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)
256 umode_t mode = inode->i_mode;
258 if (mask & MAY_WRITE) {
260 * Nobody gets write access to a read-only fs.
262 if (IS_RDONLY(inode) &&
263 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
267 * Nobody gets write access to an immutable file.
269 if (IS_IMMUTABLE(inode))
273 /* Ordinary permission routines do not understand MAY_APPEND. */
274 submask = mask & ~MAY_APPEND;
275 if (nd && (mask & MAY_WRITE) && MNT_IS_RDONLY(nd->mnt) &&
276 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
278 if ((retval = xid_permission(inode, mask, nd)))
280 if (inode->i_op && inode->i_op->permission)
281 retval = inode->i_op->permission(inode, submask, nd);
283 retval = generic_permission(inode, submask, NULL);
287 return security_inode_permission(inode, mask, nd);
291 * get_write_access() gets write permission for a file.
292 * put_write_access() releases this write permission.
293 * This is used for regular files.
294 * We cannot support write (and maybe mmap read-write shared) accesses and
295 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
296 * can have the following values:
297 * 0: no writers, no VM_DENYWRITE mappings
298 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
299 * > 0: (i_writecount) users are writing to the file.
301 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
302 * except for the cases where we don't hold i_writecount yet. Then we need to
303 * use {get,deny}_write_access() - these functions check the sign and refuse
304 * to do the change if sign is wrong. Exclusion between them is provided by
305 * the inode->i_lock spinlock.
308 int get_write_access(struct inode * inode)
310 spin_lock(&inode->i_lock);
311 if (atomic_read(&inode->i_writecount) < 0) {
312 spin_unlock(&inode->i_lock);
315 atomic_inc(&inode->i_writecount);
316 spin_unlock(&inode->i_lock);
321 int deny_write_access(struct file * file)
323 struct inode *inode = file->f_dentry->d_inode;
325 spin_lock(&inode->i_lock);
326 if (atomic_read(&inode->i_writecount) > 0) {
327 spin_unlock(&inode->i_lock);
330 atomic_dec(&inode->i_writecount);
331 spin_unlock(&inode->i_lock);
336 void path_release(struct nameidata *nd)
343 * umount() mustn't call path_release()/mntput() as that would clear
346 void path_release_on_umount(struct nameidata *nd)
353 * Internal lookup() using the new generic dcache.
356 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
358 struct dentry * dentry = __d_lookup(parent, name);
360 /* lockess __d_lookup may fail due to concurrent d_move()
361 * in some unrelated directory, so try with d_lookup
364 dentry = d_lookup(parent, name);
366 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
367 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
376 * Short-cut version of permission(), for calling by
377 * path_walk(), when dcache lock is held. Combines parts
378 * of permission() and generic_permission(), and tests ONLY for
379 * MAY_EXEC permission.
381 * If appropriate, check DAC only. If not appropriate, or
382 * short-cut DAC fails, then call permission() to do more
383 * complete permission check.
385 static inline int exec_permission_lite(struct inode *inode,
386 struct nameidata *nd)
388 umode_t mode = inode->i_mode;
390 if (inode->i_op && inode->i_op->permission)
393 if (current->fsuid == inode->i_uid)
395 else if (in_group_p(inode->i_gid))
401 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
404 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
407 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
412 return security_inode_permission(inode, MAY_EXEC, nd);
416 * This is called when everything else fails, and we actually have
417 * to go to the low-level filesystem to find out what we should do..
419 * We get the directory semaphore, and after getting that we also
420 * make sure that nobody added the entry to the dcache in the meantime..
423 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
425 struct dentry * result;
426 struct inode *dir = parent->d_inode;
430 * First re-do the cached lookup just in case it was created
431 * while we waited for the directory semaphore..
433 * FIXME! This could use version numbering or similar to
434 * avoid unnecessary cache lookups.
436 * The "dcache_lock" is purely to protect the RCU list walker
437 * from concurrent renames at this point (we mustn't get false
438 * negatives from the RCU list walk here, unlike the optimistic
441 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
443 result = d_lookup(parent, name);
445 struct dentry * dentry = d_alloc(parent, name);
446 result = ERR_PTR(-ENOMEM);
448 result = dir->i_op->lookup(dir, dentry, nd);
459 * Uhhuh! Nasty case: the cache was re-populated while
460 * we waited on the semaphore. Need to revalidate.
463 if (result->d_op && result->d_op->d_revalidate) {
464 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
466 result = ERR_PTR(-ENOENT);
472 static int __emul_lookup_dentry(const char *, struct nameidata *);
476 walk_init_root(const char *name, struct nameidata *nd)
478 read_lock(¤t->fs->lock);
479 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
480 nd->mnt = mntget(current->fs->altrootmnt);
481 nd->dentry = dget(current->fs->altroot);
482 read_unlock(¤t->fs->lock);
483 if (__emul_lookup_dentry(name,nd))
485 read_lock(¤t->fs->lock);
487 nd->mnt = mntget(current->fs->rootmnt);
488 nd->dentry = dget(current->fs->root);
489 read_unlock(¤t->fs->lock);
493 static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
502 if (!walk_init_root(link, nd))
503 /* weird __emul_prefix() stuff did it */
506 res = link_path_walk(link, nd);
508 if (nd->depth || res || nd->last_type!=LAST_NORM)
511 * If it is an iterative symlinks resolution in open_namei() we
512 * have to copy the last component. And all that crap because of
513 * bloody create() on broken symlinks. Furrfu...
