd)
1317 return __lookup_hash(&nd->last, nd->dentry, nd);
1321 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1332 hash = init_name_hash();
1334 c = *(const unsigned char *)name++;
1335 if (c == '/' || c == '\0')
1337 hash = partial_name_hash(c, hash);
1339 this.hash = end_name_hash(hash);
1341 return __lookup_hash(&this, base, NULL);
1343 return ERR_PTR(-EACCES);
1349 * is used by most simple commands to get the inode of a specified name.
1350 * Open, link etc use their own routines, but this is enough for things
1353 * namei exists in two versions: namei/lnamei. The only difference is
1354 * that namei follows links, while lnamei does not.
1357 int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1358 struct nameidata *nd)
1360 char *tmp = getname(name);
1361 int err = PTR_ERR(tmp);
1364 err = do_path_lookup(dfd, tmp, flags, nd);
1370 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1372 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1376 * It's inline, so penalty for filesystems that don't use sticky bit is
1379 static inline int check_sticky(struct inode *dir, struct inode *inode)
1381 if (!(dir->i_mode & S_ISVTX))
1383 if (inode->i_uid == current->fsuid)
1385 if (dir->i_uid == current->fsuid)
1387 return !capable(CAP_FOWNER);
1391 * Check whether we can remove a link victim from directory dir, check
1392 * whether the type of victim is right.
1393 * 1. We can't do it if dir is read-only (done in permission())
1394 * 2. We should have write and exec permissions on dir
1395 * 3. We can't remove anything from append-only dir
1396 * 4. We can't do anything with immutable dir (done in permission())
1397 * 5. If the sticky bit on dir is set we should either
1398 * a. be owner of dir, or
1399 * b. be owner of victim, or
1400 * c. have CAP_FOWNER capability
1401 * 6. If the victim is append-only or immutable we can't do antyhing with
1402 * links pointing to it.
1403 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1404 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1405 * 9. We can't remove a root or mountpoint.
1406 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1407 * nfs_async_unlink().
1409 static int may_delete(struct inode *dir, struct dentry *victim,
1410 int isdir, struct nameidata *nd)
1414 if (!victim->d_inode)
1417 BUG_ON(victim->d_parent->d_inode != dir);
1418 audit_inode_child(victim->d_name.name, victim->d_inode, dir->i_ino);
1420 error = permission(dir,MAY_WRITE | MAY_EXEC, nd);
1425 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1426 IS_IXORUNLINK(victim->d_inode))
1429 if (!S_ISDIR(victim->d_inode->i_mode))
1431 if (IS_ROOT(victim))
1433 } else if (S_ISDIR(victim->d_inode->i_mode))
1435 if (IS_DEADDIR(dir))
1437 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1442 /* Check whether we can create an object with dentry child in directory
1444 * 1. We can't do it if child already exists (open has special treatment for
1445 * this case, but since we are inlined it's OK)
1446 * 2. We can't do it if dir is read-only (done in permission())
1447 * 3. We should have write and exec permissions on dir
1448 * 4. We can't do it if dir is immutable (done in permission())
1450 static inline int may_create(struct inode *dir, struct dentry *child,
1451 struct nameidata *nd)
1455 if (IS_DEADDIR(dir))
1457 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1461 * O_DIRECTORY translates into forcing a directory lookup.
1463 static inline int lookup_flags(unsigned int f)
1465 unsigned long retval = LOOKUP_FOLLOW;
1468 retval &= ~LOOKUP_FOLLOW;
1470 if (f & O_DIRECTORY)
1471 retval |= LOOKUP_DIRECTORY;
1472 if (f & O_ATOMICLOOKUP)
1473 retval |= LOOKUP_ATOMIC;
1479 * p1 and p2 should be directories on the same fs.
