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/dnotify.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/proc_fs.h>
32 #include <linux/vserver/inode.h>
33 #include <linux/vserver/debug.h>
35 #include <asm/namei.h>
36 #include <asm/uaccess.h>
38 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
40 /* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
68 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
85 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
88 * [10-Sep-98 Alan Modra] Another symlink change.
91 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
108 * any extra contention...
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static inline int do_getname(const char __user *filename, char *page)
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
130 retval = strncpy_from_user(page, filename, len);
134 return -ENAMETOOLONG;
140 char * getname(const char __user * filename)
144 result = ERR_PTR(-ENOMEM);
147 int retval = do_getname(filename, tmp);
152 result = ERR_PTR(retval);
155 if (unlikely(current->audit_context) && !IS_ERR(result) && result)
156 audit_getname(result);
161 * generic_permission - check for access rights on a Posix-like filesystem
162 * @inode: inode to check access rights for
163 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
164 * @check_acl: optional callback to check for Posix ACLs
166 * Used to check for read/write/execute permissions on a file.
167 * We use "fsuid" for this, letting us set arbitrary permissions
168 * for filesystem access without changing the "normal" uids which
169 * are used for other things..
171 int generic_permission(struct inode *inode, int mask,
172 int (*check_acl)(struct inode *inode, int mask))
174 umode_t mode = inode->i_mode;
176 if (current->fsuid == inode->i_uid)
179 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
180 int error = check_acl(inode, mask);
181 if (error == -EACCES)
182 goto check_capabilities;
183 else if (error != -EAGAIN)
187 if (in_group_p(inode->i_gid))
192 * If the DACs are ok we don't need any capability check.
194 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
199 * Read/write DACs are always overridable.
200 * Executable DACs are overridable if at least one exec bit is set.
202 if (!(mask & MAY_EXEC) ||
203 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
204 if (capable(CAP_DAC_OVERRIDE))
208 * Searching includes executable on directories, else just read.
210 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
211 if (capable(CAP_DAC_READ_SEARCH))
217 static inline int xid_permission(struct inode *inode, int mask, struct nameidata *nd)
219 if (IS_BARRIER(inode) && !vx_check(0, VX_ADMIN)) {
220 vxwprintk(1, "xid=%d did hit the barrier.",
224 if (inode->i_xid == 0)
226 if (vx_check(inode->i_xid, VX_ADMIN|VX_WATCH|VX_IDENT))
229 vxwprintk(1, "xid=%d denied access to %p[#%d,%lu] »%s«.",
230 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
231 vxd_path(nd->dentry, nd->mnt));
235 int permission(struct inode *inode, int mask, struct nameidata *nd)
239 if (mask & MAY_WRITE) {
240 umode_t mode = inode->i_mode;
243 * Nobody gets write access to a read-only fs.
245 if (IS_RDONLY(inode) &&
246 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
250 * Nobody gets write access to an immutable file.
252 if (IS_IMMUTABLE(inode))
257 /* Ordinary permission routines do not understand MAY_APPEND. */
258 submask = mask & ~MAY_APPEND;
259 if ((retval = xid_permission(inode, mask, nd)))
261 if (inode->i_op && inode->i_op->permission)
262 retval = inode->i_op->permission(inode, submask, nd);
264 retval = generic_permission(inode, submask, NULL);
268 return security_inode_permission(inode, mask, nd);
272 * get_write_access() gets write permission for a file.
273 * put_write_access() releases this write permission.
274 * This is used for regular files.
275 * We cannot support write (and maybe mmap read-write shared) accesses and
276 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
277 * can have the following values:
278 * 0: no writers, no VM_DENYWRITE mappings
279 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
280 * > 0: (i_writecount) users are writing to the file.
282 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
283 * except for the cases where we don't hold i_writecount yet. Then we need to
284 * use {get,deny}_write_access() - these functions check the sign and refuse
285 * to do the change if sign is wrong. Exclusion between them is provided by
286 * the inode->i_lock spinlock.
289 int get_write_access(struct inode * inode)
291 spin_lock(&inode->i_lock);
292 if (atomic_read(&inode->i_writecount) < 0) {
293 spin_unlock(&inode->i_lock);
296 atomic_inc(&inode->i_writecount);
297 spin_unlock(&inode->i_lock);
302 int deny_write_access(struct file * file)
304 struct inode *inode = file->f_dentry->d_inode;
306 spin_lock(&inode->i_lock);
307 if (atomic_read(&inode->i_writecount) > 0) {
308 spin_unlock(&inode->i_lock);
311 atomic_dec(&inode->i_writecount);
312 spin_unlock(&inode->i_lock);
317 void path_release(struct nameidata *nd)
324 * umount() mustn't call path_release()/mntput() as that would clear
327 void path_release_on_umount(struct nameidata *nd)
334 * Internal lookup() using the new generic dcache.
337 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
339 struct dentry * dentry = __d_lookup(parent, name);
341 /* lockess __d_lookup may fail due to concurrent d_move()
342 * in some unrelated directory, so try with d_lookup
345 dentry = d_lookup(parent, name);
347 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
348 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
357 * Short-cut version of permission(), for calling by
358 * path_walk(), when dcache lock is held. Combines parts
359 * of permission() and generic_permission(), and tests ONLY for
360 * MAY_EXEC permission.
362 * If appropriate, check DAC only. If not appropriate, or
363 * short-cut DAC fails, then call permission() to do more
364 * complete permission check.
366 static inline int exec_permission_lite(struct inode *inode,
367 struct nameidata *nd)
369 umode_t mode = inode->i_mode;
371 if (inode->i_op && inode->i_op->permission)
374 if (current->fsuid == inode->i_uid)
376 else if (in_group_p(inode->i_gid))
382 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
385 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
388 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
393 return security_inode_permission(inode, MAY_EXEC, nd);
397 * This is called when everything else fails, and we actually have
398 * to go to the low-level filesystem to find out what we should do..