516 if (unlikely(!name)) {
520 strcpy(name, nd->last.name);
521 nd->last.name = name;
525 return PTR_ERR(link);
529 struct vfsmount *mnt;
530 struct dentry *dentry;
533 static inline int __do_follow_link(struct path *path, struct nameidata *nd)
536 struct dentry *dentry = path->dentry;
538 touch_atime(path->mnt, dentry);
539 nd_set_link(nd, NULL);
541 if (path->mnt == nd->mnt)
543 error = dentry->d_inode->i_op->follow_link(dentry, nd);
545 char *s = nd_get_link(nd);
547 error = __vfs_follow_link(nd, s);
548 if (dentry->d_inode->i_op->put_link)
549 dentry->d_inode->i_op->put_link(dentry, nd);
558 * This limits recursive symlink follows to 8, while
559 * limiting consecutive symlinks to 40.
561 * Without that kind of total limit, nasty chains of consecutive
562 * symlinks can cause almost arbitrarily long lookups.
564 static inline int do_follow_link(struct path *path, struct nameidata *nd)
567 if (current->link_count >= MAX_NESTED_LINKS)
569 if (current->total_link_count >= 40)
571 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
573 err = security_inode_follow_link(path->dentry, nd);
576 current->link_count++;
577 current->total_link_count++;
579 err = __do_follow_link(path, nd);
580 current->link_count--;
585 if (path->mnt != nd->mnt)
591 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
593 struct vfsmount *parent;
594 struct dentry *mountpoint;
595 spin_lock(&vfsmount_lock);
596 parent=(*mnt)->mnt_parent;
597 if (parent == *mnt) {
598 spin_unlock(&vfsmount_lock);
602 mountpoint=dget((*mnt)->mnt_mountpoint);
603 spin_unlock(&vfsmount_lock);
605 *dentry = mountpoint;
611 /* no need for dcache_lock, as serialization is taken care in
614 static int __follow_mount(struct path *path)
617 while (d_mountpoint(path->dentry)) {
618 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
625 path->dentry = dget(mounted->mnt_root);
631 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
633 while (d_mountpoint(*dentry)) {
634 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
640 *dentry = dget(mounted->mnt_root);
644 /* no need for dcache_lock, as serialization is taken care in
647 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
649 struct vfsmount *mounted;
651 mounted = lookup_mnt(*mnt, *dentry);
656 *dentry = dget(mounted->mnt_root);
662 static inline void follow_dotdot(struct nameidata *nd)
665 struct vfsmount *parent;
666 struct dentry *old = nd->dentry;
668 read_lock(¤t->fs->lock);
669 if (nd->dentry == current->fs->root &&
670 nd->mnt == current->fs->rootmnt) {
671 read_unlock(¤t->fs->lock);
672 /* for sane '/' avoid follow_mount() */
675 read_unlock(¤t->fs->lock);
676 spin_lock(&dcache_lock);
677 if (nd->dentry != nd->mnt->mnt_root) {
678 nd->dentry = dget(nd->dentry->d_parent);
679 spin_unlock(&dcache_lock);
683 spin_unlock(&dcache_lock);
684 spin_lock(&vfsmount_lock);
685 parent = nd->mnt->mnt_parent;
686 if (parent == nd->mnt) {
687 spin_unlock(&vfsmount_lock);
691 nd->dentry = dget(nd->mnt->mnt_mountpoint);
692 spin_unlock(&vfsmount_lock);
697 follow_mount(&nd->mnt, &nd->dentry);
701 * It's more convoluted than I'd like it to be, but... it's still fairly
702 * small and for now I'd prefer to have fast path as straight as possible.
703 * It _is_ time-critical.
705 static int do_lookup(struct nameidata *nd, struct qstr *name,
706 struct path *path, int atomic)
708 struct vfsmount *mnt = nd->mnt;
709 struct dentry *dentry = __d_lookup(nd->dentry, name);
714 if (dentry->d_op && dentry->d_op->d_revalidate)
715 goto need_revalidate;
716 inode = dentry->d_inode;
719 #ifdef CONFIG_VSERVER_FILESHARING
720 /* MEF: PlanetLab FS module assumes that any file that can be
721 * named (e.g., via a cross mount) is not hidden from another
722 * context or the admin context.
724 if (vx_check(inode->i_xid,VX_STATIC|VX_DYNAMIC|VX_ADMIN)) {
727 else /* do the following check */
729 if (!vx_check(inode->i_xid, VX_WATCH|VX_ADMIN|VX_HOSTID|VX_IDENT))
731 if (inode->i_sb->s_magic == PROC_SUPER_MAGIC) {
732 struct proc_dir_entry *de = PDE(inode);
734 if (de && !vx_hide_check(0, de->vx_flags))
739 path->dentry = dentry;
740 __follow_mount(path);
743 vxwprintk(1, "xid=%d did lookup hidden %p[#%d,%lu] »%s«.",
744 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
745 vxd_path(dentry, mnt));
751 return -EWOULDBLOCKIO;
752 dentry = real_lookup(nd->dentry, name, nd);
759 return -EWOULDBLOCKIO;
760 if (dentry->d_op->d_revalidate(dentry, nd))
762 if (d_invalidate(dentry))
768 return PTR_ERR(dentry);
773 * This is the basic name resolution function, turning a pathname into
774 * the final dentry. We expect 'base' to be positive and a directory.
776 * Returns 0 and nd will have valid dentry and mnt on success.
777 * Returns error and drops reference to input namei data on failure.
779 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
784 unsigned int lookup_flags = nd->flags;
786 atomic = (lookup_flags & LOOKUP_ATOMIC);
793 inode = nd->dentry->d_inode;
795 lookup_flags = LOOKUP_FOLLOW;
797 /* At this point we know we have a real path component. */
803 err = exec_permission_lite(inode, nd);
804 if (err == -EAGAIN) {
805 err = permission(inode, MAY_EXEC, nd);
811 c = *(const unsigned char *)name;
813 hash = init_name_hash();
816 hash = partial_name_hash(c, hash);
817 c = *(const unsigned char *)name;
818 } while (c && (c != '/'));
819 this.len = name - (const char *) this.name;
820 this.hash = end_name_hash(hash);
822 /* remove trailing slashes? */
825 while (*++name == '/');
827 goto last_with_slashes;
830 * "." and ".." are special - ".." especially so because it has
831 * to be able to know about the current root directory and
832 * parent relationships.