1481 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1486 mutex_lock(&p1->d_inode->i_mutex);
1490 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1492 for (p = p1; p->d_parent != p; p = p->d_parent) {
1493 if (p->d_parent == p2) {
1494 mutex_lock(&p2->d_inode->i_mutex);
1495 mutex_lock(&p1->d_inode->i_mutex);
1500 for (p = p2; p->d_parent != p; p = p->d_parent) {
1501 if (p->d_parent == p1) {
1502 mutex_lock(&p1->d_inode->i_mutex);
1503 mutex_lock(&p2->d_inode->i_mutex);
1508 mutex_lock(&p1->d_inode->i_mutex);
1509 mutex_lock(&p2->d_inode->i_mutex);
1513 void unlock_rename(struct dentry *p1, struct dentry *p2)
1515 mutex_unlock(&p1->d_inode->i_mutex);
1517 mutex_unlock(&p2->d_inode->i_mutex);
1518 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1522 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1523 struct nameidata *nd)
1525 int error = may_create(dir, dentry, nd);
1530 if (!dir->i_op || !dir->i_op->create)
1531 return -EACCES; /* shouldn't it be ENOSYS? */
1534 error = security_inode_create(dir, dentry, mode);
1538 error = dir->i_op->create(dir, dentry, mode, nd);
1540 fsnotify_create(dir, dentry);
1544 int may_open(struct nameidata *nd, int acc_mode, int flag)
1546 struct dentry *dentry = nd->dentry;
1547 struct inode *inode = dentry->d_inode;
1553 if (S_ISLNK(inode->i_mode))
1556 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1559 error = vfs_permission(nd, acc_mode);
1564 * FIFO's, sockets and device files are special: they don't
1565 * actually live on the filesystem itself, and as such you
1566 * can write to them even if the filesystem is read-only.
1568 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1570 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1571 if (nd->mnt->mnt_flags & MNT_NODEV)
1575 } else if ((IS_RDONLY(inode) || MNT_IS_RDONLY(nd->mnt))
1576 && (flag & FMODE_WRITE))
1579 * An append-only file must be opened in append mode for writing.
1581 if (IS_APPEND(inode)) {
1582 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1588 /* O_NOATIME can only be set by the owner or superuser */
1589 if (flag & O_NOATIME)
1590 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1594 * Ensure there are no outstanding leases on the file.
1596 error = break_lease(inode, flag);
1600 if (flag & O_TRUNC) {
1601 error = get_write_access(inode);
1606 * Refuse to truncate files with mandatory locks held on them.
1608 error = locks_verify_locked(inode);
1612 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
1614 put_write_access(inode);
1618 if (flag & FMODE_WRITE)
1627 * namei for open - this is in fact almost the whole open-routine.
1629 * Note that the low bits of "flag" aren't the same as in the open
1630 * system call - they are 00 - no permissions needed
1631 * 01 - read permission needed
1632 * 10 - write permission needed
1633 * 11 - read/write permissions needed
1634 * which is a lot more logical, and also allows the "no perm" needed
1635 * for symlinks (where the permissions are checked later).
1638 int open_namei(int dfd, const char *pathname, int flag,
1639 int mode, struct nameidata *nd)
1641 int acc_mode, error;
1646 acc_mode = ACC_MODE(flag);
1648 /* O_TRUNC implies we need access checks for write permissions */
1650 acc_mode |= MAY_WRITE;
1652 /* Allow the LSM permission hook to distinguish append
1653 access from general write access. */
1654 if (flag & O_APPEND)
1655 acc_mode |= MAY_APPEND;
1658 * The simplest case - just a plain lookup.
1660 if (!(flag & O_CREAT)) {
1661 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1669 * Create - we need to know the parent.
