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>
34 #include <asm/namei.h>
35 #include <asm/uaccess.h>
37 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static inline int do_getname(const char __user *filename, char *page)
120 unsigned long len = PATH_MAX;
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
129 retval = strncpy_from_user(page, filename, len);
133 return -ENAMETOOLONG;
139 char * getname(const char __user * filename)
143 result = ERR_PTR(-ENOMEM);
146 int retval = do_getname(filename, tmp);
151 result = ERR_PTR(retval);
154 audit_getname(result);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name)
161 if (unlikely(current->audit_context))
166 EXPORT_SYMBOL(putname);
171 * generic_permission - check for access rights on a Posix-like filesystem
172 * @inode: inode to check access rights for
173 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
174 * @check_acl: optional callback to check for Posix ACLs
176 * Used to check for read/write/execute permissions on a file.
177 * We use "fsuid" for this, letting us set arbitrary permissions
178 * for filesystem access without changing the "normal" uids which
179 * are used for other things..
181 int generic_permission(struct inode *inode, int mask,
182 int (*check_acl)(struct inode *inode, int mask))
184 umode_t mode = inode->i_mode;
186 if (current->fsuid == inode->i_uid)
189 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
190 int error = check_acl(inode, mask);
191 if (error == -EACCES)
192 goto check_capabilities;
193 else if (error != -EAGAIN)
197 if (in_group_p(inode->i_gid))
202 * If the DACs are ok we don't need any capability check.
204 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
209 * Read/write DACs are always overridable.
210 * Executable DACs are overridable if at least one exec bit is set.
212 if (!(mask & MAY_EXEC) ||
213 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
214 if (capable(CAP_DAC_OVERRIDE))
218 * Searching includes executable on directories, else just read.
220 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
221 if (capable(CAP_DAC_READ_SEARCH))
227 static inline int xid_permission(struct inode *inode, int mask, struct nameidata *nd)
229 if (IS_BARRIER(inode) && !vx_check(0, VX_ADMIN)) {
230 vxwprintk(1, "xid=%d did hit the barrier.",
234 if (inode->i_xid == 0)
236 #ifdef CONFIG_VSERVER_FILESHARING
237 /* MEF: PlanetLab FS module assumes that any file that can be
238 * named (e.g., via a cross mount) is not hidden from another
239 * context or the admin context.
241 if (vx_check(inode->i_xid,VX_STATIC|VX_DYNAMIC))
244 if (vx_check(inode->i_xid, VX_ADMIN|VX_WATCH|VX_IDENT))
247 vxwprintk(1, "xid=%d denied access to %p[#%d,%lu] »%s«.",
248 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
249 vxd_path(nd->dentry, nd->mnt));
253 int permission(struct inode *inode, int mask, struct nameidata *nd)
257 if (mask & MAY_WRITE) {
258 umode_t mode = inode->i_mode;
261 * Nobody gets write access to a read-only fs.
263 if ((IS_RDONLY(inode) || (nd && MNT_IS_RDONLY(nd->mnt))) &&
264 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
268 * Nobody gets write access to an immutable file.
270 if (IS_IMMUTABLE(inode))
275 /* Ordinary permission routines do not understand MAY_APPEND. */
276 submask = mask & ~MAY_APPEND;
277 if ((retval = xid_permission(inode, mask, nd)))
279 if (inode->i_op && inode->i_op->permission)
280 retval = inode->i_op->permission(inode, submask, nd);
282 retval = generic_permission(inode, submask, NULL);
286 return security_inode_permission(inode, mask, nd);
290 * get_write_access() gets write permission for a file.
291 * put_write_access() releases this write permission.
292 * This is used for regular files.
293 * We cannot support write (and maybe mmap read-write shared) accesses and
294 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
295 * can have the following values:
296 * 0: no writers, no VM_DENYWRITE mappings
297 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
298 * > 0: (i_writecount) users are writing to the file.
300 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
301 * except for the cases where we don't hold i_writecount yet. Then we need to
302 * use {get,deny}_write_access() - these functions check the sign and refuse
303 * to do the change if sign is wrong. Exclusion between them is provided by
304 * the inode->i_lock spinlock.
307 int get_write_access(struct inode * inode)
309 spin_lock(&inode->i_lock);
310 if (atomic_read(&inode->i_writecount) < 0) {
311 spin_unlock(&inode->i_lock);
314 atomic_inc(&inode->i_writecount);
315 spin_unlock(&inode->i_lock);
320 int deny_write_access(struct file * file)
322 struct inode *inode = file->f_dentry->d_inode;
324 spin_lock(&inode->i_lock);
325 if (atomic_read(&inode->i_writecount) > 0) {
326 spin_unlock(&inode->i_lock);
329 atomic_dec(&inode->i_writecount);
330 spin_unlock(&inode->i_lock);
335 void path_release(struct nameidata *nd)
342 * umount() mustn't call path_release()/mntput() as that would clear
345 void path_release_on_umount(struct nameidata *nd)
352 * Internal lookup() using the new generic dcache.
355 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
357 struct dentry * dentry = __d_lookup(parent, name);
359 /* lockess __d_lookup may fail due to concurrent d_move()
360 * in some unrelated directory, so try with d_lookup
363 dentry = d_lookup(parent, name);
365 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
366 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
375 * Short-cut version of permission(), for calling by
376 * path_walk(), when dcache lock is held. Combines parts
377 * of permission() and generic_permission(), and tests ONLY for
378 * MAY_EXEC permission.
380 * If appropriate, check DAC only. If not appropriate, or
381 * short-cut DAC fails, then call permission() to do more
382 * complete permission check.
384 static inline int exec_permission_lite(struct inode *inode,
385 struct nameidata *nd)
387 umode_t mode = inode->i_mode;
389 if (inode->i_op && inode->i_op->permission)
392 if (current->fsuid == inode->i_uid)
394 else if (in_group_p(inode->i_gid))
400 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
403 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
406 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
411 return security_inode_permission(inode, MAY_EXEC, nd);
415 * This is called when everything else fails, and we actually have
416 * to go to the low-level filesystem to find out what we should do..
418 * We get the directory semaphore, and after getting that we also
419 * make sure that nobody added the entry to the dcache in the meantime..