834 if (this.name[0] == '.') switch (this.len) {
838 if (this.name[1] != '.')
841 inode = nd->dentry->d_inode;
847 * See if the low-level filesystem might want
848 * to use its own hash..
850 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
851 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
855 nd->flags |= LOOKUP_CONTINUE;
856 /* This does the actual lookups.. */
857 err = do_lookup(nd, &this, &next, atomic);
862 inode = next.dentry->d_inode;
869 if (inode->i_op->follow_link) {
870 err = do_follow_link(&next, nd);
874 inode = nd->dentry->d_inode;
882 if (nd->mnt != next.mnt)
885 nd->dentry = next.dentry;
888 if (!inode->i_op->lookup)
891 /* here ends the main loop */
894 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
896 nd->flags &= ~LOOKUP_CONTINUE;
897 if (lookup_flags & LOOKUP_PARENT)
899 if (this.name[0] == '.') switch (this.len) {
903 if (this.name[1] != '.')
906 inode = nd->dentry->d_inode;
911 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
912 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
916 err = do_lookup(nd, &this, &next, atomic);
919 inode = next.dentry->d_inode;
920 if ((lookup_flags & LOOKUP_FOLLOW)
921 && inode && inode->i_op && inode->i_op->follow_link) {
922 err = do_follow_link(&next, nd);
925 inode = nd->dentry->d_inode;
928 if (nd->mnt != next.mnt)
931 nd->dentry = next.dentry;
936 if (lookup_flags & LOOKUP_DIRECTORY) {
938 if (!inode->i_op || !inode->i_op->lookup)
944 nd->last_type = LAST_NORM;
945 if (this.name[0] != '.')
948 nd->last_type = LAST_DOT;
949 else if (this.len == 2 && this.name[1] == '.')
950 nd->last_type = LAST_DOTDOT;
955 * We bypassed the ordinary revalidation routines.
956 * We may need to check the cached dentry for staleness.
958 if (nd->dentry && nd->dentry->d_sb &&
959 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
961 /* Note: we do not d_invalidate() */
962 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
969 if (nd->mnt != next.mnt)
979 * Wrapper to retry pathname resolution whenever the underlying
980 * file system returns an ESTALE.
982 * Retry the whole path once, forcing real lookup requests
983 * instead of relying on the dcache.
985 int fastcall link_path_walk(const char *name, struct nameidata *nd)
987 struct nameidata save = *nd;
990 /* make sure the stuff we saved doesn't go away */
994 result = __link_path_walk(name, nd);
995 if (result == -ESTALE) {
999 nd->flags |= LOOKUP_REVAL;
1000 result = __link_path_walk(name, nd);
1009 int fastcall path_walk(const char * name, struct nameidata *nd)
1011 current->total_link_count = 0;
1012 return link_path_walk(name, nd);
1016 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1017 * everything is done. Returns 0 and drops input nd, if lookup failed;
1019 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1021 if (path_walk(name, nd))
1022 return 0; /* something went wrong... */
1024 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1025 struct dentry *old_dentry = nd->dentry;
1026 struct vfsmount *old_mnt = nd->mnt;
1027 struct qstr last = nd->last;
1028 int last_type = nd->last_type;
1030 * NAME was not found in alternate root or it's a directory. Try to find
1031 * it in the normal root:
1033 nd->last_type = LAST_ROOT;
1034 read_lock(¤t->fs->lock);
1035 nd->mnt = mntget(current->fs->rootmnt);
1036 nd->dentry = dget(current->fs->root);
1037 read_unlock(¤t->fs->lock);
1038 if (path_walk(name, nd) == 0) {
1039 if (nd->dentry->d_inode) {
1046 nd->dentry = old_dentry;
1049 nd->last_type = last_type;
1054 void set_fs_altroot(void)
1056 char *emul = __emul_prefix();
1057 struct nameidata nd;
1058 struct vfsmount *mnt = NULL, *oldmnt;
1059 struct dentry *dentry = NULL, *olddentry;
1064 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1070 write_lock(¤t->fs->lock);
1071 oldmnt = current->fs->altrootmnt;
1072 olddentry = current->fs->altroot;
1073 current->fs->altrootmnt = mnt;
1074 current->fs->altroot = dentry;
1075 write_unlock(¤t->fs->lock);
1082 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1083 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
1087 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1091 read_lock(¤t->fs->lock);
1093 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1094 nd->mnt = mntget(current->fs->altrootmnt);
1095 nd->dentry = dget(current->fs->altroot);
1096 read_unlock(¤t->fs->lock);
1097 if (__emul_lookup_dentry(name,nd))
1098 goto out; /* found in altroot */
1099 read_lock(¤t->fs->lock);
1101 nd->mnt = mntget(current->fs->rootmnt);
1102 nd->dentry = dget(current->fs->root);
1104 nd->mnt = mntget(current->fs->pwdmnt);
1105 nd->dentry = dget(current->fs->pwd);
1107 read_unlock(¤t->fs->lock);
1108 current->total_link_count = 0;
1109 retval = link_path_walk(name, nd);
1111 if (unlikely(current->audit_context
1112 && nd && nd->dentry && nd->dentry->d_inode))
1113 audit_inode(name, nd->dentry->d_inode, flags);
1118 * Restricted form of lookup. Doesn't follow links, single-component only,
1119 * needs parent already locked. Doesn't follow mounts.