1671 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
1676 * We have the parent and last component. First of all, check
1677 * that we are not asked to creat(2) an obvious directory - that
1681 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1685 nd->flags &= ~LOOKUP_PARENT;
1686 mutex_lock(&dir->d_inode->i_mutex);
1687 path.dentry = lookup_hash(nd);
1691 error = PTR_ERR(path.dentry);
1692 if (IS_ERR(path.dentry)) {
1693 mutex_unlock(&dir->d_inode->i_mutex);
1697 if (IS_ERR(nd->intent.open.file)) {
1698 mutex_unlock(&dir->d_inode->i_mutex);
1699 error = PTR_ERR(nd->intent.open.file);
1703 /* Negative dentry, just create the file */
1704 if (!path.dentry->d_inode) {
1705 if (!IS_POSIXACL(dir->d_inode))
1706 mode &= ~current->fs->umask;
1707 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1708 mutex_unlock(&dir->d_inode->i_mutex);
1710 nd->dentry = path.dentry;
1713 /* Don't check for write permission, don't truncate */
1720 * It already exists.
1722 mutex_unlock(&dir->d_inode->i_mutex);
1728 if (__follow_mount(&path)) {
1730 if (flag & O_NOFOLLOW)
1734 if (!path.dentry->d_inode)
1736 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1739 path_to_nameidata(&path, nd);
1741 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1744 error = may_open(nd, acc_mode, flag);
1750 dput_path(&path, nd);
1752 if (!IS_ERR(nd->intent.open.file))
1753 release_open_intent(nd);
1759 if (flag & O_NOFOLLOW)
1762 * This is subtle. Instead of calling do_follow_link() we do the
1763 * thing by hands. The reason is that this way we have zero link_count
1764 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1765 * After that we have the parent and last component, i.e.
1766 * we are in the same situation as after the first path_walk().
1767 * Well, almost - if the last component is normal we get its copy
1768 * stored in nd->last.name and we will have to putname() it when we
1769 * are done. Procfs-like symlinks just set LAST_BIND.
1771 nd->flags |= LOOKUP_PARENT;
1772 error = security_inode_follow_link(path.dentry, nd);
1775 error = __do_follow_link(&path, nd);
1778 nd->flags &= ~LOOKUP_PARENT;
1779 if (nd->last_type == LAST_BIND)
1782 if (nd->last_type != LAST_NORM)
1784 if (nd->last.name[nd->last.len]) {
1785 __putname(nd->last.name);
1790 __putname(nd->last.name);
1794 mutex_lock(&dir->d_inode->i_mutex);
1795 path.dentry = lookup_hash(nd);
1797 __putname(nd->last.name);
1802 * lookup_create - lookup a dentry, creating it if it doesn't exist
1803 * @nd: nameidata info
1804 * @is_dir: directory flag
1806 * Simple function to lookup and return a dentry and create it
1807 * if it doesn't exist. Is SMP-safe.
1809 * Returns with nd->dentry->d_inode->i_mutex locked.
1811 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1813 struct dentry *dentry = ERR_PTR(-EEXIST);
1815 mutex_lock(&nd->dentry->d_inode->i_mutex);
1817 * Yucky last component or no last component at all?
1818 * (foo/., foo/.., /////)
1820 if (nd->last_type != LAST_NORM)
1822 nd->flags &= ~LOOKUP_PARENT;
1825 * Do the final lookup.
1827 dentry = lookup_hash(nd);
1832 * Special case - lookup gave negative, but... we had foo/bar/
1833 * From the vfs_mknod() POV we just have a negative dentry -
1834 * all is fine. Let's be bastards - you had / on the end, you've
1835 * been asking for (non-existent) directory. -ENOENT for you.