422 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
424 struct dentry * result;
425 struct inode *dir = parent->d_inode;
429 * First re-do the cached lookup just in case it was created
430 * while we waited for the directory semaphore..
432 * FIXME! This could use version numbering or similar to
433 * avoid unnecessary cache lookups.
435 * The "dcache_lock" is purely to protect the RCU list walker
436 * from concurrent renames at this point (we mustn't get false
437 * negatives from the RCU list walk here, unlike the optimistic
440 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
442 result = d_lookup(parent, name);
444 struct dentry * dentry = d_alloc(parent, name);
445 result = ERR_PTR(-ENOMEM);
447 result = dir->i_op->lookup(dir, dentry, nd);
458 * Uhhuh! Nasty case: the cache was re-populated while
459 * we waited on the semaphore. Need to revalidate.
462 if (result->d_op && result->d_op->d_revalidate) {
463 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
465 result = ERR_PTR(-ENOENT);
471 static int __emul_lookup_dentry(const char *, struct nameidata *);
475 walk_init_root(const char *name, struct nameidata *nd)
477 read_lock(¤t->fs->lock);
478 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
479 nd->mnt = mntget(current->fs->altrootmnt);
480 nd->dentry = dget(current->fs->altroot);
481 read_unlock(¤t->fs->lock);
482 if (__emul_lookup_dentry(name,nd))
484 read_lock(¤t->fs->lock);
486 nd->mnt = mntget(current->fs->rootmnt);
487 nd->dentry = dget(current->fs->root);
488 read_unlock(¤t->fs->lock);
492 static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
501 if (!walk_init_root(link, nd))
502 /* weird __emul_prefix() stuff did it */
505 res = link_path_walk(link, nd);
507 if (nd->depth || res || nd->last_type!=LAST_NORM)
510 * If it is an iterative symlinks resolution in open_namei() we
511 * have to copy the last component. And all that crap because of
512 * bloody create() on broken symlinks. Furrfu...
515 if (unlikely(!name)) {
519 strcpy(name, nd->last.name);
520 nd->last.name = name;
524 return PTR_ERR(link);
528 struct vfsmount *mnt;
529 struct dentry *dentry;
532 static inline int __do_follow_link(struct path *path, struct nameidata *nd)
535 struct dentry *dentry = path->dentry;
537 touch_atime(path->mnt, dentry);
538 nd_set_link(nd, NULL);
540 if (path->mnt == nd->mnt)
542 error = dentry->d_inode->i_op->follow_link(dentry, nd);
544 char *s = nd_get_link(nd);
546 error = __vfs_follow_link(nd, s);
547 if (dentry->d_inode->i_op->put_link)
548 dentry->d_inode->i_op->put_link(dentry, nd);
557 * This limits recursive symlink follows to 8, while
558 * limiting consecutive symlinks to 40.
560 * Without that kind of total limit, nasty chains of consecutive
561 * symlinks can cause almost arbitrarily long lookups.
563 static inline int do_follow_link(struct path *path, struct nameidata *nd)
566 if (current->link_count >= MAX_NESTED_LINKS)
568 if (current->total_link_count >= 40)
570 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
572 err = security_inode_follow_link(path->dentry, nd);
575 current->link_count++;
576 current->total_link_count++;
578 err = __do_follow_link(path, nd);
579 current->link_count--;
584 if (path->mnt != nd->mnt)
590 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
592 struct vfsmount *parent;
593 struct dentry *mountpoint;
594 spin_lock(&vfsmount_lock);
595 parent=(*mnt)->mnt_parent;
596 if (parent == *mnt) {
597 spin_unlock(&vfsmount_lock);
601 mountpoint=dget((*mnt)->mnt_mountpoint);
602 spin_unlock(&vfsmount_lock);
604 *dentry = mountpoint;
610 /* no need for dcache_lock, as serialization is taken care in
613 static int __follow_mount(struct path *path)
616 while (d_mountpoint(path->dentry)) {
617 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
624 path->dentry = dget(mounted->mnt_root);
630 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
632 while (d_mountpoint(*dentry)) {
633 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
639 *dentry = dget(mounted->mnt_root);
643 /* no need for dcache_lock, as serialization is taken care in
646 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
648 struct vfsmount *mounted;
650 mounted = lookup_mnt(*mnt, *dentry);
655 *dentry = dget(mounted->mnt_root);
661 static inline void follow_dotdot(struct nameidata *nd)
664 struct vfsmount *parent;
665 struct dentry *old = nd->dentry;
667 read_lock(¤t->fs->lock);
668 if (nd->dentry == current->fs->root &&
669 nd->mnt == current->fs->rootmnt) {
670 read_unlock(¤t->fs->lock);
671 /* for sane '/' avoid follow_mount() */
674 read_unlock(¤t->fs->lock);
675 spin_lock(&dcache_lock);
676 if (nd->dentry != nd->mnt->mnt_root) {
677 nd->dentry = dget(nd->dentry->d_parent);
678 spin_unlock(&dcache_lock);
682 spin_unlock(&dcache_lock);
683 spin_lock(&vfsmount_lock);
684 parent = nd->mnt->mnt_parent;
685 if (parent == nd->mnt) {
686 spin_unlock(&vfsmount_lock);
690 nd->dentry = dget(nd->mnt->mnt_mountpoint);
691 spin_unlock(&vfsmount_lock);
696 follow_mount(&nd->mnt, &nd->dentry);
700 * It's more convoluted than I'd like it to be, but... it's still fairly
701 * small and for now I'd prefer to have fast path as straight as possible.
702 * It _is_ time-critical.
704 static int do_lookup(struct nameidata *nd, struct qstr *name,
705 struct path *path, int atomic)
707 struct vfsmount *mnt = nd->mnt;
708 struct dentry *dentry = __d_lookup(nd->dentry, name);
713 if (dentry->d_op && dentry->d_op->d_revalidate)
714 goto need_revalidate;
715 inode = dentry->d_inode;
718 if (inode->i_sb->s_magic == PROC_SUPER_MAGIC) {
719 struct proc_dir_entry *de = PDE(inode);
721 if (de && !vx_hide_check(0, de->vx_flags))
724 #ifdef CONFIG_VSERVER_FILESHARING
725 /* MEF: PlanetLab FS module assumes that any file that can be
726 * named (e.g., via a cross mount) is not hidden from another
727 * context or the admin context.