1122 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1124 struct dentry * dentry;
1125 struct inode *inode;
1128 inode = base->d_inode;
1129 err = permission(inode, MAY_EXEC, nd);
1130 dentry = ERR_PTR(err);
1135 * See if the low-level filesystem might want
1136 * to use its own hash..
1138 if (base->d_op && base->d_op->d_hash) {
1139 err = base->d_op->d_hash(base, name);
1140 dentry = ERR_PTR(err);
1145 dentry = cached_lookup(base, name, nd);
1147 struct dentry *new = d_alloc(base, name);
1148 dentry = ERR_PTR(-ENOMEM);
1151 dentry = inode->i_op->lookup(inode, new, nd);
1161 struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
1163 return __lookup_hash(name, base, NULL);
1167 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1178 hash = init_name_hash();
1180 c = *(const unsigned char *)name++;
1181 if (c == '/' || c == '\0')
1183 hash = partial_name_hash(c, hash);
1185 this.hash = end_name_hash(hash);
1187 return lookup_hash(&this, base);
1189 return ERR_PTR(-EACCES);
1195 * is used by most simple commands to get the inode of a specified name.
1196 * Open, link etc use their own routines, but this is enough for things
1199 * namei exists in two versions: namei/lnamei. The only difference is
1200 * that namei follows links, while lnamei does not.
1203 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1205 char *tmp = getname(name);
1206 int err = PTR_ERR(tmp);
1209 err = path_lookup(tmp, flags, nd);
1216 * It's inline, so penalty for filesystems that don't use sticky bit is
1219 static inline int check_sticky(struct inode *dir, struct inode *inode)
1221 if (!(dir->i_mode & S_ISVTX))
1223 if (inode->i_uid == current->fsuid)
1225 if (dir->i_uid == current->fsuid)
1227 return !capable(CAP_FOWNER);
1231 * Check whether we can remove a link victim from directory dir, check
1232 * whether the type of victim is right.
1233 * 1. We can't do it if dir is read-only (done in permission())
1234 * 2. We should have write and exec permissions on dir
1235 * 3. We can't remove anything from append-only dir
1236 * 4. We can't do anything with immutable dir (done in permission())
1237 * 5. If the sticky bit on dir is set we should either
1238 * a. be owner of dir, or
1239 * b. be owner of victim, or
1240 * c. have CAP_FOWNER capability
1241 * 6. If the victim is append-only or immutable we can't do antyhing with
1242 * links pointing to it.
1243 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1244 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1245 * 9. We can't remove a root or mountpoint.
1246 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1247 * nfs_async_unlink().
1249 static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1253 if (!victim->d_inode)
1256 BUG_ON(victim->d_parent->d_inode != dir);
1258 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
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);
1298 static inline int mnt_may_create(struct vfsmount *mnt, struct inode *dir, struct dentry *child) {
1301 if (IS_DEADDIR(dir))
1303 if (mnt->mnt_flags & MNT_RDONLY)
1308 static inline int mnt_may_unlink(struct vfsmount *mnt, struct inode *dir, struct dentry *child) {
1309 if (!child->d_inode)
1311 if (mnt->mnt_flags & MNT_RDONLY)
1317 * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1320 * O_DIRECTORY translates into forcing a directory lookup.
1322 static inline int lookup_flags(unsigned int f)
1324 unsigned long retval = LOOKUP_FOLLOW;
1327 retval &= ~LOOKUP_FOLLOW;
1329 if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1330 retval &= ~LOOKUP_FOLLOW;
1332 if (f & O_DIRECTORY)
1333 retval |= LOOKUP_DIRECTORY;
1334 if (f & O_ATOMICLOOKUP)
1335 retval |= LOOKUP_ATOMIC;
1341 * p1 and p2 should be directories on the same fs.
1343 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1348 down(&p1->d_inode->i_sem);
1352 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1354 for (p = p1; p->d_parent != p; p = p->d_parent) {
1355 if (p->d_parent == p2) {
1356 down(&p2->d_inode->i_sem);
1357 down(&p1->d_inode->i_sem);
1362 for (p = p2; p->d_parent != p; p = p->d_parent) {
1363 if (p->d_parent == p1) {
1364 down(&p1->d_inode->i_sem);
1365 down(&p2->d_inode->i_sem);
1370 down(&p1->d_inode->i_sem);
1371 down(&p2->d_inode->i_sem);
1375 void unlock_rename(struct dentry *p1, struct dentry *p2)
1377 up(&p1->d_inode->i_sem);
1379 up(&p2->d_inode->i_sem);
1380 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1384 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1385 struct nameidata *nd)
1387 int error = may_create(dir, dentry, nd);
1392 if (!dir->i_op || !dir->i_op->create)
1393 return -EACCES; /* shouldn't it be ENOSYS? */
1396 error = security_inode_create(dir, dentry, mode);
1400 error = dir->i_op->create(dir, dentry, mode, nd);
1402 inode_dir_notify(dir, DN_CREATE);
1403 security_inode_post_create(dir, dentry, mode);
1408 int may_open(struct nameidata *nd, int acc_mode, int flag)
1410 struct dentry *dentry = nd->dentry;
1411 struct inode *inode = dentry->d_inode;
1417 if (S_ISLNK(inode->i_mode))
1420 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1423 error = permission(inode, acc_mode, nd);
1428 * FIFO's, sockets and device files are special: they don't
1429 * actually live on the filesystem itself, and as such you
1430 * can write to them even if the filesystem is read-only.
1432 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1434 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1435 if (nd->mnt->mnt_flags & MNT_NODEV)
1439 } else if ((IS_RDONLY(inode) || (nd && MNT_IS_RDONLY(nd->mnt)))
1440 && (flag & FMODE_WRITE))
1443 * An append-only file must be opened in append mode for writing.