400 * We get the directory semaphore, and after getting that we also
401 * make sure that nobody added the entry to the dcache in the meantime..
404 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
406 struct dentry * result;
407 struct inode *dir = parent->d_inode;
411 * First re-do the cached lookup just in case it was created
412 * while we waited for the directory semaphore..
414 * FIXME! This could use version numbering or similar to
415 * avoid unnecessary cache lookups.
417 * The "dcache_lock" is purely to protect the RCU list walker
418 * from concurrent renames at this point (we mustn't get false
419 * negatives from the RCU list walk here, unlike the optimistic
422 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
424 result = d_lookup(parent, name);
426 struct dentry * dentry = d_alloc(parent, name);
427 result = ERR_PTR(-ENOMEM);
429 result = dir->i_op->lookup(dir, dentry, nd);
440 * Uhhuh! Nasty case: the cache was re-populated while
441 * we waited on the semaphore. Need to revalidate.
444 if (result->d_op && result->d_op->d_revalidate) {
445 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
447 result = ERR_PTR(-ENOENT);
453 static int __emul_lookup_dentry(const char *, struct nameidata *);
457 walk_init_root(const char *name, struct nameidata *nd)
459 read_lock(¤t->fs->lock);
460 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
461 nd->mnt = mntget(current->fs->altrootmnt);
462 nd->dentry = dget(current->fs->altroot);
463 read_unlock(¤t->fs->lock);
464 if (__emul_lookup_dentry(name,nd))
466 read_lock(¤t->fs->lock);
468 nd->mnt = mntget(current->fs->rootmnt);
469 nd->dentry = dget(current->fs->root);
470 read_unlock(¤t->fs->lock);
474 static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
483 if (!walk_init_root(link, nd))
484 /* weird __emul_prefix() stuff did it */
487 res = link_path_walk(link, nd);
489 if (nd->depth || res || nd->last_type!=LAST_NORM)
492 * If it is an iterative symlinks resolution in open_namei() we
493 * have to copy the last component. And all that crap because of
494 * bloody create() on broken symlinks. Furrfu...
497 if (unlikely(!name)) {
501 strcpy(name, nd->last.name);
502 nd->last.name = name;
506 return PTR_ERR(link);
509 static inline int __do_follow_link(struct dentry *dentry, struct nameidata *nd)
513 touch_atime(nd->mnt, dentry);
514 nd_set_link(nd, NULL);
515 error = dentry->d_inode->i_op->follow_link(dentry, nd);
517 char *s = nd_get_link(nd);
519 error = __vfs_follow_link(nd, s);
520 if (dentry->d_inode->i_op->put_link)
521 dentry->d_inode->i_op->put_link(dentry, nd);
528 * This limits recursive symlink follows to 8, while
529 * limiting consecutive symlinks to 40.
531 * Without that kind of total limit, nasty chains of consecutive
532 * symlinks can cause almost arbitrarily long lookups.
534 static inline int do_follow_link(struct dentry *dentry, struct nameidata *nd)
537 if (current->link_count >= MAX_NESTED_LINKS)
539 if (current->total_link_count >= 40)
541 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
543 err = security_inode_follow_link(dentry, nd);
546 current->link_count++;
547 current->total_link_count++;
549 err = __do_follow_link(dentry, nd);
550 current->link_count--;
558 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
560 struct vfsmount *parent;
561 struct dentry *mountpoint;
562 spin_lock(&vfsmount_lock);
563 parent=(*mnt)->mnt_parent;
564 if (parent == *mnt) {
565 spin_unlock(&vfsmount_lock);
569 mountpoint=dget((*mnt)->mnt_mountpoint);
570 spin_unlock(&vfsmount_lock);
572 *dentry = mountpoint;
578 /* no need for dcache_lock, as serialization is taken care in
581 static int follow_mount(struct vfsmount **mnt, struct dentry **dentry)
584 while (d_mountpoint(*dentry)) {
585 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
591 *dentry = dget(mounted->mnt_root);
597 /* no need for dcache_lock, as serialization is taken care in
600 static inline int __follow_down(struct vfsmount **mnt, struct dentry **dentry)
602 struct vfsmount *mounted;
604 mounted = lookup_mnt(*mnt, *dentry);
609 *dentry = dget(mounted->mnt_root);
615 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
617 return __follow_down(mnt,dentry);
620 static inline void follow_dotdot(struct vfsmount **mnt, struct dentry **dentry)
623 struct vfsmount *parent;
624 struct dentry *old = *dentry;
626 read_lock(¤t->fs->lock);
627 if (*dentry == current->fs->root &&
628 *mnt == current->fs->rootmnt) {
629 read_unlock(¤t->fs->lock);
632 read_unlock(¤t->fs->lock);
633 spin_lock(&dcache_lock);
634 if (*dentry != (*mnt)->mnt_root) {
635 *dentry = dget((*dentry)->d_parent);
636 spin_unlock(&dcache_lock);
640 spin_unlock(&dcache_lock);
641 spin_lock(&vfsmount_lock);
642 parent = (*mnt)->mnt_parent;
643 if (parent == *mnt) {
644 spin_unlock(&vfsmount_lock);
648 *dentry = dget((*mnt)->mnt_mountpoint);
649 spin_unlock(&vfsmount_lock);
654 follow_mount(mnt, dentry);
658 struct vfsmount *mnt;
659 struct dentry *dentry;
663 * It's more convoluted than I'd like it to be, but... it's still fairly
664 * small and for now I'd prefer to have fast path as straight as possible.
665 * It _is_ time-critical.