729 if (vx_check(inode->i_xid,VX_STATIC|VX_DYNAMIC|VX_ADMIN)) {
732 else /* do the following check */
734 if (!vx_check(inode->i_xid, VX_WATCH|VX_ADMIN|VX_HOSTID|VX_IDENT))
738 path->dentry = dentry;
739 __follow_mount(path);
742 vxwprintk(1, "xid=%d did lookup hidden %p[#%d,%lu] »%s«.",
743 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
744 vxd_path(dentry, mnt));
750 return -EWOULDBLOCKIO;
751 dentry = real_lookup(nd->dentry, name, nd);
758 return -EWOULDBLOCKIO;
759 if (dentry->d_op->d_revalidate(dentry, nd))
761 if (d_invalidate(dentry))
767 return PTR_ERR(dentry);
772 * This is the basic name resolution function, turning a pathname into
773 * the final dentry. We expect 'base' to be positive and a directory.
775 * Returns 0 and nd will have valid dentry and mnt on success.
776 * Returns error and drops reference to input namei data on failure.
778 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
783 unsigned int lookup_flags = nd->flags;
785 atomic = (lookup_flags & LOOKUP_ATOMIC);
792 inode = nd->dentry->d_inode;
794 lookup_flags = LOOKUP_FOLLOW;
796 /* At this point we know we have a real path component. */
802 err = exec_permission_lite(inode, nd);
803 if (err == -EAGAIN) {
804 err = permission(inode, MAY_EXEC, nd);
810 c = *(const unsigned char *)name;
812 hash = init_name_hash();
815 hash = partial_name_hash(c, hash);
816 c = *(const unsigned char *)name;
817 } while (c && (c != '/'));
818 this.len = name - (const char *) this.name;
819 this.hash = end_name_hash(hash);
821 /* remove trailing slashes? */
824 while (*++name == '/');
826 goto last_with_slashes;
829 * "." and ".." are special - ".." especially so because it has
830 * to be able to know about the current root directory and
831 * parent relationships.
833 if (this.name[0] == '.') switch (this.len) {
837 if (this.name[1] != '.')
840 inode = nd->dentry->d_inode;
846 * See if the low-level filesystem might want
847 * to use its own hash..
849 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
850 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
854 nd->flags |= LOOKUP_CONTINUE;
855 /* This does the actual lookups.. */
856 err = do_lookup(nd, &this, &next, atomic);
861 inode = next.dentry->d_inode;
868 if (inode->i_op->follow_link) {
869 err = do_follow_link(&next, nd);
873 inode = nd->dentry->d_inode;
881 if (nd->mnt != next.mnt)
884 nd->dentry = next.dentry;
887 if (!inode->i_op->lookup)
890 /* here ends the main loop */
893 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
895 nd->flags &= ~LOOKUP_CONTINUE;
896 if (lookup_flags & LOOKUP_PARENT)
898 if (this.name[0] == '.') switch (this.len) {
902 if (this.name[1] != '.')
905 inode = nd->dentry->d_inode;
910 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
911 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
915 err = do_lookup(nd, &this, &next, atomic);
918 inode = next.dentry->d_inode;
919 if ((lookup_flags & LOOKUP_FOLLOW)
920 && inode && inode->i_op && inode->i_op->follow_link) {
921 err = do_follow_link(&next, nd);
924 inode = nd->dentry->d_inode;
927 if (nd->mnt != next.mnt)
930 nd->dentry = next.dentry;
935 if (lookup_flags & LOOKUP_DIRECTORY) {
937 if (!inode->i_op || !inode->i_op->lookup)
943 nd->last_type = LAST_NORM;
944 if (this.name[0] != '.')
947 nd->last_type = LAST_DOT;
948 else if (this.len == 2 && this.name[1] == '.')
949 nd->last_type = LAST_DOTDOT;
954 * We bypassed the ordinary revalidation routines.
955 * We may need to check the cached dentry for staleness.
957 if (nd->dentry && nd->dentry->d_sb &&
958 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
960 /* Note: we do not d_invalidate() */
961 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
968 if (nd->mnt != next.mnt)
978 * Wrapper to retry pathname resolution whenever the underlying
979 * file system returns an ESTALE.
981 * Retry the whole path once, forcing real lookup requests
982 * instead of relying on the dcache.
984 int fastcall link_path_walk(const char *name, struct nameidata *nd)
986 struct nameidata save = *nd;
989 /* make sure the stuff we saved doesn't go away */
993 result = __link_path_walk(name, nd);
994 if (result == -ESTALE) {
998 nd->flags |= LOOKUP_REVAL;
999 result = __link_path_walk(name, nd);
1008 int fastcall path_walk(const char * name, struct nameidata *nd)
1010 current->total_link_count = 0;
1011 return link_path_walk(name, nd);
1015 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1016 * everything is done. Returns 0 and drops input nd, if lookup failed;
1018 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1020 if (path_walk(name, nd))
1021 return 0; /* something went wrong... */
1023 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1024 struct dentry *old_dentry = nd->dentry;
1025 struct vfsmount *old_mnt = nd->mnt;
1026 struct qstr last = nd->last;
1027 int last_type = nd->last_type;
1029 * NAME was not found in alternate root or it's a directory. Try to find
1030 * it in the normal root:
1032 nd->last_type = LAST_ROOT;
1033 read_lock(¤t->fs->lock);
1034 nd->mnt = mntget(current->fs->rootmnt);
1035 nd->dentry = dget(current->fs->root);
1036 read_unlock(¤t->fs->lock);
1037 if (path_walk(name, nd) == 0) {
1038 if (nd->dentry->d_inode) {
1045 nd->dentry = old_dentry;
1048 nd->last_type = last_type;
1053 void set_fs_altroot(void)
1055 char *emul = __emul_prefix();
1056 struct nameidata nd;
1057 struct vfsmount *mnt = NULL, *oldmnt;
1058 struct dentry *dentry = NULL, *olddentry;
1063 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1069 write_lock(¤t->fs->lock);
1070 oldmnt = current->fs->altrootmnt;
1071 olddentry = current->fs->altroot;
1072 current->fs->altrootmnt = mnt;
1073 current->fs->altroot = dentry;
1074 write_unlock(¤t->fs->lock);
1081 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1082 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
1086 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1090 read_lock(¤t->fs->lock);
1092 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1093 nd->mnt = mntget(current->fs->altrootmnt);
1094 nd->dentry = dget(current->fs->altroot);
1095 read_unlock(¤t->fs->lock);
1096 if (__emul_lookup_dentry(name,nd))
1097 goto out; /* found in altroot */
1098 read_lock(¤t->fs->lock);
1100 nd->mnt = mntget(current->fs->rootmnt);
1101 nd->dentry = dget(current->fs->root);
1103 nd->mnt = mntget(current->fs->pwdmnt);
1104 nd->dentry = dget(current->fs->pwd);
1106 read_unlock(¤t->fs->lock);
1107 current->total_link_count = 0;
1108 retval = link_path_walk(name, nd);
1110 if (unlikely(current->audit_context
1111 && nd && nd->dentry && nd->dentry->d_inode))
1112 audit_inode(name, nd->dentry->d_inode, flags);
1117 * Restricted form of lookup. Doesn't follow links, single-component only,
1118 * needs parent already locked. Doesn't follow mounts.