1445 if (IS_APPEND(inode)) {
1446 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1452 /* O_NOATIME can only be set by the owner or superuser */
1453 if (flag & O_NOATIME)
1454 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1458 * Ensure there are no outstanding leases on the file.
1460 error = break_lease(inode, flag);
1464 if (flag & O_TRUNC) {
1465 error = get_write_access(inode);
1470 * Refuse to truncate files with mandatory locks held on them.
1472 error = locks_verify_locked(inode);
1476 error = do_truncate(dentry, 0);
1478 put_write_access(inode);
1482 if (flag & FMODE_WRITE)
1491 * namei for open - this is in fact almost the whole open-routine.
1493 * Note that the low bits of "flag" aren't the same as in the open
1494 * system call - they are 00 - no permissions needed
1495 * 01 - read permission needed
1496 * 10 - write permission needed
1497 * 11 - read/write permissions needed
1498 * which is a lot more logical, and also allows the "no perm" needed
1499 * for symlinks (where the permissions are checked later).
1502 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1504 int acc_mode, error = 0;
1509 acc_mode = ACC_MODE(flag);
1511 /* Allow the LSM permission hook to distinguish append
1512 access from general write access. */
1513 if (flag & O_APPEND)
1514 acc_mode |= MAY_APPEND;
1516 /* Fill in the open() intent data */
1517 nd->intent.open.flags = flag;
1518 nd->intent.open.create_mode = mode;
1521 * The simplest case - just a plain lookup.
1523 if (!(flag & O_CREAT)) {
1524 error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd);
1531 * Create - we need to know the parent.
1533 error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd);
1538 * We have the parent and last component. First of all, check
1539 * that we are not asked to creat(2) an obvious directory - that
1543 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1547 nd->flags &= ~LOOKUP_PARENT;
1548 down(&dir->d_inode->i_sem);
1549 path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1553 error = PTR_ERR(path.dentry);
1554 if (IS_ERR(path.dentry)) {
1555 up(&dir->d_inode->i_sem);
1559 /* Negative dentry, just create the file */
1560 if (!path.dentry->d_inode) {
1561 if (!IS_POSIXACL(dir->d_inode))
1562 mode &= ~current->fs->umask;
1563 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1564 up(&dir->d_inode->i_sem);
1566 nd->dentry = path.dentry;
1569 /* Don't check for write permission, don't truncate */
1576 * It already exists.
1578 up(&dir->d_inode->i_sem);
1584 if (__follow_mount(&path)) {
1586 if (flag & O_NOFOLLOW)
1590 if (!path.dentry->d_inode)
1592 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1596 nd->dentry = path.dentry;
1597 if (nd->mnt != path.mnt)
1601 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1604 error = may_open(nd, acc_mode, flag);
1611 if (nd->mnt != path.mnt)
1619 if (flag & O_NOFOLLOW)
1622 * This is subtle. Instead of calling do_follow_link() we do the
1623 * thing by hands. The reason is that this way we have zero link_count
1624 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1625 * After that we have the parent and last component, i.e.
1626 * we are in the same situation as after the first path_walk().
1627 * Well, almost - if the last component is normal we get its copy
1628 * stored in nd->last.name and we will have to putname() it when we
1629 * are done. Procfs-like symlinks just set LAST_BIND.
1631 nd->flags |= LOOKUP_PARENT;
1632 error = security_inode_follow_link(path.dentry, nd);
1635 error = __do_follow_link(&path, nd);
1638 nd->flags &= ~LOOKUP_PARENT;
1639 if (nd->last_type == LAST_BIND)
1642 if (nd->last_type != LAST_NORM)
1644 if (nd->last.name[nd->last.len]) {
1645 putname(nd->last.name);
1650 putname(nd->last.name);
1654 down(&dir->d_inode->i_sem);
1655 path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1657 putname(nd->last.name);
1662 * lookup_create - lookup a dentry, creating it if it doesn't exist
1663 * @nd: nameidata info
1664 * @is_dir: directory flag
1666 * Simple function to lookup and return a dentry and create it
1667 * if it doesn't exist. Is SMP-safe.
1669 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1671 struct dentry *dentry;
1674 down(&nd->dentry->d_inode->i_sem);
1676 if (nd->last_type != LAST_NORM)
1678 nd->flags &= ~LOOKUP_PARENT;
1679 dentry = lookup_hash(&nd->last, nd->dentry);
1682 error = mnt_may_create(nd->mnt, nd->dentry->d_inode, dentry);
1686 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1693 return ERR_PTR(error);
1695 EXPORT_SYMBOL_GPL(lookup_create);
1697 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1699 int error = may_create(dir, dentry, NULL);
1704 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1707 if (!dir->i_op || !dir->i_op->mknod)
1710 error = security_inode_mknod(dir, dentry, mode, dev);
1715 error = dir->i_op->mknod(dir, dentry, mode, dev);
1717 inode_dir_notify(dir, DN_CREATE);
1718 security_inode_post_mknod(dir, dentry, mode, dev);
1723 asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1727 struct dentry * dentry;
1728 struct nameidata nd;
1732 tmp = getname(filename);
1734 return PTR_ERR(tmp);
1736 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1739 dentry = lookup_create(&nd, 0);
1740 error = PTR_ERR(dentry);
1742 if (!IS_POSIXACL(nd.dentry->d_inode))
1743 mode &= ~current->fs->umask;
1744 if (!IS_ERR(dentry)) {
1745 switch (mode & S_IFMT) {
1746 case 0: case S_IFREG:
1747 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1749 case S_IFCHR: case S_IFBLK:
1750 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1751 new_decode_dev(dev));
1753 case S_IFIFO: case S_IFSOCK:
1754 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1764 up(&nd.dentry->d_inode->i_sem);
1772 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1774 int error = may_create(dir, dentry, NULL);
1779 if (!dir->i_op || !dir->i_op->mkdir)
1782 mode &= (S_IRWXUGO|S_ISVTX);
1783 error = security_inode_mkdir(dir, dentry, mode);
1788 error = dir->i_op->mkdir(dir, dentry, mode);
1790 inode_dir_notify(dir, DN_CREATE);
1791 security_inode_post_mkdir(dir,dentry, mode);
1796 asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1801 tmp = getname(pathname);
1802 error = PTR_ERR(tmp);
1804 struct dentry *dentry;
1805 struct nameidata nd;
1807 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1810 dentry = lookup_create(&nd, 1);
1811 error = PTR_ERR(dentry);
1812 if (!IS_ERR(dentry)) {
1813 if (!IS_POSIXACL(nd.dentry->d_inode))
1814 mode &= ~current->fs->umask;
1815 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1818 up(&nd.dentry->d_inode->i_sem);
1828 * We try to drop the dentry early: we should have
1829 * a usage count of 2 if we're the only user of this
1830 * dentry, and if that is true (possibly after pruning
1831 * the dcache), then we drop the dentry now.