1837 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1842 dentry = ERR_PTR(-ENOENT);
1846 EXPORT_SYMBOL_GPL(lookup_create);
1848 int vfs_mknod(struct inode *dir, struct dentry *dentry,
1849 int mode, dev_t dev, struct nameidata *nd)
1851 int error = may_create(dir, dentry, nd);
1856 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1859 if (!dir->i_op || !dir->i_op->mknod)
1862 error = security_inode_mknod(dir, dentry, mode, dev);
1867 error = dir->i_op->mknod(dir, dentry, mode, dev);
1869 fsnotify_create(dir, dentry);
1873 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1878 struct dentry * dentry;
1879 struct nameidata nd;
1883 tmp = getname(filename);
1885 return PTR_ERR(tmp);
1887 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1890 dentry = lookup_create(&nd, 0);
1891 error = PTR_ERR(dentry);
1893 if (!IS_POSIXACL(nd.dentry->d_inode))
1894 mode &= ~current->fs->umask;
1895 if (!IS_ERR(dentry)) {
1896 switch (mode & S_IFMT) {
1897 case 0: case S_IFREG:
1898 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1900 case S_IFCHR: case S_IFBLK:
1901 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1902 new_decode_dev(dev), &nd);
1904 case S_IFIFO: case S_IFSOCK:
1905 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1916 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1924 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1926 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1929 int vfs_mkdir(struct inode *dir, struct dentry *dentry,
1930 int mode, struct nameidata *nd)
1932 int error = may_create(dir, dentry, nd);
1937 if (!dir->i_op || !dir->i_op->mkdir)
1940 mode &= (S_IRWXUGO|S_ISVTX);
1941 error = security_inode_mkdir(dir, dentry, mode);
1946 error = dir->i_op->mkdir(dir, dentry, mode);
1948 fsnotify_mkdir(dir, dentry);
1952 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
1957 tmp = getname(pathname);
1958 error = PTR_ERR(tmp);
1960 struct dentry *dentry;
1961 struct nameidata nd;
1963 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1966 dentry = lookup_create(&nd, 1);
1967 error = PTR_ERR(dentry);
1968 if (!IS_ERR(dentry)) {
1969 if (!IS_POSIXACL(nd.dentry->d_inode))
1970 mode &= ~current->fs->umask;
1971 error = vfs_mkdir(nd.dentry->d_inode, dentry,
1975 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1984 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
1986 return sys_mkdirat(AT_FDCWD, pathname, mode);
1990 * We try to drop the dentry early: we should have
1991 * a usage count of 2 if we're the only user of this
1992 * dentry, and if that is true (possibly after pruning
1993 * the dcache), then we drop the dentry now.
1995 * A low-level filesystem can, if it choses, legally
1998 * if (!d_unhashed(dentry))
2001 * if it cannot handle the case of removing a directory
2002 * that is still in use by something else..
2004 void dentry_unhash(struct dentry *dentry)
2007 if (atomic_read(&dentry->d_count))
2008 shrink_dcache_parent(dentry);
2009 spin_lock(&dcache_lock);
2010 spin_lock(&dentry->d_lock);
2011 if (atomic_read(&dentry->d_count) == 2)
2013 spin_unlock(&dentry->d_lock);
2014 spin_unlock(&dcache_lock);
2017 int vfs_rmdir(struct inode *dir, struct dentry *dentry,
2018 struct nameidata *nd)
2020 int error = may_delete(dir, dentry, 1, nd);
2025 if (!dir->i_op || !dir->i_op->rmdir)
2030 mutex_lock(&dentry->d_inode->i_mutex);
2031 dentry_unhash(dentry);
2032 if (d_mountpoint(dentry))
2035 error = security_inode_rmdir(dir, dentry);
2037 error = dir->i_op->rmdir(dir, dentry);
2039 dentry->d_inode->i_flags |= S_DEAD;
2042 mutex_unlock(&dentry->d_inode->i_mutex);
2051 static long do_rmdir(int dfd, const char __user *pathname)
2055 struct dentry *dentry;
2056 struct nameidata nd;
2058 name = getname(pathname);
2060 return PTR_ERR(name);
2062 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2066 switch(nd.last_type) {
2077 mutex_lock(&nd.dentry->d_inode->i_mutex);
2078 dentry = lookup_hash(&nd);
2079 error = PTR_ERR(dentry);
2080 if (!IS_ERR(dentry)) {
2081 error = vfs_rmdir(nd.dentry->d_inode, dentry, &nd);
2084 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2092 asmlinkage long sys_rmdir(const char __user *pathname)
2094 return do_rmdir(AT_FDCWD, pathname);
2097 int vfs_unlink(struct inode *dir, struct dentry *dentry,
2098 struct nameidata *nd)
2100 int error = may_delete(dir, dentry, 0, nd);
2105 if (!dir->i_op || !dir->i_op->unlink)
2110 mutex_lock(&dentry->d_inode->i_mutex);
2111 if (d_mountpoint(dentry))
2114 error = security_inode_unlink(dir, dentry);
2116 error = dir->i_op->unlink(dir, dentry);
2118 mutex_unlock(&dentry->d_inode->i_mutex);
2120 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2121 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2129 * Make sure that the actual truncation of the file will occur outside its
2130 * directory's i_mutex. Truncate can take a long time if there is a lot of
2131 * writeout happening, and we don't want to prevent access to the directory
2132 * while waiting on the I/O.