667 static int do_lookup(struct nameidata *nd, struct qstr *name,
670 struct vfsmount *mnt = nd->mnt;
671 struct dentry *dentry = __d_lookup(nd->dentry, name);
676 if (dentry->d_op && dentry->d_op->d_revalidate)
677 goto need_revalidate;
678 inode = dentry->d_inode;
681 if (!vx_check(inode->i_xid, VX_WATCH|VX_ADMIN|VX_HOSTID|VX_IDENT))
683 if (inode->i_sb->s_magic == PROC_SUPER_MAGIC) {
684 struct proc_dir_entry *de = PDE(inode);
686 if (de && !vx_hide_check(0, de->vx_flags))
691 path->dentry = dentry;
694 vxwprintk(1, "xid=%d did lookup hidden %p[#%d,%lu] »%s«.",
695 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
696 vxd_path(dentry, mnt));
701 dentry = real_lookup(nd->dentry, name, nd);
707 if (dentry->d_op->d_revalidate(dentry, nd))
709 if (d_invalidate(dentry))
715 return PTR_ERR(dentry);
721 * This is the basic name resolution function, turning a pathname
722 * into the final dentry.
724 * We expect 'base' to be positive and a directory.
726 int fastcall link_path_walk(const char * name, struct nameidata *nd)
731 unsigned int lookup_flags = nd->flags;
738 inode = nd->dentry->d_inode;
740 lookup_flags = LOOKUP_FOLLOW;
742 /* At this point we know we have a real path component. */
748 err = exec_permission_lite(inode, nd);
749 if (err == -EAGAIN) {
750 err = permission(inode, MAY_EXEC, nd);
756 c = *(const unsigned char *)name;
758 hash = init_name_hash();
761 hash = partial_name_hash(c, hash);
762 c = *(const unsigned char *)name;
763 } while (c && (c != '/'));
764 this.len = name - (const char *) this.name;
765 this.hash = end_name_hash(hash);
767 /* remove trailing slashes? */
770 while (*++name == '/');
772 goto last_with_slashes;
775 * "." and ".." are special - ".." especially so because it has
776 * to be able to know about the current root directory and
777 * parent relationships.
779 if (this.name[0] == '.') switch (this.len) {
783 if (this.name[1] != '.')
785 follow_dotdot(&nd->mnt, &nd->dentry);
786 inode = nd->dentry->d_inode;
792 * See if the low-level filesystem might want
793 * to use its own hash..
795 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
796 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
800 nd->flags |= LOOKUP_CONTINUE;
801 /* This does the actual lookups.. */
802 err = do_lookup(nd, &this, &next);
805 /* Check mountpoints.. */
806 follow_mount(&next.mnt, &next.dentry);
809 inode = next.dentry->d_inode;
816 if (inode->i_op->follow_link) {
818 err = do_follow_link(next.dentry, nd);
824 inode = nd->dentry->d_inode;
833 nd->dentry = next.dentry;
836 if (!inode->i_op->lookup)
839 /* here ends the main loop */
842 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
844 nd->flags &= ~LOOKUP_CONTINUE;
845 if (lookup_flags & LOOKUP_PARENT)
847 if (this.name[0] == '.') switch (this.len) {
851 if (this.name[1] != '.')
853 follow_dotdot(&nd->mnt, &nd->dentry);
854 inode = nd->dentry->d_inode;
859 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
860 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
864 err = do_lookup(nd, &this, &next);
867 follow_mount(&next.mnt, &next.dentry);
868 inode = next.dentry->d_inode;
869 if ((lookup_flags & LOOKUP_FOLLOW)
870 && inode && inode->i_op && inode->i_op->follow_link) {
872 err = do_follow_link(next.dentry, nd);
877 inode = nd->dentry->d_inode;
881 nd->dentry = next.dentry;
886 if (lookup_flags & LOOKUP_DIRECTORY) {
888 if (!inode->i_op || !inode->i_op->lookup)
894 nd->last_type = LAST_NORM;
895 if (this.name[0] != '.')
898 nd->last_type = LAST_DOT;
899 else if (this.len == 2 && this.name[1] == '.')
900 nd->last_type = LAST_DOTDOT;
905 * We bypassed the ordinary revalidation routines.
906 * We may need to check the cached dentry for staleness.
908 if (nd->dentry && nd->dentry->d_sb &&
909 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
911 /* Note: we do not d_invalidate() */
912 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
926 int fastcall path_walk(const char * name, struct nameidata *nd)
928 current->total_link_count = 0;
929 return link_path_walk(name, nd);
933 /* returns 1 if everything is done */
934 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
936 if (path_walk(name, nd))
937 return 0; /* something went wrong... */
939 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
940 struct dentry *old_dentry = nd->dentry;
941 struct vfsmount *old_mnt = nd->mnt;
942 struct qstr last = nd->last;
943 int last_type = nd->last_type;
945 * NAME was not found in alternate root or it's a directory. Try to find
946 * it in the normal root:
948 nd->last_type = LAST_ROOT;
949 read_lock(¤t->fs->lock);
950 nd->mnt = mntget(current->fs->rootmnt);
951 nd->dentry = dget(current->fs->root);
952 read_unlock(¤t->fs->lock);
953 if (path_walk(name, nd) == 0) {
954 if (nd->dentry->d_inode) {
961 nd->dentry = old_dentry;
964 nd->last_type = last_type;
969 void set_fs_altroot(void)
971 char *emul = __emul_prefix();
973 struct vfsmount *mnt = NULL, *oldmnt;
974 struct dentry *dentry = NULL, *olddentry;
979 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
985 write_lock(¤t->fs->lock);
986 oldmnt = current->fs->altrootmnt;
987 olddentry = current->fs->altroot;
988 current->fs->altrootmnt = mnt;
989 current->fs->altroot = dentry;
990 write_unlock(¤t->fs->lock);
997 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
1001 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1005 read_lock(¤t->fs->lock);
1007 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1008 nd->mnt = mntget(current->fs->altrootmnt);
1009 nd->dentry = dget(current->fs->altroot);
1010 read_unlock(¤t->fs->lock);
1011 if (__emul_lookup_dentry(name,nd))
1013 read_lock(¤t->fs->lock);
1015 nd->mnt = mntget(current->fs->rootmnt);
1016 nd->dentry = dget(current->fs->root);
1018 nd->mnt = mntget(current->fs->pwdmnt);
1019 nd->dentry = dget(current->fs->pwd);
1021 read_unlock(¤t->fs->lock);
1022 current->total_link_count = 0;
1023 retval = link_path_walk(name, nd);
1024 if (unlikely(current->audit_context
1025 && nd && nd->dentry && nd->dentry->d_inode))
1027 nd->dentry->d_inode->i_ino,
1028 nd->dentry->d_inode->i_rdev);
1033 * Restricted form of lookup. Doesn't follow links, single-component only,
1034 * needs parent already locked. Doesn't follow mounts.