1833 * A low-level filesystem can, if it choses, legally
1836 * if (!d_unhashed(dentry))
1839 * if it cannot handle the case of removing a directory
1840 * that is still in use by something else..
1842 void dentry_unhash(struct dentry *dentry)
1845 if (atomic_read(&dentry->d_count))
1846 shrink_dcache_parent(dentry);
1847 spin_lock(&dcache_lock);
1848 spin_lock(&dentry->d_lock);
1849 if (atomic_read(&dentry->d_count) == 2)
1851 spin_unlock(&dentry->d_lock);
1852 spin_unlock(&dcache_lock);
1855 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1857 int error = may_delete(dir, dentry, 1);
1862 if (!dir->i_op || !dir->i_op->rmdir)
1867 down(&dentry->d_inode->i_sem);
1868 dentry_unhash(dentry);
1869 if (d_mountpoint(dentry))
1872 error = security_inode_rmdir(dir, dentry);
1874 error = dir->i_op->rmdir(dir, dentry);
1876 dentry->d_inode->i_flags |= S_DEAD;
1879 up(&dentry->d_inode->i_sem);
1881 inode_dir_notify(dir, DN_DELETE);
1889 asmlinkage long sys_rmdir(const char __user * pathname)
1893 struct dentry *dentry;
1894 struct nameidata nd;
1896 name = getname(pathname);
1898 return PTR_ERR(name);
1900 error = path_lookup(name, LOOKUP_PARENT, &nd);
1904 switch(nd.last_type) {
1915 down(&nd.dentry->d_inode->i_sem);
1916 dentry = lookup_hash(&nd.last, nd.dentry);
1917 error = PTR_ERR(dentry);
1918 if (!IS_ERR(dentry)) {
1919 error = mnt_may_unlink(nd.mnt, nd.dentry->d_inode, dentry);
1922 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1926 up(&nd.dentry->d_inode->i_sem);
1934 int vfs_unlink(struct inode *dir, struct dentry *dentry)
1936 int error = may_delete(dir, dentry, 0);
1941 if (!dir->i_op || !dir->i_op->unlink)
1946 down(&dentry->d_inode->i_sem);
1947 if (d_mountpoint(dentry))
1950 error = security_inode_unlink(dir, dentry);
1952 error = dir->i_op->unlink(dir, dentry);
1954 up(&dentry->d_inode->i_sem);
1956 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1957 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
1959 inode_dir_notify(dir, DN_DELETE);
1965 * Make sure that the actual truncation of the file will occur outside its
1966 * directory's i_sem. Truncate can take a long time if there is a lot of
1967 * writeout happening, and we don't want to prevent access to the directory
1968 * while waiting on the I/O.
1970 asmlinkage long sys_unlink(const char __user * pathname)
1974 struct dentry *dentry;
1975 struct nameidata nd;
1976 struct inode *inode = NULL;
1978 name = getname(pathname);
1980 return PTR_ERR(name);
1982 error = path_lookup(name, LOOKUP_PARENT, &nd);
1986 if (nd.last_type != LAST_NORM)
1988 down(&nd.dentry->d_inode->i_sem);
1989 dentry = lookup_hash(&nd.last, nd.dentry);
1990 error = PTR_ERR(dentry);
1991 if (!IS_ERR(dentry)) {
1992 /* Why not before? Because we want correct error value */
1993 if (nd.last.name[nd.last.len])
1995 error = mnt_may_unlink(nd.mnt, nd.dentry->d_inode, dentry);
1998 inode = dentry->d_inode;
2000 atomic_inc(&inode->i_count);
2001 error = vfs_unlink(nd.dentry->d_inode, dentry);
2005 up(&nd.dentry->d_inode->i_sem);
2007 iput(inode); /* truncate the inode here */
2015 error = !dentry->d_inode ? -ENOENT :
2016 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2020 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2022 int error = may_create(dir, dentry, NULL);
2027 if (!dir->i_op || !dir->i_op->symlink)
2030 error = security_inode_symlink(dir, dentry, oldname);
2035 error = dir->i_op->symlink(dir, dentry, oldname);
2037 inode_dir_notify(dir, DN_CREATE);
2038 security_inode_post_symlink(dir, dentry, oldname);
2043 asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
2049 from = getname(oldname);
2051 return PTR_ERR(from);
2052 to = getname(newname);
2053 error = PTR_ERR(to);
2055 struct dentry *dentry;
2056 struct nameidata nd;
2058 error = path_lookup(to, LOOKUP_PARENT, &nd);
2061 dentry = lookup_create(&nd, 0);
2062 error = PTR_ERR(dentry);
2063 if (!IS_ERR(dentry)) {
2064 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2067 up(&nd.dentry->d_inode->i_sem);
2076 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2078 struct inode *inode = old_dentry->d_inode;
2084 error = may_create(dir, new_dentry, NULL);
2088 if (dir->i_sb != inode->i_sb)
2092 * A link to an append-only or immutable file cannot be created.