2134 static long do_unlinkat(int dfd, const char __user *pathname)
2138 struct dentry *dentry;
2139 struct nameidata nd;
2140 struct inode *inode = NULL;
2142 name = getname(pathname);
2144 return PTR_ERR(name);
2146 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2150 if (nd.last_type != LAST_NORM)
2152 mutex_lock(&nd.dentry->d_inode->i_mutex);
2153 dentry = lookup_hash(&nd);
2154 error = PTR_ERR(dentry);
2155 if (!IS_ERR(dentry)) {
2156 /* Why not before? Because we want correct error value */
2157 if (nd.last.name[nd.last.len])
2159 inode = dentry->d_inode;
2161 atomic_inc(&inode->i_count);
2162 error = vfs_unlink(nd.dentry->d_inode, dentry, &nd);
2166 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2168 iput(inode); /* truncate the inode here */
2176 error = !dentry->d_inode ? -ENOENT :
2177 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2181 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2183 if ((flag & ~AT_REMOVEDIR) != 0)
2186 if (flag & AT_REMOVEDIR)
2187 return do_rmdir(dfd, pathname);
2189 return do_unlinkat(dfd, pathname);
2192 asmlinkage long sys_unlink(const char __user *pathname)
2194 return do_unlinkat(AT_FDCWD, pathname);
2197 int vfs_symlink(struct inode *dir, struct dentry *dentry,
2198 const char *oldname, int mode, struct nameidata *nd)
2200 int error = may_create(dir, dentry, nd);
2205 if (!dir->i_op || !dir->i_op->symlink)
2208 error = security_inode_symlink(dir, dentry, oldname);
2213 error = dir->i_op->symlink(dir, dentry, oldname);
2215 fsnotify_create(dir, dentry);
2219 asmlinkage long sys_symlinkat(const char __user *oldname,
2220 int newdfd, const char __user *newname)
2226 from = getname(oldname);
2228 return PTR_ERR(from);
2229 to = getname(newname);
2230 error = PTR_ERR(to);
2232 struct dentry *dentry;
2233 struct nameidata nd;
2235 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2238 dentry = lookup_create(&nd, 0);
2239 error = PTR_ERR(dentry);
2240 if (!IS_ERR(dentry)) {
2241 error = vfs_symlink(nd.dentry->d_inode, dentry,
2242 from, S_IALLUGO, &nd);
2245 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2254 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2256 return sys_symlinkat(oldname, AT_FDCWD, newname);
2259 int vfs_link(struct dentry *old_dentry, struct inode *dir,
2260 struct dentry *new_dentry, struct nameidata *nd)
2262 struct inode *inode = old_dentry->d_inode;
2268 error = may_create(dir, new_dentry, nd);
2272 if (dir->i_sb != inode->i_sb)
2276 * A link to an append-only or immutable file cannot be created.