1037 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1039 struct dentry * dentry;
1040 struct inode *inode;
1043 inode = base->d_inode;
1044 err = permission(inode, MAY_EXEC, nd);
1045 dentry = ERR_PTR(err);
1050 * See if the low-level filesystem might want
1051 * to use its own hash..
1053 if (base->d_op && base->d_op->d_hash) {
1054 err = base->d_op->d_hash(base, name);
1055 dentry = ERR_PTR(err);
1060 dentry = cached_lookup(base, name, nd);
1062 struct dentry *new = d_alloc(base, name);
1063 dentry = ERR_PTR(-ENOMEM);
1066 dentry = inode->i_op->lookup(inode, new, nd);
1076 struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
1078 return __lookup_hash(name, base, NULL);
1082 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1093 hash = init_name_hash();
1095 c = *(const unsigned char *)name++;
1096 if (c == '/' || c == '\0')
1098 hash = partial_name_hash(c, hash);
1100 this.hash = end_name_hash(hash);
1102 return lookup_hash(&this, base);
1104 return ERR_PTR(-EACCES);
1110 * is used by most simple commands to get the inode of a specified name.
1111 * Open, link etc use their own routines, but this is enough for things
1114 * namei exists in two versions: namei/lnamei. The only difference is
1115 * that namei follows links, while lnamei does not.
1118 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1120 char *tmp = getname(name);
1121 int err = PTR_ERR(tmp);
1124 err = path_lookup(tmp, flags, nd);
1131 * It's inline, so penalty for filesystems that don't use sticky bit is
1134 static inline int check_sticky(struct inode *dir, struct inode *inode)
1136 if (!(dir->i_mode & S_ISVTX))
1138 if (inode->i_uid == current->fsuid)
1140 if (dir->i_uid == current->fsuid)
1142 return !capable(CAP_FOWNER);
1146 * Check whether we can remove a link victim from directory dir, check
1147 * whether the type of victim is right.
1148 * 1. We can't do it if dir is read-only (done in permission())
1149 * 2. We should have write and exec permissions on dir
1150 * 3. We can't remove anything from append-only dir
1151 * 4. We can't do anything with immutable dir (done in permission())
1152 * 5. If the sticky bit on dir is set we should either
1153 * a. be owner of dir, or
1154 * b. be owner of victim, or
1155 * c. have CAP_FOWNER capability
1156 * 6. If the victim is append-only or immutable we can't do antyhing with
1157 * links pointing to it.
1158 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1159 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1160 * 9. We can't remove a root or mountpoint.
1161 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1162 * nfs_async_unlink().
1164 static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1168 if (!victim->d_inode)
1171 BUG_ON(victim->d_parent->d_inode != dir);
1173 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1178 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1179 IS_IXORUNLINK(victim->d_inode))
1182 if (!S_ISDIR(victim->d_inode->i_mode))
1184 if (IS_ROOT(victim))
1186 } else if (S_ISDIR(victim->d_inode->i_mode))
1188 if (IS_DEADDIR(dir))
1190 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1195 /* Check whether we can create an object with dentry child in directory
1197 * 1. We can't do it if child already exists (open has special treatment for
1198 * this case, but since we are inlined it's OK)
1199 * 2. We can't do it if dir is read-only (done in permission())
1200 * 3. We should have write and exec permissions on dir
1201 * 4. We can't do it if dir is immutable (done in permission())
1203 static inline int may_create(struct inode *dir, struct dentry *child,
1204 struct nameidata *nd)
1208 if (IS_DEADDIR(dir))
1210 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1214 * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1217 * O_DIRECTORY translates into forcing a directory lookup.
1219 static inline int lookup_flags(unsigned int f)
1221 unsigned long retval = LOOKUP_FOLLOW;
1224 retval &= ~LOOKUP_FOLLOW;
1226 if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1227 retval &= ~LOOKUP_FOLLOW;
1229 if (f & O_DIRECTORY)
1230 retval |= LOOKUP_DIRECTORY;
1236 * p1 and p2 should be directories on the same fs.
1238 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1243 down(&p1->d_inode->i_sem);
1247 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1249 for (p = p1; p->d_parent != p; p = p->d_parent) {
1250 if (p->d_parent == p2) {
1251 down(&p2->d_inode->i_sem);
1252 down(&p1->d_inode->i_sem);
1257 for (p = p2; p->d_parent != p; p = p->d_parent) {
1258 if (p->d_parent == p1) {
1259 down(&p1->d_inode->i_sem);
1260 down(&p2->d_inode->i_sem);
1265 down(&p1->d_inode->i_sem);
1266 down(&p2->d_inode->i_sem);
1270 void unlock_rename(struct dentry *p1, struct dentry *p2)
1272 up(&p1->d_inode->i_sem);
1274 up(&p2->d_inode->i_sem);
1275 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1279 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1280 struct nameidata *nd)
1282 int error = may_create(dir, dentry, nd);
1287 if (!dir->i_op || !dir->i_op->create)
1288 return -EACCES; /* shouldn't it be ENOSYS? */
1291 error = security_inode_create(dir, dentry, mode);
1295 error = dir->i_op->create(dir, dentry, mode, nd);
1297 inode_dir_notify(dir, DN_CREATE);
1298 security_inode_post_create(dir, dentry, mode);
1303 int may_open(struct nameidata *nd, int acc_mode, int flag)
1305 struct dentry *dentry = nd->dentry;
1306 struct inode *inode = dentry->d_inode;
1312 if (S_ISLNK(inode->i_mode))
1315 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1318 error = permission(inode, acc_mode, nd);
1323 * FIFO's, sockets and device files are special: they don't
1324 * actually live on the filesystem itself, and as such you
1325 * can write to them even if the filesystem is read-only.