1121 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1123 struct dentry * dentry;
1124 struct inode *inode;
1127 inode = base->d_inode;
1128 err = permission(inode, MAY_EXEC, nd);
1129 dentry = ERR_PTR(err);
1134 * See if the low-level filesystem might want
1135 * to use its own hash..
1137 if (base->d_op && base->d_op->d_hash) {
1138 err = base->d_op->d_hash(base, name);
1139 dentry = ERR_PTR(err);
1144 dentry = cached_lookup(base, name, nd);
1146 struct dentry *new = d_alloc(base, name);
1147 dentry = ERR_PTR(-ENOMEM);
1150 dentry = inode->i_op->lookup(inode, new, nd);
1160 struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
1162 return __lookup_hash(name, base, NULL);
1166 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1177 hash = init_name_hash();
1179 c = *(const unsigned char *)name++;
1180 if (c == '/' || c == '\0')
1182 hash = partial_name_hash(c, hash);
1184 this.hash = end_name_hash(hash);
1186 return lookup_hash(&this, base);
1188 return ERR_PTR(-EACCES);
1194 * is used by most simple commands to get the inode of a specified name.
1195 * Open, link etc use their own routines, but this is enough for things
1198 * namei exists in two versions: namei/lnamei. The only difference is
1199 * that namei follows links, while lnamei does not.
1202 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1204 char *tmp = getname(name);
1205 int err = PTR_ERR(tmp);
1208 err = path_lookup(tmp, flags, nd);
1215 * It's inline, so penalty for filesystems that don't use sticky bit is
1218 static inline int check_sticky(struct inode *dir, struct inode *inode)
1220 if (!(dir->i_mode & S_ISVTX))
1222 if (inode->i_uid == current->fsuid)
1224 if (dir->i_uid == current->fsuid)
1226 return !capable(CAP_FOWNER);
1230 * Check whether we can remove a link victim from directory dir, check
1231 * whether the type of victim is right.
1232 * 1. We can't do it if dir is read-only (done in permission())
1233 * 2. We should have write and exec permissions on dir
1234 * 3. We can't remove anything from append-only dir
1235 * 4. We can't do anything with immutable dir (done in permission())
1236 * 5. If the sticky bit on dir is set we should either
1237 * a. be owner of dir, or
1238 * b. be owner of victim, or
1239 * c. have CAP_FOWNER capability
1240 * 6. If the victim is append-only or immutable we can't do antyhing with
1241 * links pointing to it.
1242 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1243 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1244 * 9. We can't remove a root or mountpoint.
1245 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1246 * nfs_async_unlink().
1248 static inline int may_delete(struct inode *dir, struct dentry *victim,
1249 int isdir, struct nameidata *nd)
1253 if (!victim->d_inode)
1256 BUG_ON(victim->d_parent->d_inode != dir);
1258 error = permission(dir,MAY_WRITE | MAY_EXEC, nd);
1263 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1264 IS_IXORUNLINK(victim->d_inode))
1267 if (!S_ISDIR(victim->d_inode->i_mode))
1269 if (IS_ROOT(victim))
1271 } else if (S_ISDIR(victim->d_inode->i_mode))
1273 if (IS_DEADDIR(dir))
1275 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1280 /* Check whether we can create an object with dentry child in directory
1282 * 1. We can't do it if child already exists (open has special treatment for
1283 * this case, but since we are inlined it's OK)
1284 * 2. We can't do it if dir is read-only (done in permission())
1285 * 3. We should have write and exec permissions on dir
1286 * 4. We can't do it if dir is immutable (done in permission())
1288 static inline int may_create(struct inode *dir, struct dentry *child,
1289 struct nameidata *nd)
1293 if (IS_DEADDIR(dir))
1295 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1299 * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1302 * O_DIRECTORY translates into forcing a directory lookup.
1304 static inline int lookup_flags(unsigned int f)
1306 unsigned long retval = LOOKUP_FOLLOW;
1309 retval &= ~LOOKUP_FOLLOW;
1311 if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1312 retval &= ~LOOKUP_FOLLOW;
1314 if (f & O_DIRECTORY)
1315 retval |= LOOKUP_DIRECTORY;
1316 if (f & O_ATOMICLOOKUP)
1317 retval |= LOOKUP_ATOMIC;
1323 * p1 and p2 should be directories on the same fs.