2278 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
2280 if (!dir->i_op || !dir->i_op->link)
2282 if (S_ISDIR(old_dentry->d_inode->i_mode))
2285 error = security_inode_link(old_dentry, dir, new_dentry);
2289 mutex_lock(&old_dentry->d_inode->i_mutex);
2291 error = dir->i_op->link(old_dentry, dir, new_dentry);
2292 mutex_unlock(&old_dentry->d_inode->i_mutex);
2294 fsnotify_create(dir, new_dentry);
2299 * Hardlinks are often used in delicate situations. We avoid
2300 * security-related surprises by not following symlinks on the
2303 * We don't follow them on the oldname either to be compatible
2304 * with linux 2.0, and to avoid hard-linking to directories
2305 * and other special files. --ADM
2307 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2308 int newdfd, const char __user *newname,
2311 struct dentry *new_dentry;
2312 struct nameidata nd, old_nd;
2319 to = getname(newname);
2323 error = __user_walk_fd(olddfd, oldname, 0, &old_nd);
2326 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2330 if (old_nd.mnt != nd.mnt)
2332 new_dentry = lookup_create(&nd, 0);
2333 error = PTR_ERR(new_dentry);
2334 if (!IS_ERR(new_dentry)) {
2335 error = vfs_link(old_nd.dentry, nd.dentry->d_inode,
2339 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2343 path_release(&old_nd);
2350 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2352 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2356 * The worst of all namespace operations - renaming directory. "Perverted"
2357 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2359 * a) we can get into loop creation. Check is done in is_subdir().
2360 * b) race potential - two innocent renames can create a loop together.
2361 * That's where 4.4 screws up. Current fix: serialization on
2362 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2364 * c) we have to lock _three_ objects - parents and victim (if it exists).
2365 * And that - after we got ->i_mutex on parents (until then we don't know
2366 * whether the target exists). Solution: try to be smart with locking
2367 * order for inodes. We rely on the fact that tree topology may change
2368 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2369 * move will be locked. Thus we can rank directories by the tree
2370 * (ancestors first) and rank all non-directories after them.
2371 * That works since everybody except rename does "lock parent, lookup,
2372 * lock child" and rename is under ->s_vfs_rename_mutex.
2373 * HOWEVER, it relies on the assumption that any object with ->lookup()
2374 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2375 * we'd better make sure that there's no link(2) for them.
2376 * d) some filesystems don't support opened-but-unlinked directories,
2377 * either because of layout or because they are not ready to deal with
2378 * all cases correctly. The latter will be fixed (taking this sort of
2379 * stuff into VFS), but the former is not going away. Solution: the same
2380 * trick as in rmdir().
2381 * e) conversion from fhandle to dentry may come in the wrong moment - when
2382 * we are removing the target. Solution: we will have to grab ->i_mutex
2383 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2384 * ->i_mutex on parents, which works but leads to some truely excessive
2387 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2388 struct inode *new_dir, struct dentry *new_dentry)
2391 struct inode *target;
2394 * If we are going to change the parent - check write permissions,
2395 * we'll need to flip '..'.