1327 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1329 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1330 if (nd->mnt->mnt_flags & MNT_NODEV)
1334 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1337 * An append-only file must be opened in append mode for writing.
1339 if (IS_APPEND(inode)) {
1340 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1346 /* O_NOATIME can only be set by the owner or superuser */
1347 if (flag & O_NOATIME)
1348 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1352 * Ensure there are no outstanding leases on the file.
1354 error = break_lease(inode, flag);
1358 if (flag & O_TRUNC) {
1359 error = get_write_access(inode);
1364 * Refuse to truncate files with mandatory locks held on them.
1366 error = locks_verify_locked(inode);
1370 error = do_truncate(dentry, 0);
1372 put_write_access(inode);
1376 if (flag & FMODE_WRITE)
1385 * namei for open - this is in fact almost the whole open-routine.
1387 * Note that the low bits of "flag" aren't the same as in the open
1388 * system call - they are 00 - no permissions needed
1389 * 01 - read permission needed
1390 * 10 - write permission needed
1391 * 11 - read/write permissions needed
1392 * which is a lot more logical, and also allows the "no perm" needed
1393 * for symlinks (where the permissions are checked later).
1396 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1398 int acc_mode, error = 0;
1399 struct dentry *dentry;
1403 acc_mode = ACC_MODE(flag);
1405 /* Allow the LSM permission hook to distinguish append
1406 access from general write access. */
1407 if (flag & O_APPEND)
1408 acc_mode |= MAY_APPEND;
1410 /* Fill in the open() intent data */
1411 nd->intent.open.flags = flag;
1412 nd->intent.open.create_mode = mode;
1415 * The simplest case - just a plain lookup.
1417 if (!(flag & O_CREAT)) {
1418 error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd);
1425 * Create - we need to know the parent.
1427 error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd);
1432 * We have the parent and last component. First of all, check
1433 * that we are not asked to creat(2) an obvious directory - that
1437 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1441 nd->flags &= ~LOOKUP_PARENT;
1442 down(&dir->d_inode->i_sem);
1443 dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1446 error = PTR_ERR(dentry);
1447 if (IS_ERR(dentry)) {
1448 up(&dir->d_inode->i_sem);
1452 /* Negative dentry, just create the file */
1453 if (!dentry->d_inode) {
1454 if (!IS_POSIXACL(dir->d_inode))
1455 mode &= ~current->fs->umask;
1456 error = vfs_create(dir->d_inode, dentry, mode, nd);
1457 up(&dir->d_inode->i_sem);
1459 nd->dentry = dentry;
1462 /* Don't check for write permission, don't truncate */
1469 * It already exists.
1471 up(&dir->d_inode->i_sem);
1477 if (d_mountpoint(dentry)) {
1479 if (flag & O_NOFOLLOW)
1481 while (__follow_down(&nd->mnt,&dentry) && d_mountpoint(dentry));
1484 if (!dentry->d_inode)
1486 if (dentry->d_inode->i_op && dentry->d_inode->i_op->follow_link)
1490 nd->dentry = dentry;
1492 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode))
1495 error = may_open(nd, acc_mode, flag);
1508 if (flag & O_NOFOLLOW)
1511 * This is subtle. Instead of calling do_follow_link() we do the
1512 * thing by hands. The reason is that this way we have zero link_count
1513 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1514 * After that we have the parent and last component, i.e.
1515 * we are in the same situation as after the first path_walk().
1516 * Well, almost - if the last component is normal we get its copy
1517 * stored in nd->last.name and we will have to putname() it when we
1518 * are done. Procfs-like symlinks just set LAST_BIND.
1520 nd->flags |= LOOKUP_PARENT;
1521 error = security_inode_follow_link(dentry, nd);
1524 error = __do_follow_link(dentry, nd);
1528 nd->flags &= ~LOOKUP_PARENT;
1529 if (nd->last_type == LAST_BIND) {
1530 dentry = nd->dentry;
1534 if (nd->last_type != LAST_NORM)
1536 if (nd->last.name[nd->last.len]) {
1537 putname(nd->last.name);
1542 putname(nd->last.name);
1546 down(&dir->d_inode->i_sem);
1547 dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1548 putname(nd->last.name);
1553 * lookup_create - lookup a dentry, creating it if it doesn't exist
1554 * @nd: nameidata info
1555 * @is_dir: directory flag
1557 * Simple function to lookup and return a dentry and create it
1558 * if it doesn't exist. Is SMP-safe.