1325 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1330 down(&p1->d_inode->i_sem);
1334 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1336 for (p = p1; p->d_parent != p; p = p->d_parent) {
1337 if (p->d_parent == p2) {
1338 down(&p2->d_inode->i_sem);
1339 down(&p1->d_inode->i_sem);
1344 for (p = p2; p->d_parent != p; p = p->d_parent) {
1345 if (p->d_parent == p1) {
1346 down(&p1->d_inode->i_sem);
1347 down(&p2->d_inode->i_sem);
1352 down(&p1->d_inode->i_sem);
1353 down(&p2->d_inode->i_sem);
1357 void unlock_rename(struct dentry *p1, struct dentry *p2)
1359 up(&p1->d_inode->i_sem);
1361 up(&p2->d_inode->i_sem);
1362 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1366 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1367 struct nameidata *nd)
1369 int error = may_create(dir, dentry, nd);
1374 if (!dir->i_op || !dir->i_op->create)
1375 return -EACCES; /* shouldn't it be ENOSYS? */
1378 error = security_inode_create(dir, dentry, mode);
1382 error = dir->i_op->create(dir, dentry, mode, nd);
1384 inode_dir_notify(dir, DN_CREATE);
1385 security_inode_post_create(dir, dentry, mode);
1390 int may_open(struct nameidata *nd, int acc_mode, int flag)
1392 struct dentry *dentry = nd->dentry;
1393 struct inode *inode = dentry->d_inode;
1399 if (S_ISLNK(inode->i_mode))
1402 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1405 error = permission(inode, acc_mode, nd);
1410 * FIFO's, sockets and device files are special: they don't
1411 * actually live on the filesystem itself, and as such you
1412 * can write to them even if the filesystem is read-only.
1414 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1416 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1417 if (nd->mnt->mnt_flags & MNT_NODEV)
1421 } else if ((IS_RDONLY(inode) || MNT_IS_RDONLY(nd->mnt))
1422 && (flag & FMODE_WRITE))
1425 * An append-only file must be opened in append mode for writing.
1427 if (IS_APPEND(inode)) {
1428 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1434 /* O_NOATIME can only be set by the owner or superuser */
1435 if (flag & O_NOATIME)
1436 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1440 * Ensure there are no outstanding leases on the file.
1442 error = break_lease(inode, flag);
1446 if (flag & O_TRUNC) {
1447 error = get_write_access(inode);
1452 * Refuse to truncate files with mandatory locks held on them.
1454 error = locks_verify_locked(inode);
1458 error = do_truncate(dentry, 0);
1460 put_write_access(inode);
1464 if (flag & FMODE_WRITE)
1473 * namei for open - this is in fact almost the whole open-routine.
1475 * Note that the low bits of "flag" aren't the same as in the open
1476 * system call - they are 00 - no permissions needed
1477 * 01 - read permission needed
1478 * 10 - write permission needed
1479 * 11 - read/write permissions needed
1480 * which is a lot more logical, and also allows the "no perm" needed
1481 * for symlinks (where the permissions are checked later).
1484 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1486 int acc_mode, error = 0;
1491 acc_mode = ACC_MODE(flag);
1493 /* Allow the LSM permission hook to distinguish append
1494 access from general write access. */
1495 if (flag & O_APPEND)
1496 acc_mode |= MAY_APPEND;
1498 /* Fill in the open() intent data */
1499 nd->intent.open.flags = flag;
1500 nd->intent.open.create_mode = mode;
1503 * The simplest case - just a plain lookup.
1505 if (!(flag & O_CREAT)) {
1506 error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd);
1513 * Create - we need to know the parent.
1515 error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd);
1520 * We have the parent and last component. First of all, check
1521 * that we are not asked to creat(2) an obvious directory - that
1525 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1529 nd->flags &= ~LOOKUP_PARENT;
1530 down(&dir->d_inode->i_sem);
1531 path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1535 error = PTR_ERR(path.dentry);
1536 if (IS_ERR(path.dentry)) {
1537 up(&dir->d_inode->i_sem);
1541 /* Negative dentry, just create the file */
1542 if (!path.dentry->d_inode) {
1543 if (!IS_POSIXACL(dir->d_inode))
1544 mode &= ~current->fs->umask;
1545 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1546 up(&dir->d_inode->i_sem);
1548 nd->dentry = path.dentry;
1551 /* Don't check for write permission, don't truncate */
1558 * It already exists.
1560 up(&dir->d_inode->i_sem);
1566 if (__follow_mount(&path)) {
1568 if (flag & O_NOFOLLOW)
1572 if (!path.dentry->d_inode)
1574 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1578 nd->dentry = path.dentry;
1579 if (nd->mnt != path.mnt)
1583 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1586 error = may_open(nd, acc_mode, flag);
1593 if (nd->mnt != path.mnt)
1601 if (flag & O_NOFOLLOW)
1604 * This is subtle. Instead of calling do_follow_link() we do the
1605 * thing by hands. The reason is that this way we have zero link_count
1606 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1607 * After that we have the parent and last component, i.e.
1608 * we are in the same situation as after the first path_walk().
1609 * Well, almost - if the last component is normal we get its copy
1610 * stored in nd->last.name and we will have to putname() it when we
1611 * are done. Procfs-like symlinks just set LAST_BIND.
1613 nd->flags |= LOOKUP_PARENT;
1614 error = security_inode_follow_link(path.dentry, nd);
1617 error = __do_follow_link(&path, nd);
1620 nd->flags &= ~LOOKUP_PARENT;
1621 if (nd->last_type == LAST_BIND)
1624 if (nd->last_type != LAST_NORM)
1626 if (nd->last.name[nd->last.len]) {
1627 putname(nd->last.name);
1632 putname(nd->last.name);
1636 down(&dir->d_inode->i_sem);
1637 path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1639 putname(nd->last.name);
1644 * lookup_create - lookup a dentry, creating it if it doesn't exist
1645 * @nd: nameidata info
1646 * @is_dir: directory flag
1648 * Simple function to lookup and return a dentry and create it
1649 * if it doesn't exist. Is SMP-safe.