2397 if (new_dir != old_dir) {
2398 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2403 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2407 target = new_dentry->d_inode;
2409 mutex_lock(&target->i_mutex);
2410 dentry_unhash(new_dentry);
2412 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2415 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2418 target->i_flags |= S_DEAD;
2419 mutex_unlock(&target->i_mutex);
2420 if (d_unhashed(new_dentry))
2421 d_rehash(new_dentry);
2425 d_move(old_dentry,new_dentry);
2429 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2430 struct inode *new_dir, struct dentry *new_dentry)
2432 struct inode *target;
2435 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2440 target = new_dentry->d_inode;
2442 mutex_lock(&target->i_mutex);
2443 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2446 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2448 /* The following d_move() should become unconditional */
2449 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2450 d_move(old_dentry, new_dentry);
2453 mutex_unlock(&target->i_mutex);
2458 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2459 struct inode *new_dir, struct dentry *new_dentry)
2462 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2463 const char *old_name;
2465 if (old_dentry->d_inode == new_dentry->d_inode)
2468 error = may_delete(old_dir, old_dentry, is_dir, NULL);
2472 if (!new_dentry->d_inode)
2473 error = may_create(new_dir, new_dentry, NULL);
2475 error = may_delete(new_dir, new_dentry, is_dir, NULL);
2479 if (!old_dir->i_op || !old_dir->i_op->rename)
2482 DQUOT_INIT(old_dir);
2483 DQUOT_INIT(new_dir);
2485 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2488 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2490 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2492 const char *new_name = old_dentry->d_name.name;
2493 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2494 new_dentry->d_inode, old_dentry->d_inode);
2496 fsnotify_oldname_free(old_name);
2501 static int do_rename(int olddfd, const char *oldname,
2502 int newdfd, const char *newname)
2505 struct dentry * old_dir, * new_dir;
2506 struct dentry * old_dentry, *new_dentry;
2507 struct dentry * trap;
2508 struct nameidata oldnd, newnd;
2510 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2514 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2519 if (oldnd.mnt != newnd.mnt)
2522 old_dir = oldnd.dentry;
2524 if (oldnd.last_type != LAST_NORM)
2527 new_dir = newnd.dentry;
2528 if (newnd.last_type != LAST_NORM)
2531 trap = lock_rename(new_dir, old_dir);
2533 old_dentry = lookup_hash(&oldnd);
2534 error = PTR_ERR(old_dentry);
2535 if (IS_ERR(old_dentry))
2537 /* source must exist */
2539 if (!old_dentry->d_inode)
2541 /* unless the source is a directory trailing slashes give -ENOTDIR */
2542 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2544 if (oldnd.last.name[oldnd.last.len])
2546 if (newnd.last.name[newnd.last.len])
2549 /* source should not be ancestor of target */
2551 if (old_dentry == trap)
2554 if (MNT_IS_RDONLY(newnd.mnt))
2556 new_dentry = lookup_hash(&newnd);
2557 error = PTR_ERR(new_dentry);
2558 if (IS_ERR(new_dentry))
2560 /* target should not be an ancestor of source */
2562 if (new_dentry == trap)
2565 error = vfs_rename(old_dir->d_inode, old_dentry,
2566 new_dir->d_inode, new_dentry);
2572 unlock_rename(new_dir, old_dir);
2574 path_release(&newnd);
2576 path_release(&oldnd);
2581 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2582 int newdfd, const char __user *newname)
2588 from = getname(oldname);
2590 return PTR_ERR(from);
2591 to = getname(newname);
2592 error = PTR_ERR(to);
2594 error = do_rename(olddfd, from, newdfd, to);
2601 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2603 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2606 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2610 len = PTR_ERR(link);
2615 if (len > (unsigned) buflen)
2617 if (copy_to_user(buffer, link, len))
2624 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2625 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2626 * using) it for any given inode is up to filesystem.