1560 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1562 struct dentry *dentry;
1564 down(&nd->dentry->d_inode->i_sem);
1565 dentry = ERR_PTR(-EEXIST);
1566 if (nd->last_type != LAST_NORM)
1568 nd->flags &= ~LOOKUP_PARENT;
1569 dentry = lookup_hash(&nd->last, nd->dentry);
1572 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1577 dentry = ERR_PTR(-ENOENT);
1582 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1584 int error = may_create(dir, dentry, NULL);
1589 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1592 if (!dir->i_op || !dir->i_op->mknod)
1595 error = security_inode_mknod(dir, dentry, mode, dev);
1600 error = dir->i_op->mknod(dir, dentry, mode, dev);
1602 inode_dir_notify(dir, DN_CREATE);
1603 security_inode_post_mknod(dir, dentry, mode, dev);
1608 asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1612 struct dentry * dentry;
1613 struct nameidata nd;
1617 tmp = getname(filename);
1619 return PTR_ERR(tmp);
1621 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1624 dentry = lookup_create(&nd, 0);
1625 error = PTR_ERR(dentry);
1627 if (!IS_POSIXACL(nd.dentry->d_inode))
1628 mode &= ~current->fs->umask;
1629 if (!IS_ERR(dentry)) {
1630 switch (mode & S_IFMT) {
1631 case 0: case S_IFREG:
1632 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1634 case S_IFCHR: case S_IFBLK:
1635 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1636 new_decode_dev(dev));
1638 case S_IFIFO: case S_IFSOCK:
1639 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1649 up(&nd.dentry->d_inode->i_sem);
1657 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1659 int error = may_create(dir, dentry, NULL);
1664 if (!dir->i_op || !dir->i_op->mkdir)
1667 mode &= (S_IRWXUGO|S_ISVTX);
1668 error = security_inode_mkdir(dir, dentry, mode);
1673 error = dir->i_op->mkdir(dir, dentry, mode);
1675 inode_dir_notify(dir, DN_CREATE);
1676 security_inode_post_mkdir(dir,dentry, mode);
1681 asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1686 tmp = getname(pathname);
1687 error = PTR_ERR(tmp);
1689 struct dentry *dentry;
1690 struct nameidata nd;
1692 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1695 dentry = lookup_create(&nd, 1);
1696 error = PTR_ERR(dentry);
1697 if (!IS_ERR(dentry)) {
1698 if (!IS_POSIXACL(nd.dentry->d_inode))
1699 mode &= ~current->fs->umask;
1700 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1703 up(&nd.dentry->d_inode->i_sem);
1713 * We try to drop the dentry early: we should have
1714 * a usage count of 2 if we're the only user of this
1715 * dentry, and if that is true (possibly after pruning
1716 * the dcache), then we drop the dentry now.
1718 * A low-level filesystem can, if it choses, legally
1721 * if (!d_unhashed(dentry))
1724 * if it cannot handle the case of removing a directory
1725 * that is still in use by something else..
1727 void dentry_unhash(struct dentry *dentry)
1730 spin_lock(&dcache_lock);
1731 switch (atomic_read(&dentry->d_count)) {
1733 spin_unlock(&dcache_lock);
1734 shrink_dcache_parent(dentry);
1735 spin_lock(&dcache_lock);
1736 if (atomic_read(&dentry->d_count) != 2)
1741 spin_unlock(&dcache_lock);
1744 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1746 int error = may_delete(dir, dentry, 1);
1751 if (!dir->i_op || !dir->i_op->rmdir)
1756 down(&dentry->d_inode->i_sem);
1757 dentry_unhash(dentry);
1758 if (d_mountpoint(dentry))
1761 error = security_inode_rmdir(dir, dentry);
1763 error = dir->i_op->rmdir(dir, dentry);
1765 dentry->d_inode->i_flags |= S_DEAD;
1768 up(&dentry->d_inode->i_sem);
1770 inode_dir_notify(dir, DN_DELETE);
1778 asmlinkage long sys_rmdir(const char __user * pathname)
1782 struct dentry *dentry;
1783 struct nameidata nd;
1785 name = getname(pathname);
1787 return PTR_ERR(name);
1789 error = path_lookup(name, LOOKUP_PARENT, &nd);
1793 switch(nd.last_type) {
1804 down(&nd.dentry->d_inode->i_sem);
1805 dentry = lookup_hash(&nd.last, nd.dentry);
1806 error = PTR_ERR(dentry);
1807 if (!IS_ERR(dentry)) {
1808 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1811 up(&nd.dentry->d_inode->i_sem);
1819 int vfs_unlink(struct inode *dir, struct dentry *dentry)
1821 int error = may_delete(dir, dentry, 0);
1826 if (!dir->i_op || !dir->i_op->unlink)
1831 down(&dentry->d_inode->i_sem);
1832 if (d_mountpoint(dentry))
1835 error = security_inode_unlink(dir, dentry);
1837 error = dir->i_op->unlink(dir, dentry);
1839 up(&dentry->d_inode->i_sem);
1841 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1842 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
1844 inode_dir_notify(dir, DN_DELETE);
1850 * Make sure that the actual truncation of the file will occur outside its
1851 * directory's i_sem. Truncate can take a long time if there is a lot of
1852 * writeout happening, and we don't want to prevent access to the directory
1853 * while waiting on the I/O.
1855 asmlinkage long sys_unlink(const char __user * pathname)
1859 struct dentry *dentry;
1860 struct nameidata nd;
1861 struct inode *inode = NULL;
1863 name = getname(pathname);
1865 return PTR_ERR(name);
1867 error = path_lookup(name, LOOKUP_PARENT, &nd);
1871 if (nd.last_type != LAST_NORM)
1873 down(&nd.dentry->d_inode->i_sem);
1874 dentry = lookup_hash(&nd.last, nd.dentry);
1875 error = PTR_ERR(dentry);
1876 if (!IS_ERR(dentry)) {
1877 /* Why not before? Because we want correct error value */
1878 if (nd.last.name[nd.last.len])
1880 inode = dentry->d_inode;
1882 atomic_inc(&inode->i_count);
1883 error = vfs_unlink(nd.dentry->d_inode, dentry);
1887 up(&nd.dentry->d_inode->i_sem);
1889 iput(inode); /* truncate the inode here */
1897 error = !dentry->d_inode ? -ENOENT :
1898 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1902 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
1904 int error = may_create(dir, dentry, NULL);
1909 if (!dir->i_op || !dir->i_op->symlink)
1912 error = security_inode_symlink(dir, dentry, oldname);
1917 error = dir->i_op->symlink(dir, dentry, oldname);
1919 inode_dir_notify(dir, DN_CREATE);
1920 security_inode_post_symlink(dir, dentry, oldname);
1925 asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
1931 from = getname(oldname);
1933 return PTR_ERR(from);
1934 to = getname(newname);
1935 error = PTR_ERR(to);
1937 struct dentry *dentry;
1938 struct nameidata nd;
1940 error = path_lookup(to, LOOKUP_PARENT, &nd);
1943 dentry = lookup_create(&nd, 0);
1944 error = PTR_ERR(dentry);
1945 if (!IS_ERR(dentry)) {
1946 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
1949 up(&nd.dentry->d_inode->i_sem);
1958 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
1960 struct inode *inode = old_dentry->d_inode;
1966 error = may_create(dir, new_dentry, NULL);
1970 if (dir->i_sb != inode->i_sb)
1974 * A link to an append-only or immutable file cannot be created.