1651 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1653 struct dentry *dentry;
1655 down(&nd->dentry->d_inode->i_sem);
1656 dentry = ERR_PTR(-EEXIST);
1657 if (nd->last_type != LAST_NORM)
1659 nd->flags &= ~LOOKUP_PARENT;
1660 dentry = lookup_hash(&nd->last, nd->dentry);
1663 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1668 dentry = ERR_PTR(-ENOENT);
1672 EXPORT_SYMBOL_GPL(lookup_create);
1674 int vfs_mknod(struct inode *dir, struct dentry *dentry,
1675 int mode, dev_t dev, struct nameidata *nd)
1677 int error = may_create(dir, dentry, nd);
1682 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1685 if (!dir->i_op || !dir->i_op->mknod)
1688 error = security_inode_mknod(dir, dentry, mode, dev);
1693 error = dir->i_op->mknod(dir, dentry, mode, dev);
1695 inode_dir_notify(dir, DN_CREATE);
1696 security_inode_post_mknod(dir, dentry, mode, dev);
1701 asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1705 struct dentry * dentry;
1706 struct nameidata nd;
1710 tmp = getname(filename);
1712 return PTR_ERR(tmp);
1714 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1717 dentry = lookup_create(&nd, 0);
1718 error = PTR_ERR(dentry);
1719 if (!IS_POSIXACL(nd.dentry->d_inode))
1720 mode &= ~current->fs->umask;
1721 if (!IS_ERR(dentry)) {
1722 switch (mode & S_IFMT) {
1723 case 0: case S_IFREG:
1724 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1726 case S_IFCHR: case S_IFBLK:
1727 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1728 new_decode_dev(dev), &nd);
1730 case S_IFIFO: case S_IFSOCK:
1731 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1742 up(&nd.dentry->d_inode->i_sem);
1750 int vfs_mkdir(struct inode *dir, struct dentry *dentry,
1751 int mode, struct nameidata *nd)
1753 int error = may_create(dir, dentry, nd);
1758 if (!dir->i_op || !dir->i_op->mkdir)
1761 mode &= (S_IRWXUGO|S_ISVTX);
1762 error = security_inode_mkdir(dir, dentry, mode);
1767 error = dir->i_op->mkdir(dir, dentry, mode);
1769 inode_dir_notify(dir, DN_CREATE);
1770 security_inode_post_mkdir(dir,dentry, mode);
1775 asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1780 tmp = getname(pathname);
1781 error = PTR_ERR(tmp);
1783 struct dentry *dentry;
1784 struct nameidata nd;
1786 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1789 dentry = lookup_create(&nd, 1);
1790 error = PTR_ERR(dentry);
1791 if (!IS_ERR(dentry)) {
1792 if (!IS_POSIXACL(nd.dentry->d_inode))
1793 mode &= ~current->fs->umask;
1794 error = vfs_mkdir(nd.dentry->d_inode, dentry,
1798 up(&nd.dentry->d_inode->i_sem);
1808 * We try to drop the dentry early: we should have
1809 * a usage count of 2 if we're the only user of this
1810 * dentry, and if that is true (possibly after pruning
1811 * the dcache), then we drop the dentry now.
1813 * A low-level filesystem can, if it choses, legally
1816 * if (!d_unhashed(dentry))
1819 * if it cannot handle the case of removing a directory
1820 * that is still in use by something else..
1822 void dentry_unhash(struct dentry *dentry)
1825 if (atomic_read(&dentry->d_count))
1826 shrink_dcache_parent(dentry);
1827 spin_lock(&dcache_lock);
1828 spin_lock(&dentry->d_lock);
1829 if (atomic_read(&dentry->d_count) == 2)
1831 spin_unlock(&dentry->d_lock);
1832 spin_unlock(&dcache_lock);
1835 int vfs_rmdir(struct inode *dir, struct dentry *dentry,
1836 struct nameidata *nd)
1838 int error = may_delete(dir, dentry, 1, nd);
1843 if (!dir->i_op || !dir->i_op->rmdir)
1848 down(&dentry->d_inode->i_sem);
1849 dentry_unhash(dentry);
1850 if (d_mountpoint(dentry))
1853 error = security_inode_rmdir(dir, dentry);
1855 error = dir->i_op->rmdir(dir, dentry);
1857 dentry->d_inode->i_flags |= S_DEAD;
1860 up(&dentry->d_inode->i_sem);
1862 inode_dir_notify(dir, DN_DELETE);
1870 asmlinkage long sys_rmdir(const char __user * pathname)
1874 struct dentry *dentry;
1875 struct nameidata nd;
1877 name = getname(pathname);
1879 return PTR_ERR(name);
1881 error = path_lookup(name, LOOKUP_PARENT, &nd);
1885 switch(nd.last_type) {
1896 down(&nd.dentry->d_inode->i_sem);
1897 dentry = lookup_hash(&nd.last, nd.dentry);
1898 error = PTR_ERR(dentry);
1899 if (!IS_ERR(dentry)) {
1900 error = vfs_rmdir(nd.dentry->d_inode, dentry, &nd);
1903 up(&nd.dentry->d_inode->i_sem);
1911 int vfs_unlink(struct inode *dir, struct dentry *dentry,
1912 struct nameidata *nd)
1914 int error = may_delete(dir, dentry, 0, nd);
1919 if (!dir->i_op || !dir->i_op->unlink)
1924 down(&dentry->d_inode->i_sem);
1925 if (d_mountpoint(dentry))
1928 error = security_inode_unlink(dir, dentry);
1930 error = dir->i_op->unlink(dir, dentry);
1932 up(&dentry->d_inode->i_sem);
1934 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1935 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
1937 inode_dir_notify(dir, DN_DELETE);
1943 * Make sure that the actual truncation of the file will occur outside its
1944 * directory's i_sem. Truncate can take a long time if there is a lot of
1945 * writeout happening, and we don't want to prevent access to the directory
1946 * while waiting on the I/O.
1948 asmlinkage long sys_unlink(const char __user * pathname)
1952 struct dentry *dentry;
1953 struct nameidata nd;
1954 struct inode *inode = NULL;
1956 name = getname(pathname);
1958 return PTR_ERR(name);
1960 error = path_lookup(name, LOOKUP_PARENT, &nd);
1964 if (nd.last_type != LAST_NORM)
1966 down(&nd.dentry->d_inode->i_sem);
1967 dentry = lookup_hash(&nd.last, nd.dentry);
1968 error = PTR_ERR(dentry);
1969 if (!IS_ERR(dentry)) {
1970 /* Why not before? Because we want correct error value */
1971 if (nd.last.name[nd.last.len])
1973 inode = dentry->d_inode;
1975 atomic_inc(&inode->i_count);
1976 error = vfs_unlink(nd.dentry->d_inode, dentry, &nd);
1980 up(&nd.dentry->d_inode->i_sem);
1982 iput(inode); /* truncate the inode here */
1990 error = !dentry->d_inode ? -ENOENT :
1991 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1995 int vfs_symlink(struct inode *dir, struct dentry *dentry,
1996 const char *oldname, int mode, struct nameidata *nd)
1998 int error = may_create(dir, dentry, nd);
2003 if (!dir->i_op || !dir->i_op->symlink)
2006 error = security_inode_symlink(dir, dentry, oldname);
2011 error = dir->i_op->symlink(dir, dentry, oldname);
2013 inode_dir_notify(dir, DN_CREATE);
2014 security_inode_post_symlink(dir, dentry, oldname);
2019 asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
2025 from = getname(oldname);
2027 return PTR_ERR(from);
2028 to = getname(newname);
2029 error = PTR_ERR(to);
2031 struct dentry *dentry;
2032 struct nameidata nd;
2034 error = path_lookup(to, LOOKUP_PARENT, &nd);
2037 dentry = lookup_create(&nd, 0);
2038 error = PTR_ERR(dentry);
2039 if (!IS_ERR(dentry)) {
2040 error = vfs_symlink(nd.dentry->d_inode, dentry,
2041 from, S_IALLUGO, &nd);
2044 up(&nd.dentry->d_inode->i_sem);
2053 int vfs_link(struct dentry *old_dentry, struct inode *dir,
2054 struct dentry *new_dentry, struct nameidata *nd)
2056 struct inode *inode = old_dentry->d_inode;
2062 error = may_create(dir, new_dentry, nd);
2066 if (dir->i_sb != inode->i_sb)
2070 * A link to an append-only or immutable file cannot be created.