2628 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2630 struct nameidata nd;
2634 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2635 if (!IS_ERR(cookie)) {
2636 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2637 if (dentry->d_inode->i_op->put_link)
2638 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2639 cookie = ERR_PTR(res);
2641 return PTR_ERR(cookie);
2644 int vfs_follow_link(struct nameidata *nd, const char *link)
2646 return __vfs_follow_link(nd, link);
2649 /* get the link contents into pagecache */
2650 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2653 struct address_space *mapping = dentry->d_inode->i_mapping;
2654 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2658 wait_on_page_locked(page);
2659 if (!PageUptodate(page))
2665 page_cache_release(page);
2666 return ERR_PTR(-EIO);
2672 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2674 struct page *page = NULL;
2675 char *s = page_getlink(dentry, &page);
2676 int res = vfs_readlink(dentry,buffer,buflen,s);
2679 page_cache_release(page);
2684 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2686 struct page *page = NULL;
2687 nd_set_link(nd, page_getlink(dentry, &page));
2691 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2693 struct page *page = cookie;
2697 page_cache_release(page);
2701 int __page_symlink(struct inode *inode, const char *symname, int len,
2704 struct address_space *mapping = inode->i_mapping;
2710 page = find_or_create_page(mapping, 0, gfp_mask);
2713 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2714 if (err == AOP_TRUNCATED_PAGE) {
2715 page_cache_release(page);
2720 kaddr = kmap_atomic(page, KM_USER0);
2721 memcpy(kaddr, symname, len-1);
2722 kunmap_atomic(kaddr, KM_USER0);
2723 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2724 if (err == AOP_TRUNCATED_PAGE) {
2725 page_cache_release(page);
2731 * Notice that we are _not_ going to block here - end of page is
2732 * unmapped, so this will only try to map the rest of page, see
2733 * that it is unmapped (typically even will not look into inode -
2734 * ->i_size will be enough for everything) and zero it out.
2735 * OTOH it's obviously correct and should make the page up-to-date.
2737 if (!PageUptodate(page)) {
2738 err = mapping->a_ops->readpage(NULL, page);
2739 if (err != AOP_TRUNCATED_PAGE)
2740 wait_on_page_locked(page);
2744 page_cache_release(page);
2747 mark_inode_dirty(inode);
2751 page_cache_release(page);
2756 int page_symlink(struct inode *inode, const char *symname, int len)
2758 return __page_symlink(inode, symname, len,
2759 mapping_gfp_mask(inode->i_mapping));
2762 struct inode_operations page_symlink_inode_operations = {
2763 .readlink = generic_readlink,
2764 .follow_link = page_follow_link_light,
2765 .put_link = page_put_link,
2768 EXPORT_SYMBOL(__user_walk);
2769 EXPORT_SYMBOL(__user_walk_fd);
2770 EXPORT_SYMBOL(follow_down);
2771 EXPORT_SYMBOL(follow_up);
2772 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2773 EXPORT_SYMBOL(getname);
2774 EXPORT_SYMBOL(lock_rename);
2775 EXPORT_SYMBOL(lookup_one_len);
2776 EXPORT_SYMBOL(page_follow_link_light);
2777 EXPORT_SYMBOL(page_put_link);
2778 EXPORT_SYMBOL(page_readlink);
2779 EXPORT_SYMBOL(__page_symlink);
2780 EXPORT_SYMBOL(page_symlink);
2781 EXPORT_SYMBOL(page_symlink_inode_operations);
2782 EXPORT_SYMBOL(path_lookup);
2783 EXPORT_SYMBOL(path_release);
2784 EXPORT_SYMBOL(path_walk);
2785 EXPORT_SYMBOL(permission);
2786 EXPORT_SYMBOL(vfs_permission);
2787 EXPORT_SYMBOL(file_permission);
2788 EXPORT_SYMBOL(unlock_rename);
2789 EXPORT_SYMBOL(vfs_create);
2790 EXPORT_SYMBOL(vfs_follow_link);
2791 EXPORT_SYMBOL(vfs_link);
2792 EXPORT_SYMBOL(vfs_mkdir);
2793 EXPORT_SYMBOL(vfs_mknod);
2794 EXPORT_SYMBOL(generic_permission);
2795 EXPORT_SYMBOL(vfs_readlink);
2796 EXPORT_SYMBOL(vfs_rename);
2797 EXPORT_SYMBOL(vfs_rmdir);
2798 EXPORT_SYMBOL(vfs_symlink);
2799 EXPORT_SYMBOL(vfs_unlink);
2800 EXPORT_SYMBOL(dentry_unhash);
2801 EXPORT_SYMBOL(generic_readlink);