1976 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
1978 if (!dir->i_op || !dir->i_op->link)
1980 if (S_ISDIR(old_dentry->d_inode->i_mode))
1983 error = security_inode_link(old_dentry, dir, new_dentry);
1987 down(&old_dentry->d_inode->i_sem);
1989 error = dir->i_op->link(old_dentry, dir, new_dentry);
1990 up(&old_dentry->d_inode->i_sem);
1992 inode_dir_notify(dir, DN_CREATE);
1993 security_inode_post_link(old_dentry, dir, new_dentry);
1999 * Hardlinks are often used in delicate situations. We avoid
2000 * security-related surprises by not following symlinks on the
2003 * We don't follow them on the oldname either to be compatible
2004 * with linux 2.0, and to avoid hard-linking to directories
2005 * and other special files. --ADM
2007 asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
2009 struct dentry *new_dentry;
2010 struct nameidata nd, old_nd;
2014 to = getname(newname);
2018 error = __user_walk(oldname, 0, &old_nd);
2021 error = path_lookup(to, LOOKUP_PARENT, &nd);
2025 if (old_nd.mnt != nd.mnt)
2027 new_dentry = lookup_create(&nd, 0);
2028 error = PTR_ERR(new_dentry);
2029 if (!IS_ERR(new_dentry)) {
2030 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2033 up(&nd.dentry->d_inode->i_sem);
2037 path_release(&old_nd);
2045 * The worst of all namespace operations - renaming directory. "Perverted"
2046 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2048 * a) we can get into loop creation. Check is done in is_subdir().
2049 * b) race potential - two innocent renames can create a loop together.
2050 * That's where 4.4 screws up. Current fix: serialization on
2051 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2053 * c) we have to lock _three_ objects - parents and victim (if it exists).
2054 * And that - after we got ->i_sem on parents (until then we don't know
2055 * whether the target exists). Solution: try to be smart with locking
2056 * order for inodes. We rely on the fact that tree topology may change
2057 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2058 * move will be locked. Thus we can rank directories by the tree
2059 * (ancestors first) and rank all non-directories after them.
2060 * That works since everybody except rename does "lock parent, lookup,
2061 * lock child" and rename is under ->s_vfs_rename_sem.
2062 * HOWEVER, it relies on the assumption that any object with ->lookup()
2063 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2064 * we'd better make sure that there's no link(2) for them.
2065 * d) some filesystems don't support opened-but-unlinked directories,
2066 * either because of layout or because they are not ready to deal with
2067 * all cases correctly. The latter will be fixed (taking this sort of
2068 * stuff into VFS), but the former is not going away. Solution: the same
2069 * trick as in rmdir().
2070 * e) conversion from fhandle to dentry may come in the wrong moment - when
2071 * we are removing the target. Solution: we will have to grab ->i_sem
2072 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2073 * ->i_sem on parents, which works but leads to some truely excessive
2076 int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2077 struct inode *new_dir, struct dentry *new_dentry)
2080 struct inode *target;
2083 * If we are going to change the parent - check write permissions,
2084 * we'll need to flip '..'.
2086 if (new_dir != old_dir) {
2087 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2092 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2096 target = new_dentry->d_inode;
2098 down(&target->i_sem);
2099 dentry_unhash(new_dentry);
2101 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2104 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2107 target->i_flags |= S_DEAD;
2109 if (d_unhashed(new_dentry))
2110 d_rehash(new_dentry);
2114 d_move(old_dentry,new_dentry);
2115 security_inode_post_rename(old_dir, old_dentry,
2116 new_dir, new_dentry);
2121 int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2122 struct inode *new_dir, struct dentry *new_dentry)
2124 struct inode *target;
2127 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2132 target = new_dentry->d_inode;
2134 down(&target->i_sem);
2135 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2138 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2140 /* The following d_move() should become unconditional */
2141 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2142 d_move(old_dentry, new_dentry);
2143 security_inode_post_rename(old_dir, old_dentry, new_dir, new_dentry);
2151 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2152 struct inode *new_dir, struct dentry *new_dentry)
2155 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2157 if (old_dentry->d_inode == new_dentry->d_inode)
2160 error = may_delete(old_dir, old_dentry, is_dir);
2164 if (!new_dentry->d_inode)
2165 error = may_create(new_dir, new_dentry, NULL);
2167 error = may_delete(new_dir, new_dentry, is_dir);
2171 if (!old_dir->i_op || !old_dir->i_op->rename)
2174 DQUOT_INIT(old_dir);
2175 DQUOT_INIT(new_dir);
2178 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2180 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2182 if (old_dir == new_dir)
2183 inode_dir_notify(old_dir, DN_RENAME);
2185 inode_dir_notify(old_dir, DN_DELETE);
2186 inode_dir_notify(new_dir, DN_CREATE);
2192 static inline int do_rename(const char * oldname, const char * newname)
2195 struct dentry * old_dir, * new_dir;
2196 struct dentry * old_dentry, *new_dentry;
2197 struct dentry * trap;
2198 struct nameidata oldnd, newnd;
2200 error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2204 error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2209 if (oldnd.mnt != newnd.mnt)
2212 old_dir = oldnd.dentry;
2214 if (oldnd.last_type != LAST_NORM)
2217 new_dir = newnd.dentry;
2218 if (newnd.last_type != LAST_NORM)
2221 trap = lock_rename(new_dir, old_dir);
2223 old_dentry = lookup_hash(&oldnd.last, old_dir);
2224 error = PTR_ERR(old_dentry);
2225 if (IS_ERR(old_dentry))
2227 /* source must exist */
2229 if (!old_dentry->d_inode)
2231 /* unless the source is a directory trailing slashes give -ENOTDIR */
2232 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2234 if (oldnd.last.name[oldnd.last.len])
2236 if (newnd.last.name[newnd.last.len])
2239 /* source should not be ancestor of target */
2241 if (old_dentry == trap)
2243 new_dentry = lookup_hash(&newnd.last, new_dir);
2244 error = PTR_ERR(new_dentry);
2245 if (IS_ERR(new_dentry))
2247 /* target should not be an ancestor of source */
2249 if (new_dentry == trap)
2252 error = vfs_rename(old_dir->d_inode, old_dentry,
2253 new_dir->d_inode, new_dentry);
2259 unlock_rename(new_dir, old_dir);
2261 path_release(&newnd);
2263 path_release(&oldnd);
2268 asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2274 from = getname(oldname);
2276 return PTR_ERR(from);
2277 to = getname(newname);
2278 error = PTR_ERR(to);
2280 error = do_rename(from,to);
2287 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2291 len = PTR_ERR(link);
2296 if (len > (unsigned) buflen)
2298 if (copy_to_user(buffer, link, len))
2305 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2306 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2307 * using) it for any given inode is up to filesystem.