2072 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
2074 if (!dir->i_op || !dir->i_op->link)
2076 if (S_ISDIR(old_dentry->d_inode->i_mode))
2079 error = security_inode_link(old_dentry, dir, new_dentry);
2083 down(&old_dentry->d_inode->i_sem);
2085 error = dir->i_op->link(old_dentry, dir, new_dentry);
2086 up(&old_dentry->d_inode->i_sem);
2088 inode_dir_notify(dir, DN_CREATE);
2089 security_inode_post_link(old_dentry, dir, new_dentry);
2095 * Hardlinks are often used in delicate situations. We avoid
2096 * security-related surprises by not following symlinks on the
2099 * We don't follow them on the oldname either to be compatible
2100 * with linux 2.0, and to avoid hard-linking to directories
2101 * and other special files. --ADM
2103 asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
2105 struct dentry *new_dentry;
2106 struct nameidata nd, old_nd;
2110 to = getname(newname);
2114 error = __user_walk(oldname, 0, &old_nd);
2117 error = path_lookup(to, LOOKUP_PARENT, &nd);
2121 * We allow hard-links to be created to a bind-mount as long
2122 * as the bind-mount is not read-only. Checking for cross-dev
2123 * links is subsumed by the superblock check in vfs_link().
2126 if (MNT_IS_RDONLY(old_nd.mnt))
2128 new_dentry = lookup_create(&nd, 0);
2129 error = PTR_ERR(new_dentry);
2130 if (!IS_ERR(new_dentry)) {
2131 error = vfs_link(old_nd.dentry, nd.dentry->d_inode,
2135 up(&nd.dentry->d_inode->i_sem);
2139 path_release(&old_nd);
2147 * The worst of all namespace operations - renaming directory. "Perverted"
2148 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2150 * a) we can get into loop creation. Check is done in is_subdir().
2151 * b) race potential - two innocent renames can create a loop together.
2152 * That's where 4.4 screws up. Current fix: serialization on
2153 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2155 * c) we have to lock _three_ objects - parents and victim (if it exists).
2156 * And that - after we got ->i_sem on parents (until then we don't know
2157 * whether the target exists). Solution: try to be smart with locking
2158 * order for inodes. We rely on the fact that tree topology may change
2159 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2160 * move will be locked. Thus we can rank directories by the tree
2161 * (ancestors first) and rank all non-directories after them.
2162 * That works since everybody except rename does "lock parent, lookup,
2163 * lock child" and rename is under ->s_vfs_rename_sem.
2164 * HOWEVER, it relies on the assumption that any object with ->lookup()
2165 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2166 * we'd better make sure that there's no link(2) for them.
2167 * d) some filesystems don't support opened-but-unlinked directories,
2168 * either because of layout or because they are not ready to deal with
2169 * all cases correctly. The latter will be fixed (taking this sort of
2170 * stuff into VFS), but the former is not going away. Solution: the same
2171 * trick as in rmdir().
2172 * e) conversion from fhandle to dentry may come in the wrong moment - when
2173 * we are removing the target. Solution: we will have to grab ->i_sem
2174 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2175 * ->i_sem on parents, which works but leads to some truely excessive
2178 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2179 struct inode *new_dir, struct dentry *new_dentry)
2182 struct inode *target;
2185 * If we are going to change the parent - check write permissions,
2186 * we'll need to flip '..'.