2094 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
2096 if (!dir->i_op || !dir->i_op->link)
2098 if (S_ISDIR(old_dentry->d_inode->i_mode))
2101 error = security_inode_link(old_dentry, dir, new_dentry);
2105 down(&old_dentry->d_inode->i_sem);
2107 error = dir->i_op->link(old_dentry, dir, new_dentry);
2108 up(&old_dentry->d_inode->i_sem);
2110 inode_dir_notify(dir, DN_CREATE);
2111 security_inode_post_link(old_dentry, dir, new_dentry);
2117 * Hardlinks are often used in delicate situations. We avoid
2118 * security-related surprises by not following symlinks on the
2121 * We don't follow them on the oldname either to be compatible
2122 * with linux 2.0, and to avoid hard-linking to directories
2123 * and other special files. --ADM
2125 asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
2127 struct dentry *new_dentry;
2128 struct nameidata nd, old_nd;
2132 to = getname(newname);
2136 error = __user_walk(oldname, 0, &old_nd);
2139 error = path_lookup(to, LOOKUP_PARENT, &nd);
2143 * We allow hard-links to be created to a bind-mount as long
2144 * as the bind-mount is not read-only. Checking for cross-dev
2145 * links is subsumed by the superblock check in vfs_link().
2148 if (MNT_IS_RDONLY(old_nd.mnt))
2150 new_dentry = lookup_create(&nd, 0);
2151 error = PTR_ERR(new_dentry);
2152 if (!IS_ERR(new_dentry)) {
2153 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2156 up(&nd.dentry->d_inode->i_sem);
2160 path_release(&old_nd);
2168 * The worst of all namespace operations - renaming directory. "Perverted"
2169 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2171 * a) we can get into loop creation. Check is done in is_subdir().
2172 * b) race potential - two innocent renames can create a loop together.
2173 * That's where 4.4 screws up. Current fix: serialization on
2174 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2176 * c) we have to lock _three_ objects - parents and victim (if it exists).
2177 * And that - after we got ->i_sem on parents (until then we don't know
2178 * whether the target exists). Solution: try to be smart with locking
2179 * order for inodes. We rely on the fact that tree topology may change
2180 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2181 * move will be locked. Thus we can rank directories by the tree
2182 * (ancestors first) and rank all non-directories after them.
2183 * That works since everybody except rename does "lock parent, lookup,
2184 * lock child" and rename is under ->s_vfs_rename_sem.
2185 * HOWEVER, it relies on the assumption that any object with ->lookup()
2186 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2187 * we'd better make sure that there's no link(2) for them.
2188 * d) some filesystems don't support opened-but-unlinked directories,
2189 * either because of layout or because they are not ready to deal with
2190 * all cases correctly. The latter will be fixed (taking this sort of
2191 * stuff into VFS), but the former is not going away. Solution: the same
2192 * trick as in rmdir().
2193 * e) conversion from fhandle to dentry may come in the wrong moment - when
2194 * we are removing the target. Solution: we will have to grab ->i_sem
2195 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2196 * ->i_sem on parents, which works but leads to some truely excessive
2199 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2200 struct inode *new_dir, struct dentry *new_dentry)
2203 struct inode *target;
2206 * If we are going to change the parent - check write permissions,
2207 * we'll need to flip '..'.
2209 if (new_dir != old_dir) {
2210 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2215 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2219 target = new_dentry->d_inode;
2221 down(&target->i_sem);
2222 dentry_unhash(new_dentry);
2224 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2227 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2230 target->i_flags |= S_DEAD;
2232 if (d_unhashed(new_dentry))
2233 d_rehash(new_dentry);
2237 d_move(old_dentry,new_dentry);
2238 security_inode_post_rename(old_dir, old_dentry,
2239 new_dir, new_dentry);
2244 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2245 struct inode *new_dir, struct dentry *new_dentry)
2247 struct inode *target;
2250 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2255 target = new_dentry->d_inode;
2257 down(&target->i_sem);
2258 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2261 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2263 /* The following d_move() should become unconditional */
2264 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2265 d_move(old_dentry, new_dentry);
2266 security_inode_post_rename(old_dir, old_dentry, new_dir, new_dentry);
2274 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2275 struct inode *new_dir, struct dentry *new_dentry)
2278 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2280 if (old_dentry->d_inode == new_dentry->d_inode)
2283 error = may_delete(old_dir, old_dentry, is_dir);
2287 if (!new_dentry->d_inode)
2288 error = may_create(new_dir, new_dentry, NULL);
2290 error = may_delete(new_dir, new_dentry, is_dir);
2294 if (!old_dir->i_op || !old_dir->i_op->rename)
2297 DQUOT_INIT(old_dir);
2298 DQUOT_INIT(new_dir);
2301 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2303 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2305 if (old_dir == new_dir)
2306 inode_dir_notify(old_dir, DN_RENAME);
2308 inode_dir_notify(old_dir, DN_DELETE);
2309 inode_dir_notify(new_dir, DN_CREATE);
2315 static inline int do_rename(const char * oldname, const char * newname)
2318 struct dentry * old_dir, * new_dir;
2319 struct dentry * old_dentry, *new_dentry;
2320 struct dentry * trap;
2321 struct nameidata oldnd, newnd;
2323 error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2327 error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2332 if (oldnd.mnt != newnd.mnt)
2335 old_dir = oldnd.dentry;
2337 if (oldnd.last_type != LAST_NORM)
2340 new_dir = newnd.dentry;
2341 if (newnd.last_type != LAST_NORM)
2344 trap = lock_rename(new_dir, old_dir);
2346 old_dentry = lookup_hash(&oldnd.last, old_dir);
2347 error = PTR_ERR(old_dentry);
2348 if (IS_ERR(old_dentry))
2350 /* source must exist */
2352 if (!old_dentry->d_inode)
2354 /* unless the source is a directory trailing slashes give -ENOTDIR */
2355 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2357 if (oldnd.last.name[oldnd.last.len])
2359 if (newnd.last.name[newnd.last.len])
2362 /* source should not be ancestor of target */
2364 if (old_dentry == trap)
2367 if (MNT_IS_RDONLY(newnd.mnt))
2369 new_dentry = lookup_hash(&newnd.last, new_dir);
2370 error = PTR_ERR(new_dentry);
2371 if (IS_ERR(new_dentry))
2373 /* target should not be an ancestor of source */
2375 if (new_dentry == trap)
2378 error = vfs_rename(old_dir->d_inode, old_dentry,
2379 new_dir->d_inode, new_dentry);
2385 unlock_rename(new_dir, old_dir);
2387 path_release(&newnd);
2389 path_release(&oldnd);
2394 asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2400 from = getname(oldname);
2402 return PTR_ERR(from);
2403 to = getname(newname);
2404 error = PTR_ERR(to);
2406 error = do_rename(from,to);
2413 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2417 len = PTR_ERR(link);
2422 if (len > (unsigned) buflen)
2424 if (copy_to_user(buffer, link, len))
2431 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2432 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2433 * using) it for any given inode is up to filesystem.