2309 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2311 struct nameidata nd;
2314 res = dentry->d_inode->i_op->follow_link(dentry, &nd);
2316 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2317 if (dentry->d_inode->i_op->put_link)
2318 dentry->d_inode->i_op->put_link(dentry, &nd);
2323 int vfs_follow_link(struct nameidata *nd, const char *link)
2325 return __vfs_follow_link(nd, link);
2328 /* get the link contents into pagecache */
2329 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2332 struct address_space *mapping = dentry->d_inode->i_mapping;
2333 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2337 wait_on_page_locked(page);
2338 if (!PageUptodate(page))
2344 page_cache_release(page);
2345 return ERR_PTR(-EIO);
2351 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2353 struct page *page = NULL;
2354 char *s = page_getlink(dentry, &page);
2355 int res = vfs_readlink(dentry,buffer,buflen,s);
2358 page_cache_release(page);
2363 int page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2366 nd_set_link(nd, page_getlink(dentry, &page));
2370 void page_put_link(struct dentry *dentry, struct nameidata *nd)
2372 if (!IS_ERR(nd_get_link(nd))) {
2374 page = find_get_page(dentry->d_inode->i_mapping, 0);
2378 page_cache_release(page);
2379 page_cache_release(page);
2383 int page_symlink(struct inode *inode, const char *symname, int len)
2385 struct address_space *mapping = inode->i_mapping;
2386 struct page *page = grab_cache_page(mapping, 0);
2392 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2395 kaddr = kmap_atomic(page, KM_USER0);
2396 memcpy(kaddr, symname, len-1);
2397 kunmap_atomic(kaddr, KM_USER0);
2398 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2400 * Notice that we are _not_ going to block here - end of page is
2401 * unmapped, so this will only try to map the rest of page, see
2402 * that it is unmapped (typically even will not look into inode -
2403 * ->i_size will be enough for everything) and zero it out.
2404 * OTOH it's obviously correct and should make the page up-to-date.
2406 if (!PageUptodate(page)) {
2407 err = mapping->a_ops->readpage(NULL, page);
2408 wait_on_page_locked(page);
2412 page_cache_release(page);
2415 mark_inode_dirty(inode);
2419 page_cache_release(page);
2424 struct inode_operations page_symlink_inode_operations = {
2425 .readlink = generic_readlink,
2426 .follow_link = page_follow_link_light,
2427 .put_link = page_put_link,
2430 EXPORT_SYMBOL(__user_walk);
2431 EXPORT_SYMBOL(follow_down);
2432 EXPORT_SYMBOL(follow_up);
2433 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2434 EXPORT_SYMBOL(getname);
2435 EXPORT_SYMBOL(lock_rename);
2436 EXPORT_SYMBOL(lookup_hash);
2437 EXPORT_SYMBOL(lookup_one_len);
2438 EXPORT_SYMBOL(page_follow_link_light);
2439 EXPORT_SYMBOL(page_put_link);
2440 EXPORT_SYMBOL(page_readlink);
2441 EXPORT_SYMBOL(page_symlink);
2442 EXPORT_SYMBOL(page_symlink_inode_operations);
2443 EXPORT_SYMBOL(path_lookup);
2444 EXPORT_SYMBOL(path_release);
2445 EXPORT_SYMBOL(path_walk);
2446 EXPORT_SYMBOL(permission);
2447 EXPORT_SYMBOL(unlock_rename);
2448 EXPORT_SYMBOL(vfs_create);
2449 EXPORT_SYMBOL(vfs_follow_link);
2450 EXPORT_SYMBOL(vfs_link);
2451 EXPORT_SYMBOL(vfs_mkdir);
2452 EXPORT_SYMBOL(vfs_mknod);
2453 EXPORT_SYMBOL(generic_permission);
2454 EXPORT_SYMBOL(vfs_readlink);
2455 EXPORT_SYMBOL(vfs_rename);
2456 EXPORT_SYMBOL(vfs_rmdir);
2457 EXPORT_SYMBOL(vfs_symlink);
2458 EXPORT_SYMBOL(vfs_unlink);
2459 EXPORT_SYMBOL(dentry_unhash);
2460 EXPORT_SYMBOL(generic_readlink);