2188 if (new_dir != old_dir) {
2189 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2194 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2198 target = new_dentry->d_inode;
2200 down(&target->i_sem);
2201 dentry_unhash(new_dentry);
2203 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2206 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2209 target->i_flags |= S_DEAD;
2211 if (d_unhashed(new_dentry))
2212 d_rehash(new_dentry);
2216 d_move(old_dentry,new_dentry);
2217 security_inode_post_rename(old_dir, old_dentry,
2218 new_dir, new_dentry);
2223 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2224 struct inode *new_dir, struct dentry *new_dentry)
2226 struct inode *target;
2229 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2234 target = new_dentry->d_inode;
2236 down(&target->i_sem);
2237 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2240 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2242 /* The following d_move() should become unconditional */
2243 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2244 d_move(old_dentry, new_dentry);
2245 security_inode_post_rename(old_dir, old_dentry, new_dir, new_dentry);
2253 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2254 struct inode *new_dir, struct dentry *new_dentry)
2257 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2259 if (old_dentry->d_inode == new_dentry->d_inode)
2262 error = may_delete(old_dir, old_dentry, is_dir, NULL);
2266 if (!new_dentry->d_inode)
2267 error = may_create(new_dir, new_dentry, NULL);
2269 error = may_delete(new_dir, new_dentry, is_dir, NULL);
2273 if (!old_dir->i_op || !old_dir->i_op->rename)
2276 DQUOT_INIT(old_dir);
2277 DQUOT_INIT(new_dir);
2280 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2282 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2284 if (old_dir == new_dir)
2285 inode_dir_notify(old_dir, DN_RENAME);
2287 inode_dir_notify(old_dir, DN_DELETE);
2288 inode_dir_notify(new_dir, DN_CREATE);
2294 static inline int do_rename(const char * oldname, const char * newname)
2297 struct dentry * old_dir, * new_dir;
2298 struct dentry * old_dentry, *new_dentry;
2299 struct dentry * trap;
2300 struct nameidata oldnd, newnd;
2302 error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2306 error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2311 if (oldnd.mnt != newnd.mnt)
2314 old_dir = oldnd.dentry;
2316 if (oldnd.last_type != LAST_NORM)
2319 new_dir = newnd.dentry;
2320 if (newnd.last_type != LAST_NORM)
2323 trap = lock_rename(new_dir, old_dir);
2325 old_dentry = lookup_hash(&oldnd.last, old_dir);
2326 error = PTR_ERR(old_dentry);
2327 if (IS_ERR(old_dentry))
2329 /* source must exist */
2331 if (!old_dentry->d_inode)
2333 /* unless the source is a directory trailing slashes give -ENOTDIR */
2334 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2336 if (oldnd.last.name[oldnd.last.len])
2338 if (newnd.last.name[newnd.last.len])
2341 /* source should not be ancestor of target */
2343 if (old_dentry == trap)
2346 if (MNT_IS_RDONLY(newnd.mnt))
2348 new_dentry = lookup_hash(&newnd.last, new_dir);
2349 error = PTR_ERR(new_dentry);
2350 if (IS_ERR(new_dentry))
2352 /* target should not be an ancestor of source */
2354 if (new_dentry == trap)
2357 error = vfs_rename(old_dir->d_inode, old_dentry,
2358 new_dir->d_inode, new_dentry);
2364 unlock_rename(new_dir, old_dir);
2366 path_release(&newnd);
2368 path_release(&oldnd);
2373 asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2379 from = getname(oldname);
2381 return PTR_ERR(from);
2382 to = getname(newname);
2383 error = PTR_ERR(to);
2385 error = do_rename(from,to);
2392 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2396 len = PTR_ERR(link);
2401 if (len > (unsigned) buflen)
2403 if (copy_to_user(buffer, link, len))
2410 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2411 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2412 * using) it for any given inode is up to filesystem.
2414 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2416 struct nameidata nd;
2419 res = dentry->d_inode->i_op->follow_link(dentry, &nd);
2421 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2422 if (dentry->d_inode->i_op->put_link)
2423 dentry->d_inode->i_op->put_link(dentry, &nd);
2428 int vfs_follow_link(struct nameidata *nd, const char *link)
2430 return __vfs_follow_link(nd, link);
2433 /* get the link contents into pagecache */
2434 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2437 struct address_space *mapping = dentry->d_inode->i_mapping;
2438 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2442 wait_on_page_locked(page);
2443 if (!PageUptodate(page))
2449 page_cache_release(page);
2450 return ERR_PTR(-EIO);
2456 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2458 struct page *page = NULL;
2459 char *s = page_getlink(dentry, &page);
2460 int res = vfs_readlink(dentry,buffer,buflen,s);
2463 page_cache_release(page);
2468 int page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2471 nd_set_link(nd, page_getlink(dentry, &page));
2475 void page_put_link(struct dentry *dentry, struct nameidata *nd)
2477 if (!IS_ERR(nd_get_link(nd))) {
2479 page = find_get_page(dentry->d_inode->i_mapping, 0);
2483 page_cache_release(page);
2484 page_cache_release(page);
2488 int page_symlink(struct inode *inode, const char *symname, int len)
2490 struct address_space *mapping = inode->i_mapping;
2491 struct page *page = grab_cache_page(mapping, 0);
2497 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2500 kaddr = kmap_atomic(page, KM_USER0);
2501 memcpy(kaddr, symname, len-1);
2502 kunmap_atomic(kaddr, KM_USER0);
2503 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2505 * Notice that we are _not_ going to block here - end of page is
2506 * unmapped, so this will only try to map the rest of page, see
2507 * that it is unmapped (typically even will not look into inode -
2508 * ->i_size will be enough for everything) and zero it out.
2509 * OTOH it's obviously correct and should make the page up-to-date.
2511 if (!PageUptodate(page)) {
2512 err = mapping->a_ops->readpage(NULL, page);
2513 wait_on_page_locked(page);
2517 page_cache_release(page);
2520 mark_inode_dirty(inode);
2524 page_cache_release(page);
2529 struct inode_operations page_symlink_inode_operations = {
2530 .readlink = generic_readlink,
2531 .follow_link = page_follow_link_light,
2532 .put_link = page_put_link,
2535 EXPORT_SYMBOL(__user_walk);
2536 EXPORT_SYMBOL(follow_down);
2537 EXPORT_SYMBOL(follow_up);
2538 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2539 EXPORT_SYMBOL(getname);
2540 EXPORT_SYMBOL(lock_rename);
2541 EXPORT_SYMBOL(lookup_hash);
2542 EXPORT_SYMBOL(lookup_one_len);
2543 EXPORT_SYMBOL(page_follow_link_light);
2544 EXPORT_SYMBOL(page_put_link);
2545 EXPORT_SYMBOL(page_readlink);
2546 EXPORT_SYMBOL(page_symlink);
2547 EXPORT_SYMBOL(page_symlink_inode_operations);
2548 EXPORT_SYMBOL(path_lookup);
2549 EXPORT_SYMBOL(path_release);
2550 EXPORT_SYMBOL(path_walk);
2551 EXPORT_SYMBOL(permission);
2552 EXPORT_SYMBOL(unlock_rename);
2553 EXPORT_SYMBOL(vfs_create);
2554 EXPORT_SYMBOL(vfs_follow_link);
2555 EXPORT_SYMBOL(vfs_link);
2556 EXPORT_SYMBOL(vfs_mkdir);
2557 EXPORT_SYMBOL(vfs_mknod);
2558 EXPORT_SYMBOL(generic_permission);
2559 EXPORT_SYMBOL(vfs_readlink);
2560 EXPORT_SYMBOL(vfs_rename);
2561 EXPORT_SYMBOL(vfs_rmdir);
2562 EXPORT_SYMBOL(vfs_symlink);
2563 EXPORT_SYMBOL(vfs_unlink);
2564 EXPORT_SYMBOL(dentry_unhash);
2565 EXPORT_SYMBOL(generic_readlink);