2435 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2437 struct nameidata nd;
2440 res = dentry->d_inode->i_op->follow_link(dentry, &nd);
2442 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2443 if (dentry->d_inode->i_op->put_link)
2444 dentry->d_inode->i_op->put_link(dentry, &nd);
2449 int vfs_follow_link(struct nameidata *nd, const char *link)
2451 return __vfs_follow_link(nd, link);
2454 /* get the link contents into pagecache */
2455 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2458 struct address_space *mapping = dentry->d_inode->i_mapping;
2459 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2463 wait_on_page_locked(page);
2464 if (!PageUptodate(page))
2470 page_cache_release(page);
2471 return ERR_PTR(-EIO);
2477 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2479 struct page *page = NULL;
2480 char *s = page_getlink(dentry, &page);
2481 int res = vfs_readlink(dentry,buffer,buflen,s);
2484 page_cache_release(page);
2489 int page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2492 nd_set_link(nd, page_getlink(dentry, &page));
2496 void page_put_link(struct dentry *dentry, struct nameidata *nd)
2498 if (!IS_ERR(nd_get_link(nd))) {
2500 page = find_get_page(dentry->d_inode->i_mapping, 0);
2504 page_cache_release(page);
2505 page_cache_release(page);
2509 int page_symlink(struct inode *inode, const char *symname, int len)
2511 struct address_space *mapping = inode->i_mapping;
2512 struct page *page = grab_cache_page(mapping, 0);
2518 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2521 kaddr = kmap_atomic(page, KM_USER0);
2522 memcpy(kaddr, symname, len-1);
2523 kunmap_atomic(kaddr, KM_USER0);
2524 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2526 * Notice that we are _not_ going to block here - end of page is
2527 * unmapped, so this will only try to map the rest of page, see
2528 * that it is unmapped (typically even will not look into inode -
2529 * ->i_size will be enough for everything) and zero it out.
2530 * OTOH it's obviously correct and should make the page up-to-date.
2532 if (!PageUptodate(page)) {
2533 err = mapping->a_ops->readpage(NULL, page);
2534 wait_on_page_locked(page);
2538 page_cache_release(page);
2541 mark_inode_dirty(inode);
2545 page_cache_release(page);
2550 struct inode_operations page_symlink_inode_operations = {
2551 .readlink = generic_readlink,
2552 .follow_link = page_follow_link_light,
2553 .put_link = page_put_link,
2556 EXPORT_SYMBOL(__user_walk);
2557 EXPORT_SYMBOL(follow_down);
2558 EXPORT_SYMBOL(follow_up);
2559 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2560 EXPORT_SYMBOL(getname);
2561 EXPORT_SYMBOL(lock_rename);
2562 EXPORT_SYMBOL(lookup_hash);
2563 EXPORT_SYMBOL(lookup_one_len);
2564 EXPORT_SYMBOL(page_follow_link_light);
2565 EXPORT_SYMBOL(page_put_link);
2566 EXPORT_SYMBOL(page_readlink);
2567 EXPORT_SYMBOL(page_symlink);
2568 EXPORT_SYMBOL(page_symlink_inode_operations);
2569 EXPORT_SYMBOL(path_lookup);
2570 EXPORT_SYMBOL(path_release);
2571 EXPORT_SYMBOL(path_walk);
2572 EXPORT_SYMBOL(permission);
2573 EXPORT_SYMBOL(unlock_rename);
2574 EXPORT_SYMBOL(vfs_create);
2575 EXPORT_SYMBOL(vfs_follow_link);
2576 EXPORT_SYMBOL(vfs_link);
2577 EXPORT_SYMBOL(vfs_mkdir);
2578 EXPORT_SYMBOL(vfs_mknod);
2579 EXPORT_SYMBOL(generic_permission);
2580 EXPORT_SYMBOL(vfs_readlink);
2581 EXPORT_SYMBOL(vfs_rename);
2582 EXPORT_SYMBOL(vfs_rmdir);
2583 EXPORT_SYMBOL(vfs_symlink);
2584 EXPORT_SYMBOL(vfs_unlink);
2585 EXPORT_SYMBOL(dentry_unhash);
2586 EXPORT_SYMBOL(generic_readlink);